ANNÉE 2017

THÈSE / UNIVERSITÉ DE RENNES 1

sous le sceau de l’Université Bretagne Loire

pour le grade de
DOCTEUR DE L’UNIVERSITÉ DE RENNES 1
Mention : Sciences Économiques

École doctorale Sciences de l’Homme des Organisations

et de la Société (SHOS)

Ewen Gallic
Préparée à l’unité de recherche CREM (UMR CNRS 6211)
Centre de Recherche en Économie et Management
Faculté de Sciences Économiques

Climate Change and Thèse soutenue à Rennes

le 2 Juin 2017
Agriculture devant le jury composé de :
Jean-Paul Chavas
Anderson-Bascom Professor, University of Wisconsin
Rapporteur
Alban Thomas
Directeur de recherche, INRA, Toulouse School of
Economics
Rapporteur
Pascale Combes Motel
Professeur, Université d’Auvergne
Examinateur
Jean-Christophe Poutineau
Professeur, Université de Rennes 1
Examinateur
Katheline Schubert
Professeur, Paris School of Economics, Université
Paris 1 Panthéon–Sorbonne
Examinateur
Catherine Benjamin
Professeur, Université de Rennes 1
Directrice de thèse
This Ph.D. thesis should not be reported as representing the views of University of
Rennes 1. The views expressed are those of the author and do not necessarily reflect
those of the University.

L’Université de Rennes 1 n’entend donner aucune approbation ni improbation aux

opinions émises dans cette thèse. Ces opinions doivent être considérées comme pro-
pres à leur auteur.
Around the world, climate change is an existential threat – but if we harness the oppor-
tunities inherent in addressing climate change, we can reap enormous economic benefits

Ban Ki-moon, April 2014

 Remerciements
Comme le veut la coutume, j’aimerais remercier ceux qui ont participé de près ou
de loin à l’aboutissement de ce travail. Cet exercice me semble difficile, non pas par
absence de reconnaissance, bien au contraire, mais par crainte d’oublier de mentionner
ceux qui m’ont accompagné au long de ces dernières années. À ceux-là, je demande
de bien vouloir me pardonner.

Tout d’abord, je tiens à faire part de ma profonde reconnaissance à Catherine Benja-

min, qui a accepté d’encadrer cette thèse. Merci Catherine pour tous vos conseils, votre
soutien et pour la confiance placée en moi.

Merci à Jean-Paul Chavas, Alban Thomas, Pascale Combes-Motel, Jean-Christophe

Poutineau et Katheline Schubert de m’avoir fait l’honneur d’être membre du jury de
cette thèse.

Ma reconnaissance va à Isabelle Cadoret et Valérie Monbet, qui ont toutes deux permis
à cette thèse de débuter.

Je remercie l’école doctorale SHOS, ses présidents Franck Moraux et Jean-René Binet,
ainsi que ses gestionnaires Alexandrine Belin Brunel et Pasquale Breger, pour le cadre
de formation.

Le CREM m’a offert un excellent cadre de travail. Je tiens à exprimer ma reconnais-

sance aux directeurs actuel et passés du laboratoire, Yvon Rocaboy, Vincent Merlin et
Franck Moraux, ainsi qu’à leurs représentants de la place Hoche, Fabien Moizeau et
David Masclet. L’environnement proposé par le CREM m’a permis de rencontrer des
personnes formidables. J’ai une pensée particulière pour Marie-Hélène Hubert grâce
à qui j’ai eu la chance d’assister à de nombreux séminaires très enrichissants en éco-
nomie environnementale, et grâce à qui j’ai pu participer à des groupes de travail et
conférences. Parmi les nombreuses personnes passionnantes et passionnées croisées
au sein du CREM, j’aimerais également adresser ma reconnaissance à Olivier L’Hari-
don, pour son expertise, sa bienveillance, ainsi que ses regards éclairés sur le monde.
Je n’oublierai jamais la conversation que nous avons eue face au vent glacial alpin à
propos de la générosité, et espère pouvoir en appliquer le contenu tout au long de ma
vie. J’ai une pensée pour Elven Priour, “l’éternel doctorant”. Le laboratoire ne pour-
rait fonctionner sans tous ses membres administratifs, que je salue également pour
leur professionnalisme et leur efficacité : Anne l’Azou, Hélène Coda-Poirey, Julie Le
Diraison, Cécile Madoulet, Daniele Moret-Bailly.

Mon parcours est étroitement lié à l’Université de Rennes 1, qui après m’avoir formé
jusqu’au Master, m’a accordé une allocation de recherche puis un contrat d’ATER. Je
salue l’ensemble des enseignants qui m’ont transmis leurs connaissances depuis mes
deux premières années jusqu’au master. En particulier, je souhaite remercier celles
et ceux qui m’ont aiguillé lors de mon changement d’orientation de l’informatique
vers l’économie et les statistiques, à savoir Nathalie Colombier, Sophie Larribeau, et
Raphaël Suire. Il me serait impossible de ne pas adresser ma sympathie à Chantal Gué-
guen qui aura toujours veillé à me proposer un service compatible avec un emploi du

vii
viii Remerciements

temps partagé entre la recherche et l’enseignement, tout en me proposant une variété

dans les enseignements. Merci également à Thierry Karcher pour les TDs de statis-
tique. Un grand merci à Alice Bernard pour la relecture de l’introduction.

Qu’on me permette de mentionner la bienveillance de l’équipe administrative de la Fa-

culté de Sciences Économiques : Aurélie Allais, Muriel Benoist, Marc Borsatto, Annick
Bougeant, Sylvie Bourset, Karine Bulourde, Jocelyne Camilleri, Sylvie Chanteux, Gilles
Conan, Karine Cousin, Nadine Desodt, Angélique Dubois, Fabienne Ealet, Maddy He-
lou, Lean-Louis Le Fessant, Marie-France Lechesne, Caroline Lemoine, Sylvie Lorho-
Chevalerias, Yvelise Maret, Soizic Masson, Françoise Mazzoleni, Jean-Willy Mbuko,
Jamele Mezali, Denis Noël, Carine Philippot, Annie Quéméré, Corinne Rémondin, Cla-
risse Turuban, Christèle Viel et Annie Vignon.

Je songe également avec reconnaissance à tous les employés du restaurant universi-

taire qui m’ont nourri pendant plusieurs années, six jours sur sept, parfois midi et soir.
Merci à Claudine, Fred & Fred, Jean-Yves, Isabelle, Laurence, Lionnel et Nathalie. Vos
sourires quotidiens m’ont transmis une belle énergie.

Je souhaite remercier mes co-auteurs pour leurs discussions fructueuses. Merci à Fran-
çois Briatte, dont la capacité à travailler sans relâche sur de nombreux sujets plus inté-
ressants les uns que les autres m’inspire. Un grand merci à Jean-Christophe Poutineau,
qui m’impressionne par la clarté avec laquelle il parvient à poser un cadre d’analyse ri-
goureux, permettant d’atteindre sans peine un résultat final précis. Ses éclairages sur la
recherche, sur le monde universitaire comme sur le monde de la musique m’honorent,
et je lui suis très reconnaissant pour toute l’aide qu’il a pu m’apporter.

I would also like to thank all the participants of the seminars, workshops and confe-
rences for their helpful comments and suggestions, in particular, Stéphane Adjemian,
Zouhair Ait Benhamou, Dorothée Charlier, Pascale Combes Motel, Jean-Louis Combes,
Jean-Charles Garibal, Frédéric Karamé, Robert Kollmann, François Langot, Mengdi
Liu, Jean-Christophe Poutineau, Tovonony Razafindrabe, Christophe Tavéra, Chris-
tian Traeger, Roland C. Winkler, Stefanos Xenarios.

Parmi toutes les personnes qui m’ont accompagné durant cette aventure, il en est une
dont le nom pourrait figurer dans la plupart des paragraphes précédents, celui d’Ar-
thur Charpentier. Je ne sais par où commencer pour le remercier, tellement je lui suis
reconnaissant et redevable. Merci pour tous les conseils, les discussions, les travaux
de recherche, les séminaires, les cours, les projets et les personnes que tu m’as fait
rencontrer. Je te dois beaucoup et ne saurai jamais assez te témoigner ma gratitude.

La thèse est une formidable aventure humaine, durant laquelle j’ai partagé de nom-
breux moments mémorables avec l’ensemble des doctorants et membres de l’associa-
tion PROJECT, notamment Guillaume Beaurain, Henri Busson, Thibaud Cargoët, Tho-
mas Cadet, Clément Dheilly, Nicolas Gavoille, Charles Hamon, Ons Jedidi, Maxime
Lamandé, Gabin Langevin, Thao Nguyen, Emmanuel Peterle, Guillaume Queffelec, Si-
riki Coulibaly, Zhang Jiang. J’adresse une pensée à mes trois amis des bureaux 104 et
108, pour qui le mot “solidarité” prend tout son sens. May Atef Aly Sayd Ahmed At-
tallah, Romain Gaté et Thi Thanh Xuan Tran ; c’est un privilège d’avoir passé tant de
temps à vos côtés. Puisse notre amitié perdurer au-delà de cette thèse. J’ajoute un mot
spécialement destiné à Xuan, pour la remercier de tout ce qu’elle a pu me raconter sur
le Vietnam, et pour toutes les rencontres avec la communauté, notamment avec Uyen
Tran. Last but not least, mes pensées s’adressent aux membres de mon premier bureau,
Gauthier Vermandel et Pascaline Vincent, avec qui j’ai sûrement passé les meilleures
années de ma vie. Je suis persuadé que nos routes n’en finiront pas de se croiser, et
que nous saurons préserver ces liens d’amitié. Merci de votre soutien. Gauthier, merci
énormément pour tous tes encouragements, pour ta mansuétude, et pour l’exemple
que tu donnes. Tout ce que tu as fait pour moi restera gravé dans ma mémoire.

Mes pensées vont enfin à mes amis et ma famille. Alexis, Célia, Hicham, Isabelle &
Kévin, Jacqueline, Jérôme, Julien-Yacine, Martin, Stéphanie & Laurent, Sonia. Je songe
à vous et vous suis reconnaissant d’avoir supporté mes enthousiasmes comme mes dé-
couragements. Enfin, je souhaite remercier du fond du cœur ma famille, qui n’aura eu
de cesse de me soutenir. Merci à Henri, pour sa grande compréhension, et mes excuses
pour le peu de temps passé avec lui. J’ai une pensée pour Éliane, qui par sa force m’aura
inspiré à de nombreuses reprises pour surmonter les obstacles. Je remercie ma mère et
mon père d’avoir évité de me poser l’éternelle question adressée aux doctorants : “Et
la thèse, ça avance ?” et d’avoir plutôt formulé leur infaillible soutien par l’emploi des
mots suivants : “On ne te demande pas où tu en es rendu, mais on pense bien à toi”. Merci
à tous les deux, pour votre amour indéfectible. Il me reste une personne à remercier,
Claire, ma sœur, qui ne cessera d’être l’exemple que j’essaie de suivre, et sans laquelle
je n’aurais sûrement jamais suivi de cours d’économie.
Contents

Contents xi

Résumé en français xv

General Introduction 1
1 Climate is Changing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Modelling the Consequences of the Weather and Climate Change on
Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 The Agricultural Production Function . . . . . . . . . . . . . . 7
2.2 The Ricardian Framework . . . . . . . . . . . . . . . . . . . . . 10
2.3 A Global Analysis of the Economic Impacts . . . . . . . . . . . 12
3 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

I Climate Change in Developing Countries: the Indian Case 19

1 Climate Change and Profits: a Ricardian Analysis 23
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 Modeling Net Revenue . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1 Behavioural Model: the Role of Climatic Variables . . . . . . . 26
2.2 The Empirical Approach: the Use of Quantile Regression . . . . 28
3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1 Household Survey . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Measure of the Climatic Variables . . . . . . . . . . . . . . . . . 31
3.3 Geographic and Soil Characteristics . . . . . . . . . . . . . . . 31
3.4 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . 32
4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Climate Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Other Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2.1 Personal Characteristics . . . . . . . . . . . . . . . . 37
4.2.2 Farm Characteristics . . . . . . . . . . . . . . . . . . 38
4.2.3 Irrigation . . . . . . . . . . . . . . . . . . . . . . . . . 39
5 Climate Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2 Climate Change and Food Security: a Farm-Household Model 49

xi
xii Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2 A Farm Household Modelling . . . . . . . . . . . . . . . . . . . . . . . 52
3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1 The Production Function . . . . . . . . . . . . . . . . . . . . . . 54
3.2 The Shadow Wage . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 The Consumption Decisions . . . . . . . . . . . . . . . . . . . . 55
4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1 Labour, Income, and Farm Characteristics . . . . . . . . . . . . 59
4.2 Climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3 Household Characteristics . . . . . . . . . . . . . . . . . . . . . 63
4.4 Other characteristics . . . . . . . . . . . . . . . . . . . . . . . . 64
5 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4.1 Changes in Prices . . . . . . . . . . . . . . . . . . . . 70
5.4.2 Changes in Climate . . . . . . . . . . . . . . . . . . . 70
6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

II Climate Change in Developed Countries 79

3 Climate Change and Agricultural Yields: an European Case Study 83
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2 Modelling Framework and Estimation Procedure . . . . . . . . . . . . . 86
2.1 An Estimation by Maximum Likelihood . . . . . . . . . . . . . 86
2.2 The Production Function Assumption . . . . . . . . . . . . . . 87
3 Data Sets and Empirical Specifications . . . . . . . . . . . . . . . . . . 89
3.1 Production Data and Definition of the Agricultural Variables . 89
3.2 Environment Data and Spatial Location of the Farm Variables . 91
3.3 Weather Data and Definition of the Variables . . . . . . . . . . 92
4 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.1 Wheat Yields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 Corn Yields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4 Climate Change and Business Cycles 105

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.2 Final Goods Firms . . . . . . . . . . . . . . . . . . . . . . . . . 111
2.3 Intermediate Goods Firms . . . . . . . . . . . . . . . . . . . . . 112
2.3.1 Agricultural Production and Weather Variability . . . 112
 Contents xiii

2.3.2 Non-Agricultural Intermediate Production . . . . . . 116

2.4 Monetary Policy . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.5 Foreign Economy . . . . . . . . . . . . . . . . . . . . . . . . . . 118
2.6 Shocks, Aggregation and Equilibrium Conditions . . . . . . . . 119
3 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.1 Calibration and Prior Distributions . . . . . . . . . . . . . . . . 121
3.2 Posterior Distribution . . . . . . . . . . . . . . . . . . . . . . . 122
3.3 Do Weather Shocks Matter? . . . . . . . . . . . . . . . . . . . . 124
4 Weather Shocks as Drivers of Business Cycles . . . . . . . . . . . . . . 126
4.1 Propagation of a Weather Shock . . . . . . . . . . . . . . . . . 126
4.2 The Contributions of Weather Shocks on Aggregate Fluctuations 129
4.3 Historical Decomposition of Business Cycles . . . . . . . . . . 130
4.4 Climate Change Implications for Aggregate Fluctuations . . . . 132
4.4.1 Building Projections on Climate Shocks Variance . . 132
4.4.2 Measuring Climate Change Implications for Aggreg-
ate Fluctuations . . . . . . . . . . . . . . . . . . . . . 133
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

General Conclusion 137

A Ricardian Analysis: Data 141

1 Definitions of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
2 Weather Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
2.1 Observed Weather Data . . . . . . . . . . . . . . . . . . . . . . 143
2.2 Climate Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 146

B Agricultural Yields: Data 149

1 Agricultural Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
2 Weather Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . 151
2.3 Projected Weather Variables . . . . . . . . . . . . . . . . . . . . 151

C Climate Change and Business Cycles: Technical Appendix 157

1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
2 Model Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3 A Sketch of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.2 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.2.1 Final Firms . . . . . . . . . . . . . . . . . . . . . . . . 164
3.2.2 Agricultural Production and Weather Variability . . . 166
3.2.3 Non-Agricultural Intermediate Production . . . . . . 170
3.3 Monetary Policy . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.4 Foreign Economy . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.5 Stochastic Shock Processes . . . . . . . . . . . . . . . . . . . . 173
xiv Contents

3.6 Shocks, Aggregation and Equilibrium Conditions . . . . . . . . 174

4 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.1.1 The Weather Measure . . . . . . . . . . . . . . . . . . 178
4.1.2 The Climate Scenarios . . . . . . . . . . . . . . . . . 180
4.1.3 Macroeconomic Time Series Transformation . . . . . 181
4.2 Measurement Equations of the DSGE Model . . . . . . . . . . . 182
5 The Non-Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.1 Equations Summary . . . . . . . . . . . . . . . . . . . . . . . . 183
6 Business Cycle Facts About Climate Shocks . . . . . . . . . . . . . . . . 186
6.1 Modeling Framework . . . . . . . . . . . . . . . . . . . . . . . . 187
6.1.1 The Foreign Economy Block . . . . . . . . . . . . . . 189
6.1.2 The Domestic Climate Block . . . . . . . . . . . . . . 190
6.1.3 The Domestic Economy Block . . . . . . . . . . . . . 190
6.1.4 Estimation and Identification Issues . . . . . . . . . . 190
6.2 Macroeconomic Response to Weather Shocks . . . . . . . . . . 192

List of Figures 197

List of Tables 201

Bibliography 203
Résumé en français

La fin de l’année 2015 a été marquée par la vingt-et-unième Conférences des Parties de
la Convention cadre des Nations unies sur les changements climatiques. Cette confé-
rence a réuni les représentants de 195 nations afin de prendre des mesures pour lutter
contre le changement climatique. Ce réchauffement, à en croire l’accumulation des
preuves scientifiques, serait principalement dû à une augmentation de la concentration
en gaz à effet de serre, résultant notamment des activités humaines. Bien que des effets
positifs consécutifs à ce réchauffement pourraient être ressentis dans certaines parties
du monde, un grand nombre d’études indique qu’ils seraient compensés à l’échelle glo-
bale par les effets négatifs ressentis dans les autres régions, faisant de facto peser une
grave menace sur l’avenir de notre planète. Les attentes de la Conférence des Nations
Unies sur les changements climatiques étaient par conséquent élevées. Les chefs de
file mondiaux, y ont exprimé leur volonté de la nécessité d’agir pour éviter les effets
les plus catastrophiques du changement climatique. Les signataires du traité rédigé
lors de cette conférence ont convenu qu’il était nécessaire de limiter le réchauffement
climatique en dessous d’un seuil d’une augmentation de 2 degrés Celsius par rapport à
l’aire pré-industrielle, avec une volonté de mener des efforts encore plus poussés pour
limiter cette augmentation en-dessous de 1, 5 degrés Celsius. L’accord, ratifié par 142
pays, est entré en vigueur moins d’un an plus tard, en novembre 2016.

De manière implicite, les signataires de l’accord de Paris ont reconnu qu’en-deçà d’une
augmentation de la température globale de 2 degrés Celsius par rapport aux niveaux
pré-industriels, l’humanité pourrait s’adapter aux nouvelles conditions climatiques.
Toutefois, les moyens mis en œuvre afin de parvenir à respecter l’accord climatique
ainsi que les coûts sous-jacents sont accompagnés d’une forte incertitude. En fait,
les coûts financiers du changement climatique représentent un problème économique
ayant donné lieu à de nombreuses études scientifiques. Non seulement la littérature
s’attache à estimer les coûts liés à la mise en place de politiques d’atténuation, mais
elle tente également d’évaluer les coûts de l’inaction (OECD, 2015).

xv
xvi Résumé en français

Une manière populaire d’évaluer ces coûts a été introduite par Nordhaus (1994). Il
s’agit d’une approche globale qui consiste à représenter le monde à l’aide d’équations
mathématiques. L’idée est d’introduire une fonction de dommage climatique venant
perturber l’économie. Cette fonction de dommage lie l’augmentation de la température
au PIB, la température étant elle-même fonction du niveau de concentration en gaz à
effet de serre. Les modèles mathématiques alors construits peuvent ensuite être utili-
sés pour évaluer les bénéfices de la mise en place de certaines actions visant à limiter
l’émission des gaz à effet de serre (voir, p. ex., Hope et al., 1993 ; Tol, 2005). Toutefois,
il n’existe pas de réponse tranchée quant à la meilleure manière de s’attaquer à la di-
minution des émissions de gaz à effet de serre. Certains auteurs recommandent d’agir
rapidement (p. ex., Stern, 2007) tandis que d’autres suggèrent une approche plus pro-
gressive (p. ex., Nordhaus, 2007) impliquant moins de contrôles dans le présent et en
faveur d’un report de l’action dans le futur.

Cette approche globale pour mesurer les impacts du changement climatique est com-
plétée dans la littérature par une multitude d’approches partielles, qui se concentrent
sur un secteur économique particulier. Parmi ces secteurs économiques, l’agriculture
est un champ d’application privilégié, du fait de sa forte dépendance au conditions cli-
matiques. De manière étonnante, l’agriculture n’est pas directement mentionnée dans
les accords de Paris de 2015. Les impacts potentiels du changement climatique sur
ce secteur, malgré la faible importance relative de l’agriculture dans le PIB mondial,
posent pourtant de nombreux défis et menaces sur l’avenir de notre planète (OECD,
2015), particulièrement à l’égard de la sécurité alimentaire.

Un large pan de la littérature est ainsi dédié à l’étude des relations entre l’agriculture et
le climat, à commencer par de nombreuses études simulant la croissance des céréales
à l’aide de modèles mathématiques (voir, p. ex., Ritchie and Otter, 1985 ; Jones et al.,
1986 ; Brisson et al., 1998). En associant ces modèles à d’autres visant à simuler des
conditions climatiques, il est possible d’estimer les effets potentiels du changement
climatique sur la sécurité alimentaire (voir, p. ex., Rosenzweig and Parry, 1994 ; Jones
and Thornton, 2003 ; Parry et al., 2004). D’autres études se concentrent sur les consé-
quences de la variation du climat sur la valeur des terres (Mendelsohn et al., 1994 ;
Schlenker et al., 2005) ou encore sur les profits agricoles (Deschênes and Greenstone,
2007), afin d’évaluer les conséquences éventuelles du changement climatique, au ni-
veau mondial ou régional. Les études conduites à l’échelle mondiale indiquent que les
pays développés économiquement ne seront pas impactés de la même manière que
 Résumé en français xvii

les pays en développement, plus vulnérables aux aléas climatiques (Rosenzweig and
Parry, 1994 ; Fischer et al., 2005).

Cette thèse représente une tentative de contribuer au débat scientifique à l’égard du

changement climatique et de ses effets sur l’agriculture. Elle s’appuie sur des méthodes
empiriques et théorique au travers de quatre études, couvrant différentes échelles géo-
graphiques, tantôt concernant les pays en développement, tantôt concernant des pays
développés.

Le reste de ce résumé propose de revenir sur les concepts clés de la thèse. La section 1
décrit brièvement ce qu’est le changement climatique et fournit des détails relatifs
aux différents scénarios climatiques utilisés par les chercheurs afin d’estimer les effets
potentiels du changement climatique sur la planète. La section 2 présente de manière
plus détaillée les efforts fournis dans la littérature dans l’examen des effets du climat
sur l’agriculture, et propose un bref aperçu des méthodes principales retenues par les
chercheurs à cet effet. Enfin, la section 3 décrit la structure de la thèse, organisée en
quatre chapitres.

Le climat change
Les preuves scientifiques sont très nombreuses : le climat global se réchauffe. Le Groupe
d’experts intergouvernemental sur l’évolution du climat (GIEC), un organe internatio-
nal des Nations Unies créé en 1988 afin d’évaluer sans parti pris la question du chan-
gement climatique, et regroupant des milliers de scientifiques définit le changement
climatique comme : « variation de l’état du climat, qu’on peut déceler (par exemple au
moyen de tests statistiques) par des modifications de la moyenne et/ou de la variabilité de
ses propriétés et qui persiste pendant une longue période, généralement pendant des décen-
nies ou plus. » (Edenhofer et al., 2014). Les résultats du dernier rapport de 2014 du GIEC
prédisent une augmentation globale de la température moyenne de surface, en partie
due à l’augmentation de la concentration anthropique (c’est-à-dire causée par les acti-
vités humaines) des gaz à effet de serre. Plus cette concentration est élevée, plus sera
celle de la température moyenne de surface. Depuis la période pré-industrielle, c’est-à-
dire, depuis environ la moitié du XVIIIe siècle, la concentration en gaz à effet de serre
tels le dioxyde de carbone (CO2 ), le méthane (CH4 ) ou l’ozone (O3 ) a augmenté de ma-
nière significative. En effet, la concentration en dioxyde de carbone dans l’atmosphère
a augmenté de 43% depuis l’ère pré-industrielle pour atteindre 339.5 parties par mil-
lion (ppm) en 2016. Dans le même temps, la concentration en méthane s’est intensifiée
xviii Résumé en français

de 154% pour atteindre 1834 ppm en 2016, et celle de l’ozone troposphérique s’est ac-
crue de 42% pour atteindre 337 ppm en 2016 (Blasing, 2009). Au cours des prochaines
années, la concentration en gaz à effet de serre devrait continuer de grimper. Il existe
cependant une forte incertitude concernant à la fois la valeur de cet accroissement d’ici
la fin du XXIe siècle ainsi que le sentier conduisant à cette valeur. Ces concentrations
dépendront, entre autres, de la croissance démographique, du développement social et
du progrès technique. De nombreuses hypothèses doivent donc être émises, tout par-
ticulièrement à propos de l’activité humaine, afin d’estimer les niveaux potentiels de
concentration en gaz à effet de serre dans un proche avenir. La projection de variables
climatiques en dépend. Afin de fournir une base commune pour les chercheurs, il est
courant de prendre appui sur des scénarios climatiques bien définis, développés par la
communauté scientifique. Cela permet alors d’analyser le changement climatique et
ses impacts.

Le GIEC, dans son dernier rapport (Edenhofer et al., 2014), a adopté quatre trajectoires
différentes des émissions de gaz à effet de serre, appelées scénarios RCP (pour “Re-
presentative Concentration Pathways”), représentant différentes alternatives possible
d’émissions en fonction des scénarios auparavant utilisés dans la littérature. Les noms
de ces quatre scénarios reflètent la valeur du forçage radiatif à l’horizon auquel ils
sont simulés, c’est-à-dire 2100 : les scénarios RCP 2.6, RCP 4.5, RCP 6.0 et RCP 8.5 sont
ainsi caractérisés par un forçage radiatif en 2100 de 2, 6W /m2 , 4, 5W /m2 , 6, 0W /m2 ,
et 8, 5W /m2 respectivement. Le forçage radiatif d’un gaz correspond à la différence
entre le rayonnement solaire entrant et le rayonnement infrarouge sortant, et est in-
fluencé par la concentration de ce gaz. Plus la concentration est élevée, plus la balance
des rayonnements entrants et sortants est élevée, ce qui entraîne une augmentation
des températures de surface. Par conséquent, la température de surface globale sur
Terre devrait être la plus basse pour le scénario RCP 2.6 et la plus élevée pour le RCP
8.5. Le sentier permettant d’atteindre les niveaux de forçage radiatif diffère parmi les
scénarios, comme le montre le graphique de gauche de la fig. 3. Dans le premier scé-
nario, le RCP 2.6, c’est-à-dire le moins pessimiste en termes de concentration de gaz à
effet de serre, un pic de forçage radiatif est atteint vers 2030 et diminue lentement par
la suite. Les deux scénarios suivants, le RCP 4.5 et le RCP 6.0 sont caractérisés par des
niveaux de rayonnements plus élevés, avec une stabilisation respective sans dépasse-
ment à 4, 5W /m2 et 6, 0W /m2 . Le dernier scénario, le RCP 8.5, est plus pessimiste
et reflète des émissions de gaz à effet de serre qui augmentent continuellement, en-
traînant une valeur de forçage radiatif de 8, 5W /m2 d’ici 2100. De plus amples détails
peuvent être trouvés dans Van Vuuren et al. (2011).
 Résumé en français xix

Forçage radiatif (w/m2 ) Température moyenne (o C)

4

7 3

2
5
1

3 0

-1
2000 2025 2050 2075 2100 1900 1950 2000 2050 2100

Historique RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Notes : Chaque courbe représente la tendance du forçage radiatif (gauche) et la variation correspondante du changement de
température moyenne globale par rapport à 1986-2005 (droite) pour l’un des quatre scénarios RCP. Le graphique de gauche est
une reproduction de la Figure 10 de Van Vuuren et al. (2011) ; celui de droite de la Figure 12.1 de Collins et al. (2013).

Figure 1 : Tendances du forçage radiatif et changement global de la température

moyenne

Les tendances du forçage radiatif de chaque scénario peuvent être utilisées dans des
modèles climatiques, afin de simuler un climat potentiel jusqu’à 2100, à différentes
échelles spatio-temporelles. La variation moyenne de la température moyenne globale
pour les quatre scénarios est représentée sur le graphique de droite de la fig. 3. D’ici
la fin du XXIe siècle, comparativement aux niveaux de 1986–2005, selon les résultats
relayés par le GIEC (Collins et al., 2013), la variation de la température moyenne tablant
sur le scénario 2.6 serait vraisemblablement comprise entre 0, 3o C et 1, 7o C. Sous les
scénarios RCP 4.5 et 6.0, la variation serait plus élevée, avec des valeurs allant de 1, 1o C
à 2, 6o C et de 1, 4o C à 3, 1o C, respectivement. Dans le pire des cas, sous le scénario
RCP 8.5, il est probable que l’augmentation de la température moyenne globale soit
comprise entre 2, 6o C et 4, 8o C.

Ces valeurs de changements sont des moyennes à l’échelle mondiale. En réalité, une
forte hétérogénéité dans les projections climatiques s’observe, avec un changement
prévu sur les terres plus élevé que celui des océans. En outre, les changements sur les
terres ne devraient pas être uniformes ; certaines régions devraient en effet connaître
une augmentation de la température moyenne tandis que d’autres seraient soumises
à des climats plus froids.

Une modification des statistiques climatiques a un impact direct sur ses réalisations.
Une distinction entre ces deux notions doit être faite, comme l’ont souligné Schlenker
et al. (2006). La principale différence entre les réalisations du climat (“weather”) et le
climat lui-même (“climate”) est l’échelle temporelle. Les réalisations du climat corres-
pondent aux conditions météorologiques à un moment distinct, alors que le climat se
xx Résumé en français

réfère à une moyenne des statistiques climatiques sur une longue période de temps.
Par conséquent, si nous considérons les conditions météorologiques comme la réali-
sation de multiples variables aléatoires, la réalisation du climat peut être considérée
comme un tirage à court terme de ces variables. Le climat peut quant à lui être consi-
déré comme la moyenne de ces tirages sur le long terme. Une illustration simplifiée
peut aider à mieux appréhender cette distinction, et ainsi donner une meilleure idée
des effets sous-jacents d’une modification des statistiques du climat sur ses réalisa-
tions. Par commodité, nous pouvons supposer que la température de surface est le
résultat d’un tirage aléatoire selon une distribution gaussienne, comme représentée
par la fig. 4. Aussi, comme montré par la fig. 4(a), une augmentation de la moyenne
de la température conduirait à une augmentation de la probabilité d’occurrence de
valeurs chaudes ou extrêmement chaudes, accompagnée d’une diminution de la pro-
babilité d’occurrence de valeurs froides ou très froides. Cependant, le GIEC, dans son
dernier rapport, indique que les modifications opérées sur les systèmes climatiques
comprennent également un changement dans la variabilité du climat. Reprenons alors
notre exemple, en considérant à présent un changement uniquement dans la variance
des températures, en conservant la moyenne à son niveau initial. La fig. 4(b) illustre
ce cas. Si la variance croît, la distribution devient plus plate, impliquant de facto une
augmentation de la probabilité d’occurrence des valeurs à la fois froides et chaudes.
Une combinaison entre l’augmentation de la moyenne et de la variance, comme illus-
tré par la fig. 4(c), se solde par une augmentation de l’occurrence de valeurs chaudes
et extrêmement chaudes.

Cette illustration est une simplification de ce qui est en réalité attendu avec le chan-
gement climatique. De nombreux effets tels l’hétérogénéité spatiale et la saisonnalité
devraient être pris en compte. De plus, la loi de probabilité régissant les températures
n’est sûrement pas gaussienne. L’exemple simplifié permet toutefois de comprendre la
base des mécanismes liés au changement climatique. Il devient plus facile d’imaginer
comment le nombre et la gravité des événements extrêmes tels que les tornades, les
fortes pluies ou les sécheresses devraient augmenter d’ici la fin du siècle.

Cependant, beaucoup d’incertitude subsiste quant à l’ampleur de ces changements et

de leurs effets sur notre société, notamment sur l’économie. En particulier, sur les
systèmes agricoles qui sont au cœur des défis du changement climatique, en raison
leur forte dépendance aux conditions météorologiques.
 Résumé en français xxi

Moins de
temps Plus de
froid temps
Moins de Plus de
chaud
temps froid temps chaud
extrême extrême

(a) Augmentation de la moyenne

Plus de Plus de
Plus de temps temps
Plus de
temps frois froid chaud
temps chaud
extrême extrême

(b) Augmentation de la variance

Temps Plus de
froid temps
presque Plus de
chaud
Temps froid inchangé temps chaud
extreme extrême
presque inchangé

(c) Augmentation de la moyenne et de la variance

Notes : les lignes solides et pointillées représentent la distribution des températures avant et après une modification des moments
d’ordre 1 et 2, respectivement (sous hypothèse de normalité). Cette figure est une reproduction de la Figure 1.8 issue de Cubasch
et al. (2013).

Figure 2 : Effets du changement climatique sur la distribution des températures

Modélisation des conséquences du changement cli-

matique sur l’agriculture
Les systèmes agricoles dépendent étroitement des variables climatiques telles que la
température et les précipitations. Certaines régions ont des conditions naturelles plus
adaptées aux activités agricoles que d’autres. Par exemple, les zones tempérées sont
plus propices à la culture de céréales que les zones tropicales, les températures et les
niveaux de précipitations observés dans ces premières étant plus propices aux besoins
xxii Résumé en français

de certaines céréales telles que le blé ou le maïs. Le climat peut en fait être consi-
déré comme un intrant direct dans la fonction de production. Toutefois, contraire-
ment à d’autres intrants tels que la main-d’œuvre, les engrais, les systèmes d’irrigation
plus ou moins sophistiqués, les machines ou les méthodes agricoles, la météo ne peut
être contrôlée par les agriculteurs. Par conséquent, la production agricole est vulné-
rable aux aléas météorologiques. Cette dépendance des systèmes agricoles à la météo
pose des défis importants dans le contexte du changement climatique. Comme men-
tionné précédemment, le changement climatique devrait entraîner une modification
des conditions météorologiques, accompagnée d’une augmentation des événements
extrêmes pouvant entraîner des pertes importantes de production. Le changement
climatique est très susceptibles d’affecter la sécurité alimentaire au niveau mondial
comme à l’échelle des pays, via un accès réduit à la nourriture. Cela soulève des in-
quiétudes concernant l’un des nombreux défis auxquels le secteur agricole doit faire
face, à savoir sa capacité à produire suffisamment de nourriture pour couvrir les be-
soins de la population. En outre, le nombre de personnes vivant sur la planète devrait
continuer à croître au cours des prochaines années, rajoutant davantage de difficulté
à ce défi. En effet, la demande mondiale de récolte de 2005 à 2050 devrait augmenter
de 100 à 110 % (Tilman et al., 2011). Par conséquent, il apparaît primordial d’étudier la
relation entre l’agriculture et le climat, afin de pouvoir s’adapter aux nouvelles condi-
tions météorologiques que nous réserve l’avenir.

Pour pouvoir relever ces défis, une grande partie de la littérature se concentre directe-
ment sur les fonctions de production agricole ou, en utilisant un cadre appelé « ricar-
dien », examine les profits agricoles réalisés. Une autre partie de la littérature s’inté-
resse à la manière avec laquelle la réponse de l’agriculture aux variations climatiques
affecte le reste de l’économie.

Les fonctions de production agricole

Le nombre de contributions scientifiques documentant la relation entre le climat et la

fonction de production agricole a considérablement augmenté au cours des dernières
décennies. Deux méthodes différentes se distinguent dans cette partie de la littérature.
La première simule les rendements des cultures en utilisant des modèles agronomiques,
tandis que la seconde repose sur des modèles statistiques pour examiner les liens ob-
servés entre les rendements et le climat.
 Résumé en français xxiii

Les modèles de simulation de cultures sont utilisés pour modéliser la croissance des
cultures. Certains sont conçus pour une culture spécifique, comme CERES-Maize pour
le maïs (Jones et al., 1986) ou CERES-Wheat pour le blé (Ritchie and Otter, 1985).
D’autres peuvent être adaptables à un large éventail de cultures, telles que les mo-
dèles STICS qui peuvent modéliser le blé, le maïs, le soja, le sorgho et bien d’autres
cultures de croissance (Brisson et al., 1998). Dans les deux cas, les phases de développe-
ment des cultures sont modélisées par des équations mathématiques dans un premier
temps. Ces équations tiennent compte de nombreux facteurs telles que les conditions
du sol, le climat et les méthodes de gestion (Mearns et al., 1997). Une fois que le modèle
est établi, il doit être calibré. À cette fin, les cultures sont cultivées dans des champs
ou des terrains expérimentaux, sous différentes conditions environnementales, y com-
pris météorologiques. Des niveaux différents de concentration en dioxyde de carbone
peuvent également être testés. Comme mentionné précédemment, la concentration de
ces gaz à effet de serre dans l’atmosphère affecte le climat par le forçage radiatif, mais
elle joue également un rôle dans le processus physiologique de la photosynthèse et
de la transpiration (Field et al., 1995). Par conséquent, être en mesure de capturer les
effets fertilisants du dioxyde de carbone représente un atout considérable des modèles
de simulation des cultures. En effet, lorsque les modèles de simulations de cultures sont
ensuite utilisés, une fois calibrés, pour réaliser du contre-factuel, il est possible de leur
soumettre de nouvelles valeurs de concentration en gaz à effet de serre reflétant un
scénario climatique possible. Il est alors possible d’observer la réponse potentielle des
cultures selon différents scénarios climatiques. Les modèles de simulation des cultures
deviennent alors un outil puissant pour étudier les avantages potentiels et/ou les pré-
judices provoqués par le changement climatique.

La réponse de la production agricole au changement climatique a été largement étudiée

dans la littérature à l’échelle de régions plus ou moins vastes (voir, p. ex., Aggarwal and
Mall, 2002 pour les rendements du riz en Inde ; Xiong et al., 2009 pour la production
de maïs, de blé et de riz en Chine ; Jones and Thornton, 2003 pour les rendements de
maïs en Afrique de l’Ouest) ou à l’échelle mondiale (voir, p. ex., ; Lobell and Field, 2007
pour le blé, le riz, le maïs et le soja ; Rosenzweig et al., 2014 pour le blé, le riz et le
soja). Récemment, Bassu et al. (2014) ont montré que l’utilisation d’une combinaison
de modèles plutôt qu’un seul permet de simuler les rendements agricoles avec une
précision accrue ; ils ont également montré qu’une augmentation de la température
diminue les rendements du maïs.

Cependant, les modèles de simulation des cultures présentent certains inconvénients.

xxiv Résumé en français

Tout d’abord, ils sont calibrés dans des zones géographiques spécifiques et peuvent
donc représenter avec précision les différentes étapes de croissance de cette région
particulière. Ils peuvent dans le même temps ne pas être représentatifs de toutes les
régions du monde. Une autre lacune majeure de ces modèles est qu’ils ne tiennent pas
compte de la possibilité pour les agriculteurs de s’adapter à une nouvelle condition
climatique. En effet, ces modèles, même s’ils sont adaptables à de multiples cultures,
sont calibrés pour un type donné de culture, et ne modélisent donc pas la possibilité
pour l’agriculteur d’en changer si les conditions climatiques ne sont plus adaptées. Par
ailleurs, toutes les différences de résultats sont supposées être attribuables uniquement
à des changements dans les variables affectant la croissance des cultures, telles que la
température ou les précipitations. Cela conduit probablement les modèles de simula-
tion à surestimer les effets du changement climatique (Adams et al., 1990 ; Parry et al.,
2004). En outre, les conditions économiques telles que la variabilité des prix ne sont
pas prises en compte par ces modèles, de sorte que les agriculteurs sont considérés
comme myopes face à leur environnement économique et de facto incapables de se
tourner vers une activité offrant de meilleures opportunités.

La question de la représentativité spécifique à la localisation peut être en partie traitée

par des modèles statistiques. Dans ces cadres d’analyse, les variations des rendements
des cultures sont modélisées en fonction des variations d’autres variables telles que
les intrants utilisés, les conditions économiques auxquelles sont confrontés les agri-
culteurs, la qualité du sol, la météo ou les conditions climatiques. La méthode repose
sur des observations historiques pour estimer la forme fonctionnelle reliant la variable
d’intérêt à son prédicteur. Il est donc possible d’examiner les déterminants des rende-
ments des cultures à des échelles spatiales plus importantes que celles utilisées dans les
analyses de simulation de cultures, en s’appuyant sur les données observées à plusieurs
endroits. Une fois que le modèle statistique est estimé, il est possible de lui soumettre
de nouvelles conditions météorologiques et de prédire quelle sera la réponse de la va-
riable d’intérêt. La valeur prédite peut de facto être comparée à la réalisation historique,
et le changement peut alors être attribué à la variation des conditions climatiques.

Dans les modèles statistiques, la météo est considérée comme un apport spécial dans
la production agricole puisqu’elle ne peut être contrôlée dans des configurations na-
turelles et est donc considérée comme une variable exogène. Ainsi, les conditions cli-
matiques agissent comme une ”expérience naturelle” (Angrist and Krueger, 2001, via
Auffhammer et al., 2013). Le chercheur ne peut observer que l’issue de l’expérience
sans pouvoir contrôler manuellement la quantité de l’intrant naturel. Cependant, les
 Résumé en français xxv

effets causaux de la météo sur les rendements des cultures peuvent être examinés et
ont déjà conduit à de nombreuses études tentant de les quantifier. Il n’y a toutefois
aucun consensus quant à l’ampleur des effets du changement climatique sur les ren-
dements des cultures. Lobell et al. (2011) ont étudié la relation historique mondiale
entre la température et quatre cultures : le maïs, le riz, le blé et le soja, de 1980 à 2008.
Leurs résultats suggèrent un impact négatif de la température sur les rendements du
maïs et du blé, réduisant les rendements de 3, 8% et de 5, 5%, respectivement, avec
la présence d’hétérogénéité régionale. Leurs résultats concernant les deux autres ré-
coltes sont moins concluants, avec des baisses globales de −0, 1% pour le riz et −1, 7%
pour le soja, avec des gains observés dans certaines régions annulant les pertes su-
bies dans d’autres régions. You et al. (2009) ont trouvé des résultats similaires pour
les rendements du blé en Chine, entre 1979 et 2000. Ils ont estimé une réduction des
rendements du blé de 3% à 10% en raison d’une augmentation de la température de
la saison de croissance moyenne d’un degré. Lobell and Asner (2003) ont estimé l’im-
pact de l’augmentation des températures sur les rendements agricoles dans les comtés
aux États-Unis, en utilisant des données de 1982 à 1998. Selon les résultats de l’article,
chaque degré supplémentaire de la température moyenne entraîne une diminution de
17% des rendements du maïs et du soja. Dans la littérature, les simulations de change-
ment de climat sont plus profondément évaluées en utilisant des scénarios climatiques.
Par exemple, Schlenker and Roberts (2009) ont considéré différents scénarios pour les
États-Unis et ont conclu que l’augmentation des températures d’ici la moitié du XXIe
siècle entraînerait une diminution substantielle des rendements du maïs, du coton et
du soja par rapport aux rendements observés entre 1950 et 2005. Plus précisément,
les pertes varieraient de −30% à −46% dans un scénario optimiste, et seraient encore
plus désastreuses dans le pire des cas, allant de −63% à −82%. Schlenker and Lobell
(2010) ont utilisé des données historiques en Afrique subsaharienne pour relier les
rendements des cultures à la variation de la température, puis ont utilisé le modèle
estimé pour évaluer les effets du changement climatique sur les rendements de maïs,
de sorgho, de mil, d’arachide et de manioc. Pour l’ensemble des cultures, leurs résul-
tats prévoient des rendements en déclin, allant de −8% à −22% d’ici la moitié du XXIe
siècle, comparativement aux rendements observés entre 1961 et 2000.

Toutes ces études considèrent les effets des covariables sur la moyenne de la variable
de réponse. Certaines études (p. ex., Chen et al., 2004 ; Cabas et al., 2010) suggèrent
d’utiliser la méthode de production de Just et Pope (Just and Pope, 1978), permettant
de facto de caractériser les effets des variations climatiques à la fois sur les rendements
des cultures mais également sur leur variabilité (McCarl et al., 2008).
xxvi Résumé en français

L’avantage d’utiliser des modèles statistiques plutôt que des modèles de simulation
de culture est que les premiers indiquent clairement les incertitudes du modèle en
donnant des indicateurs statistiques concernant la qualité de l’estimation, ce qui n’est
pas la norme avec les modèles de simulation de cultures (Lobell and Burke, 2010). Ce-
pendant, les modèles statistiques présentent certaines réserves. Même s’il est fréquent
d’utiliser plusieurs variables météorologiques pour estimer les variations des rende-
ments des cultures, cette pratique pourrait causer des problèmes de multicolinearité
(Sheehy et al., 2006 ; Lobell and Ortiz-Monasterio, 2007), surtout lorsque les variables
météorologiques sont désagrégées pour refléter les effets saisonniers. Une autre lacune
des modèles statistiques est ce que Mendelsohn et al. (1994) appellent le « scénario des
agriculteurs naïfs » (“dumb farmer scenario”). Dans ces modèles, l’accent est mis sur un
seul type de culture, tout comme dans les modèles de simulation des cultures. Aussi,
il est implicitement supposé que les agriculteurs ne peuvent pas s’adapter à un en-
vironnement variable offrant de nouvelles conditions climatiques. D’où l’expression
employée par les auteurs. Par conséquent, l’analyse statistique des rendements des
cultures pourrait ne pas être un bon outil pour prédire les variations à long terme.

L’approche ricardienne

Les modèles de simulation des cultures et leur homologue statistique ne tiennent pas
compte de la capacité des agriculteurs à s’adapter à un nouvel environnement. Une
manière de contourner cet écueil est fournie par Mendelsohn et al. (1994), qui sug-
gèrent de regarder la valeur de la terre plutôt que les rendements. Dans leur travail
pionnier, ils présentent une analyse intitulée « analyse ricardienne » du nom du cé-
lèbre économiste David Ricardo.

La méthodologie ricardienne est un modèle hédonique de tarification des terres agri-

coles qui met l’accent sur la valeur foncière. L’idée fondamentale est que le climat à
long terme devrait être capitalisé en valeur foncière. Dans un marché concurrentiel, on
suppose que le prix des terres agricoles reflète la valeur actualisée de tous les bénéfices
futurs prévus qui en découlent. L’approche ricardienne mesure l’impact des variables
climatiques sur la productivité de la terre ou la valeur des terres agricoles en exploi-
tant les différences transversales dans l’utilisation des terres et les conditions météo-
rologiques. Il est à noter que les variables climatiques (moyennes sur du long terme)
sont utilisées dans l’analyse ricardienne plutôt que dans les variables météorologiques
(réalisations du climat, à court terme). En effet, un choc météorologique défavorable
comme une tempête ou une sécheresse peut entraîner des pertes sur les rendements
 Résumé en français xxvii

des cultures, mais peut avoir des effets ambigus sur les bénéfices : d’une part, la pro-
duction peut être réduite et, par conséquent, diminuer les bénéfices ; d’autre part, en
présence d’une inélasticité de la demande de produits agricoles, le marché peut s’ajus-
ter en augmentant les prix, augmentant de facto les profits agricoles. Mais à long terme,
les bénéfices devraient être réduits par un changement climatique adverse (Schlenker
et al., 2006). En outre, à court terme, les agriculteurs peuvent être vulnérables aux
chocs météorologiques, mais à long terme, ils peuvent adopter de nouvelles stratégies
agricoles et ajuster leurs décisions concernant les niveaux d’intrants en réponse au
climat auquel ils sont confronté. Par conséquent, l’utilisation de variables climatiques
plutôt que des variables météorologiques est plus adaptée au cadre ricardien.

Au cours des deux dernières décennies, le cadre ricardien original a été largement
appliqué dans de nombreux pays à travers le monde. La plupart des recherches an-
térieures utilisent des données transversales pour estimer les valeurs de la ferme sur
les variables météorologiques. Des examens approfondis des applications pour les ré-
gions d’Afrique, d’Asie, d’Amérique du Sud et d’Amérique peuvent être trouvés dans
le travail de Mendelsohn and Dinar (2009).

Dans l’article pionnier de Mendelsohn et al. (1994), l’accent a été mis sur les agricul-
teurs américains, et les résultats mettent en exergue l’hétérogénéité régionale des ef-
fets du climat sur les valeurs foncières. Les scénarios climatiques testés produisent des
résultats mitigés, soulignant la présence de gagnants et de perdants. Les pertes subies
par les perdants devraient toutefois être relativement inférieures à celles projetées en
utilisant une analyse basée sur les rendements des cultures. Des résultats mitigés sont
également observés en Afrique. Par exemple, Kurukulasuriya and Mendelsohn (2008)
ont montré, en utilisant des données au niveau de l’exploitation provenant de 11 pays
africains recueillis entre 2003 et 2004, que les effets du changement climatique sur les
revenus nets diffèrent entre les exploitations irriguées et les exploitations pluviales.
Les revenus nets de ces premières devraient augmenter jusqu’à 51% dans le meilleur
des cas, alors que les revenus nets de ces dernières devraient chuter de −43% dans le
cas d’un scénario climatique chaud et sec. Des résultats similaires entre les exploita-
tions irriguées et non irriguées ont été trouvés en Chine (Wang et al., 2009). En Europe,
Van Passel et al. (2016) ont estimé les effets du changement climatique à l’aide d’un
grand échantillon d’exploitations agricoles en Europe occidentale. Bien que l’étude
souligne différents effets régionaux, elle montre également que les exploitations agri-
coles européennes sont plus sensibles au réchauffement projeté que les exploitations
américaines.
xxviii Résumé en français

Un certain nombre de critiques à l’égard de l’approche ricardienne a été émis. Une fai-
blesse majeure réside dans le fait que les compétences non observables des agriculteurs
ne sont pas incorporées dans l’analyse. Cette lacune a conduit certains chercheurs à in-
troduire des données temporelles pour tenir compte du problème des variables omises,
en ajoutant des variables indicatrices des régions dans le modèle (voir, p. ex., Schlen-
ker et al., 2006 ; Deschênes and Greenstone, 2007 ; Kim et al., 2009 ; Cabas et al., 2010).
Contrairement aux modèles de simulation des cultures, le cadre ricardien ne tient pas
compte des effets fertilisants du CO2 , ce qui donne des résultats biaisés. Une autre fai-
blesse du modèle ricardien réside dans le fait qu’il n’intègre pas les effets sur les prix.
Dans l’approche ricardienne, la variation des valeurs foncières sur les zones clima-
tiques est due à des changements dans les variables climatiques. Une forte hypothèse
est que les prix des intrants et des produits restent inchangés. Une attention particu-
lière devrait être accordée sur ce point, car les prix des cultures tendent à être plus
volatiles de nos jours (Cline, 1996 ; Schlenker et al., 2005).

Enfin, le modèle suppose implicitement que les agriculteurs ne sont pas confrontés à
des coûts d’ajustement. Par conséquent, les résultats fournis par les modèles ricardiens
devraient être considérés comme une estimation de la limite inférieure des coûts du
changement climatique (Quiggin and Horowitz, 1999).

Les impacts économiques du changement climatique

Au lieu de se concentrer uniquement sur la fonction de production agricole, un nombre

important d’études évalue plutôt les impacts du changement climatique sur l’écono-
mie. Certaines tentatives sont faites à l’échelle mondiale ou dans des zones géogra-
phiques plus spécifiques.

Les études portant sur l’économie mondiale mettent en évidence l’existence de ga-
gnants et de perdants sous de nouvelles conditions reflétant le changement climatique.
Une méthodologie largement utilisée consiste à prédire la réponse des rendements
des cultures sous des scénarios climatiques projetés. Ces scénarios sont communé-
ment donnés par un modèle de circulation générale (GCM), c’est-à-dire un modèle
climatique. Ensuite, les rendements agricoles prévus sont incorporés dans un modèle
d’équilibre général qui évalue la production agricole ainsi que les prix des cultures. Il
est également possible de s’intéresser à des questions de sécurité alimentaire en esti-
mant le nombre de personnes exposées à la famine en fonction de la taille de la popu-
lation, de la production agricole et des prix agricoles. Rosenzweig and Parry (1994) ont
 Résumé en français xxix

fourni une contribution célèbre à cette littérature. Ils ont identifié une distinction entre
les pays en développement et les pays développés, les premiers étant plus vulnérables
au changement climatique que les derniers. Cette distinction est un résultat commun
dans la littérature (voir p. ex., Parry et al., 2004 ; Fischer et al., 2005), en raison de la
prédominance de l’agriculture dans les pays en développement (Tubiello and Fischer,
2007). Une autre explication à ce résultat est que les pays en développement sont pla-
cés de manière disproportionnée à des latitudes faibles ou proches des tropiques, où
les conditions climatiques ne sont pas optimales pour les activités agricoles, en raison
de la grande variabilité météorologique.

Une autre façon d’examiner les effets globaux du changement climatique sur l’écono-
mie est d’utiliser une autre approche d’équilibre général appelée « modèles d’évalua-
tion intégrée » (IAM, pour “Integrated Assessment Models”). Les IAM sont des modèles
mathématiques qui combinent dans un même cadre les connaissances sur le climat,
l’économie, la démographie et les décisions politiques qui influent sur les émissions
de gaz à effet de serre (voir p. ex., Nordhaus, 1991 ; Nordhaus, 1994 ; Nordhaus and
Yang, 1996 ; Tol, 2002). Ces modèles évaluent le coût social du carbone en évaluant
les changements dans le bien-être en raison des nouvelles conditions climatiques. Le
mécanisme repose sur l’existence d’une fonction de dommage qui relie la température
au PIB, de sorte qu’une augmentation de la température pourrait nuire à la produc-
tion mondiale. Les IAM peuvent être utilisés comme outil d’analyse des politiques,
en soumettant différents scénarios climatiques au modèle et en modélisant différentes
politiques basées sur le coût social du carbone, c’est-à-dire l’estimation monétaire des
dommages causés de manière directe ou non par l’émission d’une tonne de CO2 .

Les IAM, comme tout autre outil, présentent des lacunes. L’une d’elles concerne la
fonction de dommages. Comme indiqué par Weitzman (2010), la fonction de dommage
est « un lien notoirement faible dans l’économie du changement climatique1 », car des
hypothèses doivent être émises concernant sa forme fonctionnelle, et parce que ces
hypothèses peuvent modifier considérablement les conclusions apportées par les IAM.
Pindyck (2013) a également noté que les IAM ne tiennent pas compte de l’occurrence
d’événements météorologiques catastrophiques, qui peuvent avoir des répercussions
économiques très importants.

Beaucoup d’études se concentrent sur des zones géographiques plus spécifiques. La

distinction entre économie en développement et économie développée est de nouveau
1
La citation dans le papier est la suivante : « a notoriously weak link in the economics of climate
change ».
xxx Résumé en français

présente dans ces études, puisque les défis auxquels sont confrontés les pays en dé-
veloppement sont quelque peu différents de ceux auxquels sont confrontés les pays
développés.

Dans les pays en développement, le secteur agricole représente généralement une part
importante du PIB national. Ces pays dépendent donc fortement de leur secteur agri-
cole, de sorte que des conditions météorologiques défavorables peuvent avoir des ré-
percussions importantes au niveau national. Un choc météorologique néfaste affectant
la production agricole peut créer une situation de pénurie alimentaire entraînant une
augmentation des prix. En 2016, les pays d’Afrique australe ont été touchés par une
deuxième saison consécutive de sécheresse entraînant de mauvaises récoltes. Selon la
FAO, les conséquences sont dramatiques, exposant près de 40 millions de personnes à
l’insécurité alimentaire. La même année, de nombreux états indiens souffraient égale-
ment de fortes sécheresses couplées à une faible mousson. Ces conditions météorolo-
giques désastreuses ont créé un déficit en eau affectant la production agricole et en-
traînant une augmentation des prix. La situation économique de millions de personnes
s’en est vue menacée en Inde, pays dans lequel environ la moitié de la population tra-
vaille dans le secteur agricole.

Le changement climatique devrait être accompagné par une augmentation du nombre

d’événements extrêmes. Si le changement n’est pas trop important, l’être humain pourra
s’adapter aux nouvelles conditions. Un grand nombre d’études se penche sur la ques-
tion de l’adaptation, notamment celle des agriculteurs, pour déterminer des pratiques
permettant d’atténuer les effets globalement néfastes du changement climatique sur
la production alimentaire. Par exemple, Butt et al. (2005) ont montré que l’adoption
de nouvelles techniques agricoles, de cultures mixtes ou l’utilisation de variétés de
cultures plus résistantes à la chaleur peuvent aider les agriculteurs du Mali à mieux
faire face au changement climatique et à atténuer les dommages généraux causés par
ce changement. Di Falco et al. (2011) ont constaté que les ménages agricoles qui se sont
adaptés à un climat changeant ont tendance à produire plus que les ménages agricoles
qui ne se sont pas adaptés et que l’accès au crédit ainsi que la détention d’information
relative au climat sont des déterminants importants de l’adaptation. Si les coûts de
l’adaptation sont trop élevés, les personnes confrontées à des conditions météorolo-
giques trop hostiles pourraient être forcées d’émigrer, comme le montrent Black et al.
(2011).

Dans les pays à haut revenu, dans lesquels le secteur agricole n’est pas aussi impor-
tant en termes de PIB, les défis sont quelque peu différents mais méritent toutefois
 Résumé en français xxxi

d’être soigneusement étudiés. Des pays comme les États-Unis d’Amérique, l’Australie
ou les membres de l’Union européenne sont des acteurs clés du marché agricole. Selon
la FAO, au cours des trois dernières années, environ 40% de la production mondiale
de céréales provenaient de pays développés (FAO, 2017), principalement d’Amérique
du Nord et d’Europe. Les pays développés sont souvent les producteurs principaux
sur les marchés agricoles. Par conséquent, des accidents météorologiques peuvent for-
tement affecter la production mondiale et générer des problèmes de sécurité alimen-
taire. En 2003, la température moyenne d’été en Europe était jusqu’à 6o C au-dessus
de sa moyenne moyenne de 1998-2002 et les précipitations étaient 50% inférieures à la
moyenne (Ciais et al., 2005). Les pays d’Europe occidentale ont été gravement touchés
par la vague de chaleur de l’été, avec une augmentation de la mortalité et des incendies
de forêt. Le secteur agricole a également souffert de la vague de chaleur : par rapport
à l’année précédente, la production des cultures a diminué de 36% et de 30% en Ita-
lie et en France, respectivement ; le déficit fourrager était du même ordre (Easterling
et al., 2007) ; 4 millions de volailles sont morts et la production de lait a été réduite
(García-Herrera et al., 2010). Quelques années plus tard, en 2012, les États-Unis ont
également subi un choc météorologique extrêmement négatif qui a affecté sa produc-
tion de maïs. Selon le ministère de l’Agriculture des États-Unis, la production agricole
a diminué de 13% par rapport à l’année précédente. Comme les États-Unis sont parmi
les plus grands exportateurs mondiaux du maïs, et puisque les réserves mondiales de
maïs étaient faibles à ce moment-là, la baisse de la production a eu une incidence sur
les prix mondiaux de cette céréale, qui ont augmenté de 25% pour atteindre un sommet
encore supérieur à celui enregistré au cours de la crise des prix alimentaires de 2007
à 2008 (Chung et al., 2014). En somme, lorsque des acteurs importants des marchés
agricoles sont touchés par un événement climatique affectant la production, le mar-
ché s’adapte à la situation de carence par une augmentation du prix, pouvant conduire
à une situation de crise alimentaire.

Par conséquent, les conditions météorologiques affectent les pays en développement

et les pays développés de manière différente.

Plan de la thèse
Cette thèse vise à contribuer au débat théorique et empirique sur le changement cli-
matique et l’agriculture. Elle est structurée en deux parties contenant quatre essais
xxxii Résumé en français

complémentaires utilisant une variété de méthodes économétriques et théoriques ré-

pondant à la fois aux besoins des questions soulevées dans chaque étude et au type de
données disponibles.

La première partie de la thèse consiste en deux analyses microéconomiques pour un

pays en développement, l’Inde, se concentrant d’abord sur le côté de l’offre, puis se
tournant vers le côté de la demande de la production agricole. La deuxième partie
concerne les pays développés. Elle commence par une étude régionale impliquant plu-
sieurs pays d’Europe occidentale, c’est-à-dire des pays ayant un impact significatif
sur le marché mondial de l’agriculture. La dernière étape met en évidence l’intérêt de
considérer une approche d’équilibre général pour examiner les interactions qui se dé-
roulent entre le secteur agricole et le reste de l’économie, dans le contexte d’une petite
économie ouverte.

Le chapitre 1, intitulé “Climate Change and Profits : a Ricardian Analysis” (« Une

analyse ricardienne »), examine les décisions individuelles des agriculteurs d’un pays
en développement dans lequel le secteur agricole représente une part non négligeable
de l’économie nationale. L’objectif est de comprendre les effets de la variabilité mé-
téorologique sur les profits agricoles indiens et d’examiner les effets potentiels du
changement climatique sur ces profits, sous différents scénarios climatiques. L’ana-
lyse conduite dans le premier chapitre adopte l’approche ricardienne qui lie les revenus
nets par acre en fonction du climat, de la ferme et des caractéristiques des ménages.
La fonction des revenus nets par acre est estimée à l’aide de données détaillées en
coupe transversale. Bien que la plupart des études sur l’agriculture indienne soient
menées sur des données à l’échelle du district, cette étude se déroule au niveau de l’ex-
ploitation individuelle. La répartition spatiale des observations permet une couverture
raisonnable de l’ensemble du pays, fournissant des résultats généralisables à l’échelle
indienne. Les mécanismes expliquant les variations des revenus nets par acre sont exa-
minés à l’aide d’une régression quantile, ce qui permet une meilleure compréhension
des impacts des variations climatiques sur la répartition des revenus nets par acre.

Les résultats empiriques montrent que les exploitations ayant des revenus nets par
acre plus élevés semblent être plus affectées par les variables météorologiques en ma-
gnitude. Les exploitations dont les revenus nets par acre sont inférieurs ont tendance
à bénéficier davantage de la pratique de diversification des cultures que des exploi-
tations à fort revenu par acre. Dans une deuxième étape, deux scénarios climatiques
différents selon les hypothèses sur les variations de la température moyenne et des
 Résumé en français xxxiii

précipitations totales sont envisagés. Les exploitations agricoles ayant un faible re-
venu net par acre connaissent des pertes moins importantes en grandeur mais plus
importantes en pourcentage que les exploitations avec des revenus nets élevés par
acre. À l’échelle du district, les résultats montrent plus d’hétérogénéité. Dans les deux
scénarios, les districts du nord de l’Inde ont tendance à connaître une diminution des
revenus nets par acre alors qu’un effet opposé est trouvé pour les districts du sud du
pays.

Le chapitre 2, intitulé “Climate Change and Food Security : a Farm-Household

Model” (« Changement climatique et sécurité alimentaire : un modèle de ménage »),
utilise les résultats du premier chapitre comme point de départ : des effets hétéro-
gènes du climat sur les profits agricoles ont été mis en exergue, en fonction du type
de ménage. Dans le deuxième chapitre, les effets des variations climatiques sur les dé-
penses de production et de consommation agricoles sont étudiés pour différents types
de ménages agricoles, en fonction de leur participation au marché du travail. Dans un
premier temps, l’étude évalue les effets des variations climatiques sur la production
agricole des ménages agricoles indiens. Ensuite, les résultats de la première étape sont
utilisés pour estimer un système de demande quasi idéal (“Almost Ideal Demand Sys-
tem” pour différents types de ménages agricoles classés en fonction de leur régime
de travail. Différents scénarios sur les prix ainsi que sur les variables climatiques sont
ensuite testés pour évaluer les effets sous-jacents sur les consommations de décision.
Comme dans le premier chapitre, l’étude utilise des données au niveau de l’exploita-
tion agricole individuelle avec une couverture spatiale de l’Inde raisonnable. En outre,
l’analyse complète la vue donnée dans le chapitre 1 en explorant à la fois l’offre et le
côté de la demande des ménages agricoles.

La production agricole est sensible à la fois aux températures et aux précipitations. Les
décisions de consommation sont également affectées par les conditions climatiques. En
particulier, une augmentation de la pluviométrie totale entraîne une demande accrue
de biens purement marchands et de produits d’origine animale ainsi qu’une diminu-
tion de la demande de céréales et de loisirs. En outre, la demande de céréales est plus
affectée par la variation des précipitations pour les ménages autarciques par rapport
aux autres types de ménages ruraux. En revanche, la demande pour les produits agri-
coles des ménages autarciques est moins affectée par la variabilité des températures.
Les scénarios dans lesquels les précipitations et les températures sont modifiées pré-
sentent l’existence d’un arbitrage entre la demande de céréales et les produits d’origine
animale.
xxxiv Résumé en français

Le chapitre 3, intitulé entitled “Climate Change and Agricultural Yields : an Euro-

pean Case Study” (« Rendements agricoles : étude du cas européen »), considère une
vue plus agrégée pour des pays développés économique, ceux d’Europe occidentale.
Une dimension temporelle est introduite en contraste avec les deux analyses précé-
dentes. Cette caractéristique supplémentaire permet d’introduire la question de la vo-
latilité des prix, qui s’est avérée fondamentale sur les marchés agricoles ces dernières
années, bien que fréquemment ignorée dans la littérature. La volatilité des prix est
en fait une question clé pour les pays européens, en particulier dans le contexte de
l’abandon du soutien des prix pour les agriculteurs dans la Politique Agricole Com-
mune (PAC). Le troisième chapitre intègre cette volatilité des prix de production dans
une tentative d’étudier la relation entre les variations climatiques et les rendements
de deux grandes cultures céréalières, à savoir le blé et le maïs. Les effets du change-
ment climatique sur l’agriculture européenne dans différents scénarios alternatifs sont
étudiés de manière empirique.

Les résultats empiriques présentent les effets des variables météorologiques saison-
nières sur les rendements moyens ainsi que sur leur variabilité, pour le blé et le maïs.
Les prix ont un impact positif et significatif sur les rendements du blé pour le nord de
l’Europe, seulement après la réforme de la PAC. Avant cette réforme, l’effet des prix
sur les rendements n’est pas statistiquement différent de zéro. Les modèles empiriques
sont ensuite utilisés pour évaluer l’effet du changement climatique sur les rendements.
Quatre scénarios de projection du climat reflétant les trajectoires de concentration des
gaz à effet de serre sont testés. Des effets spatio-temporels mitigés sont trouvés. Les
rendements du blé augmenteraient à l’échelle européenne dans la plupart des scéna-
rios, mais les gains diminueront avec le temps pour les régions du nord, à long terme.
Les résultats sont moins optimistes pour les rendements du maïs. À court terme, cer-
taines régions du Nord connaîtraient des gains de rendement, mais ces gains se trans-
formeraient en pertes sur le long terme. Ces pertes seraient même plus élevées dans
le sud de l’Europe.

Le chapitre 4, intitulé “Climate Change and Business Cycles” (« Changement clima-

tique et cycles économiques »), fournit une approche en équilibre général, contraire-
ment aux trois premiers chapitres basés sur un cadre d’équilibre partiel. L’objectif de
ce dernier chapitre est d’étudier l’influence des chocs météorologiques sur les cycles
économiques, grâce à un modèle dynamique estimé pour une petite économie ouverte.
Un modèle d’équilibre général intertemporel stochastique (MEGIS) original avec un
 Résumé en français xxxv

secteur agricole dépendant de la météo est développé et estimé en utilisant des mé-
thodes bayésiennes et des données trimestrielles pour la Nouvelle-Zélande au cours
de la période d’échantillonnage allant de 1989 à 2014.

Les résultats du modèle suggèrent que les chocs météorologiques jouent un rôle impor-
tant en expliquant les fluctuations macroéconomiques au cours de la période d’échan-
tillonnage. Un choc météorologique mesuré par un indice de sécheresse agit comme
un choc d’offre négatif caractérisé par une baisse de la production et une hausse des
prix. Par ailleurs, les résultats montrent que les agriculteurs ne prévoient pas les chocs
météorologiques et sont plutôt surpris par la variabilité des conditions climatiques.
Enfin, l’augmentation de la variance des chocs météorologiques reflétant les change-
ments climatiques potentiels entraîne une augmentation considérable de la volatilité
des principales variables macroéconomiques, telles que la production et l’inflation.
General Introduction

By the end of 2015, nations around the world gathered in Paris for the twenty-first
yearly session of the United Nations Conference of the Parties on Climate Change.
The goal of that conference was to achieve a legally binding and universal agree-
ment on climate, with the aim of mitigating global warming. This warming is be-
lieved, on the basis of the unequivocal scientific evidence, to primarily be caused by
the increase in the concentration of greenhouse gas emissions resulting from human
activities. While climate change may have positive effects on some regions, a vast
number of scientific studies suggests that overall, it poses threats on the future of
our planet. Consequently, the expectations were high for the United Nations Climate
Change Conference in Paris. World leaders recognized the need to take action. To
avoid the most catastrophic effects of climate change, signatory governments have
agreed to sign a deal to keep the global temperature rise well below 2 degrees Celsius
above pre-industrial levels and to pursue efforts to limit it to 1.5 degree Celsius. This
agreement entered into force less than a year later, in November 2016, and was ratified
by 142 parties out of 195.

Implicitly, relying on a vast body of scientific literature, the signatories recognized

that the 2 degrees limit represents a threshold below which adaptation seems still
reasonable. However, the way of achieving these goals as well as the underlying costs
of adaptation are still surrounded by lots of uncertainties. In fact, the financial cost of
climate change is an economic issue that gave rise to a significant number of studies.
Not only the literature attempts to assess the costs and benefits of mitigation, but it
also tries to evaluate the cost of doing nothing (OECD, 2015). A review of this literature
can be found in Dell et al. (2014).

A popular approach, introduced by Nordhaus (1994), consists in using dynamic math-

ematical models to represent the world. The basic idea is to consider a damage function
linking temperature to GDP, the former being dependent on the level of greenhouse

1
2 General Introduction

gas emissions. These models can be used to evaluate the potential economic bene-
fits of policies (see, e.g., Hope et al., 1993; Tol, 2005). There is however no consensus
reached on what needs to be done, as some authors recommended urgent action (e.g.,
Stern, 2007) while others suggested a more progressive approach with more action in
the future and less control in the short run (e.g., Nordhaus, 2007).

This global approach of measuring the impacts of climate change is complemented in

the literature by partial approaches focussing on specific economic sectors. One of
the particularly vulnerable sector is agriculture, which strongly depends on climate.
Surprisingly enough, it is not explicitly mentioned in the decisions adopted by the
Conference of the Parties in Paris. Yet, the potential impacts of climate change on ag-
riculture, in spite of the relatively limited economic size of that sector, may pose sub-
stantial threats (OECD, 2015), especially regarding food security and hunger. A vast
number of studies is thus devoted to the documentation of the relationship between
agriculture and climate. A brand of the literature simulates crop growth using math-
ematical models (see, e.g., Ritchie and Otter, 1985; Jones et al., 1986; Brisson et al., 1998).
Coupled with climate models that simulate projection of the weather, crop growth
models can be used to assess the effects of climate change on food sustainability (see,
e.g., Rosenzweig and Parry, 1994; Jones and Thornton, 2003; Parry et al., 2004). An-
other part of the literature focuses on the effects of varying climate on land value (see,
e.g., Mendelsohn et al., 1994; Schlenker et al., 2005) or on agricultural profits (see, e.g.,
Deschênes and Greenstone, 2007), in an attempt to assess the potential consequences
of climate change either at the global level, or at the regional level. Studies conducted
at the global scale provide evidence that developed and developing countries will not
be impacted by climate change the same way (Rosenzweig and Parry, 1994; Fischer
et al., 2005), the latter being more vulnerable to climate variations.

This thesis consequently aims to contribute to the scientific effort of investigating

the potential impacts of climate change on agriculture. It relies on both theoretical
and empirical methods to provide four analysis covering different geographical areas
concerning either developing or developed countries.

The remainder of this introduction draws up the key concepts of this thesis. Section 1
briefly describes what climate change is and gives details on the different climate scen-
arios used by researchers to estimate the potential effects of climate change on our
planet. Section 2 presents in more details the attempts of the literature to examine the
 General Introduction 3

effects of the weather and climate on agriculture and briefly reviews the main meth-
ods employed by scholars to address the effects of climate change on the agricultural
sector. Finally, section 3 describes the structure of the thesis, organized in 4 chapters.

1 Climate is Changing
The scientific evidence is now overwhelming: global climate is warming. The Inter-
national Panel on Climate Change, an international body set up in 1988 for assessing
the science related to climate change and regrouping thousands of scientists defines
climate change as “a change in the sate of the climate that can be identified (e.g., by
using statistical tests) by changes in the mean and/or the variability of its properties, and
that persists for an extended period, typically decades or longer” (Edenhofer et al., 2014).
The results from the last 2014 report from the IPCC predict a likely global increase in
the mean surface temperature partly due to increasing anthropogenic (i.e., caused by
human activity) concentration in greenhouse gases. The higher the concentration, the
higher the increase in global mean surface temperature. Since the pre-industrial era,
i.e., since around 1750, the concentration of greenhouse gases such as carbon dioxide
(CO2 ), methane (CH4 ) or ozone (O3 ) has significantly increased. The concentration of
carbon dioxide in the atmosphere has in fact increased by 43% since the pre-industrial
era to reach 399.5 parts per million (ppm) in 2016. In the meantime, methane con-
centration has increased by 154% to reach 1834 ppm in 2016, and tropospheric ozone
increased by 42% to reach 337 ppm in 2016 (Blasing, 2009). In the forthcoming years,
greenhouse gas concentration is forecasted to rise even more. There is, however, a
great uncertainty regarding both the concentration value by the end of the 21st cen-
tury and the pathway leading to this value. These concentrations will depend, among
other things, on demographic growth, on social development, and on technological
change. Lots of assumptions thus need to be made, particularly on human activity, to
assess the possible levels of greenhouse gas concentration in the near future. The pro-
jected outcome of climate variables depends on these assumptions. In order to provide
a common ground to researchers, it is common practice to rely on well defined scen-
arios, developed by the research community. This then makes possible the analysis of
possible climate change and its impacts.

The IPCC, in its last report (Edenhofer et al., 2014), adopted four different trajectories
of greenhouse gas emissions, called Representative Concentration Pathways (RCPs),
representing possible outcomes based on the scenarios used in the literature. Their
names reflect the value of radiative forcing at the horizon at which they are simulated,
4 General Introduction

i.e., 2100 : the RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5, are characterized by a level of
radiative forcing in 2100 of 2.6W /m2 , 4.5W /m2 , 6.0W /m2 , and 8.5W /m2 , respect-
ively. The radiative forcing of a gas corresponds to the difference between incoming
solar radiation and outgoing infra-red radiation, and is influenced by the concentra-
tion of that gas. The higher the concentration, the higher the balance of incoming and
outgoing radiations, resulting in higher surface temperatures. Hence, global surface
temperature on Earth is expected to be the lowest for the RCP 2.6 scenario, and the
highest for the RCP 8.5 one. The path to reach the levels of radiative forcing differs
among the scenarios, as shown in the left panel of fig. 3. In the first scenario, the
RCP 2.6, i.e., the less pessimistic in terms of greenhouse gas concentration, a peak in
radiative forcing is reached around 2030 and then slowly declines. The two next scen-
arios, the RCP 4.5 and RCP 6.0 are characterized by higher levels of radiations, with
a stabilization without overshoot pathway to 4.5W /m2 and 6.0W /m2 , respectively.
The last scenario, the RCP 8.5, is more pessimistic and reflects continuously growing
greenhouse gas emissions leading to a radiative forcing value of 8.5W /m2 by 2100. It
is considered as a high emission scenario. More details can be found in Van Vuuren
et al. (2011).

Radiative Forcing (w/m2 ) Mean Temperature Change (o C)

4
7 3

2
5
1
3 0

-1
2000 2025 2050 2075 2100 1900 1950 2000 2050 2100

Historical RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Notes: Each curve represents the trend in radiative forcing (left panel) and the corresponding change in global mean temperature
change relative to 1986–2005 (right panel) for one of the four Representative Concentration Pathways. The graph on the left is a
reproduction of Figure 10 from Van Vuuren et al. (2011), the graph on the right is a reproduction of Figure 12.1 from Collins et al.
(2013).

Figure 3: Trends in Radiative Forcing and Global Mean Temperature Change

The trends in radiative forcing of each scenario are fed into climate models, to simulate
potential climate up to 2100, at various spatio-temporal scales. The average change
in global mean temperature for the four scenarios is reported in the right panel of
fig. 3. By the end of the 21st century, relative to 1986–2005 levels, according to the
IPCC results (Collins et al., 2013), the average temperature change based on the RCP
2.6 scenario is likely to be comprised between 0.3o C and 1.7o C. Under the RCP 4.5 and
 General Introduction 5

6.0, the change is higher, with values comprised between 1.1o C to 2.6o C and 1.4o C to
3.1o C, respectively. Under the worst-case scenario, i.e., the RCP 8.5, the likely global
average change in mean temperature is comprised between 2.6o C to 4.8o C.

These values of changes are averages at the global scale. As a matter of fact, there
is a lot of heterogeneity in the projections of climate, with an expected change over
land higher than that of the oceans. In addition, the changes over land should not be
uniform; some regions should experience higher temperatures while other should be
subject to colder climates.

A modification in climate statistics has a direct impact on the weather. A distinction

between these two notions needs to be made, as pointed out by Schlenker et al. (2006).
The main difference between weather and climate is time scale, as the weather refers
to meteorological conditions at a distinct moment in time, whereas climate refers to
an average of climate statistics over an extended period of time. Hence, if we consider
meteorological conditions as the realization of multiple random variables, the weather
can be viewed as the short-run draws from these variables while climate can be viewed
as the average of these draws in the long-run. A simplified example can be considered
to get a better idea of the underlying effects of a modification in climate statistics on
the weather. For convenience, we can suppose that observed surface temperatures are
drawn from a Normal distribution, as depicted in fig. 4. Then, as shown in fig. 4(a),
an increase of the surface temperature would lead to an increase in the probability
of occurrence of hot weather and extreme hot weather, accompanied by a decrease
in cold and extreme cold weather. However, in its last report, the IPCC predicts an
increase in the variability of the weather as well. Let us consider first the effect of an
increase in the variance of temperatures alone, keeping the mean value as its historical
average. Figure 4(b) illustrates such a case. If the variance increases, the distribution
becomes more flat, thus increasing the probability of occurrence of both cold and hot
weather, as well as extreme temperatures. Figure 4(c) illustrates the case in which both
mean and variance shift upwards. In this situation, more hot weather and much more
extreme hot weather would occur.

This illustration is a simplification of what is in reality expected with climate change,

as one should also consider spatial heterogeneity as well as seasonal effects. Yet it
enhances understanding of the basics of the possible effects of climate change. It be-
comes easier to imagine how the number and the severity of extreme events such as
tornadoes, heavy rainfall or droughts should rise by the end of the century.
6 General Introduction

Less
cold More
weather hot
Less More
weather
extreme cold extreme hot
weather weather

(a) Increase in Mean

More More
More cold hot
More
extreme cold weather weather
extreme hot
weather weather

(b) Increase in Variance

Almost More
unchanged hot
cold More
weather
Almost unchanged weather extreme hot
extreme cold weather
weather

(c) Increase in Both Mean and Variance

Notes: The solid and dotted lines represent the previous climate distribution and the new climate distribution, respectively. This
figure is a reproduction of Figure 1.8 from Cubasch et al. (2013).

Figure 4: Effects of Climate Change on the Distribution of Temperatures

However, a lot of uncertainty remains about the magnitude of these changes and their
effects on our society, especially the economic ones. Agricultural systems lie at the
very heart of the challenges of climate change, due to the particular dependency of
agriculture on the weather conditions.

2 Modelling the Consequences of the Weather

and Climate Change on Agriculture
Agricultural systems tightly depend on climatic variables such as temperature and
precipitation. Some countries have natural conditions more suitable for agricultural
activities than others. For example, crops farming is easier in temperate areas than
 General Introduction 7

in tropical zones, because the temperatures and the precipitation levels experienced
in the former area fit more the needs of certain cereals such as wheat or corn. The
weather can in fact be viewed as a direct input in the production function. However,
contrary to other inputs such as labour, fertilizers, more or less sophisticated irrigation
systems, machinery, or farming methods, the weather cannot be controlled by farmers.
Hence, agricultural production may be vulnerable to poor weather conditions. This
dependence of agricultural systems on the weather poses significant challenges in the
context of climate change. As previously mentioned, climate change is expected to
lead to a modification in the weather patterns, accompanied by an increase is extreme
events that may cause substantial losses in production. Climate change is very likely
to affect food security at the global level as well as at the country level, via a lower
access to food. This raises concerns about one of the many challenges the agricultural
sector must face, that is its capacity to produce enough food to feed the population.
Besides, the number of people living on the planet is expected to keep growing in the
forthcoming years, making this challenge even harder. The global crop demand from
2005 to 2050 is expected to increase by 100-110% (Tilman et al., 2011). Hence, it is of
primary importance to study the relationship between agriculture and the weather, to
be able to adapt to new weather conditions in the near future.

To be able to tackle these challenges, a large part of the literature focuses directly on
the agricultural production functions or, using a framework named Ricardian, looks
into the agricultural profits made by farmers. Another part of the literature looks at a
different scale and considers how the responses of the agricultural system to climate
modifications affect the rest of the economy.

2.1 The Agricultural Production Function

The number of scientific contributions documenting the relationship between the weather
and agricultural production function has considerably grown in the last decades. Two
different methodologies stand out in that part of literature. The first one simulates crop
yields using agronomic models, while the second method relies on statistical models
to examine the observed links between yields and the weather.

Crop simulation models are used to model crop growth. Some are designed for a spe-
cific crop, such as CERES-Maize for corn (Jones et al., 1986) or CERES-Wheat for wheat
(Ritchie and Otter, 1985). Others can be adaptable to a wide range of crops, such as
STICS models that can model wheat, corn, soybean, sorghum and many other crop
8 General Introduction

growth (Brisson et al., 1998). In both cases, the development stages of crops are mod-
elled through mathematical equations in a first step. These equations account for mul-
tiple factors such as soil conditions, the weather and management practices (Mearns
et al., 1997). Once the model is established, it needs to be calibrated. To that end,
crops are grown in fields or laboratory settings, subject to different environmental
conditions, including varying weather conditions and varying levels of carbon diox-
ide. As previously mentioned, the concentration of greenhouse gas in the atmosphere
affects climate through radiative forcing, but it also plays a role in the physiological
process of photosynthesis and transpiration (Field et al., 1995). Hence, being able to
capture the fertilizing effects of carbon dioxide represents a strength of crop simula-
tion models, when they are fed with new data to simulate how crop growth is affected
by new conditions. If these new conditions reflect a varying climate in accordance
with a projected scenario, crop simulation models become a powerful tool to study
the potential benefits and harms brought on by climate change.

The response of crop production to climate change has been extensively studied in
the literature at the regional scale (see, e.g., Aggarwal and Mall, 2002 for rice yields in
India; Xiong et al., 2009 for corn, wheat, and rice yields in China; Jones and Thornton,
2003 for maize yields in west Africa) ; or at the global scale (see, e.g., Lobell and Field,
2007 for wheat, rice, maize, and soybean; Rosenzweig et al., 2014 at for corn, wheat,
rice and soybean). Recently, Bassu et al. (2014) have shown that using an ensemble of
multiple models enables to simulate with better accuracy absolute yields than using a
single model, and that increasing temperature strongly diminished corn yields.

Crop simulation models however present some drawbacks. First, they are calibrated in
specific locations, and therefore may represent very well the different growth stages
for that particular region, but might not be representative of all the regions of the
world. Another major shortcoming of these models is that they fail to account for
the possibility of farmers to adapt to a new climate condition. These models, even if
adaptable to multiple crops, are calibrated for a given type of crops, and therefore do
not model the possibility of the farmer to switch crops if the climate conditions are
no longer suitable for that crop. In addition, all differences in outcomes are assumed
to be due to changes in the variables affecting crop growth, such as temperature or
precipitation. This pitfall probably leads crop simulation models to overestimate the
effects of climate change (Adams et al., 1990; Parry et al., 2004). Besides, economic
conditions such as price variability are not taken into account by these models, so that
 General Introduction 9

farmers are considered blind to their economic environment and therefore unable to
switch to an activity offering a better opportunity.

The issue of location-specific representativity can be tackled by statistical models. For

these frameworks, the variations of crop yields are modelled as the response to the
variations of other variables such as inputs used, economic conditions faced by farm-
ers, quality of soil, the weather or climate conditions, and so on. The method consists
in relying on historical observations to estimate the functional form linking the vari-
able of interest to its predictor. It is thus possible to examine the determinants of
crop yields at larger spatial scales than those used in crop simulation analyses, relying
on data observed at multiple locations. Once the statistical model is estimated, it is
possible to assess new weather or climate conditions and predict the outcome of the
variable of interest in such conditions. The predicted value can de facto be compared to
the historical realization, and the change can be attributed to the varying conditions.

In statistical models, the weather is considered as a special input in agricultural pro-

duction as it cannot be controlled in natural setups, and is therefore regarded as an
exogenous variable. Thus, the weather acts like a “natural experiment” (Angrist and
Krueger, 2001, via Auffhammer et al., 2013). researchers can only observe the output
of the experiment without being able to manually control for the amount of the nat-
ural input. However, the causal effects of the weather on crop yields can be examined,
and has already lead a great deal of studies to try to quantify it. However, there is no
consensus on the magnitude of the effects of climate change on crop yields. Lobell
et al. (2011) investigated the historical worldwide relationship between temperature
and four crops: corn, rice, wheat and soybean, from 1980 to 2008. Their results sug-
gest a negative impact of temperature on both corn and wheat yields, reducing yields
by 3.8% and 5.5%, respectively, with regional heterogeneity. Their results regarding
the two other crop yields are less conclusive, with global declines of −0.1% for rice
and −1.7% for soybean, with gains observed in some regions cancelling out losses un-
dergone in other regions. You et al. (2009) found similar results for wheat yields in
China, between 1979 and 2000. They estimated a reduction of wheat yields by 3% to
10% due to a one degree mean growing season temperature increase. Lobell and Asner
(2003) estimated the impact of increasing temperatures in counties in the United States,
using data from 1982 to 1998. According to the results of the article, each additional
degree of the average temperature leads to a decrease of 17% of both corn and soybean
yields. Simulations of changing climate are more profoundly assessed in the literature
10 General Introduction

using climate scenarios. For example, Schlenker and Roberts (2009) considered dif-
ferent scenarios for the United States, and concluded that warming temperatures by
mid 21st century would result in a substantial decrease in both corn, cotton, and soy-
bean yields, relative to observed yields from 1950 to 2005. More specifically, the losses
would range from −30% to −46% in an optimistic scenario, and would even be more
damaging in the worst-case scenario, ranging from −63% to −82%. Schlenker and
Lobell (2010) used historical data in Sub-Saharan Africa to link crop yields to weather
variation and then used the estimated model to assess the effects of climate change on
maize, sorghum, millet, groundnut, and cassava yields. For all crops, declining yields
are projected, ranging from −8% to −22% by mid 21st century relative to 1961–2000
observed yields.

All these studies consider the effects of covariables on the mean of the response vari-
able. Some studies (e.g., Chen et al., 2004; Cabas et al., 2010) suggest to use Just and
Pope’s stochastic production function approach (Just and Pope, 1978), therefore allow-
ing to characterize the effects of the weather on both crop yields and their variability
(McCarl et al., 2008).

One of the advantage of using statistical models rather than crop simulation models
is that the former clearly state the model uncertainties by giving statistical indicators
regarding the quality of the estimation, which is not the norm with crop simulation
models (Lobell and Burke, 2010). However, statistical models present some caveats.
Even if it is common to use multiple weather variables to estimate the variations of
crop yields, this practice might cause problems of multicolinearity (Sheehy et al., 2006;
Lobell and Ortiz-Monasterio, 2007), especially when weather variables are disaggreg-
ated to reflect seasonal effects. Another shortcoming of statistical models is what
Mendelsohn et al. (1994) refers to as the “dumb farmer scenario”. That is, in these
models, the focus is made on one type of crop only, just like in crop simulation mod-
els, therefore considering that farmers are not able to adapt to a varying environment
offering new climate conditions. Hence, statistical analysis of crop yields might not
be a good tool for predicting long-term variations.

2.2 The Ricardian Framework

Crop simulating models and their statistical counterpart both fail to account for the
ability of farmers to adapt to a new environment. An answer is provided by Mendel-
sohn et al. (1994), who suggest to look at land value instead of yields. In their pioneer
 General Introduction 11

work, they introduce an analysis named “Ricardian analysis” after the economist David
Ricardo.

The Ricardian methodology is a hedonic model of farmland pricing that focuses on

land value. The basic idea is that long run climate should be capitalized into land val-
ues. In a competitive market, the price of farm land is assumed to reflect the discounted
value of all the expected future profits that can derive from it. The Ricardian approach
measures the impact of climate variables on land productivity or farm land values by
exploiting cross sectional differences in land use and weather patterns. It is note-
worthy that climate variables are used in the Ricardian analysis, rather than weather
variables. As a matter of fact, an adverse weather shock such as a storm or a drought
can lead to losses on crop yields, but may have ambiguous effects on profits: on the
one hand, the output can be reduced and thereby lower profits; on the other hand, in
the presence of inelasticity of demand for agricultural goods, the market can adjust
by raising prices, thus increasing agricultural profits. But in the long-run, profits are
expected to be lowered by an adverse shift in climate (Schlenker et al., 2006). Besides,
in the short-run, farmers may be vulnerable to weather shocks, but in the long-run,
they can adopt new farming strategies and adjust their decisions regarding the levels
of inputs in response to the climate they face. Hence, using climate variables rather
than weather variables is more suitable for the Ricardian framework.

In the last two decades, the original framework has been widely applied in many coun-
tries across the world. Most previous research use cross sectional data to estimate farm
values on weather variables. Extensive reviews of applications for African, Asian,
South American, and US regions can be found in Mendelsohn and Dinar (2009).

In the pioneering article of Mendelsohn et al. (1994), the focus was made on US farmers,
and the results highlight regional heterogeneity in the effects of climate on land values.
The climate scenarios tested yield mitigated results, with winners and losers. The
losses undergone by losers are however projected to be relatively lower than those
projected using an analysis based on crop yields. Mitigated results are also found
in Africa. For instance, Kurukulasuriya and Mendelsohn (2008) showed, using farm-
level data from 11 African countries collected between 2003 and 2004, that the effects
of climate change on net revenues differ between irrigated farms and rainfed-only
farms. The net revenues of the formers are projected to increase up to 51% in the best-
case scenario, while the net revenues of the latter are projected to fall by up to −43%
in the case of a hot and dry climate scenario. Similar results between irrigated and non
irrigated farms were found in China (Wang et al., 2009). In Europe, Van Passel et al.
12 General Introduction

(2016) estimated the effects of climate changes using a large sample of farms across
Western Europe. Although the study highlights different regional effects, it also shows
that European farms are more sensitive to projected warming than US farms.

There have been a number of criticisms of the Ricardian approach. One major weak-
ness lies in the fact that unobservable skills of farmers are not incorporated in the
analysis. This shortcoming has led some scholars to introduce time-varying data to
account for the problem of omitted variables by including district or regional dummies
in the model (see, e.g., Schlenker et al., 2006; Deschênes and Greenstone, 2007; Kim
et al., 2009; Cabas et al., 2010). Unlike crop-simulation models, the Ricardian frame-
work does not consider the fertilizing effects of CO2 , yielding biased results. Another
weakness of the Ricardian model lies in the fact that it does not incorporate price ef-
fects. In the Ricardian approach, the variation in land values over climate zones is due
to changes in climate variables. A strong assumption is that input and output prices
remain unchanged. Special attention should be done on that point as crop prices tend
to be more volatile (Cline, 1996; Schlenker et al., 2005).

Finally, the model implicitly assumes that farmers do not face adjustment costs. Hence,
Ricardian results should be considered as a lower-bound estimate of the costs of cli-
mate change (Quiggin and Horowitz, 1999).

2.3 A Global Analysis of the Economic Impacts

Instead of focusing solely on the agricultural production function, a substantial num-

ber of studies rather assesses the impacts of climate change on the economy. Some
attempts are made either at the global scale or on more specific geographical areas.

Studies considering the global economy highlight the existence of winners and losers
under new conditions reflecting climate change. A widely used methodology consists
in predicting the response of crop yields to projected climate scenarios. These scen-
arios are commonly given by a general circulation model (GCM), i.e., a climate model.
Then, the predicted agricultural yields are fed into a general equilibrium model that
evaluates agricultural production as well as crops prices. It is also possible to address
food security issues by assessing the number of people at risk from hunger, depend-
ing on the population, on the agricultural production and on crops prices. Rosenzweig
and Parry (1994) provided a famous contribution to this literature. They identified a
distinction between developing and developed countries, the former being more vul-
nerable to climate change than the latter. This distinction is a common result in the
 General Introduction 13

literature (see e.g., Parry et al., 2004; Fischer et al., 2005), due to the predominance of
agriculture in developing countries (Tubiello and Fischer, 2007). Another explanation
to this result is that developing countries are disproportionately placed at low latitudes
or near the tropics, where climate conditions are not optimal for farming activities, due
to high variability in the weather.

Another way of looking at the global effects of climate change on the economy is to
use another type of general equilibrium approach called Integrated Assessment Models
(IAMs). IAMs are mathematical models that combine in a same framework knowledge
about climate, economics, demographics and political decisions that influence green-
house gas emissions (see, e.g., Nordhaus, 1991; Nordhaus, 1994; Nordhaus and Yang,
1996; Tol, 2002). These models evaluate the social cost of carbon, i.e., the monetary
estimate of the damage caused directly or indirectly by the emission of a tonne of CO2 ,
by assessing changes in welfare due to new climate conditions. The mechanism relies
on the existence of a damage function that links temperature to GDP, such that an
increase in temperature might be harmful to the global production. IAMs can be used
as a tool for policy analysis, by submitting different climate scenarios to the model and
by modelling different policies based on the social cost of carbon. IAMs, as any other
tool, come with shortcomings. One of them concerns the damage function. As stated
by Weitzman (2010), the damage function is “a notoriously weak link in the econom-
ics of climate change”, because assumptions need to be made regarding the functional
form of the damage function, and because these assumptions can greatly modify the
conclusions drawn using IAMs. Pindyck (2013) also noted that IAMs fail to account
for catastrophic weather outcomes, which might have very large impacts.

A lot of studies focus on more specific geographical areas. The distinction between
developing and developed economy still holds for these studies, as the challenges faced
by developing countries are somewhat different that that faced by developed countries.

In developing countries, the agricultural sector usually represents an important share

of the national GDP. These countries are thus heavily dependent on their agricultural
sector, so that poor weather conditions may have substantial impacts at the national
level. An adverse weather shock affecting the agricultural production may create a
situation of food shortage leading to an increase in prices. In 2016, Southern African
countries were hit by a second consecutive season of drought resulting in crop fail-
ure. According to the FAO, the consequences are dramatic, exposing almost 40 million
people to food insecurity. In the same year, many Indian states also suffered from
massive droughts coupled with weak monsoon. It created a water shortfall affecting
14 General Introduction

the agricultural production and leading to an increase in prices. The economic situ-
ation of millions of people was therefore threatened in India, a country in which about
half of the population works in the agricultural sector.

With climate change, the number of extreme episodes that have seriously negatively
affected developing countries in the past is expected to increase. Depending on the
intensity of climate change, people in developing countries may need to adapt to their
new climate environment. Many papers thus look at possible adaptation undertaken
by farmers that may help mitigating the negative effects of climate change on food
production. For example, Butt et al. (2005) showed that adopting new farming tech-
niques, mixing crops or more heat-resistant varieties of crops may help farmers in
Mali to better cope with climate change and mitigate the overall damages caused by
climate change. Di Falco et al. (2011) found that farm-households that adapted to a
changing climate tend to produce more than farm-households that did not adapt, and
that access to credit as well as information provision regarding climate are important
determinants of the adaptation. If the costs of adaptation are too high, people facing
too adverse weather conditions might be forced to emigrate, as shown by Black et al.
(2011).

In high-income countries, where the agricultural sector is not as important in terms

of GDP, the challenges are somewhat different but also deserve to be carefully in-
vestigated. Countries like the United States of America, Australia, or members of the
European Union, are key actors of world agricultural markets. According to the FAO,
during the last three years, around 40% of global cereal production came from de-
veloped countries (FAO, 2017), mostly from North America and Europe. Developed
countries are often leading producers on agricultural markets. Hence, weather acci-
dents may greatly affect world production and generate food security troubles. In 2003,
the summer average temperature in Europe was up to 6o C above its average 1998–
2002 mean and precipitation were 50% below the average (Ciais et al., 2005). Western
European countries were severely impacted by the heat wave that summer, with an
increase in mortality and in forest fires. The agricultural sector also suffered from the
heat wave: compared to the previous year, crop production decreased by 36% and 30%
in Italy and France, respectively; the fodder deficit was of the same order (Easterling
et al., 2007); 4 million broilers died and milk production was reduced (García-Herrera
et al., 2010). A few years later, in 2012, the United States also underwent an extreme
negative weather shock which affected its corn production. According to the United
States Department of Agriculture, crop production dropped by 13% compared to the
 General Introduction 15

previous year. As the United States are the exports leaders in global corn exports, and
since the world corn reserves were low at that time, the decrease in production im-
pacted world corn prices, that rose by 25% to reach a peak even higher than the one
recorded during the 2007–08 food price crisis (Chung et al., 2014). Hence, when these
countries experience poor weather conditions, it can negatively affect their produc-
tion and lead to an increase in prices, which can in turn create disturbances regarding
food security.

Therefore, weather conditions affect developing countries and developed countries in

a different way.

3 Outline of the Thesis

This thesis is intended to be a contribution to the theoretical and empirical debate re-
garding climate change and agriculture. It is structured into two parts containing four
complementary essays that employ a variety of econometric and theoretical methods
allowing to account for the main features of the countries studied.

The first part of the thesis consists of two microeconomic analyses for a developing
country, India, focusing first on the supply side and then turning to the demand side of
the agricultural production. The second part concerns developed countries. It starts
with a regional study applied on western Europe regions, i.e., in regions of countries
with a significant impact on agricultural world market. The final step highlights the
interest in considering a general equilibrium approach to examine the interactions that
take place between the agricultural sector and the rest of the economy, in the context
of a small-open economy, applied on New Zealand data.

Chapter 1, entitled “Climate Change and Profits: a Ricardian Analysis”, looks at

individual decisions of farmers in a developing country in which the agricultural sector
represents a non-negligible share of the national economy. The purpose is to under-
stand the effects of weather variability on Indian farming profits, and to examine the
potential effects of climate change on these profits, under different climate scenarios.
The analysis in the first chapter adopts the Ricardian approach that links net revenues
per acre as a function of climate, farm and households characteristics. The function
of net revenues per acre is estimated using detailed cross-sectional data. While most
studies on Indian agriculture are conducted on district-level data, this study is carried
out at the individual farm-level. The spatial distribution of the observations allows for
a fair coverage of the whole country, relying on 7, 751 individual observation within
16 General Introduction

202 districts, therefore providing generalisable results at the Indian scale. The mech-
anisms explaining the variations of net revenues per acre are examined using quantile
regression, thus allowing a deeper understanding of the impacts of climate variations
on the distribution of net revenues per acre.

Empirical results show that farms with higher net revenues per acre look to be more af-
fected by weather variables in magnitude. Farms with lower net revenues per acre tend
to benefit more from crop mixing than farms with high income per acre. In a second
step two climate scenarios which differ according to the assumptions on changes on
average temperature and total rainfall are envisaged. Farms with low net revenues
per acre experience losses less important in magnitude but larger in percent change
than farms with high net revenues per acre. At the district level, results show more
heterogeneity. Under both scenarios, districts in the North of India tend to experience
a decrease in net revenues per acre while an opposed effect is found for districts in the
South of the country.

Chapter 2, entitled “Climate Change and Food Security: a Farm-Household Model”,

uses the results of the first chapter as a starting point. We previously found heterogen-
eous effects of climate on profits among farmers, depending on the type of households.
In the second chapter, the effects of weather variation on both agricultural production
and consumption expenditures are investigated for different types of farm households,
depending on their labour market participation. As a first step, the study evaluates the
effects of climate variations on the agricultural production of Indian farm households.
Then, the results of the first step are used to estimate an Almost Ideal Demand System
for different types of farm households classified according to their labour regime. Dif-
ferent scenarios on prices as well as on climate variables are then tested to assess the
underlying effects on decision consumptions. As in the first chapter, the study uses
data at the individual farm level with a good spatial coverage of India. In addition, the
analysis complements the view given in chapter 1 by exploring both the supply and
the demand side of farm households.

We find that the agricultural production is sensitive to both temperature and rainfall.
Consumption decisions are also affected by climate conditions. In particular, an in-
crease in total rainfall leads to a higher demand for pure market goods and animal de-
rived products, and a decrease in crops and leisure. In addition, crops demand is more
affected by the variation in rainfall for autarkic households relative to other types of
rural households. On the contrary, the demand for crops products of autarkic house-
holds is less affected by varying temperatures. The scenarios in which both rainfall
 General Introduction 17

and temperatures are changed exhibit a trade-of between crops and animal-derived
products.

Chapter 3, entitled “Climate Change and Agricultural Yields: an European Case

Study”, considers a more aggregated view, for western Europe countries, i.e., de-
veloped economies. Moreover, a time dimension is introduced in contrast to the two
previous analysis. This additional feature allows us to introduce the question of the
volatility of prices, which has been a fundamental issue in agricultural markets in the
recent years, although frequently ignored in the literature. The volatility of prices is in
fact a key issue for European countries, especially in the context of the abandonment of
price support for farmers in the Common Agricultural Policy (CAP). The third chapter
incorporates the volatility of production prices in an attempt to study the relationship
between weather variations and yields of two major crops, i.e., wheat and corn. The
effects of climate change on European agriculture under different alternative scenarios
are empirically studied.

Empirical results exhibit the effects of seasonal weather variables on both mean yields
and the variance of wheat and corn yields. Prices show a positive and significant
impact on wheat yields for northern Europe, only after the CAP reform. Prior to this
reform, the effect of prices on yields were not statistically different from zero. The
empirical models are then used to assess the effect of climate change on yields. Four
climate projection scenarios reflecting greenhouse gas concentration trajectories are
tested. Mitigate spatio-temporal effects are found. Wheat yields would increase at
the European scale under most scenarios, but the gains would decrease with time for
regions in the north in the long-run. Results are less optimistic for corn yields. In the
short-run, some northern regions would experience gains in yields, but these gains
would transform into losses in the long-run. Those losses would even be higher in the
south of Europe.

Chapter 4, entitled “Climate Change and Business Cycles”, provides a general equi-
librium approach, unlike the first three chapters that are based on a partial equilibrium
framework. The aim of this last chapter is to investigate the influence of weather
shocks on business cycles, through an estimated dynamic model for a small open eco-
nomy. An original DSGE model with a weather dependent agricultural sector is de-
veloped and estimated using Bayesian methods and quarterly data for New Zealand
over the sample period 1989 to 2014.
18 General Introduction

The results from the model suggest that weather shocks play an important role in
explaining macroeconomic fluctuations over the sample period. A weather shock –
as measured by a drought index – acts as a negative supply shock characterized by
declining output and rising prices. In addition, the results show that farmers do not
anticipate weather shocks and are mostly surprised by variable climatic conditions.
Finally, increasing the variance of climate shocks in accordance with forthcoming cli-
mate change leads to a sizeable increase in the volatility of key macroeconomic vari-
ables, such as production and inflation.
 Part I

Climate Change in Developing

Countries: the Indian Case

19
 Part I. Climate Change in Developing Countries: the Indian Case 21

The first part of this thesis is devoted to the analysis of the effects of climate change
on developing economies. The focus is empirically made on an emerging economy,
namely India. India is located in Southern Asia and is the second most populated
country in the world after China, according to the United Nations, with an estimated
population of 1.33 billion in 2016. India is the 7th largest nation in the world with
a total area of 3.29 million square kilometres. According to the World Bank, around
60% of Indian’s land is devoted to agriculture. Agriculture is an important economic
sector of India, although its share in GDP has declined over years, falling from around
42% of GDP fifty years ago to 17% in 2015. Even if the value added of agriculture in
GDP is not as high as it used to be, almost one in every two workers is employed in
the agricultural sector. In addition, since the early sixties, food grain production in
India increased more than threefold – from 87 million tonnes in 1961 to 295 million
tonnes in 2014, as reported by the FAO. This increase in production was made possible
thanks to the Green Revolution that started in the early 1960s, with the introduction
of higher-yielding varieties of crops, improving water schemes, and increase use of
chemical fertilizers. This enabled India to achieve food self-sufficiency and even to
become a net food exporter.

The question of agricultural sustainability emerges as an important concern for India.

Food self-sufficiency may be challenged in the near future by the growing population
in the country. Furthermore, the increase in crop yields observed during the last dec-
ades might not be as spectacular over the coming years. In addition, climate change
places a real sword of Damocles over Indian food autonomy. The number of farms that
rely on precipitation as the only source of water for their agricultural production is still
substantial (around 40%), making production highly vulnerable to weather variations.
The increasing average temperature coupled with enhanced occurrence of droughts
that are expected due to climate change might offer inadequate climate conditions for
certain varieties of crops, therefore adding uncertainty on agricultural production.

The first part of this thesis provides two empirical analyses at the scale of rural house-
holds in India. The first chapter examines the relationship between farmers’ profits
and weather variations, and identifies how farmers may be impacted by climate change
depending on the relative magnitude of their profits and also depending on their geo-
graphical situation. The results of this first analysis are the starting point of the second
study. Climate conditions in some Indian regions are more conducive to agricultural
activities than others, and the projected climate change scenarios of the first study
highlight heterogeneous responses in farmer’s net revenues. It may be interesting to
22 Part I. Climate Change in Developing Countries: the Indian Case

enlarge the picture offered in chapter 1 by considering both the supply and the demand
side of agricultural production. From the supply side, the impacts of climate change
might create food shortages and threaten food sufficiency. From the demand side, the
consumption decisions might depend on climate, especially in the context of a devel-
oping country, in which a great number of rural households consume a substantial
share if not all of their production. Chapter 2 investigates these aspects.
Chapter 1

Climate Change and Profits: a

Ricardian Analysis

Joint work with Catherine Benjamin (University of Rennes 1

1 Introduction
The prospect of substantial climate change and its potentially huge impacts are central
concerns for the scientific community and policy makers. Research on climate impacts
has grown considerably since past years, and much has been learned regarding the
potential risk of damage associated with projected climatic change.

Climate change is expected to increase the variability of weather conditions and the
frequency of extreme weather events. Results from an International Panel on Climate
Change (IPCC) study predict increasing temperatures will lead to an increase in the
number, and severity, of extreme climates events like tornadoes and heavy rainfall in
some regions, and droughts in others (Edenhofer et al., 2014). Because food produc-
tion is fundamentally a biological process that depends in part, of temperature and
moisture, the agricultural sector’s potential vulnerability is particularly large.

There is an ongoing scientific debate over the magnitude of the effect of climate change
on overall agricultural sector. This debate partly stems from countervailing effects of
climate change, which makes possible the potential for increased production in the
short-run, and producers’ ability to adapt locally and globally through changing ag-
ricultural practices, shifting crop production, research and development and increase

23
24 Part I. Climate Change in Developing Countries: the Indian Case

trade. Some of the crucial inputs needed are the answers to the following questions:
How will climate change affect agricultural production? How will it change vari-
ability of yields? How, and under what circumstances, will climate change increase
and/or reduce production? In affected regions and countries, how difficult will it be
for producers to shift to different crops, to adopt new cropping patterns, and to adjust
production practices to the new environment?

The economic impacts of climate change on agriculture have been studied extensively
world over and in developing countries. Mixed effects are found by previous empirical
research.

Lobell et al. (2011) found that over the years 1980–2008, temperature trends were
higher than one standard deviation of historic variability in most countries. Tol (2009)
argued changes in weather patterns can have deleterious effects on agriculture, with
low-income countries being especially vulnerable to its effects. Deschênes and Green-
stone (2007) considered the effects of changes in temperature and precipitation on ag-
ricultural land rents, and conclude that climate change could lead to a slight increase
in U.S. corn and soybean profit, although Fisher et al. (2012) found higher potential
impacts of climate changes on US agriculture. Mendelsohn et al. (2006) found that
poor countries suffer much more from climate change than do rich countries. Further
analysis reveals that these adverse impacts stem from two factors. First and foremost
is the fact that the low income countries are disproportionately placed in the tropics,
where temperatures are already above the optimum for many crops. Thus, further
temperature increases bring large crop losses, whereas they bring gains in the higher
latitudes. Secondly, since agriculture is found as the most severely affected sector, the
relatively heavier reliance of these poor countries on farming results in larger losses
as a proportion of GDP.

A result shared by different studies is that the negative impacts of climate change are
more severely felt by poor people and poor countries. Developing countries, particu-
larly the least developed countries, look to be more vulnerable because of their high
dependence on natural resources, and their limited capacity to cope with climate vari-
ability and extremes climatic events.

This chapter tries to provide empirical estimate of the impact of climate change in
India. India is chosen for two major reasons. First agriculture is an important sector
of Indian economy. According to the World bank, this sector represented 17.4% of the
value added to GDP in 2014, and employed 49.7% of the workforce in 2013. This sector
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 25

is particularly vulnerable to present-day climate variability, including multiple years of

low and erratic rainfall. Scenarios generated by climatic models show that India could
experience warmer and wetter conditions as a result of climate change, particularly
if the summer monsoon becomes more intense (Mitra et al., 2002; Rupa Kumar et al.,
2002; McLean et al., 1998). About a quarter of India’s land is turning to desert and
degradation of agricultural areas (source, environment minister, 2014).

Secondly, India is one of the fastest growing economies of the world and is currently
the focus of a great deal of international attention. In 2015, the population of India
amounts to nearly 1.311 billion, which makes it the second most populous nation in
the world. The country has today emerged as a major player in the global agricul-
ture market. The leading forecasting institutions (OECD, FAPRI) expect that India
will play a bigger role in world markets in future. Weather fluctuations are potential
sources of threatening food security. In fact, “under normal weather conditions, do-
mestic output levels of rice and wheat in India are nearly sufficient to meet the domestic
demand for these grains. Due to weather fluctuations, the country faces a situation of
deficit in some years and surplus in some others and thus it is neither a regular importer
nor exporter” (Srinivasan and Jha, 2001). The unprecedented 2007–2008 world food
price crises, when agricultural prices have been highly volatile soured partly due to
decreased supply and increase demand is a clear example of world challenge. The
demographic and economic growth of emerging countries (for instance in India) and
adverse weather conditions were defined as some of the various factors which con-
tributed to this observed huge volatility.

This study uses the Ricardian approach to estimate the sensitivity of Indian farms to
climate conditions. This analysis, compared to previous ones on Indian agriculture,
uses cross-sectional individual data on farms to estimate the impact of climate and
household variables on net crop revenues.

In India, previous studies on Indian agriculture are mainly conducted on district-level

aggregate data (Kumar and Parikh, 1998, 2001; Sanghi and Mendelsohn, 2008). In case
of cross-sectional studies, they are focused on specific regions which may lack gener-
alization.

According to Mendelsohn (2008), the first attempt to assess the impact of climate on
net revenue in developing countries at the farm level concerns 11 countries of Africa
(Kurukulasuriya et al., 2006). Wang et al. (2009) also used farm-level data in China.
26 Part I. Climate Change in Developing Countries: the Indian Case

Furthermore, this analysis contributes to the existing knowledge in India by using

quantile regression. Most of the empirical studies use standard linear regression tech-
niques to summarize the average relationship between a set of regressors and the out-
come variable based on the conditional mean function. This approach provides only a
partial view of the relationship, as we might be interested in describing the relation-
ship at different points in the conditional distribution of the dependent variable (here,
net revenues per acre).

This first chapter is organized as follows. Section 2 encompasses discussions on the

specification and of the individual equations. Data are presented in section 3. Empir-
ical results are discussed in section 4. Section 5 provides climate scenarios exercises.
Some policy relevant observations emerging from this study are summarized in sec-
tion 6.

2 Modeling Net Revenue

This study is based on the Ricardian approach developed by Mendelsohn et al. (1994).
In this section, we will briefly present the Ricardian framework and then the quantile
regression methodology.

2.1 Behavioural Model: the Role of Climatic Variables

The Ricardian analysis investigates the link between farm value – or net revenue (see,
e.g., Deressa et al., 2009; Gbetibouo and Hassan, 2005; Haim et al., 2008; Kumar and
Parikh, 1998; Sanghi et al., 1998) – and climate, soils, demographic, and economic vari-
ables. The idea behind this framework is that farmers have adapted their behaviour
to face their environment.

The assumed objective of each farmer is to maximise his income subject to some con-
straints. In this sense, the farmer from exploitation i decides for each crop j ∈ J and
for each input k ∈ K the level of the input Xikj in order to maximise net revenue
(Dinar and Mendelsohn, 2011) :
∑
max πi = [pj Fj (Xikj ; Ei , Wi , Zi ) − wk Xikj ] , (1.1)
Xikj
j∈J,k∈K

where πi is the net revenue for farm i, pj and wk are vectors of output and input prices
respectively, that producers take as given. The production function associated with
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 27

each crop is denoted by Fj (·), and is a function of input choices Xikj given exogenous
environment conditions, i.e., economic control variables Ei , climate variables Wi and
farm specific variables Zi . The set of inputs that maximises net revenue is obtained by
differentiating eq. (1.1) with respect to each input, and gives the Ricardian function:

πi∗ = πi (Ei , Wi , Zi | pj ). (1.2)

Land value (Vi ) for farm i will equate the present value of the net revenue of each farm,
assuming perfect competition for land (Deschênes and Greenstone, 2007):
∫ ∞
∗ −rt
Vi = πj,t e dt, (1.3)
0

where r is the market interest rate. Then, exploring the relationship between net rev-
enue and climate enables to quantify the impact of the latter on the present value of
net revenue (Mendelsohn et al., 2001). However, in developing countries where land
market are imperfect, Ricardian analysis do not use land value as the dependent vari-
able. Instead, net revenues per unit of area are used (see, e.g., Gbetibouo and Hassan,
2005; Sanghi and Mendelsohn, 2008).

Empirically, eq. (1.3) is estimated with a linear model of the following form (see, e.g.,
Sanghi and Mendelsohn, 2008; Kumar and Parikh, 2001; Polsky, 2004):

V = α + β ⊤ E + γ ⊤ g(W ) + ζ ⊤ Z + ε, (1.4)

where V is the annual net revenue per unit of land, E is a set of economic variables,
W is a matrix of climatic variables, Z is a matrix of farm specific characteristics, ε is
a standard error vector, and α, β, γ and ζ are parameters that need to be estimated.
The function g(·) allows non-linear effects of climate on net revenues to be taken
into account. A common practice is to introduce linear and quadratic terms for each
climatic variable (see, e.g., Schlenker et al., 2006; Fleischer et al., 2008; De Salvo et al.,
2013; Deressa et al., 2009).

However the Ricardian analysis offers the advantage to implicitly assume that farm-
ers adapt to their environment, this methodology has some drawbacks. In the original
work of Mendelsohn et al. (1994), the question of irrigation is deliberately left aside.
This choice has been criticized in the literature (Cline, 1996; Schlenker et al., 2005).
In fact, irrigation is a choice that is sensitive to climate. To overcome this criticism,
some authors choose to introduce a variable related to irrigation. A dummy variable
28 Part I. Climate Change in Developing Countries: the Indian Case

indicating whether the farm is irrigated or not might be used (Kurukulasuriya and
Mendelsohn, 2008), or whether the farm has access to an irrigation technique (Des-
chenes and Kolstad, 2011). The percentage of land under irrigation might also be used
(Polsky and Easterling III, 2001; Gbetibouo and Hassan, 2005; Barnwal and Kotani,
2013). It is also possible to analyze the effects of climate conditions either on rainfed
farms or on irrigated farms, separately (Schlenker et al., 2006; De Salvo et al., 2013;
Van Passel et al., 2016). We first address the question of irrigation by introducing a
variable indicating if the farm is irrigated, and if so, which kind of irrigation technique
is used. We also compare the sensitivity of net revenues per acre to climate conditions
on either rainfed farms or irrigated farms.

The Ricardian model also suffers from its assumption of constant prices resulting from
the use of cross-sectional data. According to Cline (1996), this leads to a bias in the
analysis and thus to biased estimations of the global damages or benefits of global
warming. The same problems appear with the absence of consideration of the positive
impact of carbon dioxide fertilization highlighted by field experiments (Parry et al.,
2004; Aggarwal and Mall, 2002).

2.2 The Empirical Approach: the Use of Quantile Regres-

sion

The aim of this study is to estimate eq. (1.4) for India, using cross-sectional data. Most
studies consider the impact of climate on the conditional mean of net revenue given
some values of the independent variables, and assume a parametric distribution for the
error term. But there might be some asymmetry in these impacts across the quantiles
of net revenues. Quantile regression is one way to get to the bottom of these possible
asymmetries (Barnwal and Kotani, 2013). Also, as pointed-out by Van Passel et al.
(2016), quantile regression offers estimations that, compared to OLS, are more robust
to outliers.

Quantile regression was introduced by Koenker and Bassett Jr (1978). The τ th quantile
of a real-valued random variable V is given by:

QV (τ ) := F −1 (τ ) = inf{v : F (v) ≥ τ }, (1.5)

 Chapter 1. Climate Change and Profits: a Ricardian Analysis 29

where FV−1 (v) is the inverse of the distribution function FV (v) of V and τ can be any
value in the interval (0, 1).1

The τ th conditional quantile function is (Koenker, 2005):

QV (τ | X) = X ⊤ β τ , (1.6)

where X is a set of explanatory variables. An estimator of β τ can be obtained by

solving:

∑
n
argmin ρτ (vi − x⊤
i β), (1.7)
β∈Rp i=1

where ρτ (·) is a loss function defined as

ρτ (u) = u(τ − 1{u<0} ). (1.8)

The model writes:

V (τ | E, W , Z) = ατ + β ⊤ ⊤ ⊤
τ E + γ τ g(W ) + ζ τ Z + ετ , (1.9)
where QV (τ | E, W , Z) = ατ + β ⊤ ⊤ ⊤
τ E + γ τ g(W ) + ζ τ Z, (1.10)

where V is the annual net revenue, E is a set of economic variables, W is a matrix of

climatic variables, Z is a matrix of farm specific characteristics, ετ is a standard error
vector for a quantile τ , and ατ , β τ , γ τ and ζ τ are parameters that need to be estimated
for the set of quantiles. Function g(·) is a two-degree polynomial.

3 Data
To assess the impact of climate and other variables on net revenues across farmers in
India, this study relies on two data sources. Some descriptive statistics are provided in
table 1.2.
1
In our analysis, we choose to estimate the vector of coefficients β τ at 17 quantiles, from τ = 0.1
to τ = 0.9 by increments of 0.05. This arbitrary choice allows to draw smooth curves in section 4.
However, for clarity purposes, we only report result at three quantiles (τ = {0.25, 0.5, 0.75}) in the
tables and in some graphs.
30 Part I. Climate Change in Developing Countries: the Indian Case

3.1 Household Survey

The Indian Human Development Survey (IHDS) (Desai et al., 2005) is a nationally rep-
resentative survey conducted in India between 2004 and 2005. It was organized by
researchers from the University of Maryland and the National Council of Applied
Economic Research. Besides information about socio-economic conditions of the re-
spondents, the IHDS gives numerous details on agricultural activities of households,
such as farm structure, expenditures and revenues. Unfortunately, locations for each
household can only be traced at the district level.2

In our analysis, only households that report at least one worker in the farm and declare
that they cultivate crops are considered, the others are discarded. When information
about education is missing, data are removed as well. Individuals lying in mountain
regions of India3 are removed, since our estimations for climate variables are subject
to high variation in those areas. Finally, to have more consistent estimators, districts
with less than 20 observations are excluded. In the end, the analysis relies on 7, 751
individuals within 202 districts (see fig. 1.1).

No. Obs.
50 100 150 200

Figure 1.1: Number of Observations per District

Explanatory variables are divided in four groups: personal characteristics, farm char-
acteristics, location characteristics, and climatic factors.4
2
See Desai et al. (2010) for more details.
3
It concerns only three states in this data set: Arunachal Pradesh, Jammu and Kashmir, and
Meghalaya.
4
A description of each variable used in the analysis is provided in appendix A.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 31

3.2 Measure of the Climatic Variables

To investigate the impact of climate on agricultural net revenue, it is common practice

to use “climate normals”. “Climate normals” are the average of weather variables over
a prolonged period of time (see, e.g., De Salvo et al., 2013; Kumar and Parikh, 2001;
Polsky and Easterling III, 2001). As reminded by Schlenker et al. (2006), one needs to
distinguish between weather and climate: the weather refers to meteorological condi-
tions at a given moment while climate considers an average of climate statistics over
an extended period of time. With cross-sectional data, using “climate normals” offers
the advantage of implicitly taking into account the adaptation response of farmers to
local climate (Di Falco et al., 2011).

Unfortunately, there is no information about meteorological conditions in the sur-

vey data. So, this study relies on daily temperature and precipitation obtained from
112 weather station5 for the 30-year period from 1976 to 2005. Data are provided by
National Climatic Data Center (NCDC)/National Oceanic and Atmospheric Adminis-
tration (NOAA). As net revenues per acre are given at the district level, weather data
need to be aggregated at the same spatial level before considering computing “climate
normals”.

A four step procedure is followed.6 First, some missing data are estimated, using a
weighted average based on past and future observations. Then, a spatial interpolation
technique called thin plate splines (see, e.g., Di Falco et al., 2011; Boer, 2001; Hutchin-
son, 1995) is employed to obtain estimations of weather data at each point of a grid
covering the whole country. Once the estimation for each cell of the grid is performed,
an average by district can easily be computed. It is then possible to aggregate data by
season and district, in a fourth step. Four seasons are defined here: (i) Winter (Janu-
ary to March) (ii) Summer (April to June), (iii) Monsoon (July to September), and (iv)
Autumn (October to December).

3.3 Geographic and Soil Characteristics

To account for soil quality heterogeneity, we add a set of soil characteristics variables
from the Harmonized World Soil Database (Batjes et al., 2008). We rely on a different
set of variables, aggregated at the district level.
5
See appendix A for a spatial distribution of the weather stations.
6
The procedure is more detailed in appendix A.
32 Part I. Climate Change in Developing Countries: the Indian Case

Population density is also added at the district level to our database. The values are
obtained from the Census 2011, a national census survey conducted in all states of the
country.

It might be interesting to add information relative to the distance of the farm to the
nearest market center, to proxy transaction costs. Unfortunately, the exact location of
each farm is not provided in the database.

3.4 Descriptive Statistics

Table 1.2 summarizes descriptive statistics for the whole sample (first column) and
for two sub-samples based on net agricultural revenues per acre: the 25% observation
with lowest revenues per acre (second column) and 25% observations with highest
ones (third column).

The variable of interest is the annual agricultural net revenues per acre. As shown
in Table 1.2, the average value equals 5, 214.95 rupees per acre. The distribution of
these revenues is skewed on the right.7 Hence, most farms of the sample have modest
revenues, but some farms perform way better. If one focuses on the sub-sample of
the 25% lowest revenues on the one side, and on the sub-sample of the 25% highest
revenues on the other side, one can see quite a difference in the averages: −647.78
rupees per acre for the former and 15091.92 rupees per acre for the latter.

A positive effect of age and years of schooling is expected on net revenues. This effect
should be different for farms with low revenues and for farms with high revenues.
Indeed, head of households’ average age and years of schooling are greater for farms
with higher net revenues.

Most farms in the sample work on quite small surfaces, lower than 5 acres, though
a few individuals operate on large exploitation, which rises the mean value of the
sample. Moreover, farms with high revenues per unit of land also have higher cultiv-
ated surfaces than farms with low ones.

Farms with higher income per acre also tend to have more bullock carts, more tractors,
and more workers than farms with lower net revenues per acre. Farmers with higher
net revenues also tend to diversify their cultures. This use of crop mix is expected to
have a positive effect on agricultural revenues, as they enable a diversification of risk.
7
See in appendix A for a representation of the distribution of net revenues per acre.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 33

A large part (40%) of the sample does not use any irrigation system (table 1.1). For
farmers who benefit from irrigation source, there exists different irrigation schemes,
the most important of which is the use of tube wells. A positive effect of any irrigation
system is expected on net revenues, as opposed to the situation where the farm is not
irrigated.

Table 1.1: Irrigation Techniques in the Sample

Other Tank/ Private

None Tubewell Gov. Other
well Pond canal

No Obs. 3102 2577 892 455 377 222 126

Percentage (40.02%) (33.25%) (11.51%) (5.87%) (4.86%) (2.86%) (1.63%)

Finally, it is noteworthy that in the database, precipitation in summer and in autumn is

lower for farms with low revenues than farms with high revenues. However, summer
and autumn temperatures values are not significantly different in the two sub-samples,
according to a Welch Two Sample t-test.

Table 1.2: Summary Statistics for Data Set on Farm Households

Arithmetic Means and Standard Deviation

Whole sample Lowest 25% of Highest 25% of
net ag. revenue net ag. revenue
(n=11639) (n = 2910) (n = 2910)
Variable of interest
Net agricultural revenue (rupees per acre) 5214.95 -647.78 15091.92
(18810.32) (3936.73) (35456.49)
Climate variables (30-year average)
Summer precip. (mm) 24.69 21.89 30.78
(15.34) (12.31) (20.51)
Autumn precip. (mm) 12.4 12.62 14.23
(9.37) (10.25) (11.09)
o
Summer temp. ( C) 30.58 30.82 29.89
(1.79) (1.76) (1.8)
Autumn temp. (o C) 22.66 22.95 22.6
(2.13) (2.01) (2.48)
Personal characteristics
Age of head of household (years) 49.89 48.97 51.64
(13.39) (13.65) (13.48)
Literacy (years of schooling) 7.23 6.3 8.86
(4.72) (4.78) (4.44)
Farm characteristics
No. bullock carts 0.22 0.12 0.38
(0.42) (0.34) (0.5)
No. different cultures 2.67 2.07 3.42
(1.5) (1.23) (1.58)
No. workers in the farm 2.49 2.29 2.8
(1.27) (1.15) (1.49)
Continued on next page
34 Part I. Climate Change in Developing Countries: the Indian Case

Table 1.2 – continued from previous page

Arithmetic Means
(standard deviation)
Whole sample Lowest 25% of Highest 25% of
net ag. revenue net ag. revenue
(n=11639) (n = 2910) (n = 2910)
Area planted (acre) (acre) 6.24 3.56 12.94
(10.62) (7.15) (17.52)
Geographic and soil characteristics
Latitude (degree) 22.19 21.29 22.9
(5.4) (5.69) (5.63)
Longitude (degree) 78.96 78.24 79.2
(4.6) (4.01) (5.56)
Pop. density (hab/m2 ) 384.55 382.5 377.07
(293.01) (282.64) (311)
Gravel Content (%vol.) 8.52 8.55 8.25
(2.84) (2.71) (2.91)
Sand Fraction (%wt.) 39.55 39.35 37.92
(10.28) (9.93) (10.1)
Silt Fraction (%wt.) 30.33 30.03 31.11
(4.91) (4.74) (4.68)
Clay Fraction (%wt.) 29.66 30.17 30.62
(9.79) (9.41) (10.16)
pH (H2O) (−log(H + )) 6.83 6.83 6.98
(0.67) (0.57) (0.7)
Calcium Carbonate (%weight) 1.91 1.67 2.46
(2.12) (1.77) (2.46)
Sodicity (%) 2.47 2.28 2.94
(1.88) (1.67) (2.21)

Notes: observations are averaged on the whole sample (first column), on the sub-sample of the 25% lowest net agricultural revenue
per acre (second column), and on the sub-sample of the 25% highest net agricultural revenue per acre (third column). Standard
deviations are reported in brackets below the average value.

4 Results
The determinants of the variations of net agricultural profit per acre are estimated by
OLS (eq. 1.4) and at different quantiles using quantile regression (eq. 1.9). The results
are displayed in table 1.3. Each column reports both the estimates and standard errors
for a given quantile (0.25, 0.5, and 0.75), except for the last column which gives the
OLS estimates.8 Figures 1.2, 1.4 and 1.5 display the value of the coefficients of the
regression at a broader range of quantiles. Finally, fig. 1.3 offers a visualization of the
marginal effects of an increase in seasonal precipitation or temperature.

As we introduced quadratic and interaction terms for climate variables (as in De Salvo
et al., 2013; Kumar and Parikh, 2001; Sanghi and Mendelsohn, 2008), some simulation
8
Estimations were made at more quantiles, from 0.1 to 0.9 by increments of 0.05, but tables only
show estimates for a restricted set of quantiles, because of space limitations.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 35

were run to assess the impact of a unit change of these variables at different quantiles
of the conditional net revenue distribution, holding other variables constant. Predic-
tions computed for each individual using original data are compared to predictions
made with new data, where a given variable underwent an increase of one unit. Res-
ults are shown in boxplots for each quantile considered. The same idea is applied to
categorical variables, changing values of individuals with reference characteristic to
another one.

4.1 Climate Effects

This section looks first at the effects of rainfall on profits per acre before investigating
the effects of temperature. To avoid multicolinearity problems, monsoon and winter
climate variables were discarded from the empirical model. The correlation of these
variables are lower than seasonal precipitation and temperature observations with the
response variable.

Estimates of linear and quadratic terms of precipitation parameters vary across the
quantiles of the conditional distribution of net revenues for both summer and autumn
(fig. 1.2). Hence, the impact of a variation in total rainfall is different for farms with
small net annual revenues per acre (individuals at the lower tail of the net revenue con-
ditional distribution) than it is for farms with higher annual net revenues per income
(individuals at the upper tail).9 While the OLS coefficient for summer precipitation is
not statistically different from zero, quantile regressions tell a different story. The ef-
fects of summer precipitation on net profit per acre differ depending on the considered
quantile. At most quantiles, the quadratic term is not significant at the 5% level. The
relationship between autumn precipitation and net agricultural profit is more clear.
Precipitation in autumn have an inverted-U shape relationship with net revenues per
acre. That is, an increase in autumn precipitation increases profits up to a certain
threshold, above which the increase in precipitation leads to losses. The value of this
threshold varies according to the quantile, and is higher for farmers with smaller rev-
enues per acre, i.e., farmers at the lower tail of the distribution. In fact, the threshold
value is around 30mm for those farmers, while it drops to 21mm for farmers at the
upper tail of the distribution.

Looking at the overall effect of a one millimetre rise in 30-year average total rainfall
for summer and autumn (fig. 1.3) gives another insight. Farms are globally positively
9 Estimatesfor farms at the upper tail are less precise, as the confidence intervals become larger as
one moves up through the conditional distribution of net revenues per acre, implying a higher variance.
36 Part I. Climate Change in Developing Countries: the Indian Case

affected by an increase in summer and autumn total rainfall, and the magnitude of the
effect tends to be higher for farms with high net revenues per acre.

With OLS, a pure location effect in the distribution of net revenue is implicitly as-
sumed. If this assumption were true, farms with small net revenues per acre would be
affected in the same way as farms with high net revenues per acre (one would expect
the estimates at each quantile to be the same). But as shown in fig. 1.2, this assump-
tion is violated. The global effects of a one degree Celsius rise in 30-year average on
net revenue per acre, holding all other variables constant are plotted in fig. 1.3. In
summer, the relationship between “normal” temperatures and income per acre is such
that there exists a threshold above which a unit increase in temperature leads to losses
in profits. The value of this threshold is close to 31 degree C, implying that any tem-
perature increase above that threshold are detrimental to farmers. It is noteworthy
that the threshold value is almost identical at all quantiles, although a bit higher for
farms with higher net profits per acre. Hence, these farmers may be less affected than
farmers with low profits per acre by an increase in the average temperature. The same
inverted-U shape relationship is observed with autumn precipitation, but the value of
the threshold is around 25 degree C, and does not vary much across quantiles. Fig-
ure 1.3 shows that the average effect of an increase in summer temperature lowers
profits per acre, and the magnitude of the average loss is amplified as well as its vari-
ability as ones moves up through the distribution of net revenues per acre. The impact
of increasing temperatures in autumn mostly affects farms with high net revenues per
acre.

4.2 Other Variables

Besides climate variables, a few control variables are introduced in the model described
by eq. (1.9): personal characteristics of the household, farm characteristics, geographic,
and soil characteristics. They are briefly described in this section.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 37

Summer precip. Autumn precip. Summer temp. Autumn temp.

30000

400 0
1000 20000

200 -5000
10000
500

0 0 -10000

0
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Summer precip. sq. Autumn precip. sq. Summer temp. sq. Autumn temp. sq.
5.0 300
0
0 200
2.5
-10
0.0 -200 100

-2.5 0
-20
-400
-100
-5.0
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Notes: for summer and autumn, estimates for each quantile are represented by the solid black line ; 95% confidence intervals are
shown by grey bands ; dashed and dotted lines are OLS estimates and associated 95% confidence intervals bounds, respectively.

Figure 1.2: Estimated Climate Parameters by Quantile for Net Revenues per Acre

Summer precip. Autumn precip. Summer temp. Autumn temp.

7500
100 600
2500
5000
0 300
0
2500
-100 0 -2500
0
-300 -5000
-200
-2500
-7500
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Notes: The boxplots represent the variations of predicted yield at each quantile after a unit increase (one millimetre for precipit-
ation and one degree Celsius for temperature) above the 30-year average, all other variables kept constant. Average change for
each quantile is represented by a red rhombus.

Figure 1.3: Global Effects of a One Unit Rise in Climate Variables on Net Revenues
per Acre for each Quantile Estimated

4.2.1 Personal Characteristics

Considering the OLS estimation only, the variations of the age of the head of the house-
hold plays a positive but not significant role in the explanation of the variations of net
revenues. However, quantile regression coefficients show that up to the 70th percent-
ile, the older the head of the household, the lower net revenues per acre (fig. 1.4).
However, for farms with higher net revenues per acre, the sign of the effects changes
but becomes not significant at the 5% level. In addition to the age of the household,
we add a second personal characteristics variable, literacy, which is defined here as
the highest number of years of schooling within the members of the farm-household.
This variable is included as a proxy for knowledge of agricultural techniques. If there
38 Part I. Climate Change in Developing Countries: the Indian Case

Age of head of household Literacy No. bullock carts

150 0
20
100 -500

0 -1000
50
-1500
-20
0
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

No. different cultures No. workers in the farm

600
400
400
200
200

0
0

-200 -200
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Notes: estimates for each quantile are represented by the solid black line ; 95% confidence intervals are shown by grey bands ;
dashed and dotted lines are OLS estimates and associated 95% confidence intervals bounds, respectively.

Figure 1.4: Estimated Household Characteristics Parameters by Quantile for Net Rev-
enues per Acre

were a pure location shift effect, the mean positive effect obtained with OLS regres-
sion reported in table 1.3 would be the same at each quantile. However, results from
quantile regression show that net revenues per acre increase with years of schooling,
although the effect is not significant for farms with low net revenues per acre, i.e., for
farms at the lower tail of the distribution.

4.2.2 Farm Characteristics

Bullock carts are introduced as a proxy for technology (Sanghi and Mendelsohn, 2008).
The effect is found to be positive at each quantile, though only significant for a few
quantiles for farms with the lowest net revenues per acre. The number of different
cultures a farmer decides to grow is positive according to the least square estimate, but
not significant. For farms at the upper tail of the distribution, above the 80th percentile,
the number of different cultures has indeed no significant effect on net revenues per
acre. However, for farms with lower net revenues per acre, crop diversification is
actually increasing profits.

For individuals with low to medium net revenues per acre, the global effect of an
increase in workers per acre is positive though not significant at most quantiles. At
the upper tail of the conditional net revenues per acre distribution, this effect becomes
negative but remains not significant.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 39

4.2.3 Irrigation

As previously discussed, the choice of irrigation is sensitive to climate. One way of

addressing the question of irrigation in the Ricardian analysis is to introduce a variable
related to the choice of irrigation, leaving aside any possible underlying endogenous
bias.

Not surprisingly, irrigating one’s culture has a positive impact on net revenues per acre
(table 1.3). Results from quantile regression show that farms at almost each quantile of
the conditional net revenues per acre distribution actually realize benefits from using
any irrigation technique rather than only rely on rainfall (fig. 1.5). In addition, the
effect of irrigation on net revenues per acre grows as one moves up to the conditional
distribution of net revenues per acre.

Irrig. gov. Irrig. other Irrig. well

4000 4000

4000 3000
3000
2000
2000
2000
1000
1000
0
0
0
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Irrig. private canal Irrig. tank Irrig. tube well

20000
4000 4000
15000
3000 3000
10000 2000
2000
5000 1000
1000
0
0 0
0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9

Notes: estimates for each quantile are represented by the solid black line ; 95% confidence intervals are shown by grey bands ;
dashed and dotted lines are OLS estimates and associated 95% confidence intervals bounds, respectively.

Figure 1.5: Estimated Irigation Parameters by Quantile for Net Revenues per Acre

Instead of adding a variable linked to irrigation, some Ricardian studies choose to split
the sample of farms in two categories; one for rainfed farms, and a second for irrigated
ones. The first three columns of table 1.4 give the results of the estimation for rainfed
farms at quantiles τ = {0.25, 0.5, 0.75}. The last three columns give the results for
irrigated farms. The shape of the relationship between net revenues per acre remains
the same at each quantile both for rainfed and irrigated farms in all cases at the excep-
tion of the sensitivity to autumn temperatures for farms with the lowest net revenues
per acre. The marginal effect of an increase of either temperature or precipitation
from their 30-year average value is depicted in fig. 1.6. Irrigated and non-irrigated
40 Part I. Climate Change in Developing Countries: the Indian Case

farms tend to positively respond to an increase in autumn precipitation though the

magnitude of the effects on rainfed farms is less pronounced. An increase in summer
temperature has mitigated effects. Farms at the lower tail of the net revenue per acre
conditional distribution do not respond significantly to a unit increase in summer tem-
perature. At the upper tail, however, irrigated farms tend to be negatively impacted
by such an increase, while rainfed-only farms tend to be positively impacted by the
unit increase of summer temperature. This may be explained by the fact that irrigated
farms are already on locations where average temperature is higher than rainfed-only
farms, so that an increase in summer temperature lowers profits by either reducing
production or increasing production costs. The response of irrigated and rainfed-only
farms to an increase in autumn temperature exhibit a more contrasted difference. For
farms with low net revenues per acre, irrigated ones are negatively impacted while
rainfed-only are positively impacted. At the opposite side of the distribution of net
revenues per acre, we note that rainfed farms are not significantly sensible to autumn
temperature variations, while irrigated farms exhibit a strong negative response to the
increase in temperature.

Summer precip. Autumn precip. Summer temp. Autumn temp.

4000 1000
400 500
2000 0
200 0
0 -1000
0 -500
-2000
-2000
-200 -1000
0.25 0.5 0.75 0.25 0.5 0.75 0.25 0.5 0.75 0.25 0.5 0.75

irrigated rainfed

Notes: The boxplots represent the variations of predicted yield at each quantile after a unit increase (one millimetre for precip-
itation and one degree Celsius for temperature) above the 30-year average, all other variables kept constant, for irrigated farms
(in blue) and for rainfed farms (in yellow).

Figure 1.6: Global Effects of a One Unit Rise in Climate Variables on Net Revenues
per Acre for each Quantile Estimated for Rainfed and Irrigated Farms Separately

5 Climate Scenarios
To give an idea of the potential consequences of climate change on Indian farmers’
profits, we first envisage two climate scenarios and observe the changes in net reven-
ues per acre under the new climate conditions. More scenarios are then considered
and presented at the end of the section, by gradually altering precipitation and tem-
peratures.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 41

The scenarios exercises rely on the estimations made with all farms, irrigated or not.
Net revenues per acre under historical climate conditions are predicted and then com-
pared to predicted values under the new climate conditions given by the two scenarios.

To set up the scenarios, we follow Chaturvedi et al. (2012). The first one reflects a low
concentration of greenhouse gas (roughly corresponding to the representative con-
centration pathway (RCP) 2.6, adopted by The Intergovernmental Panel on Climate
Change for its fifth Assessment Reports in 2014), where average temperature for In-
dia is projected to globally increase by 1.7◦ C and total rainfall by 1.2%. It might be
viewed as a mitigation scenario. The second scenario reflects high concentration of
greenhouse gas (roughly corresponding to the RCP 8.5), mean temperature is projec-
ted to increase by 2.02◦ C and total rainfall by 2.4%. This scenario is more pessimistic
than the first. As the model does not take CO2 fertilization effects into consideration
nor does it account for possible technological changes, the results displayed by the
scenarios reported in table 1.5 and in figs. 1.7 and 1.8 might be biased, and should not
be viewed as a forecasting exercise.

Under both scenarios, net revenues are on average negatively impacted by the increase
in mean temperature and total rainfall. Farms with the lowest net revenues per acre
(at the 25th percentile of the distribution) experience a loss in net revenues per acre
that amounts to a median of −736 Rupees per acre under the optimistic scenario, in
which temperature increases by 1.7o C and precipitation by 1.2%. This corresponds to a
median percent change of −37.7%.10 Those losses grow higher at the upper part of the
distribution of net revenues per acre, and reach a district median of −983 Rupees per
acre. However, in terms of percent change, those losses are less important than those
at the lower tail of the distribution, with a value of −18.5%. Under the second scenario,
in which temperature and precipitation rise by 2.02o C and 2.4%, respectively, median
losses in net revenues per acre are even higher than those observed under the more
optimistic scenario. The median percent change in net revenues per acre decreases by
−46% for farms with the lowest net revenues per acre and by 22.7% for farms with the
highest net revenues per acre.

However, results given on a finer spatial resolution show some heterogeneity. As

depicted in figs. 1.7 and 1.8, some districts are positively impacted by the changes op-
erated on climate variables, while some other are getting worse. A distinction between
10
We focus on the median percent change as some farms have net revenues per acre that are com-
prised between −1 and 1, leading to inflated percent changes.
42 Part I. Climate Change in Developing Countries: the Indian Case

north and south can be made, with most districts in the south experiencing gains in
net revenues and districts in the north undergoing losses.

These findings are consistent with regional vulnerability to climate change found in
O’Brien et al. (2004). Kumar and Parikh (2001) also found losses for northern states,
but observed gains for eastern states. The few district for which observations are
available in East India exhibit gains in net revenues per acre, but the number of district
to represent East India is too small to generalize this result.

-1000 0 1000 -2000 -1000 0 1000 2000 -3000 -2000 -1000 0 1000 2000 3000

(a) τ = 0.25 (b) τ = 0.5 (c) τ = 0.75

Notes: For a each quantile (τ = {0.25, 0.5, 0.75}), the maps show the change in net revenues per acre under the first scenario,
i.e., following an increase in temperature of 1.7◦ C and a 1.2% increase in total precipitation.

Figure 1.7: Average Change in Net Revenues per Acre by Districts and Quantiles
Under Scenario 1 (Rupees per Acre)

-2000 -1000 0 1000 -2000 -1000 0 1000 2000 -2000 0 2000 4000

(a) τ = 0.25 (b) τ = 0.5 (c) τ = 0.75

Notes: For each quantile (τ = {0.25, 0.5, 0.75}), the maps show the change in net revenues per acre under the second scenario,
i.e., following an increase in temperature of 2.02◦ C and a 2.4% increase in total precipitation.

Figure 1.8: Average Changes in Net Revenues per Acre by Districts and Quantiles
Under Scenario 2 (Ruppees per Acre)

To have a better insight of the possible effects of climate change on net revenues per
acre of Indian farmers, we gradually increase 30-year average temperature from 0o C
to +3o C in steps of 0.2o C and we alter 30-year average total rainfall from −10% to
+10% in steps of 1 percentage point. Thus, we consider 336 different scenarios. In
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 43

each of them, the analysis previously done with the two climate scenarios is replic-
ated, i.e., predicting yields with historical climate conditions and comparing them to
predicted yields under new climate conditions. Figure 1.9 plots the median change in
net revenues at three different quantiles: τ = {0.25, 0.5, 0.75}, and fig. 1.10 plots the
same result expressed in percent deviation. It can be seen from the charts that temper-
ature variation generates more variation in net revenues per acre than precipitation.
Actually when precipitation increase, the losses generated by the variation in temper-
ature are mitigated. Furthermore, we note that the situation of farms with high net
revenues per acre is worsen more rapidly by an increasing temperature than that of
farms with low net revenues per acre. However, if one looks at the change occurred in
terms of percent deviations rather than in terms of magnitude, farms with the lowest
net revenues per acre suffer from higher losses due to the increase in temperature.

0.5
τ = 0.25 τ = 0.5 τ = 0.75

0.0

0.0 0.0 0.0

-0.5
Change

Change

Change

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.0

-2.0 -2.0 -2.0

3.0 3.0 3.0
2.5 10 2.5 10 2.5 10 -1.5
Te 2.0 1.5 5 Te 2.0 1.5 5 Te 2.0 1.5 5
mp n mp n mp n
era 1.0
0
tio era 1.0
0
tio era 1.0
0
tio
tur 0.5 -5
p i ta tur 0.5 -5
p i ta tur 0.5 -5
p i ta
e 0.0-10 ci e 0.0-10 c i e 0.0-10 ci
e e e
Pr Pr Pr -2.0

Notes: For each quantile (τ = {0.25, 0.5, 0.75}), the graphs show the median change in net revenues per acre following an
increase in temperature (in Degree C) and a variation in total precipitation (percent change).

Figure 1.9: Median Changes in Net Revenues per Acre by District at Selected
Quantiles (Thousand Rupees per Acre).

20
τ = 0.25 τ = 0.5 τ = 0.75

0
ange

ange

ange

0 0 0
-20
-20 -20 -20
Percent Ch

Percent Ch

Percent Ch

-40 -40 -40

-40
-60 -60 -60
-80 -80 -80
3.0 3.0 3.0 -60
2.5 10 2.5 10 2.5 10
Te 2.0 1.5 5 Te 2.0 1.5 5 Te 2.0 1.5 5
mp n mp n mp n
era 1.0
0
tio era 1.0
0
tio era 1.0
0
tio
tur 0.5 -5
p i ta tur 0.5 -5
p i ta tur 0.5 -5
p i ta -80
e 0.0-10 c i e 0.0-10 c i e 0.0-10 c i
e e e
Pr Pr Pr

Notes: For each quantile (τ = {0.25, 0.5, 0.75}), the graphs show the median percent change in net revenues per acre following
an increase in temperature (in Degree C) and a variation in total precipitation (percent change).

Figure 1.10: Median Percent Changes in Net Revenues per Acre by District at Selected
Quantiles (Thousand Rupees per Acre).
44 Part I. Climate Change in Developing Countries: the Indian Case

6 Concluding Remarks
This chapter presents an assessment of the effects of climate change on Indian agri-
culture. Agriculture represents a core part of the Indian economy and provides food
and livelihood activities to much of the Indian population (more than 60% of its popu-
lation is dependent on economic activities such as agriculture). India holds the second
largest agricultural land in the world. According to different forecasts, emergent cli-
mate phenomena seem to be aggravating the agrarian distress in India. An estimated
70% of the country’s arable land is prone to drought, 12% to floods, and 8% to cyc-
lones. Expected changes in climate, especially rainfall, are also marked by significant
regional variation, with the western and central parts witnessing a greater decrease in
rainfall days compared to the other regions of the country (Kumar, 2011). The Fifth
Assessment Report of the Intergovernmental Panel on Climate Change predicts that a
temperature rise would result in a significant drop in Indian agricultural yield.

In India, previous studies on agriculture are mainly conducted on district-level aggreg-

ate data. In case of cross-sectional data, studies focus on specific regions which may
lack generalization.

Our analysis, which is based on a representative sample of Indian farms, uses the
Ricardian approach to examine the impact of climate change on Indian agriculture
and describes farmers’ behaviour to varying environmental factors. The study uses
cross sectional data from the Indian Human Development Survey (IHDS). This survey
is a nationally representative survey conducted between 2004 and 2005 on 41, 554
households across India. The empirical method involves the specification of the net
revenues per acre as a function of climate variables and a set of economic variables.
Empirical results show that climate variables have a significant impact on the farmers’
net revenues per acre. Access to an irrigation scheme increases the net income per
acre, ceteris paribus. Crop diversification too has a positive impact.

In addition, the chapter discusses the impact of climate scenarios on farmers’ net rev-
enues per acre. In general, the results indicate that increasing temperature as well as
decreasing precipitation levels are damaging to Indian agriculture, both for small and
large farms. Increasing precipitation, on the other hand, is beneficial to Indian profits,
but when coupled with an increase in temperature, the negative effects of temperat-
ure dominate the positive effects of precipitation, leading to a global deterioration of
Indian profits.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 45

More research effort should be allocated into different points.

The empirical specification could be improved: although the analysis incorporates

some farm characteristics, the role of technology, or the change in regional prices for
the future, can be added. Another framework should therefore be considered, as the
Ricardian analysis does not allow for an integration of such variables.

The second point is linked to the irrigation. Our analysis indicates that irrigation
can and should play an important role in reducing the impacts of climate change on
farmers. As India is one of the most water stressed countries in the world, the irrigation
will be affected strongly by climate change (Edenhofer et al., 2014). We could enlarge
the different climate scenarios by implementing different assumptions on the use of
irrigation.

It would also be interesting to improve the analysis by studying the impact of climate
change on food security in India. While the magnitude of impact of climate change
looks to vary greatly by Indian region, climate change is expected to impact agricul-
tural productivity.

Hence climate change may have an impact on food availability and therefore on food
security. We can imagine that Indian agriculture cannot meet the objectives of food
self-sufficiency11 “Given that about 250 million Indians lack food security, the challenge
is to produce enough food ”sustainably” to meet the increasing demand, despite shrink-
ing resource availability” (Swain, 2014). The analysis could be improved by trying to
measure the impact on food supply at the country level.

The policy implications of our findings for climate change variability in Indian agri-
culture are important. These changes could affect water resource management, food
security, and trade policy. Policy-makers will need to address adaptive measures to
cope with changing agricultural patterns. Measures may include the introduction of
the use of alternative crops, and promotion of water conservation and irrigation tech-
niques to improve the access to water.

11
This aspect could have impact on world commodities markets, one factor which explains unanti-
cipated spike in international food prices in 2007-2008 was the growing demand of emerging countries
such as India.
46 Part I. Climate Change in Developing Countries: the Indian Case

Table 1.3: Regression Results for Agricultural Net Revenues per Acre

τ = 0.25 τ = 0.5 τ = 0.75 OLS

Intercept −79555.80∗∗∗ −94006.44∗∗∗ −139151.08∗∗∗ −11352.84
(11780.86) (15496.5) (40924.14) (103091.04)
Climate variables
Summer precip. 56.26 88.53∗∗ 134.29∗ 230.66
(35.15) (31.6) (62.4) (153.7)
Summer precip. sq. −0.09 0.02 0.23 −1.43
(0.44) (0.39) (0.76) (1.63)
Autumn precip. 164.90∗∗∗ 266.29∗∗∗ 314.50∗∗∗ 958.84∗∗∗
(32.02) (40.48) (62.14) (218.54)
Autumn precip. sq. −2.57∗∗∗ −4.64∗∗∗ −7.16∗∗∗ −16.78∗∗∗
(0.66) (0.74) (1.14) (3.35)
Summer temp. 8290.82∗∗∗ 10400.53∗∗∗ 15832.46∗∗∗ 7921.68
(1178.6) (1232.13) (2397.58) (7631.02)
Summer temp. sq. −133.31∗∗∗ −165.84∗∗∗ −250.89∗∗∗ −116.62
(18.65) (19.99) (38.69) (123.4)
Autumn temp. −3008.43∗∗∗ −4219.36∗∗∗ −5751.42∗∗ −4887.96
(757.94) (728.55) (1908.76) (4472.84)
Autumn temp. sq. 59.64∗∗∗ 86.31∗∗∗ 114.45∗∗ 94.41
(16.95) (16.68) (43.6) (101.4)
Personnal Characteristics
Age of head of household −3.30. −2.93 −1.26 4.69
(1.82) (2.49) (3.14) (15.81)
Literacy 11.52∗∗ 29.17∗∗∗ 48.75∗∗∗ 82.41.
(4.45) (7.15) (9.93) (46.23)
Farm characteristics
No. bullock carts 141.96∗ 143.58. 34.87 −828.53
(55.12) (83.27) (105.09) (535.61)
No. different cultures 137.61∗∗∗ 156.40∗∗∗ 141.87∗∗∗ 118.02
(17) (27.57) (34.67) (164.96)
No. workers in the farm 51.02∗∗ 28.05 −42.44 235.98
(18.12) (28.15) (37.08) (177.66)
Irrig. gov. 834.27∗∗∗ 1523.57∗∗∗ 2055.05∗∗∗ 2233.45∗
(88.33) (172.59) (125.03) (965.08)
Irrig. other 665.82. 1781.31∗∗∗ 2984.87∗∗∗ 2748.29∗
(367.51) (194.84) (880.65) (1299.56)
Irrig. well 214.08. 584.27∗∗∗ 1334.50∗∗∗ 1730.98∗
(114.03) (122.96) (196.01) (758.18)
Irrig. private canal 261.93 1307.39∗∗∗ 1422.10∗ 5017.65∗∗
(567.82) (298) (694.56) (1674.99)
Irrig. tank 47.82 599.21∗∗ 1279.54∗∗∗ 2401.81∗
(149.62) (220.3) (233.84) (1023.04)
Irrig. tube well 488.63∗∗∗ 1218.72∗∗∗ 1976.77∗∗∗ 3199.56∗∗∗
(93.21) (114.21) (147.64) (630.85)
Geographic and Soil Characteristics Variables
Latitude 53.19 48.96 −166.10. 108.40
(48.9) (56.94) (88.04) (368.6)
Longitude −176.55∗∗∗ −276.99∗∗∗ −425.67∗∗∗ −733.57∗∗
(26.02) (34.23) (54.26) (235.73)
Pop. density −0.18 −0.02 0.42 2.03
(0.17) (0.23) (0.46) (1.28)
Gravel Content 5.25 −19.76 −57.32∗ 86.52
(14.63) (19.95) (24.78) (122.92)
Sand Fraction −32.93∗∗ −30.10. −60.45 −192.89.
(10.85) (17.34) (51.46) (100.13)
Silt Fraction 52.27∗∗∗ 43.78∗ 11.35 −211.38.
(11.59) (20.74) (54.11) (126.31)
Clay Fraction −49.87∗∗∗ −62.16∗∗ −130.14∗ −208.25.
(13.79) (20.34) (52.97) (117.11)
pH (H2O) −2.44 324.42 570.97. 609.89
(128.43) (198.37) (313.95) (1181.85)
Calcium Carbonate −79.68∗∗∗ −142.96∗∗∗ −227.29∗∗∗ −192.67
(21.23) (37.38) (56.87) (239.66)
Sodicity 83.30∗∗ 142.63∗∗∗ 152.60∗∗∗ 236.96
(27.05) (33.06) (41.2) (191.75)
No Observations 7751

Notes: Dot (. ), asterisk (∗ ), double asterisk (∗∗ ), and triple asterisk (∗∗∗ ) denote variables significant at 10%, 5%, 1% and 0.1%,
respectively. Standard errors are in parentheses below the parameter estimates.
 Chapter 1. Climate Change and Profits: a Ricardian Analysis 47

Table 1.4: Regression Results with Only Rainfed or Irrigated Farms

Rainfed-Only Irrigated

τ = 0.25 τ = 0.5 τ = 0.75 τ = 0.25 τ = 0.5 τ = 0.75

Intercept −44046.09∗∗ −71852.06∗∗∗ −103000.17∗ −211074.32∗∗∗ −261875.32∗∗∗ −294728.72∗∗∗

(15988.14) (18464.39) (47097.59) (29302.4) (31884.44) (87114.37)
Climate variables
Summer precip. 120.97∗∗ 247.75∗∗∗ 471.22∗∗∗ 48.59 −28.99 −256.98∗
(43.38) (40.37) (80.5) (59.5) (59.47) (100.38)
Summer precip. sq. −1.23∗ −1.99∗∗∗ −3.46∗∗∗ 0.34 1.32. 4.65∗∗∗
(0.53) (0.46) (0.94) (0.68) (0.74) (1.36)
Autumn precip. 50.15 62.46 185.21. 275.89∗∗∗ 557.30∗∗∗ 807.15∗∗∗
(56.13) (39.5) (105.13) (68.09) (63.35) (105.88)
Autumn precip. sq. −0.33 −0.62 −3.83. −4.38∗∗∗ −8.16∗∗∗ −13.75∗∗∗
(1.18) (0.7) (1.98) (1.18) (1.08) (2.21)
Summer temp. 4837.10∗∗ 8010.25∗∗∗ 12163.52∗∗∗ 17061.88∗∗∗ 21953.13∗∗∗ 27674.19∗∗∗
(1544.32) (1663.46) (3595.83) (2513.09) (2512.56) (5331.02)
Summer temp. sq. −78.95∗∗ −127.25∗∗∗ −184.64∗∗ −272.97∗∗∗ −350.60∗∗∗ −447.97∗∗∗
(24.69) (26.52) (57.46) (40.37) (40.54) (85.36)
Autumn temp. −2594.36∗∗ −4356.22∗∗∗ −5887.76∗ −2343.16. −4147.26∗∗∗ −5878.41
(958.37) (932.72) (2337.52) (1417.33) (1138.76) (3757.81)
Autumn temp. sq. 58.33∗∗ 93.78∗∗∗ 122.06∗ 37.16 75.98∗∗ 100.64
(21.11) (20.59) (51) (32.89) (26.32) (86.55)

Personnal Characteristics

Age of head of HH −3.12 −2.90 6.26 −4.87 −2.68 −4.53

(2.46) (2.36) (5.1) (3.49) (3.42) (4.61)
Literacy 22.26∗∗ 35.10∗∗∗ 53.52∗∗∗ 0.20 22.14∗ 54.24∗∗∗
(7.11) (6.45) (15.08) (10.6) (10.07) (13.43)

Farm characteristics

No. bullock carts 27.22 116.09 270.73 146.38 13.45 −322.12∗∗

(103.19) (97.99) (249.58) (123.27) (122.61) (117.4)
No. different cultures 109.43∗∗∗ 111.84∗∗∗ 55.36 149.21∗∗∗ 189.84∗∗∗ 168.40∗∗∗
(29.39) (29.48) (55.09) (30.21) (32.34) (46.73)
No. workers in the farm 58.32. 103.67∗∗∗ 12.75 32.35 −20.53 −75.25
(29.91) (31.15) (63.01) (36.19) (37.41) (59.49)

Geographic and Soil Characteristics Variables

∗ ∗
Latitude 146.30 188.49 215.74. −62.01 −79.26 −559.51∗∗∗
(73.88) (75.05) (126.49) (100.79) (97.8) (150.65)
Longitude −69.07 −119.23∗∗ −412.93∗∗∗ −243.25∗∗∗ −447.77∗∗∗ −626.04∗∗∗
(42.46) (43.07) (92.92) (52.58) (47.02) (91.04)
Pop. density −0.15 −0.37 −1.33∗ −0.05 −0.07 0.76
(0.3) (0.4) (0.66) (0.33) (0.29) (0.6)
Gravel Content −11.76 −95.77∗∗∗ −152.30∗∗∗ 76.75∗ 35.75 −45.24
(22.02) (26.42) (37.76) (31.32) (33.39) (43)
Sand Fraction −10.53 −30.49. −117.49∗ −32.71 91.11∗∗ 138.60∗
(14.28) (17.74) (57.27) (30.29) (33.18) (58.92)
Silt Fraction 48.22∗ 81.61∗∗∗ 65.54 33.60 138.82∗∗∗ 159.05∗∗
(20.52) (23.74) (63.28) (34.79) (37.81) (61.44)
Clay Fraction −24.78 −76.24∗∗∗ −245.24∗∗∗ −40.39 74.55∗ 74.59
(18.18) (20.46) (64.23) (35.15) (35.86) (74.27)
pH (H2O) −172.72 213.22 1493.75∗∗ −400.78 −459.19 −84.63
(229.09) (216.25) (499.67) (350.93) (348.4) (645.43)
Calcium Carbonate −9.56 −181.83∗∗ −584.89∗∗∗ 23.95 7.16 −79.45
(51.65) (57.04) (112.69) (56.36) (60.56) (99.51)
Sodicity 50.20 83.65 87.57 73.98 183.63∗∗∗ 134.25∗
(52.76) (64.03) (118.83) (45.56) (41.68) (54.87)
No Observations 4643 4643 4643 3095 3095 3095

Notes: The estimations for rainfed-only farms at three different quantiles (τ = {0.25, 0.5, 0.75}) are provided in the first three
columns, and the estimations for irrigated-only farms on the last three columns. Dot (. ), asterisk (∗ ), double asterisk (∗∗ ), and
triple asterisk (∗∗∗ ) denote variables significant at 10%, 5%, 1% and 0.1%, respectively. Standard errors are in parentheses below
the parameter estimates.
48 Part I. Climate Change in Developing Countries: the Indian Case

Table 1.5: Changes in Net Revenues per Acre by Districts at Each Quantile (Rupees)

Median
Std
Scenario τ Min. 1st Qu. Median Mean 3rd Qu. Max. %
Dev.
Change
0.25 -1837 -1213 -736 -536 -39.3 1569 896 -37.7
Temp +1.7◦ C
0.50 -2029 -1234 -599 -364 332.4 2383 1161 -18.8
Precip. +1.2%
0.75 -3049 -1875 -983 -595 347.0 3404 1690 -18.5
0.25 -2220 -1471 -907 -667 -66.9 1852 1066 -46.0
Temp +2.02◦ C
0.50 -2447 -1487 -732 -454 379.7 2856 1382 -23.0
Precip. +2.4%
0.75 -3691 -2286 -1222 -760 373.8 4071 2010 -22.7

Notes: This tables reports changes in net revenues per acre under both climate scenarios at different quantiles (τ =
{0.25, 0.5, 0.75}), relative to predicted net revenues per acre under historical climate conditions.
Chapter 2

Climate Change and Food

Security: a Farm-Household
Model

Joint work with Catherine Benjamin (University of Rennes 1)

1 Introduction
A notable achievement of the climate change research community is that climate
change is now defined as a major topic on the political agenda. Thus, the last 2015
Paris international climate agreement marks the greatest collective effort the world
has ever seen to tackle the climate crisis.1 On that occasion, 175 parties (developed
and developing) have agreed to limit their emissions to relatively safe levels, of 2◦ C
with an aspiration of 1.5◦ C, with regular reviews to ensure these commitments can
be increased in line with scientific advice. In addition, financial aid will be provided to
poor nations to help them cut emissions and cope with the effects of extreme weather.
This agreement was possible thanks to the results of numerous studies which provide
compelling evidence of climate change. Work undertaken by an United Nations body,
the Intergovernmental Panel on Climate Change (IPCC), have produced key inform-
ation on the possible damages of climate change on biological and human systems.
Many elements have been identified in recent years (Edenhofer et al., 2014). Of the
1
[Link]

49
50 Part I. Climate Change in Developing Countries: the Indian Case

various aspects related to climate change, the possible increase in climate variability
has been recognized as one of the most critical issues.

Climate variability refers to the climatic parameters of a country or a region vary-

ing from its long-term mean. Recent IPPC report predicts that increasing frequencies
of heat stress, drought and flooding events are projected for the end of this century
and these events are to have many adverse effects over and above the impacts due to
changes in mean variables alone (Edenhofer et al., 2014). Changes in the frequency
and severity of extreme climate events and in the variability of weather patterns will
have significant consequences for human and natural systems.

Particular attention has been devoted to agricultural effects of climate change. Cli-
mate is the primary determinant of agricultural productivity; climatic variables (level
of precipitation, temperature) define key inputs in the production process. This has
motivated a number of economic assessments of the potential effects of long-term cli-
mate change on agriculture. This body of research addresses possible physical effects
of climate change on agriculture, such as changes in crop and livestock yields (Chen
et al., 2004; McCarl et al., 2008; Schlenker and Roberts, 2009), as well as the economic
consequences of these potential yield changes for instance the effects on farm income
(Sanghi and Mendelsohn, 2008), potential changes in patterns of food production and
prices (Miao et al., 2015).

Climate change affects food production and agriculture and in many ways.

First, it has a direct effect on the performance on agriculture through the impact on
crop yields. This is particularly true for rural farmers in developing countries where
agricultural production is highly dependent on rainfall and sensitive to the weather.
Land use, and changes in agricultural productivity linked to climatic variability have
been studied widely in developing countries (Khanal et al., 2014).

The general consensus of these studies is that changes in temperature and precipitation
will result in changes in land and water regimes that will subsequently affect agricul-
tural productivity. Research has also shown that specifically in tropical regions, with
many of the poorest countries, impacts on agricultural productivity are expected to be
particularly harmful (Mendelsohn, 2008). In addition, experts predict tropical regions
will see both a reduction in agricultural yields and a rise in poverty levels.

Secondly, climate change has indirect effects by affecting growth and distribution of
incomes, demand for agricultural products and thus affects different dimensions of
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 51

food security (Schmidhuber and Tubiello, 2007). Changes in climatic conditions have
already affected the production of some staple crops in some very poor countries, and
future predicted climate change threatens to exacerbate this. Given that agriculture
is highly sensitive to climate patterns, changes in temperature and rainfall can reduce
agricultural output. For the most vulnerable people, lower agricultural output would
also mean lower income. Mendelsohn et al. (2007) also showed that historically, cli-
mate is highly correlated to agricultural incomes. Under climate change, increased
temperatures and erratic precipitation are likely to result in higher rural poverty, and
lower rural income for food. Under these conditions, the poorest people – who already
use most of their income on food – would have to sacrifice additional income to meet
their nutritional requirements. Furthermore, the magnitude of the projected impacts
of climate change suggests that it is reasonable to expect that changes in production
will be large enough to drive an increase in agricultural prices in some regions. At
last, commodity price volatility is likely to intensify and become more widespread due
to climate change.

A consensus shared by different studies is that the negative impacts of climate change
are more severely felt by poor people and poor countries. Developing countries, par-
ticularly the least developed ones, look to be more vulnerable because of their high
dependence on natural resources, and their limited capacity to cope with climate vari-
ability and extreme climatic events. Furthermore, although declining, the value added
of agriculture in these countries’ GDP remains high, and and a large part of the pop-
ulation is still directly or indirectly dependent on agriculture.

Among the most significant impacts of climate change is the potential increase of food
insecurity and malnutrition. Numerous studies show the possibility of increased food
insecurity from climate change (see, e.g., Butt et al., 2005; Deaton, 1997).

There is still an extensive debate on how climate change will affect world food con-
sumption (Howard et al., 2014).

This chapter contributes to the literature in two ways. First the effects of climate
change on both production and consumption decisions are investigated in a large
sample of Indian rural households. In developing countries, farm households are of-
ten both producers and consumers of food. The impact of a price increase in any main
agricultural product goes beyond the usual income and substitution effects of basic mi-
croeconomics. A cursory look at the model of the agricultural household shows that
price changes of basic food items has winners and losers (Singh et al., 1986). In case
52 Part I. Climate Change in Developing Countries: the Indian Case

of a price increase, the winners are those farmers who are net-sellers of the crop and
who have enough resources in terms of land, labour and other inputs to benefit from
the price increase. The losers are the net-buyers of the crop. The latter are land-poor
farmers.

Secondly most studies have focused on the food security issue at the national level
which may mask food insecurity at the household level. For a better understanding of
farm household food security status, it is preferable to use methods and tools working
at micro-level, which allows to provide detailed results on a farm household scale and
to capture heterogeneity across households.

The remainder of this second chapter is organized as follows. Section 2 describes the
conceptual framework to motivate the effects of climate change both on production
and consumption behaviour. Section 3 presents the empirical strategy retained in the
analysis. Section 4 introduces the data. Section 5 focuses on estimation results. Finally,
section 6 concludes.

2 A Farm Household Modelling

The mixture of the economics of the firm and of the household is the main feature of
most agricultural families in developing countries and provides the starting point for
our analysis.

Most people in developing countries often consume at least a portion of the output
of their productive activities. As noted by De Janvry et al. (1991) farm household in
underdeveloped countries are usually located in an environment characterized by a
number of market failures for some of its products (e.g., some foods, particularly the
most perishable or bulky) and for some of its factors (e.g., child labour or family labour
with low access to the labour market). An extreme case of market failure is simply the
non-existence of a market, for example, due to cost transactions (missing means of
transportation) and so households do not in participate in markets. In that particular
case implicit food prices appear which means that food prices are endogenous and are
dependent on household preferences. The household faces wide price bands, implicit
prices levels are between the low price at which the household could sell a commodity
and the high price at which household could buy that product. Faced with this wide
price band, the household may be better off choosing self-sufficiency in that good if
its subjective price (defined as the price which equates its supply and demand) falls
inside the band.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 53

At the same time, the sale of crops produced on the family farm is an important in-
come source for the farm household. Consequently, individuals make simultaneous
decisions about production (the level of output, the demand for factors, and the choice
of technology) and consumption (labour supply and commodity demand).

Building on traditional farm household models (see Singh et al., 1986), we consider a
farm household whose objective is to maximise utility, subject to different positivity
constraints.

The household’s problem can be represented as follows:

max U(C, CL ; Z H ) (2.1)

C,CL ,LH ,LF ,H O
( ( ) )
s.t. p · q LF + LH ; Z A , W − C − wLH + wO LO ≥ 0 (2.2)
CL + LF + LO = T, (2.3)

where U(·) defines the household’s utility function, C is total household own con-
sumption, CL is leisure, Z H defines a vector household characteristics (number of
dependants, gender and age of the head of the household, …), p is the output price, q is
the quantity produced, LF is the amount of on-farm family labour, LH is the amount
of on-farm hired labour, w the cost of hired labour, wO off farm wage, Z A are the
vector of quasi fixed factors of production (e.g., cultivated surface), W a vector of exo-
genous variables which could affect farm production (such as climate variables), LO is
the amount of off-farm family labour, and T is total time available.

Following Henning and Henningsen (2007), we consider four different labour regimes.
The first regime concerns households in which members provide work outside the
family plot and hire non-family labour to work on the family plot. The second regime
applies to households that provide work outside the family plot and do not hire any
labour. The third regime includes households that do not provide work outside the
family plot and hire non-family labour to work on their farm. The fourth and last
regime is for households that neither work off-farm nor hire non-family labour.

3 Empirical Strategy
This section first presents the empirical methodology used to estimate the production
function of the farm-households. Details on the estimation of the shadow wage of
54 Part I. Climate Change in Developing Countries: the Indian Case

family work on the farm are then provided. These shadow wages are then used in the
consumption decision model, as a valuation of the opportunity cost of labour.

3.1 The Production Function

The form of the agricultural production function used in this analysis is a standard
Cobb-Douglas:

∑
J ∑
C ∑
K
ln YA,n = α0 + α1,j Xj + α2,c Wc + α3,k Zk + εYa ,n , ∀n = 1, . . . , N,
j=1 c=1 k=1

(2.4)

where YA,n is the total agricultural output value of the nth household, Xj ; j ∈ {1, . . . , J}
indicates the j th input (hours spent on farm by household members, hours spent on
farm by hired people, cultivated area), Wc ; c ∈ {1, . . . , C} refers to the cth climate
variable (temperature and rainfall, both expressed as their 30-year average value) that
include squared terms to account for non-linear effects of climate on production, and
Zk ; k ∈ {1, . . . , K} is the k th control variable. Control variables include farm-household
characteristics (mean age and gender of the head of the household), location-specific
variables (distance of the farm-household from the nearest town, distance from the
nearest public distribution system (PDS), distance from the nearest Pucca road,2 per-
centage of households with electricity within the village, state in which the household
is located, and soil characteristics). The parameters α1 , α2 , and α3 , as well as the con-
stant α0 , are parameters to be estimated, and εYa is a standard error term with zero
mean and variance σY2a .

The production function is first estimated by ordinary least square (OLS). Then, to ac-
count for the endogeneity of household labour on farm, we proceed to an instrumental
variable (IV) estimation, using household specific variables (number of working-age
members, number of children) and market prices of most important goods (pure mar-
ket goods, price of crops and price of animal products) as instruments. We finally es-
timate the production function in a two-step procedure to correct for selectivity-bias
that occurs because agricultural output is observed only if the household participates
in the agricultural market, either by providing family labour, employing labour, or
both. Thus, a two-step Heckman procedure is used (Heckman, 1979). The first step
2
There are two types of roads in India, Kutcha and Pucca roads, the latter being of better quality
than the former.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 55

consists in estimating the factors affecting the decision on whether or not to participate
in the agricultural market, by means of a probit model. The identification variables
of this first step are the number of dependants, i.e., children and teenager younger
than 15 or seniors older than 60, and number of working-age household members, i.e.,
between 15 and 60. The second step consists in estimating the production function
for the selected individuals that participate in the agricultural market, using an IV
approach, with the same instruments mentioned previously.

3.2 The Shadow Wage

Farm household’s working age members may choose to work on or off the family plot.
If they decide to work off the family farm, it is possible to observe the wage they receive
from their activities. However, when they work on the household farm, some assump-
tions need to be made to estimate the value of their work. As stated by Castagnini et al.
(2004), if one assumes perfectly competitive markets, household members are indiffer-
ent between working on or off the family plot, at the market equilibrium. However, the
underlying assumption of separability may be questioned in case of market failures,
especially in developing countries. The labour supply choices of each individual in a
family might not be separable from the labour needs of the household farm (Skoufias,
1994). In such cases, the common methodology is to estimate a shadow price of labour
(see e.g., Jacoby, 1993; Sicular and Zhao, 2004). This shadow price PL is obtained by
assuming that household farm work is remunerated at its marginal productivity, and
can thus be calculated as:

α̂1,LF YA
PL = , (2.5)
LF

where α̂1,LF is the estimated coefficient associated with family hours of work on farm
from eq. (2.4), YA is agricultural output and LF is the amount of family hours on the
household plot.

3.3 The Consumption Decisions

We follow Henning and Henningsen (2007) to model the consumption decisions of

farm households, using an Almost Ideal Demand System (AIDS) (Deaton and Muell-
bauer, 1980) framework. Each household can consume four different goods: pure-
market goods m, cereals c that can be either self-produced or bought on the market,
56 Part I. Climate Change in Developing Countries: the Indian Case

animals or animal-derived products a, and leisure L. Pure-market goods are an ag-

gregation of sugar, kerosene and oil; cereals are composed of rice, wheat, pulses, gur
and other cereals; animal goods are composed of milk, meat, chicken, fish, and eggs.
The prices associated with pure-market goods, cereals and animal-derived products
are weighted means of observed prices of each commodity used to define each of the
three categories of goods, where the weights are defined as the share of each commod-
ity in the annual consumption of the household. The price of leisure is the shadow
wage of eq. (2.5), i.e., the opportunity cost of not working. The number of hours ded-
icated to leisure is obtained by first computing total available time for the household.
We assume that family members aged between 15 and 60 has 10 hours a day available
for work, while members older than 60 can only devote a maximum of 5 hours a day to
work. Family members then allocate their available time between work (on-farm and
off-farm) and leisure. Hence, the amount of leisure time is calculated by deducting the
number of declared hours worked to the total available time.

The demand functions of the AIDS are the expenditure shares of each of the four goods
i ∈ {m, c, a, L} at the household level:

∑ Yn ∑
win = βi + γij ln Pjn + δi ln + ζic Wcn + εwi ,n , (2.6)
Pn c
j∈{m,c,a,L}
∑
s.t. βi = 1 (adding-up), (2.7)
i∈{m,c,a,L}
∑
δi = 0 (adding-up), (2.8)
i∈{m,c,a,L}
∑
γij = 0 (homogeneity in prices), (2.9)
j∈{m,c,a,L}

γij = γji , ∀i, j ∈ {m, c, a, L} (symmetry), (2.10)

∑ 1 ∑ ∑
where ln Pn = β0 + βi ln Pin + γij ln Pin ln Pjn ,
2
i∈{m,c,a,L} i∈{m,c,a,L} j∈{m,c,a,L}

(2.11)

where win = Pin Cin

Yn
;i ∈ {m, c, a, L} represents the expenditure share of the ith type
th
of good of the n household, Pin ; i ∈ {m, c, a, L} is the consumer prices, Yn the full
income, Pn is the translog consumer price index, and Wc represents the cth shifter
variable (i.e., climate). The parameters βi , γij , δi , and ζic are unknown parameters.
They are estimated using an iterative linear least squares method by the R add-on
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 57

package “micEconAids” (Henningsen, 2014).3

The consumption decisions are examined for all rural households that produce cer-
eals, and also for the different subsets based on the participation in the labour market
described in section 2. We calculate price and income elasticities for each subset.

The income elasticity measures how much the demand for a good is affected by the
changes in income. It is calculated as:

∂qi Y
ei = , (2.12)
∂Y qi

where ei is the income elasticity of good i, qi the demanded quantity of that good, and
Y represents total income.

The price-elasticity of demand measures the percent change in quantity demanded for
a good i with respect to an increase in the price of a good j:

∂qi pj
ϵij = , ∀i, j ∈ {m, c, a, L}, (2.13)
∂pj qi

where pj is the price of good j.

The eq. (2.6) can be viewed as a Marshallian demand function in budget shares. It is
possible to derive the Marshallian price elasticities of good i with respect to good j as
follows:
[ ( )]
1 Y
ϵM = −1i=j + γij − δi wj − δj ln , ∀i, j ∈ {m, c, a, L}, (2.14)
ij
wi P

where 1i=j is the indicator function that takes the value 1 when i = j and 0 otherwise.

In the presence of shifter variables, which is the case in our analysis, the coefficients
δi must be adjusted as follows:

∑
m
δi∗ = δi + ζij Wj , ∀i ∈ {m, c, a, L}. (2.15)
j=1

3
The leisure demand equation is dropped from the system to avoid singularity problems. The para-
meters βi , γij and δi for this equation are obtained using the adding-up properties.
58 Part I. Climate Change in Developing Countries: the Indian Case

From these Marshallian elasticities, it is possible to derive the compensated elasticities,

or Hicksian elasticities through the Slutsky equation:

ij = ϵij + wj × ei ,
ϵH ∀i, j ∈ {m, c, a, L}. (2.16)
M

4 Data
This study uses a nationally representative multi-topic survey of households across
India, the Indian Human Development Survey 2 (Desai and Vanneman, 2016). It was
conducted between 2011 and 2012 in 42, 152 households, and provides rich informa-
tion regarding consumption expenditures. Locations for each household can be traced
at the district level and is not provided at a finer geographical level for anonymity pur-
poses.

We only focus on rural households. The final sample covers a wide geographical sur-
face, as depicted in fig. 2.1, where the number of observation is represented. In total,
the final sample used in this study is composed of 22, 892 households from 258 dis-
tricts in 23 states. The minimum number of observation within each state ranges from
12 (Dadra and Nagar Haveli) to 2, 501 (Uttar Pradesh).

Number of observation by district

100 200 300

Note: The map shows the geographical distribution of the 22, 892 rural households from the national representative survey, at
the district level.

Figure 2.1: Number of Observations per District

As we are interested in studying the production and consumption decisions on farm-

households, we divide the sample into two categories of rural households: agricultural
and non agricultural. The former is composed of 12, 206 households while the latter
contains 10, 686 households that produce crops. Households that produce crops are
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 59

further divided into the four labour regimes previously described in section 3.1. As
shown in fig. 2.2, all categories cover a similar and large surface of India.

Number of observation by district

50 100 150
Supply and Demand Supply Only Demand Only Autarkic
(N = 3685) (N = 4835) (N = 2556) (N = 1621)

Notes: The maps show the geographical distribution of farm-households from the national representative survey, apportioned
according to the different labour regimes: “Supply and Demand”, for households that both engage in off-farm work and hire
labour on their family plot; “Supply only”, for households that engage in off-farm work and do not hire any labour on their
family plot; “Demand only”, for households that not engage in off-farm work and hire labour on their family plot; and “Autarkic”,
for households that neither engage in off-farm work nor hire labour on their family plot.

Figure 2.2: Number of Observations per District and Labour Regime

The remainder of this section describes in more details the composition of the sample
data. The descriptive statistics of the different variables are reported in table 2.1, where
households are regrouped according to their labour regime.

4.1 Labour, Income, and Farm Characteristics

Working-age members of the family can allocate their time between work, either in
or off the farm plot, and leisure. On average, farm-households have an endowment of
12, 872.54 hours, 1, 663.33 of which (13%) devoted to on-farm work, 1, 397.23 (11%)
to off-farm work, and 9, 811.98 (76%) to leisure.

The average net annual per capita income of farm-households is 22, 599 Rupees. How-
ever, it can vary by a factor of almost two according to the labour regime. In fact,
households that do not hire labour and also engage in off-farm work (Supply Only)
have an average net per capita income of 16, 0168 Rupees only, while autarkic and
households that hire labour to work on their farm and do not engage in off-farm work
(Demand Only) receive twice as much a year. There is also some regional heterogen-
eity, as depicted in fig. 2.3. The map shows that the average net per capita incomes
vary from 10, 735 Rupees (Odisha) to 62, 998 Rupees (Punjab), while the map on the
right shows that agricultural per capita net income varies from −613 Rupees (Daman
60 Part I. Climate Change in Developing Countries: the Indian Case

Table 2.1: Descriptive Statistics According to the Different Labour Regimes

Supply
Supply Demand
Not Ag. All Ag. and Autarkic
Only Only
Demand

Household Characteristics
No. Persons (Persons) 4.39 5.43 5.38 5.55 5.28 5.39
No. Children (0-14) (Persons) 1.38 1.62 1.52 1.78 1.48 1.60
No. Teens (15-20) (Persons) 0.51 0.67 0.63 0.74 0.58 0.71
No. Adults (21-60) (Persons) 2.04 2.54 2.66 2.53 2.46 2.40
No. Seniors (60+) (Persons) 0.47 0.59 0.57 0.49 0.76 0.68
No. Working Age (15-60) (Persons) 2.55 3.21 3.29 3.27 3.04 3.10
No. Dependents (<15 U >60) (Persons) 1.85 2.22 2.09 2.27 2.24 2.28
Mean Age in the HH (Years) 39.43 39.34 39.28 37.72 41.94 40.31
Labour
Total Available Time (Hours) 10,203.45 12,872.54 13,121.14 12,946.72 12,527.17 12,644.47
On-farm HH Labour (Hours) 0.00 1,663.33 1,588.68 1,407.24 2,074.00 1,952.57
Off-farm HH Labour (Hours) 2,335.48 1,397.23 2,079.74 2,094.45 0.00 0.00
On-farm Hired Labour (Hours) 9.53 322.72 454.45 0.00 969.47 0.00
On-farm Labour (Hours) 9.53 1,986.05 2,043.12 1,407.24 3,043.47 1,952.57
Leisure (Hours) 7,867.96 9,811.98 9,452.73 9,445.03 10,453.17 10,691.90
Income
Net Income (INR) 80,766.84 112,558.72 111,821.78 82,871.03 148,613.43 146,666.35
Net Income per Cap. (INR /Cap.) 19,897.11 22,599.24 22,270.44 16,168.41 30,936.11 29,545.71
Net Ag. Income (INR) 4,278.03 56,620.08 43,290.97 26,583.78 103,829.30 101,919.13
Net Ag. Income per Cap. (INR /Cap.) 1,172.25 11,278.13 8,528.62 5,161.15 21,368.14 19,849.54
Farm Characteristics
Crops Output (INR) - 4,568.68 4,020.32 2,564.06 7,771.33 6,773.32
Cultivated Area (Acre) - 4.98 4.47 2.76 8.83 6.72
Crops Kept (%) - 87.75 83.89 90.73 85.72 90.44
Dist. to Nearest Town (km) - 14.07 14.06 14.96 13.20 12.76
Dist to Nearest PDS (km) 0.32 0.58 0.46 0.74 0.47 0.56
Weather

Rainfall (30-year average, mm) 2.89 2.70 2.94 2.61 2.71 2.41
Temp. (30-year average, Deg. C) 25.61 25.29 25.86 24.92 25.57 24.72
Soil Characteristics

Gravel Content (%vol.) 8.53 8.63 8.65 8.90 8.29 8.35

Sand Fraction (%wt.) 40.19 40.84 40.06 42.23 39.26 40.84
Silt Fraction (%wt.) 31.24 30.68 30.11 30.42 31.26 31.84
Clay Fraction (%wt.) 28.16 27.96 29.43 26.66 29.22 26.73
No. obs. 10,686.00 12,206.00 3,452.00 4,715.00 2,445.00 1,594.00
% of Agricultural Households - 100% 13.1% 20% 28.3% 38.6%

Notes: Households of the sample are either “Not Ag.” if they do not engage in the agricultural sector or “All Ag.” if they do. All
agricultural households are furthermore divided into four categories according to their participation in the job market. These
four categories are as follows: “Supply and Demand”, if they both engage in off-farm work and hire labour on their family plot;
“Supply only”, if they engage in off-farm work and do not hire any labour on their family plot; “Demand only”, if they do not
engage in off-farm work and hire labour on their family plot; and “Autarkic”, if they neither engage in off-farm work nor hire
labour on their family plot.

and Diu) to 45, 800 Rupees (Punjab). Both maps exhibit a distinction between West
and East India, with lower incomes in east India than in the West part of the country.

Turning to the cropping activities of the farm-households, we also observe heterogen-

eity. First, we note that households specializing in farming activities (Demand only)
exhibit the highest value for crops output, with an average of 7, 771 Rupees. This
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 61

Per Capita Net Income

0 20,000 40,000 60,000

Total Net Income Agricultural Net Income

Note: The maps show household per capita average annual net income at the district level for total income (left) and agricultural
income (right).

Figure 2.3: Average Per capita Income by State of Farm-households)

measure includes both the amount of sold by-product and the value of the share of the
production kept for self-consumption. The value of crop production is much lower
for households that also work outside the family plot (Supply and Demand), with an
average of 4, 020 Rupees and 2, 564 Rupees for households engaging in off-farm work
and not hiring people to work on their family plot (Supply Only). However, no such
distinction can be observed regarding the share of produced crops that are kept for self-
consumption. On average, 87% of the by-product is kept by families. When looking
at the distribution of the percentage of kept production, we can distinguish between
three types of behaviour (fig. 2.4): the vast majority of households keep the complete
production (83.7%), a tiny share of households sell it (8.2%), and the remainder (8.1%)
keep some of the production and sell the rest.

Supply and Demand Supply Only Demand Only Autarkic

4,000

3,000

2,000

1,000

0
0% 25% 50% 75% 100% 0% 25% 50% 75% 100% 0% 25% 50% 75% 100% 0% 25% 50% 75% 100%

Notes: Agricultural households are divided into four labour regimes: “Supply and Demand”, if they both engage in off-farm work
and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on their family
plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if they neither
engage in off-farm work nor hire labour on their family plot.

Figure 2.4: Distribution of Percentage of Crops Kept for Self-Consumption

62 Part I. Climate Change in Developing Countries: the Indian Case

4.2 Climate

Climate conditions directly affect the agricultural production function. Consumption

decisions may also be altered by varying climate conditions. We therefore incorporate
climate data in our analysis using weather daily data from the National Climatic Data
Center (NCDC) / National Oceanic and Atmospheric Administration (NOAA). We use
two climate variables in the analysis: total rainfall and average temperature. Both are
expressed as “climate normals”, i.e., as the 30-year observed average value.

Our aim is to define climate variables both at the same geographical and temporal unit
as that for households, i.e., at the district level, for a given year. Hence, in a first step,
we interpolate daily weather station data at the district level by means of thin-plate
splines (see, e.g., Di Falco et al., 2011; Boer, 2001; Hutchinson, 1995).

In a second step, daily observations are averaged on a monthly basis.

The monthly averages are used in the third step to calculate the monthly “climate
normals”. We use a 30-year period (1980–2009) to compute monthly averages of both
rainfall and temperature, at the district level. Figure 2.5 displays these (yearly aggreg-
ated) “normals”.

Precipitation (mm) Temperatures (o C)

6 9 12 15 24 26 28 30 32
Rainfall Temperature

Note: The maps depict the district-level “climate normals”, i.e., 30-year averages, for precipitation (left panel) and temperatures
(right panel), for India.

Figure 2.5: Climate Normals (1980–2008)

Climate variations are expected to influence agricultural production, some climate

conditions being more conducive to agricultural activities than other. The correlation
between crops output (in Indian Rupees) and rainfall or temperature (30-year averages)
are reported in table 2.2. As shown in the table, crops output is negatively correlated
with temperatures, although the values are close to zero. Correlation with rainfall, on
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 63

the other hand, exhibit higher even if limited values. In addition, disparities can be
observed among the different labour regimes. Crops output in autarkic families seem
to be more sensitive to rainfall variations than the rest of agricultural households. The
same patterns are also observed in the correlation between net agricultural income and
“climate normals”.
Table 2.2: Correlations of Crops Output and Agricultural Income with Climate Vari-
ables

Crops Output Net Ag. Income

Rainfall Temperatures Rainfall Temperatures

Supply and Demand -0.16 -0.04 -0.10 -0.08

Supply Only -0.18 -0.03 -0.16 -0.11
Demand Only -0.14 -0.04 -0.11 -0.04
Autarkic -0.24 -0.04 -0.20 -0.04

Notes: The statistics provided in the table concern farm-househols only. Agricultural households are divided into four categories
according to their participation in the job market. These four categories are: “Supply and Demand”, if they both engage in off-
farm work and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on
their family plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if
they neither engage in off-farm work nor hire labour on their family plot.

4.3 Household Characteristics

The average Indian farm-household is composed of 5.4 family members, which is one
more than the average for households with no agricultural activity (table 2.1). The
decomposition by age and labour regime is depicted in fig. 2.6. The shape of the distri-
bution of the number of children, teens, adults and seniors looks the same in all labour
regimes. However, the skewness of families classified as “Supply Only”, where at least
one member engages in off-farm work and where no labour off-family is used to work
on the farm, is higher than that of other type of households.

The average age in Indian rural household is 39.4 years old, for both agricultural and
non agricultural households. It is noteworthy that this average is quite different for
families classified as “Supply Only” and “Demand Only”. In fact, in the former case,
where households engage in work outside the farm plot and do not hire non-family
labour, the average age is lower, with a value of 37.76 years old. This is explained by
the higher average number of children and teens in those households, compared to
the others, that drives the average age down. On the other hand, in “Demand Only”
households, i.e., in households that do not engage in off-farm work and hire labour
on their family plot, the average age is higher. The number of children and teens also
helps to explain why, as it is lower in those households. In addition, the number of
64 Part I. Climate Change in Developing Countries: the Indian Case

No. Children No. Teens No. Adults No. Seniors

No. Persons
(0-14) (15-20) (21-60) (60+)

6,000

Not Ag.
4,000

2,000

Supply and Demand

2,000

1,500

1,000

500

0
3,000

Supply Only
2,000

1,000

1,500

Demand Only
1,000

500

750

Autarkic
500

250

0
0 10 20 30 0 5 10 15 0 2 4 6 0 5 10 15 0 2 4

Notes: The vertical grey dashed line represents the sample mean for each category. Agricultural households are divided into four
labour regimes: “Supply and Demand”, if they both engage in off-farm work and hire labour on their family plot; “Supply only”,
if they engage in off-farm work and do not hire any labour on their family plot; “Demand only”, if they do not engage in off-farm
work and hire labour on their family plot; and “Autarkic”, if they neither engage in off-farm work nor hire labour on their family
plot.

Figure 2.6: Distribution of Number of Family Members

elderly is higher in those households: 0.76 on average, compared to 0.59 in all farm-
households.

4.4 Other characteristics

The texture or the proportions of different-sized mineral particles contained in the

soil alter the production of cereals. It is thus important to include measures of the
quality of the environment that farmers face to grow crops. Soil quality heterogeneity
is addressed using a worldwide 21, 600 × 43, 200 raster database of soil characteristics
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 65

from the Harmonized World Soil Database (Batjes et al., 2008). From the grid data,
we compute district averages of five variables relative to the texture of topsoil: gravel
content by volume, sand, silt ,clay fraction, and organic carbon as a percentage of
weight.

5 Estimation Results
This section first presents the estimation results of the agricultural production function
and the resulting shadow wage, then turns to the analysis of the consumption decisions
of farm-households, and finally presents the different demand elasticities.

5.1 Production

The results of the estimation of the agricultural production function represented by

eq. (2.4) are presented in table 2.3.

The agricultural production is estimated first by OLS. Then, the endogeneity problem
caused by labour decisions is taken into account using instrumental variables. The
sample bias linked to the decision households must face on whether or not to take
part in agricultural activities is addressed using a two-step Heckman procedure. The
determining factors of this participation are reported in table 2.4.

Overall, the estimates are mostly similar in sign across all three methods investigated,
but the magnitudes may change drastically once endogeneity and sample bias prob-
lems are considered.

In more details, we first observe that variable inputs (hours worked either by family
members or by employees) and quasi-fixed inputs (cultivated area) have, as expected, a
positive impact on the agricultural production. The magnitude of these effects differs
depending on the estimation method. Once the model accounts for the endogeneity
of the number of hours worked either by family members or by hired labour force as
well as the sample section bias, the magnitude of the effect of the number of hours
worked substantially rises. In fact, the production elasticity of family labour goes
from 0.077 to 0.268, while the elasticity of hired labour rises from 0.048 to 0.176. For
cultivated surface, the estimate obtained using the 2-steps Heckman procedure reveals
an elasticity of 0.322.
66 Part I. Climate Change in Developing Countries: the Indian Case

Household characteristics also play a significant role in the variations of the agricul-
tural production function. First, the age of the head of the household, which is com-
monly used as a proxy for the skills of the head of the family, has a significant effect
on production. Besides, the older the head of the household, the higher the probab-
ility of the household to engage in farming activities, up to a threshold of 61 years
old, above which any additional year decreases this probability. The second house-
hold characteristic that affects the production function as well as the participation in
the agricultural market is the gender of the head of the household. The OLS estimate
informs that having a man as the head of the household instead of woman increases
the agricultural production. However, once the endogeneity problem caused by the
decision of the level of labour force needed in the farm plot is addressed, and once the
sample selection biased is accounted for, the story is quite different. We first notice
that households whose head is a woman tend to engage less in agricultural activities.
For households in the selected sample, i.e., those engaged in farming activities, having
a woman as the head of the household tend to increase the agricultural output.

Regarding location characteristics, the estimation results show a negative relationship

between the output and the distance from the village to the nearest PDS shop or to
the nearest Pucca road. The percentage of households with electricity in the village
positively affects the agricultural production, but in the mean time, it reduces the prob-
ability of engaging in agricultural activities. Soil quality variables included to take soil
heterogeneity into account also exhibit significant effects.

Finally, climate variables affect the agricultural production of Indian farm-households.

The relationship between agricultural production and rainfall measured at their 30-
year average has a U-shape form, while the relationship with temperatures also meas-
ured at their 30-year average displays an inverted U-shape relationship. Hence, a one
millimetre increase in precipitation average level leads to a decrease in agricultural
outputs up to a certain threshold, of 2.80 mm (within the range of observed values),
above which the increase in precipitation has a positive effect on production. For tem-
peratures, a one degree increase in the 30-year average value becomes harmful to the
agricultural production for initial temperatures above the threshold of 17.23 Degree
C, which is within the range of observed values.

The coefficient associated with the number of hours worked by the family members
on the farm plot is used to calculate the shadow wage as explained in eq. (2.5). This
shadow wage is then used in the estimation of the demand system.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 67

Table 2.3: Production Function Estimation Results

OLS IV Probit Selection

Coef t-value Coef t-value Coef t-value

Intercept 1.379 (0.862) 2.634 (0.966) 2.522 (1.360)

Log of HH Hours on Farm 0.077∗∗∗ (8.533) 0.601∗∗∗ (3.666) 0.268∗∗∗ (27.286)
Log of Hired Hours on Farm 0.048∗∗∗ (14.479) 0.160. (1.710) 0.176∗∗∗ (49.900)
Log of Cultivated Area 0.507∗∗∗ (58.433) 0.241. (1.854) 0.322∗∗∗ (34.463)
Age of the Head of HH 0.006 (1.435) −0.012. (−1.879) −0.025∗∗∗ (−4.349)
Age Sq. of the Head of HH ×104 −0.422 (−1.161) 1.022. (1.860) 2.013∗∗∗ (3.954)
Gender (Male) 0.071∗∗ (2.653) −0.041 (−0.991) −0.107∗∗ (−2.906)
Log of Dist. to Nearest PDS −0.002 (−0.160) −0.007 (−0.418) −0.032. (−1.804)
Log of Dist. to Pucca Road −0.001 (−0.355) −0.010∗ (−2.045) −0.013∗∗ (−3.058)
Pct. of HH with Electricity 0.005∗∗∗ (12.550) 0.004∗∗∗ (6.883) 0.005∗∗∗ (11.034)
Gravel Content 0.007. (1.759) 0.011∗ (2.370) 0.005 (1.151)
Sand Fraction 0.016∗∗∗ (11.395) 0.024∗∗∗ (6.606) 0.020∗∗∗ (12.354)
Silt Fraction 0.020∗∗∗ (6.056) 0.017∗∗∗ (4.339) 0.017∗∗∗ (4.585)
Base Saturation 0.007∗∗∗ (4.685) 0.019∗∗∗ (4.671) 0.012∗∗∗ (6.881)
Rainfall −0.600∗∗∗ (−4.123) −0.981∗∗∗ (−4.948) −0.891∗∗∗ (−5.479)
Rainfall Squared 0.089∗∗∗ (3.674) 0.189∗∗∗ (4.532) 0.159∗∗∗ (5.848)
Temp. 0.439∗∗ (3.269) −0.005 (−0.025) 0.323∗ (2.144)
Temp. Squared −0.011∗∗∗ (−3.986) −0.003 (−0.633) −0.009∗∗ (−2.934)
Adj. R-Squared 0.47 0.29

Notes: N = 12, 206. Significance levels are denoted by a dot (· ) at the 10% level, one asterisk (∗ ) at the 5% level, two asterisks
(∗∗ ) at the 1% level and three asterisks (∗∗∗ ) at the .1% level. HH denotes Household. Hours on farm are instrumented using the
following variables: No. Female Children (0-14), No. Male Children (0-14), No. Female Seniors (60+), No. Male Seniors (60+),
Price of Pure-market Goods. The identification variables used in the first stage of the two-step Heckman estimation are: No.
Children (0-5), No. Children (6-14), No. Working Age (15-60). Hours on farm in the second step are instrumented as in the IV
[Link] coefficients of state dummies are not reported in the table for clarity purposes. They are however available upon
request.

5.2 Consumption

Using the shadow price estimated using the production function results, we are able to
examine the budget shares of Indian agricultural households. We analyse how house-
holds allocate the amount of their full income among four different goods: pure market
goods, crops that can be home-grown, animal products, and leisure, as as previously
stated in section 3.3. The share of each alternative in the full income as well as the
associated unit price are reported in table 2.5. On average, leisure represents 23.81%
of full income, crops represents 34.78%, animals 24.68% and pure market goods come
last with a share of 16.73% of full income. The share of crops consumption varies with
the type of household. On average, households classified as “Supply Only” devote a
larger share of their full income to crops consumption (37.97%) than the other type of
households and in the mean time, a lower share of leisure (20.74%). On the contrary,
households classified as “Demand Only”, allocate a higher share of their full income to
68 Part I. Climate Change in Developing Countries: the Indian Case

Table 2.4: Participation in Agricultural Labour Market (Probit Selection Model)

Estimate (t-value)

Coef t-value

Intercept −15.603∗∗∗ (−7.906)

No. Children (0-5) 0.014 (0.198)
No. Children (6-14) 0.421∗∗∗ (8.416)
No. Working Age (15-60) 0.148∗∗∗ (23.376)
Age of the Head of HH 0.067∗∗∗ (14.858)
Age Sq. of the Head of HH ×104 −5.222∗∗∗ (−12.210)
Gender (Male) 0.487∗∗∗ (18.435)
Log of Dist. to Nearest PDS 0.142∗∗∗ (7.271)
Log of Dist. to Pucca Road 0.046∗∗∗ (8.295)
Pct. of HH with Electricity −0.003∗∗∗ (−6.421)
Gravel Content 0.012∗∗ (2.737)
Sand Fraction 0.003 (1.541)
Silt Fraction 0.007. (1.863)
Base Saturation −0.001 (−0.819)
Rainfall 0.597∗∗∗ (3.929)
Rainfall Squared −0.107∗∗∗ (−4.358)
Temp. 0.828∗∗∗ (5.089)
Temp. Squared −0.015∗∗∗ (−4.353)
Notes: N = 10, 686. Significance levels are denoted by a dot (· ) at the 10% level, one asterisk (∗ ) at the 5% level, two asterisks
(∗∗ ) at the 1% level and three asterisks (∗∗∗ ) at the .1% level. HH denotes Household. The coefficients of state dummies are not
reported in the table for clarity purposes. They are however available upon request.

leisure (28.68%) than the rest of the households, and a lower share to crops consump-
tion (29.93%). This may be explained by the higher value of the shadow wage for
“Demand Only” households: the opportunity cost of not working is higher for these
families.

These shares and prices are used to estimate the Ideal Demand System described by
eq. (2.6). The results are reported in table 2.6.

5.3 Elasticities

The elasticities computed from the AIDS estimates are reported in table 2.7.

The demand for crops is almost inelastic. It is estimated at 0.176 for all rural house-
holds. Among the different types of households, “Demand Only”, for which the share
of crops in total expenditures is the highest, are those for which this demand elasticity
is the lowest (0.139). Animal-derived products are identified as luxury goods, since
the income elasticity for these goods is estimated at 1.3 for all rural households. The
elasticity of demand for these animal products is low for all types of rural households
(0.384). Households classified as “Supply and Demand” respond less to price variation
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 69

Table 2.5: Consumption Choices According to the Different Labour Regimes

Supply and Demand

All Ag. Supply Only Autarkic
Demand Only

Consumption Expenditures

Market Goods Exp. (INR/Cap.) 6,417.86 6,135.79 5,665.59 7,579.50 7,467.17

Crops Exp. (INR/Cap.) 13,501.60 13,821.63 12,460.19 14,807.33 13,883.87
Animals Exp. (INR/Cap.) 12,149.35 10,949.53 9,939.82 15,454.34 16,195.18
Leisure Exp. (INR/Cap.) 16,283.35 15,109.80 9,529.90 28,101.60 20,653.06
Consumption Shares

Pure-market Goods (%) 17.27 17.22 18.08 16.11 16.75

Crops (%) 35.91 36.91 39.09 31.08 31.75
Animals (%) 25.50 24.78 24.41 26.85 28.16
Leisure (%) 21.33 21.09 18.42 25.95 23.34
Unit Price

Price of Pure-market Goods (INR per unit) 63.01 62.91 63.43 62.39 62.95
Price of Crops (INR per unit) 29.38 29.91 28.22 30.39 30.13
Price of Animal Products (INR per unit) 46.10 52.95 41.38 48.56 41.50
Shadow Wage (INR per Hour) 2.51 2.53 1.98 3.42 2.65

No. obs. 12,216.00 3,454.00 4,715.00 2,445.00 1,602.00

% of Agricultural Households 100% 28.3% 38.6% 20% 13.1%

Notes: The statistics provided in the table concern farm-househols only. Agricultural households are divided into four categories
according to their participation in the job market. These four categories are: “Supply and Demand”, if they both engage in off-
farm work and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on
their family plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if
they neither engage in off-farm work nor hire labour on their family plot.

of these animal-derived products than the other types of households. Higher crops
prices lead to a weak increase in the demand for animal-derived products, and a relat-
ively higher increase in the demand for pure market goods, for all types of households.
On the contrary, higher crops prices lead to a decrease in the demand for leisure. The
case of leisure is puzzling. According to the classical theory, if leisure is a normal good,
a relative increase in its price should lead to a relative decrease in its demand. Recall
that the price of leisure is the shadow price of labour and that farmers allocate their
available time either on work or on leisure, so that the elasticity of leisure demand can
be interpreted as the elasticity of labour supply. The classical theory indicates that
this elasticity should be positive: when the wages rise, the labour supply should rise
as well. Our results suggest the opposite. The elasticity of leisure is positive for all
types of farms, implying a negative labour supply elasticity. Households classified as
“Demand Only”, and “Autarkic”, i.e., the two types of households that do not engage
in off-farm work have relatively more inelastic labour supply. Different explanations
regarding the negative sign of these supply elasticities emerge in the literature. Dess-
ing (2002) argued that at low wages, leisure can be considered as a luxury good and
that the income effect dominates, thus leading to a negative labour supply elasticity.
70 Part I. Climate Change in Developing Countries: the Indian Case

Berg (1961) argued that these negative labour supply elasticities could be explained by
the fact that once the minimum level of subsistence income is reached, poor house-
holds reduce their work. In India, Dasgupta and Goldar (2006) found a negative labour
supply elasticity for women from households below the poverty line.

5.4 Scenarios

To get a better perspective of the potential effects of either price changes or climate
changes on the consumption decisions of Indian rural households, we test different
scenarios.

The strategy we use consists in comparing the predicted consumption quantities of

each of the four goods (pure-market, crops, animals and leisure) obtained using the
estimates of the AIDS model, to the predicted quantities calculated once either prices
or climate values have been modified. The predictions are calculated for all agricul-
tural households as well as for each subset based on the households’ labour regime.
This method gives an insight of the link between prices and consumption, and on
climate and consumption, but should not be viewed as a forecasting exercise.

We first present some price scenarios and then turn to climate scenarios.

5.4.1 Changes in Prices

First, we gradually increase the price of each good, separately, and observe the impacts
on the consumption of each good. This gives a more visual representation of the
elasticities reported in table 2.7. The results are plotted in fig. 2.7. Each column of the
panel of graphs indicates how the consumption of the corresponding item is altered
by a gradual increase in the price of the good.

5.4.2 Changes in Climate

Different climate scenarios are envisaged. First, we alter precipitation statistics only,
by considering variations from −30% to +30% of the amount of total rainfall exper-
ienced on average during the last 30 years. Then, we isolate the potential effect of
temperature on consumption patterns, by assessing an increase up to 5o C. Finally, we
assess the combined effect of a change in both precipitation and temperature.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 71

Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.

Pure-market Goods
0

Price of
-5

Price of
Crops
-4

-8

Animal Products
0

Price of
-5

-10

Shadow Wage
4

-4

-8
0.0% 2.5% 5.0% 7.5% 10.0%0.0% 2.5% 5.0% 7.5% 10.0%0.0% 2.5% 5.0% 7.5% 10.0%0.0% 2.5% 5.0% 7.5% 10.0%

All Ag. Supply and Demand Supply Only Demand Only Autarkik

Notes: The graphs show the effects of an increase in each commodity prices (separately) on the share of consumption of each
good, in each different labour regimes. The x-axis gives the temperature increase in degree, while the y-axis gives the median
percent change in the consumption share. Agricultural households are divided into four labour regimes: “Supply and Demand”,
if they both engage in off-farm work and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do
not hire any labour on their family plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family
plot; and “Autarkic”, if they neither engage in off-farm work nor hire labour on their family plot.

Figure 2.7: Effects of Increasing Prices on Consumption Shares (Median Percent

Changes)

The effects of varying levels of rainfall on household decisions of consumption are

plotted in fig. 2.8. Each panel corresponds to the evolution of the share of consump-
tion of one of the four goods with respect to the modification of the level of precip-
itation. The solid red line corresponds to the median percent change in the share of
consumption,4 while the four other lines depict the change in these shares according
to the labour regime of the households. Overall, households show a similar response
to precipitation changes., i.e., an increase in the share of pure-market goods and an-
imal products and a decrease in crops consumption and in leisure time following a
decrease in precipitation, and opposed effects following an increase in precipitation.
4
The median percent change is used rather than the average percent change as some small changes
starting from low shares inflate the value of the average change.
72 Part I. Climate Change in Developing Countries: the Indian Case

Having said that, a distinction between two types of households can be drawn. The
magnitude of the response of households to an increase or a decrease in total rain-
fall clearly differs between households that both supply and demand labour, and aut-
arkic households. The latter type of households is less affected by varying climate
conditions in regards with the consumption of non-agricultural goods, but shows a
more substantial response in regards with the consumption of both crops and animal
products.

Market Goods Exp. Crops Exp.

20

-20

-20% 0% 20% -20% 0% 20%

Animals Exp. Leisure Exp.

20

-20

-20% 0% 20% -20% 0% 20%

All Ag. Supply and Demand Supply Only Demand Only Autarkik

Notes: The graphs show the effects of a variation in precipitation levels on consumption of each good, in each different labour
regimes. The x-axis gives the percent change in precipitation levels, while the y-axis gives the median percent change in con-
sumption. Agricultural households are divided into four labour regimes: “Supply and Demand”, if they both engage in off-farm
work and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on their
family plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if they
neither engage in off-farm work nor hire labour on their family plot.

Figure 2.8: Effects of Precipiation Variations on Consumption (Median Percent

Changes)

In the same way as for the precipitation changes, fig. 2.8 shows the response in the
consumption decisions of Indian agricultural households to a change in temperature.
Overall, the consumption of pure-market goods is not affected much by the increase.
However, once breaking down households according to their participation in the la-
bour market, we observe that the share of pure-market goods consumed by households
that both supply and demand labour force is negatively affected by the increase in tem-
perature while the other types of households exhibit an opposed response. Concerning
crops consumption, we observe that autarkic households seem to be unaffected by an
increase in 30-year average temperature. On the contrary, households both providing
and demanding labour as well as households that demand only labour are gradually
decreasing their share of crops consumption as temperature increases.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 73

Market Goods Exp. Crops Exp.

80

40

-40

-80
0 1 2 3 4 5 0 1 2 3 4 5

Animals Exp. Leisure Exp.

80

40

-40

-80
0 1 2 3 4 5 0 1 2 3 4 5

All Ag. Supply and Demand Supply Only Demand Only Autarkik

Notes: The graphs show the effects of an increase in temperature levels on the consumption of each good, in each different labour
regimes. The x-axis gives the temperature increase in degree, while the y-axis gives the median percent change in consumption.
Agricultural households are divided into four labour regimes: “Supply and Demand”, if they both engage in off-farm work and
hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on their family plot;
“Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if they neither engage
in off-farm work nor hire labour on their family plot.

Figure 2.9: Effects of Increasing Temperatures on Consumption (Median Percent

Changes)

Figure 2.10 offers another insight on the relationship between consumption shares
and climate. It shows how the different types of households, based on their labour
regime, respond to a modification of both precipitation and temperature. Each line of
the matrix of plots corresponds to a type of household, and each column indicates the
variation of the share of consumption of one of the four commodities (pure market
goods, crops, animal-derived products, and leisure). There is a lot of heterogeneity
in the results. Some patterns regarding agricultural products consumption seem to
emerge. When pooling all households, we note a trade-off between the consumption
of crops and animal products. If the level of rainfall decreases, the share of crops
consumption decreases as well, while, in the mean time, the share of animal products
rises. When households are divided according to their labour regime, the same trade-
off is observed for each regime.

6 Concluding remarks
The prospect of a warmer global climate with increased climate variability and chan-
ging precipitation patterns across the world defines new challenges to research on
production and consumption behaviour in developing countries. In fact, there is a
74 Part I. Climate Change in Developing Countries: the Indian Case

broad consensus that the poorest share of the world’s population is most vulnerable
to these predicted changes. The way on how climate change and climate variability
affects households in developing countries depends not only on direct impacts on pro-
ductivity of agriculture (effect on crop yields), but on indirect impacts on consumption
behaviour.

This study provides a cross-sectional analysis of the demand of Indian households

farms, relying on the household modelling to investigate the current impact of climate
variables on consumption behaviour. We first examine the response of agricultural
production to climate variations, using a large data set of agricultural households in
India. From this estimation, we calculate the shadow price of labour which can then be
fed into an Almost Ideal Demand System aiming at analysing the consumption demand
of Indian farm-households. Changes in consumption decisions following a change in
prices or in climate values reflecting different climate scenarios are then studied. The
results show that the agricultural production is sensitive to both precipitation and
rainfall. Consumption decisions are also affected by climate conditions. In particular,
an increase in total rainfall leads to a higher demand for pure market goods and animal
derived products, and a decrease in crops and leisure. In addition, crops demand is
more affected by the variation in rainfall for autarkic households relative to other
types of rural households. On the contrary, the demand for crops products of autarkic
households is less affected by varying temperatures. The scenarios in which both
rainfall and temperatures are changed exhibit a trade of between crops and animal-
derived products.

The framework used in this chapter considers first the production decisions of farm
households, and then turns to the consumption decisions. It would be interesting
to model both decisions in a single model. This strategy makes sense when market
imperfections are suspected (De Janvry et al., 1991; Sadoulet et al., 1998; Taylor and
Adelman, 2003). As long as markets are perfect, households are indifferent to con-
suming own-produced and market-purchased goods. However, when markets fail for
a households, separability does not hold any more and the household’s production and
consumption decisions should be solved simultaneously (Singh et al., 1986).

Another idea to consider is the transformation of crop production in calories as per-

formed by Auffhammer and Schlenker (2014). This conversion would provide an in-
sight on food security, and help targeting farm households that are more at risk be-
cause of climate change.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 75

Table 2.6: Almost Ideal Demand System Results

i=m i=c i=a i=L

Coef t-value Coef t-value Coef t-value Coef t-value
Supply and Demand (N = 3, 454)
β_i −0.023 (−0.216) −1.503∗∗∗ (−8.732) 0.522∗ (2.562) 2.004∗∗∗ (25.448)
δ_i −0.064∗∗∗ (−22.952) −0.025∗∗∗ (−5.594) 0.087∗∗∗ (18.275) 0.002 (0.629)
γ_im 0.067∗∗∗ (7.593) 0.056∗∗∗ (6.575) −0.007 (−0.667) −0.116∗∗∗ (−23.048)
γ_ic 0.056∗∗∗ (6.575) 0.148∗∗∗ (12.198) −0.041∗∗∗ (−5.752) −0.163∗∗∗ (−35.013)
γ_ia −0.007 (−0.667) −0.041∗∗∗ (−5.752) 0.068∗∗∗ (3.905) −0.019∗∗∗ (−3.646)
γ_iL −0.116∗∗∗ (−23.048) −0.163∗∗∗ (−35.013) −0.019∗∗∗ (−3.646) 0.299∗∗∗ (86.948)
Rainfall −0.048∗∗∗ (−4.121) 0.232∗∗∗ (11.739) −0.228∗∗∗ (−10.318) 0.044∗∗ (2.875)
Rainfall Squared (×10) 0.020 (1.129) −0.318∗∗∗ (−10.279) 0.326∗∗∗ (9.449) −0.029 (−1.211)
Temp. 0.055∗∗∗ (5.832) 0.115∗∗∗ (8.047) −0.079∗∗∗ (−4.771) −0.091∗∗∗ (−12.276)
Temp. Squared (×10) −0.011∗∗∗ (−5.558) −0.025∗∗∗ (−8.228) 0.017∗∗∗ (4.652) 0.020∗∗∗ (12.493)
Crops Kept −0.008∗ (−2.424) 0.010. (1.746) −0.008 (−1.196) 0.006 (1.282)
R-Squared 0.84 0.85 0.81 0.95

Supply Only (N = 4, 715)

β_i 0.194. (1.918) −0.115 (−0.609) −1.702∗∗∗ (−8.380) 2.623∗∗∗ (21.506)
δ_i −0.054∗∗∗ (−17.347) −0.024∗∗∗ (−5.157) 0.102∗∗∗ (24.096) −0.024∗∗∗ (−3.965)
γ_im 0.017∗ (2.121) 0.026∗∗ (3.057) 0.107∗∗∗ (9.661) −0.151∗∗∗ (−25.085)
γ_ic 0.026∗∗ (3.057) 0.151∗∗∗ (18.975) −0.010 (−0.844) −0.168∗∗∗ (−29.715)
γ_ia 0.107∗∗∗ (9.661) −0.010 (−0.844) −0.131∗∗∗ (−5.141) 0.033∗ (2.066)
γ_iL −0.151∗∗∗ (−25.085) −0.168∗∗∗ (−29.715) 0.033∗ (2.066) 0.286∗∗∗ (28.092)
Rainfall −0.010 (−1.040) 0.281∗∗∗ (17.876) −0.269∗∗∗ (−15.645) −0.002 (−0.187)
Rainfall Squared (×10) −0.037∗ (−2.421) −0.359∗∗∗ (−13.759) 0.351∗∗∗ (12.384) 0.045∗ (2.330)
Temp. 0.020∗ (2.523) −0.022 (−1.627) 0.108∗∗∗ (6.207) −0.106∗∗∗ (−16.476)
Temp. Squared (×10) −0.003. (−1.946) 0.006∗ (2.180) −0.026∗∗∗ (−6.818) 0.023∗∗∗ (16.369)
Crops Kept −0.008∗ (−2.165) 0.005 (0.713) −0.005 (−0.722) 0.008. (1.740)
R-Squared 0.78 0.77 0.73 0.93

Demand Only (N = 2, 445)

β_i 0.155 (1.106) −0.830∗∗∗ (−3.882) −0.269 (−0.984) 1.944∗∗∗ (17.566)
δ_i −0.043∗∗∗ (−16.323) −0.042∗∗∗ (−9.554) 0.081∗∗∗ (15.279) 0.004 (1.007)
γ_im 0.051∗∗∗ (6.051) 0.013 (1.433) 0.032∗∗ (3.011) −0.096∗∗∗ (−19.677)
γ_ic 0.013 (1.433) 0.155∗∗∗ (13.963) −0.013 (−1.133) −0.155∗∗∗ (−27.800)
γ_ia 0.032∗∗ (3.011) −0.013 (−1.133) 0.019 (0.824) −0.038∗∗∗ (−5.563)
γ_iL −0.096∗∗∗ (−19.677) −0.155∗∗∗ (−27.800) −0.038∗∗∗ (−5.563) 0.289∗∗∗ (68.596)
Rainfall 0.057∗∗∗ (4.448) 0.230∗∗∗ (11.765) −0.323∗∗∗ (−13.152) 0.036∗ (1.978)
Rainfall Squared (×10) −0.147∗∗∗ (−7.137) −0.305∗∗∗ (−9.541) 0.471∗∗∗ (11.704) −0.019 (−0.651)
Temp. 0.007 (0.581) 0.072∗∗∗ (4.092) 0.010 (0.435) −0.089∗∗∗ (−10.214)
Temp. Squared (×10) −0.000 (−0.153) −0.016∗∗∗ (−4.232) −0.004 (−0.742) 0.020∗∗∗ (10.697)
Crops Kept −0.006 (−1.375) 0.017∗ (2.366) −0.006 (−0.663) −0.005 (−0.768)
R-Squared 0.89 0.91 0.86 0.96

Autarkic (N = 1, 602)
β_i 0.047 (0.265) −0.212 (−0.723) −1.532∗∗∗ (−3.908) 2.696∗∗∗ (15.855)
δ_i −0.036∗∗∗ (−9.255) −0.036∗∗∗ (−6.821) 0.086∗∗∗ (13.384) −0.014∗ (−2.146)
γ_im 0.029∗∗ (2.643) 0.013 (1.089) 0.078∗∗∗ (5.332) −0.119∗∗∗ (−19.908)
γ_ic 0.013 (1.089) 0.138∗∗∗ (9.875) 0.019 (1.111) −0.170∗∗∗ (−22.699)
γ_ia 0.078∗∗∗ (5.332) 0.019 (1.111) −0.102∗∗ (−2.811) 0.005 (0.297)
γ_iL −0.119∗∗∗ (−19.908) −0.170∗∗∗ (−22.699) 0.005 (0.297) 0.284∗∗∗ (28.709)
Rainfall 0.104∗∗∗ (5.491) 0.245∗∗∗ (8.757) −0.310∗∗∗ (−8.901) −0.039. (−1.749)
Rainfall Squared (×10) −0.229∗∗∗ (−6.856) −0.297∗∗∗ (−5.912) 0.406∗∗∗ (6.466) 0.120∗∗ (3.075)
Temp. 0.004 (0.301) 0.010 (0.442) 0.110∗∗ (3.249) −0.125∗∗∗ (−10.830)
Temp. Squared (×10) 0.001 (0.171) −0.002 (−0.385) −0.026∗∗∗ (−3.560) 0.027∗∗∗ (10.996)
Crops Kept −0.002 (−0.337) 0.005 (0.503) −0.008 (−0.594) 0.005 (0.580)
R-Squared 0.92 0.94 0.91 0.98

Notes: The estimated coefficients are described in eq. (2.6). The subscripts m, c, a, and L refer to market goods, crops, animal
products, and leisure, respectively. All agricultural households are furthermore divided into four categories according to their
participation in the job market. These four categories are as follows: “Supply and Demand”, if they both engage in off-farm work
and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on their family
plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if they neither
engage in off-farm work nor hire labour on their family [Link] levels are denoted by a dot (· ) at the 10% level, one
asterisk (∗ ) at the 5% level, two asterisks (∗∗ ) at the 1% level and three asterisks (∗∗∗ ) at the .1% level.
76 Part I. Climate Change in Developing Countries: the Indian Case

Table 2.7: Elasticities

Marshallian (Compensated) Price Hicksian (Uncompensated) Price Demand

pm pc pa p∗L pm pc pa p∗L Expend.
All Agricultural Households
Cm −0.569∗∗∗ 0.417∗∗∗ 0.007 −0.561∗∗∗ −0.447∗∗∗ 0.670∗∗∗ 0.187∗∗∗ −0.410∗∗∗ 0.705∗∗∗
Cc 0.171∗∗∗ −0.499∗∗∗ −0.145∗∗∗ −0.428∗∗∗ 0.327∗∗∗ −0.176∗∗∗ 0.085∗∗∗ −0.236∗∗∗ 0.900∗∗∗
Ca −0.098∗∗∗ −0.352∗∗∗ −0.734∗∗∗ −0.185∗∗∗ 0.138∗∗∗ 0.139∗∗∗ −0.384∗∗∗ 0.107∗∗∗ 1.370∗∗∗
CL −0.520∗∗∗ −0.759∗∗∗ −0.081∗∗∗ 0.395∗∗∗ −0.353∗∗∗ −0.413∗∗∗ 0.165∗∗∗ 0.601∗∗∗ 0.965∗∗∗
Supply and Demand
Cm −0.296∗∗∗ 0.547∗∗∗ −0.306∗∗∗ −0.572∗∗∗ −0.188∗∗∗ 0.778∗∗∗ −0.151∗∗∗ −0.440∗∗∗ 0.627∗∗∗
Cc 0.209∗∗∗ −0.559∗∗∗ −0.159∗∗∗ −0.423∗∗∗ 0.370∗∗∗ −0.215∗∗∗ 0.072∗∗ −0.227∗∗∗ 0.932∗∗∗
Ca −0.325∗∗∗ −0.374∗∗∗ −0.477∗∗∗ −0.176∗∗∗ −0.092∗ 0.126∗ −0.142∗∗∗ 0.109∗∗∗ 1.353∗∗∗
∗∗∗ ∗∗∗ ∗∗
CL −0.559 −0.779 −0.086 0.415 −0.385 −0.407∗∗∗ 0.164∗∗∗ 0.628∗∗∗
∗∗∗ ∗∗∗
1.009∗∗∗
Supply Only
Cm −0.641∗∗∗ 0.347∗∗∗ 0.262∗∗∗ −0.669∗∗∗ −0.514∗∗∗ 0.621∗∗∗ 0.433∗∗∗ −0.539∗∗∗ 0.701∗∗∗
Cc 0.121∗∗∗ −0.572∗∗∗ −0.093∗∗∗ −0.396∗∗∗ 0.291∗∗∗ −0.205∗∗∗ 0.137∗∗∗ −0.223∗∗∗ 0.939∗∗∗
Ca 0.072. −0.321∗∗∗ −1.071∗∗∗ −0.099∗∗∗ 0.329∗∗∗ 0.233∗∗∗ −0.725∗∗∗ 0.163∗∗∗ 1.417∗∗∗
CL −0.705∗∗∗ −0.824∗∗∗ 0.033 0.626∗∗∗ −0.548∗∗∗ −0.484∗∗∗ 0.246∗∗∗ 0.786∗∗∗ 0.869∗∗∗
Demand Only
Cm −0.520∗∗∗ 0.270∗∗∗ 0.025 −0.506∗∗∗ −0.402∗∗∗ 0.497∗∗∗ 0.222∗∗∗ −0.317∗∗∗ 0.732∗∗∗
Cc 0.122∗∗∗ −0.408∗∗∗ −0.127∗∗∗ −0.453∗∗∗ 0.262∗∗∗ −0.139∗∗∗ 0.105∗∗∗ −0.229∗∗∗ 0.866∗∗∗
Ca −0.064. −0.262∗∗∗ −0.734∗∗∗ −0.242∗∗∗ 0.146∗∗∗ 0.143∗∗ −0.385∗∗∗ 0.096∗∗∗ 1.302∗∗∗
CL −0.378∗∗∗ −0.606∗∗∗ −0.138∗∗∗ 0.108∗∗∗ −0.215∗∗∗ −0.291∗∗∗ 0.134∗∗∗ 0.371∗∗∗ 1.014∗∗∗
Autarkic
Cm −0.694∗∗∗ 0.239∗∗∗ 0.272∗∗∗ −0.603∗∗∗ −0.562∗∗∗ 0.488∗∗∗ 0.493∗∗∗ −0.419∗∗∗ 0.785∗∗∗
Cc 0.112∗∗ −0.478∗∗∗ −0.042. −0.478∗∗∗ 0.260∗∗∗ −0.197∗∗∗ 0.208∗∗∗ −0.271∗∗∗ 0.886∗∗∗
Ca 0.083. −0.165∗∗ −1.088∗∗∗ −0.135∗∗∗ 0.302∗∗∗ 0.249∗∗∗ −0.720∗∗∗ 0.170∗∗∗ 1.306∗∗∗
CL −0.472∗∗∗ −0.682∗∗∗ −0.032 0.246∗∗∗ −0.314∗∗∗ −0.384∗∗∗ 0.232∗∗∗ 0.466∗∗∗ 0.940∗∗∗
Notes: The subscript m, c, a, and L refer to pure market, crops, animals and leisure, respectively. Significance levels are denoted
by a dot (· ) at the 10% level, one asterisk (∗ ) at the 5% level, two asterisks (∗∗ ) at the 1% level and three asterisks (∗∗∗ ) at the .1%
level. The statistics provided in the table concern farm-househols only. Agricultural households are divided into four categories
according to their participation in the job market. These four categories are: “Supply and Demand”, if they both engage in off-
farm work and hire labour on their family plot; “Supply only”, if they engage in off-farm work and do not hire any labour on
their family plot; “Demand only”, if they do not engage in off-farm work and hire labour on their family plot; and “Autarkic”, if
they neither engage in off-farm work nor hire labour on their family plot.
 Chapter 2. Climate Change and Food Security: a Farm-Household Model 77

All Ag. All Ag. All Ag. All Ag.

Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.

80
Percent Change

Percent Change

Percent Change

Percent Change
50 50 50 50

0 0 0 0

-50 -50 -50 -50

5 5 5 5
Te 4 20
30 Te 4 20
30 Te 4 20
30 Te 4 20
30
m 3 10 m 3 10 m 3 10 m 3 10
pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on 60
tu -20 i tu -20 i tu -20 i tu -20 i
re 0-30 ecip re 0-30 ecip re 0-30 ecip re 0-30 ecip
Pr Pr Pr Pr
Supply and Demand Supply and Demand Supply and Demand Supply and Demand
Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.
Percent Change

Percent Change

Percent Change

Percent Change
40
50 50 50 50

0 0 0 0

-50 -50 -50 -50

5 5 5 5
Te 4 30 Te 4 30 Te 4 30 Te 4 30
m 3 20 m 3 20 m 3 20 m 3 20
10 10 10 10
pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on
tu -20 i tu -20 i tu -20 i tu -20 i
re 0-30 ecip re 0-30 ecip re 0-30 ecip re 0-30 ecip
Pr Pr Pr Pr 20

Supply Only Supply Only Supply Only Supply Only

Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.
Percent Change

Percent Change

Percent Change

Percent Change
50 50 50 50

0 0 0 0 0

-50 -50 -50 -50

5 5 5 5
Te 4 20
30 Te 4 20
30 Te 4 20
30 Te 4 20
30
m 3 10 m 3 10 m 3 10 m 3 10
pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on
tu 0
-20
i p i tu 0
-20
ip i tu 0
-20
i p i tu 0
-20
i p i
re -30 ec re -30 ec re -30 ec re -30 ec
P r P r P r P r
Demand Only Demand Only Demand Only Demand Only -20
Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.
Percent Change

Percent Change

Percent Change

Percent Change

50 50 50 50

0 0 0 0

-50 -50 -50 -50 -40

5 5 5 5
Te 4 20
30 Te 4 20
30 Te 4 20
30 Te 4 20
30
m 3 10 m 3 10 m 3 10 m 3 10
pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on
tu -20 i tu -20 i tu -20 i tu -20 i
re 0-30 ecip re 0-30 ecip re 0-30 ecip re 0-30 ecip
P r P r P r P r
Autarkik Autarkik Autarkik Autarkik
Market Goods Exp. Crops Exp. Animals Exp. Leisure Exp.
-60
Percent Change

Percent Change

Percent Change

Percent Change

50 50 50 50

0 0 0 0

-50 -50 -50 -50

5 5 5 5
Te 4 20
30 Te 4 20
30 Te 4 20
30 Te 4 20
30 -80
m 3 10 m 3 10 m 3 10 m 3 10
pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on pe 2
ra 1 0
-10 tati
on
tu -20 i tu -20 i tu -20 i tu -20 i
re 0-30 ecip re 0-30 ecip re 0-30 ecip re 0-30 ecip
Pr Pr Pr Pr

Notes: The graphs show the effects of a combined modification of precipitation and temperature on consumption of each good,
in each different labour regimes. The x-axis gives the percent variation in temperature, the y-axis gives the temperature increase
in degree, and the z-axis gives the median percent change in consumption. Agricultural households are divided into four labour
regimes: “Supply and Demand”, if they both engage in off-farm work and hire labour on their family plot; “Supply only”, if they
engage in off-farm work and do not hire any labour on their family plot; “Demand only”, if they do not engage in off-farm work
and hire labour on their family plot; and “Autarkic”, if they neither engage in off-farm work nor hire labour on their family plot.

Figure 2.10: Effects of Climate Change on Consumption (Median Percent Changes)

 Part II

Climate Change in Developed

Countries

79
 Part II. Climate Change in Developed Countries 81

In the first two chapters, the focus is made on India, a developing country where the
agricultural sector still represents a substantial share of its GDP. The second part of
this thesis offers a complementary view on the effects of climate change on agriculture
by considering developed countries rather than developing countries.

In developed countries, although the value added of agriculture in GDP is much lower
than in developed countries, examining the effects of the climate and the weather on
production is also essential. In fact, many developed economies are key actors in the
world’s agricultural production. In the context of a changing climate, it is important
to quantify the effects of weather variations on crop production, especially as the FAO
states that in order to feed the world population in 2050, food production should in-
crease by 70 percent between 2005–07 and 2050 (Bruinsma, 2009). The third chapter
thus looks into the relationship between the weather and the agricultural yields of two
major crops in Europe, wheat and corn. These two crops account for about two third
of Europe’s cereal production. On the global market, according to the FAO, European
production of corn accounted for 12% of production on average from 1990 to 2014,
and for a third of global wheat production. It then appears necessary to investigate
how these crops react to weather variability, and to quantify this response under dif-
ferent climate scenarios. To that end, the third chapter provides an empirical study of
the response of wheat and corn yields to weather variability, using regional data from
1991 to 2009. The analysis incorporates the effects of production prices on yields in
its framework, allowing for a different effect before and after the market reform that
reduced price supports. Different climate scenarios are then envisaged to assess the
potential response of wheat and corn yields to climate change.

The fourth chapter also considers the case of developed economies, but focuses on a
smaller actor on agricultural markets. New Zealand is used as an illustrative example.
New Zealand, is a small-open economy with a relatively important agricultural sector
that accounts for around 7% of its GDP according to the World Bank. Its geograph-
ical area is relatively small (268, 838 square kilometres), so that a weather shock af-
fecting agriculture in the country can have significant impacts on national GDP. The
fourth chapter investigates how these weather shocks affect the business cycles of a
small open economy through a general equilibrium approach. To that end, a Dynamic
Stochastic General Equilibrium Model (DSGE) is developed and then estimated using
New Zealand data. Different climate scenarios are once again envisaged, to estimate
the response of the economy to an environment where the variance of climate shocks
is increased.
Chapter 3

Climate Change and

Agricultural Yields: an European
Case Study

Joint work with Catherine Benjamin (University of Rennes 1)

1 Introduction
Climate change may be one of the greatest threats facing the planet. The Intergovern-
mental Panel on Climate Change (IPCC), which includes more than 1, 300 scientists all
around the world, shows in its last study (Edenhofer et al., 2014) that over recent years
we observe increasing temperatures in various regions, and increasing occurrence and
intensity in extreme weather events such as droughts and floods.

Climate change poses a major challenge to agriculture because of the critical depend-
ence of the agricultural system on climatic variables such as temperature and pre-
cipitation. Over time, climate change is expected to increase the annual variation in
crop and livestock production because of its effects on weather patterns and because
of increases in some types of extreme weather events. There is an important literat-
ure focused on predicting and measuring the impact of climate change on agricultural
systems in many countries around the world. A brand of that literature focuses on
crop yields and adopts two complementary methods. Both are based on the produc-
tion function approach, but one relies on crop-growth biophysical modelling while the
other uses statistical methods.

83
84 Part II. Climate Change in Developed Countries

Crop simulation models represent crop growth through mathematical equations as

functions of soil conditions, the weather and management practices (Mearns et al.,
1997). The mathematical models are calibrated from carefully controlled agronomic
experiments (see, e.g., Adams et al., 1998; Rosenzweig and Parry, 1994). Crops are
grown in field or laboratory settings under different conditions. This process allows
to calibrate the parameters of the mathematical model. Then, the model can be used
to simulate crop growth under new climate conditions, reflecting climate change, for
instance. The main shortcoming of these models is that they do not take into account
adaptive behaviours of farmers. In fact, no changes are allowed to farming methods
across experimental conditions so that all differences in outcomes are assumed to be
due to only changes variables of interest (temperature, precipitation). Hence, the ef-
fects of climate change are probably overestimated (see e.g., Adams et al., 1990; Parry
et al., 2004).

This pitfall is partly overcome by statistical analyses. Rather than setting-up experi-
ments to control some input variables, statistical analyses rely on historical observa-
tions on yields, using cross-sectional or panel data (see, e.g., You et al., 2009; Schlenker
and Roberts, 2009; Lobell et al., 2011). The variation of crop yields can be linked to
the level of inputs, to farming practices, and also to varying weather conditions. The
covered geographical area can be large, and not bounded to the experiment field as in
biophysical models (Lobell and Burke, 2010).

However, both methods suffer from their failure to account for some changing be-
haviour of farmers. Neither biophysical nor statistical models will be able to catch
a switch in crops grown or a change in activity if a change in the opportunity cost
occurs. Mendelsohn et al. (1994) described this as the “dumb farmer scenario”, and
therefore recommended to explore the relationship between climate and net reven-
ues, rather than yields. The framework proposed by Mendelsohn et al. (1994), called
the “Ricardian analysis”, is a hedonic model of farmland pricing that focuses on land
value. The basic idea is that long run climate should be capitalized into land values: in
a competitive market, the price of farm land reflects the discounted value of all the ex-
pected future profits that can derive from it. In this framework, farmers adaptation is
implicitly accounted for. They adjust their decisions regarding the level of their inputs
and outputs depending on the climate conditions they face. A handful of studies use
the Ricardian analysis in European regions (Maddison, 2000; Lang, 2007; Lippert et al.,
2009; García and Viladrich, 2009; De Salvo et al., 2013; Van Passel et al., 2016). The use
of a Ricardian framework however implies a strong assumption regarding input and
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 85

output prices. Both are assumed to remain unchanged due to the use of cross-sectional
data, thus making it impossible to incorporate year-to-year fluctuations of production
prices. Special attention should be given on that point as we observe high volatility of
crop prices (Cline, 1996; Schlenker et al., 2005).

In addition, the Ricardian framework is built around agricultural profits and therefore
does not look into the question of yield variability, which is important in the case of
Europe. During the last decades, according to Eurostat data, crop yields have increased
although the cultivated area has decreased. In the meantime, the variability of crop
yields has also increased, partly due to the rise in the occurrence of adverse weather
events.

Furthermore, a major feature in the European Union is the impact of the Common
Agricultural Policy (CAP) on farmer behaviour. Over the last years, major changes
in agricultural subsidies and environmental policies have been observed. The chan-
ging policies probably have impacted yields. For example, they might have lowered
crop yields by reducing the incentive for intensive agriculture. They might also have
lowered yields by restricting the use of fertilizers.

It then appears that the question of yield variability in Europe should be addressed. In
most empirical work, the focus is made on the mean effect of the weather variability
on crop yields. Some scholars suggest to drop the hypothesis of stationarity of weather
variables and to examine how weather variables affect higher order moments of crop
yields distribution (see, e.g., McCarl et al., 2008). One way of achieving this goal is to
rely on the theoretical framework of Just and Pope’s stochastic production function
(Just and Pope, 1978).

This third chapter aims at estimating the effects of climate change on the European
agriculture in terms of wheat and corn mean and variance yields, controlling for pro-
duction prices. The estimated coefficients are then used to simulate the effects on
agricultural yields of projected changes in precipitation and temperature, under dif-
ferent scenarios.

The content of this chapter makes three contributions to the analysis of climate change
in the European Union. First, the effects of the weather on both the mean and the
variance of two major crops, wheat and corn are assessed using statistical methods.
Second, year-to-year fluctuation in prices and their potential impact on crop yields
are incorporated in the study. The structural changes brought on by the 1999 CAP
reform are accounted for. Hence, the response by farmers following the reduction in
86 Part II. Climate Change in Developed Countries

market support for prices after the reform can be examined. Third, this study covers a
large part of western Europe and gives the panorama of both historical and potential
forthcoming effects of the weather on agricultural yields.

The remainder of this chapter is organized as follows. In section 2, we describe the

modelling framework and the empirical strategy used in the analysis. In section 3 we
illustrate data and in section 4 we present the estimation results. Section 5 exposes
projected yields under different climate scenarios. Section 6 concludes.

2 Modelling Framework and Estimation Proced-

ure
To assess the effects of the weather and socio-economic variables on both the average
and variability of crop yields in the European Union, a stochastic production function
approach of the form suggested by Just and Pope (Just and Pope, 1978,9) is applied.
This section first presents the modelling framework and then describes the empirical
approached.

2.1 An Estimation by Maximum Likelihood

The basic idea of this framework is to define the production function as a sum of two
components. The first term is deterministic and linked to the output level while the
second term is linked to the variability of that output. The advantage of this approach
is that it allows to estimate the impacts of an input variable, such as the weather, on
both expected output and on its variance. In fact, these effects could be different. The
general form of the Just and Pope production function for a given crop is:

yi = f (X i , β) + ui , (3.1)
ui = h(X i , α)εi (3.2)

where yi , is crop yields of the ith observation, with i = 1, . . . , n, and f (·) is an average
production function. The matrix X contains the explanatory variables (the weather,
prices, irrigation, soil characteristics), and u is an error term with zero mean and vari-
ance σu2 = h(X, α)2 . The function h(·) for the error term accounts for heteroskedasti-
city. The vectors α and β are the unknown parameters to be estimated. In addition,
we assume that εi is a random error normally distributed with zero mean and variance
σε2i = 1.
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 87

In this framework, the effects of explanatory variables on mean yields and on the
variance of yields are assumed not to be tied so that E(y) = f (X, β) and V(y) =
h(X, α)2 . The effect of an input (e.g., weather variable) can thus be different on mean
yields and on yield variability.

The stochastic production function defined by eq. (3.1) can be estimated using a three-
step estimation procedure involving feasible generalized least squares (FGLS) under
heteroscedastic disturbances (as in Cabas et al., 2010). However, in the case of small
samples, maximum likelihood estimators are more efficient and less biased than FGLS
ones (Saha et al., 1997). We thus estimate the parameters of the regression in a single
step by maximum likelihood (as in Chen et al., 2004), using the FGLS estimates as
initial values for the optimization algorithm.

The estimation procedure encompasses three steps. The first one consists in regress-
ing yields on the set of dependent variables using ordinary least squares. Residuals
from this regression are an estimator of h(X, α)ε. They are used in the second step
to estimate the marginal effects of dependent variables on the variance (Cabas et al.,
2010). We assume that h(·) has an exponential form. Hence, we regress the logar-
ithm of squared residuals on the set of explanatory variables. Predicted values of this
second regression are used in a third step as weights in the regression by weighted
least squares of yields on the set of independent variables. The estimates from this
third step are the starting values used in the maximum likelihood estimation. Assum-
ing that εi ∼ N (0, 1), the log likelihood function reads as follows:
[ ]
1 ∑n
( ) ∑n
(yi − f (X, β))2
ln L = − nln(2π) + ln h(X, α)2 + . (3.3)
2 i=1 i=1
h(X, α)2

The maximum likelihood estimates of α and β are obtained by maximizing eq. (3.3)
with respect to α and β.

2.2 The Production Function Assumption

It is necessary to make an assumption on the production function f (·). It is common

practice to assume a linear form. The matrix X from eq. (3.1) is split in sub-groups of
variables: climate W , crop prices P , percentage of irrigated surface I, soil character-
istics S.
88 Part II. Climate Change in Developed Countries

Yields of crop c ∈ {m, w}, where m stands for for corn and w for wheat, in region r
and in year t are modeled as follows:

ycrt =βc1 W crt + βc2 Pct−1 + βc3 Icrt + βc4 Scr +

βc5 CAP1999
t + βc6 CAP1999
t × Pct−1 + µc + ucrt . (3.4)

We introduce a dummy variable, CAPt , to reflect the effects of the 1999 Common Agri-
cultural Policy reform. This variable takes the value 1 up to 1999 and zero afterwards.
An interaction term between prices and the 1999 CAP reform dummy variable is also
introduced in the equation, as we believe the effect of prices on production might dif-
fer after the market support for prices for cereals were reduced following the reform.
Furthermore, prices are introduced in the model with a one period lag to account for
their stickiness. We also add country fixed effects µ.1 The set of weather variables
W examines the effects of seasonal temperature, rainfall, and temperature deviation
– measured as the spread between the maximum and the minimum observed temper-
ature. Weather variables are disaggregated to enable their effects on yields to vary
between seasons. The weather variables whose effects on either mean yields or yield
variability are not significant are discarded from the model. As the model is used
once it is estimated to assess the potential effects of climate change on yields, keeping
variables with non-significant effects might bias the predictions. Finally, soil charac-
teristics of each region are introduced to account for the heterogeneity in soil quality.

In the presence of spatial correlation, estimates might be biased and inconsistent (Miao
et al., 2015; Auffhammer et al., 2013; Schlenker and Lobell, 2010). To overcome this
pitfall we rely on 1, 000 bootstrap runs, randomly selecting observations2 to estimate
both coefficients and standard errors.

As the data generating process of corn and wheat yields may differ for northern and
southern regions, the model described in eq. (3.4) is estimated for two subsets: one for
northern regions and the other for southern ones. We use the 45◦ parallel north to
separate northern and southern regions.3
1
Fixed effects are introduced at the country level and not at the region level to avoid an overuse that
can lead to estimators with inflated variability (Van Passel et al., 2016).
2
In each run, we randomly select 80% of available observation and get rid of the 20% left.
3
Using the 45◦ parallel north fairly separates the regions between the Mediterranean and the others
ones, as displayed in Füssel et al. (2012, p. 27).
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 89

3 Data Sets and Empirical Specifications

We focus on the effects of weather variables on yields for two important grain crops,
i.e., wheat and corn. Our study covers 19 years of observations from 1991 to 2009 in 31
European regions for wheat producers and 25 for corn producers. The main inform-
ation is collected from the Farm Accountancy Data Network. The dataset is comple-
mented by regional registers or surveys that provide information on spatial location
variables. eq. (3.4) is estimated using data gathered from five different databases.

3.1 Production Data and Definition of the Agricultural

Variables

First, we use data from the Farm Accountancy Data Network4 (FADN) apportioned
down to NUTS-3 geographic level for statistical anonymity. We only focus on repres-
entative farms whose main stated activity is crop growing, and collect three variables:
production, cultivated area, and yields. The values of each of these variables are not
necessarily provided for the entire period. Hence, we discarded regions for which less
than 15 observations were available.5 The number of NUTS-3 regions used in this
chapter as well as the number of observation are reported by country in table 3.1.
As previously mentioned, we use the 45◦ parallel North to differentiate northern and
southern regions. Regions with a centroid above this parallel are included in the north-
ern sample while the other regions are included in the southern sample.

Information at the average level for each region is reported in fig. 3.1 for production,
area and yields for both crops. The subplots on the left describe the wheat case while
those on the right side depict the corn case. tables 3.2 and 3.3 offer a complementary
view on the description of the data, for the sample used to analyse wheat and corn
yields, respectively. Descriptive statistics are reported for the entire period covered,
from 1991 to 2009, as well as for two subsets, before and after the 1999 CAP reform, i.e.,
from 1991 to 1999, and from 2000 to 2009, respectively. For each of these three differ-
ent setups, descriptive statistics are also reported for northern and southern regions,
separately.
4
The FADN database is publicly available at [Link]
consult_std_reports_en.cfm.
5
This concerns only five regions in the sample.
90 Part II. Climate Change in Developed Countries

Table 3.1: Number of NUTS-3 Regions and Observation by Country and Location

Wheat Corn

North South North South

NUTS-3 Obs. NUTS-3 Obs. NUTS-3 Obs. NUTS-3 Obs.

Austria 1 14 - - 1 14 - -
France 15 234 15 36 16 234 16 54
Italy 2 36 - - 2 36 - -
Netherlands 1 18 - - - - - -
Portugal - - 1 18 - - 2 36
Spain - - 6 107 - - 4 72
Sweden 2 28 - - - - - -
United Kingdom 3 54 - - - - - -

Total 24 384 22 161 19 284 22 162

Notes: This table gives the number of observations in the four different datasets used in this analysis (wheat yields in northern
regions, wheat yields in southern regions, corn yields in northern region, and corn yields in southern regions). The figures are
aggregated at the country level here. Observations for the datasets on wheat yields are reported on the left while observations
for the datasets on corn yields are reported on the right. Each dataset is also split between north and south regions. Regions
whose centroid is higher than 45◦ are considered on the north while the others are considered on the south. The columns untitled
“NUTS-3” give the number of NUTS-3 regions for each country in the datasets and the columns untitled “Obs.” report the number
of observation by country.

We observe major differences across northern and southern regions. Levels of all ag-
ricultural variables are higher in northern regions than in the southern ones. Remark-
able difference is found for average yields. The average value of wheat yields in the
north is about 70 quintals per hectare which is twice larger than the average value of
yields in the south. For corn, a distinction between northern and southern regions is
also observed, with the former being more productive than the latter. Even if both pro-
duction and surface are almost twice higher in northern regions, the average value of
corn yields is about the same in both northern and southern regions, and equals about
86 quintals per hectare. (table 3.3). It is noteworthy that wheat yields remained almost
unchanged after the 1999 CAP reform, with a slight increase in both production and
surface. On the other hand, corn yields in southern Europe improved substantially
and have closed the gap as compared to yields in northern Europe.

The FAO offers data on nominal producer prices, expressed in USD per ton,6 at the
country level. We divide this variable by a 2004–2006 producer price index given for
each type of crop. This yields a variable of real prices. As shown in fig. 3.2, wheat and
corn production prices decline up to 2001, then start increasing. We also observe a not-
able difference between prices observed in northern and southern regions, production
prices being relatively higher in southern regions.
6
We stick to US Dollars rather than converting prices in Euros, as the former is the currency used
in the global agricultural market.
 Chapter 3. Climate Change and Agricultural Yields: an European Case Study 91

Production (tonnes) Surface (ha) Yield (100kg/ha)

100 200 300 400 500 20 40 60 20 40 60 80 100
Wheat Corn Wheat Corn Wheat Corn

(a) Average Production (b) Average Surface (c) Average Yields

Notes: The vertical grey line represents the 45◦ parallel north used to separate northern and southern regions.

Figure 3.1: Agricultural Production of Wheat and Corn in Western Europe

Wheat Yield (100kg/ha) Corn Yield (100kg/ha)

100
70

60 90

50
80
40

30 70
1995 2000 2005 1995 2000 2005

Wheat Price (USD/tonne) Corn Price (USD/tonne)

200

180 175

150 150

125
120

100
1995 2000 2005 1995 2000 2005

Region North South

Figure 3.2: Evolution of Yields and Prices

3.2 Environment Data and Spatial Location of the Farm Vari-

ables

Locating farms within its spatial environment is very important to understand differ-
ences in production decisions. The environment affects agricultural production, via
water availability, soil composition, or the weather.

Previous studies point out the importance of irrigation. In fact, the effect of climate
change on agriculture might be different for irrigated and rain-fed farms (Schlenker
et al., 2005). This information is not given in the Public FADN database. Hence, we
use the share of irrigable agricultural area by region as a proxy (as in Polsky and
Easterling III, 2001; Gbetibouo and Hassan, 2005; Barnwal and Kotani, 2013). We divide
92 Part II. Climate Change in Developed Countries

the total irrigable area by the utilized agricultural area obtained from Eurostat7 to get
the share of irrigable agricultural area.8 Table 3.2 and table 3.3 show that the share
of irrigable area is higher in the south than it is in the north, although the standard
deviation is higher in the north.

The geographical location of the farm influences the type of crops that are cultivated,
and also affects yields. Increased latitude might be harmful to crop production, as
regions in higher latitude are less exposed to solar radiation, which are determinant of
yields according to agronomic models. On the other hand, regions closer to the equator
are subject to hotter climate that may have detrimental effects on yields. Longitude
should also be considered. As pointed-out by Vanschoenwinkel et al. (2016), Western
and Eastern Europe do not respond to climate variation in the same way.

The FADN data does not include any information on physical properties of the soil.
The characteristics of soil play an important role in the crop’s ability to extract water
and nutrients. Soil texture or the proportions of different-sized mineral particles that
soil contains must provide a good environment for plant growth. Hence we take into
account the soil’s composition using different measures. Soil quality heterogeneity is
addressed using a worldwide 21, 600 × 43, 200 raster database of soil characteristics
from the Harmonized World Soil Database (Batjes et al., 2008). We compute NUTS-3
averages for the four measures relative to the texture of topsoil: gravel content by
volume; sand , silt, and clay fraction as a percentage of weight. More details on the
values are given on table 3.2 and on table 3.3.

3.3 Weather Data and Definition of the Variables

The key climatic variables that impact crops yields are surface temperature, and level
of rainfall. In this study we adopt information provided in a global climate model, MRI-
CGCM3, which has been developed at the Meteorological Research Institute (MRI).
This framework integrates a global dataset built on a 1.125◦ × 1.12148◦ longitude lat-
itude grid since 1850. Daily precipitation and levels of temperature are available. We
select data on average temperature, total rainfall and temperature deviations. Tem-
perature deviations are computed as the difference between the maximum and the
minimum temperatures, within a day.
7
Data on irrigated areas is available at [Link]
-/ef_lu_ofirrig.
8
The information on irrigable area provided by Eurostat is given only in some NUTS-3 regions an
only every two or three years, so we estimate the missing values in the following way: we keep regions
where at least four observations are given, and impute the missing values by means of B-Splines.
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 93

Our aim is to define weather variables at the regional dimension. Hence in a first step
we aggregate data at the NUTS-3 level. In the calculus we take into account the rel-
ative importance of the cells of the grid in the region by weighting each information
by the share of the cell in the region. The second step is to change the time dimen-
sion to match with previous data collected from the FADN data set. We do not define
annual climate variables. In reality, crop development, i.e., the time from planting to
flowering and/or maturity, which takes place in one year is correlated with temper-
ature and precipitation levels. Climatic factors can promote or inhibit plant growth
and development. Hence, daily data of each weather measure are averaged by sea-
sons. For one year, we distinguish four time information: (i) Winter, from December
to February; (ii) Spring, from March to May; (iii) Summer, from June to August; and
(iv) Fall, from September to November. These different periods of time define different
steps in the development of crops. The first one defines (i) the seeding season for the
initial development, (ii) the vegetative growth stage for the stem extension, and (iii)
the generative growth stage for the ripening and harvesting of the plant.

4 Estimation Results
In this section, we present the estimates for the average crop yields production func-
tion, and for the crop yield variability function.

The coefficients of the production function, estimated by maximum likelihood, and

their associated standard errors are reported in table 3.4.9 The first four columns report
the effects on mean yields while the next four give the effects on the variance of yields.

4.1 Wheat Yields

Mean wheat and variance yields are affected by seasonal temperatures. The effects
of temperature on wheat yields are significant only in summer, for northern regions.
Any additional degree Celsius in summer is harmful to mean yields though it reduces
the variance of yields. Temperature variables of other seasons were removed from the
estimation since the effect of these variables on both mean yields and the variance of
yields were found not significant. Temperature deviation plays a more important role
in explaining variations of mean yields and the variance of yields. Temperature devi-
ation in summer negatively impact mean wheat yields in the north and increases its
variance. In fall, the opposite effect is observed, with a positive impact on mean yields
9
Country-specific coefficients are omitted for clarity purposes; they are available upon request.
94 Part II. Climate Change in Developed Countries

Table 3.2: Descriptive Statistics (Mean and Std) for Wheat Datasets

Whole Period Before CAP Reform After CAP Reform

(1991-2009) (1991-2000) 2001-2009

Variable Unit All North South All North South All North South

Production variables

Yield 100 kg per ha 59.63 70.1 34.65 58.6 69.85 32.43 60.61 70.33 36.84
(20.37) (11.95) (13.33) (21.09) (11.67) (13.48) (19.64) (12.23) (12.9)
Wheat production $100kg 1650.24 2268.4 175.87 1536.7 2130.56 155.96 1758.49 2397.88 195.54
(1675.85) (1637.37) (163.77) (1582.91) (1549.67) (142.3) (1755.9) (1709.46) (181.26)
Wheat surface ha 23.03 30.53 5.16 21.59 28.7 5.05 24.42 32.24 5.28
(20.39) (19.81) (4.27) (19.53) (19.23) (4.26) (21.12) (20.24) (4.3)
Wheat prices USD/100kg 138.44 129.74 159.18 136.51 126.88 158.89 140.28 132.43 159.45
(29.63) (24.14) (31.23) (29.39) (22.07) (32.09) (29.79) (25.7) (30.54)

Environment variables and spatial location of the farm

Irrigable area % 12.83 10.99 17.21 12.44 10.41 17.17 13.19 11.53 17.25
(14.12) (15.95) (6.44) (14.03) (15.7) (7.14) (14.21) (16.19) (5.71)
o
Latitude 47 49.49 41.05 46.83 49.31 41.05 47.16 49.66 41.06
(5.01) (3.49) (2.38) (4.85) (3.26) (2.4) (5.16) (3.69) (2.38)
o
Longitude 2.08 3.82 −2.08 1.8 3.47 −2.1 2.34 4.14 −2.05
(5.48) (5.08) (3.97) (5.22) (4.78) (3.99) (5.71) (5.33) (3.98)
Gravel Content %vol. 8.84 7.84 11.23 8.8 7.76 11.22 8.88 7.91 11.24
(2.23) (1.45) (1.93) (2.24) (1.41) (1.94) (2.23) (1.5) (1.93)
Sand Fraction %wt. 43.73 43.7 43.8 43.67 43.6 43.83 43.78 43.79 43.76
(5.32) (4.19) (7.36) (5.34) (4.19) (7.39) (5.3) (4.19) (7.37)
Silt Fraction %wt. 32.74 32.59 33.1 32.76 32.62 33.08 32.73 32.57 33.12
(3.8) (3.33) (4.73) (3.82) (3.35) (4.75) (3.78) (3.32) (4.74)
Clay Fraction %wt. 22.03 21.88 22.38 22.11 22.01 22.36 21.94 21.76 22.39
(3.15) (2.98) (3.51) (3.12) (2.93) (3.53) (3.18) (3.02) (3.51)

Weather variables
o
Temp. (winter) C 4.62 3.7 6.82 4.52 3.61 6.62 4.72 3.78 7.02
(3.06) (3) (1.85) (2.98) (2.91) (1.9) (3.14) (3.09) (1.8)
o
Temp. (spring) C 9.15 8.21 11.38 8.99 8.04 11.18 9.3 8.37 11.57
(2.64) (2.34) (1.89) (2.56) (2.18) (1.95) (2.72) (2.47) (1.81)
o
Temp. (summer) C 17.6 16.72 19.7 17.74 16.94 19.62 17.46 16.52 19.77
(2.2) (1.62) (1.99) (2.13) (1.68) (1.87) (2.27) (1.54) (2.11)
o
Temp. (fall) C 10.39 9.36 12.85 10.26 9.33 12.4 10.52 9.39 13.3
(2.79) (2.36) (2.15) (2.44) (2.03) (1.9) (3.1) (2.63) (2.3)
o
Temp. Dev. (winter) C 4.49 4.22 5.14 4.5 4.23 5.11 4.49 4.21 5.17
(0.98) (0.93) (0.76) (0.96) (0.93) (0.72) (0.99) (0.93) (0.8)
o
Temp. Dev. (spring) C 7.09 6.73 7.97 7.09 6.72 7.93 7.1 6.73 8
(1.36) (1.27) (1.16) (1.36) (1.27) (1.17) (1.37) (1.28) (1.16)
Temp. Dev. (summer) o C 7.77 7.41 8.64 7.8 7.47 8.56 7.75 7.35 8.73
(1.52) (1.43) (1.38) (1.55) (1.51) (1.36) (1.5) (1.35) (1.41)
o
Temp. Dev. (fall) C 5.32 4.93 6.26 5.2 4.86 5.97 5.44 4.99 6.54
(1.21) (1.04) (1.05) (1.08) (0.97) (0.93) (1.3) (1.1) (1.09)
Precip. (winter) m 8.1 9.24 5.39 7.64 8.72 5.14 8.54 9.73 5.64
(3.16) (2.63) (2.65) (2.95) (2.52) (2.29) (3.3) (2.64) (2.96)
Precip. (spring) m 6.68 7.06 5.77 6.37 6.73 5.52 6.97 7.37 6.01
(1.99) (1.79) (2.14) (1.97) (1.78) (2.15) (1.96) (1.75) (2.11)
Precip. (summer) m 6.35 6.47 6.05 6.55 6.58 6.47 6.15 6.37 5.62
(2.72) (2.2) (3.66) (2.83) (2.17) (3.98) (2.6) (2.23) (3.29)
Precip. (fall) m 8.75 9.39 7.22 9.59 10.22 8.13 7.94 8.6 6.32
(2.79) (2.54) (2.75) (2.78) (2.5) (2.86) (2.55) (2.34) (2.32)
No Observations 545 384 161 266 186 80 279 198 81

Notes: Descriptive statistics (mean and standard error in parenthesis) for all regions, northern regions only and souther regions
only, before and after the 1999 Common Agricultural Policy [Link] are expressed in constant 2005 USD per 100kg.

following a unit spread increase between maximum and minimum daily temperature,
and a diminution of wheat the variance of yields, although the impact on variance is
not significant at the 5% level. In winter, the sign of the effect of temperature devi-
ation on mean yields is opposed for northern and southern regions, but the coefficients
are not significant. However, for both northern and southern regions, an increase in
winter temperature deviation significantly reduces the variance of yields. Finally, pre-
cipitation also alter mean an the variance of yields. We allowed for a quadratic effect,
but it was rejected by the data for all seasons. Precipitation in spring, summer, and
winter have a negative impact on wheat yields in the north. In the south, while the
 Chapter 3. Climate Change and Agricultural Yields: an European Case Study 95

Table 3.3: Descriptive Statistics (Mean and Std) for Corn Datasets

Whole Period Before CAP Reform After CAP Reform

(1991-2009) (1991-2000) 2001-2009

Variable Unit All North South All North South All North South

Production variables

Yield 100 kg per ha 86.48 87.34 84.96 82.97 85.42 78.74 89.92 89.21 91.19
(18.05) (13.78) (23.73) (17.7) (13.71) (22.51) (17.77) (13.65) (23.42)
Corn production $100kg 674.1 807.96 439.43 604.01 728.45 388.92 742.95 885.26 489.94
(812.08) (852.06) (677.94) (777.61) (806.21) (677.93) (840.61) (890.38) (678.37)
Corn surface ha 7.36 8.77 4.88 6.74 8.06 4.47 7.96 9.46 5.3
(8.28) (8.43) (7.42) (8.04) (8.13) (7.39) (8.5) (8.69) (7.48)
Corn prices USD/100kg 143.43 133.27 161.26 141.31 130.26 160.4 145.53 136.19 162.12
(29.39) (21.51) (32.73) (28.84) (18.7) (33.12) (29.83) (23.63) (32.52)

Environment variables and spatial location of the farm

Irrigable area % 14.89 12.67 18.79 15.02 12.18 19.92 14.77 13.15 17.65
(15.27) (17.63) (8.62) (15.46) (17.25) (10.12) (15.11) (18.05) (6.67)
o
Latitude 45.54 47.8 41.58 45.52 47.8 41.58 45.56 47.8 41.58
(3.47) (1.53) (2.12) (3.49) (1.54) (2.12) (3.47) (1.52) (2.12)
o
Longitude 1.7 3.68 −1.77 1.59 3.53 −1.77 1.81 3.83 −1.77
(5.06) (4.14) (4.66) (4.95) (3.98) (4.68) (5.18) (4.29) (4.68)
Gravel Content %vol. 9.04 7.9 11.05 9.04 7.87 11.05 9.05 7.92 11.05
(2.09) (1.09) (1.89) (2.1) (1.09) (1.9) (2.08) (1.1) (1.9)
Sand Fraction %wt. 43.97 42.88 45.88 43.97 42.86 45.88 43.97 42.89 45.88
(5.24) (3.22) (7.19) (5.27) (3.25) (7.22) (5.22) (3.21) (7.22)
Silt Fraction %wt. 33.04 33.55 32.14 33.01 33.52 32.14 33.06 33.58 32.14
(3.73) (3) (4.62) (3.74) (3.02) (4.63) (3.72) (3) (4.63)
Clay Fraction %wt. 22.09 22.57 21.25 22.12 22.62 21.25 22.07 22.53 21.25
(2.9) (2.63) (3.15) (2.9) (2.63) (3.16) (2.9) (2.65) (3.16)

Weather variables
o
Temp. (winter) C 5.1 4.21 6.66 4.89 3.97 6.47 5.31 4.44 6.85
(2.53) (2.51) (1.67) (2.59) (2.56) (1.73) (2.46) (2.44) (1.61)
o
Temp. (spring) C 9.72 8.91 11.15 9.47 8.61 10.95 9.97 9.2 11.34
(1.95) (1.63) (1.64) (1.97) (1.58) (1.69) (1.91) (1.63) (1.58)
o
Temp. (summer) C 18.13 17.37 19.46 18.24 17.57 19.4 18.01 17.17 19.51
(1.6) (1.05) (1.52) (1.51) (1.09) (1.45) (1.67) (0.97) (1.6)
o
Temp. (fall) C 10.87 9.86 12.64 10.66 9.75 12.24 11.09 9.98 13.05
(2.35) (1.96) (1.87) (2.04) (1.69) (1.59) (2.6) (2.2) (2.03)
Temp. Dev. (winter) o C 4.62 4.37 5.05 4.63 4.41 5.01 4.6 4.33 5.09
(0.91) (0.9) (0.76) (0.88) (0.89) (0.73) (0.94) (0.91) (0.79)
o
Temp. Dev. (spring) C 7.34 7.06 7.83 7.34 7.06 7.81 7.34 7.05 7.85
(1.21) (1.15) (1.17) (1.2) (1.13) (1.18) (1.23) (1.17) (1.17)
Temp. Dev. (summer) o C 8.13 7.9 8.54 8.15 7.97 8.45 8.12 7.83 8.63
(1.34) (1.28) (1.34) (1.36) (1.35) (1.32) (1.31) (1.2) (1.35)
o
Temp. Dev. (fall) C 5.59 5.27 6.16 5.44 5.18 5.89 5.75 5.36 6.43
(1.08) (0.96) (1.06) (0.97) (0.88) (0.95) (1.16) (1.02) (1.09)
Precip. (winter) m 8.26 9.31 6.42 7.76 8.68 6.17 8.75 9.92 6.68
(3.33) (2.68) (3.56) (3.16) (2.6) (3.41) (3.43) (2.63) (3.71)
Precip. (spring) m 6.98 7.37 6.29 6.62 6.98 6.01 7.33 7.76 6.57
(2.01) (1.78) (2.2) (2.05) (1.84) (2.27) (1.91) (1.64) (2.11)
Precip. (summer) m 6.49 6.73 6.07 6.74 6.87 6.5 6.25 6.59 5.64
(2.67) (2.36) (3.1) (2.75) (2.3) (3.39) (2.57) (2.42) (2.74)
Precip. (fall) m 9.2 9.72 8.28 10.23 10.77 9.28 8.19 8.7 7.27
(3.02) (2.56) (3.52) (3.07) (2.41) (3.79) (2.61) (2.27) (2.92)
No Observations 446 284 162 221 140 81 225 144 81

Notes: Descriptive statistics (mean and standard error in parenthesis) for all regions, northern regions only and souther regions
only, before and after the 1999 Common Agricultural Policy [Link] are expressed in constant 2005 USD per 100kg.

effect of an increase in precipitation in summer is of the same sign as in northern re-

gions, i.e., negative, it is positive in winter, although only significant at the 10% level.

Irrigable surface, used as a proxy for irrigation has a positive impact on mean yields,
but is only significant in the north of Europe.

We introduce the effect of an important CAP reform over the studied period, the
Agenda 2000. The aim of this reform was to further pursue the 1992 reform of the
CAP and press ahead with the transition to world market prices, particularly through
a substantial drop in the common support prices for cereals. Hence, regarding the
96 Part II. Climate Change in Developed Countries

effects of price in our analysis, a distinction between two periods needs to be done:
before the 1999 reform of the PAC, and after. In the former period, the dummy vari-
able takes zero values. Since we introduced an interaction term between prices and
the dummy variable, the effect of a unit increase in prices reads as the sum of the
coefficients of the price variable and the interaction between the dummy and price
variables, i.e., βc2 + βc6 in eq. (3.4). After 1999, it reads as the coefficient associated
with the price variable only, i.e., βc2 in eq. (3.4). The results of the regressions show
that prices before the 1999 CAP reform played a positive though not significant role
in explaining the variations of mean yields of wheat in northern regions, i.e., regions
in which production in relatively higher. After the CAP reform, the effect remains
positive and becomes significant. For southern regions, prices have a significant and
positive effect on wheat the variance of yields. We note that the effect of prices are
lower after the reform. This result is linked to the fact that changing policy might re-
duce the incentive for intensive cultivation and therefore lower the rate of yields gains.
In addition, changing environmental policies included in Agenda 2000 restricted the
use of fertilizer application and may have limited yields growth.

4.2 Corn Yields

Seasonal weather variables impact both mean yields and the variance of yields. Tem-
perature in winter only impacts yields in northern regions: a unit increase is harmful
to mean yields and also increases the variance of yields. In spring, temperature does
not significantly affect corn yields in the north of Europe. However, in the south,
there is a hill-shape relationship between temperature and yields. There is a threshold
above which any additional degree Celsius becomes harmful to corn yields. The value
of this threshold is 9.24◦ C, which is comprised in the range of observed values, a bit
lower than the observed average. Temperature in summer negatively affects corn
mean yields in the south. In the north, the effect on the mean is not significant, but
an increase in temperature reduces the variance of corn yields. In fall, there is a quad-
ratic effect of temperature on corn yields. Above a threshold of 12.30◦ C, any additional
temperature increase leads to a deterioration of mean yields for regions in the north
of Europe. The value of this threshold is comprised in the range of observed values of
fall temperatures, a bit higher than the average.

Temperature deviations mainly positively affect corn yields in the south, except for
fall, where the coefficient is negative but not statistically different from zero. The
impact of an increase in the spread between the maximum and the minimum daily
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 97

temperature on the variance of corn yields is only significant in fall, for the south of
Europe, and leads to a rise in this variance.

All precipitation variables were rejected by the model, at the exception of precipitation
in fall, for southern regions only. In that particular case, an increase in total rainfall
leads to an increase in the variance of yields.

Just like for wheat, irrigation surface has a positive effect on mean corn yields in the
north of Europe. For region in the south of Europe, the share of irrigable areas only
impact the variance of corn yields, negatively.

Contrary to the regressions for wheat, the dummy variable reflecting the CAP reform
is rejected by the model. The interaction term between the dummy and prices is also
not significant. Looking at prices alone, we see that they are positively linked to corn
yields in the north of Europe; their effects in the south are not statistically different
from zero.

5 Scenarios
Impacts of several climate change scenarios on crops yields are assessed in this study.
These scenarios are conducted over the 2010–2100 period. Four Representative Con-
centration Pathways (RCPs) adopted by the IPCC for its fifth Assessment Reports in
2014 are used in our scenario exercise: RCPs 2.6, RCP4.5, RCP 6.0 and RCP 8.5. These
pathways depict different greenhouse gas concentration trajectories from the most
optimistic to the most pessimistic. The first three, i.e., the RCPs 2.6, 4.5 and 6.0 are
characterized by increasing greenhouse gas concentrations with a peak around 2030,
2040, and 2060, respectively, followed by a slow decline. The last scenario, the RCP 8.5,
is less optimistic in terms of emissions and leads to a rapidly increasing concentration
over the whole century.10 Data come from the same source as the historical weather
variables (see section 3.3).

Before implementing scenarios, we need to define projections levels for all exogenous
variables. All variables except weather ones are set to their historical average from
1991 to 2009. The weather variables vary according to the climate scenario envisaged.
Furthermore, for each of the four scenarios, two different time horizons are examined:
10 Theappendix B displays descriptive statistics of weather variables for each scenario as well as
historical values.
98 Part II. Climate Change in Developed Countries

Table 3.4: Determinants of Wheat and Corn Yield (Maximum Likelihood)

Mean Yield Yield Variance

Wheat Corn Wheat Corn

(North) (South) (North) (South) (North) (South) (North) (South)

Price Lagged 0.082∗∗∗ −0.016 0.115∗∗∗ −0.030 −0.009. 0.001 −0.004 0.024∗∗∗
(0.017) (0.027) (0.033) (0.067) (0.006) (0.007) (0.007) (0.010)
∗∗∗ ∗∗∗
Irrigable area 0.169 0.219 0.701 0.555. 0.010 0.082∗ 0.007 −0.112∗∗
(0.049) (0.204) (0.134) (0.396) (0.020) (0.051) (0.026) (0.052)
D 1999 5.287∗∗ −7.212 −1.052 9.652 −1.321. 1.384 0.448 1.473
(2.864) (6.932) (6.951) (13.397) (0.950) (1.399) (1.650) (1.382)
∗∗∗
D 1999 x Price Lagged −0.055 0.025 −0.040 −0.136. 0.012∗ −0.012. −0.0002 −0.003
(0.021) (0.040) (0.049) (0.083) (0.007) (0.009) (0.012) (0.008)
Latitude 1.691∗∗ 399.408∗∗∗ −1.114. 60.099∗∗∗ 0.048 −25.472 −0.212. −0.409
(0.660) (59.563) (0.775) (17.418) (0.139) (21.382) (0.148) (2.684)
Longitude 2.171∗∗∗ −39.432∗∗∗ 0.207 3.639. −0.150. 1.952 0.076 0.549.
(0.400) (5.890) (0.692) (2.963) (0.121) (2.192) (0.113) (0.421)
Gravel Content 0.361 52.770∗∗∗ 1.718∗∗ −59.167∗∗∗ 0.047 −2.756 −0.008 0.522
(0.540) (8.180) (0.712) (9.479) (0.127) (3.014) (0.140) (1.477)
∗∗ ∗∗∗
Sand Fraction −0.814 −2, 315.835 1.024 −812.160∗∗∗ 0.163. 152.273 −0.168 8.614
(0.378) (346.363) (1.143) (175.207) (0.112) (124.099) (0.128) (26.238)
Silt Fraction 0.539 −2, 554.501∗∗∗ 0.447 −847.371∗∗∗ 0.200. 167.941 −0.151 8.946
(0.560) (382.113) (1.319) (183.417) (0.145) (136.928) (0.162) (27.508)
Clay Fraction −0.923∗ −1, 743.921∗∗∗ 0.784 −672.498∗∗∗ 0.109 114.776 −0.032 6.817
(0.481) (260.636) (1.170) (143.252) (0.125) (93.370) (0.142) (21.487)
Temp. (winter) −2.539∗∗∗ 0.425∗∗∗
(0.657) (0.145)
Temp. (spring) 20.649∗ 2.320∗
(9.739) (1.351)
Temp. (spring) sq. −1.117∗∗ −0.084.
(0.423) (0.061)
Temp. (summer) −1.062∗∗ 0.279 −2.265∗∗ −0.463∗∗∗ −0.381 ∗∗∗
0.082
(0.491) (0.752) (1.314) (0.111) (0.114) (0.188)
Temp. (fall) 7.193∗∗ 0.826.
(3.000) (0.562)
Temp. (fall) sq. −0.292∗∗ −0.053∗∗
(0.140) (0.027)
Temp. Dev. (winter) −1.140 2.825 −0.621∗∗∗ −1.826∗∗∗
(1.108) (2.394) (0.230) (0.411)
Temp. Dev. (spring) 7.610∗∗∗ −0.150
(2.535) (0.408)
Temp. Dev. (summer) −3.096∗∗∗ 6.164∗∗∗ 0.764∗∗∗ −0.130
(0.905) (2.006) (0.208) (0.298)
∗∗∗ ∗∗
Temp. Dev. (fall) 2.423 −3.094 −3.473 −0.216 −0.169 1.205∗∗∗
(0.843) (1.506) (3.575) (0.262) (0.229) (0.507)
Precip. (winter) −0.315∗∗ 0.783. −0.022 −0.240∗∗∗
(0.150) (0.566) (0.042) (0.070)
Precip. (spring) −0.612∗∗ −0.146∗∗
(0.219) (0.071)
Precip. (summer) −0.654∗∗∗ −0.678∗∗ 0.128∗ 0.031
(0.239) (0.345) (0.075) (0.072)
Precip. (fall) −0.610 0.174∗∗
(0.538) (0.077)

Observations 384 161 284 162 384 161 284 162

Notes: Dot (. ), asterisk (∗ ), double asterisk (∗∗ ), and triple asterisk (∗∗∗ ) denote variables significant at 10%, 5%, 1% and 0.1%,
respectively.
Standard errors are in parentheses below the parameter estimates.

we refer to the short-run as the period ranging from 2009 to 2050; and to the long-run
as the period ranging from 2051 to the end of the 21st century.

Under each scenario, average temperature rises with time. In addition, the later the
projection of the peak for greenhouse gas concentration, the higher the average tem-
perature.

We compare yields predicted by our models using observed weather data from 1991
to 2009, to those predicted under different scenarios. We proceed as follows. In each
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 99

period, yearly yields are predicted by region for every bootstrap runs,11 and then av-
eraged, still by region, over the whole period. It is then also possible to compute the
percentage change from the baseline to the projected scenario.

All variables except weather ones are kept at their mean value of the baseline period,
i.e., the period ranging from 1991 to 2009. Our projections do not account for any
possible technological changes that are likely to occur, and should therefore not be
viewed as a forecasting exercise.

The evolution of annual wheat and corn yields up to 2100 are plotted for each scenario
in figs. 3.3 and 3.4. Results at the European scale of the percent change in yields of both
wheat and corn under each scenario are reported in table 3.5, while figs. 3.5 and 3.6
show the spatial variability of projected changes.

On average, our scenarios exhibit small gains in terms of annual wheat yields, at the
European scale. At a finer scale, results are more mitigated and heterogeneous changes
in yields are found. In the north, gains are relatively small, close to zero, and tend
to decrease in the long run. In the worst-case scenario, the RCP 8.5, annual wheat
yields fall after the middle of the 21st century. Some regions, mainly eastern ones
display losses under the tested climate scenarios. Gains in average yields are becoming
lower with time, ranging from +0.95% in the short run for the RCP 6.0 scenario to
even become negative in the long-run in the case of the doom-and-gloom RCP 8.5
scenario, with a loss of 1.74%. Results are more encouraging for regions in the south
of Europe, where annual wheat yields would increase under each climate scenario
envisaged. In addition, these gains are becoming even larger in the long-run, reaching
+6% compared to annual yields obtained in the benchmark period. This result can
be explained due to rising precipitation in winter under each scenario coupled with
decreasing total rainfall in summer, which are both favourable conditions for wheat
yields, according to the estimation results. However, the gains in average annual yields
are associated with relatively large variance.

Results for corn yields also show spatio-temporal disparities. In the north, corn yields
would globally increase in the short-run, ranging from +1.04% under the most optim-
istic scenario, the RCP 2.6, to +0.62% under the most pessimistic one, the RCP 8.5. In
the long-run, however, climate conditions envisaged within the scenarios are not as
hopeful. In fig. 3.4, we observe a positive trend for corn yields up to a peak whose date
differs depending on the scenario. After this peak, corn yields in northern Europe tend
11 Recall that the estimates of each model rely on 1, 000 bootstrap runs.
100 Part II. Climate Change in Developed Countries

to gradually decrease. Even if corn yields under the RCPs 2.6 scenario remain close to
their historical average of 1991–2009, they end-up below that average at the end of the
century. Moreover, under the remaining scenarios, the RCPs 4.5, 6.0 and 8.5, projected
climate values are such that corn yields substantially decrease in the long run to reach
up to an average decline of −2.65% under the RCP 8.5 scenario. Results for south-
ern regions are even more alarmist. In fact, with the exception of the optimistic RCP
2.6 scenario, where average annual corn yields would be small gains (+0.58%), every
other scenario leads to a drop in annual corn yields. However, in the short-run, those
losses are close to zero. But in the long-run, the drop in yearly annual corn yields are
more substantial, ranging from −4.40% in average under the most optimistic RCP 2.6
scenario to −19.88% in the case of the pessimistic RCP 8.5 scenario.

All in all, our scenarios exhibit a clear distinction between northern and southern
regions. This distinction is also found in Van Passel et al. (2016), which projects the
sensitivity of European farms to climate, looking at how land value reacts to a changing
climate. They find changes in land value ranging from +5 to −32% by 2100 depending
on the climate scenario. They predict farms in Southern Europe to be particularly
sensitive, suffering losses of −5% to −9% per degree Celsius. Our results for projected
corn yields are similar in terms of geographical effects. We find opposite effects for
wheat yields. But our analysis, contrary to the Ricardian model used in Van Passel et al.
(2016), focuses on a specific crop and thus does not incorporate the possibility for the
farmer to switch crop if the weather conditions become too harmful. However, we
find regional differences between western and eastern regions as in Vanschoenwinkel
et al. (2016), although our sample includes less regions in eastern Europe.

6 Conclusion
This chapter presents an assessment of the effects of climate change on European
agriculture. We employ an empirical analysis on data from 1992–2009 in European
regions. Annual yields are modelled as a function of weather variables, production
prices and control variables.

We find significant effects of climate variation both on the mean and the variance of
wheat and corn yields. Our findings show that variation of prices have no significant
impacts on wheat yields before the 1999 CAP reform in the northern regions of Europe,
i.e., regions in which production is relatively more important. For corn, prices are also
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 101

RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

72.5

North
70.0

67.5

42

40

South
38

36

34
2010 2030 2050 2070 2090 2010 2030 2050 2070 2090 2010 2030 2050 2070 2090 2010 2030 2050 2070 2090
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Note: The graphs show the evolution up to 2100 of predicted annual wheat yields for each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).
The values are aggregated for northern regions (top) and for southern regions (bottom). Loess estimates of the predicted yields
are represented by the dashed lines and accompanied by a 95% confidence interval. The dotted horizontal lines represent the
average value of historical predicted wheat yields (1991–2009).

Figure 3.3: Predicted Wheat Yields (100kg per ha) Under Different Climate Scenarios

RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

96
North
92

88

84
100

80
South

60

40
2010 2030 2050 2070 2090 2010 2030 2050 2070 2090 2010 2030 2050 2070 2090 2010 2030 2050 2070 2090
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Note: The graphs show the evolution up to 2100 of predicted annual corn yields for each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).
The values are aggregates for northern regions (top) and for southern regions (bottom). Loess estimated of the predicted yields
are represented by the dashed lines and accompanied by a 95% confidence interval. The dotted horizontal lines represent the
average value of historical predicted corn yields (1991–2009).

Figure 3.4: Predicted Corn Yields (100kg per ha) Under Different Climate Scenarios
102 Part II. Climate Change in Developed Countries

Percentage Change in Yield

-10 -5 0 5 10
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

(2009-2050)
Short-term
(2051-2100)
Long-term
Notes: Values in each region represent the percent change in average predicted wheat yields over the short run period (2009–2050,
upper panels) or the long run period (2051–2100, lower panels) compared to average predicted wheat yields over the baseline
period (1991–2009).

Figure 3.5: Regional Changes in Wheat Yields Under Different Climate Scenarios

Percentage Change in Yield

-20 0 20
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5
(2009-2050)
Short-term
(2051-2100)
Long-term

Notes: Values in each region represent the percent change in average predicted corn yields over the short run period (2009–2050,
upper panels) or the long run period (2051–2100, lower panels) compared to average predicted corn yields over the baseline period
(1991–2009).

Figure 3.6: Regional Changes in Corn Yields Under Different Climate Scenarios
Chapter 3. Climate Change and Agricultural Yields: an European Case Study 103

Table 3.5: Effect of Climate Change on Wheat and Corn Yield (percentage change)

Climate scenarios Temporal Range Model Wheat Yield Corn Yield

0.82 1.04
North
Short-term (1.10) (0.55)
(2009-2050) 3.46 0.58
South
(2.42) (2.47)
RCP 2.6
0.47 −0.12
North
Long-term (1.09) (0.87)
(2051-2100) 3.61 −4.40
South
(2.21) (2.33)

0.84 0.28
North
Short-term (0.89) (0.50)
(2009-2050) 2.53 −3.94
South
(1.71) (1.80)
RCP 4.5
0.14 0.17
North
Long-term (0.96) (1.15)
(2051-2100) 5.08 −7.15
South
(2.87) (3.84)

0.95 0.40
North
Short-term (1.19) (0.56)
(2009-2050) 4.45 −0.57
South
(2.12) (1.71)
RCP 6.0
0.20 −0.91
North
Long-term (1.12) (1.05)
(2051-2100) 5.50 −8.25
South
(3.15) (3.93)

0.31 0.62
North
Short-term (1.13) (0.63)
(2009-2050) 3.88 −3.34
South
(2.08) (2.36)
RCP 8.5
−1.74 −2.65
North
Long-term (1.21) (1.95)
(2051-2100) 6.05 −19.88
South
(3.10) (8.74)

Notes: Short-term and long-term refer to climate conditions over 2009–2050 and 2051–2100, respectively. Numbers in parentheses
are standard deviations.

positively linked to yields in northern regions, but the PAC reform has not significantly
changed this relationship.

Four different climate scenarios were proposed to observe changes in wheat and corn
annual yields. These scenarios reflect the projections of greenhouse gas concentration
until the end of the 21st century. They exhibit both spatial and temporal heterogeneity.
Overall, annual wheat yields rise under each scenario, compared to yields predicted by
our estimations based on historical climate values. Over time, these gains get higher in
the south but approach zero in the north. Projected corn yields under the four climate
104 Part II. Climate Change in Developed Countries

scenarios tested are less optimistic. Even if some small gains in corn yields would be
experienced in the short-run in northern regions, they would become losses in the
long-run. Losses would be even higher for regions in southern Europe.
Chapter 4

Climate Change and Business

Cycles

Joint work with Gauthier Vermandel (Paris Dauphine University)

1 Introduction
The prospect of considerable climate change and its potentially large impacts on eco-
nomic well-being are central concerns for the scientific community and policymakers.
Along with a forecast increase in global mean temperature of 1 to 4 degrees Celsius
above 1990 levels, the Intergovernmental Panel on Climate Change (IPCC) forecasts a
rise in both variability and frequency of extreme events, such as droughts (Edenhofer
et al., 2014). The intensification of extreme drought events is currently emerging as
one of the most important facets of global warming, which may have large macroeco-
nomic implications, particularly for the agricultural sector.

Many efforts have been made to assess the potential economic impact of climate change
(Nordhaus, 1994; Tol, 1995; Fankhauser and Tol, 2005), especially its consequences on
agricultural systems (Adams et al., 1998; Fischer et al., 2005; Deschênes and Green-
stone, 2007). As climatic factors enter into the production function as direct inputs, any
important variation in weather conditions has a large effect on agricultural produc-
tion. From a policymaker perspective, the evaluation of the economic costs incurred
from climate shocks has become a crucial element in the decision-making process to
implement measures that would offset potential harmful effects on the economy and

105
106 Part II. Climate Change in Developed Countries

in turn on social welfare. These very specific macroeconomic costs, generated by vari-
able weather conditions, are particularly challenging for agriculture-based economies,
as well as for developing countries, and may undermine world food security (Eden-
hofer et al., 2014).

Given the remaining uncertainties around the economic costs of variable weather con-
ditions, the main objective of this chapter is to provide a quantitative evaluation of
the effects of weather shocks on the business cycles of an economy. We develop
an original dynamic stochastic general equilibrium model (DSGE) that incorporates a
weather-sensitive agricultural sector into the canonical New Keynesian model. Then,
we apply state-of-art Bayesian techniques to determine the implications of weather
shocks on business cycles. In the recent literature, only a few papers examine the link
between macroeconomic variables and environmental economics using DSGE models
in an RBC framework (Fischer and Springborn, 2011; Heutel, 2012; Dissou and Karn-
izova, 2016) or a New Keynesian setup (Annicchiarico and Di Dio, 2015). While these
studies focus on optimal environmental policies, this study investigates the implica-
tions of weather shocks on short-run macroeconomic fluctuations. Once estimated,
the model is amenable to the analysis of climate change. As climate is assumed to be
a stationary process in our study, an analysis of changes in the mean of the climate
variable is irrelevant. However, changes in the variance of the climate variable and
the underlying impacts on the business cycles can be examined.

This analysis uses data from a small open economy, New Zealand. The reasons for this
choice of country are threefold. First, the agricultural sector represents a substantial
part of the country’s output (around 7% according to the World Bank), making New
Zealand highly dependent on weather conditions. In particular, Buckle et al. (2007)
underline the importance of weather variations as a source of aggregate fluctuations,
along with international trade price shocks, using a structural VAR model for New
Zealand. Second, New Zealand is small enough to be subject to relatively homogen-
eous climatic conditions over its land area.1 Third, the exercise requires an important
number of time series, some of which are not available for developing economies.

To describe New Zealand’s economic situation, we develop a standard small open eco-
nomy model, as in Galí and Monacelli (2005), which is characterized by utility max-
imizing households that consume domestically and imported goods with some degree
1
We attempted a similar exercise for a developing economy, namely, India, but the result was not as
clear, as India’s land area is too vast, resulting in strongly smoothed extreme climate events. Providing
a robust analysis for large economies requires a regional approach that is too challenging for a DSGE
model.
 Chapter 4. Climate Change and Business Cycles 107

of nominal rigidity on good prices. We depart from this framework by splitting pro-
duction into two contrasting sectors: the agricultural sector and the non-agricultural
sector. While the latter is standard, the originality of this study lies in the introduction
of an agricultural sector facing exogenous weather inputs affecting both agricultural
production function and demand for intermediate goods.

Once estimated consistently with the empirical exercise of Buckle et al. (2007), our
DSGE model reveals that weather shocks play an important role in explaining macroe-
conomic fluctuations over the sample period. Drought events are followed by a drop
in agricultural output, as in Kamber et al. (2013), and they are responsible for persistent
variations in agricultural prices and total output. Our framework, however, highlights
the absence of anticipation of weather shocks by farmers one quarter (or more) ahead.
Furthermore, the decomposition of the forecast error variance shows that weather
shocks remarkably drive the variance in agricultural production and prices. For ag-
gregate output and total demand, the main drivers are, unsurprisingly, supply and de-
mand side factors, respectively; weather shocks only play a modest role in explaining
the fluctuations. However, this role grows over time. Moreover, the historical de-
composition of business cycles shows that large drought events that occurred in New
Zealand in 2008, 2010, and 2013 contributed negatively to output fluctuations and were
accompanied by an important inflation increase that clears imbalances in the market
for agricultural goods. Finally, in an attempt to investigate the potential impact of
climate change on aggregate fluctuations, we alter the variance of the weather shock
according to four different climate scenarios. Our results show an increase in the vari-
ability of key macroeconomic variables, such as output prices, production or inflation,
in all but the optimistic scenario. Specifically, in the best-case scenario, where the
standard deviation of the weather shock diminishes by 5.29%, the standard deviations
of agricultural output and prices decrease by 2.18% and 1.63%, respectively. In the
worst-case scenario, which is characterized by a 20.46% increase in the standard de-
viation of the weather shock, the standard deviation of agricultural output and prices
rises by 11.47% and 9%, respectively.

The remainder of this chapter is organized as follows: section 2 sketches the dy-
namic stochastic general equilibrium model. Section 3 presents the estimation of the
DSGE model. Section 4 discusses the propagation of a weather shock, assesses the
consequences of a drought and its implication in terms of business cycles statistics,
presents the historical variance decomposition of the main aggregates (gross domestic
product, agricultural production, agricultural production price inflation), and provides
108 Part II. Climate Change in Developed Countries

a quantitative assessment of the implications of weather shocks under different climate

projection scenarios for aggregate fluctuations. Section 5 concludes.

2 The Model
This section is devoted to a formal presentation of the DSGE model. Our model is
two-sector, two-good economy in a small open economy setup with standard New
Keynesian nominal frictions and a flexible exchange rate regime.2 The home economy,
i.e., New Zealand, is populated by households, intermediate goods and final goods ag-
ricultural and non-agricultural firms, and a central bank. Intermediate producers in
each sector enjoy market power to maximise their profits and produce differentiated
goods. Final goods producers use a packing technology to aggregate both home and
foreign intermediate goods to produce a homogeneous good sold to households. The
final product is a composite of domestically produced and imported goods, thus cre-
ating a trading channel adjusted by the real exchange rate. Nominal rigidities in the
agricultural and non-agricultural sectors generate inflation dynamics that are damped
by the central bank though the adoption of an inflation targeting regime. This section
presents the main components of the model.3

2.1 Households

There is a continuum j ∈ [0, 1] of identical households that consume, save and work in
intermediate goods firms. The representative household maximises the welfare index
expressed as the expected sum of utilities discounted by β ∈ (0, 1):

∑
∞ [ ( )]
1 σC − 1 1+σH
Et β τ
(Cjt+τ − hCt−1+τ )1−σC exp h , (4.1)
τ =0
1 − σC 1 + σH jt+τ
2
Our small open economy setup includes two countries. The home country (here, New Zealand)
participates in international trade but is too small compared to its trading partners to cause aggregate
fluctuations in world output, price and interest rate. The foreign country, representing most of the
trading partners of the home country, is thus not affected by macroeconomic shocks from the home
country, but its own macroeconomic developments affect the home country through the trade balance
and the exchange rate.
3
A list of symbols used in this chapter is provided in appendix C, in tables C.1 to C.4, for the house-
hold bloc, the intermediate goods firms block, the final goods firms block, and the remaining equations,
respectively.
 Chapter 4. Climate Change and Business Cycles 109

where variable Cjt is the consumption index, h ∈ [0, 1] is a parameter that accounts for
external consumption habits, hjt is a labour effort index for the agricultural and non-
agricultural sectors, and σC and σH represent consumption aversion and labour dis-
utility, respectively. Following the seminal contribution of Smets and Wouters (2007),
households preferences are assumed to be non-separable in consumption, so an in-
crease in hours worked has a positive effect on the marginal utility of consumption.4

The representative household allocates total consumption Cjt between two types of
consumption goods produced by the non-agricultural and agricultural sectors denoted
N
Cjt and Cjt
A
respectively. The CES consumption bundle is determined by:

[ 1 µ−1 1
] µ
µ−1 µ−1
Cjt = (1 − φ) µ (Cjt
N µ
) A µ
+ φ µ (Cjt ) , (4.2)

where µ ≥ 0 denotes the substitution elasticity between the two types of consumption
goods, and φ ∈ [0, 1] is the fraction of agricultural goods in the household’s total con-
sumption basket. The corresponding consumption price index thus reads as follows:
1
PtC = [(1 − φ) (Py,t
N 1−µ
) ) ] , where Py,t
A 1−µ 1−µ
+ φ(Py,t N
and Py,t
A
are the final prices of
non-agricultural and agricultural goods, respectively.5

Following Iacoviello and Neri (2010), we introduce imperfect substitutability of labour

supply between the agricultural and non-agricultural sectors to explain co-movements
at the sector level by defining a CES labour disutility index:
[ ( ) ( A )1+ι ]1/(1+ι)
1+ι
hjt = n hN
jt + (1 − n) hjt . (4.3)

The labour disutility index consists of hours worked in the non-agricultural sector hN
jt
and agriculture sector hA
jt , with n denoting the relative share of employment in the
non-agricultural sector. Reallocating labour across sectors is costly and is governed
by the substitutability parameter ι ≥ 0.6
4
We refer the reader to Greenwood et al. (1988) for a discussion of the implications of non-separable
preferences on business cycles.
5
Demand for each type of final good is a fraction of the total consumption index adjusted by its
( N )−µ ( A )−µ
relative price: CjtN
= (1 − φ) Py,t /PtC Cjt and CjtA
= φ Py,t /PtC Cjt .
6
If ι equals zero, hours worked across the two sectors are perfect substitutes, leading to a negative
correlation between the sectors that is not consistent with the data. Positive values of ι capture some
degree of sector specificity and imply that relative hours respond less to sectoral wage differentials.
110 Part II. Climate Change in Developed Countries

Expressed in real terms and dividing by the consumption price index PtC , the budget
constraint for the representative household can be represented as:

∑ ∗ N
Py,t
Πjt Rt−1 Rt−1 ∗ ∗
χS wts hsjt + + b jt−1 + e t b = C jt + b jt + e t b + et ΦB (b∗jt ).
s=N,A PtC πtC πtC jt−1 jt
PtC
(4.4)
The income of the representative household is made up of labour income with a real
wage wts in each sector,7 profits Πjt generated by imperfect competition in goods, and
real riskless domestics bonds bjt and foreign bonds b∗jt . Domestic and foreign bonds
∗
are remunerated at a domestic Rt−1 and a foreign Rt−1 , respectively, nominal gross
interest rates decided by central banks of each country and adjusted by the domestic
inflation rate πtC = PtC /Pt−1
C
. Household’s foreign bonds purchases are affected by the
nominal exchange rate et (an increase in et can be interpreted as an appreciation of the
domestic exchange rate). The household’s expenditure side includes its consumption
basket Cjt , bonds and risk-premium cost Φ(b∗jt )=0.5χB (b∗jt )2 paid in terms of domestic
N 8
final goods at a market price Py,t . Parameter χB > 0 denotes the magnitude of the
cost paid by domestic households when purchasing foreign bonds.

The first-order conditions solving the household’s optimization problem are obtained
by maximizing welfare index in eq. (4.1) under the budget constraint in eq. (4.4) given
the labour sectoral re-allocation cost in eq. (4.3). First, the marginal utility of con-
sumption is determined by:9

σC − 1 1+σH
λct = exp( h ) (Cjt − hCt−1 )−σC . (4.5)
1 + σH jt

The first-order condition determines the household labour supply in each sector:
( )ι
N
n hjt
wtN = hσjtH (Cjt − hCt−1 ) , (4.6)
χN hjt
( )ι
σH (1 − n) hA
jt
A
wt = hjt (Cjt − hCt−1 ) . (4.7)
χA hjt
7
Real labour income is affected by χs > 0, a sector-specific shift parameter that allows us to calibrate
the steady state of hours worked in each sector. This is a common assumption in real business cycle
models.
8
This cost function aims at removing a unit root component that emerges in open economy models
without affecting the steady state of the model. See Schmitt-Grohé and Uribe (2003) for a discussion of
closing open economy models.
9
In equilibrium, the marginal utility of consumption equals the Lagrange multiplier λct associated
with the household budget constraint.
 Chapter 4. Climate Change and Business Cycles 111

where wtN and wtA are the real wages in the non-agricultural sector and the agricultural
sector, respectively.

The Euler condition on domestic bonds that determines the optimal consumption path
is: { }
λct+1 1 1
βEt { C } = . (4.8)
λt Et πt+1
c
Rt

Finally, the Euler condition on foreign bonds, after substituting the Lagrange mul-
tiplier, can be expressed as the real exchange rate determination under incomplete
markets: { }
et+1 Rt ∗
Et = ∗
(1 + χB pN
y,t bjt ), (4.9)
et Rt
where pN
y,t = Py,t /Pt denotes the relative price of final goods with respect to the
N C

consumption price index.

We define the real exchange rate as the ratio of final goods prices expressed in a com-
mon currency:
PtC∗
rert = et , (4.10)
PtC
where PtC∗ denotes the foreign price.

2.2 Final Goods Firms

The firm block is populated by two groups of agents: intermediate goods firms and
final goods firms. Intermediate goods firms produce differentiated goods i ∈ [0, 1],
decide on labour on a perfectly competitive inputs market, and set prices according
to a Rotemberg (1982) technology. Final goods producers act as goods bundlers by
combining national and foreign intermediate goods to produce a homogeneous non-
tradable final good that will be sold to domestic households.

In each sector s = {N, A}, where N and A denote the non-agricultural and agri-
cultural sectors, respectively, we assume that the production of the final good is per-
formed as in Rabanal and Tuesta (2010). A continuum of final goods firms purchases a
composite of intermediate home goods Xts , and a composite of intermediate foreign-
produced goods Xts∗ to produce a differentiated final good product Yts using the fol-
lowing CES technology:
[ 1
] (µµ−1)
s
1 (µs −1) (µs −1) s
Yts = (1 − αs ) µs (Xts ) µs + αs (Xts∗ )
µs µs , (4.11)
112 Part II. Climate Change in Developed Countries

where αs denotes the share of foreign-produced goods that are used for the production
of the final good, and µs is the elasticity of substitution between domestically produced
and imported intermediate goods in both countries. A value of αs = 0 implies the
autarky of this market, while αs < 0.5 reflects a home bias in the preferences of
firms. The corresponding sectoral production price index is given by: Py,t
s
= [(1 −
αs )(Pts )1−µs + αs (et Pts∗ )1−µs ]1/(1−µs ) , for s = {N, A}.

The composite intermediate goods for the non-agricultural sector bought at home and
∫n ∫n
abroad are XtN = ( 0 (XitN )(ϵN −1)/ϵN di)ϵN /(ϵN −1) and XtN ∗ = ( 0 (XitN ∗ )(ϵN −1)/ϵN di)ϵN /(ϵN −1) ,
∫1
while for the agricultural sector, we have XtA = ( n (XitA )(ϵA −1)/ϵA di)ϵA /(ϵA −1) and
∫1
XtA∗ = ( n (XitA∗ )(ϵA −1)/ϵA di)ϵA /(ϵA −1) . For each sector s = {A, N }, a parameter
ϵs > 1 is the elasticity of substitution between the types of intermediate goods. Since
goods are imperfect substitutes, firms are able to deviate from the perfectly compet-
itive equilibrium by imposing a margin on their selling prices. The packing activity
∫n
delivers the following price index for each sector: PtN = [1/n 0 PitN 1−ϵN di]1/(1−ϵN )
∫1
and PtA = [1/(1 − n) n PitA 1−ϵA di]1/(1−ϵA ) , ensuring that profits are always zero.

At the optimum, the demand functions for home and foreign goods produced in sector
s are given by:
( )−µs ( )−ϵs ( )−µs ( s∗ )−ϵs
Pts Pits Pts∗ Pit
Xits = (1 − αs ) s
Yts and Xits∗ = αs et s Yts .
Py,t Pts Py,t Pts∗

(4.12)

2.3 Intermediate Goods Firms

In each sector, intermediate goods firms produce an intermediate goods that is used
by final goods firms to produce the homegeneous non-tradable final good.

2.3.1 Agricultural Production and Weather Variability

To investigate the implications of weather variations as a source of aggregate fluc-

t , that captures
tuation, we introduce into the model a weather variable, denoted εW
variations in soil moisture affecting the production process of farmers. The measure
we use is based on soil moisture deficit observations calculated from the daily water
balance.10 A positive realization of εW
t depicts a prolonged episode of dryness that
10
The soil moisture variable measures the net impact of rainfall entering the pasture root zone in the
soil, which is then lost from this zone as a result of evapotranspiration or use of water by plants.
 Chapter 4. Climate Change and Business Cycles 113

damages agricultural output and generates inflation pressures. We assume that the
aggregate drought index follows a stochastic exogenous process driven by two shocks:

t = (1 − ρW ) + ρW εt−1 + ηt + η̃t−1 ,
εW ρW ∈ [0, 1), (4.13)
W W W

The first shock, denoted ηtW , is a traditional shock to the real business cycle that im-
pacts the level of soil moisture in the same period in which farmers see it. The second,
W
η̃t−1 , is a news shock and is differentiated from the former in that farmers observe
a weather news shock in advance (here, one quarter).11 Thus, this shock allows us
to evaluate whether farmers are anticipating drought events one quarter in advance
by capturing macroeconomic fluctuations one quarter before the realization of the
weather shock.12

To bridge weather variations with business cycle fluctuations, we define a damage

variable dt determining how variable weather conditions log(εW
t ) may induce inertial
aggregate fluctuations:

dt = ρd dt−1 + log(εW
t ), ρd ∈ [0, 1), (4.14)

where ρd captures some persistence of damage after an adverse drought event shock.
Here, it is important to disentangle the parameter ρW from eq. (4.13) and ρd from
eq. (4.14): the autoregressive component ρW captures the estimated persistence of a
drought shock, while ρd catches the persistence of its damages. The main underlying
motivation is that damages to the economy might be more persistent than the weather
shock itself, as showed by the VAR models.13

The production component of agriculture is strongly inspired by Restuccia et al. (2008)

to the extent that agricultural output is Cobb-Douglas in land, intermediate inputs, and
labour inputs.14 In addition to this modeling choice, we introduce a damage function
11
We follow the news-driven business cycles literature, as exemplified by Beaudry and Portier (2006),
Barsky and Sims (2011) and Schmitt-Grohé and Uribe (2012), to introduce climate-news shocks as a
source of macroeconomic fluctuation.
12
Anticipating the results from the estimation exercise, we have evaluated the ability of farmers to
expect weather shocks more than one quarter in advance; however, we find evidence that farmers are
not able to predict drought events and that they are rather surprised by weather shocks.
13
We refer to Buckle et al. (2007) and Kamber et al. (2013) for VAR models highlighting the hysteresis
effects of weather shocks on business cycles.
14
We refer to Mundlak (2001) for discussions of related conceptual issues and empirical applications
regarding the functional forms of agricultural production. In an alternative version of our model based
on a CES agricultural production function, the fit of the DSGE model is not improved, and the identi-
fication of the CES parameter is weak.
114 Part II. Climate Change in Developed Countries

ΓX (·) in the spirit of Integrated Assessment Models, which connects the weather to
agricultural output.

Each representative firm i ∈ [n, 1] operating in the agricultural sector has the follow-
ing production function:
(( )1−σ ( )σ )1−ω
XitA = εZt Zitω ΓX (dt , dt−1 ) L̄i κi HitA , (4.15)

where XitA is the production function of the intermediate agricultural good that com-
bines a (fixed) land endowment L̄i for each farmer i, labour demand HitA and non-
agricultural inputs Zit . Production is subject to an economy-wide technology shock
εZt .15 The parameter ω ∈ [0, 1] is the elasticity of output to intermediate inputs,
σ ∈ [0, 1] denotes the share of production/land in the production process of agri-
cultural goods, and κi > 0 is a technology parameter endogenously determined in
the steady state. The economy-wide technology shock εZt affects both agricultural
and non-agricultural sectors by capturing fluctuations associated with declining hours
worked and prices coupled with increasing output.

Agricultural production is tied up with exogenous weather conditions through a dam-

age function ΓX (·) that alters land productivity. This function has a simple form with
one lag aiming at capturing the hump-shaped response of output to weather shock:

ΓX (dt , dt−1 ) = 1 + γ0X dt + γ1X dt−1 , (4.16)

where γ0X , γ1X ∈ (−∞, +∞) are elasticities that are estimated agnostically (i.e., without
tight priors) during the estimation exercise. In our setup, we are interested in the
short-run implications of weather shocks, leaving aside the neutral long-run effects
( )
with ΓX d,¯ d¯ = 1, where d¯ denotes the (zero) deterministic steady state of damages
induced by drought events. The parameter γ1X captures the lagged response of out-
put after drought events. The introduction of this parameter is motivated by the time
prices usually take to adjust to climate shocks, as assumed by Bloor and Matheson
(2010).

In addition to this damage function for output, inputs costs are affected by a similar
function. The real costs paid by farmers read as follows:

wtA HitA + pN
t Zit ΓZ (dt , dt−1 ) , (4.17)
15
Technology is characterized as an AR(1) shock process: εZ t = 1 − ρZ + ρZ εt−1 + ηt with ηt ∼
Z Z Z

N (0, σZ ), where ρA ∈ [0, 1) denotes the AR(1) term in the technological shock process.
 Chapter 4. Climate Change and Business Cycles 115

where wtA is the real wage offered to households hired in the agricultural sector, and
t = Pt /Pt denotes the relative price of intermediate goods, with Pt as the con-
pN N C C

sumer price index. The demand for intermediate goods Zit is affected by ΓZ (dt , dt−1 ),
which aims at capturing extra consumption of intermediate goods following a drought
event. A drought shock increases the feed budget, as dairy cattle require more water as
temperature, humidity and production levels rise. Farming activities also demand more
water to offset soil dryness by increasing field irrigation. This damage function cap-
tures the demand effects in the intermediate sector, and the shape of this damage func-
tion reads as in eq. (4.16) with different elasticities denoted γ0Z and γ1Z ∈ (−∞, +∞).

To introduce nominal rigidities, we assume that firms must solve a two-stage problem.
In the first stage, the real input price wtN is taken as given, firms rent inputs HitN and
Zit in a perfectly competitive factor market in order to minimize costs subject to the
production constraint. Each firm maximises profits:

max mcA it Xit − wt Hit − ΓZ (dt , dt−1 ) pt Zit

A A A N

{Zit ,Hit }
N

under the supply constraint in eq. (4.15). The variable mcA

it denotes the real marginal
cost of producing an additional agricultural good.

The cost-minimization problem ensures that the real agricultural wage is directly
driven by the marginal product of labour:

XtA
t (1 − ω) σ
wtA = mcA . (4.18)
HtA

The second cost-minimizing condition is obtained from the marginal product of in-
termediate consumption Zt and provides the optimal demand for intermediate goods
from the farmer:
mcA
Zt = ω t
N
XtA . (4.19)
ΦZ (dt , dt−1 ) pt

In the second stage, the intermediate goods firm operates monopolistically and sets the
retail price according to a Rotemberg (1982) technology. Intermediate goods firms face
adjustment costs with price changes ACitA defined according to: ACitA = 0.5κA (PitA /Pit−1
A
−
A
(πt−1 )ξA )2 , where κA is the cost of adjusting prices, and ξA is the coefficient that
measures the rate of indexation to the past rate of inflation of intermediate goods,
A
πt−1 A
= Pt−1 A
/Pt−2 . These costs are paid in terms of the final goods at a market price of
N
Py,t . Given this price adjustment cost specification and replacing the demand function
116 Part II. Climate Change in Developed Countries

for final goods firms, the problem of the representative firms becomes dynamic:
[ ( )−ϵA ]
∑
+∞ c
λt+τ τ A
Pit+τ A
Pit+τ
Et β A
Xt+τ − mcA
it+τ Xit+τ − py,t+τ Yt+τ ACit+τ .
A N A A

τ =0
λct C
Pt+τ A
Pt+τ
(4.20)
The variables mcA
it and py,t are the real marginal cost and the relative price of non-
A

agricultural final goods. Since firms are owned by households, they discount expected
profits using the same discount factor as households (β τ λct+τ /λct ).16 Anticipating sym-
metry between firms with PtA = PitA , the first-order condition is:

YtA ( A ) A
t + ϵA mct − py,t
(1 − ϵA ) pA A N
κA π t − (π A
t−1 )ξA
πt
XtA
{ c A (
}
λt+1 N Yt+1 ) A
+ κA βEt p π − (πt ) πt+1 = 0. (4.21)
A A ξA
λct y,t+1 XtA t+1

2.3.2 Non-Agricultural Intermediate Production

Each representative intermediate goods firm i ∈ [0, n] has the following technology:

XitN = εZt HitN , (4.22)

where XitN is the production of the ith intermediate goods firm that combines labour
demand Hit and technology εZt .

Intermediate goods producers solve a two-stage problem. In the first stage, the real
input price wtN is taken as given, and these firms rent inputs HitN in a perfectly com-
petitive factor markets in order to minimize costs subject to the production constraint:

[ N ]
it Xit − wt Hit + λt Xit − εt Hit .
max mcN N N N n Z N
{Xit
N ,H N }
it

The first-order condition leads to i) mcN

it = λt and ii) the real marginal cost expres-
n

sion:
wtN
mcN N
it = mct = . (4.23)
εZt

In the second stage, the intermediate goods firm operates monopolistically and sets the
retail price according to a Rotemberg (1982) technology. Intermediate goods firms face
16
The stochastic discount factor is endogenously determined by the Euler condition of households.
In equilibrium, the stochastic discount is inversely related to the real interest rate.
 Chapter 4. Climate Change and Business Cycles 117

adjustment costs on price changes ACitN defined according to ACitN = 0.5κN (PitN /Pit−1
N
−
) ) , where κN is the cost of adjusting prices, and ξN is the coefficient that
N ξN 2
(πt−1
measures the rate of indexation to the past rate of inflation of intermediate goods
N
πt−1 N
= Pt−1 N
/Pt−2 . These costs are paid in terms of final goods at a market price Py,t
N
.
Given this price adjustment cost specification, the problem of the representative firms
becomes dynamic:
[ ( )−ϵN ]
∑
+∞ c
λt+τ τ N
Pit+τ N
Pit+τ
Et β N
Xt+τ − εN
t+τ mcit+τ Xit+τ − py,t+τ Yt+τ ACit+τ ,
N N N N N

τ =0
λct C
Pt+τ N
Pt+τ
(4.24)
where εN
t is an AR(1) markup shock that aims at capturing the external factors driving
the inflation rate, which are not included in the model such as commodity prices.

Anticipating symmetry between firms with PtN = PitN , the first- order condition reads:

N Yt
N ( ) N
(1 − ϵN ) pN
t + ϵ ε
N t
N
mc N
t − p y,t κ πtN − (πt−1
N N
N ξN
) πt
Xt
{ c N (
}
λt+1 N Yt+1 ) N
+ κN βEt p π − (πt )
N N ξN
πt+1 = 0. (4.25)
λct y,t+1 XtN t+1

2.4 Monetary Policy

The central bank reacts to fluctuations in price, activity and external imbalance. The
general expression of the linear interest rule implemented by the central bank can be
expressed as:

( )1−ρ [ ](1−ρ) ( )
D ϕ∆Y R
Rt = R̄ (Rt−1 )ρ (πt )ϕπ (RERt )ϕE YtD /Yt−1 εt , (4.26)

where R̄ is the steady-state interest rate, YtD is gross domestic product, εR

t is an exo-
genous AR(1) monetary policy shock,17 ϕπ , ϕE and ϕ∆Y denote inflation, real ex-
change rate and GDP growth gap parameters, respectively, that aim to stabilize the
economy when it deviates from its steady state. In a small economy context, we fol-
low the definition of monetary policy rules in open economies of Clarida et al. (1998)
and estimate ϕE . A positive value of ϕE induces a reduction in the variance of the real
exchange rate.
17
The monetary policy shock follows a standard AR(1) stochastic process: εR R
t = (1−ρR )+ρR εt−1 +
t ,
εR with ηt ∼ N (0, σR ), and 0 ≤ ρR < 1 the autoregressive term.
R
118 Part II. Climate Change in Developed Countries

2.5 Foreign Economy

Our foreign economy is characterized by a set of five equations that aims at capturing
the standard business cycle patterns of the foreign economy. Four equations are taken
from the standard New Keynesian framework, namely, the Phillips curve, the IS curve,
the Taylor rule and the CES substitution curve between two types of goods. These
equations provide the structural relations between aggregate output Yt∗ , agricultural
output YtA∗ , inflation πt∗ and the nominal interest rate Rt∗ . Most of the parameters and
the steady state are symmetric between domestic and the foreign economy for clarity
purposes.

The foreign inflation rate is determined by the firm’s price setting equation under
Rotemberg price adjustment costs:

{( ) ∗ }
(1 − ϵN ) + ϵN χ∗ Yt∗ − κ∗ (πt∗ − 1) πt∗ + κ∗ βEt ∗
πt+1 − 1 πt+1 = 0. (4.27)

Foreign non-agricultural output is determined by the following Euler equation:

{ }
Yt∗ 1 εt
βEt ∗ ∗
= (4.28)
Yt+1 πt+1 Rt∗

where εYt ∗ is a demand shock characterized by an iid AR(1).

The third relation is the Taylor rule, which is analogous to eq. (4.26):

( ) ∗
∗ 1−ρ
( ∗ )ρ∗ ( ∗ n ( A∗ A∗ )1−n ϕ∗ ∗ ∗ ϕ∗ )(1−ρ∗ )
Rt∗ = R̄ Rt−1 ((πt ) pt /pt−1 ) π (Yt /Ȳ ) y . (4.29)

In this expression, ρ∗ is the autocorrelation parameter, ϕ∗π is the elasticity of the nom-
inal interest rate to the inflation rate, and ϕ∗y is the elasticity of the nominal interest
t /pt−1 denotes the variation of the relative price
rate to the output gap. Expression pA∗ A∗

index of agricultural goods weighted by the size of the agricultural sector 1-n.

The fourth equation determines the demand for agricultural goods by foreign house-
holds. This equation is a reduced-form equation of eq. (4.2), modeling households pref-
erences by substituting agricultural and non-agricultural goods via:
( )−µ
YtA∗ φ pA∗
t
∗
= ,
Yt 1−φ A∗
pt−1
 Chapter 4. Climate Change and Business Cycles 119

where pA∗
t is the relative price of agricultural goods, parameter φ is the share of
agricultural goods in the consumption basket, and µ is the substitution parameter.
This equation shows that the household’s consumption allocation is determined by
the gap between variations in the relative price index between agricultural and non-
agricultural goods.

Finally, the foreign agricultural price is too volatile to be determined by a New Keyne-
sian Phillips curve. We assume the relative price of foreign agricultural goods is de-
termined by an AR(1) shock process:

∗ ∗ A∗ ∗2
t = 1 − ρA + ρA εt−1 + ηt with ηt ∼ N (0, σA ),
εA∗ (4.30)
A∗ A∗

with relative agricultural prices directly driven by the shock pA∗

t = εt .
A∗

In addition, the second exogenous shock affecting the IS curve reads:

εYt ∗ = 1 − ρ∗Y + ρ∗Y εYt−1

∗
+ ηtY ∗ with ηtY ∗ ∼ N (0, σY∗2 ). (4.31)

2.6 Shocks, Aggregation and Equilibrium Conditions

After (i) aggregating all agents and varieties in the economy, (ii) imposing market
clearing on all markets, and (iii) substituting the relevant demand functions, we can
deduct the general equilibrium conditions of the model.

First, total demand for non-agricultural goods is as follows:

( )−µ
N
Py,t
YtN = (1 − φ) Ct + ΦB (b∗jt ) + ACtN YtN + ACtA YtA , (4.32)
PtC

while the equilibrium in the intermediate goods market after aggregation is determ-
ined by:
( )−µN ( )−µN
PtN 1 PtN
nXtN = (1 − αN ) N
YtN + αN N∗
YtN ∗ + (1 − n) Zt , (4.33)
Py,t et Py,t
∫n ∫1
where nXtN = 0
XitN di is the aggregate supply, and (1 − n) Zt = n
Zit di denotes
the aggregate demand for domestic intermediate goods from farmers. In this equation,
the left side denotes the aggregate production of intermediate goods, while the right
120 Part II. Climate Change in Developed Countries

side denotes respectively demands from home and foreign (i.e. imports) final goods
firms, and also demand for intermediate goods from farmers.

Similarly, for the agricultural sector, the aggregate demand is:

( )−µ
A
Py,t
YtA = φ Ct , (4.34)
PtC

and equilibrium in the intermediate market is achieved by the following clearing mar-
ket condition:
( )−µA ( )−µA
PtA 1 PtA
(1 − n) XtA = (1 − αA ) A
YtA + αA A∗
YtA∗ . (4.35)
Py,t et Py,t

Turning to the labour market, the market clearing condition between household labour
supply and demand from firms in each sector is:
∫ 1 ∫ n ∫ 1 ∫ 1

jt dj
hN = HitN di and jt dj
hA = HitA di. (4.36)
0 0 0 n

The law of motion for the total amount of real foreign debt is:
∗
Rt−1 ( ) ( )
b∗jt = C
∆et b∗jt−1 + n pN
t Xt − py,t Yt
N N N
+ (1 − n) pA
t Xt − py,t Yt − pt Zt .
A A A N
πt
(4.37)
Real domestic absorption (YtD ) and aggregate production (YtS ) are given by:

YtD = npN
y,t Yt + (1 − n) py,t Yt ,
N A A
(4.38)
( A A )
YtS = npN
t X t
N
+ (1 − n) p t X t − p N
t Zt . (4.39)

Finally, the general equilibrium condition is defined as a sequence of quantities {Qt }∞

t=0
and prices {Pt }∞ ∞
t=0 such that for a given sequence of quantities {Qt }t=0 and the real-
ization of shocks {St }∞ ∞
t=0 , the sequence {Pt }t=0 guarantees simultaneous equilibrium
on all markets previously defined.

3 Estimation
The model is estimated using 7 time series with Bayesian methods and quarterly data
for New Zealand over the sample period 1989:Q1 to 2014:Q2.18 We estimate the struc-
tural parameters and the sequence of shocks following the seminal contributions of
18
See appendix 4.1 for more details on the series used in the estimation.
 Chapter 4. Climate Change and Business Cycles 121

Smets and Wouters (2007) and Christiano et al. (2005). For a detailed description, we
refer the reader to the original papers.

3.1 Calibration and Prior Distributions

We fix a small number of parameters that are commonly used in the literature of real
business cycle models, including β=0.99, the discount factor; h̄N =h̄A =1, the steady
state hours worked per firm normalized to one; and σH =2, the labour effort disutility.
Regarding the sectoral labour reallocation costs, we fix ι=2.5 (higher than σH ) in order
to obtain a positive correlation between the sectors, as reported by Iacoviello and Neri
(2010). Following Smets and Wouters (2007), the substitution parameters for agricul-
tural goods and non-agricultural goods are ϵN =ϵN =10, implying a quarterly markup
of 11%. Regarding the international business cycle parameters, we employ a calibra-
tion in line with the small open economy literature. The portfolio adjustment cost on
foreign debt is set close to that in Schmitt-Grohé and Uribe (2003), with χB = 0.01.19
The current account is balanced in steady state assuming b̄∗ = ca = 0, and sub-
stitution between home and foreign varieties is set at µN =µA =1.5 for both sectors.
Regarding the openness of the goods market, our calibration is strongly inspired by
Liu (2006), with a share αN of exported non-agricultural goods set to 25% and to 30%
for agricultural goods αA to account for the greater internationalization of the agricul-
tural sector. Turning to agricultural sector parameterization, the share of agricultural
goods is set to φ = 14%, as observed in the consumption basket of New Zealander
households, while the relative share of firms operating in the non-agricultural sector
is fixed to n = 0.93 to obtain an agricultural output-to-GDP ratio that is consistent
with the data. In addition, our calibration of the production function relies on Res-
tuccia et al. (2008) with a land-to-employment ratio L̄i =1.4, an income share of labour
in agriculture σ=0.70, and an intermediate input-to-output ratio ω=0.40. Finally, for
the foreign economy, we select parameters that replicate US business cycles, as it is a
leading trading partner of New Zealand: ρ∗ =0.8, κ∗ = 40, ϕ∗π =1.5 and ϕ∗y =0.125.

The rest of the parameters are estimated using Bayesian methods. Table 4.1 and fig. 4.1
report the prior (and posterior) distributions of the parameters for New Zealand.20
19
The value of this parameter marginally affects the dynamic of the model, but it allows us to remove
a unit root component induced by the open economy setup.
20
The posterior distribution combines the likelihood function with prior information. To calculate
the posterior distribution to evaluate the marginal likelihood of the model, the Metropolis-Hastings
algorithm is employed. We compute the posterior moments of the parameters using a total generated
sample of 400, 000, discarding the first 40, 000, and based on four parallel chains. The scale factor was
set in order to deliver acceptance rates of between 20% and 25%. Convergence was assessed by means of
122 Part II. Climate Change in Developed Countries

Overall, our prior distributions are either relatively uninformative or consistent with
earlier contributions to Bayesian estimations. For a majority of New Keynesian model
parameters, i.e., σC , h, ξN , ξA , ρ, ϕπ , and ϕ∆y , and shock processes parameters, we use
the prior distributions imposed by Smets and Wouters (2007). However, some priors
have been marginally adjusted to match New Zealand’s situation (with more unin-
formative priors). We discuss the prior information that is not taken from Smets and
Wouters (2007). First, since we do not have empirical evidence regarding the substi-
tution parameter on goods µ, we give diffuse prior information on a positive support
characterized by a gamma distribution centered on 10, with a deviation of 2. Second,
sectoral price adjustment costs κN , and κA are assumed to be gamma distributed with
mean 50 and standard deviation 7.50, which corresponds to an average contract dura-
tion of approximately 4.5 quarters in the Calvo model. Third, we evaluate the perspect-
ive of real exchange rate targeting by imposing the corresponding policy weight ϕE on
a prior similar to the output growth weight. Finally, regarding the damage parameter
induced by weather shocks, we adopt an agnostic approach using very uninformative
prior information. Damage parameters γ0X , γ1X , γ0Z and γ1,P , are given a diffuse normal
distribution, with the mean and standard deviation set to 0 and 50, respectively. The
autoregressiveness coefficient ρd of the damage variable is also diffuse, with a beta
distribution with mean 0.50 and standard deviation 0.2.

3.2 Posterior Distribution

In addition to the prior distributions, table 4.1 reports the estimation results that sum-
marize the means and the 5th and 95th percentiles of the posterior distributions, while
the latter are illustrated in fig. 4.1. According to fig. 4.1, the data were fairly inform-
ative, as their posterior distributions did not stay very close to their priors. While
our estimates of the standard parameters are in line with the business cycle literature
(see, for instance, Smets and Wouters (2007) for the US economy or Liu (2006) for New
Zealand), several observations are worth making regarding the means of the posterior
distributions of structural parameters.

Contrasting our results with the estimates of Smets and Wouters (2007) for the US
economy, consumption appears to be less persistent in New Zealand and more elastic
to variations in the real interest rate. Nominal rigidities appear to be stronger in New
Zealand. This outcome is consistent with a small open economy characterized by
the multivariate convergence statistics taken from Brooks and Gelman (1998). We estimate the model
using the dynare package Adjemian et al. (2011).
 Chapter 4. Climate Change and Business Cycles 123

Table 4.1: Prior and Posterior Distributions of Structural Parameters and Shock Pro-
cesses

Prior distributions Posterior distribution

Shape Mean Std. Mean [5%:95%]

SHOCK PROCESS AR(1)

Productivity σN IG 0.1 ∞ 1.99 [1.75:2.16]
Preference σD IG 0.1 ∞ 8.39 [7.66:8.93]
Markup non-agricultural σN IG 0.1 ∞ 4.69 [4.19:5.07]
Foreign demand σY ∗ IG 0.1 ∞ 0.62 [0.54:0.69]
Foreign agricultural prices σA ∗ IG 0.1 ∞ 3.89 [3.56:4.14]
Monetary policy σR IG 0.1 ∞ 0.30 [0.26:0.33]
Weather (surprise) σW IG 0.1 ∞ 0.42 [0.45:0.61]
Weather (news-1q) σ̂W IG 0.1 ∞ 0.21 [0.02:0.24]
Productivity (AR term) ρZ B 0.5 0.2 0.76 [0.74:0.80]
Preference (AR term) ρD B 0.5 0.2 0.82 [0.79:0.87]
Markup non-agricultural (AR term) ρN B 0.5 0.2 0.97 [0.96:0.99]
Foreign demand (AR term) ρ∗Y B 0.5 0.2 0.99 [0.99:1.00]
Foreign agricultural prices (AR term) ρ∗A B 0.5 0.2 0.96 [0.95:0.98]
Monetary policy (AR term) ρR B 0.5 0.2 0.19 [0.14:0.25]
Weather (AR term) ρW B 0.5 0.2 0.33 [0.25:0.40]

STRUCTURAL PARAMETERS
Consumption aversion σC G 1.5 0.37 1.34 [1.15:1.47]
Consumption habits h B 0.7 0.10 0.42 [0.37:0.47]
Price adjustment cost sector N κN G 50 7.5 39.47 [33.51:44.94]
Price adjustment cost sector A κA G 50 7.5 59.14 [48.07:66.77]
Price indexation sector N ξN B 0.5 0.15 0.25 [0.15:0.31]
Price indexation sector A ξA B 0.5 0.15 0.23 [0.11:0.29]
Policy rule smoothing ρ B 0.8 0.05 0.76 [0.73:0.79]
Policy weight - inflation ϕπ G 2 0.25 2.49 [2.27:2.69]
Policy weight - output growth ϕ∆y N 0.15 0.05 0.10 [0.08:0.12]
Policy weight - real exchange rate ϕE N 0.15 0.05 0.02 [0.01:0.02]
Consumption substitution µC G 10 5 2.20 [1.78:2.64]
Climatic damage - inertia ρd B 0.5 0.2 0.98 [0.97:0.99]
Climatic damage - output γ0X N 0 50 -45.15 [-63.37:-26.78]
Climatic damage - output (lag) γ1X N 0 50 9.51 [-2.13:22.76]
Climatic damage - cost γ0Z N 0 50 -20.96 [-29.83:-13.08]
Climatic damage - cost (lag) γ1Z N 0 50 9.16 [3.65:15.23]

Marginal log-likelihood -1123.46

Notes: The column entitled “Shape” indicates the prior distributions using the following acronyms: N describes a Normal distri-
bution, G a gamma one, B a beta one, and IG an inverse gamma one.

weaker competition making prices stickier. In addition, we find that monetary poli-
cymaking had been more oriented toward inflation stabilization in New Zealand than
in the US economy. For real exchange rate targeting, we obtain a value that lies in the
ballpark of estimates obtained from the open economy monetary policy rule literature
of Clarida et al. (1998) (they find ϕE =0.07 for G7 economies). Turning to the damage
parameters, we find that the weather shocks have a strong negative immediate effect
on agricultural output (γ0X <0) and depress demand for intermediate goods (γ0Z <0).
However, the damage fuels demand for intermediate goods one period after the real-
ization of the climate shock (with γ1Z >0) and generates inflation pressures in both
sectors. This lag in the response after damage, modeled by γ1X , γ1Z >0, is explained by
the observed dynamic of agricultural prices, which adjusts to weather shocks after one
period. Finally, we find that the damage process is very inertial (with ρd =0.98), and this
 124 Part II. Climate Change in Developed Countries

σC risk consumption h consumption habits κN price cost non-farm κA price cost farm
·10−2 ·10−2
2 6
6 4
4 4
1 2
2 2
0 0 0 0
0.5 1 1.5 2 2.5 0.2 0.4 0.6 0.8 20 40 60 80 0 20 40 60 80 100120
ξN indexation non-farm ξA indexation farm ρ taylor smoothing φπ inflation weight

4 4
10 1.5
3
2 1
2 5
1 0.5
0 0 0 0
−0.2 0 0.2 0.4 0.6 0.8 −0.2 0 0.2 0.4 0.6 0.8 0.6 0.7 0.8 0.9 1.5 2 2.5 3 3.5
φ∆y gdp growth weight φe rer weight µ CES substitution ρd damage inertia
60 40
0.8
15 30
40 0.6
10 0.4 20
5 20 10
0.2
0 0 0 0
05 · 10−2
[Link].250.3 05 · 10−2
[Link].250.3 0 5 10 15 20 25 30 0.2 0.4 0.6 0.8 1
γ0X output damage γ1X output dam. (lag) γ0P input damage γ1P input dam. (lag)
·10−2 ·10−2 ·10−2 ·10−2
2 3 6
4
1.5 2 4
1 2
0.5 1 2
0 0 0 0
−150
−100−50 0 50 100150 −150
−100−50 0 50 100150 −150
−100−50 0 50 100150 −150
−100−50 0 50 100150

Prior Posterior

Figure 4.1: Prior and Posterior Distributions of Structural Parameters for New Zeal-
and (Excluding Shocks)

results is consistent with the findings of Kamber et al. (2013). These authors find, using
a long-run VAR for New Zealand, that drought shocks have persistent feedback effects
on inflation and output. This pattern of weather-driven business cycles is captured
here by ρd . The estimated standard deviation of surprise weather shocks is much lar-
ger than anticipated weather shocks, thus showing that unanticipated weather shocks
play a larger role in the business cycle.

3.3 Do Weather Shocks Matter?

A natural question to ask at this stage is whether weather shocks significantly explain
part of the business cycle. More broadly, it is also relevant to question whether the
number of lags for the damage functions ΓZ (·) and ΓX (·) are consistent with data. To
that aim, we consider four (nested) models Mq (θ) differing from the functional forms
of the weather costs. Each model is given by:
{
Γs (·) = 1 for q = 0
Mq (θ) = ∑q , s = {X, Z}, (4.40)
1 γq−1 dt+1−q for q > 0
s
Γs (·) = 1 +
 Chapter 4. Climate Change and Business Cycles 125

Table 4.2: Prior and Posterior Model Probabilities

No damage Damage Damage (+1 lag) Damage (+2 lags)

M0 (θ) M1 (θ) M2 (θ) M3 (θ)
Prior probability 1/4 1/4 1/4 1/4
Laplace approximation -1125.30 -1128.59 -1123.45 -1126.98
Posterior odds ratio 1.00 0.037 6.341 0.186
Posterior model probability 0.132 0.005 0.838 0.025

where θ is the vector of the estimated parameters of the model Mq (θ). We focus on
four different versions of the model for q = {0, 1, 2, 3}. M0 (θ) denotes the version
of the model without damages (i.e., weather variations do not incur macroeconomic
fluctuations). The second model M1 (θ) is a version where contemporaneous weather
shocks generate fluctuations. The third version of the model M2 (θ) is the version
previously presented in the study with a damage function including one lag. Finally,
the last version M3 (θ) includes an additional lag.

We estimate these four versions of the model (using the same data and priors). Table 4.2
reports for four different nested models the corresponding data density (Laplace ap-
proximation), posterior odds ratio and posteriors model probabilities, which allow us
to determine the model that best fits the data from a statistical standpoint. Using
an uninformative prior distribution over models, we compute both posterior odds ra-
tios and model probabilities taking the model M0 (θ) without damages as the bench-
mark.21 We conduct a formal comparison between models and refer toGeweke (1999)
for a presentation of the method to perform the standard Bayesian model comparison
employed in table 4.2 for our four models. Briefly, one should favor a model whose
data density, posterior odds ratios and model probability are the highest compared to
other models.

Contrasting the results of our four models, we find that the model M2 (θ) employed in
the present study is preferred to any other model, as its posterior probability is 83.8%.
The result is confirmed in terms of the data density and posterior odds ratio. This result
highlights that weather shocks play an important role in explaining macroeconomic
fluctuations over the sample period. In addition, the lag number of the cost function
(q = 2) employed in the present study is preferred to any other tested lag structure
(q=1 or 3).
21
As underlined by Rabanal (2007), it is important to stress that the marginal likelihood already takes
into account that the size of the parameter space for different models can be different. Hence, more
complicated models will not necessarily rank better than simpler models, and M3 (θ) will not inevitably
be favored to other models.
126 Part II. Climate Change in Developed Countries

Similarly, we have applied the same standard Bayesian model comparison applied
to weather news shocks in order to examine whether farmers are able to anticipate
weather shocks before their realization.22 Estimating different versions of the model
where farmers anticipate a weather shock one, two or eight periods in advance, we
find that anticipated weather shocks account for a very small fraction of weather fluc-
tuations (< 1%) and, more broadly, of business cycle fluctuations. Thereby, this result
shows that farmers are not able to anticipate weather shocks.

4 Weather Shocks as Drivers of Business Cycles

This section discusses the propagation of a weather shock and its implications in terms
of business cycle statistics.

4.1 Propagation of a Weather Shock

In the model, the measure of drought is assumed to be a stochastic exogenous process

driven by two shocks: a standard surprise shock (ηtW ) and an anticipated one (η̃t−1
W
),
which we to refer as the weather news shock. The insight behind the weather news
shock is that farmers may be able to anticipate weather variations one quarter before
they occurs. To evaluate how an average drought event in New Zealand propagates
in the economy, we first report the simulated Bayesian system responses of the main
macroeconomic variables following a standard weather shock in fig. 4.2 and a weather
news shock in fig. 4.3. The impulse response functions (IRFs) and their 90% highest
posterior density intervals are obtained in a standard way when parameters are drawn
from the posterior distribution, as reported in fig. 4.1.

A drought event strongly affects business cycles through a large decline in agricul-
tural output, as weather is a direct input in the production process of agricultural
goods. This result is in line with Kamber et al. (2013), as New Zealand’s farmers rely
extensively on rainfall to support the agricultural sector. This shock acts as a standard
negative supply shock through a combination of rising prices and falling output. The
damages incurred by the weather shock to agricultural output are rather persistent,
as output requires more than 10 periods to return to the steady state. This persistence
can be explained by the deterioration of the competitiveness of farmers on interna-
tional commodity markets, and thus, they require time to recover their market share.
22
Formally, in the weather shock process, a shock anticipated Q quarters in advance is given by:
∑Q
q=1 η̃t−q .
W
 Chapter 4. Climate Change and Business Cycles 127

GDP ∆log(Ytd ) CPI πtC consumption ∆log(Ct ) interest rate log(rt )

·10−2 ·10−2
0 0
−0.05 −0.1 2
4
−0.1 −0.2 1
2 −0.3
−0.15 0
0 −0.4
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

agricultural output ∆log(XtA ) farm prod. inflation πtA non-farm output ∆log(XtN ) non-farm pp inflation πtN
0
0.3 0.1
−0.5 0.3
−1 0.2 0.2 0.05
−1.5 0.1 0.1
−2 0
0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

weather (drought) εW
t intermediate inputs ∆log(Zt ) current account cat real exchange rate log(rert )
·10−2
0.6 0 0
10
0.4 −2 −0.1
5
0.2
−4 −0.2
0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
Notes: Blue lines are the medians of the distributions of the Impulse Response Functions (IRFs) generated when parameters are
drawn from the posterior distribution, as reported in fig. 4.1. Grey areas are the 90 percent highest posterior density interval.
IRFs are reported in percentage deviations from the deterministic steady state. “pp” stands for producer price.

Figure 4.2: System Response to an Estimated Weather Shock ηtS Measured in Per-
centage Deviations from the Steady State

In reaction to inflation pressures, the central bank increases its nominal interest rate,
which deteriorates consumption and output. Monetary policy tightening combined
with a decline in real agricultural production depresses aggregate GDP growth for
four periods before returning to equilibrium.

A drought shock increases the feed budget, since dairy cattle require more water as
temperature, humidity, and production levels rise. In extreme cases, farmers might
even be forced to slaughter their cattle. Farming activities also require more water to
irrigate lands to offset the soil dryness. This demand effect in terms of intermediate
inputs is captured by our model through the increase of 8% in intermediate inputs
(Zt ), which has a positive side effect on non-farm output (XtN ). Regarding interna-
tional economics, the decline in domestic agricultural production generates current
accounts deficits. Two factors might explain this. First, a substantial part of New Zea-
land’s exports is accounted for by agricultural commodities. As agricultural output is
depressed, this might also negatively affect exports. Second, the need for intermediate
goods might be such that it increases imports. These two mechanisms lead to a deteri-
oration in the current account. In the meantime, the real exchange rate depreciates
 128 Part II. Climate Change in Developed Countries

driven by the depressed competitiveness of farmers.

GDP ∆log(Ytd ) CPI πtC consumption ∆log(Ct ) interest rate ∆log(rt )

·10−2 ·10−2 ·10−2
0 0
2
−2 1 −0.05 0
−4 0
−0.1 −1
−6 −1
−2 −0.15 −2
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

agricultural output ∆log(XtA ) farm prod. inflation πtA non-farm output ∆log(XtN ) non-farm pp inflation πtN
·10−2
0
0.1
−0.2 0.1 2
0.05
−0.4
0.05 0 0
−0.6
−0.8 0 −0.05 −2
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

weather (drought) εW
t intermediate inputs ∆log(Zt ) current account cat real exchange rate rert
·10−2 ·10−2
4 0
1
0.2 −2
0 −4
2
0.1 −6
−1 −8
0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
Notes: Blue lines are the medians of the distributions of the Impulse Response Functions (IRFs) generated when parameters are
drawn from the posterior distribution as reported in fig. 4.1. Grey areas are the 90 percent highest posterior density intervals.
IRFs are reported as percentage deviations from the deterministic steady state. “pp” stands for producer price.

Figure 4.3: System Response to an Estimated Weather News Shock η̂t−1

W Measured

as Percentage Deviations from Steady State

Turning to the weather news shock, i.e., the one farmers might anticipate one period
before it occurs, the transmission mechanism is quite similar to the weather surprise
shock overall, with declining output and increasing inflation. Anticipating a shock
one quarter in advance, farmers decide to reduce their production to avoid losses due
to poor weather conditions. The market reacts to this decline in agricultural output
by adjusting prices. Firms operating in the non-agricultural sector anticipate a reces-
sion and reduce their production in advance, which causes non-agricultural prices and
interest rates to drop.

Overall, a representative weather shock has a clear depressing effect on both agricul-
tural output and aggregate GDP. Our model identifies the weather shock as a negative
productivity shock, characterized by a large and durable decline in output and an in-
crease in inflation.
 Chapter 4. Climate Change and Business Cycles 129

4.2 The Contributions of Weather Shocks on Aggregate

Fluctuations

Figure 4.4 reports the forecast error variance decomposition for the four variables of
interest, i.e., aggregate production (YtS ), real domestic absorption (YtD ), agricultural
production (XtA ) and agricultural prices (πtA ). Five different time horizons are con-
sidered, ranging from one quarter (Q1) to ten years (Q4) along with the unconditional
forecast error variance decomposition (Q∞). In each case, the variance is decomposed
into five main components related to supply shocks (technology and markup shocks),
demand shocks (preference and monetary policy shocks), foreign price shocks, foreign
demand shocks and weather shocks.

Real domestic absorption YtS Intermediate production YtD

100% 100%

80% 80%

60% 60%

40% 40%

20% 20%

0% 0%
Q1 Q2 Q4 Q10 Q40 Q∞ Q1 Q2 Q4 Q10 Q40 Q∞

Agricultural production XtA Agricultural production prices πtA

100% 100%

80% 80%

60% 60%

40% 40%

20% 20%

0% 0%
Q1 Q2 Q4 Q10 Q40 Q∞ Q1 Q2 Q4 Q10 Q40 Q∞

Supply shocks (ηtZ + ηtN ) Demand shocks (ηtD + ηtR )

Foreign Price shocks (ηt∗A ) Foreign Demand shocks (ηt∗Y )
Weather shocks (ηtW +η̂t−1
W
)

Figure 4.4: Forecast Error Variance Decomposition at the Posterior Mean for Differ-
ent Time Horizons (one, two, four, ten, forty and unconditional).

As observed for aggregate production (YtS ), home and foreign demand shocks are the
main drivers of the variance in the very short run. However, by increasing the time
horizon, supply and weather shocks become the leading forces of production cycles
to the detriment of demand shocks. In the long run, the unconditional variance error
130 Part II. Climate Change in Developed Countries

forecast decomposition shows that weather shocks explains up to 9% of New Zealand’s

production.

For the total demand (YtD ), unsurprisingly, demand shocks are the main drivers of
fluctuations with a predominance of foreign demand shocks regardless of time hori-
zon. Both types of shocks account for 84% of fluctuations on a one-quarter horizon
and up to 93% on a forty-quarter horizon. In contrast to the aggregate production,
the contribution of weather shocks to macroeconomic fluctuations is quite modest,
although increasing over time.

Turning to agricultural output, demand shocks account for most fluctuations in the
short run, while their importance declines in the long run, although remaining non-
negligible. Weather shocks remarkably drive the variance of agricultural output, and
increasing the time horizon magnifies this result in a similar fashion to the aggregate
production.

Conversely, for agricultural prices, the variance of the forecast error is almost entirely
explained by both supply and demand shocks, which combined represent 98% of the
fluctuations in agricultural prices in the very short run and a slightly lower share in
the long run: 84% at a ten-year horizon. Interestingly, the importance of weather
shocks is also growing with time. However, agricultural prices are mainly driven by
foreign shocks. This result is quite consistent with empirical evidence showing that
food prices are internationally determined.

Overall, we find that weather variations cause important macroeconomic fluctuations,

especially in the long run because of the estimated inertia of the damage function. The
prospect of the increasing variance of drought events caused by climate change is a
challenging issue for New Zealand policymakers, as it can have large implications for
stabilizing policies.

4.3 Historical Decomposition of Business Cycles

An important question one can ask of the estimated model is how important the
weather shocks were in shaping the recent New Zealand macroeconomic experience.
Figure 4.5 provides an answer by reporting the time paths of aggregate output, agri-
cultural production and production prices on a quarter-to-quarter basis. The solid line
depicts the time path of the ratio of deviation from the steady state, while the bars
 Chapter 4. Climate Change and Business Cycles 131

depict the contribution of the shocks (gathered by groups) in the corresponding point
deviation (at the mean of the estimated parameters).

Intermediate production YtS

10 %

0%

−10 %
1989 1991 1994 1997 1999 2002 2005 2008 2010 2013

Agricultural production XtA

50 %

0%

−50 %

1989 1991 1994 1997 1999 2002 2005 2008 2010 2013

Agricultural production prices inflation πtA

4%

2%

0%

−2 %

−4 %
1989 1991 1994 1997 1999 2002 2005 2008 2010 2013

Supply shocks (ηtZ + ηtN ) Demand shocks (ηtD + ηtR )

Foreign Price shocks (ηt∗A ) Foreign Demand shocks (ηt∗Y )
Weather shocks (ηtW +η̂t−1
W
) Residual
Variable Path

Figure 4.5: Historical Decomposition of Aggregate Output, Agricultural Production

and Agricultural Production Price Inflation

In fig. 4.5, we can distinguish between two time periods for output, (YtS ) and (XtA ).
First, up to 2006-2007, variations in aggregate production positively entailed weather
shocks. Over this period, New Zealand did not experience any significant drought
events, with important soil moisture surpluses favoring agricultural production and
132 Part II. Climate Change in Developed Countries

decreasing production prices (πtA ). In fact, during this period, approximately two-
thirds of the increase in agricultural output was driven mostly by positive weather
shocks.

However, major drought events in 2008, 2010 and 2013 contributed negatively to out-
put fluctuations accompanied by an important inflation increase that cleared the im-
balances in the market for agricultural goods. After 2008, almost one-third of the de-
cline in agricultural output is driven by adverse weather drought shocks. Regarding
production price inflation, weather shocks have a limited impact on fluctuations over
the sample period. We find that agricultural prices are mainly driven by home and
international demand shocks, as underlined by the previous forecast variance error
decomposition analysis.

4.4 Climate Change Implications for Aggregate Fluctuations

We now turn to the implications of climate change for aggregate fluctuations. The
IPCC defines climate change as “a change in the state of the climate that can be iden-
tified (e.g., by using statistical tests) by changes in the mean and/or the variability of
its properties, and that persists for an extended period, typically decades or longer” (Ed-
enhofer et al., 2014). In our framework, climate is supposed to be stationary, which
makes our set-up irrelevant for analyzing changes in mean climate values. However,
it allows for changes in the variance of climate. As a first step, we assess the change in
the variance of the weather shock by estimating it under different climate scenarios.
Then, in a second step, we use the estimates of these variances for each scenario and
investigate the effects on aggregate fluctuations.

4.4.1 Building Projections on Climate Shocks Variance

To investigate the potential impact of climate change on aggregate fluctuations, we

assume that the volatility of climatic shocks (ηtW and η̂t−1
W
) (eq. (4.13)) will be affected
by climate change. Instead of arbitrarily setting a value for this shift, we provide an
approximation using a proxy for the drought index. To do so, we rely on monthly cli-
matic data simulated from a circulation climate model, the Community Climate Sys-
tem Model (CCSM). The resolution of the dataset is a 0.9◦ × 1.25◦ grid. Simulated data
are divided into two sets: one of “historical” data up to 2005 and one of “projected”
data from 2006 to 2100. The projected data are given for four scenarios of greenhouse
gas concentration trajectories, the so-called Representative Concentration Pathways
 Chapter 4. Climate Change and Business Cycles 133

(RCPs). The first three, i.e., the RCPs of 2.6, 4.5 and 6.0, are characterized by increasing
greenhouse gas concentrations, which peak and then decline. The date of this peak
varies among scenarios: around 2020 for the RCP 2.6 scenario, around 2040 for the
RCP 4.5 and around 2080 for the RCP 6.0. The last scenario, the doom and gloom 8.5
pathway, is based on a quickly increasing concentration over the whole century.

For these four scenarios, soil moisture deficit data are not available. We therefore use a
strongly correlated variable as a proxy: total rainfall. Simulated data for each scenario
are provided on a grid on a monthly basis. We aggregate them at the national level on
a quarterly basis. More details on the aggregation can be found in appendix 1.

These data are then used to estimate the evolution of the volatility of the climatic
shock. We do so using a rolling window approach. In the DSGE model, we assume
that the weather shock is autoregressive of order one. We therefore fit an AR(1) model
on each window. The size of the latter is set to 25.5 years, i.e., the length of the sample
data used in the DSGE model, so each regression is estimated using 102 observations.
The standard error of the residuals are then extracted to give a measure of the evolution
of the volatility of the climatic shock. Figure 4.6 illustrates them for each scenario. It
is then possible to compute the average growth rate of the standard error over the
century depending on the climate scenario.23 In the best-case scenario, RCP 2.5, the
variance of the climate measure is reduced by 5.29%; under the RCP 4.5 and RCP 6.0
scenarios, it increases by 5.34% and 6.42%, respectively; under the pessimistic RCP 8.5
scenario, it drastically increases by 20.46%.

4.4.2 Measuring Climate Change Implications for Aggregate Fluctu-

ations

We use the estimated DSGE model to assess the effects of a shift in the variability of
the weather shock process. We do so in a two-step procedure. First, the simulations
are estimated with the value of the standard error of the weather shock that is estim-
ated during the fit exercise, which corresponds to historical variability. Second, new
simulations are made after altering the variability of the weather shock so it corres-
ponds to the one associated with climate change, using the values obtained from the
previous section. Hence, we proceed to four different alterations of the variance of
the weather process.
23
More details on the procedure can be found in appendix 1.
134 Part II. Climate Change in Developed Countries

65

60

55

50

2010 2015 2020 2025 2030 2035 2040 2045 2050 2055 2060 2065 2070 2075 2080 2085 2090 2095 2100
End Date of Rolling Window
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Notes: Each curve represents the standard errors of climate shock resulting from the rolling window estimation of an AR(1)
model using data from the 25.5 previous years for a Representative Concentration Pathways scenario.

Figure 4.6: Estimations of the Standard Error of Climatic Shock Under Four Different
Climate Scenarios

Table 4.3: Changes in Standard-Errors of Simulated Observables Under Climate

Change Scenarios

1989-2014 2015-2100 (projections)

Benchmark RCP 2.5 RCP 4.5 RCP 6.0 RCP 8.5
sd(ηtW ) 100 94.71 105.34 106.42 120.46
sd(YtS ) 100 99.57 100.68 100.93 102.37
sd(YtD ) 100 99.88 100.19 100.26 100.66
sd(XtA ) 100 97.92 103.27 104.50 111.47
sd(πtA ) 100 98.37 102.57 103.53 109.00
sd(Ct ) 100 99.88 100.19 100.26 100.66
sd(rt ) 100 99.96 100.06 100.09 100.23
sd(rert ) 100 99.97 100.04 100.06 100.15
Notes: The model is first simulated as described in section 3. Theoretical standard errors of each variable are then estimated and
normalized to 100. Then, standard errors of both weather (ηtW ) and weather-news shocks (η̃tW ) are modified to reflect different
climate scenarios (compared to the reference 1986–2014 period, changes in the standard error are as follows: RCP 2.5, −5.29%;
RCP 4.5, +5.34%; RCP 6.0, +6.42%; RCP 8.5, +20.46%). New simulations are estimated using the modified standard errors of
these shocks, and the theoretical standard errors of the variables of interest are then compared to those of the reference period.
The variables of interest are aggregate production (YtS ), total demand (YtD ), agricultural production (Xt ), agricultural prices
(πtA ), consumption (Ct ), interest rate (r), and real exchange rate (rert ).

To measure the implications of climate change on aggregate fluctuations of a repres-

entative open economy, we compare the volatility of some macroeconomic variables
under historical weather conditions (for the 1989–2014 period) to their volatility un-
der future climate scenarios (for the 2015–2100 period), normalizing the values of the
historical period of each variable to 100.

Table 4.3 reports these variations for some key variables. The first scenario is clearly
 Chapter 4. Climate Change and Business Cycles 135

optimistic, as the standard deviation of drought events is declining by 5.29%. Macroe-

conomic fluctuations are clearly affected by the concrete reduction of the volatility of
outputs and prices for New Zealand, ranging from roughly -0.1% to -2%. In contrast,
we consider two intermediate pessimistic scenarios characterized by weather shock
volatility rising between 5% and 6% (RCPs 4.5 and 6.0). For these scenarios, our model
predicts a strong increase in the standard deviations of macroeconomic variables. In
particular, the standard deviation of agricultural output increases by 4.5% and that of
agricultural prices by 3.53%. Finally, in the worst-case scenario (RCP 8.5), a 20.46%
increase in the standard deviation of weather shocks has important implications for
the business cycles of New Zealand. It leads agricultural output variability to rise by
11% and inflation to rise by 9%, and it affects the variability of aggregate production
by 2%.

5 Conclusion
In this chapter, we have investigated the business cycle evidence on weather shocks.
We have developed and estimated a DSGE model for a small open economy, New
Zealand. Our model includes an agricultural sector that faces an exogenous weather
input affecting the production function and demand for the intermediate goods. Using
our estimated model, we find that weather shocks play an important role in explaining
macroeconomic fluctuations over the sample period. This finding is robust to different
specifications of the lag structure of the damage function affecting agricultural produc-
tion. However, our framework shows that farmers do not anticipate weather shocks
and are mostly surprised by variable climatic conditions. Our results also show that in
the medium run, weather shocks are important drivers of agricultural production and
are responsible for variations in agricultural prices and total output. These findings
suggest that weather shocks should not be neglected in the conduct of macroeconomic
policies, especially since one-third of the decline in agricultural output after 2008 is
driven by adverse weather drought shocks, as illustrated by our historical decompos-
ition of business cycles. Finally, our framework is amenable to the analysis of climate
change. Our simulations show that altering the variability of weather shocks in line
with what is expected to happen with climate change leads to an increase in the vari-
ability of key macroeconomic variables, such as output prices, production or inflation,
in all but the most optimistic scenario.
136 Part II. Climate Change in Developed Countries

The analysis of weather-driven business cycles is a burgeoning research area given the
important context of climate change. In this chapter, we have analysed the import-
ance of weather shocks on the macroeconomic fluctuations of a developed economy.
However, the application of our framework to developing countries could highlight
the high vulnerability of their primary sectors to weather shocks. In addition, from a
policymaker’s perspective, our framework could be fruitfully employed to evaluate the
optimal conduct of monetary, fiscal and environmental investment policies to mitigate
the destabilizing effects of weather shocks for different scenarios of climate change.
General Conclusion

Scientists predict that climate change will result in an increase of the average temperat-
ure accompanied by more severe weather events, such as drought and desertification,
intense rainfall, flooding, etc. As the list of potentially harming events goes on and
on, so does the need for documentation on their effects on our planet. The aim of this
thesis is to contribute to this effort by focusing on a specific area, agriculture. Many
challenges lay ahead with respect to climate change. The last 2015 Paris international
climate agreement, signed by 175 parties, sets goals to keep a global temperature rise
well below 2 degrees Celsius above pre-industrial levels and to pursue efforts to limit
it to 1.5 degree Celsius. It is however up to the respective governments to take action
to achieve this ambitious goal, with no incentive to prevent the free rider problem
from occurring. In any case, whether the goal of the limiting temperature is reached
or not, even an increase of 1.5 degree Celsius above the pre-industrial level is expec-
ted to significantly affect agriculture. The impacts are expected to affect developing
and developed countries differently. In developing countries, although the process of
catching up to the developed countries mitigates the importance of agriculture in the
economy, this sector still represents a relatively important part of total output and
provides jobs to millions of people. Hence, threats posed by climate change on agri-
culture are likely to have substantial consequences on production, and therefore on
profits and on food security. These problems also apply to developed countries, even
if their economy depends less on agriculture. In fact, a large part of the global produc-
tion comes from developed countries. On the basis of the estimates of the FAO, around
40% of global cereal production comes from developed countries at this time. Such a
large share implies that the impact of climate change on developing countries should
not be overlooked. Both for developing and developed countries, adverse weather ac-
cidents might drastically reduce production resulting in disastrous situations of food
shortage, as witnessed during the last decades. A better understanding on the under-
lying mechanisms is therefore necessary to help mitigate the harmful effects of climate

137
138 Chapter 4. Climate Change and Business Cycles

change, to be able to feed the growing population of our planet.

The four chapters of this thesis are an attempt to provide some insight into these
challenges. The first part focuses on developing countries and provides two empirical
studies based on Indian data at the individual farm level. The second part considers
developed countries with a first look at crop yields in Europe followed by an analysis
of weather shocks on business cycles.

The main results can be summarised as follows. The empirical studies provided in
chapters 1 and 2 show heterogeneous impacts of climate on profits, on production,
and on consumption choices in India. The agricultural production of some farmers as
well as their profits is projected to increase under the climate scenarios tested. But
these gains are more than offset by the losses projected for the other farmers, so that
the overall projected effect of climate change is damaging to Indian agriculture. Fur-
thermore, climate change scenarios underscore a contrasting difference between the
north and the south of the country, the former being more vulnerable to the scen-
arios considered. Some farming practices however enable farmers to better cope with
changing climate conditions. Irrigated farms tend to be less affected by a marginal
change in temperature and in total precipitation. As around 40% of Indian farms still
rely on water from precipitation only to irrigate crops, further efforts can be made to
improve irrigation schemes. Mixing crops is another possible way of mitigating the
overall damaging effects of climate change, especially for small farms.

The relationship between the weather and crop yields in Europe is investigated in
Chapter 3. Wheat yields tend to rise under the projected climate scenarios in south-
ern Europe, although this rise is accompanied by a relatively high variability. In the
northern regions, i.e., in regions in which wheat production is currently relatively
higher, small gains are projected, but also accompanied by a lot of variability. Projec-
tions for corn yields are more pessimistic. In the short-run, small gains are projected,
but in the long-run, these gains turn to relatively high losses, especially for southern
regions.

Finally, the DSGE model presented in chapter 4 stresses the role of the weather in
explaining the impact of weather shocks in the short-run of macroeconomic fluctu-
ations. The model is developed for a small-open economy with two sectors, one of
which – the agricultural sector – being affected by a weather shock, more specifically,
by a drought. The model identifies the weather shock as a negative supply shock that
 Chapter 4. Climate Change and Business Cycles 139

depresses both GDP and agricultural production, followed by a rise in prices. The po-
tential effects of climate change are addressed by altering the variance of the weather
shock process, depending on different climate scenarios. The results show that the
modification in the variability of the weather shock leads to an increase in the variab-
ility of key macroeconomic variables such as output prices, production or inflation, in
all but the most optimistic scenario.

Looking forward, some efforts could be directed towards several points for further
research. The analyses on the impacts of climate on profits could be extended to a
wider geographical area, to provide a more complete picture. The production and
consumption decisions of rural households could also be further documented, using a
framework modelling these two decisions in a single model, to account for market im-
perfections. Some more research could also be done regarding the short-run impacts
of weather shocks on the economy. For example, it might be interesting to delving
deeper into the empirical facts, by developing in more details a VAR-type model. Fo-
cusing on the welfare aspect of the question might also be a starting point for further
work. Finally, the literature finds evidence that developing countries are and will be
more vulnerable to climate change than developed countries. However, some dispar-
ities are also observed within countries. Hence, it may be interesting to investigate
how climate change can affect social inequalities.
Appendix A

Ricardian Analysis: Data

1 Definitions of Variables
Table A.1 provides a definition of each variable used in the analysis.

Table A.1: Description of Variables

Variable Description Type

Variable of interest
Net ag. revenue Net agricultural revenue (rupees) numeric

Personal characteristics
Age of head of householdage of the head of the household numeric
Literacy Years of schooling (∈ [0, 15]) numeric

Farm characteristics
Area planted Planted surface (acres) numeric
No. bullock carts Number of bullock carts numeric
No. tractors Number of tractors numeric
No. different cultures Number of different cultures numeric
No. workers in the farm Number of workers in the farm numeric
Irrigation Most important type of irrigation factor
(none, tube well, other well, govern-
ment, tank, private canal, other)

Location
Latitude Latitude (degrees) numeric

Continued on next page

141
142 Appendix A Ricardian Analysis: Data

Variable Description Type

Pop. density Population density (number of inhab- numeric

itants per square meter)

Climatic factors
Winter precip. January to March total rainfall aver- numeric
age (mm)
Summer precip. April to June total rainfall average numeric
(mm)
Monsoom precip. July to September total rainfall aver- numeric
age (mm)
Autumn precip. October to December total rainfall av- numeric
erage (mm)
Winter temp. January to March mean temperature numeric
average
Summer temp. April to June mean temperature aver- numeric
age
Monsoom temp. July to September mean temperature numeric
average
Autumn temp. October to December mean temperat- numeric
ure average

Figure A.1 shows the distribution of net revenues per acre, while Figure A.2 presents
the distribution of cultivated area and fig. A.3 the distribution of irrigation types.

900

600

300

0
-10000 0 10000 20000 30000

Note: the average value is represented by the dashed vertical grey line. The y-axis gives the number of farms in the sample in
each bin.

Figure A.1: Net Revenues per Acre Distribution

 Appendix A Ricardian Analysis: Data 143

1500

1000

500

0
0 10 20 30 40 50

Notes: the average value is represented by the dashed vertical grey line. The y-axis gives the number of farms in the sample in
each bin.

Figure A.2: Cultivated Surface

6000

4000
40.02%
33.25%
2000
11.51%
5.87% 4.86% 2.86% 1.63%
0
None Tubewell Other well Gov. Tank/pond/nala Other Private canal

Notes: The y-axis gives the number of farms in the sample for each type of irrigation.

Figure A.3: Most Important Mode of Irrigation (Number of Observations)

2 Weather Data

2.1 Observed Weather Data

This annex provides more details on the methods used to compute climate “normals”
at the district level from weather station data. Figure A.4 summarizes the proced-
ure. We rely on daily mean temperature and total rainfall data, provided by the Na-
tional Climatic Data Center (NCDC)/National Oceanic and Atmospheric Administra-
tion (NOAA).

The first step consists in roughly estimating missing values. We follow a simple rule:
if no more than 4 observations are missing, both during the previous and the next 14
days, the missing value is estimated by a weighted mean. Weights are computed in
the following way:

d2
wi = ∑n i , (A.1)
j=1 d2j
144 Appendix A Ricardian Analysis: Data

where di = (15 − i)δ, with δ = 1 if the observation i days away is available, δ = 0

otherwise. These weights give thereby more importance to close observations. As a
result, we can rely on more observation (see fig. A.6) for the estimation in the following
step.

The second consists in using the available data to make predictions where there is
no station. The idea is to consider a grid covering India. For each cell of this grid,
a prediction is made, using an interpolation method called thin plate splines (see e.g.
Di Falco et al. (2011); Boer (2001); Hutchinson (1995)), implemented in the statistical
software R in the package fields (Nychka et al., 2015). On average, there are 61 stations
providing data for each day (see fig. A.6). The reliability of the estimation was assessed
by cross-validation for each day, leaving one observation out and checking if the ac-
tual value lies in the confidence interval. The interpolation predicts 84.46% of actual
weather station precipitation and 76.79% of actual weather station precipitation.1

Once the estimation for each cell of the grid is performed, an average by district can
easily be computed. It is then possible to aggregate data by season and district. Results
are displayed in figs. A.7 and A.8 for precipitation and precipitation, respectively.

Raw data

Estimation of missing values

Thin-plate splines

Aggregation

Figure A.4: Process from Raw Data to District Level Climatic Data
1
A way to improve this accuracy would be to add elevation data to realize the interpolation.
 Appendix A Ricardian Analysis: Data 145

0km 500km 1000km

Notes: Each dot represents the location of a weather station.

Figure A.5: Meteorological Stations Locations in India

80 Mean Temperature Precipitation

80

60
60

40
40

20 20

Mean Temperature Precipitation 1980 1990 2000 1980 1990 2000

(a) Boxplot (b) Overview

Notes: The distribution of the number of weather stations used to compute daily weather data is displayed on the left panel. The
evolution through time of the number of weather stations used is displayed on the right panel.

Figure A.6: Number of Observations Used to Estimate Weather Data per Day
146 Appendix A Ricardian Analysis: Data

Precipitation (mm)

50 100 150 200

Winter Summer Monsoon Autumn

Notes: Precipitation “normals” correspond to 30-year average over the period ranging from 1976–2005.

Figure A.7: Precipitation “Normals”

Temperatures (o C)

10 15 20 25 30
Winter Summer Monsoon Autumn

Notes: Temperature “Normals” Correspond to 30-year Average Over the Period Ranging from 1976–2005.

Figure A.8: Temperature “Normals”

2.2 Climate Scenarios

To give an idea of the potential consequences of climate change on Indian net revenues,
we envisage two climate scenarios and observe the changes in net revenues in each of
them under the new weather conditions. To set up the scenarios, we follow Chaturvedi
et al. (2012). The first scenario reflects a low concentration of greenhouse gas (roughly
corresponding to the representative concentration pathway (RCP) 2.6, adopted by The
Intergovernmental Panel on Climate Change for its fifth Assessment Reports in 2014),
where average temperature for India is projected to increase by 1.7◦ C and total rainfall
by 1.2%. It might be viewed ad a mitigation scenario. The second scenario reflects
 Appendix A Ricardian Analysis: Data 147

high concentration of greenhouse gas (roughly corresponding to the RCP 8.5), mean
temperature is projected to increase by 2.02◦ C and total rainfall by 2.4%. This scenario
is more pessimistic than the first.
Appendix B

Agricultural Yields: Data

1 Agricultural Data
The agricultural data are from the Farm Accountancy Data Network1 (FADN). For
statistical anonymity, the observations collected at the level of European farms are
aggregated at the NUTS-3 geographic level. The NUTS-3 regions are subdivisions
of the European territory established by Eurostat. All in all, the European Union is
divided in 1, 294 economic territories, and the FADN data report values for 139 NUTS-
3 regions in 28 countries, from 1989 to 2009.

Farm plots are classified according to their main activity, in 8 categories: (i) Fieldcrops,
(ii) Horticulture, (iii) Wine, (iv) Other permanent crops, (v) Milk, (vi) Other grazing
livestock, (vii) Granivores, (viii) Mixed. In our analysis, we focus on farms whose
main activity is crop growing, i.e., the first category in the dataset.

In the study, some restrictions are imposed on the dataset to ensure the convergence
properties of the estimates. First, we only keep regions for which at least 15 observa-
tion are provided in the historical period ranging from 1989 to [Link], we discard
observations from isolated areas such as small [Link], we create two datasets,
one for regions reporting values for wheat production, and the other one giving values
for corn production. This leaves us with the following specifications:

• Wheat dataset: 545 observations in 31 NUTS-3 regions from 8 countries (Austria,

France, Italy, Netherlands, Portugal, Spain, Sweden, United Kingdom)
1
The FADN database is publiclu available at [Link]
consult_std_reports_en.cfm

149
150 Appendix B Agricultural Yields: Data

– 384 of which are located north of the 45th parallel

– 161 of which are located south of the 45th parallel

• Corn dataset: 446 observations in 25 NUTS-3 regions from 5 countries (Austria,

France, Italy, Portugal, Spain)

– 284 of which are located north of the 45th parallel

– 162 of which are located south of the 45th parallel

2 Weather Data
This section describes historical and projected weather data used in the main docu-
ment.

2.1 Definitions

Weather data are obtained from the Meteorological Research couples ocean-atmosphere
model MRI-CGCM3, and aggregated at regional level. Raw data are given at a daily
frequency, on a 1.125◦ × 1.12148◦ grid. We aggregate them at the NUTS-3 level by
means of an area-weighted means, such that for each region, the aggregate value cor-
responds to the weighted average of all grids where the weight is proportional to the
grid area in the region. We then aggregate data by season as follows: (i) Winter, from
December to February; (ii) Spring, from March to May; (iii) Summer, from June to
August; and (iv) Fall, from September to November.

Three weather variables are considered: temperature, temperature deviation (daily

spread between maximum and minimum temperature), and precipitation. Four cli-
mate scenarios are used: the RCPs 2.6, 4.5, 6.0, and 8.5. The first three, i.e., the RCPs
2.6, 4.5 and 6.0 are characterized by increasing greenhouse gas concentrations with a
peak around 2020, 2040, and 2060, respectively, followed by a slow decline. The last
scenario, the RCP 8.5, is less optimistic in terms of emissions and leads to a rapidly
increasing concentration over the whole century.

In addition to historical values, ranging from 1991 to 2009, two time horizon are high-
lighted: a short one, ranging from 2009 to 2050, and a long one, ranging from 2051 to
to end of the 21st century.
 Appendix B Agricultural Yields: Data 151

2.2 Descriptive Statistics

The analysis of wheat and corn yield is made on sub-samples of data, considering
northern and southern European regions separately. Descriptive statistics of weather
variables used in these sub-samples are reported in Table B.1.

Table B.1: Descriptive Statistics of Climate Variables

Data for Wheat Yield Model

Historical RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Short-run Long-run Short-run Long-run Short-run Long-run Short-run Long-run

Variable Unit Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD
o 4.62 3.06 4.63 3.19 5.2 3.04 4.86 3.09 5.41 2.99 4.71 3.26 5.74 2.94 4.83 3.22 6.54 2.91
Temp. (winter) C
o 9.15 2.64 9.08 2.58 9.61 2.51 9.64 2.56 9.93 2.57 9.37 2.59 10.13 2.63 9.62 2.68 10.93 2.68
Temp. (spring) C
o 17.6 2.2 17.77 2.34 18.14 2.36 17.96 2.35 18.65 2.53 17.8 2.4 18.81 2.58 18.14 2.43 19.81 2.71
Temp. (summer) C
o 10.39 2.79 10.76 2.69 11.31 2.69 10.97 2.7 11.77 2.71 10.9 2.75 11.95 2.63 10.89 2.73 13.05 2.76
Temp. (fall) C
o 4.49 0.98 4.37 1 4.38 0.96 4.32 0.96 4.32 0.98 4.45 1.04 4.32 1 4.4 1.04 4.31 0.98
Temp. Dev. (winter) C
o 7.09 1.36 6.98 1.4 7.01 1.34 7 1.37 7.05 1.43 7.08 1.43 7.07 1.44 7.13 1.44 7.07 1.47
Temp. Dev. (spring) C
o 7.77 1.52 7.63 1.55 7.6 1.53 7.62 1.52 7.66 1.62 7.66 1.57 7.71 1.69 7.67 1.58 7.62 1.69
Temp. Dev. (summer) C
o 5.32 1.21 5.2 1.2 5.26 1.18 5.22 1.16 5.21 1.23 5.28 1.21 5.26 1.27 5.24 1.27 5.24 1.27
Temp. Dev. (fall) C
Precip. (winter) m 8.1 3.16 7.78 2.63 8.26 2.84 8.13 2.95 8.32 2.84 8.29 2.99 8.31 2.96 8.15 2.95 8.71 3.12
Precip. (spring) m 6.68 1.99 6.67 2.13 6.68 2.17 6.6 2.06 6.49 2.11 6.44 2.16 6.29 2.11 6.28 2.1 6.65 2.45
Precip. (summer) m 6.35 2.72 6.3 2.63 6.44 2.7 6.27 2.6 6.04 2.68 6.16 2.58 5.95 2.74 6.31 2.65 6.09 2.8
Precip. (fall) m 8.75 2.79 8.89 2.44 8.6 2.47 8.47 2.5 9.01 2.51 8.58 2.55 8.55 2.59 8.7 2.6 8.56 2.64
Data for Corn Yield Model
Historical RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

Short-run Long-run Short-run Long-run Short-run Long-run Short-run Long-run

Variable Unit Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD
o 5.1 2.53 5.26 2.44 5.76 2.43 5.44 2.45 5.95 2.41 5.37 2.43 6.3 2.33 5.45 2.49 7.08 2.37
Temp. (winter) C
o 9.72 1.95 9.7 1.82 10.22 1.81 10.28 1.8 10.56 1.85 10 1.83 10.75 1.94 10.26 1.94 11.55 2
Temp. (spring) C
o 18.13 1.6 18.31 1.76 18.72 1.75 18.55 1.74 19.26 1.89 18.38 1.8 19.4 1.97 18.74 1.78 20.41 2.14
Temp. (summer) C
o 10.87 2.35 11.29 2.16 11.83 2.2 11.49 2.18 12.31 2.18 11.45 2.22 12.44 2.17 11.4 2.24 13.57 2.3
Temp. (fall) C
o 4.62 0.91 4.49 0.93 4.51 0.89 4.43 0.9 4.48 0.91 4.54 0.96 4.47 0.93 4.53 0.97 4.49 0.9
Temp. Dev. (winter) C
o 7.34 1.21 7.2 1.29 7.24 1.23 7.22 1.25 7.31 1.31 7.32 1.31 7.33 1.32 7.37 1.32 7.35 1.35
Temp. Dev. (spring) C
o 8.13 1.34 7.97 1.39 7.93 1.38 7.96 1.35 8.01 1.44 8 1.42 8.07 1.51 8.01 1.42 7.98 1.51
Temp. Dev. (summer) C
o 5.59 1.08 5.46 1.09 5.53 1.07 5.48 1.05 5.5 1.1 5.56 1.09 5.56 1.15 5.52 1.15 5.54 1.15
Temp. Dev. (fall) C
Precip. (winter) m 8.26 3.33 8.07 2.97 8.59 3.24 8.4 3.33 8.68 3.27 8.55 3.29 8.62 3.43 8.44 3.24 9.11 3.61
Precip. (spring) m 6.98 2.01 6.95 2.2 7.01 2.31 6.88 2.1 6.71 2.16 6.67 2.23 6.51 2.16 6.44 2.16 6.88 2.55
Precip. (summer) m 6.49 2.67 6.45 2.63 6.56 2.71 6.41 2.61 6.07 2.68 6.26 2.61 6.01 2.8 6.46 2.66 6.08 2.82
Precip. (fall) m 9.2 3.02 9.37 2.62 9.03 2.72 8.95 2.83 9.44 2.75 8.95 2.7 8.87 2.81 9.09 2.85 8.87 2.75

Notes: For each weather variable, average and standard error are given for the historical period (1992–2009), for projections in
the short-run (2009–2050), and in the long-run (2051–2100). Four scenarios are considered: RCP 2.6, RCP 4.5, RCP 6.0 and RCP
8.5. Descriptive statistics are given for the sample used for wheat yield model (top) and for corn yield (bottom).

2.3 Projected Weather Variables

Using the historical period (1991–2009) as a benchmark, we compute the average re-
gional and seasonal change in the short-run (2009–2050) and in the long-run (2051–
2100) for the three weather variables used in the analysis, under each of the four cli-
mate scenarios. Figures B.1 to B.4 display the average change in temperature, in degree
Celsius. Finally, figures B.5 to B.8 show the average percentage change in precipita-
tion.
152 Appendix B Agricultural Yields: Data

Change in Temp. (winter), (o C)

-4 0 4
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

(2009-2050)
Short-term
(2051-2100)
Long-term
Note: Each map shows the change in temperature in the short-run (2009-2050) (top panels) and in the long-run (2051-2100)
(bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).

Figure B.1: Projected Changes in Winter Temperature

Change in Temp. (spring), (o C)

-4 0 4
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5
(2009-2050)
Short-term
(2051-2100)
Long-term

Note: Each map shows the change in temperature in the short-run (2009-2050) (top panels) and in the long-run (2051-2100)
(bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).

Figure B.2: Projected Changes in Spring Temperature

 Appendix B Agricultural Yields: Data 153

Change in Temp. (summer), (o C)

-4 0 4
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

(2009-2050)
Short-term
(2051-2100)
Long-term
Note: Each map shows the change in temperature in the short-run (2009-2050) (top panels) and in the long-run (2051-2100)
(bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).

Figure B.3: Projected Changes in Summer Temperature

Change in Temp. (fall), (o C)

-4 0 4
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5
(2009-2050)
Short-term
(2051-2100)
Long-term

Note: Each map shows the change in temperature in the short-run (2009-2050) (top panels) and in the long-run (2051-2100)
(bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and 8.5).

Figure B.4: Projected Changes in Fall Temperature

154 Appendix B Agricultural Yields: Data

Percentage Change in Precip. (winter)

-25 0 25
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

(2009-2050)
Short-term
(2051-2100)
Long-term
Note: Each map shows the percentage change in precipitation in the short-run (2009-2050) (top panels) and in the long-run
(2051-2100) (bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and
8.5).

Figure B.5: Projected Percentage Changes in Winter Precipitation

Percentage Change in Precip. (spring)

-25 0 25
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5
(2009-2050)
Short-term
(2051-2100)
Long-term

Note: Each map shows the percentage change in precipitation in the short-run (2009-2050) (top panels) and in the long-run
(2051-2100) (bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and
8.5).

Figure B.6: Projected Percentage Changes in Spring Precipitation

 Appendix B Agricultural Yields: Data 155

Percentage Change in Precip. (summer)

-25 0 25
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5

(2009-2050)
Short-term
(2051-2100)
Long-term
Note: Each map shows the percentage change in precipitation in the short-run (2009-2050) (top panels) and in the long-run
(2051-2100) (bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and
8.5).

Figure B.7: Projected Percentage Changes in Summer Precipitation

Percentage Change in Precip. (fall)

-25 0 25
RCP 2.6 RCP 4.5 RCP 6.0 RCP 8.5
(2009-2050)
Short-term
(2051-2100)
Long-term

Note: Each map shows the percentage change in precipitation in the short-run (2009-2050) (top panels) and in the long-run
(2051-2100) (bottom panels) compared to the average historical value (1991-2009), under each scenario (RCPs 2.6, 4.5, 6.0, and
8.5).

Figure B.8: Projected Percentage Changes in Fall Precipitation

Appendix C

Climate Change and Business

Cycles: Technical Appendix

1 Data
Agricultural production price inflation data begins in 1994:Q2. We backcast the miss-
ing data by fitting an ARM A(1, 1) model with CPI inflation and soil moisture deficit
index as external regressors.

Climate data are obtained from weather stations at a monthly rate. The measure we
use is based on soil moisture deficit observations.

• Gross domestic product: real per capita output, expenditure approach, season-
ally adjusted. Source: Statistics New Zealand.

• CPI inflation: all groups index, Source: Statistics New Zealand.

• Agricultural output: real agriculture, fishing and forestry gross domestic product,
seasonally adjusted. Source: Statistics New Zealand.

• Agricultural producer price inflation: agriculture, fishing and forestry produ-

cer price index. Source: Statistics New Zealand.

• Population: actual population of working age, in thousands, seasonally adjus-

ted. Source: Statistics New Zealand.

• Interest rate: 3-Month rates and yields: bank bills for New Zealand, not season-
ally adjusted. Source: Main Economic Indicators, OECD.
157
158 Appendix C Climate Change and Business Cycles: Technical Appendix

• Real exchange rate: real trade weighted index. Source: Reserve Bank of New
Zealand.

• Climate: soil moisture deficit at the station level. Source: National Climate Data-
base, National Institute of Water and Atmospheric Research.

2 Model Description
Agricultural Sector

Goods XtA?
Foreign Firms

Central Bank

Goods XtA
Intermediate Firms Final Firms

Rt
Labor hA
t

Inputs Zt
Households

Labor hN
t
Non-Agricultural Sector

Goods XtN
Intermediate Firms Final Firms

Goods Xt?
Foreign Firms

Consumption Ct

Figure C.1: A Small Open Economy Featuring Two Sectors: Agricultural and Non-
agricultural Sectors

As shown in fig. C.1, the domestic economy is populated by:

• Households that consume save and supply labour;

• Intermediate and final firms operating in two sectors: the agricultural sector and
the non-agricultural one. In both of them:

– intermediate firms supply differentiated goods in a monopolistically com-

petitive market and set prices in a staggered basis,
– final producers aggregate the differentiated goods from both domestic and
foreign intermediate firms and sell it to households. The agricultural final
firm uses intermediate goods from firms of both sectors;
 Appendix C Climate Change and Business Cycles: Technical Appendix 159

• A Central bank that decides the interest rate according to a Taylor rule.

3 A Sketch of the Model

This section is devoted to a formal presentation of the DSGE model. Our model is
two-sector two-good economy in a small open economy setup with standard New
Keynesian nominal frictions and a flexible exchange rate regime. Small open economy
models usually include two countries. The home country participates in international
trade, but is too small compared to its trading partners to cause aggregate fluctuation
to world output, price and interest rate. The foreign country, gathering most of the
trading partners of the home country, is thus not affected by macroeconomic shocks
from the home country but its own macroeconomic developments affect the home
country through the trade balance and the exchange rate.

The home economy, i.e., New Zealand, is populated by households, final firms, ag-
ricultural and non-agricultural intermediate firms and a central bank. Intermediate
producers in each sector enjoy market power to maximize their profits and produce
differentiated goods. Final goods producers use a packing technology to aggregate
both home and foreign intermediate goods to produce a homogeneous good sold to
households. The final product is a composite of domestically produced and imported
goods, thus creating a trading channel adjusted by the real exchange rate. Nominal
rigidities in the agricultural and non-agricultural sectors generate inflation dynamics
that are damped by the central bank though the adoption of an inflation targeting
regime.

The foreign economy is modelled through six structural and linear equations that aim
at capturing the key patterns of business cycles of New Zealand’s trading partners.
The description of these equations are left in appendix 3.4.

3.1 Households

There is a continuum of identical households who consume, save and work in inter-
mediate firms. The total number of households is normalized to 1. The representative
household j ∈ [0, 1] maximizes the welfare index expressed as the expected sum of
160 Appendix C Climate Change and Business Cycles: Technical Appendix

utilities discounted by β ∈ (0, 1):

{∞ [ ( )]}
∑ (Cjt+τ − hCt−1+τ )1−σC σC − 1 1+σH
Wjt = Et βτ exp h , (C.1)
τ =0
1 − σC 1 + σH jt+τ

where variable Cjt is the consumption index, h ∈ [0, 1] is a parameter that accounts for
external consumption habits, hjt is a labour effort index for the agricultural and non-
agricultural sectors, and σC and σH represent consumption aversion and labour dis-
utility, respectively. Following the seminal contribution of Smets and Wouters (2007),
households preferences are assumed to be non-separable in consumption, so an in-
crease in hours worked has a positive effect on the marginal utility of consumption.1

The representative household allocates total consumption Cjt between two types of
consumption goods produced by the non-agricultural and agricultural sectors denoted
N
Cjt and Cjt
A
respectively. The CES consumption bundle is determined by:

( 1 ( ) µ−1 ( A ) µ−1
1
) µ−1
µ

Cjt = (1 − φ) µ N
Cjt µ
+ (φ) Cjt
µ µ
, (C.2)

where µ ≥ 0 denotes the substitution elasticity between the two types of consumption
goods, and φ ∈ [0, 1] is the fraction of agricultural goods in the household’s total con-
sumption basket. The corresponding consumption price index thus reads as follows:
[ ( N )1−µ ( A )1−µ ] 1−µ
1

PtC = (1 − φ) Py,t + φ Py,t , (C.3)

where Py,t
N
and Py,t
A
are the final prices of non-agricultural and agricultural goods re-
spectively. Demand for each type of final good is a fraction of the total consumption
index adjusted by its relative price:

( N C )−µ ( A C )−µ
N
Cjt = (1 − φ) Py,t /Pt Cjt and Cjt
A
= φ Py,t /Pt Cjt . (C.4)

Following Iacoviello and Neri (2010), we introduce imperfect substitutability of labour

supply between the durable and non-durable sector to explain co-movements at the
sector level by defining a CES labour disutility index:
[ ( ) ( A )1+ι ]1/(1+ι)
1+ι
hjt = n hN
jt + (1 − n) hjt (C.5)
1
We refer the reader to Greenwood et al. (1988) for a discussion of the implications of non-separable
preferences on business cycles.
 Appendix C Climate Change and Business Cycles: Technical Appendix 161

The labour disutility index consists of hours worked in the non-agricultural sector hN
jt
and agriculture sector hA
jt , with n denoting the relative share of employment in the
non-agricultural sector. Reallocating labour across sectors is costly and is governed
by the substitutability parameter ι ≥ 0.2

Expressed in real terms and dividing by the consumption price index PtC , the budget
constraint for the representative household can be represented as:

∑ ∗ N
Py,t
Πjt Rt−1 Rt−1 ∗ ∗
χS wts hsjt + + b jt−1 + b = C jt + b jt + e b
t jt + et ΦB (b∗jt ).
s=N,A PtC πtC πtC jt−1 PtC
(C.6)
The income of the representative household is made up of labour income with a real
wage wts in each sector,3 profits Πjt generated by imperfect competition in goods, and
real riskless domestics bonds bjt and foreign bonds b∗jt . Domestic and foreign bonds
∗
are remunerated at a domestic Rt−1 and a foreign Rt−1 , respectively, nominal gross
interest rates decided by central banks of each country and adjusted by the domestic
inflation rate πtC = PtC /Pt−1
C
. Household’s foreign bonds purchases are affected by the
nominal exchange rate et (an increase in et can be interpreted as an appreciation of the
domestic exchange rate). The household’s expenditure side includes its consumption
basket Cjt , bonds and risk-premium cost Φ(b∗jt )=0.5χB (b∗jt )2 paid in terms of domestic
N 4
final goods at a market price Py,t . Parameter χB > 0 denotes the magnitude of the
cost paid by domestic households when purchasing foreign bonds.
2
If ι equals zero, hours worked across the two sectors are perfect substitutes, leading to a negative
correlation between the sectors that is not consistent with the data. Positive values of ι capture some
degree of sector specificity and imply that relative hours respond less to sectoral wage differentials.
3
Real labour income is affected by χs > 0, a sector-specific shift parameter that allows us to calibrate
the steady state of hours worked in each sector. This is a common assumption in real business cycle
models.
4
This cost function aims at removing a unit root component that emerges in open economy models
without affecting the steady state of the model. See Schmitt-Grohé and Uribe (2003) for a discussion of
closing open economy models.
162 Appendix C Climate Change and Business Cycles: Technical Appendix

The dynamic Lagrangian reads as follows:

∑∞ ( )
τ (Cjt+τ − hCt−1+τ ) σC − 1 1+σH
1−σC
Lt = Et { β [ exp h
1 − σC 1 + σH jt+τ
[
τ =0
]
∑ ∗
Π R R
b∗jt−1+τ
jt+τ t−1+τ
+ λct+τ s
χS wt+τ hsjt+τ + C + C bjt−1+τ + t−1+τ C
s=N,A Pt+τ πt+τ πt+τ
[ ]
N
Py,t+τ
− λct+τ Cjt+τ + bjt+τ + et+τ b∗jt+τ + C et ΦB (b∗jt+τ )
Pt+τ
[ ( )1+ι ( A )1+ι ]
h
+ λt+τ n hjt+τN
+ (1 − n) hjt+τ − hjt+τ ]}
1+ι

where λct and λht are Lagrange multipliers associated to each constraint and can be
interpreted as the marginal utility of consumption and marginal disutility of labour
supply, respectively. Constraints are included in the discounted sum as they bind every
periods.

First-order conditions are:

( )
−σC σC − 1 1+σH
Cjt : (Cjt+τ − hCt−1+τ ) exp h − λct = 0
1 + σH jt
hιjt
hjt : (σC − 1)hσjtH Ujt − λht =0
( )ι1 + ι
N c N h
n hN jt
hjt : λt χN wt + λt =0
1+ι ( )
ι
A c A h
(1 − n) hA jt
hjt : λt χA wt + λt =0
1+ι
Rt
bjt : −λct + βλct C = 0
πt+1
( )
P N
R∗
b∗jt : −λct et 1 + C χB b∗jt + βλct+1 Ct
y,t
Pt πt+1

The first-order conditions solving the household’s optimization problem are obtained
by maximizing welfare index in eq. (C.1) under the budget constraint in eq. (C.6) given
the labour sectoral re-allocation cost in eq. (4.3). First, the marginal utility of con-
sumption is determined by:5
( )
σC − 1 1+σH
λct = exp h (Cjt − hCt−1 )−σC . (C.7)
1 + σH jt
5
In equilibrium, the marginal utility of consumption equals the Lagrange multiplier λct associated
with the household budget constraint.
 Appendix C Climate Change and Business Cycles: Technical Appendix 163

The first-order condition determines the household labour supply in each sector:
( )ι
n hN
jt
wtN = hσjtH (Cjt − hCt−1 ) , (C.8)
χN hjt
( )ι
(1 − n) hA
jt
wtA = hσjtH (Cjt − hCt−1 ) , (C.9)
χA hjt

where wtN and wtA are the real wages in the non-agricultural sector and the agricultural
sector, respectively.

The Euler condition on domestic bonds that determines the optimal consumption path
is: { }
λct+1 1 1
βEt { C } = . (C.10)
λt Et πt+1
c
Rt

Finally, the Euler condition on foreign bonds, after substituting the Lagrange mul-
tiplier, can be expressed as the real exchange rate determination under incomplete
markets: { }
et+1 ( ∗
) Rt
Et = 1 + χB pN b , (C.11)
et y,t jt
Rt∗

y,t = Py,t /Pt denotes the relative price of final goods with respect to the
where pN N C

consumption price index.

We define the real exchange rate as the ratio of final goods prices, expressed in a
common currency:
PtC∗
rert = et C , (C.12)
Pt
where PtC∗ denotes the foreign price.

3.2 Production

The firm block is populated by two groups of agents: intermediate goods firms and
final goods firms. Intermediate goods firms produce differentiated goods i ∈ [0, 1],
decide on labour on a perfectly competitive inputs market, and set prices according
to a Rotemberg (1982) technology. Final goods producers act as goods bundlers by
combining national and foreign intermediate goods to produce a homogeneous non-
tradable final good that will be sold to domestic households.
164 Appendix C Climate Change and Business Cycles: Technical Appendix

Intermediate firms are divided in two sectors s = {N, A}, where the non-agricultural
N and agricultural sectors A are of size n and 1 − n, respectively. Firms operating in
the non-agricultural sector are standard with real business cycle models, they combine
hours worked and total factor productivity (TFP) to produce a differentiated type of
goods. In addition, firms operating in the agricultural sectors combine labour, TFP,
intermediate goods as well as land to produce differentiate types of agricultural goods.
With this technology, an adverse weather shock on farm business triggers a lower
demand for final goods which in turn depresses output. Assuming n = 1 and the
model boils down to a very standard (one sector) small open economy New-Keynesian
model such as Galí and Monacelli (2005).

3.2.1 Final Firms

In each sector s = {N, A}, where N and A denote non-agricultural and agricultural
sectors, respectively, we assume that the production of the final good is performed
as in Rabanal and Tuesta (2010). A continuum of final firms purchases a composite
of intermediate home goods Xts , and a composite of intermediate foreign-produced
goods Xts∗ to produce a differentiated final good product Yts using the following CES
technology:
( )µs /(µs −1)
Yts = (1 − αs ) 1/µs
(Xts )(µs −1)/µs + αs1/µs (Xts∗ )(µs −1)/µs , for s = {N, A} ,
(C.13)
where αs denotes the share of foreign-produced goods that are used for the production
of the final good, and µs is the elasticity of substitution between domestically produced
and imported intermediate goods in both countries. A value of αs = 0 implies the
autarky of this market, while αs < 0.5 reflects a home bias in the preferences of firms.

The composite intermediate goods for the non-agricultural sector bought at home and
abroad are:
(∫ )ϵN /(ϵN −1) (∫ )ϵN /(ϵN −1)
n ( )(ϵ −1)/ϵN N∗
n ( N ∗ )(ϵN −1)/ϵN
XtN = XitN N di and Xt = Xit di ,
0 0
(C.14)
while for the agricultural sector:
(∫ )ϵA /(ϵA −1) (∫ )ϵA /(ϵA −1)
1 ( A )(ϵA −1)/ϵA 1 ( A∗ )(ϵA −1)/ϵA
XtA = Xit di and Xt =
A∗
Xit di
n n
(C.15)
 Appendix C Climate Change and Business Cycles: Technical Appendix 165

where ϵs for s = {A, N } is the elasticity of substitution between the types of inter-
mediate goods in each sector.

We consider a two-stage problem for final firms. In the first stage, they decide the
amount of imports and domestic intermediate goods by maximizing profits under the
CES technology constraint in eq. (C.13):

max s
Py,t Yts − Pts Xts − Pts∗ Xts∗ . (C.16)
{Yts ,Xts ,Xts∗ }

where Py,t
s
denotes the price of final goods Yts produced in sector s, while Xts and Xts∗
are intermediate inputs involved in the production process of final goods.

In the second stage of the problem, final goods producers decide the optimal amount
of varieties produced by intermediates firms in each sector using their packing tech-
nologies in eq. (C.14) and eq. (C.15). For domestically produced goods, the problem
reads as follows:6

∫n ∫1
max PtN XtN − 0
PitN XitN di and max PtA XtA − n
PitA XitA di. (C.17)
{ XtN ,Xit
N
} { XtA ,Xit
A
}

After solving the two-stage problem in each sector, the following first-order conditions
emerge for both sectors:
( )−µs ( s )−ϵs
Pts Pit
Xits = (1 − αs ) s
Yts , (C.18)
Py,t Pts
( )−µs ( s∗ )−ϵs
Pts∗ Pit
Xits∗ = αs et s s∗
Yts , for s = {N, A} (C.19)
Py,t Pt

Finally concerning price indexes. The price index for final goods reads as follows:

( )1/(1−µs )
s
Py,t = (1 − αs ) (Pts )1−µs + αs (et Pts∗ )1−µs , for s = {N, A} , (C.20)

while the zero-profit assumption for intermediate goods varieties packing activity de-
∫n
livers the following price index for each sectors PtN = [ n1 0 PitN 1−ϵN di]1/(1−ϵN ) and
∫ 1 A 1−ϵ
1
PtA = [ 1−n P
n it
A
di]1/(1−ϵA ) .

6
The same result is symmetrically obtained for foreign goods which are not developed here for
clarity purposes.
166 Appendix C Climate Change and Business Cycles: Technical Appendix

3.2.2 Agricultural Production and Weather Variability

To investigate the implications of weather variations as a source of aggregate fluc-

tuation, we introduce into the model a weather variable, denoted εW
t , that captures
variations in soil moisture affecting the production process of farmers. The measure
we use is based on soil moisture deficit observations calculated from the daily wa-
ter balance.7 A positive realization of εW
t depicts a prolonged episode of dryness that
damages agricultural output and generates inflation pressures. We assume that the ag-
gregate drought index follows a stochastic exogenous process driven by two shocks:

t = (1 − ρW ) + ρW εt−1 + ηt + η̃t−1 ,
εW ρW ∈ [0, 1) (C.21)
W W W

The first shock, denoted ηtW , is a traditional shock to the real business cycle that im-
pacts the level of soil moisture in the same period in which farmers see it. The second,
W
η̃t−1 , is a news shock and is differentiated from the former in that farmers observe
a weather news shock in advance (here, one quarter).8 Thus, this shock allows us
to evaluate whether farmers are anticipating drought events one quarter in advance
by capturing macroeconomic fluctuations one quarter before the realization of the
weather shock.9

To bridge weather variations with business cycle fluctuations, we define a damage

variable dt determining how variable weather conditions log(εW
t ) may induce inertial
aggregate fluctuations:

dt = ρd dt−1 + log(εW
t ), ρd ∈ [0, 1) (C.22)

where ρd captures some persistence of damage after an adverse drought event shock.
Here, it is important to disentangle parameters ρW from eq. (C.21) and ρd from eq. (C.22):
the autoregressive component ρW captures the estimated persistence of a drought
7
The soil moisture variable measures the net impact of rainfall entering the pasture root zone in the
soil which is then lost from this zone as a result of evapotranspiration or use of water by plants.
8
We follow the news-driven business cycle literature, as exemplified by Beaudry and Portier (2006),
Barsky and Sims (2011), and Schmitt-Grohé and Uribe (2012), to introduce climate-news shocks as a
source of macroeconomic fluctuation.
9
Anticipating the results from the estimation exercise, we have evaluated the ability of farmers to
expect weather shocks more than one quarter in advance ; however we find evidence that farmers are
not able to predict drought events and that they are rather surprised by weather shocks.
 Appendix C Climate Change and Business Cycles: Technical Appendix 167

shock, while ρd catches the persistence of its damages. The main underlying mo-
tivation is that damages to the economy might be more persistent than the weather
shock itself, as showed by the VAR models.10

The production component of agriculture is strongly inspired by Restuccia et al. (2008)

to the extent that agricultural output is Cobb-Douglas in land, intermediate inputs, and
labour inputs.11 In addition to this modeling choice, we introduce a damage function
ΓX (·) in the spirit of Integrated Assessment Models, which connects weather to agri-
cultural output.

Each representative firm i ∈ [n, 1] operating in the agricultural sector has the follow-
ing production function:
(( )1−σ ( ) )1−ω
A σ
XitA = εZt Zitω ΓX (dt , dt−1 ) L̄i κi Hit , (C.23)

where XitA is the production function of the intermediate agricultural good that com-
bines a (fixed) land endowment L̄i for each farmer i, labour demand HitA and non-
agricultural inputs Zit . Production is subject to an economy-wide technology shock
εZt .12 The parameter ω ∈ [0, 1] is the elasticity of output to intermediate inputs,
σ ∈ [0, 1] denotes the share of production/land in the production process of agri-
cultural goods, and κi > 0 is a technology parameter endogenously determined in the
steady state. The economy-wide technology shock εZt affects both sectors agricultural
and non-agricultural sectors by capturing fluctuations associated with declining hours
worked and prices coupled with increasing output.

Agricultural production is tied up with exogenous weather conditions through a dam-

age function ΓX (·) that alters land productivity. This function has a simple form with
one lag aiming at capturing the hump-shaped response of output to weather shock:

ΓX (dt , dt−1 ) = 1 + γ0X dt + γ1X dt−1 , (C.24)

where γ0X , γ1X ∈ (−∞, +∞) are elasticities that are estimated agnostically (i.e., without
tight priors) during the estimation exercise. In our setup, we are interested in the
10
We refer to Buckle et al. (2007) and Kamber et al. (2013) for VAR models highlighting the hysteresis
effects of weather shocks on business cycles.
11
We refer to Mundlak (2001) for discussions of related conceptual issues and empirical applications
regarding the functional forms of agricultural production. In an alternative version of our model based
on a CES agricultural production function, the fit of the DSGE model is not improved, and the identi-
fication of the CES parameter is weak.
12
Technology is characterized as an AR(1) shock process: εZ t = 1 − ρZ + ρZ εt−1 + ηt with ηt ∼
Z Z Z

N (0, σZ ), where ρA ∈ [0, 1) denotes the AR(1) term in the technological shock process.
168 Appendix C Climate Change and Business Cycles: Technical Appendix

short-run implications of weather shocks, leaving aside the neutral long run effects
( )
with ΓX d,¯ d¯ = 1, where d¯ denotes the (zero) deterministic steady state of damages
induced by drought events. The parameter γ1X captures the lagged response of out-
put after drought events, the introduction of this parameter is empirically motivated
by the time prices usually take to adjust to climate shocks, as assumed by Bloor and
Matheson (2010).

In addition to this damage function for output, inputs costs are affected by a similar
function. The real costs paid by farmers read as follows:

wtA HitA + pN
t Zit ΓX (dt , dt−1 ) , (C.25)

where wtA is the real wage offered to households hired in the agricultural sector, and
t = Pt /Pt denotes the relative price of intermediate goods, with Pt as the con-
pN N C C

sumer price index. The demand for intermediate goods Zit is affected by ΓX (dt , dt−1 )
which aims at capturing extra-consumption of intermediate goods following a drought
event. A drought shock increases the feed budget, as dairy cattle requires more water as
temperature, humidity and production levels rise. Farming activities also demand more
water to offset soil dryness by increasing field irrigation. This damage function cap-
tures the demand effects in the intermediate sector, and the shape of this damage func-
tion reads as in eq. (C.24) with different elasticities denoted γ0Z and γ1Z ∈ (−∞, +∞).

To introduce nominal rigidities, we assume that firms must solve a two-stage problem.
In the first stage, the real input price wtN is taken as given, firms rent inputs HitN and
Zit in a perfectly competitive factor markets in order to minimize costs subject to the
production constraint. Each firm maximizes profits:

it Xit − wt Hit − ΦZ (dt , dt−1 ) pt Zit

max mcA A A A N

{Zit ,HitN }
[ (( ]
)1−σ ( A σ
) )1−ω
+ λA
t εZt Zitω ΦX (dt , dt−1 ) L̄i κi Hit − Xit ,
A

under the supply constraint in eq. (C.23). The variable mcA

it denotes the real marginal
cost of producing an additional agricultural good.

The cost-minimization problem ensures that the real agricultural wage is directly
driven by the marginal product of labour:

XtA
t (1 − ω) σ
wtA = mcA . (C.26)
HtA
 Appendix C Climate Change and Business Cycles: Technical Appendix 169

The second cost-minimizing condition is obtained from the marginal product of in-
termediate consumption Zt and provides the optimal demand for intermediate goods
from the farmer:
mcA
Zt = ω t
XtA . (C.27)
ΦZ (dt , dt−1 ) pN
t

In the second stage, the intermediate firm operates monopolistically and sets the re-
tail price according to a Rotemberg (1982) technology. Intermediate good firms face
adjustment costs with price changes ACitA defined according to:
( )2
κA PitA
ACitA = A
− (πt−1
A
)ξA ,
2 Pit−1

where κA is the cost of adjusting prices, and ξA is the coefficient that measures the rate
of indexation to the past rate of inflation of intermediate goods, πt−1
A A
= Pt−1 A
/Pt−2 .
These costs are paid in terms of the final goods at a market price of Py,t
N
. Given this
price adjustment cost specification, the problem of the representative firms becomes
dynamic:

∑
+∞ c τ [ A
]
λt+τ Pit+τ
Et β Xit+τ − mcit+τ Xit+τ − py,t+τ Yt+τ ACit+τ = 0,
A A A N A A
(C.28)
τ =0
λct C
Pt+τ

The variables mcA

it and py,t are the real marginal cost and the relative price of non-
A

agricultural final goods, respectively. Since firms are owned by households, they dis-
count expected profits using the same discount factor as households (β τ λct+τ /λct ).13
The firm faces the downward sloping constraint from final good producers obtained
from eq. (C.17):
( )−ϵA
PitA
XitA = XtA . (C.29)
PtA
Anticipating symmetry between firms with PtA = PitA , first-order condition is:

YtA ( A ) A
(1 − ϵA ) pA
t + ϵA mct − py,t
A N
κA π t − (π A
t−1 )ξA
πt
XtA
{ c A (
}
λt+1 N Yt+1 ) A
+ κA βEt p π − (πt ) πt+1 = 0. (C.30)
A A ξA
λct y,t+1 XtA t+1

13
The stochastic discount factor is endogenously determined by the Euler condition of households.
In equilibrium, the stochastic discount is inversely related to the real interest rate.
170 Appendix C Climate Change and Business Cycles: Technical Appendix

3.2.3 Non-Agricultural Intermediate Production

Each representative intermediate firm i ∈ [0, n] has the following technology:

XitN = εZt HitN , (C.31)

where XitN is the production of the ith intermediate firm that combines labour demand
Hit and technology εZt .

Intermediate goods producers solve a two-stage problem. In the first stage, the real
input price wtN is taken as given, and these firms rent inputs HitN in a perfectly com-
petitive factor markets in order to minimize costs subject to the production constraint:

[ N ]
it Xit − wt Hit + λt Xit − εt Hit
max mcN N N N n Z N
{Xit
N ,H N }
it

The first-order condition leads to i) mcN

it = λt and ii) the real marginal cost expression:
n

wtN
mcN N
it = mct = . (C.32)
εZt

In the second stage, the intermediate firm operates monopolistically and sets the re-
tail price according to a Rotemberg (1982) technology. Intermediate good firms face
adjustment costs on price changes, ACitN defined according to:
( )2
κN PitN
ACitN = N
− (πt−1
N ξN
)
2 Pit−1

where κN is the cost of adjusting prices and ξN is the coefficient that measures the
rate of indexation to the past rate of inflation of intermediate goods πt−1
N N
= Pt−1 N
/Pt−2 .
These costs are paid in terms of final goods at a market price Py,t
N
. Given this price ad-
justment cost specification, the problem of the representative firms becomes dynamic:

∑
+∞ c τ [ N
]
λt+τ Pit+τ
Et β Xit+τ − εt+τ mciN ont+τ Xit+τ − py,t+τ Yt+τ ACit+τ ,
N N N N N N N
(C.33)
τ =0
λct C
Pt+τ

t is an AR(1) markup shock that aims at capturing the external factors driving
where εN
the inflation rate, which are not included in the model such as commodity prices.
it and py,t+τ are the real marginal cost and the relative price of non-
The variables mcN N

agricultural final goods, respectively.

 Appendix C Climate Change and Business Cycles: Technical Appendix 171

Firms operate in a monopolistically competitive market as in Dixit and Stiglitz (1977).

Hence, the amount of firm-specific output, XitN , is demand-determined in response to
its relative price PitN /PtN and to the aggregate demand for goods, XtN , as obtained
from eq. (C.17):
( )−ϵN
PitN
XitN = XtN . (C.34)
PtN

Anticipating symmetry between firms with PtN = PitN , the first-order condition reads:

N Yt
N ( ) N
(1 − ϵN ) pN
t + ϵN ε N
t mc N
t − p y,t κ πtN − (πt−1
N N
N ξN
) πt
Xt
{ c N (
}
λt+1 N Yt+1 ) N
+ κN βEt p π − (πt )
N N ξN
πt+1 = 0 (C.35)
λct y,t+1 XtN t+1

3.3 Monetary Policy

The central bank reacts to fluctuations in price, activity and external imbalance. The
general expression of the linear interest rule implemented by the central bank can be
expressed as:

( )1−ρ [ ](1−ρ) ( )
D ϕ∆Y R
Rt = R̄ (Rt−1 )ρ (πt )ϕπ (RERt )ϕE YtD /Yt−1 εt , (C.36)

where R̄ is the steady-state interest rate, YtD is gross domestic product, εR

t is an exo-
genous AR(1) monetary policy shock,14 ϕπ , ϕE and ϕ∆Y denote inflation, real ex-
change rate and GDP growth gap parameters, respectively, that aim to stabilize the
economy when it deviates from its steady state. In a small economy context, we fol-
low the definition of monetary policy rules in open economies of Clarida et al. (1998)
and estimate ϕE . A positive value of ϕE induces a reduction in the variance of the real
exchange rate.

3.4 Foreign Economy

Our foreign economy is characterized by a set of five equations that aims at capturing
the standard business cycle patterns of the foreign economy. Four equations are taken
from the standard New Keynesian framework, namely, the Phillips curve, the IS curve,
the Taylor rule, and the CES substitution curve between two types of goods. These
14
The monetary policy shock follows a standard AR(1) stochastic process: εR R
t = (1−ρR )+ρR εt−1 +
t ,
εR with ηt ∼ N (0, σR ), and 0 ≤ ρR < 1 the autoregressive term.
R
172 Appendix C Climate Change and Business Cycles: Technical Appendix

equations provide the structural relations between aggregate output Yt∗ , agricultural
output YtA∗ , inflation πt∗ and the nominal interest rate Rt∗ .

The first relation is the New Keynesian Phillips Curve that links current inflation (πt∗ )
∗
to expected inflation (Et πt+1 ) and to the output gap (Yt∗ ):

{( ) ∗ }
(1 − ϵN ) + ϵN χ∗ Yt∗ − κ∗ (πt∗ − 1) πt∗ + κ∗ βEt ∗
πt+1 − 1 πt+1 =0 (C.37)

In this expression, β is the psychological discount factor, χ∗ is the Rotemberg price

adjustment cost, and ϵN is the imperfect substitution between varieties. This relation
comes from the aggregation of the supply decision of firms that have a market power
and can optimize prices with adjustment costs generating nominal rigidities.

The second relation is the intertemporal (dynamic) IS curve. This schedule is a log-
linearization of the Euler bond equation that describes the intertemporal allocation of
consumption of agents in the economy over the cycles. This relation plays the same
role as the IS curve in the IS-LM model by determining the output gap equation Yt∗
through: { }
Yt∗ 1 εt
βEt ∗ ∗
= (C.38)
Yt+1 πt+1 Rt∗
where εYt ∗ is a demand shock characterized by an iid AR(1). This time-preference
shock lowers the discount factor and forces the foreign household to increase its spend-
ing in terms of consumption goods.

The third relation is the Taylor rule that links the nominal interest rate Rt∗ controlled
by monetary authorities to the average inflation rate (πt∗ and PtA∗ /Pt−1
A∗
with sec-
∗
toral weight n) and to the output gap (Yt+1 ). Monetary policy inertia is accounted
∗
for through the previous period rate of interest (Rt−1 ). This rule is defined as:

( )ρ∗ ( ( ) )(1−ρ∗ )
A∗ 1−n ϕ∗π ∗
βRt∗ = ∗
βRt−1 ((πt∗ )n pA∗
t /p t−1 ) (Yt∗ /Ȳ ∗ )ϕy . (C.39)

In this expression, ρ∗ is the autocorrelation parameter, ϕ∗π is the elasticity of the nom-
inal interest rate to the inflation rate, and ϕ∗y is the elasticity of the nominal interest
rate to the output gap. We multiply by β the interest rate to get a balanced steady state
(as R∗ =β −1 ). In addition, pA∗
t /pt−1 denotes the variations of the relative price index of
A∗

agricultural goods.
 Appendix C Climate Change and Business Cycles: Technical Appendix 173

The fourth equation determines the demand for agricultural goods by foreign house-
holds. This equation is a reduced-form equation modelling households preferences by
substituting agricultural and non-agricultural goods via:

( )−µ
YtA∗ φ pA∗
t
∗
= ,
Yt 1−φ A∗
pt−1
where pA∗
t is the relative price between goods, parameter φ is the share of agricultural
goods in the consumption basket, and µ is the substitution parameter as defined in
eq. (C.2). This equation shows that the household’s consumption allocation is determ-
ined by the gap between variations in the relative price index between agricultural
and non-agricultural goods.

Finally, the foreign agricultural price is too volatile to be determined by a New Keyne-
sian Phillips curve. We assume the relative price of foreign agricultural goods is de-
termined by an AR(1) shock process:

∗ ∗ A∗ ∗2
t = 1 − ρA + ρA εt−1 + ηt with ηt ∼ N (0, σA ),
εA∗ (C.40)
A∗ A∗

t = εt .
with pA∗ A∗

In addition, the second exogenous shock affecting the IS-curve reads:

εYt ∗ = 1 − ρ∗Y + ρ∗Y εYt−1

∗
+ ηtY ∗ with ηtY ∗ ∼ N (0, σY∗2 ). (C.41)

In the long run, the economies are perfectly symmetric with Ȳ =Ȳ ∗ , Ȳ A =Ȳ A∗ , R̄=R̄∗ ,
π=π ∗ and pA =pA∗ .

3.5 Stochastic Shock Processes

To be in line with the seminal contribution of Smets and Wouters (2003), all our random
processes s = {Z, N, D, R, Y ∗ , A∗ } follow an AR(1) specification defined by:

εst = ρs εst−1 + ηts with ηts ∼ N (0, σs ). (C.42)

174 Appendix C Climate Change and Business Cycles: Technical Appendix

3.6 Shocks, Aggregation and Equilibrium Conditions

After (i) aggregating all agents and varieties in the economy, (ii) imposing market
clearing on all markets, and (iii) substituting the relevant demand functions, we can
deduct the general equilibrium conditions of the model.

First, total demand for non-agricultural goods is as follows:

( )−µ
N
Py,t
YtN = (1 − φ) Ct + ΦB (b∗jt ) + ACtN YtN + ACtA YtA , (C.43)
PtC

while the equilibrium in the intermediate goods market after aggregation is determ-
ined by:
( )−µN ( )−µN
PtN 1 PtN
nXtN = (1 − αN ) N
YtN + αN N∗
YtN ∗ + (1 − n) Zt , (C.44)
Py,t et Py,t
∫n ∫1
where nXtN = 0
XitN di is the aggregate supply, and (1 − n) Zt = n
Zit di denotes
the aggregate demand for domestic intermediate goods from farmers.

Similarly for the agricultural sector, the aggregate demand is:

( )−µ
A
Py,t
YtA = φ Ct , (C.45)
PtC

and equilibrium in the intermediate market is achieved by the following clearing mar-
ket condition:
( )−µA ( )−µA
PtA 1 PtA
(1 − n) XtA = (1 − αA ) A
YtA + αA A∗
YtA∗ . (C.46)
Py,t et Py,t

Turning to the labour market, the market clearing condition between household labour
supply and demand from firms in each sector is:
∫ 1 ∫ n ∫ 1 ∫ 1

jt dj
hN = HitN di and jt dj
hA = HitA di. (C.47)
0 0 0 n

The law of motion for the total amount of real foreign debt is:
∗
Rt−1 ( ) ( )
b∗jt = C ∆et b∗jt−1 + n pN
t Xt − py,t Yt
N N N
+ (1 − n) pA
t Xt − py,t Yt − pt Zt .
A A A X
πt
(C.48)
 Appendix C Climate Change and Business Cycles: Technical Appendix 175

Table C.1: Notations of Variables and Parameters for the Household Block

Variable Description

Variables

Cjt Consumption index

N
Cjt Consumption of non-agricultural goods
A
Cjt Consumption of agricultural goods
PtC Consumption price index
PtC∗ Foreign consumption price index
N
Py,t Price index of final non-agricultural goods
A
Py,t Price index of final agricultural goods
hjt labour disutility index
hN jt Hours worked in the non-agricultural sector
hA jt Hours worked in the agricultural sector
wtN Real hourly wage in the non-agricultural sector
wtA Real hourly wage in the agricultural sector
Πjt Profits generated by imperfect competition in goods in both sectors
bjt Domestic bonds
b∗jt Foreign bonds
Rt Interest rate on domestic bonds
Rt∗ Interest rate on foreign bonds
πtC Inflation rate of Consumption goods
et Nominal exchange rate
pNy,t Relative price of final non-agricultural goods wrt consumption price index
rert Real exchange rate
λct Marginal utility of consumption

Parameters

µ Substitution elasticity between the two goods

h Consumption habits parameter
σC Consumption risk aversion
σH Labour disutility
φ Share of agricultural goods in household consumption basket
n Relative share of employment in the non-agricultural sector
ι Hours worked sectoral adjustment cost
χB International financial market cost
χN Scale parameter in non-agricultural sector
χA Scale parameter in agricultural sector

And the real GDP can be computed either by the demand side (YtD ) or by the supply
side (YtS ):

YtD = npN
y,t Yt + (1 − n) py,t Yt
N A A
(C.49)
( A A )
YtS = npN
t X t
N
+ (1 − n) p t X t − p N
t Zt . (C.50)

Finally, the general equilibrium condition is defined as a sequence of quantities {Qt }∞

t=0
and prices {Pt }∞ ∞
t=0 such that for a given sequence of quantities {Qt }t=0 and the real-
ization of shocks {St }∞ ∞
t=0 , the sequence {Pt }t=0 , guarantees simultaneous equilibrium
on all markets previously defined.
176 Appendix C Climate Change and Business Cycles: Technical Appendix

Table C.2: Notations of Variables and Parameters for the Final Firms Block

Variable Description

Variables

N
Py,t Consumption price index of goods produced in the non-agricultural sector
A
Py,t Consumption price index of goods produced in the agricultural sector
PtC∗ Foreign consumption price index
PtN Intermediate price of non-agricultural goods
PtA Intermediate price of agricultural goods
XtN Intermediate production of non-agricultural goods
XtA Intermediate production of agricultural goods
XtN ⋆ Foreign intermediate goods in the non-agricultural sector
XtA⋆ Foreign intermediate goods in the agricultural sector
YtN Final non-agricultural goods
YtA Final agricultural goods

Parameters

αN Share of imports in the production of final non-agricultural goods

αA Share of imports in the production of final agricultural goods
ϵN Elasticity of substitution between non-agricultural varieties
ϵA Elasticity of substitution between agricultural varieties
Elasticity of substitution between domestically produced and imported intermediate non-agricultural
µN
goods
µA Elasticity of substitution between domestically produced and imported intermediate agricultural goods
n Relative share of employment in the non-agricultural sector

4 Estimation
We apply standard Bayesian estimation techniques as in Smets and Wouters (2003,0).
In this section, we describe the data sources and transformations.

4.1 Data

The model is estimated using 7 time series with Bayesian methods and quarterly data
for New Zealand over the sample time period 1989:Q1 to 2014:Q2. All data is in
log-difference except interest rate and climate. The time reference for all indexes is
2010:Q1. Transformed data is shown in fig. C.3.

• Gross domestic product: real per capita output, expenditure approach, season-
ally adjusted. Source: Statistics New Zealand.

• CPI inflation: all groups index, Source: Statistics New Zealand.

• Agricultural output: real agriculture, fishing and forestry gross domestic product,
seasonally adjusted. Source: Statistics New Zealand.
 Appendix C Climate Change and Business Cycles: Technical Appendix 177

Table C.3: Notations of Variables and Parameters for the Intermediate-good Firms
Block

Variable Description

Variables

ACit N Adjustment costs faced by non-agricultural intermediate good firms

dt Weather damage variable
HitN labour demand in non-agricultural intermediate good firms
Hit A labour demand in agricultural intermediate good firms
L̄i Land endowment
mcN t Marginal cost in the non-agricultural sector
mctA Marginal cost in the agricultural sector
Pt C Consumption price index
PtN Intermediate price index of non-agricultural goods
PtA Intermediate price index of agricultural goods
wtN Real hourly wage in the non-agricultural sector
wtA Real hourly wage in the agricultural sector
pNt Relative price of intermediate non-agricultural goods wrt consumption price index
pA
t Relative price of intermediate agricultural goods wrt consumption price index
XtN Intermediate non-agricultural goods
XtA Intermediate agricultural goods
Zit Non-agricultural input in the agricultural production
ηtW Weather shock
W
η̃t−1 News shock (weather shock observed in advance)
εWt Drought index
εZ
t Technology shock hitting the agricultural production function
πtN Inflation rate of production price of non-agricultural goods
πtA Inflation rate of production price of non-agricultural goods

Parameters

β Discount factor
X, Parameters of the contemporaneous and lagged effects of the weather damage, respectively (for the
γO γ1X
damage function directly hitting the agricultural production function)
Z , γZ Parameters of the contemporaneous and lagged effects of the weather damage, respectively (for the
γO 1 damage function affecting the demand for inputs)
ϵN Elasticity of substitution between non-agricultural varieties
ϵA Elasticity of substitution between agricultural varieties
κN Cost of adjusting prices for non-agricultural intermediate firms
κA Cost of adjusting prices for agricultural intermediate firms
κi Technology scale parameter in the agricultural intermediate good firm
ξN Rate of indexation of the past rate of inflation of intermediate non-agricultural goods
ξA Rate of indexation of the past rate of inflation of intermediate agricultural goods
ρW Autoregressive parameter capturing the persistence of a drought
ρd Autoregressive parameter capturing the persistence of the damages of the drought
σ Share of production/land in the production process of agricultural goods
ω Elasticity of output to intermediate input in agricultural intermediate good firms

• Agricultural producer price inflation: agriculture, fishing and forestry produ-

cer price index. Source: Statistics New Zealand.

• Population: actual population of working age, in thousands, seasonally adjus-

ted. Source: Statistics New Zealand.

• Interest rate: 3-Month rates and yields: bank bills for New Zealand, not season-
ally adjusted. Source: Main Economic Indicators, OECD.
178 Appendix C Climate Change and Business Cycles: Technical Appendix

Table C.4: Notations of Variables and Parameters for Remaining Blocks

Variable Description

Variables

pA∗
t Relative price index of foreign agricultural goods
Rt Nominal interest rate
Rt∗ Nominal foreign interest rate
R̄ Steady-state interest rate
R̄∗ Steady-state foreign interest rate
rert Real exchange rate
YyD Gross domestic product (demand side)
Yt∗ Aggregate foreign output
YtA∗ Aggregate foreign agricultural output
ϵRt∗ Monetary policy shock
ϵYt Foreign demand shock
πt∗ Foreign inflation rate

Parameters

β Discount factor
ϵN Elasticity of substitution between non-agricultural varieties
ϵA Elasticity of substitution between agricultural varieties
µ Substitution elasticity between the two types of goods
ϕπ Inflation reaction parameter
ϕ∗π Foreign inflation reaction parameter
ϕE Real exchange rate reaction parameter
ϕ∆Y Output-gap growth reaction parameter
ϕ∗y Elasticity of the nominal interest rate to the output gap (for foreign authorities)
κ∗ Cost of adjusting prices in foreign firms
rho∗ Foreign Taylor rule smoothing parameter
φ Share of agricultural goods consumed in total consumption
n Relative share of employment in the non-agricultural sector

• Real exchange rate: real trade weighted index. Source: Reserve Bank of New
Zealand.

• Climate: soil moisture deficit at the station level. Source: National Climate Data-
base, National Institute of Water and Atmospheric Research.

4.1.1 The Weather Measure

The measure of weather we use is an index of drought constructed following the meth-
odology of Kamber et al. (2013). It is based on soil moisture deficit observations15 and
is collected from the National Climate Database from National Institute of Water and
Atmospheric Research. Raw data is obtained from weather stations at a monthly rate.
The spatial covering of these stations is depicted in fig. C.2(a), while its temporal cov-
ering is represented in fig. C.2(b). To get quarterly national representative data, both
spatial and time scales need to be changed. In a first step, we average the monthly
values of mean soil moisture deficit at the region level. We then remove a seasonal
15
Named “MTHLY: MEAN DEFICIT (WBAL)” in the database.
 Appendix C Climate Change and Business Cycles: Technical Appendix 179

trend by simply subtracting long term monthly statistics. Long term statistics are eval-
uated as the average value over the 1980 to 2015 period. Then, we follow Narasimhan
and Srinivasan (2005) to create the soil moisture deficit index. In a nutshell, for each
m = {1, . . . , 12} month in each t = {1980, . . . , 2015} year, we compute monthly soil
water deficit (expressed in percent) as:

SWt,m − M ed(SWm )
SDt,m = . (C.51)
M ed(SWm )

The index for any given month is then computed as:

SDt,m
SM DIt,m = 0.5 × SM DIt,m−1 + , (C.52)
50
SD1980,m
using SM DI1980,m = 50
, m = {1, . . . , 12} as initial values for the series.

Then, we aggregate the monthly values of the index at the national level by means of
a weighted mean, where the weights reflect the share of yearly agricultural GDP of
each region.16 In a final step, monthly observations are quarterly aggregated.

500

-35 Auckland
Bay of Plenty
Canterbury 450
Gisborne
Hawke’s Bay
Manawatu-Wanganui
-40
Marlborough 400
Northland
Otago
Southland
Taranaki
350
Tasman/Nelson
-45

Waikato
Wellington
300
West Coast
170 175

1982 1986 1990 1994 1998 2002 2006 2010 2014

(a) Spatial covering (b) Temporal covering

Figure C.2: Covering of Weather Stations Used to Construct the Soil Moisture Deficit
Index

16
The regional agricultural GDP data we use ranges from 1987 to 2014. The weight before 1987 and
after 2014 is set to the average contribution of the region to the total agricultural GDP over the whole
covered period.
180 Appendix C Climate Change and Business Cycles: Technical Appendix

4.1.2 The Climate Scenarios

To estimate the variability of the weather process ηtW , we rely on simulated weather
data from a circulation climate model, the Community Climate System Model (CCSM).
We consider the data simulated under the four well-employed Representative Con-
centration Pathways (RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5). They are given on a
0.9◦ × 1.25◦ grid, at a monthly rate, for two distinct periods. The first one corresponds
to “historical” values, and ranges from 1850 to 2005. The second one gives observa-
tions for “future” values up to 2100. Since our DSGE models is fed-up with quarterly
data at the national level, we need to aggregate the raw data provided by the CCSM.
To do so, we compute the average values of total rainfall at the region level by means
of a weighted mean. The weight put on each cell of the grid in a given region is the
proportion of the region covered by the cell. Values are then averaged for each month,
at the national level. The aggregation is done using a weighted mean, where weights
are set according to the share of agricultural GDP of the region.17 Resulting data is
then converted to quarterly data, by summing the monthly values of total rainfall.
The final dataset of simulated data contains quarterly data of rainfall at the national
level for the historical period (ranging from 1983 to 2005) and for the future period
(covering 2006 to 2100) for each RCP scenario.

We then need to estimate how the variance of the weather shock changes through
time in each of the i = {RCP 2.6, RCP 4.5, RCP 6.0, RCP 8.5} scenario. We proceed by
rolling window regression, the size of each window being set to 102 quarters, matching
the size of the number of observations used to estimate the DSGE model. In each step
of the rolling window regression, we fit an AR(1) model to the data and compute
the standard deviation of the residuals. We estimate the growth rate of the standard
deviation ∆σi,εW by least squares, regressing the natural logarithm of the standard
deviation previously obtained on time. Then, we estimate the average growth rate
∆σi,εW of the standard deviation over the 1989–2100 period for the ith scenario as:

∆σi,εW = (1 + σi,εW )q − 1, (C.53)

where σi,εW is the estimated compound quarterly rate of growth for the standard error
of the weather shock process under the ith climate change scenario, and q is the number
of quarter in the whole sample, i.e., 347. table C.5 summarizes the estimates.
17
The regional agricultural GDP data we use ranges from 1987 to 2014. The weight before 1987 and
after 2014 is set to the average contribution of the region to the total agricultural GDP over the whole
covered period.
 Appendix C Climate Change and Business Cycles: Technical Appendix 181

Table C.5: Estimations of Growth Rates of Standard Errors of the Weather Process
Under Different Scenarios

Compound quarterly rate Average growth rate

Scenario
(σi,εW ) (∆σi,εW )
RCP 2.6 −0.1271890 × 103 −4.269057
RCP 4.5 0.1897154 × 103 6.722968
RCP 6.0 0.2581432 × 103 9.256886
RCP 8.5 0.6176223 × 103 23.587480
Notes: For each Representative Concentration Pathways, we estimate the quarterly rate of growth of the standard deviation of
the weather measure (σi,εW ), and the corresponding average growth rate over the whole 1989–2100 period (∆σi,εW ).

4.1.3 Macroeconomic Time Series Transformation

Concerning the transformation of series, the point is to map non-stationary data to a

stationary model (namely, here, GDP Yt ). The data that are known to have a trend or
unit root are made stationary in two steps. First, we divide the sample by the civilian
population, denoted Ni,t . Second, data are taken in log and we use a first difference
filtering to obtain growth rates. Real variables are deflated by GDP deflator price index
denoted Pt .

As an illustration, the calculation method used to detrend real GDP growth per capita
is as follows: ( ) ( )
Ybt Ybt−1
∆Ybtrl = log − log ,
P̂t Nt P̂t−1 Nt−1
bt denote the real, the per capita, and the log value of Xt , respect-
where Xtr , Xtl , and X
ively.

Hours worked are divided by civilian population to improve the identification of labour
demand, as in Smets and Wouters (2007):

btl = log(Ht ) − log(Nt )

H

Turning to the weather index, we simply apply the logarithm function:

Sbt = log(SM DIt )

Finally, we demean the data because we do not incorporate trends in our model. We are
aware that the introduction of trends could affect our estimation results. However for
tractability reasons, we have chosen to focus on short run macroeconomic fluctuations
 182 Appendix C Climate Change and Business Cycles: Technical Appendix

and to neglect long run effects involved with trends. Such an approach has also been
chosen by Smets and Wouters (2003).

Foreign Output ∆log (Yt∗ ) Foreign CPI Inflation πt∗ Foreign Interest Rate rt∗ Foreign Ag. Price Infl. ∆log pA∗


t
6
3 3

1
2 3

2
0 1
0

-1 0
1
-3
-1
-2
1990 2000 2010 1990 2000 2010 1990 2000 2010 1990 2000 2010

Foreign Ag. Output ∆log YtA∗

Output ∆log Ytd


CPI Inflation πtC Ag. Output ∆log XtA

3 10
10
3

2 5

0 0
0
1

-3 -5
-10
0

-10
-6 -1
1990 2000 2010 1990 2000 2010 1990 2000 2010 1990 2000 2010
A


Ag. Inflation log πx,t Real Exchange Rate rert Interest Rate rt Climate εC
t
4 2

5 4
1
2
0 3
0

-5 2
0 -1

-10
1
-2
1990 2000 2010 1990 2000 2010 1990 2000 2010 1990 2000 2010

Figure C.3: Observable Variables Used in the BSVAR and the DSGE Estimations

4.2 Measurement Equations of the DSGE Model

The final dataset includes ten times series: real per capita growth rate of the GDP
(demand side) ∆Ybrl , real capita growth rate of agricultural output ∆YbArl , per capita
t t
hours worked btl ,
H
as well as quarterly money market rate Rt (on an annual basis),
inflation rate ∆P̂t , inflation rate for agricultural goods ∆P̂tA , and a weather index Sbt .
fig. C.3 plots the transformed data.

Measurement equations read as follows:

   
100*Hbtl t /h̄ ) + (1 − n) log(ht /h̄ )
n log(hN N A A
   
 100*∆Ybtrl   log(YtD /Yt−1
D
) 
   
 bt   
 100*∆P   log(πt ) 
   
 100*∆YbArl = log(XtA /Xt−1
A
) .
 t   
 b   
 100*∆PtA   log(πy,t
A
) 
   
 100*Sb   log(εSt ) 
 t   
100*Rt (Rt − R̄)
 Appendix C Climate Change and Business Cycles: Technical Appendix 183

5 The Non-Linear Model

5.1 Equations Summary

The marginal utility of consumption is given by:

( )
σC − 1 1+σH
λC
t = exp ht (C t − h C t−1 )(−σC ) .
1 + σH

The Euler equation is given by:

rt
λC C
t = β λt+1 .
πt+1

The real exchange rate is obtained by:

πt∗
RERt = ∆et RERt−1 .
πt

Variations of the expected nominal exchange rate is obtained by:

rt ( N ∗
)
∆et+1 = 1 + χ B y, t t .
p b
rt∗

The labour supply equation in each sector is:

( )ι
n σH hN t
wtN= (ct − h ct−1 ) ht ,
χN ht
( )ι
1 − n hA t
wt = ht (ct − h ct−1 )
A σH
.
χA ht

The labour effort disutility index generating costly cross-sectoral labour re-allocation:

1+ι 1+ι
h1+ι
t = n (hN
t ) + (1 − n) (hA
t ) .

Non-agricultural and agricultural production functions are given by:

XtN = hN t εZt ,
((( ) )1−σ ( A )σ )1−ω
XtA = εZt Zt ω 1 + γ0X Dt + γ1X Dt−1 L̄ h tκ .
184 Appendix C Climate Change and Business Cycles: Technical Appendix

Real marginal products of labour determining the real wage for each sector are:

mcNt
wtN = XtN ,
hN t
X A σ (1 − ω) mcA
wtA = t t
.
hA t

Real marginal product of intermediate goods is:

XtA ω mcA t
Zt = .
(1 + γ0Z dt + γ1Z dt−1 ) pN
t

The Rotemberg sticky price device:

( )
N ξN
x,t (1 − ϵN ) + Xt n mct ϵN εt − κN py,t πt
XtN n pN πtN − (πt−1
N N N N N
) ytN
λC ( )
N ξN
+ κN β t+1
py,t+1 πt+1 πt+1 − (πt )
N N N N
yt+1 =0
λC
t

The Rotemberg sticky price device for the other sector:

( )
ξA
XtA (1 − n) (1 − ϵA ) pA
x, t + X A
t (1 − n) mc A
ϵ
t A − κ p N A
π
A y,t t π A
t − (π A
t−1 ) ytA
λC ( )
A ξA
+ κA β C py,t+1 πt+1 πt+1 − (πt )
t+1 N A A A
yt+1 =0
λt

The intermediate goods equilibrium on each sector:

( )−µN ( N )−µN
pN pt
n XtN = (1 − αN )
ytN t
N
+ αN yt∗ + (1 − n) Zt ,
py,t RERt
( A )−µA ( A )−µA
pt pt 1
(1 − n) Xt = yt (1 − αA )
A A
+ αA ytA∗
pA
y,t ε A∗
t RER t

Relative production and final price indexes are respectively given by:

pNt πtN
= ,
pNt−1 πt
pAt πtA
= ,
pAt−1 πt
pN
y,t
N
πy,t
= ,
pN
y,t−1 πt
A A
py,t πy,t
= ,
pA
y,t−1 πt
 Appendix C Climate Change and Business Cycles: Technical Appendix 185

while CES price indexes for final goods in real terms are:

1−µN 1−µN
pN
y,t = (1 − αN ) pN
t + αN RERt 1−µN ,
1−µA 1−µA ( )1−µA
pA
y,t = (1 − αA ) ptA + αA RERt pA∗
t ,

and the consumption price index in real terms:

1−µ 1−µ
1 = (1 − φ) pN
y,t + φ pA
y,t ,

and the law of motion of net foreign assets:

∗
rt−1
b∗t = ∆et b∗ + n XtN pN
t − py,t yt
N N
πt t−1
t + (1 − n)Xt pt − yt py,t ,
− Zt (1 − n) pN A A A A

which allows us to get the quarterly current account, computed as the variations of
the external assets position:
cat = b∗t − b∗t−1 .

The resource constraint for final goods is given by:

( )2
−µ N ξN
ytN = (1 − φ) pN
y,t ct + ytN κN 0.5 πtN − πt−1
( )2
A ξA
+ 0.5κA πtA − πt−1 ytA + 0.5χB (b∗t )2 ,

and for agricultural goods:

−µ
ytA = φ εD A
t py,t ct .

The two-sector set-up allows us to disentangle GDP computation by the demand and
the supply side:

YD N N A A
t = py,t yt + py,t yt ,

t xt + (1 − n)(pt xt − pt Z t ).
Y St = npN N A A N

The damage law of motion is given by:

( )
dt = ρd dt−1 + log εSt .
186 Appendix C Climate Change and Business Cycles: Technical Appendix

The interest rate monetary policy rule:

(r ) (r ) ( ( d) )
t t−1 yt
log = ρ log + (1 − ρ) ϕy log ¯d + ϕπ log (πt )
r̄ r̄ y
( d )
yt ( )
+ ϕE log (RERt ) + ϕ∆y log d
+ log εR
t
yt−1

The foreign economy structural equations are given by (i) the following Euler equa-
tion:
∗
yt+1 rt∗ 1
= β ,
yt∗ ∗
πt+1 εYt ∗
as well as (ii) the following supply equation:

( ∗ )
1 − ϵN + ϵN χ∗ yt∗ − κ∗ (πt∗ − 1) πt∗ + β κ∗ πt+1
∗
πt+1 − 1 = 0,

(iii) the following monetary policy equation:

( ( ))
( ∗ ) εA∗
log (β rt∗ ) = ρ∗ log β rt−1 + (1 − ρ∗ ) ϕ∗π n log (πt∗ ) + (1 − n) log t
εA∗
t−1
( ∗)
∗ ∗ y
+ (1 − ρ ) ϕy log t∗ ,
y¯

(iv) the following substitution equation determining the demand for agricultural goods:
( )−µ
ytA∗ φ pA∗
t
∗
= ,
yt 1−φ pA∗
t−1

and (v), two stochastic shock processes:

∗ ∗ A∗
t = 1 − ρA + ρA εt−1 + ηt
εA∗ P A∗
,
∗
εYt ∗ = 1 − ρ∗Y + ρ∗Y εYt−1 + ηtY ∗

6 Business Cycle Facts About Climate Shocks

To observe how the economy responds to a weather shock, we develop an empirical
framework, a Bayesian Structural VAR, and analyze the impulse response functions
following a drought shock.
 Appendix C Climate Change and Business Cycles: Technical Appendix 187

6.1 Modeling Framework

Let us consider three blocks of equations: a first one representing a domestic small
open economy, a second one representing domestic weather and a third one repres-
enting the international economy. The model writes:
       
A11 A12 A13 XtD B 11
B 12
B 13
X D
ηD
   ∑p
 l l l
  t−l   t 
 0 A22 0  X W  =  0 B 22 0  X W  + CIt + η W  ,
  t   l   t−l   t 
∗ ∗
0 0 A 33
Xt l=1
0 0 Bl 33
Xt−l ηt∗
(C.54)

where t = 1, . . . , T is the time subscript, p is the lag length,18 XtD , XtW and Xt∗ are
column vectors of variables for the small open economy, the climatic block and the rest
of the world respectively. The column vector It contains the j exogenous variables,
including the constant. The error terms ηtD , ηtW and ηt∗ are exogenous and independent
D W ∗
with zero mean and variance σ η , σ η , and σ η , respectively. The coefficients in A11
to A33 , Bl11 to Bl33 , and C are the parameters of interest.

For our New Zealand economy model, the domestic block is:
[ ( ) ( ) ( A) ]′
XtD = ∆log Ytd πtC ∆log XtA log πx,t rert rt ,
( ) ( ) ( A)
where ∆log Ytd is for real GDP growth, πtC for prices, ∆log XtA and log πx,t for
agricultural real output growth and prices, respectively, rert for the nominal exchange
rate, and rt for interest rate. The weather block writes:
[ ]′
XtW = εW
t ;

t is the weather measure, i.e., the drought index. Finally, the international
where εW
economy block writes: [ ]′
Xt∗ = ∆log (Yt∗ ) πt∗ rt∗ ,
( )
where ∆log Ytd stands for foreign real output growth, πt∗ for foreign prices and rt∗
for foreign interest rate.
18
We use a lag of four in the model basing our choice on the value of the Akaike information criterion.
188 Appendix C Climate Change and Business Cycles: Technical Appendix

The framework represented by eq. (C.54) imposes bloc exogeneity (see e.g. Cushman
and Zha (1997); Kim and Roubini (2000)). To be consistent with the small open eco-
nomy setup, variables from the foreign economy block can impact variables from the
domestic block, but not the other way around. In addition, we impose restrictions
regarding the weather block, such that weather shocks may impact the domestic eco-
nomy only.

For clarity purposes, eq. (C.54) can be rewritten in the following way:

∑
p
AXt = Bl Xt−l + CIt + ηt , (C.55)
l=1

 
11 12 13
A A A
 
where A =   0 A
22
0  is the n × n matrix of contemporaneous effects with n
0 0 A33
[ ]′
the number of endogenous variables, Xt = Xt Xt Xt is the n × 1 vector of
D W ∗
 
Bl11 Bl12 Bl13
 
endogenous variables at time t, Bl =  0 B l
22
0 , for l = 1, . . . , p are the n × n

33
0 0 Bl
matrices of lagged parameters to be estimated, C is[ the n × j matrix
] of parameters
′
associated with the exogenous variables, and ηt = ηtD ηtW ηt∗ , the n × 1 vector
contains white noise structural errors, normally distributed with zero mean and both
serially and mutually uncorrelated. The model of eq. (C.55) can be written in a more
compact way:

AXt = BZt + ηt , (C.56)

[ ]
where B = B1 · · · Bp C is the n × (np + j) matrix of lagged restrictions and
[ ]′
Zt = Xt−1 · · · Xt−p It , is the (np + j) × 1 column vector of lagged endogenous
variables.

Restrictions imposed on contemporaneous relationships (A matrix) are summarized

in table C.6, while those imposed on lagged relationships (Bl matrix) are reported in
table C.7. These restrictions ensure the block exogeneity. Additional restrictions are
set following Buckle et al. (2007). In both tables, structural equations are written in
columns so that the lines show the variables appearing in each equation.
 Appendix C Climate Change and Business Cycles: Technical Appendix 189

Table C.6: Contemporaneous Structure of the Model

Dependent Variable ( )
( ) ( )
∆log (Yt∗ ) πt∗ rt∗ ∆log Ytd πtC ∆log XtA log πx,tA rert rt εW
t
∆log (Yt∗ ) a11 a12 a13 0 0 0 0 a18 0 0
πt∗ 0 a22 a23 0 a25 0 a27 a28 0 0
Independent Variables

rt∗( ) 0 0 a33 0 0 0 0 a38 a39 0

∆log Ytd 0 0 0 a44 0 0 0 a48 0 0
π(tC ) 0 0 0 0 a55 0 0 a58 a59 0
∆log( XtA) 0 0 0 0 0 a66 0 a68 0 0
A
log πx,t 0 0 0 0 0 0 a77 a78 a79 0
rert 0 0 0 0 a85 0 a87 a88 0 0
rt 0 0 0 0 0 0 0 a98 a99 0
εW
t 0 0 0 a104 0 a106 0 a108 a109 a1010
Note: Structural equations are written in columns. Zeros represent contemporaneous restrictions imposed ( ) in matrix A from
eq. (C.55). ∆log (Yt∗ ): Foreign Output, πt∗ : Foreign CPI Inflation, rt∗ : Foreign Interest Rate, ∆log Ytd : Output, πtC : CPI
( ) ( )
Inflation, ∆log XtA : Ag. Output, log πx,t A : Ag. Inflation, rert : Real Exchange Rate, rt : Interest Rate, εW
t : Climate.

Table C.7: Lagged Structure of the Model

Dependent Variable ( )
( ) ( )
∆log (Yt∗ ) πt∗ rt∗ ∆log Ytd πtC ∆log XtA log πx,tA rert rt εW
t
∆log (Yt∗ ) b11 b12 b13 b14 0 b16 0 b18 0 0
πt∗ b21 b22 b23 0 b25 0 b27 b28 0 0
Independent Variables

rt∗( ) b31 b32 b33 0 0 0 0 b38 b39 0

∆log Ytd 0 0 0 b44 0 b46 0 b48 0 0
π(tC ) 0 0 0 0 b55 0 b57 b58 b59 0
∆log( XtA) 0 0 0 b64 0 b66 0 b68 0 0
log πx,tA 0 0 0 0 b75 0 b77 b78 b79 0
rert 0 0 0 b84 b85 b86 b87 b88 0 0
rt 0 0 0 0 0 0 0 b98 b99 0
εW
t 0 0 0 b104 b105 b106 b107 b108 b109 b1010
Note: Structural equations are written in columns. Zeros represent lagged restrictions imposed ( ) in matrix Bl from eq. (C.55).
∆log (Yt∗ ): Foreign Output, πt∗ : Foreign CPI Inflation, rt∗ : Foreign Interest Rate, ∆log Ytd : Output, πtC : CPI Inflation,
( ) ( )
∆log XtA : Ag. Output, log πx,t A : Ag. Inflation, rert : Real Exchange Rate, rt : Interest Rate, εW
t : Climate.

6.1.1 The Foreign Economy Block

The foreign economy block comprises three variables: real output ∆log (Yt∗ ), prices
πt∗ , and interest rate rt∗ . All three measures are computed as a weighted average of the
respective value observed for New Zealand’s most important historical trading part-
ners: Australia, United States, United Kingdom and Japan. Weights are fixed according
to the share of imports and exports with New Zealand at each quarter.

We follow Rohe and Hartermann (2015) and restrict the foreign structural equations
in a recursive order to guarantee the block structure of the contemporaneous matrix.
Prices respond to contemporaneous variations of output. They also respond to vari-
ations in interest rate, but only with a lag. Interest rate is assumed to be affected by
both output and prices movements.
190 Appendix C Climate Change and Business Cycles: Technical Appendix

6.1.2 The Domestic Climate Block

The VAR model estimated contains a domestic weather block to study the impact of
climatic conditions on business cycle fluctuations. We rely on the same weather vari-
able as in the DSGE model whose construction is explained in appendix 4.1.1. When it
takes positive values, the weather variable depicts a prolonged episode of dryness. It
is the only variable in the exogenous domestic weather block and it is assumed to have
significant contemporaneous effects on GDP growth, agricultural output growth, real
exchange rate and interest rate. No restrictions are set for lagged effects except those
ensuring the exogeneity between blocks.

6.1.3 The Domestic Economy Block

( )
The domestic economy block comprises real output growth ∆ Ytd , prices πtC , real
( ) ( A)
agricultural output growth ∆log XtA , agricultural prices log πx,t , exchange rate
rert , and interest rate rt .

Real output growth is supposed to respond contemporaneously and with lags to weather
variations. Foreign output growth, agricultural output growth, real exchange rate, and
the weather variable also appear in the lagged relationships. Prices respond to contem-
poraneous movements of foreign prices. We set a zero restriction on contemporaneous
effects weather variations on prices to reflect the idea that prices will only react with a
lag following a climatic shock. However, we allow prices to contemporaneously react
to variations of exchange rate.

6.1.4 Estimation and Identification Issues

The order condition given by Rothenberg (1971) is a necessary condition for the struc-
tural VAR to be identified. The model from eq. (C.56) has n = 10 endogenous variables
and hence requires n × (n − 1)/2 = 45 restrictions. We impose 68 zero restrictions on
the contemporaneous matrix (A) and each of the lag-restriction matrices Bl contains
55 zero restrictions. The model we estimate is therefore overidentified. Using the
same procedure as in Rohe and Hartermann (2015), we ensure that the rank condition
for overidentified models (Rubio-Ramirez et al., 2010) is satisfied.
 Appendix C Climate Change and Business Cycles: Technical Appendix 191

Model given in eq. (C.56) is estimated as in Rohe and Hartermann (2015)19 with Bayesian
techniques. Priors are set to reflect the idea of Litterman (1986) that each of the time
series follows a random walk.

In a nutshell, for the ith structural equation, 1 ≤ i ≤ n, the prior is formed following
Sims and Zha (1998) and Waggoner and Zha (2003). Denoting ai and fi the ith row of
the contemporaneous-coefficient matrix A and the ith row of the lagged-coefficient
matrix B, respectively, the general form of the prior is given by:

ai ∼ N (0, S̄i )
. (C.57)
f | a ∼ N (P̄ a , H̄ )
i i i i i

The prior covariance matrix of the contemporaneous parameters is specified as a n×n

diagonal matrix whose diagonal elements are given as:
([( )2 ( )2 ])
Ŝi = diag λ0
σ1
··· λ0
σn
. (C.58)

The prior covariance matrix of the parameters in fi | ai is also a (np + j) × (np + j)

diagonal matrix such that:
([( )2 ( )2 ( )2 ( )2 ])
H̄i = diag λ0 λ1
1λ3 σ1
··· λ0 λ1
1λ3 σn
λ0 λ1
2λ3 σ1
··· λ0 λ1
pλ3 σ1
λ4 · · · λj .
(C.59)

The prior means of contemporaneous parameters are supposed to be null, while the
prior means of lagged parameters incorporate the random walk assumption by setting
P̄i such that:
[ ]
I
P̄i = . (C.60)
0

Prior information in the Bayesian estimation is weighted according to the values se-
lected for the hyperparameters of eq. (C.58) and eq. (C.59).20
19
We thank Matthias Hartermann for the R code he provided.
20
We follow Rohe and Hartermann (2015), Sims and Zha (1998) and Bhuiyan (2012) in the choice of
the values given to the hyperparameters: λ0 = 1, λ1 = 0.5, λ3 = 0.1 and λ4 = · · · = λj = λ0 ×λ4 = 1,
with λ4 = 1.
 192 Appendix C Climate Change and Business Cycles: Technical Appendix

6.2 Macroeconomic Response to Weather Shocks

We now present the empirical results of the impulse responses to a one standard devi-
ation shock to the weather variable i.e., the drought indicator to assess the macroeco-
nomic response following this shock.21 These IRFs are reported in fig. C.4. The solid
lines are the responses while the grey areas are the eighty-four percent confidence
bands obtained from 20, 000 iterations of the Gibbs sampler and 5, 000 more iterations
for the burn-in. The responses are computed for 20 periods.

Output ∆log Ytd Ag. Output ∆log XtA A

CPI Inflation πtC Ag. Inflation log πx,t

0.2 0.1 0.5

0.2
0.05 0
0 0
0 −0.5

−0.2 −0.05 −1 −0.2

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

Real Exchange Rate rert Interest Rate rt Weather εW

t

0 0.8

0 0.6
−0.05 0.4
−0.5
0.2
−1 −0.1 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Notes: The blue line is the median of the distribution of the IRFs generated when parameters are drawn from the posterior
distribution. The grey areas represent the 16th and 84th percentiles from the draws. The response horizon is in quarters.

Figure C.4: Impulse Responses of the VAR Model due to Weather Shock

Figure C.4 shows multiple channels affecting the business cycles after a climate shock.
In line with (Buckle et al., 2007), domestic output immediately falls. The fall in the
agricultural output is also instantaneous, and for the agricultural output, the effects
remain until four quarters before they become insignificant. In reaction to the GDP
decline, the central bank decreases the interest rate. Agricultural production prices
eventually rise in reaction to the adverse supply shock. This expansion may be due
to the rise in costs sustained by farmers. To offset the adverse effects of a drought,
farmers may use more inputs, such as water that is needed to feed livestock or irrig-
ate cultures. The exchange rate initially appreciates following the climate shock, as
specified in the model, but eventually depreciates before it reverts back to trend after
two years and a half. The agricultural sector represents a substantial portion of New
Zealand’s exports, so the decline in agricultural output may lead to a decline in exports
followed by a depreciation of the exchange rate.

21
We focus on the shock to the weather variable. The complete set of IRFs is available upon request.
 Appendix C Climate Change and Business Cycles: Technical Appendix 193

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )

1 0 0
0.4
−0.1
0.2 0.5 −0.2
−0.2
0 0 −0.4 −0.3

2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

0 1
0
0.5
−0.1
−0.5 0
−0.2
−0.5
−1
−0.3
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.5: IRF to an Estimated Productivity Shock ηtZ Affecting Both Sectors

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )

1 0 0
0.4
−0.1
0.2 0.5 −0.2
−0.2
0 0 −0.4 −0.3

2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

0 1
0
0.5
−0.1
−0.5 0
−0.2
−0.5
−1
−0.3
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.6: IRF to a Preference Shock ηtD affecting the Consumption Index of House-
holds
 194 Appendix C Climate Change and Business Cycles: Technical Appendix

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )
·10−2
0 0
0.2 5
−0.2 −0.1
0
−0.2 0.1
−0.4 −5
−0.6 −0.3
0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

0.1 0 1
0.08 0.5
−0.5
0.06
0
0.04 −1
0.02 −0.5
0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.7: IRF to a Markup Shock ηtN Affecting Prices of Non-agricultural Goods

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )
0 0
0 0 −0.1 −0.1
−0.2 −0.2
−0.2
−0.1 −0.3
−0.4 −0.3
−0.4
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

·10−2
0 1
4 −0.1 0.5

2 −0.2 0
−0.5
−0.3
0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.8: IRF to a Monetary Policy Shock ηtR

 Appendix C Climate Change and Business Cycles: Technical Appendix 195

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )
·10−2
0 0
0.2
−0.05 4
−0.2
−0.1 0.1
−0.4 2
−0.15
−0.6 0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

·10−2
0.6
0.3
2
0.2 0.4
1
0.1 0.2
0
0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.9: IRF to a Weather Surprise Shock ηtS

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )
·10−2 ·10−2 ·10−2
0 0 2 8
−2 1 6
−4 −0.1 0 4
−1 2
−6
−0.2 0
−2
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

·10−2
0.1
0 0.05 0.2

−1 0 0.1
−0.05
−2 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.10: IRF to a Weather-news Shock η̂t−1

S (Expected one Quarter in Advance)
 196 Appendix C Climate Change and Business Cycles: Technical Appendix

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )

0 1 0.3 0.4
−0.5 0.2
0.5 0.2
0.1
−1
0 0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

1 1

0.2 0.5
0.5 0
0.1
−0.5
0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.11: IRF to a Foreign Demand Shock ηt∗Y

Output ∆log(Ytd ) Agriculture ∆log(XtA ) CPI Inflation πtC Agriculture Inflation log(πtA )
·10−2 ·10−2
0 2 8 1
6
−5 0 4 0.5
2
−2 0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Interest Rate rt Hours Worked log(hdt ) climate log(st )

·10−2
0.1 1
4 0.5
0
0
2 −0.1
−0.5
−0.2
0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Note: Colored lines are the posterior means, grey areas are the 90 percent HPD intervals. IRF are reported as percentage
deviations from the deterministic steady state.

Figure C.12: IRF to a Foreign Agricultural Price Shock ηt∗A

List of Figures

1 Tendances du forçage radiatif et changement global de la température

moyenne . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
2 Effets du changement climatique sur la distribution des températures . xxi
3 Trends in Radiative Forcing and Global Mean Temperature Change . . 4
4 Effects of Climate Change on the Distribution of Temperatures . . . . . 6

1.1 Number of Observations per District . . . . . . . . . . . . . . . . . . . 30

1.2 Estimated Climate Parameters by Quantile for Net Revenues per Acre . 37
1.3 Global Effects of a One Unit Rise in Climate Variables on Net Revenues
per Acre for each Quantile Estimated . . . . . . . . . . . . . . . . . . . 37
1.4 Estimated Household Characteristics Parameters by Quantile for Net
Revenues per Acre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.5 Estimated Irigation Parameters by Quantile for Net Revenues per Acre 39
1.6 Global Effects of a One Unit Rise in Climate Variables on Net Revenues
per Acre for each Quantile Estimated for Rainfed and Irrigated Farms
Separately . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.7 Average Change in Net Revenues per Acre by Districts and Quantiles
Under Scenario 1 (Rupees per Acre) . . . . . . . . . . . . . . . . . . . . 42
1.8 Average Changes in Net Revenues per Acre by Districts and Quantiles
Under Scenario 2 (Ruppees per Acre) . . . . . . . . . . . . . . . . . . . 42
1.9 Median Changes in Net Revenues per Acre by District at Selected Quantiles
(Thousand Rupees per Acre). . . . . . . . . . . . . . . . . . . . . . . . . 43
1.10 Median Percent Changes in Net Revenues per Acre by District at Se-
lected Quantiles (Thousand Rupees per Acre). . . . . . . . . . . . . . . 43

2.1 Number of Observations per District . . . . . . . . . . . . . . . . . . . 58

2.2 Number of Observations per District and Labour Regime . . . . . . . . 59
2.3 Average Per capita Income by State of Farm-households) . . . . . . . . 61
2.4 Distribution of Percentage of Crops Kept for Self-Consumption . . . . 61
2.5 Climate Normals (1980–2008) . . . . . . . . . . . . . . . . . . . . . . . . 62
2.6 Distribution of Number of Family Members . . . . . . . . . . . . . . . . 64
2.7 Effects of Increasing Prices on Consumption Shares (Median Percent
Changes) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.8 Effects of Precipiation Variations on Consumption (Median Percent
Changes) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.9 Effects of Increasing Temperatures on Consumption (Median Percent
Changes) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
197
198 List of Figures

2.10 Effects of Climate Change on Consumption (Median Percent Changes) 77

3.1 Agricultural Production of Wheat and Corn in Western Europe . . . . 91

3.2 Evolution of Yields and Prices . . . . . . . . . . . . . . . . . . . . . . . 91
3.3 Predicted Wheat Yields (100kg per ha) Under Different Climate Scenarios 101
3.4 Predicted Corn Yields (100kg per ha) Under Different Climate Scenarios 101
3.5 Regional Changes in Wheat Yields Under Different Climate Scenarios . 102
3.6 Regional Changes in Corn Yields Under Different Climate Scenarios . . 102

4.1 Prior and Posterior Distributions of Structural Parameters for New

Zealand (Excluding Shocks) . . . . . . . . . . . . . . . . . . . . . . . . 124
4.2 System Response to an Estimated Weather Shock ηtS Measured in Per-
centage Deviations from the Steady State . . . . . . . . . . . . . . . . . 127
4.3 System Response to an Estimated Weather News Shock η̂t−1 W
Measured
as Percentage Deviations from Steady State . . . . . . . . . . . . . . . . 128
4.4 Forecast Error Variance Decomposition at the Posterior Mean for Dif-
ferent Time Horizons (one, two, four, ten, forty and unconditional). . . 129
4.5 Historical Decomposition of Aggregate Output, Agricultural Produc-
tion and Agricultural Production Price Inflation . . . . . . . . . . . . . 131
4.6 Estimations of the Standard Error of Climatic Shock Under Four Dif-
ferent Climate Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 134

A.1 Net Revenues per Acre Distribution . . . . . . . . . . . . . . . . . . . . 142

A.2 Cultivated Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
A.3 Most Important Mode of Irrigation (Number of Observations) . . . . . 143
A.4 Process from Raw Data to District Level Climatic Data . . . . . . . . . 144
A.5 Meteorological Stations Locations in India . . . . . . . . . . . . . . . . 145
A.6 Number of Observations Used to Estimate Weather Data per Day . . . 145
A.7 Precipitation “Normals” . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
A.8 Temperature “Normals” . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

B.1 Projected Changes in Winter Temperature . . . . . . . . . . . . . . . . 152

B.2 Projected Changes in Spring Temperature . . . . . . . . . . . . . . . . 152
B.3 Projected Changes in Summer Temperature . . . . . . . . . . . . . . . 153
B.4 Projected Changes in Fall Temperature . . . . . . . . . . . . . . . . . . 153
B.5 Projected Percentage Changes in Winter Precipitation . . . . . . . . . . 154
B.6 Projected Percentage Changes in Spring Precipitation . . . . . . . . . . 154
B.7 Projected Percentage Changes in Summer Precipitation . . . . . . . . . 155
B.8 Projected Percentage Changes in Fall Precipitation . . . . . . . . . . . . 155

C.1 A Small Open Economy Featuring Two Sectors: Agricultural and Non-
agricultural Sectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
C.2 Covering of Weather Stations Used to Construct the Soil Moisture De-
ficit Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
C.3 Observable Variables Used in the BSVAR and the DSGE Estimations . . 182
C.4 Impulse Responses of the VAR Model due to Weather Shock . . . . . . 192
 List of Figures 199

C.5 IRF to an Estimated Productivity Shock ηtZ Affecting Both Sectors . . . 193
C.6 IRF to a Preference Shock ηtD affecting the Consumption Index of House-
holds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
C.7 IRF to a Markup Shock ηtN Affecting Prices of Non-agricultural Goods . 194
C.8 IRF to a Monetary Policy Shock ηtR . . . . . . . . . . . . . . . . . . . . 194
C.9 IRF to a Weather Surprise Shock ηtS . . . . . . . . . . . . . . . . . . . . 195
C.10 IRF to a Weather-news Shock η̂t−1 S
(Expected one Quarter in Advance) 195
∗Y
C.11 IRF to a Foreign Demand Shock ηt . . . . . . . . . . . . . . . . . . . . 196
C.12 IRF to a Foreign Agricultural Price Shock ηt∗A . . . . . . . . . . . . . . . 196
List of Tables

1.1 Irrigation Techniques in the Sample . . . . . . . . . . . . . . . . . . . . 33

1.2 Summary Statistics for Data Set on Farm Households . . . . . . . . . . 33
1.3 Regression Results for Agricultural Net Revenues per Acre . . . . . . . 46
1.4 Regression Results with Only Rainfed or Irrigated Farms . . . . . . . . 47
1.5 Changes in Net Revenues per Acre by Districts at Each Quantile (Rupees) 48

2.1 Descriptive Statistics According to the Different Labour Regimes . . . . 60

2.2 Correlations of Crops Output and Agricultural Income with Climate
Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3 Production Function Estimation Results . . . . . . . . . . . . . . . . . . 67
2.4 Participation in Agricultural Labour Market (Probit Selection Model) . 68
2.5 Consumption Choices According to the Different Labour Regimes . . . 69
2.6 Almost Ideal Demand System Results . . . . . . . . . . . . . . . . . . . 75
2.7 Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.1 Number of NUTS-3 Regions and Observation by Country and Location 90

3.2 Descriptive Statistics (Mean and Std) for Wheat Datasets . . . . . . . . 94
3.3 Descriptive Statistics (Mean and Std) for Corn Datasets . . . . . . . . . 95
3.4 Determinants of Wheat and Corn Yield (Maximum Likelihood) . . . . . 98
3.5 Effect of Climate Change on Wheat and Corn Yield (percentage change) 103

4.1 Prior and Posterior Distributions of Structural Parameters and Shock

Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.2 Prior and Posterior Model Probabilities . . . . . . . . . . . . . . . . . . 125
4.3 Changes in Standard-Errors of Simulated Observables Under Climate
Change Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

A.1 Description of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

B.1 Descriptive Statistics of Climate Variables . . . . . . . . . . . . . . . . . 151

C.1 Notations of Variables and Parameters for the Household Block . . . . 175
C.2 Notations of Variables and Parameters for the Final Firms Block . . . . 176
C.3 Notations of Variables and Parameters for the Intermediate-good Firms
Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
C.4 Notations of Variables and Parameters for Remaining Blocks . . . . . . 178

201
202 List of Tables

C.5 Estimations of Growth Rates of Standard Errors of the Weather Process

Under Different Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . 181
C.6 Contemporaneous Structure of the Model . . . . . . . . . . . . . . . . . 189
C.7 Lagged Structure of the Model . . . . . . . . . . . . . . . . . . . . . . . 189
Bibliography

Adams, R., Hurd, B., Lenhart, S., and Leary, N. (1998). Effects of global climate change
on agriculture: an interpretative review. Climate Research, 11(1):19–30.

Adams, R. M., Rosenzweig, C., Peart, R. M., Ritchie, J. T., McCarl, B. A., Glyer, J. D.,
Curry, R. B., Jones, J. W., Boote, K. J., and Allen, L. H. (1990). Global climate change
and US agriculture. Nature, 345(6272):219–224.

Adjemian, S., Bastani, H., Juillard, M., Mihoubi, F., Perendia, G., Ratto, M., and Ville-
mot, S. (2011). Dynare: Reference manual, version 4. Dynare Working Papers, 1.

Aggarwal, P. K. and Mall, R. (2002). Climate change and rice yields in diverse agro-
environments of India. II. effect of uncertainties in scenarios and crop models on
impact assessment. Climatic Change, 52(3):331–343.

Angrist, J. D. and Krueger, A. B. (2001). Instrumental variables and the search for
identification: From supply and demand to natural experiments. The Journal of
Economic Perspectives, 15(4):69–85.

Annicchiarico, B. and Di Dio, F. (2015). Environmental policy and macroeconomic

dynamics in a new Keynesian model. Journal of Environmental Economics and Man-
agement, 69:1–21.

Auffhammer, M., Hsiang, S. M., Schlenker, W., and Sobel, A. (2013). Using weather
data and climate model output in economic analyses of climate change. Review of
Environmental Economics and Policy, 7(2):181–198.

Auffhammer, M. and Schlenker, W. (2014). Empirical studies on agricultural impacts

and adaptation. Energy Economics, 46:555–561.

Barnwal, P. and Kotani, K. (2013). Climatic impacts across agricultural crop yield dis-
tributions: An application of quantile regression on rice crops in Andhra Pradesh,
India. Ecological Economics, 87(0):95 – 109.
203
204 Bibliography

Barsky, R. B. and Sims, E. R. (2011). News shocks and business cycles. Journal of
monetary Economics, 58(3):273–289.

Bassu, S., Brisson, N., Durand, J.-L., Boote, K., Lizaso, J., Jones, J. W., Rosenzweig,
C., Ruane, A. C., Adam, M., Baron, C., et al. (2014). How do various maize crop
models vary in their responses to climate change factors? Global change biology,
20(7):2301–2320.

Batjes, N., Dijkshoorn, K., van Engelen, V., Fischer, G., Jones, A., Montanarella, L.,
Petri, M., Prieler, S., Teixeira, E., Wiberg, D., and Shi, X. (2008). Harmonized world
soil database (version 1.2).

Beaudry, P. and Portier, F. (2006). Stock prices, news, and economic fluctuations. The
American economic review, 96(4):1293–1307.

Berg, E. J. (1961). Backward-sloping labor supply functions in dual economies—the

Africa case. The Quarterly Journal of Economics, 75(3):468–492.

Bhuiyan, R. (2012). Monetary transmission mechanisms in a small open economy: a

Bayesian structural VAR approach. Canadian Journal of Economics/Revue canadienne
d’économique, 45(3):1037–1061.

Black, R., Bennett, S. R., Thomas, S. M., and Beddington, J. R. (2011). Climate change:
Migration as adaptation. Nature, 478(7370):447–449.

Blasing, T. (2009). Recent greenhouse gas concentrations.

Bloor, C. and Matheson, T. (2010). Analysing shock transmission in a data-rich envir-

onment: a large BVAR for New Zealand. Empirical Economics, 39(2):537–558.

Boer, E. (2001). Kriging and thin plate splines for mapping climate variables. Interna-
tional Journal of Applied Earth Observation and Geoinformation, 3(2):146–154.

Brisson, N., Mary, B., Ripoche, D., Jeuffroy, M. H., Ruget, F., Nicoullaud, B., Gate, P.,
Devienne-Barret, F., Antonioletti, R., Durr, C., et al. (1998). STICS: a generic model
for the simulation of crops and their water and nitrogen balances. I. Theory and
parameterization applied to wheat and corn. Agronomie, 18(5-6):311–346.

Brooks, S. P. and Gelman, A. (1998). General methods for monitoring convergence of

iterative simulations. Journal of computational and graphical statistics, 7(4):434–455.

Bruinsma, J. (2009). The resource outlook to 2050. In Expert meeting on how to feed the
world in, volume 2050, pages 1–33.
 Bibliography 205

Buckle, R. A., Kim, K., Kirkham, H., McLellan, N., and Sharma, J. (2007). A structural
VAR business cycle model for a volatile small open economy. Economic Modelling,
24(6):990–1017.

Butt, T. A., McCarl, B. A., Angerer, J., Dyke, P. T., and Stuth, J. W. (2005). The economic
and food security implications of climate change in Mali. Climatic change, 68(3):355–
378.

Cabas, J., Weersink, A., and Olale, E. (2010). Crop yield response to economic, site and
climatic variables. Climatic Change, 101(3-4):599–616.

Castagnini, R., Menon, M., and Perali, F. (2004). Extended and full incomes at the
household and individual level: an application to farm households. American
Journal of Agricultural Economics, 86(3):730–736.

Chaturvedi, R. K., Joshi, J., Jayaraman, M., Bala, G., and Ravindranath, N. (2012).
Multi-model climate change projections for India under representative concentra-
tion pathways. Current Science, 103(7):791–802.

Chen, C.-C., McCarl, B. A., and Schimmelpfennig, D. E. (2004). Yield variability as

influenced by climate: a statistical investigation. Climatic Change, 66(1-2):239–261.

Christiano, L. J., Eichenbaum, M., and Evans, C. L. (2005). Nominal rigidities and
the dynamic effects of a shock to monetary policy. Journal of political Economy,
113(1):1–45.

Chung, U., Gbegbelegbe, S., Shiferaw, B., Robertson, R., Yun, J. I., Tesfaye, K., Hoo-
genboom, G., and Sonder, K. (2014). Modeling the effect of a heat wave on maize
production in the USA and its implications on food security in the developing world.
Weather and Climate Extremes, 5:67–77.

Ciais, P., Reichstein, M., Viovy, N., Granier, A., Ogée, J., Allard, V., Aubinet, M., Buch-
mann, N., Bernhofer, C., Carrara, A., et al. (2005). Europe-wide reduction in primary
productivity caused by the heat and drought in 2003. Nature, 437(7058):529–533.

Clarida, R., Galí, J., and Gertler, M. (1998). Monetary policy rules in practice: some
international evidence. european economic review, 42(6):1033–1067.

Cline, W. R. (1996). The impact of global warming of agriculture: comment. The

American Economic Review, 86(5):1309–1311.
206 Bibliography

Collins, M., Knutti, R., Arblaster, J., Dufresne, J.-L., Fichefet, T., Friedlingstein, P., Gao,
X., Gutowski, W., Johns, T., Krinner, G., et al. (2013). Long-term climate change:
projections, commitments and irreversibility.

Cubasch, U., Wuebbles, D., Chen, D., Facchini, M., Frame, D., Mahowald, N., and Win-
ther, J. (2013). Climate change 2013: the physical science basis. Contribution of
Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Cli-
mate Change Cambridge, United Kingdom and New York, NY, USA, pages 119–158.

Cushman, D. O. and Zha, T. (1997). Identifying monetary policy in a small open eco-
nomy under flexible exchange rates. Journal of Monetary economics, 39(3):433–448.

Dasgupta, P. and Goldar, B. (2006). Female labour supply in rural India: An economet-
ric analysis. Indian Journal of Labour Economics, 49(2):293–310.

De Janvry, A., Fafchamps, M., and Sadoulet, E. (1991). Peasant household beha-
viour with missing markets: some paradoxes explained. The Economic Journal,
101(409):1400–1417.

De Salvo, M., Raffaelli, R., and Moser, R. (2013). The impact of climate change on
permanent crops in an Alpine region: A Ricardian analysis. Agricultural Systems,
118(0):23 – 32.

Deaton, A. (1997). The analysis of household surveys: a microeconometric approach to

development policy. World Bank Publications.

Deaton, A. and Muellbauer, J. (1980). An almost ideal demand system. The American
economic review, 70(3):312–326.

Dell, M., Jones, B. F., and Olken, B. A. (2014). What do we learn from the weather? The
new climate–economy literature. Journal of Economic Literature, 52(3):740–798.

Deressa, T. T., Hassan, R. M., Ringler, C., Alemu, T., and Yesuf, M. (2009). Determinants
of farmers’ choice of adaptation methods to climate change in the Nile Basin of
Ethiopia. Global Environmental Change, 19(2):248–255.

Desai, S., Dubey, A., Joshi, B., Sen, M., Shariff, A., and Vanneman, R. (2005). India
human development survey (IHDS). University of Maryland and the National Council
of Applied Economic Research (NCAER), New Delhi.
 Bibliography 207

Desai, S. and Vanneman, R. (2016). India human development survey-II (IHDS-II), 2011-
12. University of Maryland and the National Council of Applied Economic Research
(NCAER), New Delhi.

Desai, S. B., Dubey, A., Joshi, B. L., Sen, M., Shariff, A., and Vanneman, R. (2010).
Human Development in India. Oxford University, New York.

Deschênes, O. and Greenstone, M. (2007). The economic impacts of climate change:

evidence from agricultural output and random fluctuations in weather. The Amer-
ican Economic Review, 97(1):354–385.

Deschenes, O. and Kolstad, C. (2011). Economic impacts of climate change on Califor-

nia agriculture. Climatic Change, 109(1):365–386.

Dessing, M. (2002). Labor supply, the family and poverty: the S-shaped labor supply
curve. Journal of Economic Behavior & Organization, 49(4):433–458.

Di Falco, S., Veronesi, M., and Yesuf, M. (2011). Does adaptation to climate change
provide food security? A micro-perspective from Ethiopia. American Journal of
Agricultural Economics, 93(3):829–846.

Dinar, A. and Mendelsohn, R. O. (2011). Handbook on climate change and agriculture.

Edward Elgar Publishing.

Dissou, Y. and Karnizova, L. (2016). Emissions cap or emissions tax? A multi-sector

business cycle analysis. Journal of Environmental Economics and Management.

Dixit, A. K. and Stiglitz, J. E. (1977). Monopolistic competition and optimum product

diversity. The American Economic Review, 67(3):297–308.

Easterling, W. E., Aggarwal, P. K., Batima, P., Brander, K. M., Erda, L., Howden, S. M.,
Kirilenko, A., Morton, J., Soussana, J.-F., Schmidhuber, J., et al. (2007). Food, fibre and
forest products. In Parry, M., Canziani, O. F., Palutikof, J. P., van der Linden, P. J.,
Hanson, C. E., et al., editors, Climate Change 2014, Mitigation of Climate Change.,
chapter 5, pages 273–313. Cambridge University Press, Cambridge, United Kingdom
and New York, NY, USA.

Edenhofer, O., Pichs-Madruga, R., Sokona, Y., Farahani, E., Kadner, S., Seyboth,
K., Adler, A., Baum, I., Brunner, S., Eickemeier, P., Kriemann, B., Savolainen, J.,
Schlömer, S., Stechow, C. v., Zwickel, T., and Minx, J. (2014). IPCC, 2014: Summary
208 Bibliography

for policymakers. In of Working Group III to the Fifth Assessment Report of the In-
tergovernmental Panel on Climate Change, C., editor, Climate Change 2014, Mitig-
ation of Climate Change. Cambridge University Press, Cambridge, United Kingdom
and New York, NY, USA.

Fankhauser, S. and Tol, R. S. (2005). On climate change and economic growth. Resource
and Energy Economics, 27(1):1–17.

FAO (2017). Crop prospects and food situation. Technical report, Food and Agriculture
Organization of the United Nations, Rome.

Field, C. B., Jackson, R. B., and Mooney, H. A. (1995). Stomatal responses to increased
CO2: implications from the plant to the global scale. Plant, Cell & Environment,
18(10):1214–1225.

Fischer, C. and Springborn, M. (2011). Emissions targets and the real business cycle:
Intensity targets versus caps or taxes. Journal of Environmental Economics and Man-
agement, 62(3):352–366.

Fischer, G., Shah, M., Tubiello, F. N., and Van Velhuizen, H. (2005). Socio-economic
and climate change impacts on agriculture: an integrated assessment, 1990–2080.
Philosophical Transactions of the Royal Society B: Biological Sciences, 360(1463):2067–
2083.

Fisher, A. C., Hanemann, W. M., Roberts, M. J., and Schlenker, W. (2012). The eco-
nomic impacts of climate change: evidence from agricultural output and random
fluctuations in weather: comment. The American Economic Review, 102(7):3749–
3760.

Fleischer, A., Lichtman, I., and Mendelsohn, R. (2008). Climate change, irrigation, and
Israeli agriculture: Will warming be harmful? Ecological economics, 65(3):508–515.

Füssel, H.-M., Jol, A., et al. (2012). Climate change, impacts and vulnerability in Europe
2012 an indicator-based report. Technical report, Luxembourg: Publications Office
of the European Union.

Galí, J. and Monacelli, T. (2005). Monetary policy and exchange rate volatility in a
small open economy. The Review of Economic Studies, 72(3):707–734.

García, M. and Viladrich, M. (2009). The economic relevance of climate variables in

agriculture: The case of Spain. Economía agraria y recursos naturales, 9(2):149–180.
 Bibliography 209

García-Herrera, R., Díaz, J., Trigo, R., Luterbacher, J., and Fischer, E. (2010). A review of
the European summer heat wave of 2003. Critical Reviews in Environmental Science
and Technology, 40(4):267–306.

Gbetibouo, G. and Hassan, R. (2005). Measuring the economic impact of climate change
on major South African field crops: a Ricardian approach. Global and Planetary
Change, 47(2-4):143 – 152.

Geweke, J. (1999). Using simulation methods for Bayesian econometric models: infer-
ence, development, and communication. Econometric reviews, 18(1):1–73.

Greenwood, J., Hercowitz, Z., and Huffman, G. W. (1988). Investment, capacity utiliz-
ation, and the real business cycle. The American Economic Review, pages 402–417.

Haim, D., Shechter, M., and Berliner, P. (2008). Assessing the impact of climate change
on representative field crops in Israeli agriculture: a case study of wheat and cotton.
Climatic Change, 86(3-4):425–440.

Heckman, J. J. (1979). Sample selection bias as a specification error. Econometrica,

47(1):153–161.

Henning, C. H. and Henningsen, A. (2007). Modeling farm households’ price responses

in the presence of transaction costs and heterogeneity in labor markets. American
Journal of Agricultural Economics, 89(3):665–681.

Henningsen, A. (2014). micEconAids: Demand Analysis with the Almost Ideal Demand
System (AIDS). R package version 0.6-16.

Heutel, G. (2012). How should environmental policy respond to business cycles? Op-
timal policy under persistent productivity shocks. Review of Economic Dynamics,
15(2):244–264.

Hope, C., Anderson, J., and Wenman, P. (1993). Policy analysis of the greenhouse
effect: an application of the PAGE model. Energy Policy, 21(3):327–338.

Howard, P., Sterner, T., et al. (2014). Raising the temperature on food prices: Climate
change, food security, and the social cost of carbon.

Hutchinson, M. (1995). Interpolating mean rainfall using thin plate smoothing splines.
International journal of geographical information systems, 9(4):385–403.

Iacoviello, M. and Neri, S. (2010). Housing market spillovers: evidence from an estim-
ated DSGE model. American Economic Journal: Macroeconomics, 2(2):125–164.
210 Bibliography

Jacoby, H. G. (1993). Shadow wages and peasant family labour supply: an econometric
application to the Peruvian Sierra. The Review of Economic Studies, 60(4):903–921.

Jones, C. A., Kiniry, J. R., and Dyke, P. (1986). CERES-Maize: A simulation model of
maize growth and development. Texas A& M University Press.

Jones, P. G. and Thornton, P. K. (2003). The potential impacts of climate change on

maize production in Africa and Latin America in 2055. Global environmental change,
13(1):51–59.

Just, R. E. and Pope, R. D. (1978). Stochastic specification of production functions and

economic implications. Journal of Econometrics, 7(1):67–86.

Just, R. E. and Pope, R. D. (1979). Production function estimation and related risk
considerations. American Journal of Agricultural Economics, 61(2):276–284.

Kamber, G., McDonald, C., Price, G., et al. (2013). Drying out: Investigating the eco-
nomic effects of drought in New Zealand. Technical report, Reserve Bank of New
Zealand Analytical Note Series No. AN2013/02, Reserve Bank of New Zealand, Wel-
lington, New Zealand.

Khanal, A. R., Mishra, A. K., and Bhattarai, M. (2014). Weather risk and cropping
intensity: A non-stationary and dynamic panel modeling approach. In 2014 Annual
Meeting, July 27-29, 2014, Minneapolis, Minnesota, number 170603. Agricultural and
Applied Economics Association.

Kim, M.-K., Pang, A., et al. (2009). Climate change impact on rice yield and production
risk. Journal of Rural Development, 32(2):17–29.

Kim, S. and Roubini, N. (2000). Exchange rate anomalies in the industrial countries: A
solution with a structural VAR approach. Journal of Monetary Economics, 45(3):561–
586.

Koenker, R. (2005). Quantile regression. Number 38. Cambridge university press.

Koenker, R. and Bassett Jr, G. (1978). Regression quantiles. Econometrica: journal of

the Econometric Society, pages 33–50.

Kumar, K. and Parikh, J. (1998). Climate change impacts on Indian agriculture: the
Ricardian approach. In Dinar, A., Mendelsohn, R., Evenson, R., Parikh, J., Sanghi, A.,
Kumar, K., McKinsey, J., and Lonergan, S., editors, Measuring the impact of climate
 Bibliography 211

change on Indian agriculture, pages 141–184. World Bank Technical Paper No. 402,
World Bank, Washington, DC.

Kumar, K. and Parikh, J. (2001). Indian agriculture and climate sensitivity. Global
environmental change, 11(2):147–154.

Kumar, K. K. (2011). Climate sensitivity of Indian agriculture: do spatial effects matter?

Cambridge Journal of Regions, Economy and Society, 4(2):221–235.

Kurukulasuriya, P. and Mendelsohn, R. (2008). A Ricardian analysis of the impact

of climate change on African cropland. African Journal Agriculture and Resource
Economics, 2(1):1–23.

Kurukulasuriya, P., Mendelsohn, R., Hassan, R., Benhin, J., Deressa, T., Diop, M., Eid,
H. M., Fosu, K. Y., Gbetibouo, G., Jain, S., et al. (2006). Will African agriculture
survive climate change? The World Bank Economic Review, 20(3):367–388.

Lang, G. (2007). Where are Germany’s gains from Kyoto? Estimating the effects of
global warming on agriculture. Climatic Change, 84(3-4):423–439.

Lippert, C., Krimly, T., and Aurbacher, J. (2009). A Ricardian analysis of the impact of
climate change on agriculture in Germany. Climatic Change, 97(3-4):593–610.

Litterman, R. B. (1986). Forecasting with Bayesian vector autoregressions?five years

of experience. Journal of Business & Economic Statistics, 4(1):25–38.

Liu, P. (2006). A small new Keynesian model of the New Zealand economy. Reserve
Bank of New Zealand, 3.

Lobell, D. B. and Asner, G. P. (2003). Climate and management contributions to recent

trends in US agricultural yields. Science, 299(5609):1032–1032.

Lobell, D. B. and Burke, M. B. (2010). On the use of statistical models to pre-

dict crop yield responses to climate change. Agricultural and Forest Meteorology,
150(11):1443–1452.

Lobell, D. B. and Field, C. B. (2007). Global scale climate–crop yield relationships and
the impacts of recent warming. Environmental research letters, 2(1):014002.

Lobell, D. B. and Ortiz-Monasterio, J. I. (2007). Impacts of day versus night temperat-

ures on spring wheat yields. Agronomy journal, 99(2):469–477.
212 Bibliography

Lobell, D. B., Schlenker, W., and Costa-Roberts, J. (2011). Climate trends and global
crop production since 1980. Science, 333(6042):616–620.

Maddison, D. (2000). A hedonic analysis of agricultural land prices in England and

Wales. European Review of Agricultural Economics, 27(4):519–532.

McCarl, B. A., Villavicencio, X., and Wu, X. (2008). Climate change and future analysis:
is stationarity dying? American Journal of Agricultural Economics, 90(5):1241–1247.

McLean, R., Sinha, S., Mirza, M., and Lal, M. (1998). Tropical Asia. In Watson, R. T.,
Zinyowera, M. C., and Moss, R. H., editors, The regional impact of climate change:
an assessment of vulnerability, pages 381–407. Cambridge University Press.

Mearns, L. O., Rosenzweig, C., and Goldberg, R. (1997). Mean and variance change in
climate scenarios: methods, agricultural applications, and measures of uncertainty.
Climatic Change, 35(4):367–396.

Mendelsohn, R. (2008). The impact of climate change on agriculture in developing

countries. Journal of Natural Resources Policy Research, 1(1):5–19.

Mendelsohn, R., Basist, A., Kurukulasuriya, P., and Dinar, A. (2007). Climate and rural
income. Climatic Change, 81(1):101–118.

Mendelsohn, R., Dinar, A., and Sanghi, A. (2001). The effect of development on the
climate sensitivity of agriculture. Environment and Development Economics, 6:85–
101.

Mendelsohn, R., Dinar, A., and Williams, L. (2006). The distributional impact of cli-
mate change on rich and poor countries. Environment and Development Economics,
11(02):159–178.

Mendelsohn, R., Nordhaus, W. D., and Shaw, D. (1994). The impact of global warming
on agriculture: A Ricardian analysis. American Economic Review, 84(4):753–71.

Mendelsohn, R. O. and Dinar, A. (2009). Climate change and agriculture: an economic

analysis of global impacts, adaptation and distributional effects. Edward Elgar Pub-
lishing.

Miao, R., Khanna, M., and Huang, H. (2015). Responsiveness of crop yield and acreage
to prices and climate. American Journal of Agricultural Economics, 98(1):191–211.
 Bibliography 213

Mitra, A., Kumar, M. D., Kumar, K. R., Abrol, Y., Kalra, N., Velayutham, M., and Naqvi,
S. (2002). Global change and biogeochemical cycles: the South Asia region. In
Global-Regional linkages in the earth system, pages 75–107. Springer.

Mundlak, Y. (2001). Production and supply. Handbook of agricultural economics, 1:3–85.

Narasimhan, B. and Srinivasan, R. (2005). Development and evaluation of soil moisture

deficit index (SMDI) and evapotranspiration deficit index (ETDI) for agricultural
drought monitoring. Agricultural and Forest Meteorology, 133(1):69–88.

Nordhaus, W. D. (1991). To slow or not to slow: the economics of the greenhouse

effect. The economic journal, 101(407):920–937.

Nordhaus, W. D. (1994). Managing the global commons: the economics of climate change,
volume 31. MIT press Cambridge, MA.

Nordhaus, W. D. (2007). A review of the Stern review on the economics of climate

change. Journal of economic literature, 45(3):686–702.

Nordhaus, W. D. and Yang, Z. (1996). A regional dynamic general-equilibrium model

of alternative climate-change strategies. The American Economic Review, pages 741–
765.

Nychka, D., Furrer, R., and Sain, S. (2015). fields: Tools for Spatial Data. R package
version 8.2-1.

O’Brien, K., Leichenko, R., Kelkar, U., Venema, H., Aandahl, G., Tompkins, H., Javed,
A., Bhadwal, S., Barg, S., Nygaard, L., et al. (2004). Mapping vulnerability to multiple
stressors: climate change and globalization in India. Global environmental change,
14(4):303–313.

OECD (2015). The economic consequences of climate change. Technical report.

Parry, M. L., Rosenzweig, C., Iglesias, A., Livermore, M., and Fischer, G. (2004). Ef-
fects of climate change on global food production under SRES emissions and socio-
economic scenarios. Global Environmental Change, 14(1):53–67.

Pindyck, R. S. (2013). Climate change policy: What do the models tell us? Journal of
Economic Literature, 51(3):860–872.

Polsky, C. (2004). Putting space and time in Ricardian climate change impact studies:
Agriculture in the US Great Plains, 1969–1992. Annals of the Association of American
Geographers, 94(3):549–564.
214 Bibliography

Polsky, C. and Easterling III, W. E. (2001). Adaptation to climate variability and change
in the US Great Plains: A multi-scale analysis of Ricardian climate sensitivities.
Agriculture, Ecosystems & Environment, 85(1):133–144.

Quiggin, J. and Horowitz, J. K. (1999). The impact of global warming on agriculture: a

Ricardian analysis: comment. The American Economic Review, 89(4):1044.

Rabanal, P. (2007). Does inflation increase after a monetary policy tightening? An-
swers based on an estimated DSGE model. Journal of Economic Dynamics and Con-
trol, 31(3):906–937.

Rabanal, P. and Tuesta, V. (2010). Euro-dollar real exchange rate dynamics in an estim-
ated two-country model: an assessment. Journal of Economic Dynamics and Control,
34(4):780–797.

Restuccia, D., Yang, D. T., and Zhu, X. (2008). Agriculture and aggregate productivity:
A quantitative cross-country analysis. Journal of Monetary Economics, 55(2):234–
250.

Ritchie, J. and Otter, S. (1985). Description and performance of CERES-Wheat: a user-

oriented wheat yield model. ARS-United States Department of Agriculture, Agricul-
tural Research Service (USA).

Rohe, C. and Hartermann, M. (2015). The role of external shocks for monetary policy
in Colombia and Brazil: A Bayesian SVAR analysis. Technical report, Center for
Quantitative Economics (CQE), University of Muenster.

Rosenzweig, C., Elliott, J., Deryng, D., Ruane, A. C., Müller, C., Arneth, A., Boote, K. J.,
Folberth, C., Glotter, M., Khabarov, N., et al. (2014). Assessing agricultural risks of
climate change in the 21st century in a global gridded crop model intercomparison.
Proceedings of the National Academy of Sciences, 111(9):3268–3273.

Rosenzweig, C. and Parry, M. L. (1994). Potential impact of climate change on world

food supply. Nature, 367(6459):133–138.

Rotemberg, J. J. (1982). Monopolistic price adjustment and aggregate output. The

Review of Economic Studies, 49(4):517–531.

Rothenberg, T. J. (1971). Identification in parametric models. Econometrica: Journal of

the Econometric Society, pages 577–591.
 Bibliography 215

Rubio-Ramirez, J. F., Waggoner, D. F., and Zha, T. (2010). Structural vector autoregres-
sions: Theory of identification and algorithms for inference. The Review of Economic
Studies, 77(2):665–696.

Rupa Kumar, K., Krishna Kumar, K., Ashrit, R., Patwardhan, S., and Pant, G. (2002).
Climate change in India: Observations and model projections. Climate Change and
India: IIssues, Concerns and Opportunities. Tata McGraw-Hill Publishing Company
Limited, New Delhi.

Sadoulet, E., De Janvry, A., and Benjamin, C. (1998). Household behavior with imper-
fect labor markets. Industrial Relations: A Journal of Economy and Society, 37(1):85–
108.

Saha, A., Havenner, A., and Talpaz, H. (1997). Stochastic production function estima-
tion: small sample properties of ML versus FGLS. Applied Economics, 29(4):459–469.

Sanghi, A. and Mendelsohn, R. (2008). The impacts of global warming on farmers in

Brazil and India. Global Environmental Change, 18(4):655–665.

Sanghi, A., Mendelsohn, R., and Dinar, A. (1998). The climate sensitivity of Indian
agriculture. In Dinar, A., Mendelsohn, R., Evenson, R., Parikh, J., Sanghi, A., Kumar,
K., McKinsey, J., and Lonergan, S., editors, Measuring the impact of climate change
on Indian agriculture, pages 69–139. World Bank Technical Paper No. 402, World
Bank, Washington, DC.

Schlenker, W., Hanemann, W. M., and Fisher, A. C. (2005). Will US agriculture really
benefit from global warming? Accounting for irrigation in the hedonic approach.
The American Economic Review, 95(1):395–406.

Schlenker, W., Hanemann, W. M., and Fisher, A. C. (2006). The impact of global warm-
ing on US agriculture: an econometric analysis of optimal growing conditions. The
Review of Economics and Statistics, 88(1):113–125.

Schlenker, W. and Lobell, D. B. (2010). Robust negative impacts of climate change on

African agriculture. Environmental Research Letters, 5(1):014010.

Schlenker, W. and Roberts, M. J. (2009). Nonlinear temperature effects indicate severe

damages to US crop yields under climate change. Proceedings of the National
Academy of sciences, 106(37):15594–15598.

Schmidhuber, J. and Tubiello, F. N. (2007). Global food security under climate change.
Proceedings of the National Academy of Sciences, 104(50):19703–19708.
216 Bibliography

Schmitt-Grohé, S. and Uribe, M. (2003). Closing small open economy models. Journal
of international Economics, 61(1):163–185.

Schmitt-Grohé, S. and Uribe, M. (2012). What’s news in business cycles. Econometrica,

80(6):2733–2764.

Sheehy, J. E., Mitchell, P. L., and Ferrer, A. B. (2006). Decline in rice grain yields with
temperature: models and correlations can give different estimates. Field Crops Re-
search, 98(2):151–156.

Sicular, T. and Zhao, Y. (2004). China and the WTO: accession, policy reform, and poverty
reduction strategies, chapter Earnings and Labor Mobility in Rural China: Implica-
tions for China’s Accession to the WTO. World Bank Publications.

Sims, C. A. and Zha, T. (1998). Bayesian methods for dynamic multivariate models.
International Economic Review, pages 949–968.

Singh, I., Squire, L., and Strauss, J. (1986). Agricultural household models: Extensions,
applications, and policy. The World Bank.

Skoufias, E. (1994). Using shadow wages to estimate labor supply of agricultural house-
holds. American Journal of Agricultural Economics, 76(2):215–227.

Smets, F. and Wouters, R. (2003). An estimated dynamic stochastic general equilibrium

model of the euro area. Journal of the European Economic Association, 1(5):1123–
1175.

Smets, F. and Wouters, R. (2007). Shocks and frictions in US business cycles: A

Bayesian DSGE approach. American Economic Review, 97(3):586–606.

Srinivasan, P. and Jha, S. (2001). Liberalized trade and domestic price stability. the case
of rice and wheat in India. Journal of Development Economics, 65(2):417–441.

Stern, N. H. (2007). The economics of climate change: the Stern review. cambridge
University press.

Swain, A. K. (2014). Can India reform its agriculture? | the diplomat.

Taylor, J. E. and Adelman, I. (2003). Agricultural household models: genesis, evolution,

and extensions. Review of Economics of the Household, 1(1-2):33–58.
 Bibliography 217

Tilman, D., Balzer, C., Hill, J., and Befort, B. L. (2011). Global food demand and the
sustainable intensification of agriculture. Proceedings of the National Academy of
Sciences, 108(50):20260–20264.

Tol, R. S. (1995). The damage costs of climate change toward more comprehensive
calculations. Environmental and Resource Economics, 5(4):353–374.

Tol, R. S. (2002). Estimates of the damage costs of climate change. Part 1: Benchmark
estimates. Environmental and resource Economics, 21(1):47–73.

Tol, R. S. (2005). The marginal damage costs of carbon dioxide emissions: an assess-
ment of the uncertainties. Energy policy, 33(16):2064–2074.

Tol, R. S. (2009). The economic effects of climate change. The Journal of Economic
Perspectives, pages 29–51.

Tubiello, F. N. and Fischer, G. (2007). Reducing climate change impacts on agriculture:

Global and regional effects of mitigation, 2000–2080. Technological Forecasting and
Social Change, 74(7):1030–1056.

Van Passel, S., Massetti, E., and Mendelsohn, R. (2016). A Ricardian analysis of the
impact of climate change on European agriculture. Environmental and Resource
Economics, pages 1–36.

Van Vuuren, D. P., Edmonds, J., Kainuma, M., Riahi, K., Thomson, A., Hibbard, K.,
Hurtt, G. C., Kram, T., Krey, V., Lamarque, J.-F., et al. (2011). The representative
concentration pathways: an overview. Climatic change, 109(1-2):5.

Vanschoenwinkel, J., Mendelsohn, R., and Van Passel, S. (2016). Do Western and East-
ern Europe have the same agricultural climate response? Taking adaptive capacity
into account. Global Environmental Change, 41:74–87.

Waggoner, D. F. and Zha, T. (2003). Likelihood preserving normalization in multiple

equation models. Journal of Econometrics, 114(2):329–347.

Wang, J., Mendelsohn, R., Dinar, A., Huang, J., Rozelle, S., and Zhang, L. (2009). The
impact of climate change on China’s agriculture. Agricultural Economics, 40(3):323–
337.

Weitzman, M. L. (2010). What is the” damages function” for global warming—and what
difference might it make? Climate Change Economics, 1(01):57–69.
218 Bibliography

Xiong, W., Declan, C., Erda, L., Yinlong, X., Hui, J., Jinhe, J., Ian, H., and Yan, L. (2009).
Future cereal production in China: the interaction of climate change, water avail-
ability and socio-economic scenarios. Global Environmental Change, 19(1):34–44.

You, L., Rosegrant, M. W., Wood, S., and Sun, D. (2009). Impact of growing season
temperature on wheat productivity in China. Agricultural and Forest Meteorology,
149(6):1009–1014.
 VU : VU :
Le Directeur de Thèse Le Responsable de l'École Doctorale
(Nom et Prénom)

VU pour autorisation de soutenance

Rennes, le

Le Président de l'Université de Rennes 1

David ALIS

VU après soutenance pour autorisation de publication :

Le Président de Jury,
(Nom et Prénom)
 Changement climatique et agriculture
Le climat de la planète se réchauffe et ses effets sont entachés d’une forte incertitude. Une hausse
de la température et de la fréquence d’événements extrêmes tels des inondations ou des séche-
resses est prévue. La forte dépendance de l’agriculture aux conditions climatiques en fait de facto
un champ d’application privilégié. Cette thèse se destine ainsi à étudier la relation entre climat
et agriculture, afin d’évaluer les conséquences potentielles du changement climatique, en mêlant
travaux empiriques et théoriques. Les deux premiers chapitres se concentrent sur les pays en déve-
loppement au travers de deux études examinant la production et les profits agricoles ainsi que les
décisions de consommation des ménages agricoles indiens. Les divers scénarios climatiques en-
visagés montrent un effet global négatif sur la production et les profits, particulièrement pour les
ménages agricoles du sud du pays. L’irrigation tout comme le mélange des cultures permettent
toutefois de réduire les dommages subis, notamment pour les petits exploitants. Les deux cha-
pitres suivants considèrent des pays économiquement développés, en commençant par une étude
des rendements céréaliers européens. Les projections sous les différents scénarios climatiques in-
diquent une faible croissance des rendements du blé d’ici à la fin du XXIe siècle, comparativement
aux observations des 25 dernières années. Ces gains faibles sont toutefois accompagnés d’une
forte hétérogénéité régionale. Pour le maïs, des faibles gains d’ici la moitié du XXIe s’effacent
derrières de plus fortes pertes dans le long terme. L’approche partielle est ensuite délaissée pour
laisser place à une analyse en équilibre général s’attachant à étudier les effets de court terme des
chocs climatiques sur les cycles économiques, à travers leur impact sur l’agriculture. Une hausse
de la variance des chocs climatiques conformément à celle prévue par des scénarios climatiques
entraîne un accroissement substantiel de variables macroéconomiques telles la production et l’in-
flation.

Mots clés : Agriculture, Changement Climatique, Cycles Économiques, Ménages Agricoles, Petite
Économie Ouverte, Régression Quantile, Rendements Céréaliers, Système de Demande (AIDS).

Climate Change and Agriculture

Global climate is warming, and the effects of climate change are associated with a lot of uncer-
tainty. Not only average temperatures are expected to rise, but also the occurrence of extreme
events such as floods or droughts. Agriculture is particularly at risk, due to the importance of
weather conditions in production. This thesis therefore aims at investigating the relationship
between weather variations and agricultural production, to better assess the potential effects of
climate change on agriculture, relying on both theoretical and empirical methods. The first two
chapters focus on developing countries and provide two empirical studies based on Indian data
at the individual farm level that link climate to agricultural production and profits and to con-
sumption decisions. We find contrasted results, with an overall damaging effect of climate change
scenarios on Indian agricultural production and profits, especially for farmers in southern India.
Irrigation may however help mitigating the losses, as well as crop mixing, particularly for small
farms. The last two chapters consider developed countries. The first step focuses on crop yields
in Europe. Under the tested climate scenarios, wheat yields are projected to slightly increase by
the end of the 21st century relative to the observed yields from the past 25 years. These small
gains are however accompanied by a lot of regional heterogeneity. For European corn yields,
the projections highlight small gains in by the middle of the 21st century, followed by relatively
higher losses in the long run. The second step relies on a general equilibrium approach, and
aims at investigating the short-run impacts of weather shocks on business cycles, through their
damaging effects on agriculture. Increasing the variance of climate shocks in accordance with
forthcoming climate change leads to a sizeable increase in the volatility of key macroeconomic
variables, such as production and inflation.

Keywords : Agriculture; Business Cycles; Climate Change; Demand System (AIDS); Crop Yields;
Household Behaviour; Quantile Regression; Small Open Economy.