An Introduction to

Market Microstructure Theory

UOB 1:32:18

57 11/16 –1 3/16 -2.02% 1,238,723

Last prices:
1:32:16 57 11/16 –1/16 500
1:32:12 57 3/4 +1/8 1,400
1:32:11 57 5/8 –1/8 925
1:32:05 57 3/4 –1/16 2,300
1:32:01 57 13/16 +1/8 270

Bid Ask
57 5/8 1,125 57 3/4 852
57 9/16 8,200 57 13/16 2,150
57 1/2 5,127 57 15/16 13,760
57 7/16 17,942 58 25,487
57 3/8 21,474 58 1/16 47,723

Andreas Krause
An Introduction to Market Microstructure Theory

Andreas Krause
School of Management
University of Bath
Andreas Krause
University of Bath
School of Management
Claverton Down
Bath BA2 7AY
Great Britain

E-Mail: [email protected]

First draft
Version of June 21, 2005

c Andreas Krause 2005

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any forms or by any means, without the
prior permission of the author, or as expressively permitted by law. Enquiries
concerning reproduction outside the scope of the above should be sent to the
author at the address above.

Typeset using LATEX

This is a preliminary draft of the book.
All comments are welcome.
Preface

There are three main fields in finance: asset pricing, corporate finance and market
microstructure. Asset pricing encompasses not only the theories on the valua-
tion of securities such as shares, corporate bonds, foreign exchange or derivatives,
but also how prices may systematically deviate from these valuations. The field
of corporate finance investigates financial decisions of companies such as invest-
ments, capital structure and dividend payouts. Finally market microstructure
analyzes the effect trading rules have on market prices, it thus deals with the
institutional setting of the trading process.
It is now widely acknowledged among academics as well as practitioners that
the rules governing the trading on exchanges can have a significant impact on
the success of a market. It is thus of importance to assess the trading rules of
an exchange and how they affect the behavior of investors, e.g. through implicit
and explicit trading costs. In addition competition between market participants
as well as different market forms has to be taken into account. In traditional
neoclassical models the trading process is assumed to be frictionless and instan-
taneous. As a consequence of this assumption the way trades are conducted in
markets is not relevant for the outcome. Once this assumption is lifted, however,
the trading rules become very much relevant for the outcome.
Over the last three decades models have been developed to investigate the im-
pact trading rules have on market participants. This line of research has become
known as market microstructure theory.1
1
The term market microstructure has first been introduced by Garman (1976). According
to Easley and O’Hara (1995, p. 357) it is
”the study of the process and outcomes of exchanging assets under explicit trading
rules.”
ii Preface

Market microstructure theories are not interested in the fundamental value of

the assets traded on an exchange, which is assumed to be determined exogenously,
but to model the price formation in asset markets with a given fundamental value
and how this is affected by the trading rules.
After some pioneering work by Demsetz (1968) and Bagehot (1971) the
first models developed by Stoll (1978) focused on dealer markets and inven-
tory effects that influence the price setting of market makers. Starting with
Copeland and Galai (1983) for dealer markets and Kyle (1985) for auction
markets, the attention changed from inventory effects to that of different infor-
mation between market participants. These models dominate the literature since
this time, addressing important questions like informational efficiency of prices
and market liquidity. More recently increased attention has been paid to limit
order markets and the order submission strategies of traders therein.

Structure of the book

The aim of this book is to provide an introduction to the main models used in
market microstructure theory upon which much of the more advanced literature
relies. A common topic throughout this text will be the presence of asymmetric
information between market participants and other direct and indirect trading
costs.
We will start in chapter 1 with an overview of the trading rules commonly
found in stock markets before the following three chapters then cover different
market forms. Chapter 2 focuses on auction markets and the effect asymmetric
information between traders has on the evolution of the market price. Theories on
dealer markets in chapter 3 deal with the price setting behavior of market makers
with and without the presence of asymmetric information between them and
traders. In chapter 4 limit order markets are considered and the optimal strategy
for investors to submit limit orders is discussed; where models usually abstract
from the presence of asymmetric information for simplicity. These theoretical
chapters bring up many questions which can be investigated empirically, such as
 iii

the relative importance of different aspects for the spread, informed trading or the
relevance of market microstructure elements for daily asset returns. Chapter 5
provides an overview of the key empirical methods used in market microstructure
analysis.
While in most cases the market form is taken as given, the differences between
them can be significant such that the choice of the market form is potentially im-
portant for the prices of traded assets. How the main market forms may evolve
endogenously is modeled in chapter 6. The final chapter 7 shows the impor-
tance of market microstructure for the prices of assets by linking its outcomes to
asset pricing theory. An outlook towards more advanced and specialized topics
in market microstructure concludes the main body of the text.
Besides a detailed overview of the regulation of the NASDAQ Stock Mar-
ket in appendix A, the appendix provides a brief introduction to important
mathematical methods in appendix B and key economic and financial concepts
in appendix C. The reader unfamiliar with the presented ideas might wish to
consult them prior to working through the main part of the text.
Throughout this book the emphasis is laid on theoretical models rather than
empirical methods to estimate them. Readers more interested in empirical meth-
ods are referred to the original articles, many of which are mentioned throughout
the book and which in many cases provide more details on these aspects. As many
derivations of models are presented in more detail than the original articles, it
allows the reader to fully appreciate the mathematical foundations of these mod-
els. It has however to be emphasized that it is of more relevance to understand
the reasoning behind a result rather than its mathematical derivation.

Use of the book

This book is aimed at taught postgraduate students specializing in finance, al-
though advanced undergraduate students might find much of the material ac-
cessible. It will also be a valuable introduction for beginning PhD students as a
help in their exploration of the field. The detailed mathematical derivation of the
iv Preface

models will provide them also with insights towards the problem of solving often
complex models. Taught postgraduate students and their teachers will also find
this feature of the book useful as it allows the class discussion to focus more on
the interpretation of results rather than their derivation. From chapter 1 onwards
the key readings at the beginning of each chapter represent the most important
original articles in which the presented models have been developed.

In order to successfully use this book readers should be familiar with basic
microeconomic theories as well as key financial concepts. Although appendix C
provides a short introduction to these theories, readers without such knowledge
are advised to consult basic books on microeconomics and finance in parallel. It is
furthermore required that readers have been introduced to differentiation and in-
tegration as well as matrix algebra and stochastics. Although some slightly more
advanced concepts are covered in appendix B, it does not serve as a substitute
for such a background.

Each chapter features a set of ten review questions of varying difficulty

which allow the reader to his knowledge. Although all these questions address
problems in the respective chapters, some require additional reading and research
before they can be answered properly. The aim of these questions is to enable
the reader to test his own knowledge on the topics learnt. Chapters 2, 3, 4, 6
and 7 also feature an application, where the applications in chapters 2 to the
first application in chapter 6 all build on each other. These applications require
the reader to combine different and sometimes contrasting theories to solve a
given problem. In most cases there will not be a single ’correct’ answer but
different foci will lend themselves towards producing very different conclusions.
Much can be gained from discussing the applications in small groups to explore a
range of opinions thoroughly using several market microstructure models where
appropriate.
 v

An appeal for help

The copy of the book you have in front of you is a first draft. Inevitably there
will be a large number of spelling mistakes, grammatical errors and much more
importantly mistakes in formulae, be it in forms of wrong indices, wrong symbols,
missing brackets or even more serious mistakes. You can also expect the one or
other cross-reference to be incorrect or references not being up-to-date.
Even more annoyingly to the reader than all these errors, some explanations
given might not be as clear as they could be or in the worst case might even be
wrong due some mix up during writing. You might also find that some topics are
under-represented or over-represented and you would like to see a change.2
Please feel free to send me any comments you have, ideally with a suggestion
for improvement but it is often also of value just to point out that you think
there might be problem although you do not have a solution at hand. Sending
me your comments and thoughts on this draft is an easy way to enter the hall
of fame, aka getting your name mentioned in the acknowledgements in the final
version.

Thank you very much for your help to improve this book.

June 2005 A.K.

Bath

2
You also may just think that I should cite your paper when discussing an aspect of the
theory.
vi Preface
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

1. The organization of trading in stock markets . . . . . . . . . . . 1

1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The importance of organized stock markets . . . . . . . . . . . . . 5
1.3 The different market forms . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Market participants and order submission . . . . . . . . . . . . . 17
1.5 Trading priority rules . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.6 Electronic trading mechanisms . . . . . . . . . . . . . . . . . . . . 22
1.7 Recent developments . . . . . . . . . . . . . . . . . . . . . . . . . 25
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2. Auction Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1 Auctions with informed investors and noise traders . . . . . . . . 31
2.1.1 Auctions with a single informed investor . . . . . . . . . . 31
2.1.2 Auctions with multiple informed investors . . . . . . . . . 41
2.2 Auctions with strategic uninformed investors . . . . . . . . . . . . 46
2.2.1 Markets with a single asset . . . . . . . . . . . . . . . . . . 46
2.2.2 Auction markets with multiple assets . . . . . . . . . . . . 59
2.3 The informational content of trading volume . . . . . . . . . . . . 64
2.4 Explaining short-term movements of asset prices . . . . . . . . . . 70
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3. Dealer Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.1 Inventory-Based Models of Market Making . . . . . . . . . . . . . 76
3.1.1 The costs of market making . . . . . . . . . . . . . . . . . 79
3.1.2 Competitive price setting . . . . . . . . . . . . . . . . . . . 87
3.1.3 The price setting of a monopolistic market maker . . . . . 93
3.2 Information-Based Models of Market Making . . . . . . . . . . . . 99
3.2.1 Determination of adverse selection costs . . . . . . . . . . 102
viii Contents

3.2.2 Simultaneous trading at different stock exchanges . . . . . 111

3.2.3 Comparing competitive and monopolistic market makers . 112
3.2.4 Explaining return patterns . . . . . . . . . . . . . . . . . . 114
3.3 Dealer markets with multiple assets . . . . . . . . . . . . . . . . . 116
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

4. Limit Order Markets . . . . . . . . . . . . . . . . . . . . . . . . . . 125

4.1 Static models of limit order placement . . . . . . . . . . . . . . . 127
4.2 Price dynamics with limit order trading . . . . . . . . . . . . . . . 130
4.3 Dynamic models of limit order submissions . . . . . . . . . . . . . 137
4.4 Informational efficiency with limit order trading . . . . . . . . . . 150
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

5. Empirical investigations of market microstructure models . . . 153

5.1 Estimating spread components . . . . . . . . . . . . . . . . . . . . 154
5.2 Estimating informed trading . . . . . . . . . . . . . . . . . . . . . 161
5.3 Properties of daily returns . . . . . . . . . . . . . . . . . . . . . . 164
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6. The emergence of different market forms . . . . . . . . . . . . . 169

6.1 Demand and Supply of Immediacy . . . . . . . . . . . . . . . . . 171
6.2 Determination of the optimal market form . . . . . . . . . . . . . 178
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

7. Asset Pricing with Trading Frictions . . . . . . . . . . . . . . . . 193

7.1 Required returns in the presence of bid-ask spreads . . . . . . . . 194
7.2 Compensation for adverse selection costs . . . . . . . . . . . . . . 198
7.3 The price impact of trading . . . . . . . . . . . . . . . . . . . . . 202
7.4 Pricing liquidity risk . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.5 The equity premium puzzle . . . . . . . . . . . . . . . . . . . . . 208
Review questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

A. The NASDAQ Stock Market . . . . . . . . . . . . . . . . . . . . . 221

A.1 History of the NASDAQ Stock Market . . . . . . . . . . . . . . . 222
A.2 The organization of the NASD . . . . . . . . . . . . . . . . . . . . 230
A.3 Listing requirements . . . . . . . . . . . . . . . . . . . . . . . . . 232
A.4 Registration as broker and market maker . . . . . . . . . . . . . . 235
A.5 Trading rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Contents ix

A.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

B. Mathematical methods . . . . . . . . . . . . . . . . . . . . . . . . . 243

B.1 Taylor series expansion . . . . . . . . . . . . . . . . . . . . . . . . 243
B.2 Dynamic programming . . . . . . . . . . . . . . . . . . . . . . . . 244
B.3 Constrained optimization . . . . . . . . . . . . . . . . . . . . . . . 245
B.4 Conditional moments . . . . . . . . . . . . . . . . . . . . . . . . . 247
B.4.1 Truncated distributions . . . . . . . . . . . . . . . . . . . . 247
B.4.2 Joint distributions . . . . . . . . . . . . . . . . . . . . . . 248
B.5 Stochastic processes . . . . . . . . . . . . . . . . . . . . . . . . . . 249
B.5.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
B.5.2 Itô’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 250

C. Economic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

C.1 Utility theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
C.1.1 The expected utility hypothesis . . . . . . . . . . . . . . . 251
C.1.2 Risk aversion . . . . . . . . . . . . . . . . . . . . . . . . . 255
C.2 Portfolio selection theory . . . . . . . . . . . . . . . . . . . . . . . 257
C.2.1 The mean-variance criterion . . . . . . . . . . . . . . . . . 258
C.2.2 The Markowitz frontier . . . . . . . . . . . . . . . . . . . . 264
C.3 The Capital Asset Pricing Model . . . . . . . . . . . . . . . . . . 274
C.4 The Rational Expectations Approach . . . . . . . . . . . . . . . . 279
C.5 Bayesian learning . . . . . . . . . . . . . . . . . . . . . . . . . . . 282

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284

Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
x Contents
List of Figures

1.1 Classification of market forms . . . . . . . . . . . . . . . . . . . . 11

1.2 Order flow in different market forms . . . . . . . . . . . . . . . . . 18

2.1 Liquidity in the Kyle (1985) model . . . . . . . . . . . . . . . . . 42

2.2 Liquidity with different number of auctions . . . . . . . . . . . . . 47
2.3 Informativeness of prices with different number of auctions . . . . 47
2.4 Liquidity with different number of informed investors . . . . . . . 48
2.5 Informativeness of prices with different number of informed investors 48
2.6 Market liquidity with varying number of uninformed investors . . 57
2.7 Market liquidity with varying number of informed investors . . . . 57
2.8 Market liquidity with different uncertainties about the liquidation
value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.9 Trading volume with different precision of information and changes
of beliefs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.10 Trading volume with different shares of informed investors and
changes of beliefs . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.1 Demand and Supply in Dealer Markets . . . . . . . . . . . . . . . 78

3.2 Costs of a market maker . . . . . . . . . . . . . . . . . . . . . . . 80
3.3 Competitive price setting . . . . . . . . . . . . . . . . . . . . . . . 87

4.1 The placement of limit orders . . . . . . . . . . . . . . . . . . . . 131

4.2 Order submission strategies . . . . . . . . . . . . . . . . . . . . . 135
4.3 Influence of the trader composition and trade frequency on the
spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.4 Influence of the trader composition and trade frequency on the
waiting time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.5 Influence of the trader composition and trade frequency on the
duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.6 Influence of the tick size and waiting costs on the spread with θ = 0.7146
4.7 Influence of the tick size and waiting costs on the waiting time
with θ = 0.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.8 Influence of the tick size and waiting costs on the duration with
θ = 0.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.9 Influence of the tick size and waiting costs on the spread with θ = 0.3148
4.10 Influence of the tick size and waiting costs on the waiting time
with θ = 0.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
xii List of Figures

4.11 Influence of the tick size and waiting costs on the duration with
θ = 0.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

5.1 Sequences of transaction prices with the initial price at the bid . . 157
5.2 Evolution of the order flow with informed trading . . . . . . . . . 162

A.1 Distribution of the number of market makers for NASDAQ secu-

rities in 1996 Data: NASDAQ Factbook 1997 . . . . . . . . . . . . . . 222
A.2 Daily US-Dollar trading volume on the NYSE and the NASDAQ . 228
A.3 Daily share trading volume on the NYSE and the NASDAQ . . . 229
A.4 Development of the Dow Jones Industrial Average and NASDAQ
Composite Index Data: Datastream . . . . . . . . . . . . . . . . . . 229
A.5 Organization of the NASD . . . . . . . . . . . . . . . . . . . . . . 230

C.1 The Arrow-Pratt measure of risk aversion . . . . . . . . . . . . . 257

C.2 The mean-variance criterion . . . . . . . . . . . . . . . . . . . . . 259
C.3 The efficient frontier . . . . . . . . . . . . . . . . . . . . . . . . . 260
C.4 Determination of the optimal alternative . . . . . . . . . . . . . . 263
C.5 Efficient portfolios with two assets . . . . . . . . . . . . . . . . . . 267
C.6 Determination of the optimal portfolio with two assets . . . . . . 268
C.7 Portfolio selection with three assets . . . . . . . . . . . . . . . . . 269
C.8 The optimal portfolio with N > 2 assets . . . . . . . . . . . . . . 269
C.9 The optimal portfolio with a riskless asset . . . . . . . . . . . . . 271
C.10 Portfolio selection with short sales . . . . . . . . . . . . . . . . . . 272
C.11 The Capital Asset Pricing Model . . . . . . . . . . . . . . . . . . 278
List of Tables

1.1 Market forms of selected stock exchanges . . . . . . . . . . . . . . 14

1.2 Overview of alliances between major stock and futures exchanges 26

5.1 Parameter constellations for the spread components . . . . . . . . 157

5.2 Estimates of the spread components . . . . . . . . . . . . . . . . . 160

A.1 Main historical events of the NASDAQ . . . . . . . . . . . . . . . 227

xiv List of Tables
Chapter 1

The organization of trading in stock

markets

———————————————————————————————————
The main aim of this chapter is to introduce the reader to
• the different ways assets can be traded in markets
• the different types of orders that traders can use in asset markets
• the way orders are filled in asset markets
• recent developments in stock market trading

Key readings:
Ruben Lee: What is an Exchange? The Automation, Management and Regula-
tion of Financial Markets, Oxford: Oxford University Press, 1998
———————————————————————————————————
2 Chapter 1. The organization of trading in stock markets

When investigating the market microstructure it is essential to be aware about

the way prices are determined during trading in real markets and the incentives
of individuals to submit their orders. Although rules can vary significantly across
markets, we can derive some general properties upon which markets base their
actual rules.
The aim of this chapter is not to give a survey of all possible regulations
and special forms found at different stock exchanges, but to explain the basic
principles.1 A special focus is laid on those aspects that have an impact on the
price setting in these markets.2
After some preliminary definitions in chapter 1.1, the importance of stock
markets for investors, companies and the society as a whole will be pointed out
in chapter 1.2; chapter 1.3 describes the different forms of trading on an exchange
and chapter 1.4 shows how orders are submitted to an exchange, who participates
in trading and what forms of orders exist; according to which rules orders are
executed is explained in chapter 1.5; chapter 1.6 deals with the influence of com-
puterization on the organization of stock markets and trading; finally, chapter
1.7 points out some recent developments in financial markets.

1.1 Definitions
A market is a place where supply and demand for a good meet, i.e. potential
sellers and buyers of a good submit their supply and demand schedules. This
place has not to be a certain location,

”... but the entire territory of which the parts are united by the
relations of unrestricted commerce that prices there take the same
level throughout with ease and rapidity.”3
1
Lee (1998) gives a very detailed and theoretically based overview of the current regulation
of stock markets in the United States, including recent developments in electronic trading.
2
An overview of the regulation of stock markets is given in Schwartz (1993, ch. 2 and 4)
for the NYSE and the London Stock Exchange. A detailed description of the regulations in
the United Kingdom, United States, Switzerland, France, Netherlands, Austria, and Japan can
be found in Hopt et al. (1997, Part III) and in Hopt and Baum (1997) for Germany. A
history of stock exchange regulations in Europe is given in Merkt (1997).
3
Cournot (1838, p. 42).
1.1. Definitions 3

To achieve this tendency towards an equal price, Jevons (1911) emphasizes

the importance of communication between market participants, i.e. the trans-
mission of information, especially prices:

”The traders may be spread over a whole town, or region of country,

and yet make a market, if they are (...) in close communication with
each other.”4

Marshall states the characteristics a good must have to be traded in a

market:5

• The good has to be widely demanded. A good that is not widely demanded
cannot be traded frequently, hence there is only infrequent intercourse and
too few prices that are needed to form a market as stated in the above
definition of Cournot.

• The good has to be homogeneous.6 If a good is not homogeneous, e.g.

pieces of art or special machines, the pieces are not interchangeable with
each other, hence they will fail to have a tendency towards equal prices as
needed by the definition of a market.

• The good has to be transferable and storable at low costs, compared to its
value. High costs of transferring the rights or storing the good would hinder
the free intercourse wanted by Cournot to form a market. The gains that
have to be made from trading the good to offset these costs would be too
high.

Technological innovations reduced the costs of transfer and storage signifi-

cantly in recent years, thereby increasing the number of goods that can be traded
in markets. New information technologies enable traders to communicate at low
costs over long distances, enlarging the region that can be viewed as a market.
4
Jevons (1911, p. 81).
5
See Marshall (1920, pp. 141f.).
6
Goods are homogeneous if they are interchangeable, i.e. one piece of a good has exactly
the same characteristics as another.
4 Chapter 1. The organization of trading in stock markets

Further improvements in standardization made more goods homogeneous, so that

they can be traded in markets nowadays. The reduction in costs and standardiza-
tion, besides further improvements, also increased the demand for many goods.
Hence today more and more goods can be and are traded in markets.
Securities are rights that are chartered in a document. To execute or transfer
this right the document has to be presented. Financial securities are securities,
where the rights are a sequence of future cash flows. The future cash flow can
consist of money (interest, dividends) or other financial securities (e.g. stocks or
bonds). Most securities fit into one of the following categories: stocks, bonds,
derivatives7 or a combination of these. When addressing securities in economics,
financial securities are principally being referred to. We will stick to this tradition
in the remaining parts of this chapter.8
Securities are typically divided into smaller parts, each part representing a
fraction of the whole, hence securities are homogeneous. They are transferable at
very low costs and storage costs can be neglected.9 As securities are also widely
demanded as a means for investments, they fulfill all characteristics to be traded
in a market.10 These markets are called security markets.11
We have to distinguish two forms of security markets: primary and secondary
markets. In primary markets new shares of a security are issued, i.e. for the first
time sold to an investor,12 while in secondary markets already issued shares of
securities are traded among investors.
7
A derivative is a security whose future cash flow depends on the value of another security.
8
Another frequently used expression for financial securities is asset. In relation with prices
and valuation this term is the most frequently used, while in relation with market regulations
the term securities is more common. Therefore, starting in chapter 2, we will use the expression
asset for securities.
9
Many securities do not exist physically nowadays and if they do, they are stored at a
clearing house, imposing very low storage costs. Formerly securities had to be handed over,
but today the transfer usually is only booked into accounts kept at the clearing house.
10
There are sometimes securities issued that are not widely demanded. They typically vanish
after a short period of time.
11
Often the term financial markets also refers only to security markets, although they in
general encompass also rights that are no securities, such as bank loans or deposits. Often
foreign exchange markets and even the commodity markets are included into the term financial
markets, as the characteristics of these markets are very similar.
12
We call investors all those, who are invested in securities or are interested in being so
(potential investors).
1.2. The importance of organized stock markets 5

If the amount of an existing security is increased, the new shares can be issued
either by organizing a separate auction or by selling them in the open market, i.e.
they are issued by placing a sell order from the issuer in the secondary market.
In this case the issuer behaves like an investor. Although this form of issuing
formally belongs to the primary market, it is referred to as an operation in the
secondary market. Sometimes this form of increasing the amount of an existing
security is also applied for issuing a new security, especially in OTC markets for
derivatives.
In futures and options exchanges there does not exist a secondary market.
All transactions take place in the primary market according to the definition
above. If an investor buys a derivative, another investor has to issue a new
unit of this security, an existing derivative cannot be bought. The outstanding
amount of these derivatives is not fixed as in the case of stocks or bonds. If an
investor wants to sell a derivative he has bought, he issues a derivative which
exactly offsets the derivative he wants to sell.13 Formally, all these operations
must be placed into the category of primary markets. But since they have all
the characteristics of a typical trading activity, they are assigned to secondary
markets.
In most cases, security markets are being referred to the category of secondary
markets. This convention will also be applied in this text. The remaining analysis
will be concentrating on stock markets, but most concepts and findings can easily
be adapted to other security markets.

1.2 The importance of organized stock markets

In most cases the time horizons of investors and the issuer of a security do not
coincide. While companies have a need for capital of very long, even infinite
13
For every derivative such an offsetting derivative exists because it can be found in two
forms: as a long and as a short position. An investor is long if he has bought a security, he is
short if has issued the security. In stock and bond markets investors usually are long and the
company who has issued the security is short. However, there is the possibility of short sales
by investors in many markets, so that only on average investors have to hold a long position.
Adding a short and a long position of the same security exactly offsets the investor.
6 Chapter 1. The organization of trading in stock markets

disposability, investors on the other hand may want to change their investments
to adjust for new information, changed tastes, or liquidity needs. As the amount
of outstanding shares is fixed (disregarding periodic increases and repurchases
of capital) all shares have to be held by investors. To adjust their investments,
investors have to trade them with each other. Keynes (1936, p. 151) pointed
this aspect out as follows:

”In the absence of security markets, there is no object in frequently

attempting to revalue an investment to which we are committed. But
the Stock Exchange revalues many investments every day and the
revaluations give a frequent opportunity to the individual (though
not the community as a whole) to revise his commitments.”

If no securities markets exist, it is very difficult to find another investor who

wants to take the counterpart in a trade. The lack of communication between
investors makes the search for a counterpart very costly.14 When a counterpart
is finally found (e.g. by advertising) the next problem is to find a price at which
both are willing to trade. The valuation of a stock depends on the information
an investor has, inconsistent information will make it difficult to find a suitable
price.15 Even if both can agree on a price, the trade may occur at a ”wrong”,
i.e. informationally inefficient, price16 when neither participant has precise in-
formation. This gives wrong incentives to investors, resulting in an inefficient
allocation of resources.
To overcome these difficulties markets provide investors with two services:
liquidity and the aggregation and revelation of information.
14
Chapter 6.2 presents a formal model showing how these costs can give rise to the formation
of organized markets.
15
An overview of asset valuation can be found in nearly all textbooks on finance. Good
synopsis are Ingersoll (1987), Duffie (1996) and Cochrane (2000), besides others.
16
Prices are called informationally efficient if they fully reflect all available information.
Fama (1970) distinguishes three forms of efficiency, depending on the information available:
weak, semi-strong and strong efficiency. Prices are weakly efficient if they reflect only informa-
tion derived from previous prices or returns; if all publicly available information is reflected in
the price, it is called semi-strong efficient. In cases where also private information available to
any market participant is reflected, prices exhibit strong efficiency.
1.2. The importance of organized stock markets 7

By providing liquidity, markets facilitate the exchange of assets between in-

vestors. As investors meet on markets, the costs of searching a counterpart for
a trade is reduced significantly, a counterpart can easily be found by address-
ing the market. Competition between investors for a trade will further ensure
that a better, informationally more efficient price will be charged. These reduced
costs of trading will allow investors to adjust their investment decisions more
frequently to their information and tastes, resulting in a more efficient allocation
of resources.17

As markets generate prices, they can be used to reveal and aggregate infor-
mation on a security.18 Compared to other sources of information, prices can be
observed at nearly no costs. Without having much additional costs, an investor
can increase his information and in this way reduce the risk to trade at a disad-
vantageous price resulting from a lack of information. This further reduces his
costs of trading.

By holding a portfolio that fits better his tastes and information, an investor
reaches a higher level of utility. The reduced costs of adjusting his investment
decisions increase his returns and hence the price he is willing to pay for an asset.
This benefits also the issuers of assets as they can issue their assets at higher
prices, reducing their costs of capital and increasing profits.19 Increased profits
give incentives for further investments and hence promote economic growth.

From the existence of security markets investors profit from higher returns
and a better allocated portfolio, issuers of securities from lower costs of capital
and higher profits and the society as a whole from a more efficient allocation of
resources, higher investments and growth. Therefore everyone benefits from the
existence of securities markets.

17
Arrow (1964) stresses the importance of asset markets for an efficient allocation of risks
between individuals.
18
It was Hayek (1945) to point out the importance of prices as a source of information. How
prices can aggregate and reveal information to investors is discussed in more detail in chapters
2 and 3.2.1.
19
See Keynes (1930, Vol. II, p. 195).
8 Chapter 1. The organization of trading in stock markets

A further reduction in the costs of trading can be achieved by applying uniform

rules to the trading process. Security markets that apply a fixed set of rules
governing trading are called organized security markets or an exchange.20 The
rules of an exchange should regulate21

The admission of stocks. The stocks to be traded on an exchange have to

meet certain standards in order to guarantee a minimum of investor protec-
tion and ensure regular trading. These standards differ very much between
exchanges.

The access to the market. It has to be determined who is allowed to trade

directly in the market and what conditions have to be met for this access.
By applying certain standards, the risk of a counterpart failing to fulfill his
duties after a trade has been negotiated, can be reduced.

The types of orders that can be submitted. A standardization of the or-

der types simplifies the trading process and the set of rules can be held
small.

The execution of orders. Rules on the execution of an order include the rules
to determine the prices at which a trade occurs and when trades are ex-
ecuted. These rules can avoid the problem of some investors gaining the
advantage at the cost of others, e.g. as a consequence of personal links to
other market participants.

The clearing of trades. A standardization of the clearing process, i.e. the

way and time trades are settled, avoids a separate negotiation on this point.
Most exchanges do not only set rules for the clearing process, but organize
it by themselves through special clearing houses.
20
A more formal definition of an exchange, based on the current legislation can be found
in Lee (1998). Rudolph and Röhrl (1997, p. 168) also apply a similar definition of an
exchange and point out that recent innovations, especially the computerization of exchanges
(see also chapter 1.6), made it necessary to adapt the traditional legal definition of an exchange.
21
Appendix A describes thee regulation of the NASDAQ Stock Market in more detail as an
example of how these principles are applied in reality.
1.3. The different market forms 9

Besides the reduction of trading costs, the standardization of trading also

makes prices more comparable to each other, both over time and between assets.
The prices therewith can better aggregate and reveal information.
The rules imposed on the trading process are called the market structure.22
They can be imposed by public regulation or self-regulation of the market. In
most cases a combination of public regulation and self-regulation can be found.
The structure of an exchange will influence the costs of trading, which in turn
will affect the price formation. We will therefore investigate the process of price
formation in more detail in this chapter by giving a short overview of the basic
market forms.

1.3 The different market forms

The market form describes the way trades occur in a market. In standard neoclas-
sical theory the market form implicitly assumed is that of an Walrasian auction-
eer. The auctioneer suggests a price to the investors. The investors23 determine
which amount they are willing to buy or to sell at the stated price and submit
their decisions, called orders, to the auctioneer. If demand and supply do not
exactly equal, no trade occurs. The auctioneer suggests a new price and the
investors revise their decisions. This process continues until aggregate demand
exactly equals aggregate supply. The price at which aggregate demand and sup-
ply are balanced, is called the equilibrium price or market clearing price. If the
equilibrium price is found, all orders are executed in a single multilateral trade.
This process mostly is assumed to be finished in an instant of time, imposing
no costs on investors. In reality, however, the revision of orders and the an-
nouncement of a new price would take considerable time and impose high costs,
especially if many investors are trading on the market. Furthermore, it would
be pure incident if a price could be found that clears the market exactly. Both,
22
See O’Hara (1995, p. 1).
23
In this section use the term investor for all market participants and assume that they
directly interact at the exchange. However, in reality most investors have to use an agent for
trading. This part of the structure of an exchange will further be discussed in the following
sections.
10 Chapter 1. The organization of trading in stock markets

prices and quantities are discrete, either by definition as no fractions of a stock

can be traded or by convention, e.g. discrete prices. Additional rules have to be
applied to determine the price with these small imbalances.
For these reasons a Walrasian auctioneer seldom exists in real markets, other
market forms have been established over time.24 There exists a wide variety of
market forms around the globe, every exchange has its unique way of executing a
trade. Also within an exchange there often exists different market segments, each
having its own rules.25 Despite these differences in detail, the market forms can
be organized in one of six main market forms. Figure 1.1 shows the classification
of market forms as will be used below.
In all market forms orders26 can be submitted to the market at any time, what
differs is the time and way these orders are executed. A market form which is very
close to the concept of an Walrasian auctioneer is the batch system or (periodic)
call market. In batch systems incoming orders are not executed immediately, but
stored and executed in a multilateral trade at a predetermined point of time. The
price that will be applied for this trade, is the price at which most orders can be
executed, i.e. the price with the highest trading volume. We find batch systems
in two forms: à la criée and par cassier.
Trading à la criée allows investors to revise their orders until the time of
execution. To give additional information for the revision of orders, the price that
would be applied if all orders were to be executed immediately, is continuously
published.27
24
The only examples of markets that use the concept of a Walrasian auctioneer are the
London Gold Fixings at 10.30 am and 3 pm and the Frankfurt Foreign Exchange Fixing for
selected currencies at 12 am that has been ceased to exist after December 30, 1998 with the
introduction of the EURO.
25
Different rules can often be found for frequently and infrequently traded stocks. The
Frankfurt Stock Exchange with its official quotation (Amtliche Notierung), regulated unofficial
market (Freiverkehr) and Neuer Markt is a good example for an exchange with different market
segments applying different sets of rules.
26
We find two forms of orders: orders that specify the worst price (highest price for a buy
order, lowest price for a sell order) at which they can be executed (limit orders) and orders to
be executed at any price (market orders). These order forms are presented in more detail in
section 1.3.
27
The term à la criée refers to the verbal order submission to the auctioneer that has been
used at the Paris Stock Exchange prior to the introduction of electronic trading in 1986. How-
 Security Markets

? ?
Batch Systems Continuous Markets

?
Matching Systems
1.3. The different market forms

? ? ? ? ? ?
à la criée par cassier Order Book Board Trading Crowd Trading Dealer Markets

? ?
Open Closed

Fig. 1.1: Classification of market forms

11
12 Chapter 1. The organization of trading in stock markets

Trading par cassier does not allow investors to revise their orders and typically
the price that would be applied in case of immediate execution of all orders is
not published.28
Batch trading can rarely be found in regular markets. It is only frequently
used to determine the opening and sometimes the closing price of a trading day
and to determine the price of some infrequently traded stocks.29
In continuous markets trades can occur not only at predetermined points of
time, as in batch trading, but at any time two orders can be executed. For every
submitted order it is immediately checked whether there exists another order in
the market, such that these orders can be executed in a bilateral trade. If no
such order exists, the order is stored and executed with the next matching order
arriving in the market. We find two forms of continuous markets: dealer markets
and matching systems.
In dealer markets special market participants, called dealers, have the obli-
gation to ”make the market”, hence they are also called market makers. Every
market maker has publicly to set prices at which he is willing to sell (ask price)
and to buy (bid price) the security.30,31 The bid and the ask prices have not to
be, and will not be, equal, as we will see in chapter 3. At the stated price the
market maker has to sell (buy) the security immediately from (to) any investor
demanding this. The market maker trades on his own account, i.e. he forms the
counterpart of the investor.
In matching systems no special market participants exist to form the counter-
part by trading on their own account. The trades are only bilaterally executed
between two investors. Three different forms of matching systems are known:
order book systems, board trading and crowd trading.
ever, it is not relevant whether the order is submitted verbal or written, the important feature
of this market form is the publication of the price and the possibility to revise orders. The
name is only kept for historical reasons.
28
Like à la criée the term par cassier is kept for historical reasons, as in this market form
orders were submitted written. Important is only the impossibility to revise orders.
29
See also table 1.1.
30
When setting these prices, the market maker does not know whether the next order arriving
at the market is a buy or a sell order, what its size is and when an order will arrive.
31
The prices a market maker sets are also called quotes or quoted prices.
1.3. The different market forms 13

In order book systems all submitted orders are stored in an order book. If
two orders cross, they are immediately executed and the price and volume of the
trade are published. In some cases the order book keeper also acts as a market
maker, like on the NYSE. A further distinction can be made whether the order
book is open or closed to the public, i.e. if the investors can look into the order
book or not. Mixed forms can also be found where the order book keeper can
give some information, as on the Frankfurt Stock Exchange.

In board trading the prices at which investors are willing to trade are also
entered into an order book, but only the best (highest) bid and the best (lowest)
ask prices are published on the board, e.g. the two best prices on the Hong Kong
Stock Exchange. If an order arrives in the market and accepts the best price
stated on the board, they are immediately executed. The price at which the
trade takes place is published on a separate board and the executed orders are
cancelled from the order book. In this system the trade size is fixed to what is
called a lot. This facilitates trading because the size of an order has not to be
considered in matching them.

The third form of matching systems is crowd trading. Investors meet on the
trading floor of the exchange and discuss the prices at which they are willing to
conduct trades. If two investors agree upon a trade, their orders are executed
and the price is published.

In another classification matching and batch systems are also called auction
markets, to distinguish in this classification between matching and batch systems,
matching systems are called continuous auctions.

As can be seen from table 1.1 every market form can be found on least at one of
the leading stock exchanges.32 There is no dominating market form to be found,
what suggests that every market form has its advantages and disadvantages,
although currently we observe a tendency towards order book systems.

32
Only crowd trading cannot be found since the Swiss Exchange Zürich changed to an elec-
tronic trading platform in 1996. Of the more important exchanges nowadays only the London
Metal Exchange applies crowd trading, but it is also planned to introduce electronic trading in
the near future and hence change the market structure.
Chapter 1. The organization of trading in stock markets

Batch Systems Continuous Markets

Matching Systems
à la criée par cassier Order Book Board Trading Crowd Trading Dealer Markets
Frankfurt Stock Exchange
Amtliche Notierung opening closed
Freiverkehr X
Neuer Markt closed (X)
Hong Kong Stock Exchange X
London Stock Exchange X
NASDAQ X
NYSE opening closed X
Tokyo Stock Exchange opening open
Swiss Exchange Zürich X
Paris Stock Exchange X
Vienna Stock Exchange X
Tab. 1.1: Market forms of selected stock exchanges
14
1.3. The different market forms 15

Batch systems have the advantage of collecting orders over a longer period
of time. Large order imbalances that may occur over time, e.g. a large order
arriving in the market, will have a smaller effect on prices than with an immediate
execution of the order. Often an imbalance is reduced over time and the volatility
of prices diminishes. If the trading is à la criée, investors can react to this
imbalance as they can observe that the price would change significantly with
execution. On the other hand, to revise an order imposes not only costs on
investors, but also on the exchange. The order flow becomes difficult to handle
and errors are more likely to occur than in trading par cassier. On many stock
exchanges with matching systems all orders accumulated over night are cleared
immediately at the beginning of the trading session in one multilateral trade at
a single price. This is advantageous for the determination of opening prices. If
they had to be executed in subsequent bilateral trades, these orders would hinder
the handling of orders submitted at the beginning of the trading session.33
On the other hand, batch systems have the disadvantage that trades occur
only a few times per day (once or twice). Investors have to wait a considerable
time until they are able to trade the next time. A reaction to new information is
not possible immediately, imposing waiting costs on investors. As less prices are
available to investors, the aggregation and revelation of information through the
price system cannot be assured as good as in continuous markets.34
The advantage of faster execution of an order and therewith reduced waiting
costs is one of the main arguments for continuous markets. In dealer markets the
market maker guarantees immediate execution of an order, but, as we will see
in chapter 3, he will not provide this service for free. The fees charged by the
market maker may counteract the advantages of this market form.
Matching systems impose no additional fees on investors35 , but therefore im-
33
The NASDAQ has no call auction at the opening, hence trading volume is very high and
the execution of orders submitted at the beginning of the trading hours takes a considerable
time. Therefore they are currently considering to introduce a call auction at the opening.
34
With trading à la criée prices are published continuously, but they will be biased up to short
before the execution. In order to save costs, investors will not adjust their orders permanently,
but only once just prior to the execution.
35
We neglect here for simplicity any direct fees levied by exchanges, brokers and any taxes
16 Chapter 1. The organization of trading in stock markets

mediate execution is not guaranteed as a matching order has to be found before

an order can be executed, hence investors may also face waiting costs. Especially
for infrequently traded stocks, the time until an order can be executed may be
very long. This makes the differences to batch systems in terms of waiting costs
less important, whereas the volatility of prices will in general be higher, as an
order may have a substantial effect on prices. This favors batch systems for in-
frequently traded stocks, whereas for frequently traded stocks the impact of not
too large orders on the price can be neglected, favoring matching systems as the
result of more frequent trading.
Even for frequently traded stocks, where no considerable waiting costs exist,
matching systems have the disadvantage that the price at which the trade will
be executed, is not known in advance. It depends also on the order with which it
is matched, whereas in dealer markets the investor knows the stated price of the
market maker at which he will trade. This uncertainty on the price may favor
dealer markets despite the fee market makers charge.
Crowd systems allow only for small trading volumes. The time to negoti-
ate the price does not allow for too many investors trading. But this system
ensures the best price in continuous markets, as all orders compete directly for
a trade. Board trading is capable of handling larger trading volumes and or-
der book systems are more flexible to handle orders of different sizes. With the
ability of handling larger volumes due to advanced computerization, the costs of
conducting a trade decreases for the exchange and hence for investors.
As can be seen from the above discussion, every market form has its advan-
tages and disadvantages. The optimal market form depends on the nature of the
stock, e.g. the frequency it is traded, and the personal tastes of the investors,
e.g. their risk aversion or time preferences. A discussion of these aspects can
be found in Angel (1997), where he suggests optimal trading rules for stocks of
small companies.
In addition to the market forms presented above, many mixtures, like the

to be paid.
1.4. Market participants and order submission 17

NYSE, and variations of these pure forms exist. Special trading forms, e.g. off-
exchange trades for large orders (block trading) or special trading facilities for
small orders, complete the list of market forms.

1.4 Market participants and order submission

In the last sections it has been assumed for the sake of simplicity that investors
trade directly on the exchange. In reality, however, they have to use an agent,
who trades for them on the exchange. This agent is called a broker.36 A broker
transmits the order he receives from an investor, his customer, to the exchange,
where the order is treated according to the rules of the exchange. He does not
trade on his own account. The broker also informs the investor about the execu-
tion and the applied price of his order and settles the accounts. He also is liable
for fulfilling the trade, hence the counterpart risk is reduced significantly.
In dealer markets the broker has to identify the market maker37 offering the
best price and transmit the order to this market maker for immediate execution.
In matching and batch systems the broker only transmits the order to a match
maker 38 and waits for its execution. Figure 1.2 visualizes the way orders are
submitted in different market forms.
Often the roles of market participants change. A market maker or broker
may want to trade for his own account and hence by our definition become an
investor, or on the NYSE the market maker also is keeper of the order book, i.e.
match maker. Despite these mixtures of roles individual market participants have
at a particular time, their activities will fit into one of the following categories:
investor, broker, market maker or match maker.
36
There are special conditions that must be met to act as broker. In our context these
conditions are of no interest and are therefore omitted here. Appendix A.4 gives a detailed
description of the conditions to be met for being granted access as broker to the NASDAQ.
37
To become a market maker very strict conditions have to be fulfilled. In some markets
only one market maker per asset is allowed, e.g. the specialists at the NYSE. Appendix A.4
gives a detailed description of the conditions to be met for being granted access as broker to
the NASDAQ.
38
A match maker is a person that stores the order, i.e. keeps the order book, and initiates
the matching of the orders.
18 Chapter 1. The organization of trading in stock markets

Investor Trading Decision Investor Trading Decision Trading Decision

Submission Submission

Broker ? Broker ? ?

Transmission Transmission
? ? ?
Market maker Execution Match maker Matching Matching

?
Execution

(a) Dealer markets (b) Matching system

Investor Trading Decision ... Trading Decision

Submission

Broker ? ? ? ? ? ? ?

Transmission
? ? ? ? ? ? ?
Match maker Wait for execution ... Wait for execution

?
Execution

(c) Batch system

Fig. 1.2: Order flow in different market forms

1.4. Market participants and order submission 19

The size of an order is not fixed, except in board trading. A market maker has
to accept any order size at the stated price as long as the order is not too large.
Large orders are normally divided into several smaller orders and executed over
a longer time period, ranging between hours and months to avoid a significant
influence on the price.
In most markets there exists a ”normal” order size, as in board trading called
a lot.39 Orders with the size of a multiple of a lot can be divided into smaller,
separate orders with the minimum size of one lot. These smaller orders will be
executed separately with different matching orders or market makers at different
points of time and prices. This splitting of large orders facilitates the execution
of such an order and increases liquidity. If the orders were to be executed as a
whole, it may be difficult to find a matching order with exactly the same size.
Orders with a size smaller than one lot (so called odd-lots)40 are sometimes
traded by special market makers or match makers, respectively, or a batch system
is introduced for these small orders. The Fixing on the Frankfurt Stock Exchange
at 12.00 noon is an example for such a batch system.
So far we only considered orders that were for execution at the best available
price of the market. Such an order is called a market order. A market order will
be executed at any price, in frequently traded stocks with continuous markets
it normally will be executed within a short time after submission, as any offset-
ting order matches. In batch systems market orders are executed nearly with
certainty.41
There exists another frequently used type of order, limit order. When sub-
mitting a limit order, the investor sets a maximum (minimum) price, the limit,
at which he is willing to buy (sell) the security. Of course, he also will buy (sell)

39
The typical lot on the Hong Kong Stock Exchange is 5,000 shares. In 1998 the Frankfurt
Stock Exchange changed its lot size from 100 shares to 1 share. On the NASDAQ the lot sizes
vary between 100 and 1000 shares, depending on the stocks.
40
Orders that are larger than one lot, but are not a multiple of a lot are split into a part
consisting of a multiple of a lot and a part with the odd-lot. They are then treated as different
orders, see Keenan (1987, p. 23).
41
There may be some special situations in which the execution is not guaranteed, but these
situations have no practical relevance.
20 Chapter 1. The organization of trading in stock markets

at lower (higher) prices. The execution of limit orders is not guaranteed, an

offsetting order has to be found that fulfills these conditions. In most matching
and batch systems the trade between two market orders is given priority over
the trade of a limit and a market order. Therefore it can take a considerable
time until a limit order can be executed. Limit orders will be canceled either at
expiration, e.g. at the end of the day or upon withdrawal.
There exist many other order forms, but they are rarely applied in security
markets. One order form that has been important in the past is the stop loss
order. Stop loss orders are executed as a market sell order if the market price
falls below a certain limit, otherwise the order is not executed. This type of order
has been used in the past, when an investor had not the possibility to observe the
market continuously or contact his broker immediately. The aim was to prevent
greater losses in these cases. Nowadays it is possible to receive recent news on
security prices and communicate without problems nearly all over the world and
stop loss orders can rarely be found.42

1.5 Trading priority rules

We can frequently run into the situation where there is more than one order
unexecuted in the market which matches an incoming order, e.g. two limit orders
submitted which could be filled by an offsetting market order arriving at the
market. In this case it is important to establish rules deciding which of these
orders will be executed, called trading priority rules. They not only have an
impact on the time an investor has to wait until his order is executed and hence
his waiting costs, but also on the price applied.43
The most important rule is price priority. With price priority market orders
42
Schwartz (1988, pp. 45 ff) also gives an overview is given of other order forms that are
possible to submit to US stock exchanges. However, these order forms are only rarely found
and for this reason not further considered here.
43
Moulin (2000) provides an axiomatic treatment of priority rules. He shows that in each
market there exists only a single optimal priority rule, however, his analysis does not allow him
to determine this rule. Domowitz (1993) gives an overview of the priority rules applied on
several important stock markets.
1.5. Trading priority rules 21

are executed first and only after all market orders have been executed, the limit
order with the best price, i.e. lowest price for a buy order and highest price for
a sell order, is executed, then the limit order with the second best price, and so
forth. This rule ensures that securities can be bought at the lowest and be sold
at the highest available price, reducing trading costs. Price priority is found as
the first priority rule at all stock exchanges.
The price priority rule will in many cases not be sufficient to distinguish
between all unexecuted orders in the market. It will often be found that more
than one order has been submitted at the same price or is a market order, so that
we need additional rules for choosing the order that is executed with a matching
order. These rules are called secondary trading priority rules.
The most common rule is time priority. An order that has been transmitted
earlier by the broker is executed before an order transmitted later.44 Another
rule that frequently can be found, often in combination with time priority, is size
priority. With size priority a larger order is executed before a smaller order. This
rule in combination with time priority can be found on the NYSE and since 1996
on the Toronto Stock Exchange, when the secondary priority rule was change
from pure time priority.
In dealer markets the rule public before dealer is very important. If a limit
order45 submitted by an investor has the same limit as the price quoted by the
market maker, the limit order is executed first.46 Public before dealer is applied
by the NASDAQ since 1997, while most other dealer market do not have this
rule.
There are many other rules that have only minor importance in leading stock
markets, such as pro rata partial execution. If at a certain price there is an

44
The time at which an order is transmitted, in most cases at which it is entered into the
computer system of the exchange, is measured in hundredth of seconds to ensure a clear dis-
tinction between all orders, also in times of high volume. In dealer markets similar rules can
be established to determine the market maker executing an incoming order.
45
Market orders have to be executed immediately either by the market maker or by a limit
order. Consequently, there can be no unexecuted market orders.
46
If more than one limit order is unexecuted at this price they are distinguished by another
rule, e.g. time priority.
22 Chapter 1. The organization of trading in stock markets

imbalance in the orders, i.e. not all orders can be executed on one side, all
orders on the larger side are only partially executed with the same fraction. This
rule may be applied in batch systems. The least complicated rule is the random
selection of the order that is executed.
Not every rule can be applied in all market forms. There exists a wide variety
of further rules, exceptions and modifications that are specific to certain stock
exchange.
The rules and market forms presented here form only a part of the market
structure. Additional rules not yet mentioned encompass maximum price change
limits, lower transaction costs for certain groups of market participants, besides
others. An exhaustive description of all possibilities to form a market structure
lies beyond the scope of this chapter.47

1.6 Electronic trading mechanisms

In recent years exchanges have computerized more and more functions. At the
beginning of this process brokers only used computers to facilitate their order
handling, e.g. the settlement of trades with their clients, the supervision of
order execution and the clearing of trades. Exchanges used computers only for
displaying and storing publicized data and for the clearing process. Orders had
to be transmitted in conventional ways, i.e. verbal or written, from the broker
to the exchange. Market makers and match makers as well as brokers had to be
physically present on the trading floor of the exchange.48
Later the computer was used to assist market makers and match makers in
handling the order flow by ordering the orders according to the priority trading
rules and displaying the relevant information. The orders had to be entered into
the computers by the market makers and match makers themselves.
47
Appendix A gives a more detailed overview of the market structure of the NASDAQ,
including some, but not all, features omitted here. Rudolph and Röhrl (1997) give a more
detailed overview of the economics and current state of stock exchange regulation.
48
There existed also OTC markets that had no trading floor, but used a telephone for com-
munication (telephone markets). On these markets in most cases only very infrequently traded
stocks were listed, an exception has been the NASDAQ.
1.6. Electronic trading mechanisms 23

In 1973 the Frankfurt Stock Exchange had developed the first automated
order handling system. It allowed brokers to transmit their orders electronically
from their internal computer systems into the computer system of the exchange.
The order was then automatically routed to the appropriate match maker. The
execution of the orders still had to be done manually. Although the system has
never been introduced in Frankfurt, other exchanges developed similar systems.
These systems, e.g. the DOT -System of the NYSE implemented in 1976, were
at the beginning only able to handle a small number of orders. For this reason
the use was restricted to small orders. With time computer systems were able to
handle more orders and the use was extended.49
The last step towards a fully automated trading system (or electronic ex-
change) was first taken in 1977 with the introduction of CATS on the Toronto
Stock Exchange. It enabled not only the electronic transmission of orders, but
without any interference was able to execute orders.50 Similar to automated or-
der handling systems introduced earlier, these systems were at the beginning only
able to handle small amounts of orders and were therefore only used for the trade
of infrequently traded stocks or the use was restricted to small orders. In 1991 the
Frankfurt Stock Exchange introduced the IBIS trading system, where all major
stocks could be traded.51 But this trading system is not regarded as the official
stock exchange, which still is a conventional exchange, it was initially established
as a system for inter-bank trading. With the reduction of the minimum order
size to 100 shares in 1998 this trading platform is now also open to the general
public. The first official stock exchange to introduce a fully automated trading
system for all traded securities and order sizes, including an electronic clearing
of trades, was the Swiss Exchange in Zürich (EBS) in 1996. In the mean time,
more and more exchanges have introduced an electronic exchange, at least for a
49
The SuperDOT 250, implemented in November 1984 on the NYSE could handle a daily
volume of 250 million shares. The capacity has been extended since then. Nowadays more than
1 billion shares can be traded on this system without problems. See Schwartz (1988, p. 27).
50
In 1982 a similar system was introduced on the Tokyo Stock Exchange, see Schwartz
(1988, p. 27), and 1986 on the Paris Stock Exchange, see Schwartz (1993, p. 90).
51
A new version with the name XETRA is used since November 1997. Unser and Oehler
(1998) provide a concise introduction into the features of this system.
24 Chapter 1. The organization of trading in stock markets

part of their trading activities.52

In dealer markets, also with a fully automated trading system, a minimum of
inference is needed. The system can automatically route the orders to the market
maker quoting the best price, who then only has to confirm the trade and enter
his new prices into the system. In matching or batch systems a match maker is
no longer needed, his duties can be taken over entirely by the computer system.
With the introduction of a fully automated trading system, a trading floor
is no longer needed. All market participants only have to be connected to the
computer system of the exchange. Their locations have no importance, they
can trade from any place around the globe. If market participants gain access
from another country this is called remote access and enables twenty-four hour
trading on the same exchange if the market participants are located in different
time zones.53
Computerization per se does not change the market structure or forces to
do so.54 In many cases, however, the introduction of a new computer system
is used to establish a new market structure to increase the efficiency of trading.
In many cases order book systems are introduced, replacing or complementing
market makers, like on the London Stock Exchange.
Nevertheless, computerization has had a great impact on trading: orders can
be executed more accurately to the rules, as errors are reduced, they furthermore
can be executed faster by a computer system, trading costs are lowered and
trade information (volume and prices) can be transmitted faster. Investors are
also able to react more quickly to changes in the market, as access to real-time
data has become affordable to a wide public55 and investors can submit their
orders electronically to their broker, e.g. by using the internet.
52
Especially the futures exchanges e.g. EUREX or LIFFE have introduced electronic trading.
53
The GLOBEX system of the CME, introduced in 1992, was the first, and by now the only
system that enables a twenty-four hour trading on the same exchange.
54
As has been pointed out above, only exchanges with crowd trading are not able to preserve
their market form, as this form needs the physical presence of the brokers on the trading floor
to negotiate the price.
55
Several brokers display real-time data for their customers on the internet for free. The costs
of data providers also have decreased substantially in recent years due to competition from the
internet.
1.7. Recent developments 25

Before the computerization of exchanges, a broker had to find the best price
for an order, what in dealer markets with many competing market makers was
a difficult task, because the quotes may change after every trade. In matching
markets it was important to transmit the orders as soon as possible to the ex-
change before the prices changed too much. With a computerized exchange the
best price is found by the computer system, the broker only has to transmit the
order he receives - more and more electronically - from his client. His service is
reduced to a pure transmission of the order. Due to this computerization and the
increased global competition of brokers, broker fees have decreased significantly
in the last years, reducing the costs to investors.56
As the broker no longer plays an active role, the submission of an order via
a broker, who immediately passes this order without any interferences to the
exchange, has the same effect as if the order would be directly submitted to the
exchange, i.e. as if the investors were directly interacting.57 Therefore the process
of computerization is also called the disintermediation of financial markets.

1.7 Recent developments

Stock and futures exchanges are currently challenged by a number of changes tak-
ing place. First of all the computerization has increased the competition between
exchanges. Due to the possibility of remote access in computerized markets,
the physical location of an exchange close to the most important financial in-
stitutions has become less important. This development increased competition
globally between exchanges for the listing of securities as well as for order flow.
In this competition trading costs are a very important factor. Therefore ex-
changes have to improve their market structure to reduce trading costs for in-
vestors. Increased fixed costs for developing, improving and maintaining the
56
For a trade of about USD 10,000 several brokers offer fees of less than USD 10 when the
order is placed using the internet, compared to fees of more than USD 100 not long ago and
still applied by conventional brokers.
57
The only reason, besides regulatory restrictions, that brokers are still used as intermediaires,
is to reduce the counterpart risk of a trade as brokers guarantee to fulfill the trades of their
customers.
26 Chapter 1. The organization of trading in stock markets

Partners Year Features

DTB, SOFFEX 1998 common trading platform
EUREX
NASDAQ, AMEX 1998 merger, separate markets
Paris, Zürich 1999 reciprocal access of members
EUREX, CBoT 1999 reciprocal access of members
LIFFE, CME 1999 reciprocal access of members
Paris, Lissabon 1999 reciprocal access of members
Paris, Brussels, Amsterdam 2000 merged to Euronext
Frankfurt, London 2000,2005 abandoned merger talks
NASDAQ, Instinet 2005 talks about merger
NYSE, Archipelago 2005 talk about merger

Tab. 1.2: Overview of alliances between major stock and futures exchanges

computer systems of an exchange with fast growing trading volumes lead to nu-
merous alliances between exchanges, seeking economies of scale. The number of
these alliances, cooperations and mergers are advancing very fast. Table 1.2 lists
some of the most important alliances, which in most cases guarantee reciprocal
access to the partners for all members of one exchange. In some cases a common
listing of securities, a common trading platform or even a merger are planned.
However, with exception of the EUREX trading platform,58 Euronext and the
NASDAQ/AMEX-merger none of the listed alliances is currently in operation.
The agreed alliances were often quietly dissolved or let to no further cooperations
between the exchanges after the burst of the internet boom in mid 2000 due to
the reduced trading activity during that time period.
Typically stock exchanges are organized as non-profit organizations supported
by its members, in general the financial institutions having direct access to the
market. Growing investments into the computerization made it more and more
difficult for exchanges to raise the necessary capital from its members. Therefore
several exchanges discuss the transformation into for-profit corporations which
would give them more flexibility in raising capital.59

58
A brief description of the EUREX trading system is given in Schiller and Marek (2000).
59
Such plans are under discussion at the NYSE, the NASDAQ, and the CBoT. The members
of the LSE, Frankfurt Stock Exchange, Euronext and the CME already have approved such
plans and the exchanges are quoted.
1.7. Recent developments 27

This development is further accelerated by the increasing number of Electronic

Communications Networks (ECNs), which are private-owned trading platforms
offering similar services as an exchange. These ECNs, which are mostly financed
and operated by large financial institutions, have gained a substantial market
share in trading NASDAQ securities.60 ECNs have a gained a market share of
more than a quarter in dollar trading volume of securities listed on the NAS-
DAQ.61 The importance of ECNs is witnessed in particular by the fact that both
the NASDAQ and NYSE have both planned to merge with one of their main
ECN competitors. We also have seen a small number of mergers between ECNs,
e.g. Island and Instinet, confirming the general trend towards consolidation of
exchnages.
A final development is the extension of trading hours. Especially institutional
investors are demanding longer trading hours, which enable them to react faster
on news they receive. On May 15, 2000 the Milano Stock Exchange extended
trading 65 large stocks until 8.30 pm and the Frankfurt Stock Exchange extended
their trading hours for all listed stocks until 8 pm on June 2, 2000.62 The lack of
demand in the aftermath of the internet bubble, however, let again to a reduction
of the trading hours.

60
The market share for securities listed on the NYSE can be neglected as regulation does gen-
erally not allow members of the NYSE to trade securities listed on the NYSE off exchange. Up
to now in Europe only a single ECN concentrating on securities listed on the LSE, Tradepoint,
exists, which has a negligible market share of less than 1% in trading volume.
61
Gomber (2000) gives an overview of the requirements for a successful electronic trading
system based on economic considerations.
62
The traditional trading hours on the Frankfurt Stock Exchange have been from 10.30 am
to 1.30 pm, currently most European markets operate from 9 am to 5 pm, the NYSE operates
from 9.30 am to 4 pm. Extensions to 8 or 10 pm were planned by most leading stock exchanges,
but after the burst of the internet bubble have never been implemented.
28 Chapter 1. The organization of trading in stock markets

Review questions
1. What are the characteristics a good must have to be traded in a market?

2. What is the role of security markets in an economy?

3. What characterizes an exchange?

4. Compare call and auction markets.

5. What are the relative advantages and disadvantages of dealer markets and
limit order markets?

6. What are the differences between market and limit orders?

7. Why do markets usually have fixed lot sizes?

8. Why is it important to have secondary priority rules?

9. What are the implications of computerized trading?

10. Why is there a tendency by exchanges to merge?

Chapter 2

Auction Markets

———————————————————————————————————
This chapter will introduce the reader to theories on auction markets. Particular
attention will be paid to the implications of asymmetric information between
investors. The main contents of this chapter evolves around
• the optimal behavior of informed investors
• the implications of this behavior for asset prices and trading volume
• the timing of trading

Key readings:
Albert S. Kyle: Continuous Auctions and Insider Trading, Econometrica, 53,
1315-1335, 1985

Craig W. Holden and Avanidhar Subrahmanyam: Long-Lived Private Informa-

tion and Imperfect Competition, Journal of Finance, 47, 247-270, 1992

Matthew Spiegel and Avanidhar Subrahmanyam: Informed Speculation and

Hedging in a Noncompetitive Securities Market, Review of Financial Studies, 5,
307-329, 1992

Lawrence Blume, David Easley and Maureen O’Hara: Market Statistics and
Technical Analysis: The Role of Volume, Journal of Finance, 49, 153-181, 1994
———————————————————————————————————
30 Chapter 2. Auction Markets

This chapter will give an overview of the main contributions in market mi-
crostructure theory on order-driven or auction markets. Prices are not analyzed
on a trade-by-trade basis, but aggregated over a given period of time, e.g. sev-
eral hours of a trading day or an entire trading day. These models allow to
explain short-term variations in price and trading volume, like the day-of-the-
weak-anomaly.

We assume initially that a single asset is traded in T trading rounds by two

groups of investors, informed and uninformed investors. After the final round of
trading the asset is liquidated and the proceedings are consumed. The informed
investors, called insiders,1 receive a signal on the liquidation value of the asset
before the first round of trading. Informed investors are assumed to be risk
neutral. The uninformed investors do not receive a signal, hence their trades are
not based on superior information, but on exogenous needs for liquidity (noise
traders) as in the first two models to be presented or to rebalance their portfolios
as a result of changing prices (hedgers) in the final model.

In each trading round (auction) investors can submit their orders to a match
maker.2 The match maker arranges the trades at a single price. This price
has to be set such that it equals the expected liquidation value given the match
makers information. The information the match maker has, as he is assumed to be
uninformed, is the order imbalance of that auction. As the match maker does not
know whether the orders submitted come from informed or uninformed investors
(the use of brokers ensures anonymity of the investors to the match maker) more
orders to buy than to sell could mean that informed investors received a signal
indicating that the liquidation value is above the last observed price. But it could
also be the result of pure incident. This price setting behavior is identical to the

1
The term insider as used in this context has to be distinguished from the legal definition
of an insider. Here an insider is an investor having acquired publicly available information. In
the legal definition an insider has access to not yet publicly available information due to his
position in a company, e.g. as member of the board.
2
We allow only for market orders, the submission of limit orders, i.e. of demand schedules,
is not allowed. When submitting an order the orders submitted by the other investors for this
auction are not known to any investor, i.e. the choices have to be simultaneous and cannot be
conditioned on the behavior of other investors.
2.1. Auctions with informed investors and noise traders 31

efficient market hypothesis, where the price equals the expected value of the asset
given a set of information.3 For simplicity it is assumed that the match maker
charges no fee for his service.
The next section presents the Kyle (1985) model. Although the model pre-
sented in section 2.1.1 is included as a special case in section 2.1.2 it is treated
separately due to its outstanding position in the development of theories on auc-
tion markets. It assumes a single informed investor and a large number of noise
traders. In section 2.1.2 this model is extended to include more than one in-
formed investor. A last extension in 2.2 assumes that the uninformed investors
are no longer noise traders, but risk averse hedgers maximizing their own utility.
Section 2.3 shows the information trading volume reveals. Finally in 2.4 it is
shown how the developed models can be used to explain some effects observed
in stock markets. The presentation of the models have been adopted to use the
same framework for all models.

2.1 Auctions with informed investors and noise

traders
2.1.1 Auctions with a single informed investor

Kyle (1985) presents a model where a single informed investor trades a single
asset together with N uninformed noise traders. It is assumed that the informed
investor becomes to know the liquidation value v of the asset with certainty.4 For
uninformed investors the liquidation value is a random variable ve that is normally
distributed with initial mean p0 and variance Σ0 :

(2.1) ve ∼ N (p0 , Σ0 ) .
3
Although this behavior is very restrictive, it captures some of the behaviors in real stock
exchanges applying auction markets. For example the Frankfurt Stock Exchange urges the
match makers to rise the price if more buy than sell orders are in the market. See Deutsche
Boerse Group (ed.): Rules and Regulations, Part 6, 3.22 and 3.3.1.2.
4
The original models assume that the signal is not perfect, but that the liquidation value
has a positive variance given this information. As informed investors are assumed to be risk
neutral this remaining risk has not to be considered in the optimal behavior as maximizing
expected profit and expected utility are equivalent.
32 Chapter 2. Auction Markets

Uninformed investors are assumed to trade for purely exogenous reasons, they do
ei , are random variables
not maximize any objective function. Their order sizes, u
that are assumed to be independently identically normally distributed with mean
zero and variance σu2i .5 They are independent of the order sizes of other unin-
formed investors, the behavior of the informed investor, independent over time
and of ve:

ei ∼ N 0, σu2i .


(2.2) u

The total relevant order flow from uninformed investors is their order imbalance,
all other orders can directly be matched:
N
X
ei ∼ N 0, N σu2i = N 0, σu2 .



(2.3) u
e= u
i=1

We assume that T , N , σu2i , p0 and Σ0 are known by all market participants as

well as the normal distribution of the variables.
The order size of the informed investor is denoted x
e. The match maker cannot
distinguish between orders from informed and uninformed investors, hence he only
can observe the aggregated order flow u
e+x
e. This order flow is used to determine
the price according to

(2.4) v |u + x].
pe = E[e

Kyle (1985) first investigates the optimal behavior of the informed investor
by choosing an optimal x
e in a single auction, i.e. for T = 1. The informed
investor maximizes his expected profits from trading given his information on
the liquidation value of the asset. His profits are

(2.5) π v − pe)e
e = (e x.

Only linear equlibria are considered by Kyle (1985). This assumption gives rise
5
In deriving the results the assumption of normality is central. Relaxing this assumption
to the class of elliptical functions gives similar results, but further generalizations may give
different results.
2.1. Auctions with informed investors and noise traders 33

to the following pricing rules:

(2.6) pe = µ + λ(e
x+u
e),

(2.7) x
e = α + βe
v.

With (2.5) - (2.7) we get for the expected profits

(2.8) π |v] = E[(e

E[e v − µ − λ(e
x+u x|v]
e))e

= (v − µ − λx)x.

Maximizing (2.8) to find the optimal order size of the informed investor gives the
following first order condition:

(2.9) v − µ − λx − λx = v − µ − 2λx = 0.

Rearranging results in
µ v
(2.10) x=− + .
2λ 2λ
Comparing coefficients with (2.7) we get
1
(2.11) β = ,
2λ
µ
α = − = −µβ.
2λ
The second order condition for a maximum

(2.12) − 2λ < 0,

states that we only have to consider positive λ. Using (2.4) we get with (2.1)
and (2.3) and the results of the conditional mean of jointly normally distributed
random variables:
Cov[ev, x
e+u e]
(2.13) v |e
pe = E[e x+u
e] = E[e
v] + (x + u − E[e
x+u e])
V ar[ex+u e]
Cov[ev , α + βe v+u e]
= p0 + (e
x+u e − E[α + βe v+ue])
V ar[α + βe v+u e]
βCov[e v , ve] + Cov[e v, u
e]
= p0 + 2 ×
β V ar[e v ] + V ar[e u] + 2Cov[e v, ue]
×(ex+u e − α − βE[e v − E[e u]])
βΣ0 βΣ0
= p0 + 2 2
(α + βp0 ) + 2 (e
x+ue).
β Σ0 + σu β Σ0 + σu2
34 Chapter 2. Auction Markets

By comparing coefficients with (2.6) we see that

βΣ0
(2.14) λ = ,
β 2Σ 2
0 + σu
βΣ0
µ = p0 − 2 (α + βp0 ) = p0 − λ(α + βp0 ).
β Σ0 + σu2

Solving (2.11) and (2.14) we get with (2.12):

s
σu2
(2.15) β = ,
Σ0
s
Σ0
λ = 2 ,
σu2
µ = p0 ,

α = −βp0 .

Therewith we can rewrite (2.6) and (2.7) as

(2.16) p = p0 + λ(x + u),

(2.17) x = β(v − p0 ).

The linear equilibrium exists and is unique. Nothing can be said about the
existence of further nonlinear equilibria.

Uninformed investors cannot observe the order flow, only the price that is set
by the match maker. Using this information they can update their beliefs on the
2.1. Auctions with informed investors and noise traders 35

distribution of the liquidation value:

Cov[e v , pe]2
(2.18) v |p] = V ar[e
Σ1 = V ar[e v] −
V ar[e p]
2
Cov[ev , p0 + λ(e x+u e)]
= Σ0 −
V ar[p0 + λ(e x+u e)]
2 2
λ Cov[e v, xe+u e]
= Σ0 −
λ2 V ar[e x+u e]
(Cov[e v, xe] + Cov[e v, u e])2
= Σ0 −
V ar[e
x] + V ar[e u + 2Cov[e x, u
e]]
Cov[e v , β(e v − p0 )]2
= Σ0 −
V ar[β(e v − p0 )] + σu2 + 2Cov[β(e v − p0 ), u
e]
2 2
β Cov[e v , ve]
= Σ0 − 2 2
β V ar[e v ] + σu + 2βCov[e v, u
e]
2 2
β Σ0
= Σ0 −
β Σ0 + σu2
2
2
σu
Σ2
Σ0 0
= Σ0 − 2
σu
Σ0 0
Σ + σu2
σu2 Σ0
= Σ0 −
2σu2
1
= Σ0 ,
2

Cov[ev , pe]
v |p] = E[e
(2.19) p1 = E[e v] + (p − E[ep])
V ar[ep]
Cov[ev , β(e v − p0 )]
= p0 + (p0 + λ(x + u) − E[p0 + λ(e
x+u
e)])
V ar[β(e v − p0 )]
βCov[e v , ve]
= p0 + 2 (λ(x + u − E[e x]))
β V ar[e v]
1
= p0 + λ(β(v − p0 ) + u − E[β(e v − p0 )])
β
λ
= p0 + λ(v − p0 ) + u − λ(E[e v − p0 ])
β
= p0 + λ(v − p0 ) + 2λ2 u

= p + 2λ2 u.

u|p] = 0 we find that E[e

As E[e p1 |p] = p, hence the posterior distribution of ve is
 
1
(2.20) ve|p ∼ N p, Σ0 .
2
36 Chapter 2. Auction Markets

The variance the uninformed investors attribute to the liquidation value of the
asset can be interpreted on how much information is incorporated into the price.
A variance of zero has to be interpreted as perfect revelation of the information
through prices, the closer this variance is to Σ0 the less informative the price is.
As by observing only the price the variance halves, it can be said that half of the
information is incorporated into prices. The variance of the liquidation value we
can view as a measure for the informativeness of prices.
λ measures the influence an additional unit of an order has on the price:

∂p
(2.21) = λ.
∂(x + u)

If we define a market as liquid if by placing an additional order the price does

not change. Therefore λ measures the liquidity of a market, the closer it is to
zero the more liquid the market is. Usually 1/λ is taken as a measure of liquidity
as by this definition a larger value corresponds to higher liquidity.
From (2.15) we see that a market is more liquid the more liquidity traders
are in the market, i.e. the larger N is, or the more their order sizes vary, i.e the
larger σu2i is.
This framework can now be extended to trades occurring in 1 < T < ∞
auctions in a given time period [0, 1].This setting allows the informed investor
to time his trades such that over time his expected profits are maximized. The
order flow from an investor for auction 1 ≤ t ≤ T is denoted uit for the order of an
uninformed investor. The variance of the order flow from uninformed investors,
σu2i ,t , remains constant over the entire period, i.e.

T
X
(2.22) σu2i = σu2i ,t = T σu2i ,t .
t=1

Hence we have
 
1
(2.23) uit ∼N 0, σu2i
T

for all t = 1, . . . , T . The total order flow from the informed investor in the first t
auctions is denoted xt and the order flow for a specific auction k ∆xk . Hence we
2.1. Auctions with informed investors and noise traders 37

have for all t = 1, . . . , T :

t
X
(2.24) xt = ∆xk .
k=1

When again considering only linear equilibria we get in analogy to (2.6) and (2.7)
for all t = 1, . . . , T :

(2.25) pet = µt + λt (∆xt + u

et ),

(2.26) ∆xt = αt + βt v.

The match maker sets his price again such that it equals the liquidation value of
the asset given the order flows observed in the past:

(2.27) v |Ωt ],
pt = E[e

where Ωt = {∆x1 + u1 , . . . , ∆xt + ut }. The informed investor maximizes his

expected profits by choosing the optimal order size given the liquidation value
and past prices. His profits from the remaining auctions are given by6
T
X
(2.28) π
et = v − pk )xk = (e
(e v − pt )∆xt + π
et+1 .
k=t

We assume the expected profits to be quadratic in v − pt :

(2.29) πt+1 |p1 , . . . , pt , v] = γt (v − pt )2 + δt .

E[e

Using (2.28) we get as the objective function of the informed investor:

πt+1 |p1 , . . . , pt , v] = E[(e

(2.30) E[e v − pt )∆xt + π
et+1 |p1 , . . . , pt , v]

v − pt )∆xt |p1 , . . . , pt , v]
= E[(e

+γt (v − pt )2 + δt

v − µt − λt (∆xt + u
= E[(e et ))∆xt |p1 , . . . , pt , v]

+γt (v − µt − λt (∆xt + ut ))2 + δt

= (v − µt − λt ∆xt )∆xt + δt + γt λ2t σu2i ,t

+γt (v − µt − λt (∆xt + ut ))2 .

6
It is worth noting that it is assumed here that future profits are not discounted to their
present value.
38 Chapter 2. Auction Markets

Maximizing this expression for the optimal order size to submit in auction t gives
the following first order condition:

(2.31) 0 = v − µt − λt ∆xt − λt ∆xt + 2γt (v − µt − λt ∆xt )(−λt )

= (v − µt )(1 − 2γt λt ) − 2λt ∆xt (1 − γt λt ).

Solving for ∆xt we get

1 − 2γt λt 1 − 2γt λt
(2.32) ∆xt = − µt + v.
2λt (1 − γt λt ) 2λt (1 − γt λt )
Comparing coefficients with (2.26) gives
1 − 2γt λt
(2.33) βt = ,
2λt (1 − γt λt )
1 − 2γt λt
(2.34) αt = − µt = −βt µt .
2λt (1 − γt λt )
The second order condition

(2.35) − 2λt (1 − γt λt ) < 0

implies that an equilibrium exists only if λt (1 − γt λt ) > 0. Using (2.27) we get

(2.36) v |Ωt ]
pt = E[e

v |Ωt−1 ]|∆xt + ut ]
= E[E[e

= E[ev |Ωt−1 ]
Cov[e v , ∆ext + u et |Ωt−1 ]
+ (∆xt + ut − E[∆e et |Ωt−1 ])
xt + u
V ar[∆e xt + u et |Ωt−1 ]
Cov[e v , αt + βt ve + u et |Ωt−1 ]
= pt−1 + ×
V ar[αt + βt ve + u et |Ωt−1 ]
×(∆xt + ut − E[αt + βt ve + u et |Ωt−1 ])
βt Cov[e v , ve|Ωt−1 ]
= pt−1 + 2 (∆ext + uet − αt − βt pt−1 )
βt V ar[e v |Ωt−1 ] + σu2i ,t
βt Σt−1
= pt−1 + 2 (∆e xt + uet − (αt + βt pt−1 )).
βt Σt−1 + σu2i ,t
Comparing coefficients with (2.25) we receive
βt Σt−1
(2.37) λt = ,
βt Σt−1 + σu2i ,t
2

βt Σt−1
µt = p t − 1 − (αt + βt pt−1 )
βt Σt−1 + σu2i ,t
2

= pt−1 − λt (αt + βt pt−1 ).

2.1. Auctions with informed investors and noise traders 39

Solving (2.33) and (2.37) we get

βt Σt−1
(2.38) λt = ,
βt Σt−1 + σu2i ,t
2

1 − 2λt γt
βt = ,
2λt (1 − λt γt )
µt = pt−1 ,

αt = −βt pt−1 .

Therewith we have found a unique linear equilibrium:

(2.39) pt = pt−1 + λt (∆xt + ut ),

(2.40) ∆xt = βt (v − pt−1 ).

By comparing the coefficients in (2.30) and (2.29) we see that

(2.41) δt−1 = δt + γt λ2t σu2i ,t ,

(2.42) γt−1 (v − pt−1 ) = (v − µt − λt ∆xt )∆xt + γt (v − µt − λt ∆xt )2

= (v − pt−1 − λt βt (v − pt−1 )) βt (v − pt−1 )

+γt (v − pt−1 − λt βt (v − pt−1 ))2

= (v − pt−1 )2 (1 − λt βt )βt

+γt (v − pt−1 )2 (1 − λt βt )2

= (v − pt−1 )2 (1 − λt βt )βt + γt (1 − λt βt )2 .


40 Chapter 2. Auction Markets

Hence

(2.43) γt−1 = (1 − λt βt )βt + γt (1 − λt βt )

   2
1 − 2γt λt 1 − 2γt λt
= 1 − λt βt + 1 − λt γt
2λt (1 − γt λt ) 2λt (1 − γt λt )
1 1
= βt + γt
2(1 − γt λt ) 4(1 − γt λt )2
2(1 − γt λt )βt + γt
=
4(1 − γt λt )2
2(1 − γt λt ) 2λ1−2γ t λt
t (1−γt λt
+ γt
=
4(1 − γt λt )2
1−2γt λt
λt
+ γt
=
4(1 − γt λt )2
1 − γt λt
=
4λt (1 − γt λt )2
1
= .
4λt (1 − γt λt )
The new variance of the liquidity value for the uninformed investors is determined
as

(2.44) v |pt ]
Σt = V ar[e

v |Ωt ]
= V ar[e
Cov[e v , ∆e
xt + uet |Ωt−1 ]2
v |Ωt−1 ] −
= V ar[e
V ar[∆e et |Ωt−1 ]
xt + u
Cov[e v − pt−1 )|Ωt−1 ]2
v , βt (e
= Σt−1 −
v − pt−1 ) + σu2i ,t |Ωt−1 ]
V ar[βt (e
βt2 Σ2t−1
= Σt−1 − 2
βt Σt−1 + σu2i ,t
= Σt−1 − λt βt Σt−1

= (1 − λt βt )Σt−1 .

Equations (2.38) - (2.41), (2.43) and (2.44) completely characterize the equilib-
rium. In order to avoid profits of the informed investor after the last auction, we
have to impose the boundary condition δT = γT = 0. It can now be shown that
with these two conditions the equilibrium exists, i.e. (2.35) is fulfilled, and that
it is a unique linear equilibrium.7
7
See Kyle (1985, pp. 1325 f.) for a formal proof. Again, there can exist other, nonlinear
2.1. Auctions with informed investors and noise traders 41

As we see from (2.37) 1 − λt βt is always between zero and one, hence the
variance of the liquidity value for the uninformed investors is strictly decreasing
as long as 0 < σu2i ,t < ∞. Therewith the information is gradually incorporated
into the price. An efficient market in the strong form is not achieved immediately,
i.e. prices do not reveal fully the information, including the private information
of the insider. They tend to become fully revealing over time. This behavior is
the result of the profit maximization of the insider, he exploits his informational
advantage over time by placing only small orders to hide his trades in the trades
of noise traders.
Figure 2.1 illustrates the behavior of the liquidity parameter λt depending on
the time and the number of auctions.8 It can be seen that the liquidity is falling
with time for all T , but as T increases λt becomes nearly constant over time. It is
optimal for the informed investor is to hold λt constant, the more auctions there
are, the better he can achieve this situation. Hence by placing an order the price
is always equally affected, i.e. the costs of trading (adverse selection costs) are
constant over time and prices adjust gradually.

2.1.2 Auctions with multiple informed investors

Holden and Subrahmanyam (1992) extended the model of Kyle (1985) to

incorporate multiple informed investors. The only change that has to be made
in the notation is to view the informed investors order flow, ∆xt , to be composed
of the order flows of M informed investors, ∆xit . Denoting ∆xit the conjecture of
informed investor i about the average order flow of the other informed investors,
we have
M
X
(2.45) ∆xt = ∆xkt = ∆xit + (M − 1)∆xit .
k=1

With the other assumptions identical to Kyle (1985) the derivation follows ex-
actly the steps already presented in the last subsection if we assume all informed
equilibria.
8
A method to solve the dynamic equations has been proposed by Holden and Subrah-
manyam (1992, pp. 253 f.).
42 Chapter 2. Auction Markets

Fig. 2.1: Liquidity in the Kyle (1985) model

investors to be equal with respect to all characteristics. An informed investor

maximizes in analogy to (2.30) with Ωp = {p1 , . . . , pt , v}:

(2.46) E[e v − pt )∆xit + π

πt+1 |Ωp ] = E[(e et+1 |Ωp ]

v − pt )∆xit |Ωp ] + γt (v − pt )2 + δt
= E[(e

v − µt − λt (∆xit + (M − 1)∆xit + u
= E[(e et ))∆xit |Ωp ]

+γt (v − µt − λt (∆xit + (M − 1)∆xit + ut ))2 + δt

= (v − µt − λt ∆xit + (M − 1)∆xit )∆xit

+γt (v − µt − λt (∆xit + (M − 1)∆xit + ut ))2

+δt + γt λ2t σu2i ,t .

This gives the first order condition

(2.47) 0 = v − µt − λt (∆xit + (M − 1)∆xit ) − λt ∆xit

+2γt (v − µt − λt (∆xit + (M − 1)∆xit ))(−λt )

= (v − µt − λt (M − 1)∆xit )(1 − 2γt λt ) − 2λt ∆xit (1 − γt λt ).

2.1. Auctions with informed investors and noise traders 43

As all informed investors are assumed to be equal, the only reasonable conjecture
about the other informed investor’s behavior is that they behave exactly in the
same way, i.e. ∆xit = ∆xit . Inserting this relation and solving for ∆xit gives the
optimal order size for informed investors:
1 − 2γt λt 1 − 2γt λt
(2.48) ∆xit = − µt + v.
λt (1 + M (1 − 2γt λt )) λt (1 + M (1 − 2γt λt ))
Comparing coefficients with the linear equilibrium from (2.26) we see that
1 − 2γt λt
(2.49) βt = ,
λt (1 + M (1 − 2γt λt ))
1 − 2γt λt
αt = − µt = −βt µt .
λt (1 + M (1 − 2γt λt ))
the second order condition for a maximum

(2.50) − 2λt (1 − γt λt ) < 0

remains unchanged from Kyle (1985).

From the price setting behavior of the match maker we get in analogy to
(2.36):

v |Ωt ]
pt = E[e

= E[E[e v |Ωt−1 ]|Ωt ]

Cov[e v , M ∆e xit + u et |Ωt−1 ]
= E[ev |Ωt−1 ] + i
×
V ar[M ∆e xt + u et |Ωt−1 ]
×(∆xt + ut − E[M ∆e xit + u et |Ωt−1 ])
M Cov[e v , αt + βt ve + u et |Ωt−1 ]
= pt−1 + 2
×
M V ar[αt + βt ve + u et |Ωt−1 ]
×(∆xt + ut − M E[αt + βt ve + u et |Ωt−1 ])
M βt Cov[e v , ve|Ωt−1 ]
= pt−1 + 2 2 ×
M βt V ar[e v |Ωt−1 ] + σu2i ,t
×(∆e et − M (αt − βt pt−1 ))
xt + u
M βt Σt−1
= pt−1 + 2 2 (∆e et − M (αt + βt pt−1 ))
xt + u
M βt Σt−1 + σu2i ,t
!
M βt Σt−1
= pt−1 1 − M βt 2 2
M βt Σt−1 + σu2i ,t
M βt Σt−1 M βt Σt−1
+ M αt + (δxt + ut ).
M 2 βt2 Σt−1 + σu2i ,t M 2 βt2 Σt−1 + σu2i ,t
44 Chapter 2. Auction Markets

Comparing coefficients with (2.25) we get

M βt Σt−1
(2.51) λt =
M βt2 Σt−1 + σu2i ,t
2
!
M βt Σt−1
µt = pt−1 1 − M βt 2 2
M βt Σt−1 + σu2i ,t
M βt Σt−1
+ M αt
M βt2 Σt−1 + σu2i ,t
2

= pt−1 (1 − M βt λt ) − M αt λt ,

which with (2.49) solves to

M βt Σt−1
(2.52) λt = ,
M 2 βt2 Σt−1 + σu2i ,t
1 − 2γt λt
βt = ,
λt (1 + M (1 − 2γt λt ))
µt = pt−1 ,

αt = −βt pt−1 .

The variance of the liquidation value for the uninformed investors from observing
the prices is given by

(2.53) v |pt ] = V ar[e

Σt = V ar[e v |Ωt ]
Cov[e v , ∆e et |Ωt−1 ]2
xt + u
v |Ωt−1 ] −
= V ar[e
V ar[∆e xt + uet |Ωt−1 ]
2
M Cov[e v , ∆e xt |Ωt−1 ]2
i
= Σt−1 − 2
M V ar[∆e xit |Ωt−1 ] + σu2i ,t
M 2 Cov[e v − pt−1 )|Ωt−1 ]2
v , βt (e
= Σt−1 −
v − pt−1 ) + σu2i ,t |Ωt−1 ]
M 2 V ar[βt (e
M 2 βt2 Σ2t−1
= Σt−1 −
M 2 βt2 Σt−1 + σu2i ,t
= Σt−1 − M λt βt Σt−1

= (1 − M λt βt )Σt−1 .

As in Kyle (1985) the information is gradually incorporated into the price as

2.1. Auctions with informed investors and noise traders 45

the variance is strictly decreasing over time. From (2.46) we see that

(2.54) δt−1 = δt + γt λt σu2i ,t

(2.55) γt−1 (v − pt−1 )2 = (v − µt − λt (∆xit + (M − 1)∆xit ))∆xit

+γt (v − µt − λt (∆xit + (M − 1)∆xit ))2

= (v − µt − M λt ∆xit )∆xit

+γt (v − µt − M λt ∆xit )2

= (v − µt − M λt βt (v − pt−1 ))βt (v − pt−1 )

+γt (v − µt − M λt βt (v − pt−1 ))2

= (v − pt−1 )2 (1 − M λt βt )βt

+γt (v − pt−1 )2 (1 − M λt βt )2

= (v − pt−1 )2 ((1 − M λt βt )βt

+γt (1 − M λt βt )βt )2 ).

Hence we find that

(2.56) γt−1 = (1 − M λt βt )βt + γt (1 − M λt βt )βt )2

1 − 2γt λt
= (1 − M λt )βt
λt (1 + M (1 − 2γt λt ))
1 − 2γt λt
+γt (1 − M λt )2
λt (1 + M (1 − 2γt λt ))
1 1
= βt + γt
1 + M (1 − 2γt λt ) (1 + M (1 − 2γt λt ))2
(1 + M (1 − 2γt λt ))βt + γt
=
(1 + M (1 − 2γt λt ))2
1−2γt λt
(1 + M (1 − 2γt λt )) λt (1+M (1−2γt λt ))
+ γt
= 2
(1 + M (1 − 2γt λt ))
1 − 2γt λt + γt λt
=
λt (1 + M (1 − 2γt λt ))2
1 − γt λt
= .
λt (1 + M (1 − 2γt λt ))2
Holden and Subrahmanyam (1992, pp. 253 ff.) provide a method how to
solve the dynamic system of (2.52), (2.53), (2.54) and (2.56). While a single
46 Chapter 2. Auction Markets

informed investor uses its monopoly power to hold λt constant over time, com-
petition between informed investors forces them to trade more aggressively on
their information in the first auctions. In consequence much information will
be revealed in these first auctions, resulting in a lower liquidity of the market
and a more quickly dropping variance. Information is revealed much faster than
with a single informed investor. As nearly all information has been revealed in
the first auctions later trades are not very informative and λt fast drops near
zero, i.e. the market becomes very liquid in later auctions. As the number of
informed investors increases, information is revealed faster. In the limit as M
reaches infinity, prices are fully revealing in an instant, i.e. with the first trade,
which corresponds to the case of perfect competition.
With perfect competition therefore the market is efficient in the strong form,
otherwise only in the semistrong form. If competition is too strong, profits from
trading on this information are very low and may not cover the costs of acquiring
information, although the prices will never be fully revealing as Σt > 0 in all
auctions due to the noise of uninformed investors. This leads to the problem
pointed out by Grossman and Stiglitz (1980) that markets cannot be fully
revealing if information is costly. If the profits from information acquisition are
too low to cover these costs, no investor will acquire information and prices are
not informative. But this on the other hand gives incentives for investors to
acquire information and make profits (monopolistic case), hence no equilibrium
will exist.
Figures 2.2 - 2.5 illustrate the findings of this model.

2.2 Auctions with strategic uninformed

investors
2.2.1 Markets with a single asset

Thus far we assumed that uninformed investors trade for exogenous reasons and
do not respond to price changes, i.e. they were noise traders. We will now assume
2.2. Auctions with strategic uninformed investors 47

Fig. 2.2: Liquidity with different number of auctions

Fig. 2.3: Informativeness of prices with different number of auctions

48 Chapter 2. Auction Markets

Fig. 2.4: Liquidity with different number of informed investors

Fig. 2.5: Informativeness of prices with different number of informed investors

2.2. Auctions with strategic uninformed investors 49

that they are risk averse investors holding a portfolio consisting of the risky asset
and a riskless asset. We assume that the uninformed investors face a portfolio
imbalance, wj . This imbalance can be due to changed prices, new information
or liquidity needs. They will not only trade this imbalance, but act to maximize
their expected utility, i.e. they are hedgers.
We assume that the portfolio imbalance is a normally distributed random
variable with mean zero and variance σw2 > 0. The imbalance is assumed to be
independent between investors and from any other relevant variable.9

(2.57) wj ∼ N (0, σw2 )

for all j = 1, . . . , N , where N denotes the number of uninformed investors.

There is only a single auction before the asset is liquidated. Again only linear
equilibria are considered, the orders of uninformed investors are also linear in
their portfolio imbalance:

(2.58) uj = η + ξwj j = 1, . . . , N,

(2.59) xi = α + βv i = 1, . . . , M,

(2.60) p = µ + λ(x + u),

PM PN
where x = i=1 xi and u = j=1 uj .
The derivation of the equilibrium follows the same steps as before. The in-
formed investors, assumed to be risk neutral, maximize their expected profits:

(2.61) π v − p)xi = (e
ei = (e v − µ − λ(e
x+u
e)) xi
N
!!
X
= ve − µ − λ xi + (M − 1)xi + N η + ξ w
ej xi .
j=1

(2.62) πi |v] = (v − µ − λxi − λ(M − 1)xi − ληN ) xi .

E [e

Maximizing this expression gives the following first order condition:

(2.63) 0 = v − µ − λxi − λ(M − 1)xi ληN − λxi

= v − µ − 2λxi − λ(M − 1)xi ληN.

9
The model presented here follows Spiegel and Subrahmanyam (1992). Although they
provide a more general framework, we restrict it here to make the results comparable to the
models presented in the previous sections.
50 Chapter 2. Auction Markets

The second order condition −2λ = 0 states again the need for a positive solution
for the liquidity parameter λ. With xi = xi as the only rational conjecture of an
informed investor about the other informed investors trading decisions this solves
to

1 1
(2.64) xi = − (µ + ληN ) + .
(M + 1)λ (M − 1)λ

By comparing with (2.59) we see that

1
(2.65) β = ,
λ(M + 1)
1
α = − (µ + ληN ) = −β(µ + ληN ).
(M + 1)λ

The price setting of the match maker gives us

(2.66) v |x + u]
p = E [e
Cov[e v, x
e+ue]
= E[ev] + (x + u − E[e x+u e])
V ar[e x+u e]
h i
v) + N η + ξ N
P
Cov ve, M (α + βe w
j=1 j
e
= p0 + h PN i ×
V ar M (α + βe v ) + N η + ξ j=1 w ej
N
!
X
× x + u − M (α + βE[e v ]) − N η − ξ E[w ej ]
j=1
M βCov[e v , ve]
= p0 + (x + u − M (α + βp0 ) − N η)
M 2β 2V
v ] + ξ 2 N σw2
ar[e
M βΣ0
= p0 + 2 2 (x + u − M (α + βp0 ) − N η)
M β Σ0 + N ξ 2 σw2
M βΣ0
= p0 − 2 2 M (α + βp0 ) + N η)
M β Σ0 + N ξ 2 σw2
M βΣ0
+ 2 2 (x + u).
M β Σ0 + N ξ 2 σw2

Comparing coefficients with (2.60) results in

M βΣ0
(2.67) λ = ,
M 2 β 2 Σ0 + N ξ 2 σw2
M βΣ0
µ = p0 − 2 2 (M (α + βp0 ) + N η)
M β Σ0 + N ξ 2 σw2
= p0 − λ(M (α + βp0 ) + N η).
2.2. Auctions with strategic uninformed investors 51

By inserting for β we get

M Σ0
λ(M +1)
λ = M 2 Σ0
λ2 (M +1)2
+ ξ 2 N σw2
M Σ0 (M + 1)λ
=
M 2Σ 2 2 2 2
0 + (M + 1) λ ξ N σw

M 2 Σ0 + (M + 1)2 λ2 ξ 2 N σw2 = M Σ0 (M + 1)

M (M + 1)Σ0 − M 2 Σ0
λ2 =
(M + 1)2 ξ 2 N σw2
M Σ0
= .
(M + 1)2 ξ 2 N σw2

The second order condition implies λ to be positive, i.e. we find

s
Σ0 M
(2.68) λ = ,
N σw (M + 1)2 ξ 2
2
s
σw2 N ξ 2
β = .
Σ0 M

With these results we get from (2.65)

(2.69) α = −βp0 ,

(2.70) µ = p0 − N λη.

Uninformed investors maximize their expected utility from holding a portfolio.

For simplicity we do not consider the utility derived from holding the optimal
portfolio, but only the utility from a deviation from this portfolio. The deviation
consists of the size of the portfolio imbalance, wj and the amount traded, uj
adjusted by the price paid. The wealth of a deviation from the optimal portfolio
52 Chapter 2. Auction Markets

is:

(2.71) W ej ) − uj p
fj = ve(uj + w

ej ) − uj (p0 + λ(x + u) − N λη)

= ve(uj + w

= v(uj + wj ) − uj p0 − N λη
!!
X
v − p0 ) + uj + (N − 1)η + ξ
+λ M β(e w
ej )
k6=j

ej ) − uj
= ve(uj + w p0 − N λη + λβM ve − λβp0
!
X
+λuj − ληN − λξ w
ej
k6=j

ej ) − uj
= ve(uj + w p0 (1 − λβM ) − N λη + λβM ve
!
X
+λuj − ληN − λξ w
ej .
k6=j

Therewith we find
fj |wj ] = E[e
(2.72) E[W v |wj ](wj + uj ) − uj p0 (1 − M λβ)
!
X
v |wj ] + λuj − N λη − λξ
+M λβE[e ek |wj ]
E[w
k6=j
= p0 (wj + uj ) − uj (p0 (1 − M λβ) + M λβp0

+λuj − N λη)

= p0 wj + p0 uj − p0 uj − λu2j − N ληuj

= −λu2j + p0 wj − N ληuj ,

(2.73) fj |wj ] = (wj + vj )2 Σ0

V ar[W

+u2j M 2 λ2 β 2 Σ0 + (N − 1)λ2 ξ 2 σw2

−2Cov[(uj + wj )v, M uj vλβ]

= (wj + vj )2 Σ0

+u2j M 2 λ2 β 2 Σ0 + (N − 1)λ2 ξ 2 σw2

−2(uj + wj )λβuj M Σ0 .
2.2. Auctions with strategic uninformed investors 53

The expected utility of the uninformed investors is given by

 
1
(2.74) E [U (Wj )|wj ] = U E[Wj |wj ] − zV ar[Wj |wj ] .
2
Maximizing (2.74) after having inserted (2.72) and (2.73) gives the following first
order condition:

1
(2.75) 0 = −2λuj − N λη − z 2(wj + uj )Σ0
2
+2uj M λ β Σ0 − 2M λβΣ0 + (N − 1)λ2 ξ 2 σw2
2 2 2



−2(wj M λβΣ0 + 2M uj λβΣ0 )

= uj −2λ − zΣ0 M 2 λ2 β 2 − 2M λβ + 1 + (N − 1)λ2 ξ 2 σw2

−zwj Σ0 (1 − M λβ) − N λη

= uj −2λ − zΣ0 (1 − M λβ)2 − z(N − 1)λ2 ξ 2 σw2

−zwj Σ0 (1 − M λβ) − N λη.

The second order condition for a maximum −2λ − zΣ0 (1 − M λβ)2 − z(N −
1)λ2 ξ 2 σw2 < 0 is fulfilled as λ > 0 and all other terms are positive. Solving for
the optimal trade size of uninformed investors gives
N λη
(2.76) uj = −
2λ + zΣ0 (1 − M λβ)2 + z(N − 1)λ2 ξ 2 σw2
zΣ0 (1 − M λβ)
− wj .
2λ + zΣ0 (1 − M λβ)2 + z(N − 1)λ2 ξ 2 σw2
By comparing coefficients with (2.58) we see that
zΣ0 (1 − M λβ)
(2.77) ξ = − ,
2λ + zΣ0 (1 − M λβ)2 + z(N − 1)λ2 ξ 2 σw2
N λη
η = − .
2λ + zΣ0 (1 − M λβ)2 + z(N − 1)λ2 ξ 2 σw2
As from the second order condition λ > 0 and the denominator is positive the
last equation implies that

(2.78) η = 0.

The first equation has to be solved for ξ:

(2.79) z(N − 1)λ2 ξ 3 σw2 + (zΣ0 (1 − M λβ)2 + 2λ)ξ + zΣ0 (1 − M λβ) = 0

As all coefficients are positive10 the solution for ξ has to be negative. Inserting
10 M 1
From (2.65) we know that 1 − M λβ = 1 − M +1 = M +1 > 0.
54 Chapter 2. Auction Markets

for β and λ we get

 2 !
M
(2.80) 0 = z(N − 1)λ2 σw2 ξ 3 + zΣ0 1− + 2λ ξ
M +1
 
M
+zΣ0 1 −
M +1
Σ0 M (N − 1)σw2 3
= z ξ
N σw2 (M + 1)2 ξ 2
s !
zΣ0 Σ0 M zΣ0
+ 2
+2 2 2 2
ξ+
(M + 1) N σw (M + 1) ξ M +1
q
M (N − 1) + N zΣ0 − 2 Σσ20 M
wN
= zΣ0 2
ξ+ .
N (M + 1) M +1


p √ 
zΣ0 σw2 N − 2 Σ0 M N (M + 1)
(2.81) ξ=− p .
σw2 N zΣ0 (M (N − 1) + N )

As ξ has to be negative, a linear equilibrium only exists if the numerator is

positive as the denominator always is positive, i.e. if
p p
zΣ0 σw2 N − 2 Σ0 M > 0,
1 2
(2.82) M < N z Σ0 σw2 .
4

The existence depends on a not too large fraction of informed investors in the
market. The higher the risk aversion, uncertainty about the liquidation value, Σ0 ,
and dispersion of portfolio imbalances the more informed investors can participate
in the market. This can be explained by the behavior of the uninformed investors,
as in these cases their need to trade is higher and they are willing to make larger
losses to rebalance their portfolios.
The equilibrium can be written as

(2.83) p = p0 + λ(x + u),

(2.84) xi = β(v − p0 ),

(2.85) uj = ξwj ,

where λ, β and ξ are given by (2.68) and (2.81).

2.2. Auctions with strategic uninformed investors 55

The variance of the liquidation value after observing the price is given by

v , pe]2
Cov[e
(2.86) v |p] = V ar[e
Σ1 = V ar[e v] −
V ar[ep]
2
Cov[e
v , p0 + λ(e
x+u e)]
= Σ0 −
V ar[p0 + λ(ex+u e)]
λ2 Cov[e v − p0 ) + ξ N ej ]2
P
v , M β(e j=1 w
= Σ0 −
v − p0 ) + ξ N
P
λ2 V ar[M β(e j=1 wej ]
M 2 β 2 Σ20
= Σ0 −
M 2 β 2 Σ20 + ξ 2 N σw2
ξ 2 N σw2 Σ0
=
M 2 β 2 Σ20 + ξ 2 N σw2
ξ 2 N σw2 Σ0
= M βΣ0
λ
λξ 2 N σw2
=
Mβ
λ σΣ2 0N (M +1)
M 2
2 λ2 N σw
w
=
Mβ
Σ0
=
(M + 1)2 λβ
Σ0
= .
M +1

The informativeness of the price does only depend on the number of informed
investors. Competition between them to trade on their information results in a
high revelation of information.

λ is decreasing in the risk aversion of uninformed investors. If they are more

risk averse their wish to trade is larger, although they make losses from trading.
This increased trading of uninformed investors enables informed investors to hide
their trades better, the order flow becomes less informative. The same holds if the
portfolio imbalance increases, i.e. σw2 increases. The need to offset the imbalance
at least partially is increased and trades of uniformed investors increase.

If the number uninformed investors increases the volatility of the price in-
56 Chapter 2. Auction Markets

creases:

(2.87) V ar[e
p] = V ar[p0 + λ(e
x+u
e)]
" N
#
X
= λ2 V ar M β(e
v − p0 ) + ξ w
ej
j=1
2 2 2 2 2
= M λ β Σ0 + λ ξ N 2 σw2 .

As the price of the trade is not known at the time the order is submitted, the
risk from trading is increased for uninformed investors. If they are not too risk
averse this effect will be more than compensated by the additional orders from
uninformed investors, i.e. adverse selection costs are smaller and λ will be de-
creasing in the risk aversion. If the uninformed investors are more risk averse,
however, the reduced order from every single uninformed investor dominates the
additional order flow generated by more uninformed investors in the market and
λ increases with more uninformed investors. But if the number is sufficiently
enlarged the additional order flow will dominate again and λ decreases. Figure
2.6 illustrates this finding.
By increasing the number of informed investors, competition between them is
increased as has been shown in the previous section, resulting in an increased λ.
On the other hand as we saw in (2.86) the informativeness of prices is increased,
i.e. the adverse selection costs of the uniformed investors are reduced. If the
uninformed investors are very risk averse this will induce them to trade more
actively, thereby compensating the increased λ, which becomes decreasing. If
they are less risk averse their increased trading cannot compensate the effect of
competition and λ is increasing as illustrated in figure 2.7.
If the uncertainty about the liquidation value, Σ0 , increases, the adverse selec-
tion costs and the variance of the price will increase, hence uninformed investors
will scale back their trades, this increases λ. On the other hand the benefits
from offsetting an imbalance is increased. If the uninformed investors are very
risk averse the first effect may dominate and λ will be increasing in Σ0 . If the
uninformed investors are less risk averse the second effect will dominate for small
uncertainties and λ decreases, if the uncertainty is too large, however, the first
2.2. Auctions with strategic uninformed investors 57

Fig. 2.6: Market liquidity with varying number of uninformed investors

Fig. 2.7: Market liquidity with varying number of informed investors

58 Chapter 2. Auction Markets

Fig. 2.8: Market liquidity with different uncertainties about the liquidation
value

effect again will dominate and λ increses again. This behavior is illustrated in
figure 2.8.11
Spiegel and Subrahmanyam (1992) provide a more general framework
by allowing the signals the informed investors receive to be imperfect and differ
between investors. Compared to the more restrictive version presented here no
additional insights can be gained, whereas the tractability of the derivation is
reduced significantly. Introducing risk averse informed investors does also not
change the results derived here, but becomes in a general framework unhand-
able.12
A central element for calculating conditional moments and hence deriving the
equilibrium is the assumption of normally distributed random variables. This
assumption cannot be lifted without facing the problem that the results may
11
These results are formally shown in Spiegel and Subrahmanyam (1992).
12
Subrahmanyam (1991) presents a model with risk averse informed investors. He uses only
noise traders, but no hedgers in his model. The results he derives are very similar to those with
risk neutral informed investors, so that no new insights can be expected.
2.2. Auctions with strategic uninformed investors 59

change. Foster and Viswanathan (1993) derive similar results within the
more general class of elliptically contoured distributions instead of using the spe-
cial case of a normal distribution, a generalization to other distributions cannot
be made.

All models presented in this part considered only linear equilibria, nothing
can be said about the existence and properties of non-linear equilibria. Kyle
(1985, p. 1322) suspects that there exist no non-linear equilibrium, but he is not
able to proof his suspicion.

There exist several extensions of these models, which we will not consider in
detail here. Bondarenko (1999) relaxes the assumption of only a single match
maker by investigating a model with a given number of match makers that are
not restricted to quote prices such that they make zero expected profits, e.g. by
applying strategic price setting to maximize their profits. He finds the results as
derived above to hold with the number of match makers reaching infinity, i.e. if
they behave competitively.

By adding another source of information on the value of the asset for the
market maker, Jain and Mirman (1999) show that the match maker sets more
informative prices and reduces the profits of the informed investor. Chau (1998)
considers the trading of many pure noise traders, a single risk averse informed
investor and a risk averse market maker. Although this model assumes a dealer
rather than an auction market, it helps understanding the adjustment of prices
to new information. In contrast to models with risk neutral market participants,
he finds that risk aversion cause prices to overshoot before gradually adjusting
to the new fundamental value.

2.2.2 Auction markets with multiple assets

Using the framework developed by Holden and Subrahmanyam (1992) for

a single trade, Caballe and Krishnan (1994) extend the model to the more
realistic case of investors being able to trade multiple assets.
60 Chapter 2. Auction Markets

Each investor receives a noisy signal of the liquidation values ve, such that13

(2.88) sei = ve + εei ,

εei ∼ N (0, Σε ) ,

s1i , . . . , seLi ) denotes the vector of signals investor i receives for each of
where sei = (e
v 1 , . . . , veL ) the vector of the liquidation values, εei = (e
the L assets, ve = (e ε1i , . . . , εeLi )
the vector of noise terms as received by investor i and Σε the covariance matrix
of noise terms. The liquidation value is multivariate normally distributed with
mean p0 = (p10 , . . . , pL0 ) and covariance matrix Σ0 :

(2.89) ve ∼ N (p0 , Σ0 ).

The imbalance of orders from uninformed investors, u u1 , . . . , u

e = (e eL ) is normally
distributed with mean zero and covariance matrix Σu :

(2.90) e ∼ N (0, Σu ).
u

p1 , . . . , peL ) denoting the vector

The match maker sets prices such that with pe = (e
of prices and the aggregate order flow from informed investors x x1 , . . . , x
e = (e eL ),
ej = M eji
P
x i=1 x

(2.91) v |e
pe = E[e x+u
e].

The profits from trading for informed investors are the sum of the profits made
from trading each asset:

(2.92) π v − pe)0 x
ei = (e ei .

As before the analysis is restricted to a linear equilibrium:

(2.93) pe = µ + Λ(e
x+u
e),

(2.94) x si − p0 ),
ei = α + B(e
13
With the exception of this assumption, where Kyle (1985) assumes perfect knowledge of
the liquidation value, the assumptions are similar to those in Kyle (1985). It has to be noticed
that all variables represent vectors or matrices of the considered assets.
2.2. Auctions with strategic uninformed investors 61

where µ = (µ1 , . . . , µL )0 and α = (α1 , . . . , αL )0 are vectors of constants and

Λ = (λ11 , . . . , λ1L ; . . . ; λL1 , . . . , λLL ) and B = (β 11 , . . . , β 1L ; . . . ; β L1 , . . . , β LL ) are
positive definite and nonsingular matrices of constants.
Informed investors maximize expected utility from trading, if we assume them
to be risk neutral this is equivalent to maximizing expected profits. These are
given by

(2.95) E[e v − pe)0 x

πi |si ] = E [(e ei |si ]

v − µ − Λ(e
= E [(e x+ue))0 xi |si ]
" M
!!0 #
X
= E ve − µ − Λ xi + xj + u
e xi |si
j=1,j6=i
"
= E ve − µ

M
! !0 #
X
−Λ xi + sj − p 0 ) + u
(α + B(e e xi |si .
j=1,j6=i

Differentiating gives us the first order condition

(2.96) v |si ] − µ − Λxi

0 = E[e
XM
−Λ sj |si ] − p0 ) − Λxi
(α + B(E[e
j=1,j6=i
v |si ] − µ − 2Λxi
= E[e
XM
−ΛB sj |si ] − p0 ) − (M − 1)Λα.
(E[e
j=1,j6=i

The second order condition −2Λ < 0 is fulfilled by the assumption of Λ being
positive definite. We further get

(2.97) v |si ] = E[e

E[e si − εei |si ] = si ,

sj |si ] − p0 = E[e
(2.98) E[e sj − p0 |si − p0 ]

sj − p0 ] + Cov[e
= E[e sj − p0 )−1
sj − p0 , sei − p0 ]V ar(e

×(si − p0 − E[e
si − p0 ])

= Cov[e sj ]−1 (si − p0 )

sj , sei ]V ar[e

= Σs Σ−1
ε (si − p0 ),
62 Chapter 2. Auction Markets

where Σs denotes the covariance matrix of the signals received. Inserting (2.97)
and (2.98) into (2.96) we get

(2.99) 0 = si − µ − (M − 1)Λα
XM
−ΛB (Σs Σ−1
ε (si − p0 )) − 2Λxi
j=1,j6=i
= si − µ − (M − 1)Λα

−(M − 1)ΛBΣs Σ−1

ε (si − p0 ) − 2Λxi

= (si − p0 ) − (M − 1)Λα

−(M − 1)ΛBΣs Σ−1

ε (si − p0 ) + p0 − µ − 2Λxi .

Solving for 2Λxi and inserting for xi from (2.94) we get

I − (M − 1)ΛBΣs Σ−1


(2.100) ε (si − p0 ) + p0 − µ − (M − 1)Λα

= 2Λα + 2ΛB(si − p0 ),

where I denotes the identity matrix. By comparing coefficients we see that

(2.101) p0 − µ − (M + 1)Λα = 0,

(2.102) I − (M − 1)ΛBΣs Σ−1

ε = 2ΛB.

From (2.91) we get

(2.103) v |x + u]
pe = E[e

= E[e
v ] + Cov[e
v, x
e+u
e]V ar[e e]−1 (e
x+u e − E[e
x+u x+u
e]),

where

(2.104) E[e
v ] = p0 ,

(2.105) E[e
x+u
e] = E[e
x] + E[e
u]
"M # "M #
X X
= E x
ei = E si − p0 ))
(α + B(e
i=1 i=1
M
X
= Mα + B si − p0 ] = M α,
E[e
i=1
2.2. Auctions with strategic uninformed investors 63

(2.106) V ar[e
x+u
e] = V ar[e
x] + V ar[e
u]
"M #
X
= V ar x
ei + Σu
i=1
" M
#
X
= V ar si − p0 )) + Σu
(α + B(e
i=1
si − p0 ]B 0
= M BV ar[e

si − p0 , sej − p0 ]B 0 + Σu
+M (M − 1)BCov[e

= M BΣε B 0 + M (M − 1)BΣs B 0 + Σu ,

(2.107) Cov[e
v, x
e+u
e] = Cov[e
v, x
e] + Cov[e v, u
e]
" M
#
X
= Cov sei − εei , B sj − p 0 )
(e
j=1
0
= M Σε B .

Inserting these relations into (2.103) we get

−1
(2.108) pe = p0 + M Σε B 0 [M BΣε B 0 + M (M − 1)BΣs B 0 + Σu ] (e e − M α).
x+u

Comparing coefficients with (2.93) gives

−1
(2.109) Λ = M Σε B 0 [M BΣε B 0 + M (M − 1)BΣs B 0 + Σu ] ,

(2.110) µ = p0 − M Λα.

we can now solve (2.101), (2.102), (2.109) and (2.110) for the unknown parameters
µ, Λ, α and B. This calculation is conducted in Caballe and Krishnan (1994,
pp. 701 f.). It turns out that

√
M
(2.111) p = p0 + Γx,
2
(2.112) x = Θ(si − p0 ),
64 Chapter 2. Auction Markets

where

−1 1 −1
(2.113) Γ = Σ u 2 H 2 Σu 2 ,
1 1
(2.114) H = Σu2 GΣu2
 −1
−1 M − 1 −1 −1
(2.115) G = Σε + Σε Σs Σε
2
 −1
2 −1 −1 M − 1 −1 −1
− Σ + 2Σε + Σε Σ s Σε ,
M −1 s 2
1
(2.116) Θ = √ −1 .
M Γ−1 I + M2−1 Σs Σ−1

ε

Although this result is very difficult to interpret due to its complexity, some
general properties can be derived. It can be shown that Γ is positive definite
and symmetric. This can be used to state that the order flow of the pth asset
affects the price of the qth asset in the same degree as the order flow of the qth
asset influences the price of the pth asset, i.e. the informativeness of trades in
the other asset is equal for all assets.14
Further insights should be gained by analyzing the correlation structure in
more detail, i.e. how different covariances in error terms or liquidity trade im-
balances influence the covariance of the observed prices. The complexity of the
result makes such an analysis very difficult to conduct in general. It should be
noted that as

−1 −1
v |p] = Σ0 − M 2Σ−1 −1
 
(2.117) V ar[e ε + (M − 1)Σε Σs Σε ,

the beliefs after observing prices can exhibit a completely different correlation
structure than the initial beliefs.

2.3 The informational content of trading

volume
Thus far prices have been used by uninformed investors to revise their beliefs of
the fundamental value of the asset. Another variable is widely used in markets
14
Caballe and Krishnan (1994, p.700) report that this result can be confirmed in empirical
investigations.
2.3. The informational content of trading volume 65

as a source of information: trading volume. Using this additional information

enables uninformed investors to learn the information more quickly. Blume
et al. (1994) provide a simple model of trading volume.
They assume the fundamental value to be normally distributed with mean p0
and variance σ 2 :

(2.118) pe ∼ N (p0 , σ 2 ).

A share of γ of the total N investors are assumed to be informed, but unlike in

the previous models their information is not perfect. They only observe a noisy
signal of the fundamental value:15

(2.119) ψe1i = pe∗ + εei1 ,

where

(2.120) pe∗ ∼ N (p, σp2 )

is a common signal. There remains uncertainty about the transformation of the

information into the fundamental value. This information is not observed purely,
i.e. there is some noise with

(2.121) εei1 ∼ N (0, σ

eε21 ).

eε21 is itself a random variable, i.e. the quality of information

The error variance σ
eε21
varies randomly over time. The informed investors know the realization of σ
when they receive their information, uninformed investors only know its distri-
bution. Uninformed investors are assumed to receive also the signal pe∗ but with
another error term εe2 :

(2.122) ψe2i = pe∗ + εei2 ,

εei1 ∼ N (0, σ
eε22 ),
15
This modification is necessary as otherwise uninformed investors would be able to deduct
the information completely by only observing price and volume. In this case informed investors
would not be able to make profits from their information and if information is costly there
would be no incentives to become informed. If no one is informed it would be profitable to be
informed and no equilibrium exists. See Grossman and Stiglitz (1980) for more details on
this problem.
66 Chapter 2. Auction Markets

eε22 is known to all investors, informed and uninformed. With the assump-
where σ
eε22 > σ
tion that σ eε21 the information of informed investors is more precise, that is
why they are called informed. We therewith have for the distribution of the true
value ψei of group i, if we assume the errors terms to be independent of all other
relevant variables:

(2.123) ψei ∼ N (p, σp2 + σ

eε2i ).

The demand of the investors for the asset has been derived by Diamond and
Verrecchia (1981) and Brown and Jennings (1989). When we restrict the
equilibrium to be linear we receive with risk averse investors j the demand for
every individual in groups 1 (informed investors) and 2 (uninformed investors) as

p0 − p1 ψ1j − p1
(2.124) dj1 = + ,
zσ 2 z(σp2 + σε21 )
p0 − p1 ψ2j − p1
dj2 = + ,
zσ 2 z(σp2 + σε22 )

where z denotes the Arrow-Pratt measure of risk aversion and p1 the price applied.
The total demand of all investors has to equal zero as long as the amount of the
asset is fixed to achieve an equilibrium. Aggregation over all individuals gives us
after dividing by N and multiplying by z:

γN
!
1 X p0 − p1 ψ1j − p1
(2.125) 0 = + 2
N i=1
σ2 σp + σε21
N
!!
X p0 − p1 ψ2j − p1
+ + 2
i=γN +1
σ2 σp + σε22
γN
p0 − p1 1 1 X
= 2
+ 2 2
(ψ1i − p1 )
σ N σp + σε1 1=1
N
!
1 X
+ (ψ2i − p1 )
σp2 + σε22 i=γN +1

p0 − p1 γ(ψ 1 − p1 ) (1 − γ)(ψ 2 − p1 )
= + 2 + ,
σ2 σp + σε21 σp2 + σε22

1
PγN 1
PN
where ψ 1 = N i=1 ψ1i and ψ 2 = N i=γN +1 ψ2i denote the average realization of
2.3. The informational content of trading volume 67

the information. Solving for p1 gives

p0 γ 1−γ
σ2
+ ψ
σp2 +σε21 1
+ ψ
σp2 +σε22 2
(2.126) p1 = 1 γ 1−γ .
σ2
+ σp2 +σε21
+ σp2 +σε22

If we let N → ∞ then the law of large numbers can be used to show that ψ j → p
and (2.126) becomes
 
p0 γ 1−γ
σ2
+ σp2 +σε21
+ σp2 +σε22
p
(2.127) p1 = 1 γ 1−γ .
σ2
+ σp2 +σε21
+ σp2 +σε22

The total trading volume is the absolute value of the demands given in (2.124). To
avoid double counting this has to be divided by two. Hence the trading volume,
adjusted to per capita average volume is with N → ∞:
γN N
!
11 X i X
i
(2.128) V = |d | + |d2 |
2 N i=1 1 i=γN +1
γN N
γ 1 X i 1−γ 1 X
= |d1 | + |di2 |
2 γN i=1 2 (1 − γ)N i=γN +1
γ 1−γ
→ E[|d1 |] + E[|d2 |].
2 2

Inserting from (2.124) and (2.127) gives the expression for the volume, which is
explicitly stated in Blume et al. (1994, p. 165), but not reproduced here. The
evaluation is best conducted with numerical examples.16
If the information received by informed investors is only of low precision and
suggests only a small deviation from prior beliefs, i.e. σε21 is relative large com-
pared to σε22 , informed investors have not much confidence in their information.
Although their beliefs are widely spread they do not trade much on their infor-
mation, hence trading volume will be relatively small. If the precision increases,
i.e. σε21 decreases, they become more confident in their information and if their
beliefs are wide enough dispersed, trading volume increases. For σε21 = σp2 the
trading volume reaches its maximum. If the precision of information is further
increased, their confidence also increases, but on the other hand the dispersion of
16
The results of the illustrations below can be shown to be valid in all economically relevant
situations.
68 Chapter 2. Auction Markets

beliefs is reduced, they do not find many informed investors to trade with, they
are forced to trade nearly only with uninformed investors, hence trading volume
reduces.

If the information received suggests a large deviation from the prior belief of
p0 , the rebalancing of the portfolios will dominate, even if the precision is not
large. The trade volume will increase with the precision of the information, even
if the dispersion of beliefs is reduced.

The larger the signal suggests the deviation from prior beliefs is the more need
the investors to rebalance their portfolios, this need increases with the precision
of information. As p1 denotes also the new belief of the informed investors, p1 −p0
denotes the change in belief. Therewith the volume should be V-shaped in the
change of beliefs, where the V becomes more pronounced the more precise the
information is. The lowest trading volume should be found at p0 = p1 .

Figure 2.9 visualizes these findings using the expression for trading volume
provided by Blume et al. (1994). Virtually all empirical investigations on
price changes and volume find V-shaped pattern between the price change and
volume. With this simple model this pattern can be explained.

Figure 2.10 shows furthermore that the V-shape does not only depend on the
precision of information, but also on the dissipation of information, i.e. the share
of informed investors γ. The more investors are informed the less pronounced
the V-shape is until it vanishes if all investors are informed. This pattern can be
explained with the fact that with a price change due to new information only few
investors can expect to make a profits as their beliefs deviate from the current
price. Hence the trading volume does not respond so sensitive to price changes.
With only minor price changes, or equivalently changes in beliefs, the volume is
increasing for σε21 > σp2 and decreasing for σε21 < σp2 due to the effect of rebalancing
portfolios as described above.

While prices reveal information about the magnitude of a signal, i.e. the
change in beliefs of the informed investors, trading volume reveals information
about the precision and dissemination of information. By observing prices and
2.3. The informational content of trading volume 69

Fig. 2.9: Trading volume with different precision of information and changes of
beliefs

Fig. 2.10: Trading volume with different shares of informed investors and
changes of beliefs
70 Chapter 2. Auction Markets

trading volume an investor can find out the beliefs of the informed investors
and their changes by observing prices and the persistence of the movement, i.e.
the confidence of informed investors by observing trading volume. This can be
observed in markets where large price changes in association with low trading vol-
ume in most cases are viewed as being not very persistent, whereas the same price
change with a large trading volume is viewed as persistent by market observers.

2.4 Explaining short-term movements of asset

prices
In recent years many empirical studies have been conducted to detect patterns
in asset markets. Many such patterns have been found between trading days as
well as within a trading day. Various explanations and models have been offered
to explain the found behavior. In this subsection we will not cover this literature
in detail, but concentrate on two widely cited contributions, whose results can
be explained intuitively with the models presented above, such that there is no
need to develop the models that are actually chosen by the authors in detail.
Foster and Viswanathan (1990) use the above derived framework to ex-
plain the variation of asset price volatility and trading volume within a trading
week. Trading takes place typically from Monday to Friday, while they assume
that information is acquired at every day of the week, i.e. from Monday to Sun-
day, but portfolio imbalances are not aggregated over the weekend. Therewith
informed investors should have a larger informational advantage from trading
at Mondays, on the other hand adverse selection costs are highest on Mondays.
Hence uninformed investors will not trade so actively on this day. If we assume
further that a part of the information becomes public by other sources than
prices, e.g. publications in newspapers, rumors, over the following trading days,
informed investors are forced trade on their information also on Mondays in or-
der to make profits, hence the fraction of informed trades will be relative high
on these days and trading volume should be low due to the reduced participa-
tion of uninformed investors. A high fraction of informed trades implies a large
2.4. Explaining short-term movements of asset prices 71

change in the price, hence we should expect to find a larger volatility on Mondays.
Empirical investigations support this effect as suggested by this model.
Due to adverse selection costs uninformed investors always make losses when
trading with informed investors. The more uninformed investors are in the mar-
ket, the more these costs are distributed among uninformed investors, reducing
costs of trading for a single uninformed investor. It would therefore be a bene-
ficial strategy for uninformed investors to concentrate their trading in a certain
trading round. Also informed investors would be better able to hide their trades
and would be able to trade more and hence make more profits in these trading
rounds. It is reasonable to assume that for most investors it is of no importance
at which time of the day they trade as the settlement of the orders depends only
on the day of trading and not on the time of trading within a day. Admati and
Pfleiderer (1988) show that with such assumptions trading will be concen-
trated at certain points of time in the trading day, provided competition between
informed investors is large enough and adverse selection costs do not increase too
much by the presence of more informed investors. Again the presence of many
informed and uninformed investors increases the volatility of prices in those times
of higher trading activity, a result that can be confirmed in empirical studies.
The two models shortly presented above can explain some of the effects ob-
served in asset market regarding volatility and trading volume. However, they
cannot explain observed patterns in returns. Admati and Pfleiderer (1989)
provide a model how to explain such patterns, but they have to introduce another
rule how the match maker determines prices. They assume the match maker to
set prices strategically, what leaves the framework of the models presented above.
72 Chapter 2. Auction Markets

Review questions
1. In the single period model of Kyle (1985), why is not all information
included into the price after trading?

2. Why is liquidity increasing over time in the multi-period version of the

model of Kyle (1985)?

3. How does the informational efficiency of the market evolve over time in
Kyle (1985)?

4. Why do a larger number of traders increase the informational efficiency of

prices in Kyle (1985)?

5. A larger number of auctions increases the efficiency of the market. Why?

6. How can you explain the increasing liquidity of the market for more risk
averse uninformed investors?

7. Why is the liquidity not monotone decreasing in the number of risk averse
uninformed traders?

8. Explain the V-shaped trading volume in the change of belief.

9. What role does the quality of information play for trading volume?

10. Why do investors cluster their trading?

Application
In the morning of Wednesday, 21 November 2004 a rumor spread in the mar-
ket about an imminent approach for a takeover of Nanobot Inc. by its larger
rival TechAppliance Inc. at a significant premium. TechAppliance Inc. made it
publicly known on Thursday, 22 November 2004 at 12.00 noon that it was not
planning a takeover bid in Nanobot Inc. As usual in such cases the market reg-
ulator took a closer look at the trading activity during these days, in particular
as the stock had seen significant price movements.
For a preliminary analysis the regulator has produced a chart of stock prices
(represented by the line) and trading volumes (represented by the bars) in five
minute intervals. This chart is reproduced below. Given the theories presented in
this chapter, what can be inferred from this on the trading activity, particularly
during Tuesday and Thursday morning?
 Stock price

0.2
0.6
1.4
1.8
2.2

1
08:00
Application

10:00
12:00
14:00

Monday
16:00
09:00
11:00
13:00

Tuesday
15:00
08:00
10:00
12:00
14:00

Wednesday
16:00
09:00
11:00
13:00
Thursday 15:00
08:00
10:00
12:00
Friday

14:00
16:00
0
10000
20000
30000
40000
50000
60000

Trading volume (shares)

73
74 Chapter 2. Auction Markets
Chapter 3

Dealer Markets

———————————————————————————————————
This chapter will introduce the main theories on dealer markets. The focus in
this chapter will be on the determination of the spread in such markets, and the
main contents will cover
• costs of market making without asymmetric information
• competitive and monopolistic spreads
• adverse selection costs in markets with asymmetric information
• the interaction of multiple assets and their effects on the spread

Key readings:
Hans R. Stoll: The Supply of Dealer Services in Securities Markets, Journal of
Finance, 33, 1133-1151, 1978

Thomas Ho and Hans R. Stoll: Optimal Dealer Pricing under Transactions and
Return Uncertainty, Journal of Financial Economics, 9, 47-73, 1981

Lawrence R. Glosten and Paul R. Milgrom: Bid, Ask and Transaction Prices in
a Specialist Market with Heterogenously Informed Traders, Journal of Financial
Economics, 14, 71-100, 1985

Thomas Gehrig and Matthew Jackson: Bid-Ask Spreads with Indirect Competi-
tion Among Specialists, Journal of Financial Markets, 1, 89-119, 1998
———————————————————————————————————
76 Chapter 3. Dealer Markets

The last chapter analyzed how prices adjust to new information, therefore the
order flow has been aggregated over a given period of time. The match maker
had no other role than to determine the equilibrium price, he played no active
role in the trading process. In the following sections we will now investigate the
behavior of prices on a trade-by-trade basis. We therefore introduce a market
maker into the trading process, replacing the match maker. The market maker
directly influences the prices by quoting prices at which he is willing to buy and
sell the asset. But in general he also has no active role in the trading process as
he does not initiate or actively search for a trade, but waits for an order to arrive
and then clears this order at the stated price on his own account. It is intuitively
clear that transaction prices are not only influenced by the orders, but also by
the behavior of the market maker.
Two main groups of theories modeling the behavior of market makers have
evolved, inventory and information-based models. Information-based models are
close to the models of auction markets, they assume a risk-neutral market maker
and two groups of investors, informed and uninformed investors. Inventory-based
models assume a risk averse market maker and all investors have the same in-
formation and agree on the implications of this information on the fundamental
value, i.e. they agree on the fundamental value. While in information-based mod-
els trades can be motivated by exploiting informational advantages (informed
investors) and need for liquidity (uninformed investors, or noise traders), in
inventory-based models need for liquidity is the only source of trade. The coming
section analyzes inventory-based models and the following section information-
based models.

3.1 Inventory-Based Models of Market Making

A market maker provides the service of enabling an investor to trade immediately
at a given price by acting as counterpart of the order. He then waits for another
order offsetting his position. Therewith he takes the risk of not knowing when
and at which price he can offset this position. A trade that is typically conducted
3.1. Inventory-Based Models of Market Making 77

between two investors is divided into two parts that occur at different prices at
different times. Not knowing when and at which price an offsetting order arrives
imposes costs on the market maker that he has to cover by quoting different
prices at which he willing to buy (bid price) and to sell (ask price) the asset.
As the investor has not to bear these risks, he is willing to pay for this service
by accepting a less favorable price for the trade until the costs imposed by the
market maker equals his costs from waiting for an offsetting order and trading at
an unknown price (waiting costs).1
We now have to make some assumptions on the the determination of the order
flow.2 At first we assume that all investors place their orders independently of
each other. Each investor either submits a buy or a sell order, but not both.
Therewith the order submissions to buy and to sell are independent of each other.
The need for trading is exogenously given by a liquidity event, which determines
the waiting costs of an investor. This liquidity event is assumed to be a random
variable (hence waiting costs are a random variable) that is independent and
identical distributed between investors. If the waiting costs are higher than the
costs imposed by the market maker,3 he will submit his order to the market maker,
otherwise he will wait for an offsetting order by himself. The higher the costs
for trading with the market maker, the less orders are submitted. We can now
aggregate all orders (separately for buy and sell orders) and see that the orders
submitted in a given period of time follow a Poisson process or, equivalently, that
the probability of an order arriving is Poisson distributed.4
A Poisson distribution is characterized by the order arrival rate λ. Let λa
denote the order arrival rate for buy orders (for trades at the ask) and λb for sell
orders (for a trade at the bid). Let further denote p∗ the fundamental value all
investors agree on, pa the ask and pb the bid price. The costs the market maker
1
Demsetz (1968, p. 37) was the first to introduce the concept of waiting costs into literature.
His pathbreaking article lead the way to the following literature on market microstructure
theory.
2
These assumptions have first explicitly been stated by Garman (1976, pp. 258 ff.).
3
The costs are the difference between the value of the asset and the quoted price.
4
To see this it has to be noted that submitting an order or not is a binomial variable, by
aggregating binomial variables they converge to a Poisson process.
78 Chapter 3. Dealer Markets

p
6
Q 

Q 
Q 
pa Q
Q 

Q 
Q 
Q 
Q 
Q  Sell orders
Q 
Q 
Q 
Q 
Q 
Q 
0
p Q
Q
 Q
 Q
 Q
 Q
 Q
pb  Q
Q Buy orders

 Q
 Q
 Q
 Q
 Q
 Q
 Q
 Q
 QQ
-
λa λb λ0 λ

Fig. 3.1: Demand and Supply in Dealer Markets

imposes on the investors are pa − p∗ and p∗ − pb .5 Therewith we find that

∂λb ∂λa
(3.1) > 0, < 0,
∂pb ∂pa
∂λb ∂λa
= 0, = 0.
∂pa ∂pb
Figure 3.1 visualizes these findings. At (p0 ,λ0 ) we have a stochastic equilibrium
of individual order arrival rates.6 However by choosing pa = pb = p0 the market
maker would not be able to cover his costs. He has to choose two prices on the
curves such that pa > pb , e.g. those marked in figure 3.1. As in this case λb > λa ,
it is more likely that a sell order arrives next in the market, i.e. the market maker
expects to increase his position in the asset by offsetting the order. Similarly he
will choose λb < λa to decrease his position in the asset and λa = λb if he does
not want his position to change.
Before determining the costs of market making it is necessary to characterize
5
It has already been pointed out that pb ≤ pa .
6
The equilibrium is stochastic as only expected demand and supply equal by having the
same order arrival rates. The realized demand and supply may not equal.
3.1. Inventory-Based Models of Market Making 79

the trading process and the behavior of the market maker in more detail. We
assume that only a single trade per period of time is submitted to the market
maker, either a buy or a sell order.7 When quoting prices at which he is willing
to buy and sell the asset he does not know whether the next order will be a buy
or a sell order.
We consider an economy with only a single risky asset that is traded with
the market maker and a riskless asset bearing no interest, e.g. money. Assume
further that the market maker has an optimal portfolio consisting of the single
risky asset and the riskless asset, chosen according to portfolio selection theory.
Any deviations from this optimal portfolio are denoted as inventory.8 After a
single trading round we assume the risky asset to be liquidated at the fundamental
value.
The market maker is assumed to be risk averse and maximize his expected
utility of terminal wealth that occurs after a single round of trading by setting
optimal bid and ask prices.

3.1.1 The costs of market making

The first to provide a model how to determine the costs of a market maker
was Stoll (1978). Suppose that the market maker holds his optimal portfolio,
denoted E in figure 3.2. By accepting a trade the portfolio actually held deviates
from his optimal portfolio, suppose it is located at E 0 , i.e. a buy order arrives
at the market and the share of the asset in the portfolio is reduced, while the
share of the riskless asset is increased. The utility level from holding portfolio
E 0 instead of the optimal portfolio E decreases from U0 to U1 . This difference in
utility are the costs that the market maker faces from his service. He has to be
compensated for this loss in utility, which is done by holding a larger portfolio,
7
We could also allow for many orders to be submitted and concentrate on the order imbalance
receiving the same results. In this sense we can interpret this situation as an auction market
presented above, where the order imbalance has not been served and now is served by the
market maker.
8
As the optimal portfolio is fixed we can concentrate our analysis on the inventory that is a
linear transformation of the entire portfolio held.
80 Chapter 3. Dealer Markets

µ
...
...
6
...  ..
U 0 .... ...  ...
...
.
. U  1 ...
..
.. ...
...
 ...
... ..
..
....
.  .
.
..
...  ...
.... ...
µ ....  ...
r ...... .
....
....
.  ...
...
.... ...
.... ....

......  ......
.. ....
.... ....
....

.... ....
... ....
.
...
 ..
.
.. ..
 ..... ....
..... ....
.....
...... ....

..
.......
. .....
.
......
 ..
...... ....
 ...... ....
...... ....
 ...... ...
E u .
......
.......
...
.
....
....
.
....... ....
....... ......
.......
 .
. .....
.
...
... ....
.
........ ......

........ .....
 ........ .....
........ ......
 ...
.............. ..
........
.
... ......
......... ......
.........  .......
.......... .......
.......... .......
............
...
.....
 ..
..
.
.... ...
 .............. .......
.............. .......
................. .......
.................. .......

6 ...................
.......
...
...................... ......
.........
............................................... ..
 ........
........
.........

.........
 Costs .........
..........
..........
..........

..........
 ............
.
.... .....
............
...
.............

..............
................
 ? 0
Eu ..................
.
................
... ....
.....................
.
.......
..................................................................


 -
σr σ

Fig. 3.2: Costs of a market maker

i.e. a higher total wealth. For holding a non-optimal portfolio he is compensated

by an increase in the level of holdings.
It is also obvious from figure 3.2 that, if the market maker does currently not
hold his optimal portfolio, i.e. his inventory is nonzero, he may gain in utility
from accepting an order. Suppose as an example that he holds portfolio E 0 , be
accepting a sell order of the same size as before the buy order, he would gain
utility by reaching portfolio E again, hence his costs would be negative. In this
case pb would lie above p∗ in figure 3.1.
We only assume these costs to occur which are also called inventory costs.
Other costs, e.g. for order processing, are not included here, but they can easily
incorporated into this framework without changing the argument significantly.9
Early contributions to market microstructure theory were concerned with the

9
Stoll (1978, pp. 1144 ff.) also provides a more general framework by assuming more than
a single risky asset to be in the market. But as he also assumes the market maker to act as
market maker only for a single asset the results obtained are identical to those in this restricted
version, they only come with more notational complications.
3.1. Inventory-Based Models of Market Making 81

ability of the market maker to deliver the asset or the money. By allowing no
short sales of the market maker he faces the risk of running out of stock, of assets
as well as of money.10 Although such a situation cannot be ruled out by any
model, it has been found to be not relevant in practice. By allowing short sales
a bankruptcy of the market maker can be avoided and these concerns have not
to be considered. As also the results of the models focusing on the threat of
bankruptcy give similar results, models of inventory costs have attracted more
attention in the literature.
For bearing the above described inventory costs the market maker has to be
compensated, i.e. his expected utility of terminal wealth from holding the initial
portfolio and from holding the new portfolio after having accepted a trade have to
f ∗ denote the terminal wealth without a trade and W
be equal. Let W f the terminal

wealth after having accepted a trade, then the costs are determined such that
h i h i
(3.2) E U (Wf ∗ ) = E U (W f) .

The initial portfolio has not necessarily to be the optimal portfolio, it is the
optimal portfolio plus the inventory of the market maker, which can be either
positive or negative. Let k denote the fraction of total wealth that is invested
into the risky asset in the optimal portfolio, the optimal holding of the risky
asset then is kW0 , where W0 denotes the initial total wealth. The total amount
of the risky asset actually held is kW0 + I, with I denoting the inventory. With R
e

denoting the return of the fundamental value of the risky asset, the final wealth
of the portfolio is

(3.3) f ∗ = W0 + (kW0 + I)R

W e
   
I
= W0 1 + k + R
e .
W0
e and σ 2 = V ar[R]
With µ = E[R] e we get
   
∗ I
(3.4) E[W
f ] = W0 1 + k + µ ,
W0
 2
∗ 2 I
(3.5) V ar[W ] = W0 k +
f σ2.
W0
10
See e.g. Garman (1976) or Amihud and Mendelson (1980).
82 Chapter 3. Dealer Markets

f ∗ ) by a second order Taylor series around E[W

Approximating U (W f ∗ ] we get

h i 
(3.6) f∗
E U (W ) = E U (E[W f ∗ ]) + U 0 (E[W
f ∗ ])(W
f ∗ − E[W
f ∗ ])

1 00 ∗ ∗ ∗ 2
+ U (E[W ])(W − E[W ])
f f f
2
 
∗ 1 00 ∗ ∗
= U E[W ] + U (E[W ])V ar[W ] .
f f f
2

Let Q denote the trade size, measured in value not in numbers of assets traded,
where Q > 0 for a sell order and Q < 0 for a buy order. We further denote C
as the costs of the market maker to conduct a trade of size Q, transformed from
utility into value. We then have for the terminal wealth with accepting a trade:

(3.7) W
f = W0 + (kW0 + I)R e − (Q − C)
e + Q(1 + R)
   
I e − (Q − C),
= W0 1 + k + R
e + Q(1 + R)
W0

where the first term denotes the part of terminal wealth that arises from holding
the initial portfolio, the second term the part that has been affected by the change
in inventory due to accepting an order of size Q and the last term the benefits in
money from conducting this trade.

From (3.7) we get with assuming the order size Q to be known11

   
I
(3.8) E[W
f] = W0 1 + k + µ + Q(1 + µ) − (Q − C)
W0
   
I
= W0 1 + k + µ + Qµ + C,
W0
 2  
2 I 2 2 2 I
(3.9) V ar[W
f] = W0 k + σ + Q σ + 2W0 k + Qσ 2
W0 W0
   2
I
= σ 2 W0 k + +Q .
W0

11
If we do not know the order size, but only its expected value and variance the results do
not change, instead of the order size only its expected values have to be used if we assume the
order size to be independent of other factors, e.g. costs.
3.1. Inventory-Based Models of Market Making 83

Approximating U (W
f ) by a second order Taylor series around E[Wf ] gives
h i 
(3.10) E U (W ) = E U (E[W
f f ]) + U 0 (E[W f − E[W
f ])(W f ])

1 00 2
+ U (E[W ])(W − E[W ])
f f f
2
 
1 00
= U E[W ] + U (E[W ])V ar[W ] .
f f f
2

If Q is relatively small compared to the initial wealth of the market maker we

can assume that

(3.11) U 0 (E[W
f ∗ ]) = U 0 (E[W
f ]),

(3.12) U 00 (E[W
f ∗ ]) = U 00 (E[W
f ]).

f 0 between W
The mean-value theorem states that there exists a W f ∗ such
f and W

that
f ∗ ])
f ]) − U (E[W
U (E[W
= U 0 (E[W
f 0 ]).
E[W ] − E[W ]
f f ∗

With (3.11) we can rewrite this as

f ∗ ])
f ]) − U (E[W
U (E[W
(3.13) = E[W f ∗ ].
f ] − E[W
0
U (E[W ])
f ∗

By inserting (3.6) and (3.10) into (3.2) we get with (3.12) after rearranging

f ∗ ]) = 1 U 00 (E[W
f ]) − U (E[W
U (E[W f ∗ ])(V ar[W
f ∗ ] − V ar[W
f ]).
2

With z as the Arrow-Pratt measure of absolute risk aversion we get after dividing
by U 0 (E[W
f ∗ ]) and inserting (3.4), (3.5), (3.8), (3.10) and (3.13):
  
1 2 2 I
(3.14) Qµ + C = zσ Q + 2QW0 k + .
2 W0

f ∗ = W0 (1 + k R),
For the optimal portfolio, i.e. I = 0, we see from (3.3) that W e

hence
 
f ∗ )] = U ∗ 1 ∗
(3.15) E[U (W E[W ] − zV ar[W ]
f f
2
 
1 2 2 2
= U W0 (1 + kµ) − zW0 k σ
2
84 Chapter 3. Dealer Markets

such that

∂U
(3.16) = W0 kU 0 (.),
∂µ
∂U
(3.17) = −zW02 k 2 U 0 (.).
∂σ 2

By totally differentiating (3.15) we get

(3.18) f ∗ )] = ∂U dµ + ∂U dσ 2
dE[U (W
∂µ ∂σ 2
= (W0 kdµ − zW02 k 2 dσ 2 )U 0 (.).

Setting (3.18) equal to zero we get the slope of the indifference curve at the
optimal portfolio:

dµ
(3.19) = zW0 k.
dσ 2

The slope of the security market line is known from portfolio selection theory to
be

dµ µ
(3.20) 2
= 2.
dσ σ

As we know from portfolio selection theory these two slopes have to be identical,
hence we get from these two relations after rearranging:

µ
(3.21) k= .
zW0 σ 2

Inserting (3.21) into (3.14) we get

  
1 2 2 µ I
(3.22) C = −Qµ + zσ Q + 2QW0 +
2 zW0 σ 2 W0
1
= −Qµ + zσ 2 Q2 + Qµ + zσ 2 QI
 2 
2 1 2
= zσ Q + QI .
2

This is the expression for the costs a market maker faces when accepting orders
of size Q. The costs depend on a characteristic of the asset, the variance of the
fundamental value of the asset σ 2 , a characteristic of the trade, the trade size Q,
3.1. Inventory-Based Models of Market Making 85

and two characteristics of the market maker, his risk aversion z and his inventory
position I.
It is more natural to assume the trade size to be always positive. Define
Q0 = |Q| as the trade size and we get the costs for a trade at the ask, Ca , and at
the bid, Cb , by
 
2 1 2 0
(3.23) Cb = zσ Q +QI ,
2
 
2 1 2 0
(3.24) Ca = zσ Q −QI .
2
The relative costs are given by
 
Cb 2 1 0
(3.25) cb = = zσ Q +I ,
Q0 2
 
Ca 2 1 0
(3.26) ca = = zσ Q −I .
Q0 2
When buying the asset the market maker reduces the price at which he is willing
to buy compared to the fundamental value and increases it when selling the asset.
Therefore the reservation prices of a market maker for the entire trade of size Q0
are given by

(3.27) pb = p∗ − Cb ,

(3.28) pa = p∗ + Ca .

The difference between the bid and the ask price is denoted the spread, s. From
(3.23) - (3.28) we get

(3.29) s = pa − pb = Ca + Cb = zσ 2 Q2 .

The spread does not depend on the inventory of the market maker, but only on
his risk aversion, the trade size, and the variance of the fundamental value of the
asset.12 Unless the market maker is risk neutral, i.e. z = 0 or the asset is riskless,
i.e. σ 2 = 0, the spread will always be positive. We therewith have verified the
result already earlier stated that

(3.30) pa > p b .
12
The reservation prices for trading a single unit of the asset, as usually published, are given
86 Chapter 3. Dealer Markets

The inventory does not influence the spread, but only the level of prices. It can
easily be verified that if I 6= 0 the spread is not located symmetrically around p∗ .
If the inventory is sufficiently large the costs can even become negative, implying
that pb > p∗ or pa < p∗ .
These results derived by Stoll (1978) made use of the assumption that the
market maker does know the trade size before quoting the price or that he is
allowed to quote different prices for every trade size. In reality, however, the
market maker does not know the trade size, he has to accept any trade at a
single stated price up to a certain limit.13 This adds another uncertainty to the
market maker, the order size. It can be shown that the results above do not
change if the order size is independent of the costs and expected return of the
asset. The order size becomes a random variable and therefore instead of Q0 and
Q2 we have to insert E[Q0 ] and E[Q2 ] into the above derived formulas. No further
insight can be gained from this generalization.
The costs and reservation prices derived here are those that occur if the market
maker has a time horizon of a single trade, i.e. he makes his consideration on
a trade-by-trade basis. Such a short-term behavior can be justified if there is a
fierce competition between market makers.
If the market maker has a time horizon longer than a single trade the costs
are reduced. By accepting a trade with which the inventory position becomes
less favorable there exists the chance that in one of the next trades an offsetting
order arrives, what reduces his costs. Nevertheless the influences of the above
parameters on the costs do not change significantly.
by

pb = p∗ (1 − cb ),
pa = p∗ (1 + ca ).

and the spread is

s = pa − pb = p∗ (ca + cb ) = p∗ zσ 2 Q0 .
In this representation we can see that the higher the fundamental value of the asset is, the
higher we expect the spread to be.
13
At the NASDAQ Rule 4613 requires a market maker has to accept orders of at least an
equivalent of USD 50,000-100,000, depending on several characteristics of the asset and the
market.
3.1. Inventory-Based Models of Market Making 87

B
A

-
pA
b pB
b p∗ pA
a pB
a p

Fig. 3.3: Competitive price setting

After having derived the costs and reservation prices of a market maker we
will in the following subsections discuss the price setting of market makers, first
under competition and then for a monopolistic market maker.

3.1.2 Competitive price setting

Let us for simplicity assume throughout this subsection that all market makers
have the same risk aversion and the order size is fixed to Q0 , i.e. the spread
quoted by all market makers is identical. If all market makers have the same
inventory we can easily see from (3.23) - (3.28) that they all have the same costs
and reservation prices. Hence, if they act competitively they all will quote their
reservation prices. Quoting a lower bid or a higher ask price would make this
market maker not to offer the best price and hence he is excluded from the order
flow by the assumption of strict price priority and would not receive a trade and
thereby make no profits, like in the case when he quotes his reservation price.
Quoting a higher bid or a lower ask price would bring him a loss, so that he will
not quote such prices.
After a single trade however, the inventory position of the market maker
executing the trade will change and thereby the costs and reservation prices. Ho
and Stoll (1980) provide a framework how competitive market makers set their
quotes in such a situation.
Figure 3.3 illustrates the situation where two market makers, A and B, are in
the market and have different costs. Suppose the price setting of the bid price,
the market maker with the highest bid price receives an incoming order, if both
88 Chapter 3. Dealer Markets

quote the same price it is assigned to one of them randomly. Market maker B
has the lower costs for a trade at the bid, but by quoting his reservation price he
would make no profits. If he lowers his quote he would be able to make a profit.
As the costs of market maker A do not allow him to quote a higher price than pA
a

he cannot prevent him from doing so, market maker B still has the best price in
the market and will receive all orders that arrive at the market until he quotes a
price just a fraction above pA
a.
14
As quoting a price just a fraction above pA
a gives

him the highest profits, he will quote this price. By quoting the same price, pA
a,

he would have to share the order flow with the other market maker, reducing his
expected profits.15 For simplicity we neglect the fraction that has to be quoted
above pA A
a and say that he will quote pa , only keeping in mind this arbitrary small

fraction.

As is obvious with the same risk aversion it is not possible for a market maker
to quote the best price as well on the ask as the bid side, unless all market makers
have the same costs and quote their reservation prices. An inventory position
allowing him to have smaller costs on one side of the trade gives rise to larger
costs on the other side as can easily be seen from (3.23) and (3.24). Generalizing
the example with two market makers to M market makers, we derive easily in a
similar way the result that the market maker with the lowest costs for this side
of the trade receives the order. He quotes the reservation price (minus a fraction)
of the market maker with the second lowest costs. If we define C 1 and C 2 as the
costs of the market maker with the lowest and second lowest costs, the profits

14
We assume here that prices can be set continuously for simplicity. In case of discrete prices
he would have to quote the last discrete price above pA B
a but below pa . We also rule out the
possibility that the profit maximum lies between pA a and p B
a as λ b is falling with pb . In this
case we get a behavior comparable to a monopolistic market maker to be treated in 3.1.3.
15
The possibility that both market makers collude to make higher profits by quoting prices
that are below the reservation prices of both market makers is neglected here.
3.1. Inventory-Based Models of Market Making 89

from a trade are given by

(3.31) πb1 = Cb1 − Cb2

   
2 1 2 0 1 2 1 2 0 2
= zσ Q + Q I − zσ Q +QI
2 2
= zσ 2 Q0 I 1 − I 2 ,

πa1 = Ca1 − Ca2

   
2 1 2 0 1 2 1 2 0 2
= zσ Q − Q I − zσ Q −QI
2 2
= zσ 2 Q0 I 2 − I 1 .

Therewith the spread observed in the market (market spread ) is the difference
between the reservation prices of the market makers with the second lowest costs
on each side of the trade. Following Ho and Stoll (1983) we will investigate
how the market spread is related to the reservation spread. Until now we assumed
the market makers to be passive, i.e. to wait for an order arriving at the market,
to offset any undesirable inventory position. Another possibility for the market
makers would be to initiate a trade by themselves with another market maker
(interdealer trading).
The expected utility to wait for an offsetting order to arrive at the market
is for the market maker with the lowest costs (the other market makers will not
receive a trade)

(3.32) E[U (W1 )] = E[U (W0 )] + U 0 (W0 )λa (Ca2 − Ca1 ),

where W1 denotes the wealth of the market maker in the next period, W0 the
initial wealth, λa the probability that an order arrives at the market in the next
period and Ca1 − Ca2 the gain from the trade according to (3.31). This gain in
wealth is transformed into utility by multiplying with the first derivative of the
utility function by the concept of a first order Taylor series approximation.
For interdealer trading we assume that it takes place at the beginning of a
period and that the market maker has enough time to update his quotes for
his new inventory and then waits for orders to arrive at the market. We further
90 Chapter 3. Dealer Markets

assume that the market maker with the second lowest costs quotes his reservation
price.
The expected utility of wealth after interdealer trading, W00 , is composed of
the expected utility from his initial wealth, W0 , and an adjustment for the fee
he has to pay the other market maker, % = Cb2 , and the gain from offsetting his
inventory, −Ca1 :

(3.33) E[U (W00 )] = E[U (W0 )] + U 0 (W0 )(−Ca1 − %).

After this interdealer trade his inventory has changed from I 1 to I 1 − Q and
0
therewith costs have changed from Ca1 to Ca1 . The expected utility from a trade
is now

0
(3.34) E[U (W10 )] = E[U (W00 )] + U 0 (W0 )(Ca2 − Ca1 )λa .

If we assume W0 and W00 to differ not too much such that U 0 (W0 ) = U 0 (W00 ) we
get from (3.33) and (3.34):
 0

(3.35) E[U (W10 )] = E[U (W0 )] + U 0 (W0 ) (Ca2 − Ca1 )λa − Ca1 − % .

For choosing interdealer trading rather than waiting for an offsetting order it is
necessary that E[U (W10 )] > E[U (W1 )]. Inserting from (3.32) and (3.35) we get
after rearranging:

0
λa (Ca2 − Ca1 ) < λa (Ca2 − Ca1 ) − Ca1 − %.

0
Inserting for Ca1 and Ca1 we get after eliminating Ca2 :
    
2 1 2 0 2 1 2 0 0
−% > λa −zσ Q − Q I1 + zσ Q − Q (I1 − Q
2 2
 
1 2
+zσ 2 Q − Q0 I1
2
 
2 2 1 2 0
= zσ Q λa + Q − I1 Q ,
2

  
2 0 2 1
(3.36) % < zσ I1 Q − Q λa + .
2
3.1. Inventory-Based Models of Market Making 91

If there are only two market makers the market maker who wants to trade with
his only competitor will be charged the fee that equals the second best reservation
price, i.e. his own reservation price at the bid. Inserting from (3.23) for % we get
    
2 0 1 2 2 0 2 1
zσ I1 Q + Q < zσ I1 Q − Q λa +
2 2

which solves for

λa > 1.

As λa is the probability that an order arrives at the market, it cannot exceed 1,

hence with two market makers no interdealer trading occurs.
With more than two market makers we get with % = Cb2 = zσ 2 I2 Q0 + 21 Q2

from (3.36):
    
2 0 1 0 1
zσ I2 Q + Q2 < zσ 2 2
I1 Q − Q λa + ,
2 2

which solves for

(3.37) I1 − I2 > Q0 (1 + λa ),

i.e. if the difference in the inventory between the best and the second best market
maker is large enough, interdealer trading is induced. For the second best market
maker we find λa = 0 as he will never serve a trade, therewith he induces a trade
only if I1 − I2 > Q0 . For the best market maker we find 0 ≤ λa ≤ 1, hence for him
the divergence has to be even larger before he initiates a trade as he can hope to
offset his inventory by an incoming order. If I1 − I2 > 2Q0 he will always induce
an interdealer trade.
If there is the situation that for all market makers I1 − I2 < Q0 (1 + λa ), i.e. no
market maker wants to induce an interdealer trade, there will be no interdealer
trading in the future provided that the parameters Q0 and λa do not change. To
see this suppose that all market makers have the same inventory, an order arriving
can lead to a deviation of Q0 for this market maker from the others, no market
maker wants to induce a trade. Further orders arriving at the market affect only
the inventories of the market makers with the most deviating inventories as they
92 Chapter 3. Dealer Markets

quote the best prices. They can therewith only narrow the divergence and no
interdealer trading takes place.
The market spread, sM , is the difference between the best quoted prices on
each side of the trade. The best quoted prices are the reservation prices of the
second best market makers, hence we have

(3.38) sM = Ca2 + Cb2

   
2 1 2 0 2 0 1 2
= zσ Q − I1 Q + zσ I2 Q + Q
2 2
2 2 2 0
= zσ Q + zσ Q (I2 − I1 )

= s + zσ 2 Q0 (I2 − I1 ).

With only two market makers it has been shown that no interdealer trading
occurs, therewith I2 − I1 ≤ Q0 , hence we find that

(3.39) s ≤ sM ≤ 2s.

With three market makers, one of them is the best on the ask, one at the bid
and the third is the second best on both sides, hence I2 = I1 and we get

(3.40) sM = s.

With more than three market makers, the second best market maker on the bid
side will have a lower inventory than the second best at the ask side, i.e. I2 ≤ I1 .
As the difference is allowed to be maximal Q0 we find that with more than three
market makers

(3.41) 0 ≤ sM ≤ s.

Incoming orders have the tendency to balance the inventory of the market makers
as always the market maker with the most deviating inventory serves the order.
Therewith we have the tendency of I2 − I1 to converge to zero and the market
spread to converge to the reservation spread. If the inventories are quite similar
another order increases the divergence and the process of convergence can start
again.
3.1. Inventory-Based Models of Market Making 93

Allowing the trade size to vary, adds many more possibilities how the spread
can behave over time, but the general finding of the mechanism is not changed.
As all market makers have to pose their quotes for a period simultaneously,
it has implicitly been assumed that market makers know the inventory positions
of their competitors and can therewith calculate their reservation prices and set
their prices accordingly, especially the reservation price of the second best market
maker has to be known. By observing past prices of the other market makers it
is also possible to determine the inventory position by inverting equation (3.23)
and (3.24) without initially knowing their inventory position.
If other factors also influence the quoted prices and reservation prices, the
exact inventory position cannot so easily be determined. It may be impossible
to infer the exact reservation prices. If the market makers can infer only the
probability distribution of the reservation prices instead of the the exact reser-
vation prices, Biais (1993) has shown that the bid and ask quotes, on average,
are identical to those quoted with full knowledge of the other market makers’
reservation prices. He further finds that in this case the spread and the quoted
prices are more volatile than with full knowledge.
After having investigated the competitive price setting, we now turn to the
price setting of a monopolistic market maker.

3.1.3 The price setting of a monopolistic market maker

In several markets, e.g. the NYSE, a single market maker is granted a monopoly
in providing his services. The main focus of a monopolistic market maker is not
to cover his costs, but to maximize his expected utility of terminal wealth by
choosing optimal bid and ask prices. Ho and Stoll (1981) provide a model,
how a monopolistic market maker sets his prices.
As the demand for the service of the market maker will play an important
role, at first we will model this side in more detail. The order arrival rates λa
and λb can be interpreted as the probability that an order arrives at the market
within a given time period [t, t + 1[. Approximating by a first order Taylor series
94 Chapter 3. Dealer Markets

around zero gives us

∂λa (0)
(3.42) λa (pa ) = λa (0) + pa ,
∂pa
∂λb (0)
(3.43) λb (pb ) = λb (0) + pb .
∂pb

If we denote the quoted prices to be the fundamental value of the asset adjusted
by the fees the market maker charges, xa and xb , respectively, we get in analogy
to (3.27) and (3.28):

(3.44) pa = p∗ + xa ,

(3.45) pb = p∗ − xb .

Inserting these relations into (3.42) and (3.43) we get

∂λa (0) ∗ ∂λa (0)

(3.46) λa (pa ) = λa (0) + p + xa ,
∂pa ∂pa
∂λb (0) ∗ ∂λb (0)
(3.47) λb (pb ) = λb (0) + p + xb .
∂pb ∂pb

The first two terms can be interpreted as the demand of the market maker if
he would charge no costs, i.e. it is a measure for the size of the liquidity event
investors face. We will denote these terms by αa and αb , respectively. The last
term can be interpreted as an adjustment in the demand due to the sensitivity
of the demand to fees charged, the absolute values of this sensitivities will be
denoted βa and βb , respectively:

(3.48) λa (pa ) = αa − βa xa ,

(3.49) λb (pb ) = αb − βb xb .

We assume that there are T time periods in which trading can take place, at
every point of time t ∈ [0, T ] the market maker chooses bid and ask fees that are
optimal in the sense that they maximize his expected utility of terminal wealth.
At time T the asset is liquidated at the fundamental value and the proceedings
are consumed. As before the total wealth of the market maker consists of two
3.1. Inventory-Based Models of Market Making 95

components, his inventory and money:16

(3.50) Wt = It + Mt

for all t = 1, . . . , T . The inventory changes with the rate of return and with
trading. By assuming again a fixed trade size Q0 we get

(3.51) et+1 It ∆t + p∗ Qb + p∗ Qa ,
∆It+1 = R

where Qb equals Q0 if a trade at the bid occurs and is zero otherwise. Qa equals
Q0 if a trade at the ask occurs and is zero otherwise. Money holdings change with

(3.52) ∆Mt+1 = (p∗ − xb )Qb + (p∗ + xa )Qa = pb Qb + pa Qa .

Let W
f denote the terminal wealth of the market maker, he then has to maximize

E[U (W
f )] by choosing optimal fees xa and xb . We can define a performance

function J as

(3.53) J(t, Mt , It ) = max E[U (W

f )|t, Mt , It ].
xa ,xb

After the last trade at time t = T has taken place the portfolio is liquidated.
Therefore the market maker faces no uncertainty and (3.53) has to fulfill the
boundary restriction that

(3.54) J(T, MT , IT ) = U (WT ).

With optimal fees no further increase in J can be achieved by changing fees, i.e.

(3.55) max dJ(t, Mt , It ) = 0.

xa ,xb

From the principle of optimality in dynamic programming we can use the funda-
mental recurrence relation and rewrite (3.53) as

(3.56) J(t, Mt , It ) = max {L(t, Mt , It , xa , xb )∆t

xa ,xb

+J(t + ∆t, Mt + ∆Mt+1 , It + ∆It+1 )} ,

16
Ho and Stoll (1981) also consider the optimal portfolio as part of the wealth. But as he
is compensated for the risks associated with holding the optimal portfolio by the market it is
not necessary to take into account this part of his wealth, but only deviations.
96 Chapter 3. Dealer Markets

where L(t, Mt , It , xa , xb ) denotes the expected gain in utility in the current period
and J the performance in the remaining periods. The expected gain in utility in
the present period consists of the expected gain from a trading at the bid and at
the ask, what can be interpreted as the difference in the performance functions
with and without a trade:

(3.57) L(t, Mt , It , xa , xb ) = λa (J(t, Mt + pa Q0 , It − Q0 ) − J(t, Mt , It ))

+λb (J(t, Mt − pb Q0 , It + Q0 ) − J(t, Mt , It )) .

We can further approximate J(t + ∆t, Mt + ∆Mt+1 , It + ∆It+1 ) by a first order

Taylor series around (t, Mt , It ):

(3.58) J(t + ∆t, Mt ∆Mt+1 , It + ∆It+1 ) = J(t, Mt , It ) + Jt ∆t

+JM ∆Mt+1 + JI ∆It+1 ,

where the subscripts denote the partial derivatives with respect to this variable
evaluated at the appropriate point. Inserting (3.57) and (3.58) into (3.56) we get
after dividing by ∆t and rearranging:

∆It+1 ∆Mt+1
(3.59) − Jt = max JI + JM
xa ,xb ∆t ∆t
+λa (J(t, Mt + pa Q , It − Q0 ) − J(t, Mt , It ))
0

0 0
+λb (J(t, Mt − pb Q , It + Q ) − J(t, Mt , It )) .

If we let ∆t → 0, we see that

∆It+1 ∂It
(3.60) → ,
∆t ∂t
∆Mt+1 ∂Mt
(3.61) → = 0.
∆t ∂t

As with ∆t also λa and λb converge to zero, the probability of a trade arriving

within a very short period of time also approaches zero. We now assume that the
return of the risky asset in a given period of time follows an Itô process:

(3.62) R
et+1 = µdt + σdz
3.1. Inventory-Based Models of Market Making 97

with dz denoting a standard Wiener process. Therewith we have from (3.51)

with ∆t → 0:

(3.63) dIt = µIdt + σIdz.

From Itô’s lemma we get

1
(3.64) dJ(t, Mt , It ) = JI dIt + Jt dt + JII (dIt )2 ,
2

where

(3.65) (dIt )2 = σ 2 It2 dt,

(3.66) JI dIt = µIt JI dt.

As time is no own variable in the model we find

(3.67) Jt dt = 0.

Inserting (3.65) - (3.67) into (3.64) gives after dividing by dt:

1
(3.68) Jt = µIt JI + JII σ 2 It2 .
2

Using these results we can rewrite (3.58) as

1
(3.69) − Jt = µIt JI + JII σ 2 It2
2
+ max {λa (J(t, Mt + (p∗ + xa )Q0 , It − Q0 ) − J(t, Mt , It ))
xa ,xb

+λb (J(t, Mt − (p∗ − xb )Q0 , It + Q0 ) − J(t, Mt , It ))} .

We denote τ as the remaining time or time horizon, i.e. τ = T − t.. Therewith

we have

∂τ
(3.70) Jt = Jτ = −Jτ .
∂t

Define further LJ = µIt + 12 σ 2 It2 JII , BJ = J(t, Mt +p∗ Q0 , It −Q0 ), SJ = J(t, Mt −

p∗ Q0 , It + Q0 ) and J = J(t, Mt , It ). Using this notation we can approximate
98 Chapter 3. Dealer Markets

J(t, Mt + (p∗ + xa )Q0 , It − Q0 ) and J(t, Mt − (p∗ + xb )Q0 , It + Q0 ) by a first order

Taylor series around (t, Mt +p∗ Q0 , It −Q0 ) and (t, Mt −p∗ Q0 , It +Q0 ), respectively:

(3.71) J(t, Mt + pa Q0 , It − Q0 ) = J(t, Mt + p∗ Q0 , It − Q0 ) + JM xa Q0

= BJ + JM xa Q0 ,

(3.72) J(t, Mt − pb Q0 , It + Q0 ) = J(t, Mt − p∗ Q0 , It + Q0 ) + JM xb Q0

= SJ + JM xa Q0 .

Therewith (3.69) can be written as

(3.73) Jτ = LJ + max {λa (BJ − J) + λb (SJ − J)

xa ,xb

+λa xa Q0 BJM + λb xb Q0 SJM } .

Inserting for λa and λb we can conduct this maximization and obtain the following
first order conditions for the optimal fees xa and xb :

(3.74) αa − 2βa xa Q0 BJM − βa (BJ − J) = 0,

αb − 2βb xb Q0 SJM − βb (SJ − J) = 0.

The second order condition for a maximum can be shown to be fulfilled as the
Hesse-Matrix is negative definite. Solving for the optimal fees we get:

αa J − BJ
(3.75) xa = + ,
2βa 2Q0 BJM
αb J − SJ
xb = + .
2βb 2Q0 SJM

The first term can easily be verified to be the optimal fee at which benefits are
maximized, provided that a predetermined trade occurs with certainty. The sec-
ond term adjusts the fees for the risk in the return of the asset and the uncertainty
of the order flow. To derive an explicit expression for the fees we have to find
a solution for J. Ho and Stoll (1981) provide a separate mathematical ap-
pendix where they show how to derive a solution for J by further Taylor series
approximations. We do not track down this long and tedious derivation, but only
state the results they achieve if the time horizon is not too long.
3.2. Information-Based Models of Market Making 99

The fees a monopolistic market maker charges to compensate for the risks he
faces, i.e. the second terms in (3.75), depend positively on his risk aversion, the
variance of the fundamental value and the trade size. The expressions are similar
to those obtained when deriving the costs of market making. As a fourth factor
the profits he can make influence his fees, the larger the expected profits are, the
higher fees he charges. These profits depend on the size of the liquidity event (αa
and αb ) and the sensitivity to a higher fee (βa and βb ). A final factor is the time
horizon. The longer the time horizon the higher the fees. With a longer time
horizon the market maker faces a higher total risk from holding the inventory,
this effect offsets the benefits from the possibility of an offsetting order arriving
at the market.
The formula provided in the appendix to Ho and Stoll (1981) gives no
further substantial insight. The exact relations are subject to many approxima-
tions, such that it has to be handled with care, qualitative reasonings seem more
appropriate than exact interpretations.
In comparing the results of the price setting for a monopolistic market maker
and competitive market makers it can be seen that the main difference lies in
the fact that a monopolistic market maker is not only compensated for the risk
directly connected to his inventory, but also for the risks of the future order flow,
i.e. for the risk of an unfavorable shift in his inventory. As this risk increases the
longer his time horizon is, the higher the fee he charges.

3.2 Information-Based Models of Market

Making
The last section focused on the influence of inventory costs on bid and ask prices
and the spread. It has been assumed that all investors and market makers agree
on the fundamental value of the asset, i.e. have equal information. In this section
we will take up the line already laid in section 2 by assuming that there exist two
groups of investors, informed and uninformed investors. Unlike in the models
building on Kyle (1985) we will not aggregate the order flow over a given period
100 Chapter 3. Dealer Markets

of time, but examine the behavior on a trade-by-trade basis by introducing a

market maker instead of a match maker.
The basis for the information-based models of market making has been laid
by Bagehot (1971) with his distinction of market and trading gains. A market
gain arises from the price change and dividends of an asset in a given time period.
An investor can realize the market gain by holding the asset during the entire
time period without trading. Let p1 denote the price of the asset at the end of
the period and p0 at the beginning, then with assuming that no dividends are
paid, the market gain is

(3.76) ∆p = p1 − p0 .

The market gain is the same for all investors holding the asset at the beginning
and at the end of the period. Let us now assume that all investors hold the
same amount of assets at the beginning and at the end of the period.17 Investors
can either only hold the asset over the entire period or they can trade with each
other. The gain associated with this trading activity is the trading gain, denoted
f i . The total gain of investor i is

(3.77) π i = ∆p + f i .

The total gain of all investors can only be the market gain by comparing the
wealth at the beginning and at the end of the period, i.e. with N investors it is
N
X N
X
(3.78) π= πi = ∆p = N ∆p.
i=1 i=1

By aggregating (3.77) over all investors we get

N
X N
X N
X
i i
(3.79) π= π = (∆p + f ) = N ∆p + f i.
i=1 i=1 i=1

Inserting from (3.78) we receive

N
X
(3.80) f i = 0,
i=1
17
If some investors decide to liquidate their position in the asset there has to be found another
investor taking his position as the number of outstanding shares is fixed. We therefore could
aggregate these investors such that the assumption is fulfilled.
3.2. Information-Based Models of Market Making 101

i.e. the total trading gains are zero. For every investor making trading prof-
its there must be another making a loss. If there are two groups of investors,
informed and uninformed, where the uninformed investors have to trade for ex-
ogenous reasons, the informed investors will only trade if they expect to profit
from trading, while the uninformed are forced to trade even if they make trading
losses.

With a market maker acting as counterpart in every trade, the informed

investors will only trade with him if he sets a price at which they make an
expected profit, otherwise they would refuse to trade. Hence, if we allow only for
a single trade per period of time, we see from (3.80) that as there are only two
market participants involved, the market maker will make a loss from trading with
an informed investor. He will try to reduce the loss he receives from trading with
an informed investor by quoting a less favorable price than what he thinks the
fundamental value is, but he will never be able to make a gain. This remaining loss
he has to offset from another source. This other source are uninformed investors,
he has to charge a price to them such that he makes a gain from trading with
them and they make a loss.

By the anonymity of the two investors provided by a broker, the market maker
does not know whether the investor he trades with is informed or uninformed.
We will see below that this problem of loosing to one and gaining from the other
group gives rise to the spread. The losses to informed investors are also called
adverse selection costs.

Bagehot (1971) further states that market makers will be uninformed. They
observe the order flow and try to balance the buy and sell orders by quoting
appropriate prices. By observing the order flow they aggregate the information
available in the market as well as the errors. Errors cancel out by the law of large
numbers if the number of informed investors is sufficiently high and they make
no systematic errors. If the market maker would become informed and quote his
prices according to his own information, he faces the risk of relying on a large
error in his considerations, causing him a large loss. Therefore he will not invest
102 Chapter 3. Dealer Markets

his wealth to become informed.18

3.2.1 Determination of adverse selection costs

The first to formalize the idea of Bagehot (1971) were Copeland and Galai
(1983). They provide a simple framework in which only a single trade takes
place before the information is fully revealed to all market participants, e.g.
by liquidation of the asset at the fundamental value. The liquidation value of
the asset is a random variable, pe, which has a known distribution F and an
expected value of E[e
p] = p0 . It is assumed that informed investors know the
exact liquidation value, p∗ , before trading takes place. Uninformed investors
trade for exogenous reasons, but their demands depend on the fee charged by the
market maker, the larger the fee, the smaller their demand. The market maker
is assumed to be risk neutral, i.e. he faces no inventory costs.
Trading takes place as follows: At the beginning informed investors become
to know the liquidation value and the market maker sets his prices. Knowing
these prices informed and uninformed investors place their orders. Only one of
these orders will be served by the market maker, this order is chosen randomly.
The probability that the order chosen is from an informed investor is γI . This
probability depends on the prices he quotes. The larger the fee from the point of
an uninformed investor, the less uninformed investors submit orders and hence
γI increases. An informed investor will submit a buy order if pa < p∗ and a sell
order if p∗ < pb , if pb < p∗ < pa he will not submit an order and all orders in
the market are from uninformed investors. As the market maker does not know
p∗ , he does not know what his loss is, he only knows that he makes a loss. His
expected profit from trading with an informed investor is
Z ∞ Z pb
(3.81) E[πI ] = (pa − p)dF (p) + (p − pb )dF (p) ≤ 0.
pa 0

18
Although the assumption of uninformed market makers is common in dealer markets, it
can reasonably be assumed that market makers can get private information from other sources,
such that he will not be completely uninformed. Calcagno and Lovo (1998) provide a model
where market makers are at least partially informed and have different information. However,
we consider this line of research not further at this point.
3.2. Information-Based Models of Market Making 103

If he trades with an uninformed investor the agreed value of the asset is p0 , he

will make a profit of pa − p0 from a trade at the ask and of p0 − pb from a trade
at the bid. Uninformed investors can either buy or sell the asset, we assign a
probability of γa for a trade at the ask and γb for a trade at the bid, such that
γa + γb = 1. These probabilities depend on the fees charged by the market maker,
or equivalently by the prices he sets, such that

∂γU ∂γU
(3.82) > 0, < 0,
∂pa ∂pb
∂γa ∂γb
< 0, > 0,
∂pa ∂pb
∂γa ∂γb
= 0, = 0.
∂pb ∂pa

The expected profits from trading with an uninformed investor are therewith

(3.83) E[πU ] = γa (pa − p0 ) + (1 − γa )(p0 − pb ) ≥ 0.

The total expected profits of the market maker are

(3.84) E[π] = γI E[πI ] + (1 − γI )E[πU ].

The costs or reservation prices are obtained by setting E[π] = 0. A monopolistic

market maker would maximize (3.84).

Solving the problem to determine the adverse selection costs would rely on
many assumptions and is therefore not conducted here. Nevertheless we can use
(3.84) to derive some implications of the model. It can be shown that we always
have a positive spread if γI > 0, i.e. if there are informed investors. Suppose that
104 Chapter 3. Dealer Markets

pa = pb = p0 , we then get from (3.81):

Z ∞ Z p0
0
(3.85) E[πI ] = (p − p)dF (p) + (p − p0 )dF (p)
p0 0
Z ∞ Z ∞ Z p0 Z p0
0 0
= p dF (p) − pdF (p) + pdF (p) − p dF (p)
p0 p0 0 0
Z p0 Z p0
0 0
= p −p dF (p) − E[p] + pdF (p)
0 0
Z p0 Z p0 Z p0
0 0
+ pdF (p) − p dF (p) − p dF (p)
0 0 0
Z p0
0
= p − p0 + 2 (p − p0 )dF (p).
0

R p0 R p0
As 0
(p − p0 )dF (p) < 0 we get for p0 < p0 that E[πI ] < 0 and as also 0
(p −
p0 )dF (p) < p0 − p0 we have for p0 ≥ p0 E[πI ] < p0 − p0 < 0. Hence we always have
E[πI ] < 0. From (3.83) we get

(3.86) E[πU ] = γa (p0 − p0 ) + (1 − γa )(p0 − p0 ) = (p0 − p0 )(2γa − 1).

If p0 < p0 we need γa < 1

2
for E[πU ] > 0, what is necessary to compensate for the
loss from trading with informed investors. This would imply that the probability
of a trade at the bid is higher than a trade at the ask, although the fee for
a trade at the bid is positive while it is negative for a trade at the ask. If we
reasonably assume that the probabilities for a trade are equal for the same fee this
contradicts (3.82), hence we cannot set p0 < p0 . With the same argumentation
the case p0 > p0 can be ruled out. Hence the only solution is to set p0 = p0 , for
which E[πU ] = 0. Inserting these findings into (3.84) we get

(3.87) E[π] = γI E[πI ] < 0.

As a market maker will not accept to make a loss the spread always has to be
positive if γI > 0.19
Further we can see from (3.84) that the more informed investors are present,
i.e. the higher γI is, the larger the first term becomes. To compensate the losses
19
A negative spread has earlier been pointed out to be not possible by arguments of arbitrage.
3.2. Information-Based Models of Market Making 105

from a higher probability of trading with informed investors a larger spread has
to be quoted.20
If the uncertainty about the liquidation value increases, i.e. more probability
is put on the tails of F , e.g. with a larger variance, we can see from (3.81)
that the expected losses from trading with informed investors increase. This is
compensated by the market maker with quoting a larger spread.
The results found are very similar to those of Kyle (1985) in the case of a
single auction. As with the Kyle model it would therefore be interesting to extend
this static model to a dynamic model by allowing more than a single trade in
a given time period before the asset is liquidated. This generalization has been
undertaken by the model of Glosten and Milgrom (1985).
With only a single trade before the asset is liquidated there is no need to
exploit information from the order flow. If there are more trades, information
from the order flow will be used by the market maker to minimize his losses from
trading with an informed investor through updating his beliefs.
The market maker uses all information available to him from former trades,
denoted Ωt . He knows that an informed investor only trades if the quoted price
is above (trade at the bid) or below (trade at the ask) the liquidation value, while
an uninformed investor trades independent of the quotes and fundamental value.
We only assume that the frequency with which he trades is sensitive to the fee
the market maker charges. His belief on the fundamental value equals that of
the market maker as both are uninformed, but the market maker can observe the
former order flows.
The probability that a buy or a sell order arrives at the market are given by

(3.88) P r(Buy order|Ωt ) = γI P r(p∗ > pta |Ωt ) + (1 − γI )γa ,

(3.89) P r(Sell order|Ωt ) = γI P r(p∗ < ptb |Ωt ) + (1 − γI )(1 − γa ).

20
In general it cannot be determined how this larger spread is achieved, by increasing only
the ask, decreasing only the bid price or a combination of these. Admati and Pfleiderer
(1988) provide an interesting strategy of the market makers in setting their prices as will be
presented in section 3.2.4.
106 Chapter 3. Dealer Markets

Let us denote the order form by ot , where


1 if the order is a buy order
(3.90) ot = ,
−1 if the order is a sell order

then Ωt = {o1 , . . . , ot−1 }. The expected profit of the market maker is given by

(3.91) E[πt |Ωt ] = (pta − E[p|Ωt , ot = 1])P r(ot = 1|Ωt )

+(E[p|Ωt , ot = 1] − ptb )P r(ot = −1|Ωt ).

In order to determine the costs of market making his expected profits have to
equal zero. He can achieve this either by setting both prices so that he makes no
profit on either side or he can set the prices so that he makes a profit on one side
of the trade and a loss on the other. If we assume competitive market makers
the profits on one side of the trade will be deteriorated by other market makers
undercutting the price by applying another price strategy, hence the second al-
ternative is ruled out as all market makers have the same costs. Therewith the
bid and ask prices are determined as follows:

(3.92) pta = E[p|Ωt , ot = 1],

ptb = E[p|Ωt , ot = −1].

With Bayes rule and (3.88) and (3.89) we get

(3.93) P r(p∗ > pt ) = P r(p∗ > pta |Ωt )P r(ot = 1|p∗ > pta , Ωt ) ×

× P r(p∗ > pta |Ωt )P r(ot = 1|p∗ > pta , Ωt )
−1
∗ t ∗ t
+P r(p ≤ pa |Ωt )P r(ot = 1|p ≤ pa , Ωt )

= P r(p∗ > pta |Ωt )(γI + (1 − γI )γa ) ×


×P r(p∗ > pta |Ωt )(γI + (1 − γI )γa )
−1
∗ t
+(1 − P r(p > pa |Ωt ))(1 − γI )γa
γI + (1 − γI )γa
= P r(p∗ > pta |Ωt ) ∗
P r(p > pta |Ωt )γI + (1 − γI )γa
= P r(p∗ > pta |Ωt ) if γI = 0

.
> P r(p∗ > pta |Ωt ) if γI > 0
3.2. Information-Based Models of Market Making 107

As long as there are informed investors a buy order increases the probability that
the fundamental value is above the ask. A similar result can be obtained for a
trade at the bid, it increases the probability that the fundamental value is below
the bid. With this result we get

(3.94) pta = E[pta |ot = 1] = E[E[p|Ωt , ot = 1]|ot = 1]

= E[p|Ωt , ot = 1]

= E[p|Ωt ] if γI = 0
,
> E[p|Ωt ] if γI > 0

i.e. as long as there is the threat of adverse selection the ask will exceed the fun-
damental value assigned by the market maker. The same result can be obtained
for the bid such that

(3.95) pta = E[p|Ωt ] = ptb if γI = 0,

pta > E[p|Ωt ] > ptb if γI > 0.

If there is the possibility of trading with an informed investor the spread will
always be positive due to adverse selection costs. As we can easily see, (3.93) is
increasing in the fraction of informed investors, γI . Therewith from (3.94) we see
that pta increases and analogue ptb decreases, hence the spread increases in γI .

Let pt denote the price at which a trade occurs at time t, i.e. pt = pta if ot = 1
and pt = ptb if ot = −1. By using (3.92) we get

(3.96) E[pt+1 |Ωt ] = E[E[p|Ωt , ot ]|Ωt ] = E[p|Ωt ] = E[p|Ωt−1 , ot ] = pt .

This implies that expected price changes between two subsequent trades are inde-
pendent of each other if the fundamental value does not change. These findings
are in contrast to the behavior with inventory costs where price changes are
negatively correlated.
108 Chapter 3. Dealer Markets

From (3.92), (3.94) and (3.95) we get

(3.97) pt+1
a = E[p|Ωt+1 , ot+1 = 1]

= E[E[p|Ωt+1 ]|ot+1 = 1]

= E[E[p|Ωt , ot ]|ot+1 = 1]

E[pta |ot+1 = 1] if ot = 1
=
E[ptb |ot+1 = 1] if ot = −1

E[p|Ωt+1 ] if ot = 1
>
E[p|Ωt+1 ] if ot = −1
 t
pa if ot = 1
=
ptb if ot = −1
pt+1
b = E[p|Ωt+1 , ot+1 = −1]

= E[E[p|Ωt+1 ]|ot+1 = −1]

= E[E[p|Ωt , ot ]|ot+1 = −1]


E[pta |ot+1 = −1] if ot = 1
=
E[ptb |ot+1 = −1] if ot = −1

E[p|Ωt+1 ] if ot = 1
<
E[p|Ωt+1 ] if ot = −1
 t
pa if ot = 1
= .
ptb if ot = −1

If the previous transaction has taken place at the ask (bid) the new ask (bid)
price is higher (lower) than the previous ask (bid) price. We see further that
always pt+1
a > ptb and pt+1
b < pta , i.e. the prices will never be revised so much
that both prices are outside the former spread. This price behavior can easily
be explained by looking at (3.92). The former ask and bid prices were the best
guess of the market maker given the transactions he was waiting for, hence from
(3.95) we see that pt+1
a > E[p|Ωt+1 ] = pt > pt+1
b .

1 1
PT
Define the spread as st = pta −ptb and ξt = P r(ot =1|Ωt )P r(ot =−1|Ωt )
, ξ= T t=1 ξt ,
3.2. Information-Based Models of Market Making 109

1
PT
s= T t=1 st and et = E[(pt − pt−1 )2 |Ωt ]. We then find with p0 = E[p]
" T #
X
(3.98) V ar[p] ≥ V ar[pT ] = V ar (pt − pt−1 )
t=1
T
X T
X
t t−1
= V ar[p − p ]= E[(pt − pt−1 )2 ]
t=1 t=1
" T
#
X
= E E[(pt − pt−1 )2 |Ωt ]
t=1
" T
#
X
= E et .
t=1

We further know that the bid and ask prices are increased following a trade
at the ask in the previous period and decreased following a trade at the bid.
Therewith we get

(3.99) et = (pat−1 − ptb )2 P r(ot−1 = 1|Ωt−1 )P r(ot = −1|Ωt−1 )

+(pbt−1 − pta )2 P r(ot−1 = −1|Ωt−1 )P r(ot = 1|Ωt−1 )

+(pat−1 − pta )2 P r(ot−1 = 1|Ωt−1 )P r(ot = 1|Ωt−1 )

+(pbt−1 − ptb )2 P r(ot−1 = −1|Ωt−1 )P r(ot = −1|Ωt−1 )

≥ (pta − ptb )2 P r(ot = 1|Ωt )P r(ot = −1|Ωt )

s2
= t.
ξt
Therewith we have found that

(3.100) et ξt ≥ s2t .

Aggregating over all T trades we get

T
X T
X
(3.101) et ξt ≥ s2t .
t=1 t=1

With the Cauchy-Schwartz inequality this can be transformed into

T T T
!2
X X X
(3.102) et ξt ≥ st .
t=1 t=1 t=1

Hence we get with the above definitions

T
X T 2 s2 T s2
(3.103) et ≥ = ,
t=1
Tξ ξ
110 Chapter 3. Dealer Markets

and by taking expectations we get with (3.98):

ξV ar[p]
(3.104) E[s2 ] ≤ .
T

If ξ is bounded21 we find the average squared spread to be independent of any

trade patterns. We can use (3.104) to derive a relationship between the average
spread and the average trading volume in a given period of time. As we have
assumed that within a given period of time T trades of a fixed size occur, a large
T implies a high trading volume. As can be seen from (3.104) and as ξt and
V ar[p] are independent of T we expect a small average spread.
With a large number of trades the information of informed investors is revealed
much faster to the market maker than with only few trades. This reduces the
adverse selection costs and hence the spread. The initial adverse selection costs
at the first trades, represented by V ar[p] increase the average spread.
As we saw in the Holden and Subrahmanyam (1992) model, trading will
be especially intense shortly after information has been released to informed
investors, in our case the early trades of the time period, causing adverse selection
costs to increase with γI in the first trades. The result obtained here can explain
why spreads in actively traded stocks are smaller than in less actively traded
stocks.
Krishnan (1992) showed that the Kyle (1985) and Glosten and Mil-
grom (1985) models are equivalent if we restrict the Kyle (1985) model slightly.
Although Kyle (1985) allows for no market makers and no spread, the prices
at which trades occur turn out to be identical if we allow only for fixed order
imbalances of the same size as an order size here. In this case the price deter-
mined by the match maker equals the bid and ask price of the market maker.
The properties of the prices are identical in both cases.
Although the model of Glosten and Milgrom (1985) model gave further
insights into the behavior of bid and ask prices and the spread, no explicit formula
PT
21
For ξ to be bounded T1 t=1 ξt has to converge, what is achieved if the market maker sets
prices always low enough such that both, P r(ot = 1|Ωt ) and P r(ot = −1|Ωt ) are bounded away
from zero, i.e. if he can assure a sufficient large trading of uninformed investors.
3.2. Information-Based Models of Market Making 111

could be derived directly, only the equivalence with the model of Kyle (1985)
and its further extensions, such formulas can easily be provided.

3.2.2 Simultaneous trading at different stock exchanges

The result that competitive market makers set prices equal to the expected fun-
damental value given the order flow is due to the assumption of Bertrand com-
petition between market makers. Although not explicitly stated in the models
presented thus far, it has been assumed that investors route their entire order
flow to a single market maker quoting the most favorable price. As market mak-
ers quoting less favorable prices do not participate in the order flow, they make
zero profits, ignoring fixed costs. Hence Bertrand competition requires market
makers to quote prices such that they make expected profits of zero, which has
been shown in the previous section to imply quotes that equal market maker’s
inference about the fundamental value.
In the models considered the assumption of strict price priority ensured the
entire order flow to be routed to a single market maker quoting the best price
for the entire order. As has been shown that the quoted prices become less
favorable the larger the order size is, it can be profitable for investors to split
their orders between market makers to reduce the order size submitted to each
market maker and therewith receive more favorable prices.22 For this reason also
market makers quoting not the most favorable price will receive a fraction of the
order flow, the less favorable the quotes are, the smaller this fraction will be. The
optimal splitting of the order flow will give an investor equal costs for trading
with every market maker.
Splitting the order flow requires that investors are able to route their orders
directly to different market makers, violating strict price priority. A possibility to
achieve such a situation is the assumption that an asset is traded at several stock
exchanges and that trading at these stock exchanges can take place simultane-
22
It is common to assume that market makers condition their quotes only on the order flow
they receive, but not on the order flow the other market makers receive, hence such a splitting
of the order flow would not influence the quoted prices.
112 Chapter 3. Dealer Markets

ously. Given this possibility to split the order flow Bernhardt and Hughson
(1997) show that competitive market makers are able to make positive expected
profits by quoting less competitive prices.
The intuition behind their result is that when quoting not the most favorable
price, market makers still will receive a fraction of the order flow, whereas if
investors are not allowed to split the order flow they would receive no orders.
This situation enables market makers, as formally shown in Bernhardt and
Hughson (1997), to quote less competitive prices, i.e. higher ask and lower bid
prices, and make positive expected profits. These less favorable prices increase
the trading costs for investors. They further show that prices become more
competitive by increasing the number of competing market makers, in the limiting
case, with an infinite number of competing market makers, the quoted prices are
competitive.23
Furthermore Dennert (1993) points out that as a result of splitting the
order flow the share of informed trades increases. Therewith adverse selection
costs increase as shown in the previous subsection, what increases the trading
costs further.
With these two effects due to the possibility to split the order flow, e.g. by
trading simultaneously at different stock exchanges, competition among market
makers not necessarily reduces trading costs as usually expected.

3.2.3 Comparing competitive and monopolistic market

makers

If the number of informed investors is increased, it has been shown above that
the adverse selection costs increase and therewith the fee charged by the market
maker. If we assume that uninformed investors demand responds to the fee
charged, their demand reduces further, the share of informed investors increases
and again the fee increases. If the adverse selection costs are high enough, the
23
Bernhardt and Hughson (1997) also show that a linear equilibrium as derived in Kyle
(1985) only exists with enabling investors to split the order flow, if the demand of liquidity
traders for trading are price elastic.
3.2. Information-Based Models of Market Making 113

demand of uninformed investors nearly vanishes and the market maker cannot
offset his losses from trading with the informed investors. He has to quote such
high fees that the market breaks down.

The market breakdown as the result of too much informed investors is due to
the need of the market maker to achieve zero expected profits. It has been shown
in Holden and Subrahmanyam (1992) that especially shortly after new in-
formation is available, informed investors trade very actively on this information.
Afterwards the information is revealed through prices, the trades of informed in-
vestors reduce and with it the adverse selection costs. However if at the beginning
the market breaks down, prices never will be able to reveal information and the
threat remains ad infinitum.

Glosten (1989) derives a model where, with further assumptions like a nor-
mally distributed liquidation value, conditions are derived under which the mar-
ket breaks down as the result of too high adverse selection costs with compet-
itively acting market makers. These results are in line with the argumentation
thus far and are therefore not presented here in more detail.

To overcome the problem of a market breakdown he proposes to establish a

monopolistic market maker. As such a market maker maximizes his total profits
over a given time period, he not necessarily has to avoid losses from every trade,
he can be compensated for losses by larger profits in future trades. Glosten
(1989) shows that the optimal price setting strategy of a risk neutral monopolistic
market maker is to set an average price that enables him to make extraordinary
profits in normal trading environments and compensates him for losses in times
of high adverse selection costs.

By incurring a loss shortly after new information has become available to in-
formed investors he keeps the market open and learns the information through
orders he receives. After having learnt this information sufficiently well, his ad-
verse selection costs are reduced and he can make larger profits compensating
him for the incurred loss. By shutting the market down he would not be able to
learn the information and hence would make no profits.
114 Chapter 3. Dealer Markets

As long as the adverse selection costs are higher than a certain threshold,
a monopolistic market maker quotes prices more favorable to investors than a
competitive market maker would be able to do. Only if adverse selection costs
are low, competing market makers charge a smaller fee. This result is different
to the result obtained for inventory-based models, where a monopolistic market
maker always quotes less favorable prices than competitive market makers.
We should therefore find competing market makers in markets with only small
adverse selection costs, i.e. only few informed investors, while for markets with
high adverse selection costs, i.e. many informed investors, a monopolistic market
maker would be preferred.
The NASDAQ has a system of competing market makers, while the NYSE
has a monopolistic market maker (specialist). The many analysts following the
companies at the NYSE make adverse selection costs much higher at the NYSE
than at the NASDAQ, where especially the information of the informed investors
are much less precise than at the NYSE, as the companies mostly work in a
more dynamic environment. This would be a rationale for the NYSE to have a
monopolistic market maker and the NASDAQ to have competing market makers.

3.2.4 Explaining return patterns

There exists significant evidence of systematic patterns in asset returns, as well

within a trading day as across trading days. The most prominent of this effect
is the Monday-effect, where the returns on Mondays are negative on average.
Admati and Pfleiderer (1989) present a model that is able to explain such
a pattern.
In section 2.3 the model of Admati and Pfleiderer (1988) has been pre-
sented, that showed the concentration of trading in a few periods. This pattern
in trading volume arises endogenously out of the model as uninformed investors
want to reduce their adverse selection costs from trading with informed investors.
Nevertheless such behavior was not able to explain patterns in returns.
By introducing a market maker we are able to explain such patterns in ex-
3.2. Information-Based Models of Market Making 115

pected returns. As the uninformed investors react sensitive to the fee charged by
the market maker, quoting a high fee reduces the trading of uninformed investors
and increases adverse selection costs. If the market maker quotes a low fee his
profits are small, but also his adverse selection costs are reduced by the increased
number of uninformed investors trading.

By quoting a very low fee on one side of the trade, e.g. the ask side, and a
high fee on the other side of the trade, in our example the bid side, the mar-
ket maker will induce many uninformed trades on the ask side and virtually a
breakdown of the market on the bid side. Although this price setting behavior
causes a large order imbalance, the ask price does not change significantly after
every trade as most trades are from uninformed investors. On the bid side nearly
no uninformed investors trade due to the high costs, nearly all trades come from
informed investors, hence an order submitted to the bid side is much more infor-
mative than at the ask side. The high fee charged and the information revelation
reduces the profits of informed investors from trading on their information. If
it is known that the market maker will reverse his strategy in the next period
of time, e.g. the next trading day, by quoting low fees at the bid and high fees
at the ask side, an informed investor will prefer to wait until this period as his
profits are increased, provided that his information is not revealed beforehand
by other means. The market maker will have to reverse his strategy in order to
offset his large inventory position he had to acquire as a result of the large order
imbalance in the first period.

With this quoting strategy the market maker does not only induce a pattern
in trading volume, but also a pattern in the types of trades that occur. In the
example above the market maker induces more trades at the ask in the first
and more trades at the bid in the second period. Admati and Pfleiderer
(1989) show that such a strategy is an equilibrium for a monopolistic market
maker as well as for competitive market makers. They furthermore show that
the concentration of trading at the bid and at the ask under certain assumptions
do not concentrate in the same period like in Admati and Pfleiderer (1988).
116 Chapter 3. Dealer Markets

If the trading at the ask concentrates in a period and the trading at the bid
in the next period, it is more likely to observe an ask price in the first and a bid
price in the second period. Therewith the expected return can easily be shown
to be negative between these two periods. If in the next period trading again
concentrates at the ask, the expected return will be positive.
The model presented here shows that patterns in expected returns can arise,
but not how they are timed, neither that they have to be timed equally by
all market makers. This timing has to be explained by exogenous behavior of
investors. The Monday-effect suggests with this model a concentration of buy
orders on Fridays and a concentration of sell orders on Mondays. As over the
weekend no trading is possible and there are many investment funds that have to
invest money newly acquired, they invest it on Fridays. Over the past years there
has been a massive flow into funds, such that they have to invest, i.e. buy. Their
need to buy can be justified by the frequent benchmarking with an index, as the
closing of an index is taken as their benchmark, they best can track the index
by trading near the close, i.e. the end of the day and the end of the week. This
causes a concentration of trading at the ask on Fridays, resulting in an expected
loss on Mondays as well as for the beginning of the next trading day, an effect
that also is confirmed empirically.

3.3 Dealer markets with multiple assets

We assume a market with L > 1 different risky assets. Each asset is assigned
to a single market maker, who is granted a monopoly to act as market maker
for this asset. This market structure can be found, e.g. at the NYSE, where
these market makers are called specialists. We allow each specialist to make the
market for more than a single asset, i.e. be specialist in more than one asset.24
We have K ≤ L specialists. The specialists are assumed to be risk neutral, i.e.
they face no inventory costs. It is further assumed that no private information is
24
Hagerty (1991) allows every specialist only to make the market in one asset, so that the
framework presented here is a generalization of her setting.
3.3. Dealer markets with multiple assets 117

in the market to avoid the problem of adverse selection costs. The aim of these
assumptions is to concentrate on the effect of execution of monopoly power in
this environment. In such a market the spread of competitive market makers
would be zero as we know from sections 3.1 and 3.2.

Following Gehrig and Jackson (1998) we have two groups of investors,

indexed i = {1, 2}. Non-optimal endowments in their portfolio holdings induce
them to trade, i.e. trading takes place as the result of a liquidity event. One
group buys an asset and the other sells it. Let the endowment of investors in
group i of asset j be denoted eij and ei = (ei1 , . . . , eiL ) is the vector of endowments.
The total supply is e = e1 + e2 .

Specialist k controls a subset Lk of the L assets. Specialists set prices such

that the market clears, i.e. demand equals supply. Such prices exist if we assume
the specialist to know the endowments of the investors and all other relevant
variables.

The risky assets are traded in a single round of trading, after the trade has
occurred they are liquidated at the fundamental value. The fundamental value
of asset j is denoted vej and has a mean a mean of µj and a covariance matrix Σ,
which we assume to be positive definite and non singular.

A risk averse investor will determine the demand such that he maximizes his
expected utility of terminal wealth, which is given by

L
X
fi = xij vej + pbj max(eij − xij , 0) + paj min(eij − xij , 0)) ,


(3.105) W
j=1

where xij denotes the demand of investor i for asset j, pbj the bid and paj the ask
price of asset j.The final terms denote the change in money holding as a result of
trading and the first term the return from holding the asset. With a competitive
market maker we have shown earlier that in this environment pbj = paj = pcj .
118 Chapter 3. Dealer Markets

Inserting this we get

L
X
fi = xij vej + pcj (eij − xij ) ,


(3.106) W
j=1
L
X
fi ] = xij µj + pcj (eij − xij = xi 0 µ + pc 0 (ei − xi ),


(3.107) E[W
j=1
L
X L
X L
X
(3.108) fi ] =
V ar[W (xij )2 Σjj + xij xik Σjk
j=1 j=1 k=1,k6=j
i 0 i
= x Σx .

The expected utility is then given by

(3.109) E[U (W f i ] − 1 zV ar[W

f i )] = U (E[W f i ]),
2
inserting and differentiating gives the first order condition for a maximum:

(3.110) µ − pc − zΣxi = 0.

Solving for the optimal demand we get

1
(3.111) xi = Σ−1 (µ − pc ).
z
The second order condition −zΣ < 0 is fulfilled with the assumption of a positive
definite covariance matrix. The market clearing condition x1 + x2 = e gives us
the competitive price as
1
(3.112) pc = µ − zΣe.
2
Let θi = xi −ei denote the vector of trades investor i conducts. If the specialist has
market power he is able to charge different prices for the two groups of investors,
denoted pi . Therewith (3.111) becomes
1
(3.113) xi = Σ−1 (µ − pi ).
z
Market clearing requires that θ1 + θ2 = 0:

(3.114) 0 = θ 1 + θ 2 = x1 + x2 − e
1 −1
= Σ (2µ − p1 − p2 ) − e.
z
3.3. Dealer markets with multiple assets 119

Solving (3.112) for e and inserting into (3.114) gives

1 −1 2 1
(3.115) Σ (2µ − p1 − p2 ) − Σ−1 (µ − pc ) = Σ−1 (2pc − p1 − p2 ) = 0,
z z z
what requires

(3.116) p1 − pc = pc − p2 .

The specialist as being risk neutral maximizes his total profits. With Rjl denoting
the revenues from asset j for specialist l we get with the market clearing condition
θ1 + θ2 = 0 and (3.116):

(3.117) Rjl = θj1 p1j + θj2 p2j = θj1 p1j − θj1 (2pcj − p1j ) = 2θj1 (p1j − pcj ).

The total revenues of specialist l are given by

X X X
(3.118) Rl = Rjl = 2 θj1 (p1j − pcj ) = 2 (x1j − e1j )(p1j − pcj ).
jLl jLl jLl

Differentiating with respect to p1j for all j ∈ Ll we get with (3.113) the first order
condition
1 −1
0 1 X −1 1
(3.119) Σ j (µ − p1 ) − e1j − Σ (p − pci ) = 0,
z z iL ij i
l

where (Σ−1 )j denotes that jth row of Σ−1 and Σ−1 −1

ij the (i, j)th element of Σ .

Defining

2Σ−1

ij if i = j or i, j ∈ Ll
(3.120) Aij =
Σ−1
ij else

we can rewrite (3.119) in vector form as

1 −1 1
(3.121) Σ (µ − pc ) − e1 − A(p1 − pc ) = 0,
z z
where from (3.111) and (3.113)
1
(3.122) µ − pc = zΣx1 = zΣ Σ−1 (µ − p1 ) = µ − p1 .
z
Solving (3.121) for p1 − pc we get

p1 − pc = A−1 Σ−1 (µ − pc ) − ze1 .

(3.123)
120 Chapter 3. Dealer Markets

As from (3.112) we get

2
(3.124) e = Σ−1 (µ − pc )
z
we find
 
1 c 1 −1 2 −1 c 1
(3.125) p −p = zA Σ (µ − p ) − 2e
2 z
1 −1
= zA (e − 2e1 )
2
1 −1 2
= zA (e − e1 ).
2
If an investor of group 1 buys the asset, i.e. e1j < e2j , we find that p1j > pcj > p2j
and we can interpret p1j as the ask price and p2j as the bid price, as this is the
price at which an investor of group 2 will sell the asset. If an investor of group 2
buys the asset the relations change. The spread is given by

(3.126) s = |p1 − p2 | = z|A−1 (e2 − e1 )| > 0.

The bid and ask prices are symmetric around pc as can be seen from (3.116),
because the market maker has the same market power on both sides of the trade.
The spread is increasing in the risk aversion of the investors, as with a higher
risk aversion demand reacts less elastic to price changes, because the risk of
holding a non-optimal portfolio has a larger influence on the expected utility than
trading costs. In the same manner differences in endowment, i.e. the portfolio
imbalance, increases the spread. The exact influence on the spread depends on
the covariances of the assets, that form A. If two assets are very similar, i.e.
their correlation is close to 1, the investor can trade the other asset instead and
receive only a small reduction in expected utility (substitution). This indirect
competition between assets (if they are not assigned to the same market maker)
forces the specialists to compete against each other. This reduces their market
power and hence the spread.
Using this result Gehrig and Jackson (1998) derive some further results by
analyzing the correlation structure of the assets and the resulting spreads in more
detail. It turns out that for assets with a high correlation of the fundamental
3.3. Dealer markets with multiple assets 121

value, spreads are lower if they are assigned to different market makers. This
is the result of the mentioned indirect competition. With a negative correlation
the assets become complements and they cannot be traded for each other, which
increases the market power of the specialists. Lower spreads result if such assets
are assigned to the same market maker.
Comparing the situation with two specialists each being assigned to one asset
and a single market maker being responsible for both assets, the endowments of
the investors have to be considered. If the two groups of investors are both well
engaged in the assets and only want to rebalance their portfolio, it turns out that
a single market maker quotes a lower spread for positively correlated assets, while
two specialists would give lower spreads for negatively correlated assets.
If the investors however want to change their overall engagement in the asset
market without changing the composition of their portfolios the relation exactly
reverses. For positively correlated assets two specialists would give lower spreads
and a single specialist for negatively correlated assets.
For the optimal allocation of the responsibilities of the specialists it is not
only important to take into account the correlation structure of the assets, but
also the motives of trading, portfolio rebalancing of well engaged investors or a
change in the engagement in the asset market.
It can be preferable to assign a monopoly in market making for several assets
to the same market maker in order to prevent indirect competition, while in other
situations this indirect competition reduces the spread.
We can summarize these findings by stating that if assets trade as substitutes
the spread is lower if the assets are assigned to different market makers, while for
assets trading as complements the reverse is true.
122 Chapter 3. Dealer Markets

Review questions
1. Why should the spread increase with the risk of the asset?

2. Why will the best quoted spread under competition always be larger than
the costs of market making?

3. What is the importance of the time horizon of a monopolistic market maker?

4. Why do uninformed traders always make losses in the presence of informed

traders?

5. Why does in the presence of asymmetric information the spread decrease

with trading volume?

6. Compare the model on adverse selection with Kyle (1985).

7. Why can market makers quote non-competitive prices when investors can
split their orders?

8. Under which conditions are monopolistic market makers preferable to com-

petitive market makers?

9. How can market makers induce clustering of trading?

10. When does allocating two assets to the same market maker result in a lower
spread than allocating it to two different market makers?

Application
Consider the same case as in the application of chapter 2 on page 72. After
the results of this initial investigation the market regulator sought to consider
additional evidence and compiled a graph with the average spread in each five-
minute interval, which is reproduced below.
Given the theories on dealer markets detailed above and the details of the case
presented in chapter 2, what can be said about the development of the spread
and its causes and how does this development fit into the analysis of the case
provided previously?
 Spread (in % of share price)

0.1
0.2
0.3
0.4
08:00 0.5
Application

10:00

12:00

14:00

16:00

Monday
09:00

11:00

13:00

Tuesday
15:00

08:00

10:00

12:00

14:00

Wednesday
16:00

09:00

11:00

13:00
Thursday 15:00

08:00

10:00

12:00
Friday

14:00

16:00
123
124 Chapter 3. Dealer Markets
Chapter 4

Limit Order Markets

———————————————————————————————————
In this chapter we will focus on the submission strategies of limit order traders.
Our emphasis will be to understand the trade-off between the price a trader
receives and the waiting time until the order is filled. The main aspects in this
chapter will be
• existence of a spread in competitive markets
• price properties in limit order trading
• dynamic limit order submission strategies
• the informational efficiency of limit order markets

Key readings:
Kalman J. Cohen, Steven F. Maier, Robert A. Schwartz and Daniel K. Whit-
comb: Transactions Costs, Order Placement Strategy, and Existence of the
Bid-Ask Spread, Journal of Political Economy, 89, 287-305, 1981

Christine A. Parlour: Price Dynamics in Limit Order Markets, Review of

Financial Studies, 11, 789-816, 1998

Thierry Foucault, Odean Kadan and Eugene Kandel: Limit order book as a
market for liquidity, Review of Financial Studies, 18, forthcoming, 2005
———————————————————————————————————
126 Chapter 4. Limit Order Markets

In all models presented so far investors were only allowed to submit market
orders to a match maker or market maker. The market maker has been the
sole provider of immediacy. As has already been mentioned in section 1.4 many
different order forms, beside market orders, exist, the most important being the
limit order.1 Limit orders can be viewed as an alternative to market makers for
providing immediacy. Like the quotes of a market maker it enables an investor to
trade at a fixed price, the limit price, with certainty. For such an investor there
is no difference between a limit order and the quotes of a market maker. We can
therefore interpret the quotes of a market maker also as a limit order.

Despite these similarities between market makers and limit orders, the con-
cepts of inventory costs and adverse selection costs, cannot easily be applied to
investors submitting limit orders. The most important difference between a mar-
ket maker and a limit order trader is that the market maker is obliged to quote
both bid and ask prices, while the limit order trader is free to submit either a limit
order, a market order or not to trade at all. He therefore will only submit a limit
order if this is the most profitable alternative for him. Despite these differences
their similarity for investors submitting market orders suggests that limit orders
are competitors to the quotes of market makers and that they will influence each
other. In this section it shall be investigated what the consequences are from the
introduction of limit orders.

Throughout this section we assume strict price priority not only between
quotes of market makers, but also between their quotes and limit orders and
within limit orders, i.e. an incoming market order is executed at the best available
price, a limit order or the quote of a market maker.2

1
The importance of limit orders even exceeds the of market orders. In the second half of 1993
62% of all orders submitted to the SuperDOT system of the NYSE were limit orders, further-
more the market maker has only been involved in 17% of all transactions. See Chakravarty
and Holden (1995, pp. 213 ff.).
2
Not all exchanges apply such strict rules. The NASDAQ goes even a step further to enhance
competition between limit orders and market makers in their Rule 2110 and interpretation IM-
2110-2 by forcing market makers to give limit orders priority at the same price as the quotes
of a market maker.
4.1. Static models of limit order placement 127

4.1 Static models of limit order placement

Despite their importance in trading, limit orders have only recently attracted
more attention. One of the few early exemptions are Cohen et al. (1981), who
investigate the optimal order placement decision of an investor and its implica-
tions for the spread.
They do not distinguish between limit orders and quotes of a market maker,
as only the decision of an investor is considered. It is assumed that investors are
free to submit an order at every time, the order can either be a market or a limit
order for a fixed size of trade. The arrival of market orders is assumed to follow
a Poisson process with order arrival rate λ, i.e. per period of time the expected
number of market orders equals λ. A limit order can be submitted at any price,
it cancels if the limit order has to execute a market order. The investor is then
free to submit a new order, either a limit or market order.3
If the limit order is not executed it remains valid for the whole period of time,
e.g. a trading day and cancels automatically afterwards. All limit orders are
assumed to be published, i.e. the limit order book is open. Investors see not only
the best prices available, but also all other prices that can occur.
We assume now an investor who wants to buy the asset.4 He can do so either
by submitting a market order that executes with certainty, i.e. probability one,
at the best available price pa . The alternative he has now is to submit a limit
bid order. When submitting a limit bid order, the execution of this order is not
guaranteed as it has to wait for a market sell order to arrive at the market. Before
such an order executes there must be no other limit bid order offering a more
favorable price. This uncertainty of execution imposes costs on the investor,
which earlier have already considered as waiting costs, reducing his expected
utility from trading by a limit order. But on the other hand he will be able to
trade at a more favorable price.

3
We can introduce a market maker to ensure at least one limit order on both sides of the
trades.
4
The same considerations can be made for an investor selling the asset.
128 Chapter 4. Limit Order Markets

It is obvious from previous sections, the higher he sets his limit bid price
pLb , the higher the probability of execution, φ. If he sets a limit bid price above
pa his order will execute with probability one as investors could make arbitrage
profits by selling at pLb and buying at the lower price pa . This negative spread
cannot happen as the investor would be able to buy the asset at a lower price by
submitting a market order. If pLb increases from below to pa the probability of
execution increases, but does not converge to one. This can easily be shown as
below.
The number of trades in a time period in which the limit order remains valid,
N (λ), is finite as long as λ < ∞. If the probability of an order arriving at the
market never equals zero, despite a very high fee charged, the probability that
all N (λ) trades are at the ask is strictly positive, hence the limit bid order will
never execute. It is not necessary that all trades have to occur at the ask. As
the change in the ask price upwards due to the many trades at the ask, causes
the market maker also to revise his bid quotes upwards as the result of inventory
costs or update of believes. This may result in bid quotes exceeding the limit
bid price even if it has initially been the best available bid price. In the same
manner other investors can place limit orders that offer a higher price. Hence the
probability of execution is always smaller than one, even if the limit bid is very
close to the ask, while for pLb ≥ pa the order executes with certainty. We have
found a probability jump in execution at p = pa :

(4.1) lim φ(pLb , λ) < 1.

pL
b →pa

If more orders arrive at the market, i.e. if λ is increased, the probability that the
order will not execute is reduced as can easily be seen from the above example.
For λ < λ0 we have for all pLb < pa :

(4.2) φ(pLb , λ) < φ(pLb , λ0 ).

When λ → ∞ the probability of non-execution of the limit order goes to zero,

i.e. for all pLb < pa we have

(4.3) lim φ(pLb , λ) = 1,

λ→∞
4.1. Static models of limit order placement 129

the jump in probability vanishes. A continuous trade, i.e. λ = ∞, requires the

investors to revise their portfolios at every instant of time. This is a reasonable
assumption as long as trading is costless, if trading imposes costs investors will
not revise their portfolios too often as this would exceed the benefits. Hence in
the presence of trading costs (other than the fee charged by the market maker),
like costs for submitting orders, we find that λ < ∞ and a probability jump can
be found at p = pa .

The investor would never submit a limit bid order such that pLb > pa as has
been stated above. The expected utility of submitting such an order would be
falling and its maximum would be at pLb = pa , i.e. by submitting a market order.

Assuming the utility function to be continuous, the expected utility will make
a jump downwards by lowering pLb below pa as a result of the probability jump in
execution. Lowering pLb further increases the profits from trading at the stated
prices, but also the probability of execution reduces as the fee charged increases.
Depending on the slope of the utility curves and the sensitivity of investors to
changes in the fee the expected utility can either increase or decrease, also a
maximum or minimum at some point is possible. In general no shape can be
predicted, Cohen et al. (1981) assume that the probability of execution at
first reduces only slowly, increasing expected utility, and then it decreases faster,
decreasing also expected utility. There exists some point where the expected
utility reaches its maximum.

If pLb reaches pb , i.e. the best available bid price, the expected utility jumps
again downwards, if submitting a limit order at the same price the order flow
has to shared, hence the probability of execution is reduced by its share in the
order flow. If the limit order is submitted at prices even lower as pb at first
the limit orders with a better bid price have to be executed before this order is
executed. By lowering the bid price further and further, an increasing number of
limit orders have to be executed before. Depending on the utility function, the
sensitivity of investors to changes in fees and the distribution of other limit bid
orders the expected utility may be falling or increasing.
130 Chapter 4. Limit Order Markets

Figure 4.1 presents an example how the expected utility may look like. In the
first panel the investor will choose to submit a limit bid order at pLb = p0b . The
spread is reduced from s to s0 . If the expected utility at pLb = pa is higher, as in
the second panel, the investor would submit a market buy order.5
The jump in expected utility at p = pa prevents the spread from converging
to zero with an increased number of limit orders submitted. There always exists
a price below pa such that the expected utility from submitting a market order
is higher, hence the spread will never sink below a certain level. The spread that
can be achieved depends on the size of the probability jump at pa . With a low
λ this jump will be larger, confirming the empirical finding that more frequently
traded assets have smaller spreads than less actively traded assets. For this result
no adverse selection costs are needed as in the previous models to explain this
result.
Although this analysis of Cohen et al. (1981) is very intuitive, it faces sev-
eral problems for the analysis of the behavior of spreads. For determination of the
probability of execution dynamic aspects have to be taken into account. In every
N trades new limit orders can be submitted, reducing the execution probability
of existing limit orders by offering a more favorable price. Also adverse selection
costs may become a severe problem if the limit order cannot be withdrawn. The
next section addresses some of these aspects.

4.2 Price dynamics with limit order trading

The order submission strategy does not only depend on the unexecuted limit
orders in the market, but also on the expectations of future order submissions
during the remaining trading rounds. Limit orders submitted at more favorable
prices will have priority and therefore reduce the probability of execution. But
also limit orders submitted at the other side of the trade or at less favorable
prices influence the execution probability as the submitted orders are not market
5
In all cases it has to be compared with the expected utility from submitting no order at
all. If the expected utility from submitting no order is higher, the investor will not submit an
order as he has no obligation to do so.
 4.2. Price dynamics with limit order trading 131

E[U (p)]6

....................................
........... ........
........ ......
....... .....
.
..
...... ....
....
. ....
..
... ....
.
... ....
. .
.. ....
.. ...
..... ...
..
.. ...
....
. ...
... ...
. ...
..
. ...
... ...
... ...
... ...
... ...
... ...
.. ...
s
. ...
... ...
... ... ......
.. ....
. ... ....
... ... ....
... ... ....
...
... ... ...
... ... ...
... ... ...
... ... ...
.. ... ...
c
.. ... ...
.
.
. ... ...
.. ...
.. ...
.. ...
..
..
c
.. ...
.. ...
...
...
...
...
...
s
...
...
...
...
...
...

c
.
..
..
....... ...
....... ...
........ ..
........
.........
..........
............ ....
...
...  s0 -
........................................
 s -
0
-
pb pLb = pb pa p

E[U (p)]6

....
...
s
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
c
...
......................... ...
.................... ...
.................. ...
............... ...
...............
............. ...
............. ...
............
........... ..
........... ..
........... ..
.......... ..
.......... ..
.......... ..
......... ..
......... ..
......... ..
......... ..
........ ..
........ ..
..
........ ..
........
.... c ..
..
..
..
..
..
..
..
..
s
..
..
..
..
..
..
..
..
.

c
..
..
..
....... ..
........ ...
.
........
........ ...
.........
.......... ....
............ ....
.......................................
 s = s0 -
0
-
p b = pL
b = pb pa p

Fig. 4.1: The placement of limit orders

132 Chapter 4. Limit Order Markets

orders that have to be executed against a limit order. Most models of limit order
trading presented in the literature thus far are static and therefore do not allow to
model these dynamic aspects. The first to model the order submission strategy
in a dynamic environment is the recent work by Parlour (1998), although
the dynamic aspects are incomplete as no equilibrium submission strategies are
considered. She models not only the order submission strategy, but also its
implications for the movement of prices, on which we concentrate here.
The time structure of the model consists of two periods. Every investor has
an initial endowment of the asset which is optimal for him. In the first period he
can choose to trade the asset at given bid and ask prices, pb and pa , respectively.
He can buy the asset and finance this by reducing his consumption in period 1,
C1 , by the price he is charged for the asset, or he can sell the asset and consume
the amount received. In period 2 the asset is liquidated at the fundamental value
p∗ , which is known to all market participants, hence there is no informational
asymmetry, adverse selection or even uncertainty about this value. All market
participants have to reverse their trades of period 1 by reducing their consumption
in period 2, C2 , by p∗ if they have sold the asset or to increase consumption by
the same amount if they have bought the asset before. Hence the asset is a mean
to delay or advance consumption between the two time periods. The trade size
is assumed to be fixed.
The investors are assumed to be risk neutral with a utility function

(4.4) U (C1 , C2 ) = C1 + βC2 ,

where β ≥ 0 is a parameter that denotes the preferences for consumption in

periods 1 and 2.6 For β < 1 consumption period 1 is preferred and for β > 1
consumption in period 2. The more β deviates from 1 the more urgent is the
need for trading in period 1 as we will see. If β ≈ 1 the investors are very patient
with trading. In another interpretation βC2 is the value the investor assigns to
consumption in period 2, as this consumption consists of the asset, it is the value
6
In terms of microeconomics it is the rate of substitution between consumption in these two
periods as ∂C
∂C2 = β.
1
4.2. Price dynamics with limit order trading 133

the investor assigns to the asset in period 1.

Investors differ only in their value for β, i.e. their time preferences or the value
they assign to the asset,7 they know their own β and they know the distribution
F of the β of the other investors.
Trading takes place only in period 1, there is a fixed number of T ≥ 1 trades
in this period. For every trade an investor is chosen randomly to arrive at the
market. Which investor, i.e. which β will be selected, depends on the distribution
function F .8 The chosen investor has to decide whether he wants to submit a
market buy, limit buy, limit sell, market sell order or not to trade at all.
If an investor submits a limit order he is not free to choose his limit price. The
limit price is assumed to be fixed at pa for a limit ask order and at pb for a limit
bid order, which are the current bid and ask prices. This strong assumption is
made to isolate the effect limit orders have on the behavior of prices. The absence
of any inventory or adverse selection costs prevent the bid and ask prices from
having any dynamics due to the behavior of a market or match maker. But it also
restricts the competition of limit orders through the price. The orders submitted
in this way are executed with arriving market orders according to time priority,
i.e. all orders that were submitted in earlier trading rounds will be executed first.
It is obvious that the more limit orders are already unexecuted in the limit order
book, the less likely is the execution of a newly submitted limit order. With a
finite number of trading rounds the probability of execution will always be lower
than one.
An investor submitting a market order receives the amount of pb in period
1 and has to repurchase the asset in period 2 at p∗ . When submitting a limit
sell order he receives the amount pa in period 1 and has to repurchase the asset
in period 2 also at p∗ , but only if his limit order executes, what happens with
probability π s . When submitting a market buy order he pays pa in period 1 and
7
It would not be correct to assign different beliefs on the value of the asset and that its
value revealed to the investors in period 2. The value is not a random variable as every investor
knows his own β and here therefore receives exactly the amount he expects.
8
We assume that there are much more investors than trading rounds, so that we do not have
to care about an investor to be chosen twice to trade.
134 Chapter 4. Limit Order Markets

receives p∗ in period 2. With a limit buy order he pays only pb in period 1 and
also receives p∗ in period 2, provided that his limit order executes, what has a
probability of π b . If the limit orders do not execute, he will receive no payment
and also has to make no payments, the same situation if he decides not to trade
at all. The expected utility is therewith given by
 ∗
 pbs − βp ∗
 market sell order
 π (pa − βp ) limit sell order


(4.5) E[U (C1 , C2 )] = 0 no order .
π b (−pb + βp∗ ) limit buy order




−pb + βp∗ market buy order


The optimal strategy is to maximize (4.5). Comparing market and limit sell
orders gives the condition for preferring a market sell order if pb − βp∗ > π s (pa −
βp∗ ) or

s pa pa − pb pb
(4.6) βM < ∗
− ∗ s
< ∗.
p p (1 − π ) p
s s
If βM turns out to be negative then we have to set βM = 0. Hence if β <
s
βM submitting a market sell order is preferred to submitting a limit sell order.
Comparing a limit sell order with not trading gives π s (pa − βp∗ ) > 0 or

pa
(4.7) βLs < ,
p∗

if β < βLs submitting a limit order will be preferred to not trading. The submission
of a limit buy order is preferred to not trading if π b (−pb + βp∗ ) > 0 or

pb
(4.8) βLb > .
p∗

If β < βLb then not to submit an order is preferred to submitting a limit buy
order. For preferring a limit buy to a market buy order we need π b (−pb + βp∗ ) >
−pa + βp∗ or

b pb πa − πb pa
(4.9) βM > ∗
+ ∗ b
> ∗.
p pi (1 − π ) p

The final comparison that has to be made is for comparing a limit sell and a limit
buy order as they are both preferred to the other alternatives for βLb < β < βLs .
4.2. Price dynamics with limit order trading 135

market sell limit sell limit buy market buy

 - - -
-
s
0 βM βLb βLbs 1 βLs s
βM β

Fig. 4.2: Order submission strategies

The condition that a limit sell order is preferred, π s (pa − βp∗ ) > π b (−pb + βp∗ ),
solves to
pa π b pa − pb
(4.10) βLbs < − .
p∗ π b + π s p∗
By the transitivity of preferences all alternatives can now be clearly ordered.
Figure 4.2 illustrates the order submission strategy. For extreme values of β the
investors prefer to submit market orders to ensure consumption in their preferred
period. If β is close enough to one the need to consume in the preferred period is
less important than the possible gain from receiving a more favorable price, hence
limit orders are submitted. In all cases the alternative not to trade is dominated.
The probability of execution of a limit order, π b and π s , plays a central role like
in Cohen et al. (1981). Not only the past order submissions, which are known
to all investors from the open order book, are of importance, but also the future
behavior of the investors. As the distribution of β is known, we could in general
use (4.7) - (4.10) to determine the probabilities for the different order types to
be submitted in the remaining trading rounds. Unfortunately the important
s b
parameters βM and βM depend themselves on these probabilities. Therefore
Parlour (1998) proposes an indirect approach to address this problem.
Let bB A
t and bt denote the number of unexecuted limit buy and sell orders in
s
the market. As the execution probability depends on these parameters, also βM
b
and βM will depend on them. Parlour (1998, pp. 809 ff.) gives a detailed proof
that
s
(4.11) βM (bB A s B A
t , bt ) ≥ βM (bt , bt − 1),

s
βM (bB A s B A
t , bt ) ≥ βM (bt + 1, bt ),

s
βM (bB A s B A
t , bt ) ≤ βM (bt + 1, bt + 1),
136 Chapter 4. Limit Order Markets

b
(4.12) βM (bB A b B A
t , bt ) ≤ βM (bt − 1, bt ),

b
βM (bB A b B A
t , bt ) ≤ βM (bt , bt + 1),

b
βM (bB A b B A
t , bt ) ≥ βM (bt + 1, bt − 1).

By noting that
s
∂βM
(4.13) ≤ 0,
∂π s
b
∂βM
≥ 0
∂π b

as can easily be seen by differentiating (4.6) and (4.9), we can relate the exe-
cution probabilities to the number of outstanding limit orders. Surprisingly the
execution probabilities depend on both sides of the limit order book and not only
on the side at which the order executes.
For observing a transaction at the ask price a market buy order has to be
submitted. A market buy order is submitted if the investor chosen has a β such
b B
that β > βM , which will be observed with probability 1 − F (βM ). With pt
denoting the price observed in trading round t and ot the order type at time t we
get

s
(4.14) P r(pt+1 = pa |ot ) = 1 − F (βM (bB A
t+1 , bt+1 )|ot ).

If in trading round t also a market order has been submitted, we find that bB
t+1 =

bB A A
t and bt+1 = bt − 1. With (4.11) we get

s
(4.15) βM (bB A s B A s B A
t+1 , bt+1 ) = βM (bt , bt − 1) ≤ βM (bt , bt ).

If the order has been a market sell order we have bB B A A

t+1 = bt − 1 and bt+1 = bt .

Again with (4.11) we get

s
(4.16) βM (bB A s B A s B A
t+1 , bt+1 ) = βM (bt − 1, bt ) ≥ βM (bt , bt ).

s
Obviously the expression βM (bB A
t+1 , bt+1 ) in equation (4.15) is smaller than in equa-

tion (4.16). By the monotonicity of the distribution function we have

s
(4.17) F (βM (bB A s B A
t+1 , bt+1 )|pt = pa ) ≤ F (βM (bt+1 , bt+1 )|pt = pb ).
4.3. Dynamic models of limit order submissions 137

Inserting into (4.14) gives the results that

(4.18) P r(pt+1 = pa |pt = pa ) ≥ P r(pt+1 = pa |pt = pb ).

(4.19) P r(pt+1 = pb |pt = pa ) ≤ P r(pt+1 = pb |pt = pb ),

It turns out that it is more likely to observe subsequent trades both at the bid or
at the ask rather than a change of the side of the trade. We now can calculate
the covariance of subsequent price changes. It will remain negative, but as the
probability that the price change equals zero is increased owing to (4.18) and
(4.19), we get
s2
(4.20) Cov(∆pt+1 , ∆pt ) ≥ − .
4

4.3 Dynamic models of limit order submissions

In order to obtain a fully dynamic equilibrium model of limit order submissions
we have to take into account that orders submitted later but at a more favorable
price are filled first. Hence any costs associated with such prolonged waiting has
to be taken into account when submitting an order. Foucault et al. (2005)
provide such a model which abstracts from the problem of a fundamental value
changing over time or asymmetric information. This allows them to focus purely
on the dynamics of the trading process itself.
Suppose risk neutral buyers and sellers arrive at the market alternating in
fixed time intervals of length ∆t.9 Each trader submits either a market or a limit
order upon arrival in the market. There are two groups of traders, patient and
impatient traders. The only difference between these two trader groups is the
waiting costs they face when their order is not filled immediately. Patient traders
face waiting costs of c ≥ 0 per time period and impatient traders of c0 > c. The
fraction of patient traders is fixed at θ.
Let each trader i have a valuation of the asset at Vi and the current bid and
ask prices prior to the submission of the order are denoted Pb and Pa , respectively.
9
Changing the requirement that buyers and seller arrive alternating at the market and
introducing stochastic time intervals between order submissions does not change the derived
results significantly.
138 Chapter 4. Limit Order Markets

With a tick size of d the price a buyer pays is Pbi = Pa − zi d and a seller receives
Pai = Pb + zi d, where zi denotes the spread expressed in the number of tick sizes
after he has submitted his order. The case of zi = 0 corresponds to a market
order and zi > 0 to the submission of a limit order, where we assume that limit
orders have to reduce the prevailing spread.10 Thus the expected profits for a
buyer and seller will be

(4.21) ΠB i
i (zi ) = Vi − Pb − ci t(zi ) = (V − Pa ) + zi d − ci t(zi ),

ΠB i
i (zi ) = Pa − Vi − ci t(zi ) = (Pb − Vi ) + zi d − ci t(zi ),

where t(zi ) denotes the expected time until the order is filled. As the current bid
and ask prices are given when submitting the order the trader seeks to maximize

(4.22) πi (zi ) = zi d − ci t(zi ).

The equilibrium obviously requires the expected waiting time for all possible
limit prices to be determined. While it is obvious that t(0) = 0 from the fact
that market orders are filled immediately, for zi > 0 this will be more difficult to
determine.
Suppose that the trader submits a limit order when the spread is s. We
are showing later that under certain conditions the trader will only submit limit
orders that improve the spread, i.e. zi < s. If the subsequent order to a limit
order is a market order the spread reverts back from zi to s, while for another
limit order the spread becomes zj < zi . This procedure continues until we have
zk = 1, in which case the following order has to be a market order. Hence once
an order is cleared the spread will return to what it has been before the order
was submitted.
Thus the expected waiting time for an order submitted at a spread of zi will be
the time until a market order arrives multiplied the probability of this happening
and the expected time it takes to clear any better limit orders submitted later
with a lower limit price. Let pz0 (z) denote the probability that, given a spread of
10
Below we will derive conditions for which this assumption is an equilibrium outcome.
4.3. Dynamic models of limit order submissions 139

z, an order with a spread of z 0 is submitted. In this case we can write

z−1
X
(4.23) t(z) = p0 (z)∆t + pz0 (z) (∆t + t(z 0 ) + t(z)) ,
z 0 =1

and furthermore t(1) = ∆t. The first term denotes the time for a market order to
arrive and the second term for better limit orders to be cleared. In the brackets
the first term accounts for the time until the next order arrives in the market,
second term is the time this subsequent order takes to be filled, after which we
face the original problem again. As we know that zz−1
P
0 =0 pz 0 (z) = 1, we find that

z−1
X
t(z) = p0 (z)∆t + pz0 (z)t(z 0 ) + t(z) (1 − p0 (z)) ,
z 0 =1

which easily solves for

Pz−1
∆t + z 0 =1pz0 (z)t(z 0 )
(4.24) t(z) = .
p0 (z)

Let us define the equilibrium spreads such that s1 < s2 < . . . < sN , where
ck t(s1 ) ck ∆t
s1 = d
= d
as the lowest spread giving a positive profit for a limit order as
defined in (4.22) and noting that for s1 we only observe the submission of limit
limit orders by patient traders. Assume finally that we never observe a spread
exceeding sN .
We can indeed show that impatient traders always submit market orders while
patient traders submit limit orders if s ≥ s2 . Thus the probability of observing
a market order is 1 − θ, hence p0 (z) = 1 − θ. As a patient trader will submit a
limit order such that the new spread is just one equilibrium spread smaller than
the existing spread, we observe that psi−1 (si ) = θ and for any k > 1 we have
psi−k (si ) = 0. We can therefore rewrite (4.24) as

∆t + θt(si−1 )
(4.25) t(si ) = .
1−θ
∆t+θt(si )
Shifting by one spread we obtain that t(si+1 ) = 1−θ
, which yields

θ
(4.26) t(si+1 ) − t(si ) = (t(si ) − t(si−1 )) .
1−θ
140 Chapter 4. Limit Order Markets

1+θ
As t(s1 ) = ∆t and from (4.24) we have t(s2 ) = ∆t 1−θ we get from iterating (4.26)
that
N −1  k !
X θ
(4.27) t(si ) = ∆t 1 + 2 .
k=1
1−θ

Having now established the expected waiting times for each equilibrium spread,
it remains to be determined what these equilibrium spreads are. Suppose the
spread is currently si+1 , then the limit order submitted by a patient trader is si
rather then si−1 , thus it is πi (si ) ≥ πi (si−1 ), which after inserting from (4.22)
becomes

c
(4.28) si − si−1 ≥ (t(si ) − t(si−1 )) .
d

The trader also does not submit an order si − 1 when the spread is si , but prefers
si−1 , unless the two are identical, thus πi (si−1 ) ≥ πi (si − 1) or equivalently

c
(4.29) si − si−1 ≤ (t(si ) − t(si − 1)) + 1.
d

In equilibrium it must thus be either t(si − 1) = t(si ) or t(si − 1) = t(si−1 ) as

these are the only two equilibrium spreads. As furthermore we see from (4.27)
that t(si ) > t(si−1 ), we can rewrite (4.29) as

c
(4.30) si − si−1 < (t(si ) − t(si−1 )) + 1.
d

Combining this expression with (4.28) we see that the spread improvement ∆si
between equilibrium spreads is given by
&  i−1 '
l cm θ c
(4.31) ∆si = si − si−1 = (t(si ) − t(si−1 )) = 2 ∆t ,
d 1−θ d

where the last equality follows from (4.27) and we note that the improvement
must be in whole ticks, indicated by the ceiling function d·e. Noting that s1 =
c 
d
∆t , we obviously obtain that
n
X
(4.32) si = s1 + ∆sn .
k=2
4.3. Dynamic models of limit order submissions 141

It is worth noting at this stage that the spread improvements can be larger than
a single tick, i.e. ∆si > 1 for sufficiently large waiting costs, low tick sizes or
infrequent trading. In all these cases the waiting costs outweigh the less favorable
price received. Finally we note that the spread improvement increases the larger
the spread is if θ > 12 , i.e. we have many patient traders. Biais et al. (1995)
confirm the existence of limit orders improving the spread by more than a single
tick.
The spread improvement can also be used to evaluate the resiliency of the
market. Suppose a liquidity shock reaches the market causing a large number
of market orders to be submitted. Naturally the spread widens as limit orders
get filled and are not replaced. The market will return to the original spread
only after the same number of patient limit order traders arrive at the market.
The spread prevalent at any point of time until the order book is refilled to its
previous state will thus be larger and by how much the spread is increased will
depend on the spread improvement. Thus the larger the spread improvement the
quicker the market spread will return to its previous level.
We are now in a position to show that all limit orders are spread improving.
Suppose that a trader could also submit limit order at the current spread level
and let us denote the expected waiting time by t0 . Thus our assertion requires
that

(4.33) si−1 d − ct(si−1 ) ≥ si d − ct0 (si ),

(si − si−1 )d ≤ c (t0 (si ) − t(si−1 )) .

With the limit order being submitted later we have by time priority that the first
order at that price has to be filled first, thus t0 (si ) > t(si ). After this order being
filled there are at least two more orders that have to be filled before the order
finally can be filled. The next trade will be on the same side of the trade and only
a market order does not increase the order book and thus the waiting time. After
that at least another market order is required to fill the order. We have thereby
an additional expected waiting time of at least 2(1 − θ)∆t. In case of a limit
142 Chapter 4. Limit Order Markets

order being submitted subsequently, at least an additional two more trades are
required and the expected waiting time increases by at least 4 (1 − (1 − θ)2 ) ∆t.
Hence

t0 (si ) ≥ t(si ) + 2(1 − θ)∆t + 4 1 − (1 − θ)2 ∆t

(4.34)

= t(si ) + 2 (1 + θ(2 − θ)) ∆t.

Inserting this result into (4.33) we obtain that

(4.35) (si − si−1 )d − c (t(si ) − t(si−1 )) ≤ 2c∆t (1 + θ(2 − θ)) .

Noting the expression for ∆si = si − si−1 from (2.11) we see immediately that

(4.36) d ≤ 2c∆t (1 + θ(2 − θ)) .

Thus for a sufficiently small tick size we always observe spread improving limit
orders. But we also see that for very active markets, i.e. a small ∆t, this is
unlikely to be fulfilled and we would observe queuing of limit orders at the inside
spread as often observed in real markets. A small tick size allows to avoid queuing
at low costs given that the required spread improvement can be very small indeed.
We can now finally investigate the time between two orders being filled, i.e.
the time between two trades taking place. Let the first order being filled have
a spread of si . We would thus require to know the expected number of traders
arriving in the market until the next trade takes place. The next trade takes
place only once a market order has been submitted, thus an impatient trader
arrives at the market. The probability to observe m patient traders is given by

(4.37) P rob(Ni = m) = θm−1 (1 − θ),

thus the expected value is

i
X 1 − θi+1
(4.38) E[Ni ] = mθm−1 (1 − θ) + (i + 1)θi = ,
m=1
1−θ

where a maximum of i limit orders are possible before the spread is so small that
all investors submit a market order, which is captured by the final term. Thus
4.3. Dynamic models of limit order submissions 143

the duration becomes

1 − θi+1
(4.39) Di = ∆t.
1−θ
We see that the duration increases with the spread and the fraction of patient
traders, besides the trading activity.
As we know which spreads are to be found in equilibrium, it would be of
relevance to establish in which frequency we find which spread and finally to
determine the average spread, average waiting time and duration to analyze the
properties of the market further.
Suppose the current spread is si , then the following two scenarios can happen.
A market order arrives with probability 1 − θ and the spread increases to si+1
or with probability θ a new limit order is submitted and the spread reduces to
si−1 . If i = 1 only market orders can be submitted and if i = N the spread does
not increase further with a market order but remains constant. The transition
matrix becomes
 
1 0
0 0 ··· 0 0 0

 0 1−θ
θ 0 ··· 0 0 
 0

 θ 0
0 1−θ ··· 0 0 
 0
 0 0
θ 0 ··· 0 0  0
(4.40) Λ= .
 
.. ..
.. .. .. .. .. ..
 . .
. . . . .  .
 
 0 0
 0 0 ··· 0 1 − θ 0 
 0 0 0 0 ··· θ 0 1−θ 
0 0 0 0 ··· 0 θ 1−θ
With the probability of observing a spread of si denoted ui and its vector u,
a stable equilibrium requires that after a trade the probabilities do not change,
thus Λu = u. With the usual restriction that N
P
i=1 ui = 1 we obtain that

θN −1
(4.41) u1 = PN ,
θN −1 + j=1 θN −j (1 − θ)j−2
N −i
θ (1 − θ)i−2
ui = .
θN −1 + N
P N −j (1 − θ)j−2
j=1 θ
θ

j−i u1 1
We see that ui = uj 1−θ
for i, j ≥ 2 and u2 = θ
. For θ = 2
we obtain a
uniform distribution of spreads across the range, apart from the smallest equi-
librium spread. For θ < 21 , i.e. a market dominated by impatient traders, the
144 Chapter 4. Limit Order Markets

Fig. 4.3: Influence of the trader composition and trade frequency on the spread

probability of high spreads is increasing due to the liquidity demands of these

1
traders while for θ > 2
the dominance of patient traders leads to less likely large
spreads as they liquidity supply outstrips liquidity demand.

Rather than using conditional spreads, durations and waiting times which
are difficult to assess empirically , it is straightforward to evaluate the expected
spread, duration and waiting times. Figures 4.3-4.11 show how these properties
are affected by selected variables. We observe typically two different behaviors
1
for θ < 2
and θ > 12 , i.e. markets dominated by patient and impatient traders,
respectively.
1
Firstly we observe that for θ < 2
the average spread is not really affected by
the trader composition unless the fraction of impatient traders is very close to
1
2
. The impatient traders begin then to become important and quickly fill the
submitted limit orders, thus waiting times are becoming quite short too. With the
waiting costs thus being small, the spreads set by limit order traders, i.e. patient
traders, can become quite small. As their fraction of the market increases, the
4.3. Dynamic models of limit order submissions 145

Fig. 4.4: Influence of the trader composition and trade frequency on the waiting
time

Fig. 4.5: Influence of the trader composition and trade frequency on the dura-
tion
146 Chapter 4. Limit Order Markets

Fig. 4.6: Influence of the tick size and waiting costs on the spread with θ = 0.7

increased competition reduces the spread which is balanced against the reduced
number of impatient traders. If however the time interval between trader arrivals
increases , the total waiting costs increase. This induces patient traders to seek
larger benefits and increasing the spread at which they post consequently. As
impatient traders are not much affected by this behavior, orders are still filled
within the same time as before. Obviously the time between trades, the duration,
is increasing in the time between trader arrivals and with more patient traders
as market orders become less common.
1
For θ > 2
the picture is however, very different. The increased number
of patient traders submitting limit orders reduces the spread and consequently
the time to fill orders as more market orders are submitted once the spread is
reduced sufficiently. Only once the fraction of patient traders becomes very large,
the lack of market orders becomes relevant and the time to fill an order increases.
The increased total waiting costs require a larger incentive for patient traders to
submit limit orders, thus increasing the expected spread. As these two effects
4.3. Dynamic models of limit order submissions 147

Fig. 4.7: Influence of the tick size and waiting costs on the waiting time with
θ = 0.7

Fig. 4.8: Influence of the tick size and waiting costs on the duration with θ = 0.7
148 Chapter 4. Limit Order Markets

Fig. 4.9: Influence of the tick size and waiting costs on the spread with θ = 0.3

Fig. 4.10: Influence of the tick size and waiting costs on the waiting time with
θ = 0.3
4.3. Dynamic models of limit order submissions 149

Fig. 4.11: Influence of the tick size and waiting costs on the duration with θ =
0.3

balance, the duration between trades remains largely unaffected.

We can finally observe that the waiting costs and tick size are of limited
relevance in market dominated by impatient traders as their considerations are
irrelevant with respect to these variables. Only in markets dominated by patient
traders do the costs of undercutting the spread become relevant and increase the
spread while having no big effect on the duration. The increased spread then
makes the waiting time until an order is filled longer.

Empirical work testing this or similar models have not yet been conducted,
but the results reported in Biais et al. (1997), Cho and Nelling (2000),
Lo et al. (2004) and Hollifield et al. (2004) all find that a limit order
submitted at a larger spread takes longer to fill, in accordance with our results
and Lo et al. (2004) also report that increased trading reduces this time for
which the above model provides weak evidence.

The model presented here provides a first stepping stone towards a fully dy-
150 Chapter 4. Limit Order Markets

namic model of limit order submissions. Clear drawbacks in this model are the
absence of asymmetric information or any other form of uncertainty about the
fundamental value. Also the requirement that traders do not queue was excluded
in condition (4.30), but we see immediately that for even modestly liquid assets
the spread has to be unrealistically small. As we observe such queuing in reality,
it is save to assume that this condition is not fulfilled. Rosu (2003) provides an
alternative model to explain this observation while Luckock (2001) does not
account for endogenous demand.

4.4 Informational efficiency with limit order

trading
A formal model of the informational efficiency of prices allowing for limit order
trading has been proposed by Brown and Zhang (1997). As their model re-
turns to the batch framework of chapter 2 we do not derive their results explicitly,
but give the intuition behind their findings.
If only market orders are allowed to be submitted to the market, informed
investors face the risk that due to a large imbalance in the orders of uninformed
investors the price is less favorable than expected, they may even make losses.
Hence informed investors will trade much less aggressive and prices do not reveal
so much information.
If informed investors and risk averse hedgers are allowed to submit limit orders
to protect themselves from a too unfavorable shift in prices, both groups will trade
more actively. As informed investors can rule out losses from trading, uninformed
investors still face the risk of making a loss as they do not have access to the same
information. Therefore informed investors will trade much more aggressively than
uninformed investors to exploit their informational advantage.
If all informed investors had perfect knowledge of the liquidation value as
assumed in section 2, the match maker would observe a bundling of limit orders
at this value and would be able to deduct the true value with certainty from
observing the limit order flow and by quoting this price the match maker would
4.4. Informational efficiency with limit order trading 151

hinder the informed investors from making profits. As there are no possible profits
from being informed, there are no incentives to become informed, hence with the
results of Grossman and Stiglitz (1980) no equilibrium would exist.
To avoid such a situation it is assumed that the informed investors do not
have perfect knowledge of the fundamental value, but observe only a noisy signal.
These different observations of the information results in different limit prices and
the match maker cannot directly deduct the fundamental value of the asset.
The more aggressive trades of informed investors, nevertheless, make it easier
to deduct the fundamental value of the asset and with the possibility to submit
limit orders, prices become more efficient than if only market orders were allowed
to be submitted.
152 Chapter 4. Limit Order Markets

Review questions
1. Why does the spread in limit order markets never collapse?

2. What aspects of the limit order price do traders have to consider when
making their decision?

3. How do investors decide whether to submit a market or a limit order?

4. What is the impact of a more filled order book on the order submission
strategy?

5. What are the properties of transaction prices in limit order markets?

6. Why is the average spread very low when the market consists of many
patient traders?

7. Explain the influence of the tick size and the spread and waiting time.

8. Why is the fraction of informed traders so important for the properties of

the market?

9. Why do limit order potentially reveal more information than market orders?

10. Why can markets never become fully informationally efficient?

Application
In light of the theories presented in this chapter, re-examine your analysis of
the spread in the applications presented in chapters 2 and 3 on pages 72 and
122. Is there an alternative explanation for the observed pattern and can the two
explanations be reconciled?
Chapter 5

Empirical investigations of market

microstructure models

———————————————————————————————————
This chapter will provide a short introduction to empirical methods in market
microstructure research. The main topics of this chapter will comprise of
• estimating the spread components
• estimating informed trading
• estimating the impact on daily returns

Key readings:
Hans R. Stoll: Inferring the Components of the Bid-Ask spread: Theory and
Empirical Tests, Journal of Finance, 44, 115-134, 1989

David Easley, Nicholas M. Kiefer, Maureen O’Hara and John B. Paperman:

Liquidity, information and infrequently traded stocks, Journal of Finance, 51,
1405-1436, 1996.

Andreas Krause: Inventory effects on daily returns in financial markets, Inter-

national Journal of Theoretical and Applied Finance, 6, 739-765, 2003
———————————————————————————————————
154 Chapter 5. Empirical investigations of market microstructure models

We thus far considered mostly theoretical results of the models investigated,

besides the application of the results to explain some empirical observations. In
this chapter we want to establish how important market microstructure effects
are empirically. For this reason we firstly consider ways to estimate the size
of inventory and adverse selection costs in dealer markets and informed trading
before then continuing to investigate the implications for daily returns.

5.1 Estimating spread components

In the preceding chapters we identified two components of the spread in dealer
markets, s: inventory costs, cI , and adverse selection costs, cA . Thus far ignored
have been order processing costs, cO , which arise from maintaining the infrastruc-
ture of a market maker, e.g. his computer system and staff, which are mostly
fixed costs, as well as variable costs arising from conducting a trade, like fees
charged by the exchange. We assume that all costs include normal profits as has
been determined previously. We then find a final component of the spread, which
is mostly neglected in the literature, excess profits, cπ .
The spread can therewith be written as the sum of its components:

(5.1) s = cI + cA + cO + cπ .

In this section we will briefly review some contributions on the estimation of

these components. No investigation addresses excess profits, for which reason we
neglect this spread component in the remainder.1
At first we will briefly consider the estimation technique as introduced by
Stoll (1989) and then show the results of several empirical investigations.
There exists a large variety of estimation techniques for the components of
the spread. Many of these techniques concentrate on the determination of the
1
Although the literature usually uses the described techniques to estimate the spread com-
ponents in dealer markets, the liquidity suppliers, i.e. limit order traders, will face similar costs
and we can reasonably conclude that the results will also hold in auction markets without a
market maker.
5.1. Estimating spread components 155

adverse selection component and do not further distinguish between inventory and
order processing costs. We will here concentrate on the approach developed by
Stoll (1989), which explicitly uses the results of market microstructure models
as described before.
We will at first investigate the properties of prices if only one of the com-
ponents is present and finally combine these properties into a single framework
used for estimation. The model of Stoll (1989) requires to observe every single
trade in an asset to conduct the estimation.2 We assume for simplicity that the
first observed trade has taken place at the bid, results for the first trade at the
ask can be derived in the same manner.
At first we investigate the case that only order processing costs are present,
i.e. s = cO , which are supposed to be constant. With the fundamental value not
changing over time the bid and ask price are the same throughout the sampling
period, hence the price change is either zero if the next trade is also at the bid
or s if the next trade is at the ask. We assume furthermore that prices are set
such that the market clears on average, hence the probability for a trade at the
bid and at the ask are both .5. We then have for the change of the transaction
pt = pet − pet−1 :
price at time t, ∆e

s with probability .5
(5.2) ∆e
pt = .
0 with probability .5

When only inventory costs are present, i.e. s = cI , we know from the inventory
based models of market making that after a trade at the bid both, the bid and
ask prices, decrease to adjust for the larger inventory, while the spread is held
constant. The linearity of inventory costs in inventory and the symmetry of the
price changes implies the price to fall by .5s. To see this, notice that the spread
compensates the market maker for his costs to trade at the bid and at the ask.
With the symmetry the costs of trading on a single side of the market are .5s.
2
As such data have only been available recently, early contributions, like Roll (1984),
developed models using daily data to estimate the components. A major problem many of
them address is to determine the spread as quoted by market makers, as such data also have
not readily been available. We do not consider these models here in more detail as it has become
a standard nowadays to use data on transaction basis.
156 Chapter 5. Empirical investigations of market microstructure models

For the next transaction these costs do not change relative to the new inventory
because of the linearity, hence costs also change by .5s and therewith prices. The
probabilities for a trade at the bid and ask, respectively, are no longer equal. The
market maker decreases the ask price and hence the costs for investors of trading
at the ask, while he increases these costs for a trade at the bid, the probability
of receiving such a trade reduces, hence

.5s with probability .5 < γ < 1
(5.3) ∆e
pt = .
−.5s with probability 0 < 1 − γ < .5

A similar argumentation we can use in the presence of solely adverse selection

costs, i.e. s = cA . As we know from information based models of market making
prices are set such that they are the expected fundamental value in case a trans-
action takes place at this side of the market. Assuming adverse selection costs do
not to change over time and therewith a constant spread is applied, prices also
change by .5s. When assuming the market to clear on average we find

.5s with probability .5
(5.4) ∆ept = .
−.5s with probability .5

Let us now define a δ ∈ [0, 1] such that (1 − δ)s measures the price change if the
side of the trade changes in subsequent trades, −δs then is the price change if
the side does not change. Let further γ ∈ [0, 1] measure the probability that the
side of the trade changes. Using this notation we can rewrite equations (5.2) -
(5.4) as

(1 − δ)s with probability γ
(5.5) ∆e
pt = .
−δs with probability 1 − γ

The sequence of possible transaction prices together with their transition prob-
a
abilities are shown in figure 5.1 for convenience, where the superscripts and
b
, respectively, denote the ask and bid price. Table 5.1 exhibits the parameter
constellations for the different spread components as identified above.
The expected price change between two transactions with the initial trade
being at the bid is given by

(5.6) pt ] = (1 − δ)sγ + (−δs)(1 − γ) = (γ − δ)s.

E[∆e
5.1. Estimating spread components 157

pa2 = pa1 + δs
1−γ


pa1 = pb0 + (1 − δ)s

 @
@
γ γ@
@
pb0 R pb
@
0
@ 
@
@1 − γ γ
@
R
@ pb1 = pb0 − δs
@
@
1 − γ@
@
R pb = pb − δs
@
2 1

Fig. 5.1: Sequences of transaction prices with the initial price at the bid

Spread component δ γ

1
Order processing costs 0 2

1 1
Adverse selection costs 2 2

1 1
Inventory costs 2 2
<γ<1

Tab. 5.1: Parameter constellations for the spread components

158 Chapter 5. Empirical investigations of market microstructure models

Due to the symmetry, the expected price change with the initial trade at the
ask is given by −(γ − δ)s as easily can be shown. We can now define the realized
spread as the gross revenue made by the market maker from a round trip, i.e. a
trade at the bid followed by a trade ask and vice versa. In this case the market
maker has the same inventory position and belief on the fundamental value of
the asset before and after these two trades have taken place. The realized spread
therewith can be calculated as the expected price change after a trade at the bid
has taken place, minus the expected price change after a trade at the ask has
taken place, hence we find the realized spread, sr , to be

(5.7) sr = 2(γ − δ)s.

As we see from table 5.1 the realized spread is zero with only adverse selection
costs being present, while in the presence of order processing and inventory costs
we find 0 < sr ≤ s. Hence we can use the realized spread to determine the part of
the spread due to adverse selection costs. However, at first we have to determine
the parameters γ and δ.
If the probability of the initial trade to take place at the bid is as likely as
taking place at the ask, we see that the unconditional expected price change is
zero. We then can determine the first order autocovariance of transaction prices
with help of the sequences denoted in figure 5.1 to be

(5.8) ρT = Cov[∆e
pt , ∆e
pt−1 ]

= E[∆e pt ∆e
pt−1 ]
1
= [(1 − δ)sγδs(1 − γ) + (1 − δ)sγ(−(1 − δ))sγ
2
+(1 − γ)(−δs)(1 − γ)(−δs) + (1 − γ)(−δs)(1 − δ)sγ]
1
+ [(1 − δ)sγδs(1 − γ) + (1 − δ)sγ(−(1 − δ))sγ
2
+(1 − γ)(−δs)(1 − γ)(−δs) + (1 − γ)(−δs)(1 − δ)sγ]

= δ (1 − γ)2 − γ 2 (1 − δ)2 s2
 2 

= δ 2 (1 − 2γ) − γ 2 (1 − 2δ) s2 .
 

In the same manner we can derive the first order serial covariance of the bid
5.1. Estimating spread components 159

price:3

(5.9) pbt , ∆e
ρQ = Cov[∆e pbt−1 ]

= E[∆e pbt−1 ]
pbt ∆e
1
= [−δs(1 − γ)(−δs)(1 − γ) + (−δs)(1 − γ)(1 − δ)sγ]
2
1
+ [δs(1 − δ)s(1 − γ) + δsγ(−δs)γ]
2
= δ (1 − γ)2 − δ 2 γ 2 s2
 2 

= δ 2 (1 − 2γ)s2 .

We can now estimate these covariances from the data sample and determine the
parameters γ and δ accordingly. The nonlinearities implied by (5.8) and (5.9)
cause severe biases in small sample sizes as reported by Brooks and Masson
(1996). Therefore large sample sizes are needed for conducting this estimation.
Having estimated the parameters γ and δ we can finally determine the spread
components. The share of the adverse selection component is given by4

s − sr
(5.10) sA = = 1 − 2(γ − δ).
s

Using the realized spread we can determine the inventory cost component with
the results from table 5.1 by inserting the relevant parameter values into (5.7) as

(5.11) sI = 2γ − 1.

The remainder is the order processing component:

(5.12) sO = 1 − 2δ.

Besides these estimations several other techniques are used in empirical investi-
gations, in many cases modifications of this estimator. A common approach in
most techniques is to decompose the spread into different components and use
3
For the ask price it can be shown to yield the same result such that it is of no interest
which quoted price is observed.
4
We here directly define the share a cost component has at the spread as the spread com-
ponent. We could instead also define the component in absolute values without changing the
meaning of the results.
160 Chapter 5. Empirical investigations of market microstructure models

Authors Exchange Period sA sI sO

Glosten and Harris (1988) NYSE Dec 1981 .35 .65
Stoll (1989) NASDAQ/NMS Oct-Dec 1984 .43 .10 .47
George et al. (1991) NASDAQ 1983-1987 .09 .91
Affleck-Graves et al.
NYSE/AMEX Mar/Apr 1985 .59 .29 .12
(1994)
Affleck-Graves et al.
NASDAQ Apr 1985 .35 .24 .41
(1994)
Schmidt and Treske (1996) Frankfurt Feb-Sep 1995 .22 .55 .23
Porter and Weaver (1996) NYSE 1990 .02 .98
Porter and Weaver (1996) AMEX 1990 .16 .84
Porter and Weaver (1996) NASDAQ 1990 -.01 1.01
Huang and Stoll (1997) DJIA 1992 .10 .29 .61

Tab. 5.2: Estimates of the spread components

the properties of each component as a reference point to determine the influence

of this component on the entire spread. We will present the results of several
empirical investigations in the next section.
A large number of empirical investigations are reported in the literature ad-
dressing the composition of the spread, some are reported in table 5.2. The
evidence shows large differences between the authors, an observation that cannot
only be attributed to the investigation of different assets in different periods of
time. Several authors, e.g. Huang and Stoll (1997), use different estimation
techniques for the same data set and derive results that differ substantially. Neal
and Wheatley (1995) compare two widely applied estimation techniques, those
of Glosten and Harris (1988) and George et al. (1991) for various stocks
and mutual funds and show large differences in the estimates of adverse selection
components, with the estimates of Glosten and Harris (1988) being smaller
throughout.
We will therefore not interpret the results found in the literature in too much
detail. Despite all problems one faces when estimating the spread components,
it can be seen that adverse selection components form a considerable part of
the spread, estimates for many investigations, including several not included in
the table, ranging from about a third to a half of the observed spread. Other
investigations, e.g. Hansch et al. (1998) and Snell and Tonks (1999) find
significant evidence of inventory costs, although they do not estimate the share
5.2. Estimating informed trading 161

of the spread to be attributed to these costs.

In chapters 2.3 and 3.2.1 we showed that a larger share of informed investors
increases trading volume and that adverse selection costs increase. We should
therefore expect to find a positive relation between trading volume and adverse
selection costs. Indeed, Glosten and Harris (1988) report the adverse selec-
tion costs and therewith the entire spread, to be increasing in trading volume. A
similar result is reported by Lin et al. (1995) and Laux (1993).
We may summarize these empirical results shortly by stating that support is
found for all spread components to be relevant in markets and adverse selection
costs increase with trading volume. However, determining the effect of each
component on the total spread is very difficult and subject to the model applied
for estimation. Up to now there exists no generally accepted framework for the
estimation of spread components and therewith no reliable estimation of the
spread components.

5.2 Estimating informed trading

Apart from evaluating the origins and composition of the spread it is also of
interest to determine the amount of informed trading in the market. As usually
not only the motivations for a trade are not known, but databases do not even
disclose the identity of a trader, it is not possible to make direct inferences.
However, Easley et al. (1996) developed a simple model that uses the number
of trades at the bid and ask to determine the probability of informed trading
(PIN).
Suppose a market consists of two types of traders, informed traders and un-
informed traders. Uninformed traders are arriving at the market following a
Poisson process, independently for buy and sell orders.5 Suppose that in any
time period, e.g. a trading day, there are ε orders submitted by uninformed in-
vestors, for sell as well as buy orders. Informed investors only trade when they
5
A Poisson process states that the probability of x orders arriving at the market in a given
x −x
period of time to be P rob(x orders arriving) = λ x!
e
, where λ denotes the expected number
of arrivals, i.e. E[x] = λ.
162 Chapter 5. Empirical investigations of market microstructure models

1 ε+µ

 Buy



* PP

 PP
 δ  PP
q ε
P
P Sell

HH
HH 1 − δ 1 ε

 Buy
H  
HH 
jP
H
α PP
PP
PP
q ε+µ
P Sell
@
@
@1 − α
@ : ε
 Buy
@  

@
 
@R
@ 
XXX
XXX
XXX
XXX
XXX
z ε Sell

Fig. 5.2: Evolution of the order flow with informed trading

receive new information, which we assume to be happening with probability α.

Upon receiving good news they will buy the asset and upon receiving bad news
they will sell the asset. Good news arrive with probability δ and the order arrival
rate of informed investors in that case is µ. Figure 5.2 illustrates this process.
We easily see that the expected number of orders is given by

(5.13) E[v] = α (δ(ε + ε + µ) + (1 − δ)(ε + ε + µ)) + (1 − α)(ε + ε) = αµ + 2ε,

and the informed trading is given by

(5.14) E[vI ] = α (δµ + (1 − δ)µ) = αµ.

Hence the fraction of informed trading is given by

αµ
(5.15) PI = ,
αµ + 2ε
which is obviously also the probability of informed trading for each individual
trade. In an empirical investigation we now simply have to estimate the pa-
rameters α, µ and ε. To achieve this we use the assumption that order arrivals
5.2. Estimating informed trading 163

follow an independent Poisson process. In the absence of any information the

probability of observing exactly B buy and S sell orders6 is given by
B S B+S
−ε ε −ε ε −2ε ε
(5.16) P robN (B, S) = P robN (B)P robN (S) = e e =e .
B! S! B!S!
In the case of good and bad news we obtain
(µ + ε)B εS
(5.17) P robG (B, S) = P robG (B)P robG (S) = e−(µ+2ε) ,
B!S!
(µ + ε)S εB
P robB (B, S) = P robB (B)P robB (S) = e−(µ+2ε) .
B!S!
Hence the total probability which we can observe is given by
(5.18)
εB+S (µ + ε)B εS (µ + ε)S εB
P rob(B, S) = (1−α)e−2ε +αδe−(µ+2ε) +α(1−δ)e−(µ+2ε) .
B!S! B!S! B!S!
We can now use this probability function to estimate the required parameters.
Investigations of trading on the NYSE by Easley et al. (1996), Chung and
Li (2003) and Chung et al. (2005) show that the PIN is about 14% with more
liquid stocks having a lower and less liquid stocks a higher fraction of informed
trading. Similar results are obtained in Hanousek and Podpiera (2002) and
Němeček and Hanousek (2002) for the Prague Stock Exchange with a PIN
of about 30% and Ma and Yang (2002) find also 14% for Taiwan.
The model as presented here is very restrictive as it assumes firstly that the
order arrival rates remain constant over time which is not a realistic assertion as
pointed out by Easley et al. (2001). Lei and Wu (2001) and Lei and Wu
(2005) allow for time varying order arrival rates as well as different rates for buy
and sell orders. This specification leads to slightly higher estimates of the PIN
of about 22%. Venter and de Jongh (2002) also extend the model to allow
uninformed liquidity traders reacting to the fact of information arrival without
changing the outcome significantly.
In order to estimate our model above, trades have to be classified as initiated
by a buyer or a seller. Obviously simply looking at the bid and ask prices as a
6
More precisely we have to evaluate buyer and seller initiated trades, respectively. In simple
terms, a buyer (seller) initiated trade is one which was conducted as the result of buy (sell)
order being subnitted.
164 Chapter 5. Empirical investigations of market microstructure models

guide is insufficient as a seller might submit a low ask price in order to obtain
a quick trade. Ellis et al. (2000) point out that common ways to determine
the initiation of trades fail in 20-25% of cases. Grammig and Theissen (2002)
point out that in this case the PIN will be biased downwards.

5.3 Properties of daily returns

As this section will show, microstructure elements do not only affect prices be-
tween subsequent trades but also affect the properties of daily and to a much
lesser degree extend to weekly and even monthly returns. Building on the simple
inventory based model of market making, Krause (2003) derives such properties.
Suppose a dealer market with a single competitive market maker in which
investors trade without the ability to use limit orders. The fundamental value of
the asset is assumed to be common knowledge and between subsequent trading
days the log-price evolves as

(5.19) Pt∗ = Pt−1

∗
+ εt ,

where εt ∼ iidN (0, σ 2 ). As derived in chapter 3.1 the prices quoted by the market
maker will depend on his inventory position at the trade τ on trading day t, this
adjustment is denoted ηt,τ . The mid-price quote of the market maker will be

(5.20) M
Pt,τ = Pt∗ + ηt,τ ,

and the bid and ask prices are then given as

1
(5.21) pbt,τ = Pt,τ
M
− st,τ ,
2
1
pat,τ = Pt,τ
M
+ st,τ .
2

Using the result from chapter 3.1, denoting the trade size with Qt,τ , measured
in nominal value, and an inventory with the value of It,τ we obtain from the
equations above that
 
2 1 1
(5.22) ηt,τ = zσ Qt,τ − It,τ − st,τ ,
2 2
5.3. Properties of daily returns 165

which becomes after inserting for st,τ

1
(5.23) ηt,τ = − zσ 2 It,τ .
2

Define α = zσ 2 as the inventory effect and noting that It,τ = It,τ −1 − χt,τ −1 Qt,τ −1 ,
where

 +1 if the order is a sell order
(5.24) χt,τ = 0 if no order arrives ,
−1 if the order is a buy order


and we easily get that

(5.25) ηt,τ = ηt,τ −1 + αχt,τ −1 Qt,τ −1 .

The final assumption necessary is regarding the behavior of investors. Suppose

they are pure liquidity traders and their trading is affected only by the prices
charged by the market maker. Their trading depends on the costs imposed on
a
them by the market maker which is defined as Ct,τ a
= Pt,τ −Pt∗ and Ct,τ
b
= Pt∗ −Pt,τ
b

for trades at the bid and ask, respectively. The probability of trading is then for
x = a, b given by

1
(5.26) λxt,τ = x .
1 + eCt,τ
x 1
Any ex-ante expectations would imply that on average Ct,τ = s .
2 t,τ
Let us
assume for simplicity that the trade size is fixed and thus the spread is constant
and simply denoted by s. When we investigate the daily return we have to focus
on the final trade of the day which determines the relevant price. The above
model can now be used to derive properties of the moments of the daily price
returns. Krause (2003) shows that for daily returns we obtain approximately
the following result:

(5.27) V ar[∆Pt ] = 2α + σ 2 ,

Cov[∆Pt , ∆Pt−1 ] = −α,

V ar[∆Pt2 ] = 9.44α2 + 8ασ 2 ,

Cov[∆Pt2 , ∆Pt−1
2
] = 2α2 .
166 Chapter 5. Empirical investigations of market microstructure models

p
We firstly note from this result that the volatility of the asset price, V ar[∆Pt ], is
higher than the volatility of the fundamental value, σ. Krause (2003) estimates
the price volatility to be about three times the volatility of the fundamental value.
Secondly, returns should exhibit a negative first order autocorrelation, a result
not necessarily confirmed empirically. Thirdly, returns exhibit heteroscedasticity
as observed empirically and finally the volatility is positively correlated over time,
also in accordance with empirical evidence.
Thus we can conclude that the model provides some very realistic results,
apart from the negative autocorrelation of returns. The introduction of long-lived
private information would however easily introduce a positive autocorrelation.
Informed traders spreading their trades over several days to optimally exploit
their information, would easily cause this change in combination with uninformed
traders revising their expectations on the fundamental value.
The model by Roll (1984) can be seen as a special case which neglects
inventory effects by setting α = 0 and assuming a constant fundamental value,
1
i.e. σ 2 = 0. Setting λ = 2
by proposing that traders are not elastic to trading
costs, we recover the result that V ar[∆Pt ] = 12 s2 and Cov[∆Pt , ∆Pt−1 ] = − 41 s2
and thus Corr[∆Pt , ∆Pt−1 ] = − 12 .
Review questions 167

Review questions
1. Why do we need transaction data for estimating microstructure models?

2. What can be said about the composition of the spread in stock markets?

3. Why is the adverse selection component of the spread not a measure of

informed trading?

4. What is the difference between observing a transaction at the ask and a

buyer initiated trade?

5. Why can informed trading not be estimated directly from the trades?

6. What can be said about the amount of informed trading in stock markets?

7. What causes the negative autocorrelation in daily returns?

8. Why is the daily volatility of assets higher than the volatility of the funda-
mental value?

9. What causes heteroskedasticity in daily returns?

10. How is informed trading likely to affect the properties of daily returns?
168 Chapter 5. Empirical investigations of market microstructure models
Chapter 6

The emergence of different market

forms

———————————————————————————————————
The aim of this chapter is to show under which conditions different market forms
emerge endogenously. The focus will be on the number of market makers that
might emerge and the emergence of dealer or limit order markets. The main
aspects considered in this chapter will be
• the decision by market participants to act as market makers
• the relevance of search costs in markets
• the decision to become a match maker

Key readings:
Sanford J. Grossman and Merton H. Miller: Liquidity and Market Structure,
Journal of Finance, 43, 617-637, 1988

A. Yavas: Market Makers versus Match Makers, Journal of Financial Interme-

diation, 2, 33-58, 1992
———————————————————————————————————
170 Chapter 6. The emergence of different market forms

Chapter 1.3 gave an overview of the different market forms and their advan-
tages and disadvantages from the view of an investor. It was intuitively shown
that investors may demand different market forms, depending on their prefer-
ences. These different market forms also have to be supplied, i.e. there have to
be match makers and market makers. In this section we will derive conditions
under which auction and dealer markets arise in an unregulated market.
In both market forms, auction and dealer markets, we find specialized market
participants, match makers and market makers. Contrary to investors they do
not actively trade in the market, they only react to the demand of investors.
A match maker receives an order1 from an investor and has to find an offsetting
order in the market. When he has found such an order he arranges the trade and
determines the price. For these efforts he charges a fee to both investors. At the
time investors submit their orders to the match maker neither the price at which
the trade is conducted nor the time the trade occurs is known. Investors face
the risk that prices may change significantly in the mean time and the trading
decisions may not longer be optimal.
A market maker, on the other hand, publishes prices at which he is willing
to buy and sell the asset on his own account from any investor. At the time the
market maker publishes these prices, he does not know whether the next order
arriving will be a buy or a sell order, but any order is immediately executed at
the stated prices. Investors know the price they will receive upon submitting an
order and also the time of trading. They face no risk after having submitted
the order. This risk is now taken by the market maker, he does not know when
another, offsetting order will arrive at the market and at which price he will be
able to trade. This risk imposes costs on the market maker for which he has to
be compensated. Typically a market maker is not allowed to charge any fee to
investors, therefore he has to incorporate his costs into the prices he charges. He
will quote a higher price for selling the asset to an investor (ask price) and a lower
price for buying the asset from an investor (bid price). As has been shown in

1
Unless otherwise stated only market orders will be considered throughout this chapter.
6.1. Demand and Supply of Immediacy 171

sections 3.1 and 3.2 the ask price will always exceed the bid price, the difference
between these two prices is called the spread.
The possibility for investors to trade at a fixed price without any delay is
called immediacy. The willingness to trade with a market maker is the demand for
immediacy and the willingness to act as market maker is the supply of immediacy.2
The following subsection will provide an equilibrium of immediacy before in
6.2 a model is presented to determine the equilibrium market form.

6.1 Demand and Supply of Immediacy

Grossman and Miller (1988) assume a market with N participants, a single

risky asset3 and a riskless asset with a zero rate of return.4 Of the N market
participants n < N face an exogenously given liquidity event, i.e. their holding of
the risky asset is no longer optimal and they have to trade in order to rebalance
their portfolio by trading the risky against the riskless asset. What causes this
liquidity event is of no importance here, we could easily assume the liquidity
event to be caused by an exogenous need for changing the investment position,
but also informational asymmetries to induce trading. The other N − n market
participants face no need for changing their portfolios. The n market participants
facing a liquidity event are called investors.
We assume that there exist three time periods. At the beginning of period 1
the liquidity event occurs and at the end of this period a first round of trading
takes place. At the beginning of period 2 new information is released on the fun-
damental value of the asset and afterwards a second round of trading occurs. In
period 3 the fundamental value is fully revealed and the portfolios are liquidated
at the fundamental value without further trading.
2
Another form of supplying immediacy is by submitting limit orders. As the effect of this
form of immediacy is not important in this setting, limit orders are not considered here. A
formal treatment of limit orders is given in chapter 4.
3
This single asset can also be interpreted as the optimal risky portfolio as determined by
portfolio theory.
4
The assumption of a zero return can be justified by using the return of the riskless asset as
a normalization, i.e. all returns have to be interpreted as excess returns.
172 Chapter 6. The emergence of different market forms

The price formation has to be such that after the second trading round all
market participants hold their optimal portfolio, i.e. all market participants have
adjusted their portfolios according to the liquidity event and the new information
released in period 2.
Investors can offset their liquidity event either by trading in period 1, in period
2 or in both periods. Assume without loss of generality that they trade either in
period 1 or in period 2, but not in both periods. When an investor decides to
trade depends on his exogenously given preferences and the costs in each period.
Assume that n1 investors trade in period 1 and n2 investors in period 2, where
n1 + n2 = n. Let the number of assets that have to be traded as a result of the
liquidity event by investor i be denoted xi0 , where with a positive amount the
investor buys additional units of the asset and with a negative amount he sells
the asset in exchange for the riskless asset. The sum of all orders has to equal
zero because the overall holding of the asset remains constant:
n
X
(6.1) xi0 = 0.
i=1

But if some investors decide to trade in period 1 and others in period 2, the orders
arriving in each period will not necessarily be balanced. In order to clear the
market there have to be some market participants facing no liquidity event that
are willing to take offsetting positions enabling investors to trade. These market
participants do not have to trade as they face no liquidity event, but they can
take the offsetting positions in period 1 and can these positions offset by trading
again with the remaining investors in period 2. Those market participants that
take voluntary offsetting position in period 1 are called market makers. Let there
be M ≤ N − n market makers.
With X01 denoting the order imbalance in period 1 we define
n1
X
(6.2) X01 ≡ xi0 .
i=1

Market makers have only to be concerned with the net order imbalance as all
other orders can immediately be offset in the same period by matching the orders
6.1. Demand and Supply of Immediacy 173

directly and only the excess is routed to a market maker. Hence they are only
concerned about trades in one direction, either they buy or they sell the asset.
As market makers have to quote fixed prices at which they are willing to trade,
but only trades at one side occur, it has not to be distinguished between bid and
ask prices. The price of the asset in period t, Pt , has to be interpreted as a bid
or ask price, depending on the net order flow.
We assume that all market participants maximize their expected utility of
final, i.e. period 3, wealth. Denote the wealth of investor i in period t by Wti ,
the amount of the riskless asset he holds by Bti , the units of risky assets held by
qti . Let further W0i be the initial wealth of an investor before the liquidity event
occurs. We therewith can determine the final wealth as follows:

(6.3) W1i = W0i + P1 xi0

= B1i + P1 q1i ,

(6.4) W2i = B1i + P2 q1i

= B2i + P2 q2i ,

(6.5) W3i = B2i + P3 q2i .

Inserting B2i = B1i + P2 (q1i − q2i ) and B1i = W0i − P1 (q1i − xi0 ) from manipulating
(6.3) and (6.4), (6.5) becomes

(6.6) W3i = B1i + P2 (q1i − q2i ) + P3 q2i

= W0i + P1 (xi0 − q1i ) + P2 (q1i − q2i ) + P3 q2i

= W0i + P1 (xi0 − q1i ) + P2 (q1i − xi0 ) − P2 (q1i − xi0 − q1i + q2i )

+P3 (q2i − xi0 + xi0 )

= W0i + (P2 − P1 )(q1i − xi0 ) + (P3 − P2 )(q2i − xi0 ) + P3 xi0 .

Define the excess demand over the liquidity event in period t = 1, 2 by investor i
as

(6.7) ξti = qti − xi0 .

174 Chapter 6. The emergence of different market forms

Replacing (6.7) in (6.6) gives

(6.8) W3i = W0i + (P2 − P1 )ξ1i + (P3 − P2 )ξ2i + P3 xi0 .

By using the approximation of the expected utility we get with risk aversion z i :
 
i i 1 i i
(6.9) E[U (W3 )] = U E[W3 ] − z V ar[W3 ] .
2

By using all available information, Ω2 , at time t = 2 we get


(6.10) E[U (W3 )|Ω2 ] = U W0i + (P2 − P1 )(q1i − xi0 )
i

+(E[P3 |Ω2 ] − P2 )(q2i − xi0 ) + E[P3 |Ω2 ]xi0


1 i i i 2
− z (ξ2 + x0 ) V ar[P3 |Ω2 ] .
2

Maximizing (6.10) for the optimal excess demand in period 2, ξ2i , gives the fol-
lowing first order condition:

E[P3 |Ω2 ] − P2 − z i (ξ2i + xi0 )V ar[P3 |Ω2 ] U 0 (.) = 0.

(6.11)

With U 0 > 0 and U 00 < 0 as required for risk averters the second order condition
can easily be shown to be fulfilled. Solving for ξ2i gives the optimal excess demand
in period 2 as

E[P3 |Ω2 ] − P2
(6.12) ξ2i = − xi0 .
z i V ar[P3 |Ω2 ]

As only for investors we find that xi0 6= 0, for all other market participants the
excess demand simplifies to

E[P3 |Ω2 ] − P2
(6.13) ξ2i = ,
z i V ar[P3 |Ω2 ]

where only market makers trade, hence this demand is only valid for market
makers. For market participants that are neither investors nor market makers
we find ξ2i = 0. Assuming all market participants to have the same risk aversion,
i.e. z i = z for all i = 1, . . . , N ,5 we can aggregate (6.12) and (6.13) and get with
5
Assuming equal risk aversion does not change the arguments to be derived below. We could
easily proceed with different degrees of risk aversion at the cost of additional notation.
6.1. Demand and Supply of Immediacy 175

(6.1):6
n
X E[P3 |Ω2 ] − P2
(6.14) ξ2I = ξ2i = n ,
i=1
zV ar[P3 |Ω2 ]
M
X E[P3 |Ω2 ] − P2
(6.15) ξ2M = ξ2i = M .
i=1
zV ar[P3 |Ω2 ]

Market clearing in period 2 requires

E[P3 |Ω2 ] − P2
(6.16) 0 = ξ2I + ξ2M = (n + M ) ,
zV ar[P3 |Ω2 ]
which implies

(6.17) E[P3 |Ω2 ] = P2

unless M = n = 0.
By inserting (6.17) into (6.12) and (6.13) we receive

(6.18) ξ2i = −xi0 for investors,

(6.19) ξ2i = 0 for market makers.

With using (6.17) - (6.19) it is now possible to determine the optimal excess
demand in period 1 by returning to (6.9) and evaluationg this expression using
all available information from period 1, Ω1 :

(6.20) E[U (W3 )|Ω1 ] = U W0i + (E[P2 |Ω1 ] − P1 )ξ1i
i

+(E[P3 |Ω1 ] − E[P2 |Ω1 ])ξ2i + E[P2 |Ω1 ]xi0


1 i i i 2
− z (ξ1 + x0 ) V ar[E[P3 |Ω2 ]|Ω1 ]
2

= U W0i + (E[P3 |Ω1 ] − P1 )ξ1i + E[P3 |Ω1 ]xi0

1 i i i 2
− z (ξ1 + x0 ) V ar[E[P3 |Ω2 ]|Ω1 ] .
2
The first order condition for a maximum is

E[P3 |Ω1 ] − P1 − z i (ξ1i + xi0 )V ar[E[P3 |Ω2 ]Ω1 ] U 0 (.) = 0.

(6.21)
6
It is necessary to aggregate over all investors in period 2 and not only over those trading
in period 2, because the order imbalance of the investors having traded in period 1 has to be
offset in period 2.
176 Chapter 6. The emergence of different market forms

The second order condition can easily be shown to be fulfilled. Solving for ξ1i
gives the optimal excess demand in period 1:

E[P3 |Ω1 ] − P1
(6.22) ξ1i = − xi0 .
ziV ar[E[P3 |Ω2 ]|Ω1 ]

For market makers this again reduces to

E[P3 |Ω1 ] − P1
(6.23) ξ1i = .
ziV ar[E[P3 |Ω2 ]|Ω1 ]

Aggregating over all investors and market makers we get with (6.2) and the
assumption of equal risk aversion for all market participants:

E[P3 |Ω1 ] − P1
(6.24) ξ1I = n1 − X01 ,
zV ar[E[P3 |Ω2 ]|Ω1 ]
E[P3 |Ω1 ] − P1
(6.25) ξ1M = .
zV ar[E[P3 |Ω2 ]|Ω1 ]

Market clearing in period 1 requires

E[P3 |Ω1 ] − P1
(6.26) 0 = ξ1I + ξ1M = (n1 + M ) − X01 .
ziV ar[E[P3 |Ω2 ]|Ω1 ]

The rate of return of the market makers from their activity is given by

P2 − P 1 E[P3 |Ω2 ] − P1
(6.27) r= = .
P1 P1

With (6.26) the expected value and variance of this return is given by

1
(6.28) V ar[r|Ω1 ] = V ar[E[P3 |Ω2 ]|Ω1 ],
P12
E[E[P3 |Ω2 ]|Ω1 ] − P1 E[P3 |Ω1 ] − P1
(6.29) E[r|Ω1 ] = =
P1 P1
1
X0 z
= V ar[E[P3 |Ω2 ]|Ω1 ]
P1 (n1 + M )
P1 X01 z
= V ar[r|Ω1 ].
n1 + M

To derive the equilibrium number of market makers, i.e. the supply of immediacy,
suppose that there exists a fixed cost of C for becoming a market maker, e.g. costs
of the back office. The expected utility from not participating in the market
for all market participants not facing a liquidity event is E[U (W0i )|Ω1 ]. The
6.1. Demand and Supply of Immediacy 177

profits from acting as market maker are (P2 − P1 )ξ1i , hence the expected utility
is E[U (W0i − C + (P2 − P1 )ξ1i )|Ω1 ].
In equilibrium the expected utility from acting as market maker and not
participating in the market have to be equal:

(6.30) E[U (W0i )|Ω1 ] = E[U (W0i − C + (P2 − P1 )ξ1i )|Ω1 ].

As the risk from holding the initial portfolio is equal in both cases it can be
neglected in the further analysis and only the additional risk that arises from
acting as market maker has to be considered.
Inserting (6.23) and (6.26) we get from (6.30):
   
E[P3 |Ω1 ] − P1
E[U (W0i )|Ω1 ] = E U W0i
− C + (P2 − P1 ) Ω1
zV ar[E[P3 |Ω2 ]|Ω1 ]
X01
   
i
= E U W0 − C + (P2 − P1 ) Ω1
n1 + M
2 !
1 1

X 0 1 X 0
= U W0i − C + (E[P2 |Ω1 ] − P1 ) − z V ar[P2 |Ω1 ] .
n1 + M 2 n1 + M

By comparing coefficients we get

2
X01 X01

1
(6.31) C = (E[P2 |Ω1 ] − P1 ) − z V ar[P2 |Ω1 ]
n1 + M 2 n1 + M
X01
= (E[E[P3 |Ω2 ]|Ω1 ] − P1 )
n1 + M
 1
2
1 X0
− z V ar[P2 |Ω1 ]
2 n1 + M
2
X01 X01

1
= (E[P3 |Ω1 ] − P1 ) − z V ar[P2 |Ω1 ]
n1 + M 2 n1 + M
2
P1 X01 X01

1
= E[r|Ω1 ] − z V ar[P2 |Ω1 ]
n1 + M 2 n1 + M
2 2
X01 X01
 
2 1
= z P1 V ar[r|Ω1 ] − z V ar[P2 |Ω1 ]
n1 + M 2 n1 + M
2 2
X01 X01
 
1
= z V ar[P2 |Ω1 ] − z V ar[P2 |Ω1 ]
n1 + M 2 n1 + M
2
X01

1
= z V ar[P2 |Ω1 ].
2 n1 + M
178 Chapter 6. The emergence of different market forms

Solving for M as the number of market makers gives the optimal supply of im-
mediacy:
r
zV ar[P2 |Ω1 ]
(6.32) M= |X01 | − n1 .
2C

The number of market makers is larger the lower the costs of market making, the
higher the order imbalance, the higher the risk aversion and uncertainty about
the price in period 2 and the lower the number of investors trading in the first
period. For a high number of market makers it would be optimal that only a few
investors were trading in the first period having a large order imbalance.7
When determining the optimal number of market makers, the restriction M ≤
N − n has further to be taken into account. In this framework immediacy is
provided because of the possible profits that can be derived from a temporary
order imbalance. How large this order imbalance is and how many investors
decide to trade in period 1 also depends on the costs charged by market makers
for trading in the first period, i.e. the demand for immediacy. The higher these
costs, the lower demand will be.

6.2 Determination of the optimal market form

The last section presented a model to determine the number of market makers
endogenously. It was assumed that only the order imbalance in periods 1 and
2 are offset by market makers. This implies that the total order flow at first
is matched and only those orders that could not be matched in this process
are routed to market makers. Implicitly therewith it has been assumed that this
matching process is costless and takes place in an instant. In reality, however, this
matching process will be costly. Searching for an offsetting order is time intensive
and it is not ensured that an offsetting order can be found in a given time period,
even if such an order exists. These costs can give rise to the emergence of a
specialized market participant, the match maker. He can be used by investors
7
Grossman and Miller (1988, pp. 626) derive a similar result by assuming that the order
imbalance is not known to the market makers in advance. For solving this problem they have
to assume an exponential utility function.
6.2. Determination of the optimal market form 179

to match their orders. As many investors choose this match maker, he can more
easily find an offsetting order. Unlike the market maker he does not offset the
order by trading on his own account, he only matches two orders of investors.
The match maker, as well as the market maker, makes it more easy to execute
an order, hence they both facilitate trading and provide liquidity, what has been
identified in chapter 1.2 to be one reason for the emergence of markets.

In chapter 6.1 it has been assumed that a market participant not facing a
liquidity event only can choose to become a market maker or not to participate
in the market. In this section a model will be presented in which a single market
participant can choose to become a match maker or market maker.8 Yavas
(1992) provides a model how this market participant chooses between becoming
a market maker or a match maker and hence whether an auction or a dealer
market emerges.

We only have two groups of market participants, besides the match maker or
market maker. One group assigns a value of P1 and the other of P2 to the asset.
All market participants are risk neutral, implying that maximizing expected util-
ity is equivalent to maximizing expected profits. Every investor knows the value
he assigns to the asset, but not the value the other groups assigns. Their value
is a random variable whose distribution function Fi (Pi ) is known to all market
participants.

There is not a fixed buyer and a fixed seller group. If the price used for a
transaction is below the value assigned to the asset, it is bought by the investor,
otherwise it is sold.9 If two investors meet they reveal their values of the asset.
The joint surplus of a trade, |P1 − P2 |, is assumed to be shared equally, each
1
investor receiving 2
|P1 − P2 |.10 For the following we assume without loss of
generality that P1 > P2 .

8
The lack of competition between market makers or match makers does not effect the model
to be presented here. We can similarly assume that all market participants have to make the
same decision, i.e. that we are considering a group of market participants acting competitively.
9
If the two groups assign the same value to the asset no transaction occurs, but for simplicity
we neglect this case without changing results.
10
Yavas (1992, p. 36) shows that other divisions of the surplus do not affect the main results.
180 Chapter 6. The emergence of different market forms

Let further 0 ≤ θ ≤ 1 denote the probability at which two investors of dif-

ferent groups meet. This probability depends on the search intensity of the two
investors, A1 and A2 , i.e. θ = θ(A1 , A2 ). We assume that

∂θ(A1 , A2 )
(6.33) > 0,
∂Ai
∂ 2 θ(A1 , A2 )
< 0 i = 1, 2.
∂A2i

The costs of searching for investor i are denoted Ci (Ai ), where

∂Ci (Ai )
(6.34) > 0 i = 1, 2,
∂Ai
2
∂ Ci (Ai )
> 0 i = 1, 2,
∂A2i
∂Ci (Ai )
= 0 i = 1, 2,
∂A3−i
∂Ci (Ai ) ∂θ(A1 , A2 )
> i = 1, 2,
∂Ai ∂Ai
Ci (0) = 0.

The functions θ(A1 , A2 ) and Ci (Ai ) are assumed to be known by all market
participants. As for an investor the value his trading partner assigns to the asset
is random, we get the expected surplus from a trade in the absence of a market
or match maker:
Z ∞
1
(6.35) Si (Pi ) = |Pi − Pe3−i |dF3−i (Pe3−i ).
0 2

The search intensity of the other group of investors is not known to investors of
the other group, hence they have to be conjectured, this value is denoted A03−i .
The expected profits from searching are
Z ∞
1
(6.36) πi (Ai , Pi , A03−i ) = θ(Ai , A03−i ) |Pi − Pe3−i |dF3−i (Pe3−i ) − Ci (Ai ).
0 2

Maximizing expected profits by choosing an optimal search intensity gives the

6.2. Determination of the optimal market form 181

following first order condition:

∂θ(Ai , A03−i ) ∞ 1
Z
∂πi ∂Ci (Ai )
(6.37) = |Pi − Pe3−i |dF3−i (Pe3−i ) −
∂Ai ∂Ai 0 2 ∂Ai
0
∂θ(Ai , A3−i ) ∂Ci (Ai )
= Si (Pi ) −
∂Ai ∂Ai
= 0.

The second order condition for a maximum can easily be shown to be fulfilled
∂ 2 Ci (Ai ) ∂ 2 θ(Ai ,A03−i )
if ∂A2i
> ∂A2i
Si (Pi ). The term on the left side is positive from the
assumption in (6.34), the first term on the right side is negative from (6.33)
and the second term positive as can be seen from its definition in (6.35), hence
the left side is negative. The second order condition is always fulfilled with the
assumptions stated above.
As we assume all functions to be common knowledge, investors can correctly
infer the search intensity of the investors of the other group by solving their
equation (6.37). Inserting this result, they can then solve (6.37) to find their own
optimal search activity.
So far we have not introduced a match maker or market maker, the investors
were supposed to search for each other on their own behalf. Without introduc-
ing such an intermediaire this search equilibrium is subject to two inefficiencies.
The first inefficiency is that investors may not meet, although they would like
to trade with each other. The second inefficiency is that the probability that
investors meet, depends on the search activities of both groups, hence each group
produces positive externalities to the other group. Therefore the search activity
in equilibrium will be lower than in the social optimum.
These inefficiencies enable match or market makers to be established in the
market. It will first be analyzed how investors behave in such markets and then
these two market forms are compared.
We first analyze the case where the intermediaire is a market maker. Like
the investors we assume him to be risk neutral. He is also assumed to know the
distributions of the fundamental values each group of investors assigns to the
asset, but not the valuation itself. Furthermore he knows the cost functions for
182 Chapter 6. The emergence of different market forms

searching and the probability function of meeting upon searching.

At the beginning of the period a market maker quotes his prices at which he
is willing to sell the asset, Pa , and at which he is willing to buy the asset, Pb ,
where Pa ≥ Pb .11 These prices become known to all investors at no costs. They
then have to decide whether they want to search for an offsetting order on their
own or whether they want to trade directly with the market maker at the stated
prices. By searching on their own investors hope to receive a better price than
the quotes of the market maker. If the search is not successful, i.e. they find no
offsetting order, or upon meeting the price is less favorable than the quotes of
the market maker, they can yet trade with the market maker at the same stated
prices. But as the search process takes time this trade will take place at the end
of the period, hence the surplus achieved has to be discounted by a rate of ρ.12

If Pi > Pa the investor will buy the asset from the market maker and if Pi < Pb
he will sell it to him. If Pb > Pi > Pa the investor would make a loss from trading
with the market maker and hence will not trade with him.

The surplus from trading with the market maker, Pi − Pa and Pb − Pi , re-
spectively, is now not divided as the market maker quotes firm prices. In the
presence of a market maker the surplus from a direct trade between investors
has to be divided as follows: To prevent the investors to refuse the trade and
trade with the market maker instead, they have at first to receive the discounted
surplus they would get from trading with him afterwards. After both investors
have received this compensation for the surplus they could receive from trading
with the market maker afterwards, the remaining surplus is divided equally be-
tween them. If the surplus from a direct trade with each other cannot give this
minimum compensation, it is refused. Therewith they only trade with each other

11
If Pa < Pb an investor could buy the asset at Pa from the market maker and sell it to
him in an instant of time at Pb making a profit, while the market maker would make a loss.
To prevent this arbitrage it is required that Pa ≥ Pb . See chapters 3.1 and 3.2 for more
sophisticated models of market making.
12
This discount factor could also be interpreted as an adoption of utility to the risk that
prices have changed unfavorably in the mean time if we assume risk averse market participants.
6.2. Determination of the optimal market form 183

if

(6.38) P2 − P1 ≥ ρ (max{0, P1 − Pa , Pb − P1 } + max{0, P2 − Pa , Pb − P2 }) .

Without an intermediaire the reservation prices of the investors have been their
valuation of the asset. With the existence of a market maker these reservation
prices change. For trading with each other the price of an investor to buy the
asset from another has to be reduced by the surplus he could earn from trading
with the market maker. Similarly for the investor selling the asset the reservation
price is increased by this amount. The reservation prices become

(6.39) P1r = P1 − ρ max{0, P1 − Pa , Pb − P1 },

P2r = P2 + ρ max{0, P2 − Pa , Pb − P2 }.

The negotiated price will always be between these two reservation prices. If
P1r ≥ P2r we see that

(6.40) P1 ≥ P1r ≥ P2r ≥ P2 .

The prices are allowed to change only in a smaller interval by subsequent trades
with the presence of a market maker, hence the price dispersion is reduced.
The additional surplus from trading directly with each other instead with the
market maker is 21 (P1r − P2r ) if P1r > P2r , otherwise no trade occurs. There also
does not occur a trade if the investors do not meet. They will not meet if one of
them decides to trade with the market maker directly instead of searching first,
i.e. his search activity is zero (Ai = 0). Therefore the probability distribution of
the other investors valuation changes to

D Fi (Pi ) if Ai > 0
(6.41) Fi (Pi ) = .
0 if Ai = 0

The expected surplus from a trade with each other now becomes by using (6.39):
Z ∞
D 1
(6.42) Si (Pi , Pa , Pb ) = max{0, P1r − P2r }dF3−i
D
(P3−i )
0 2
As we see from (6.40) and (6.41) that P1r − P2r ≤ P1 − P2 and FiD (Pi ) ≤ Fi (Pi ),
it is obvious that the expected surplus from a trade with each other is reduced
184 Chapter 6. The emergence of different market forms

in the presence of a market maker:

(6.43) Si (Pi ) ≥ SiD (Pi , Pa , Pb ).

The expected profits are determined by the expected surplus from trading with
each other, the costs of searching,13 and the expected surplus from trading with
the market maker:

(6.44) πi (Ai , Pi , A03−i , Pa , Pb )

Z ∞
1
= D
θ(Ai , A03−i ) max{0, P1r − P2r }dF3−i (P3−i )
0 2
−Ci (Ai ) + (1 − θ(Ai , A03−i ))ρ max{0, Pi − Pa , Pp − Pi }.

Maximizing to determine the optimal search activity, Ai , gives the following first
order condition:14

∂πi ∂θ(Ai , A03−i ) D ∂Ci (Ai )

(6.45) = Si (Pi , Pa , Pb ) −
∂Ai ∂Ai ∂Ai
0
∂θ(Ai , A3−i )
− ρ max{0, Pi − Pa , Pb − Pi }
∂Ai
∂θ(Ai , A03−i ) D

= Si (Pi , Pa , Pb ) − ρ max{0, Pi − Pa , Pb − Pi }
∂Ai
∂Ci (Ai )
−
∂Ai
= 0.

By inspection of (6.43) we see that the term in brackets has to be smaller than
Si (Pi ), but still is positive. Let us for notational simplicity rewrite (6.37) and
(6.45) as

(6.46) θ0 Si − Ci0 = 0,
0 0 0
(6.47) θD SiD − CD = 0,

0
where SiD = SiD (Pi , Pa , Pb ) − ρ max{0, Pi − Pa , Pb − Pi } ≤ Si . Solving for Si and
13
These costs also have to be beared if no trade occurs as the investors do not meet or find
it more profitable to trade with the market maker later, they are sunk costs.
14
The second order condition can be proofed to be fulfilled in the same manner as in (6.37).
6.2. Determination of the optimal market form 185

0
SiD we get
Ci0
Si = ,
θ0
0
0 CD
SiD = 0
.
θD
Hence we have
Ci0 CD0
(6.48) ≥ 0
.
θ0 θD
From (6.34) we can find a k and a kD such that Ci0 = kθ0 and CD
0 0
= kD θD , where
because of (6.33) and (6.34) k and kD are strictly increasing in Ai . Inserting
these relations into (6.48) we get

(6.49) k ≥ kD .

Hence the optimal search activity is smaller in the presence of a market maker
than without. This is the result of the possibility to trade with the market maker
instead of only trading directly with other investors.
The optimal search intensity can be determined using (6.45). Based on these
interferences of the optimal search activities the market maker can set his prices
optimal to maximize his expected profits. Yavas (1992, pp. 42 ff.) shows that
an overall equilibrium exists and assigns the problems of multiple equilibria that
are of no importance for our purpose. It can easily be shown that the expected
profits of the investors are higher in the presence of a market maker, hence it is
beneficial for investors to have a market maker. A market maker will also make
profits from his activity and therefore there will be market participants acting as
market maker.
We now turn to the case where the intermediaire has decided to become a
match maker. Like in the presence of a market maker investors can either decide
to search for an offsetting order by themselves, facing the risk that they do not
meet, or use the match maker where we assume that he has the certainty that the
order will be executed.15 If they decide to search and do not find an offsetting
15
We could also assume that the probability of execution is significantly higher by using the
match maker without changing the argument.
186 Chapter 6. The emergence of different market forms

order, they can afterwards turn to the match maker for execution and receive
the discounted surplus. For his service the match maker charges a fee from both
investors whose orders he matches, the fee is a fraction c of the transaction price.
The price is determined by the match maker such that the surplus is equally
distributed between the investors.
The reservation prices for a direct trade with each other are again the val-
uations of the asset as no other market participant offers a better price. The
match maker only matches the orders and offers no fixed price. The surplus from
trading will P1 − P2 , that will be divided equally. When trading with the help
of a match maker the surplus however first has to cover the fees that have to be
paid to the match maker. If the surplus does not cover these costs no trade will
occur. With P denoting the transaction price, we get

(6.50) P1 − P ≥ cP,

P − P2 ≥ cP,

which implies
P1
(6.51) P ≤ P1r = ≤ P1 ,
1+c
P2
P ≥ P2r = ≥ P2 ,
1−c
where Pir denote the reservation prices. A trade through the match maker will
only occur at prices between P2r and P1r , with P1r > P2r . As this interval is smaller
than [P2 , P1 ] the price dispersion is reduced like in the presence of a market maker.
The surplus from trading with a match maker is given by

(6.52) P1 − P2 − c(P1 + P2 ) = (1 + c)P1r − (1 − c)P2r

−c(1 + c)P1r + (1 − c)P2r

= P1r (1 + c)(1 − c) − P2r (1 + c)(1 − c)

= (1 − c2 )(P1r − P2r )

> 0.

If P2r > P1r the surplus would be negative and no trade would occur. Like in
the presence of a market maker investors can also directly use the match maker
6.2. Determination of the optimal market form 187

instead of searching for an offsetting order by themselves, i.e. Ai = 0. In this

case an investor searching an offsetting order cannot find a counterpart and his
probability distribution changes to

M Fi (Pi ) if Ai > 0
(6.53) Fi (Pi ) = .
0 if Ai = 0

The expected surpluses from trading directly with each other and by using the
match maker are given by
Z ∞
1 M
(6.54) Si (Pi ) = |Pi − P3−i |dF3−i (P3−i ),
2
Z0 ∞
1
(6.55) SiM (Pi , c) = M
max{0, |Pi − P3−i | − c(P1 + P2 )}dF3−i (P3−i ).
0 2

Therewith the expected profit of an investor is

(6.56) πiM (Pi , Ai , AM 0

3−i , c) = θ(Ai , A3−i )Si (Pi )

+(1 − θ(Ai , A03−i ))ρSiM (pi , c) − Ci (Ai ).

Maximizing the expected profits to find the optimal search activity gives the
following first order condition:

∂πi ∂θ(Ai , A03−i ) ∂Ci (Ai )

(6.57) = Si (Pi ) −
∂Ai ∂Ai ∂Ai
0
∂θ(Ai , A3−i ) M
− ρSi (Pi , c)
∂Ai
∂θ(Ai , A03−i ) ∂Ci (Ai )
= (Si (Pi ) − ρSiM (Pi , c)) −
∂Ai ∂Ai
= 0.

As from inspection of (6.54) and (6.55) we see that Si (Pi ) ≥ SiM (Pi , c) ≥
ρSiM (Pi , c) ≥ 0, it can easily be verified that 0 ≤ Si (Pi ) − ρSiM (Pi , c) ≤ Si (Pi ).
This is in line with the results in the presence of a market maker, the second
order condition is fulfilled and it turns out that the search activities are reduced
in the presence of a match maker. The optimal search activity can be derived
from (6.57) and given these results that match maker can determine his optimal
fee. Yavas (1992, pp. 50 f.) again shows the existence of such an equilibrium.
As in the presence of a market maker the expected profits of the investors are
188 Chapter 6. The emergence of different market forms

higher with the presence of a match maker, hence it is beneficial for investors to
have a match maker. A match maker will also make profits from his activity and
therefore there will be market participants acting as match maker.
Thus far it has only been shown that a market maker as well as a match maker
are beneficial for investors and that both forms will be provided as they make
a profit from their activities. Nothing has been said about which form will be
preferred. We therefore now turn to the choice a market participant will make,
whether to become a market maker or a match maker. The decision is made on
the basis of the expected returns one can earn from these two activities. Yavas
(1992) provides some conditions for this choice.16
A market participant will decide to become a market maker if the value he
assigns to the asset differs largely from that of the investors, i.e. if his valuation
is at the tail of the distribution of the investors. By offering a very favorable
price to both investor groups he still makes a sufficient profit from his activity
with a high probability as it is very unlikely that two investors will find it more
profitable to trade with each other even if they meet. If search is very efficient,
i.e. the costs are relatively small, two investors searching are very likely to meet,
but they in most cases will not trade with each other as the market maker can
offer a more favorable price and even with very efficient search the market maker
will make a considerable profit.
However, if the valuation of the asset does not differ much between investors
and intermediaire and search is inefficient, i.e. imposes high costs, he will choose
to become a match maker. As market maker he would in many cases not offer a
more favorable price to the investors than if they trade directly with each other, so
that he is unlikely to serve an order, hence his expected profits are low. Whereas
a match maker charges fees from both sides independent of the price he charges
giving him higher profits.
These conditions to become a market maker or a match maker may change
over time and therefore he may want to revise his decision. As a market should
16
The long and tedious proofs are not presented here to limit space, but they are provided
in detail by Yavas (1992, pp. 51 ff.).
6.2. Determination of the optimal market form 189

apply the same rules over a longer period of time to give investors stable envi-
ronments to trade and enable the price mechanism to work properly, frequent
changes of the market forms should not be allowed. When taking into account
that a decision binds him over a given period the above criteria have to be made
in a dynamic setting, taking into account changes that are expected. The general
idea is not affected, it has to be taken the (discounted) profits over the time
period he is bound.
The decision in reality often is restricted by rules stating that all assets in a
certain market have to be traded with the same rules. For the determination of
the optimal market form in this case the costs and benefits have to aggregated. A
certain variety can be achieved by defining market segments that can be traded
according to different rules. The assets are then traded into one of these segments
according to their characteristics.
190 Chapter 6. The emergence of different market forms

Review questions
1. Why are some traders willing to act as market makers?

2. What drives the number of market makers?

3. Who of the traders is likely to become a market maker?

4. Why is searching inefficient?

5. What are the types of inefficiency in search?

6. How can intermediaries overcome the inefficiencies in search?

7. Why is the search of traders reduced in the presence of an intermediary?

8. What is the main difference between a match and a market maker?

9. What drives the decision whether an intermediary becomes a market or

match maker?

10. How can a stock exchange decide on its optimal market form?

Application
Application 1

On continuation of the application in chapters 2, 3 and 4 the regulator has pro-

duced additional information on the trading activity in the market. He has
analyzed the fraction of traders who within an hour buy and sell the stock at
least once, the graph is shown below.
What can be said about the evolution of the market structure and how does
this fit within the wider context of the case as presented and analyzed before?

Application 2

The state of Galaria has set up a stock exchange after embarking on a programme
of economic liberalizations in the mid-1990s. Since then the stock market has
developed considerably and the number of listed stocks has increased to 534 last
month with a total market capitalization of the equivalent of $85bn, equal to
nearly 26% of GDP.
Application 191

0.16

0.14
Fraction of short term traders

0.12

0.1

0.08

0.06

0.04
08:00
10:00
12:00
14:00
16:00
08:00
10:00
12:00
14:00
16:00
08:00
10:00
12:00
14:00
16:00
08:00
10:00
12:00
14:00
16:00
08:00
10:00
12:00
14:00
16:00
Monday Tuesday Wednesday Thursday Friday

The stocks listed on the stock exchange fall mainly into two categories. Firstly
large formerly state-owned companies that have been privatized and whose shares
are widely held by private investors. Holdings are in most cases small, although
a few investors, mostly investment funds, have accumulated significant stakes in
these companies. The companies in this category are mostly utilities and a few
well established mining and oil exploration companies.

The second group of companies listed can best be described as small and
medium-sized technology companies. Their origin lies in the drive of the state to
become a leader in computer technology and they are to a large degree engaged
in computer and chip manufacturing. Only recently have they become engaged
also in the production of LCD and plasma screens, mostly as subcontractors
to the leading companies in this sector. Another characteristic of these stocks
is that while they are much more actively traded than the former state-owned
companies, their volatility is also significantly higher. In contrast to the formerly
state-owned companies their prospects are subject to very contrasting views in
the investment community.
192 Chapter 6. The emergence of different market forms

Thus far the stock exchange has relied on an open out cry mechanism on the
floor of the stock exchange. But repeatedly this trading mechanism has been
criticized for not being able to cope with the steadily increasing trading volume.
In response to these criticisms the government has in cooperation with the stock
exchange set up a task force to review the market structure of the stock exchange.
Given the information you have been given above, what would your recom-
mendation to the task force be on the best market structure?
Chapter 7

Asset Pricing with Trading Frictions

———————————————————————————————————
This chapter will provide an overview of the influence market microstructure
elements have on asset prices. The emphasis will be on the equilibrium rate of
return an asset has to generate in the market. We will investigate
• the impact of the bid-ask spread
• the impact of adverse selection
• the impact of imperfect liquidity in the market
• the fraction of the equity premium these trading frictions can explain

Key readings:
Yakov Amihud and Haim Mendelson: Asset Pricing and the Bid-Ask Spread,
Journal of Financial Economics, 17, 223-249, 1986

Maureen O’Hara (2003): Liquidity and price discovery, Journal of Finance, 58,
1335-1354, 2003

W.J. Breen, L. S. Hodrick and R. A. Korajczyk: Predicting equity liquidity,

Management Science, 48, 470-483, 2003

V. V. Acharya and Lasse Pedersen: Asset pricing with liquidity risk. Mimeo,
London Business School, 2004
———————————————————————————————————
194 Chapter 7. Asset Pricing with Trading Frictions

The models considered thus far focused exclusively on the trading decisions
of individual investors and the behavior of special intermediaries present in the
market. The costs faced by market makers and other traders, such as inventory
costs or adverse selection costs were borne indirectly by investors through the
bid-ask spread or the price impact of their trade, i.e. illiquidity. These trading
costs of investors will reduce their net returns from holding the asset. Assuming
them to be exogenously given and the required rate of return determined by the
risk of the asset, the market returns have to be higher to compensate for these
costs. Therefore we can reasonably argue that the market returns of assets will
depend on microstructure elements as identified above.
The benefits of including market microstructure elements into the determi-
nants of asset returns and thus asset prices is not only restricted to a better
explanation of observed prices but can directly be used by companies to evaluate
the optimal market form for their listing. Hence it will not only be investors seek-
ing to trade in a market structure that reduces their trading costs but companies
will seek to list on exchanges whose microstructure ensures the highest possible
price for their shares.
The main sources of transaction costs for investors are the bid-ask spread and
the threat that as an uninformed investors he will make losses when trading with
an informed investor (adverse selection), a problem similar to that of the market
maker. The final element of the trading costs is the illiquidity of the market
through which an investor affects the price when trading. In this chapter we will
address each of these costs in turn and investigate how these costs affect asset
prices.

7.1 Required returns in the presence of

bid-ask spreads
Let us consider a market in which M > 1 investors can trade N + 1 assets.
Investors differ only in the time horizons of their investments, denoted Tj for
j = 1, 2, . . . , M . The assets differ in the spread that is exogenously imposed on
7.1. Required returns in the presence of bid-ask spreads 195

investors, si for i = 1, 2, . . . , N . Furthermore there exists an asset for which no

spread is charged, i.e. s0 = 0.
To avoid any effect arising from the uncertainty of prices in the future, Ami-
hud and Mendelson (1986a) assume that the price of all assets remains con-
stant for all time horizons of investors with the ask price at Pi and the bid price
at Pi (1 − si ) for all i = 0, 1, 2, . . . , N . With each asset paying a fixed dividend of
di in each time period the return for investor j from investing into asset i is for
a time horizon of Tj given by
1 Pi (1 − si ) + di Tj − Pi di si
(7.1) rij = = − ,
Tj Pi Pi Tj
thus consisting of the gross return adjusted by the costs of the spread for each
unit of time the investment is held. The return of the asset in the market will now
be determined by those investors willing to pay the highest price, thus requiring
the lowest return. Conversely, any investor will seek to hold that asset which
provides him with the highest return.
Consider now two assets, l and m, which are held by different investors, p and
q. With the foresaid we obviously require that rlp ≥ rmp and rlq ≤ rmq . Inserting
from (7.1) we obtain the conditions that
dl sl dm sm
(7.2) − ≥ − ,
Pl Tp Pm Tp
dl sl dm sm
− ≤ − .
Pl Tq Pm Tq
We can now easily combine these two inequalities to obtain

(7.3) (sl − sm ) (Tp − Tq ) ≥ 0.

Thus if sl > sm we require that Tp > Tq , which means that assets with a larger
spread are held by investors with longer investment horizons. This result obtained
here has been confirmed empirically by Atkins and Dyl (1990).
The market return will be the minimum return any investor requires as his
demand will be served first due to him willing to pay the highest price:
 
si
(7.4) ri = min rij + .
j=1,2,...,M Tj
196 Chapter 7. Asset Pricing with Trading Frictions

We see that this return is increasing in the spread and as with the spread the
time horizon increases this increase becomes less and less pronounced, i.e. ri is
concave in the spread.1
Using the asset with a spread as a benchmark, we see that the asset price is
decreasing with the spread. Capitalizing the dividend at the respective market
d0 di
returns we obtain that P0 = r0
and Pi = ri
. By assuming that d0 = di for
simplicity we get

Pi r0
(7.5) = ,
P0 ri

which is decreasing in the spread as we argued above that ri was increasing in

the spread. In an alternative interpretation, we can determine that the price of
the asset is reduced by the present value of future trading costs:
 
di si
(7.6) = ri = min rij + ,
Pi j=1,2,...,M Tj
!
di
(7.7) Pi = max ,
j=1,2,...,M rij + Tsij
si
di T
(7.8) Pi = − Pi j .
ri ri

We can thus conclude this analysis by stating that a higher spread increases the
required gross rate of return companies have to achieve which in turn reduces
the value of shares. This increases the costs of equity to companies and provides
them an incentive to seek a listing at a stock exchange that is able to provide
the lowest spread as is discussed in Amihud and Mendelson (1988). Another
aspect apparent from the above analysis is that the required gross rate of return is
also increasing the shorter the time horizon of investors is. Here we find incentives
for companies to attract more long-term investors as this actually reduces the cost
of equity.
Empirical evidence overwhelmingly support the finding that a higher bid-ask
spread increases the required rate of return. Amihud and Mendelson (1986a)
1
It can be shown that in equilibrium the increase in Tj does not lead to the adjustment term
to reduce, although with ri = Pdii the price will also change.
7.1. Required returns in the presence of bid-ask spreads 197

find evidence that an increase in the spread by one percentage point increases the
return by about 2.5% p.a., a similar result is obtained in Amihud and Mendel-
son (1986b), Amihud and Mendelson (1989) and Amihud and Mendelson
(1991) with additional evidence for the concavity of the relationship between the
spread and the return. Eleswarapu (1997) also finds a positive relationship
but points out that this result is mostly driven by a strong relationship in Jan-
uary. This seasonality is confirmed in Rubio and Tapia (1998) for the Spanish
market.

The proposed effect on the value of assets is also confirmed in Amihud and
Mendelson (1991) reporting that a doubling of the spread causes prices to drop
by about 6%. For the Swiss stock exchange and the NASDAQ Loderer and
Roth (2003) find that a spread of 1% causes a discount of the asset of about
9.4% relative to an asset trading without a spread. The average discount they
find on the Swiss exchange is 12% and 28% for the NASDAQ. Mixed evidence
on the effect of the spread can be obtained by investigating a change of exchange
listings. While Baker and Edelman (1992) report a positive effect of a reduced
spread on the asset value, the results of Kadlec and McConnell (1994) and
Barclay et al. (1998) are inconclusive. Finally Benveniste et al. (2001)
report a premium of 12-22% for liquid real estate investment trusts compared to
illiquid vehicles trading at a significant spread.

Other models using a transaction cost approach to trading yield similar re-
sults to those obtained above, we could easily extend the model to include other
relevant costs for investors such as broker fees. In particular Lo et al. (2004)
develop a model which also shows a concave premium for higher transaction costs
and a discount of the asset price increasing with trading costs. This discount is
very low in Constantinides (1986) using a similar model, but this can be at-
tributed to the fact that he only allows for very infrequent trading of the asset,
thus requiring no significant compensation for trading costs. This is similar to
Vayanos (1998) where the time horizon is endogenously determined. Other ex-
amples of model variations are Huang (2003) who also considers borrowing as
198 Chapter 7. Asset Pricing with Trading Frictions

an alternative to selling or Swan (2002) using a similar model to that used above
but allowing for endogenous trading.

7.2 Compensation for adverse selection costs

The spread is not the only cost faced by investors in the market. As has been
pointed out in chapter 3.2, informed traders always achieve an expected profit at
the expense of uninformed traders. These losses have to be compensated for to
the uninformed traders or they would not be willing to participate in the market.
This compensation is not included in the spread as the spread only compensates
the market maker. In this section we present a simplified model first developed
in Easley et al. (2001) and O’Hara (2003) which allows us to determine the
premium investors require for adverse selection.

Suppose a market with a single asset and a risk free asset B paying a fixed
return of r. The risky asset is traded in a single trading round after which it is
distributed to its holders at the fundamental value v. This fundamental value
is not known to traders at the time of trading, but its distribution is common
knowledge: v ∼ N (v, σv2 ). The market has three types of traders, noise traders,
uninformed strategic traders and informed strategic traders. A fraction λ of the
strategic traders are informed and receive a common signal s ∼ N (v, σs2 ) about
the true value of the asset.

Each trader i has an initial wealth Wi = Bi + qi p, consisting of the risk free

asset Bi and an investment of qi units into the risky asset at price p. The wealth
in the period after trading becomes

(7.9) Wi0 = (1 + r)Bi + qi v

= (1 + r)(Wi − qi p) + qi v

= (1 + r)Wi + qi (v − (1 + r)p).

Traders are risk averse with a common absolute risk aversion z and seek to max-
7.2. Compensation for adverse selection costs 199

imize the expected utility using their information set Ωi :

 
0 0 1 0
(7.10) E [U (Wi )|Ωi ] ≈ U E [Wi |Ωi ] − zV ar [Wi |Ωi ]
2
= U ((1 + r)Wi + qi (E [v|Ωi ] − (1 + r)p)

1 2
− zqi V ar [v|Ωi ] .
2

The optimal demand can easily be derived as

E [v|Ωi ] − (1 + r)p
(7.11) qi = .
zV ar [v|Ωi ]

With informed traders receiving the common signal s we can use Bayes’ theorem
such that
v s
σv2
+ σs2 vσs2 + sσv2
(7.12) vI = E [v|Ωi ] = 1 = ,
σv2
+ σ12 σv2 + σs2
s

σv2 σs2
(7.13) σI2 = V ar [v|Ωi ] = .
σv2 + σs2

Thus the demand of informed investors is determined as

vσs2 + sσv2 − (1 + r)(σv2 + σs2 )p

(7.14) qI = .
zσv2 σs2

Uninformed investors do not receive the signal directly, but can infer some infor-
mation from the prevailing price. Suppose that

(7.15) p = αv + βs + γq + δx,

+ u, with u ∼ N (0, σu2 ) the demand of noise traders, is the total

P
where q = i qi

demand of all traders and x the total supply of the asset in the market. Define

p − αv + (γ − δ)q δ
(7.16) θ= = s + (x − q),
β β

which can be inferred from the available information by uninformed traders. We

easily see that in equilibrium with E[q] = x we obtain

δ
(7.17) E[θ] = E[s] + (x − E[q]) = v,
β
δ2 δ2
V ar[θ] = V ar[s] + 2 V ar[q] = σs2 + 2 σu2 = σθ2 .
β β
200 Chapter 7. Asset Pricing with Trading Frictions

Using θ as a signal the uninformed traders form their beliefs as follows:

v θ
σv2
+ σθ2 vσθ2 + θσv2
(7.18) vU = E [v|θ] = 1 1 = ,
σv2
+ σθ2
σv2 + σθ2
σv2 σθ2
(7.19) σU2 = V ar [v|θ] = ,
σv2 + σθ2
giving rise to the demand of uninformed investors of
vσθ2 + θσv2 − (1 + r)(σv2 + σθ2 )p
(7.20) qU = .
zσv2 σθ2
In equilibrium we obviously require that demand and supply equal, thus

(7.21) λqI + (1 − λ)qU + E[u] = λqI + (1 − λ)qU = x.

Inserting from (7.14), (7.16) and (7.20) we easily obtain that

λσθ2 + (1 − λ)σs2
 
1 δ 1−λ δ 1−λ
(7.22) 0 = v+ s− q+ −1 x
zσv2 zσs2 σθ2 β σθ2 β zσθ2
σ 2 σ 2 + λσv2 σθ2 + (1 − λ)σv2 σs2
−(1 + r) s θ p.
zσv2 σs2 σθ2
By comparing coefficients we see immediately that from (7.15)
δ 1−λ
δ β zσθ2
−1 δ
(1 − λ) − zσθ2
2 β
(7.23) = λσθ2 +(1−λ)σs2
= σs 2 ,
β λσθ + (1 − λ)σs2
zσs2 σθ2

which solves for

δ zσ 2
(7.24) = s
β λ
and implying that
z2 4
(7.25) σθ2 = σs2 + σ + σu2 .
λ2 s
Now define
σs2 σθ2 + λσv2 σθ2 + (1 − λ)σv2 σs2
(7.26) ε = (1 + r) ,
σv2 σs2 σθ2
1
α = ,
σv2 ε
λσθ2 + (1 − λ)σs2
β = ,
σs2 σθ2 ε
σ 2 (1 − λ)
γ = −z s ,
σθ ελ
σ 2 (1 − λ) − σθ2 λ
δ = z s
σθ2 ελ
7.2. Compensation for adverse selection costs 201

as the coefficients for (7.15). We know that ex ante E[s] = v and E[q] = x from
the equilibrium condition, hence with the above parameters we have

(7.27) E[p] = αv + βE[s] + γE[g] + δx

= (α + β)v + (γ + δ)x
v z
= − x,
1+r ε
vσθ2 + E[θ]σv2
E[vU ] = = v.
σv2 + σθ2
Thus we obtain the expected return on this asset as
E[vU ] − E[p] rεv + z(1 + r)x
(7.28) µ= = .
E[p] εv − z(1 + r)x
It is now straightforward, although tedious, to show that the expected return is
increasing in the risk of the asset, σv2 , as well as the variance of the signal, σs2 ,
and noise trading σu2 . While the increasing returns for the risk of the asset are a
direct consequence of the compensation risk averse traders require for holding the
asset, the total compensation also includes the risk arising from the imperfect in-
formation. A less precise signal causes informed traders to trade more cautiously
and thus the price reveals less information, increasing the risk of uninformed in-
vestors. Similarly does increased noise trading cause more uncertainty about the
price emerging in equilibrium, thus increasing the risk to all traders.
Of more interest to determine the effect of adverse selection is the behavior
of the expected return as the fraction of informed traders, λ, changes. We can
derive that the expected return is indeed decreasing in λ. Obviously there cannot
be any adverse selection if λ = 1, i.e. all information is public, a case which serves
as a useful benchmark. With λ < 1 the uninformed investors make an expected
loss when trading and as the price tends to reveal less and less information this
risk increases, resulting in an a higher expected return to compensate for this
effect. We can show that the adverse selection premium, using the case of λ = 1
as a benchmark, is concave in λ.
In an alternative approach to this problem, Gârleanu and Pedersen
(2004) come to similar conclusions although their model focuses on distortions
202 Chapter 7. Asset Pricing with Trading Frictions

arising from adverse selection as the result of prices not fully reflecting all avail-
able information. In their approach traders may in the future make decisions
based on prices that are not fully revealing and thus decisions might be subopti-
mal. The resulting utility loss has to be compensated by higher expected returns
and obviously the more severe potential distortions are, the higher this return
has to be, similar to the model we used above.
Empirical evidence confirms the relationship between adverse selection and
expected returns. Easley et al. (2002) find that for every 10% of informed
trading, used as a proxy for adverse selection, the expected return increases by
2.5% p.a. An earlier investigation by Brennan and Subrahmanyam (1996)
also showed a positive relationship between adverse selection and expected re-
turns. Evaluating the listing choice of companies, Baruch and Saar (2004)
find evidence that companies seek listing at an exchange such that adverse se-
lection costs are reduced, reducing their costs of equity and increasing company
value. Thus empirical research confirms the results obtained here.

7.3 The price impact of trading

With prices essentially determined by supply and demand, it is apparent that a

trader buying an asset increases demand and a trader selling increases supply of
the asset in the market. With standard economic theory it is thus reasonable
to assume that these trading decisions affect prices, even without the implica-
tions arising from asymmetric information. Empirical evidence in Breen et al.
(2002) confirms this assertion by finding that a trade of 0.1% of outstanding
shares affects the returns by about 2.65% in the direction of the trade, at least
temporarily. It has to be noted that these price effects are not necessarily asso-
ciated with information, but merely reflect the illiquidity of the market.
Suppose now that when executing an order of size x the price P0 changes to
Pb0 as follows:

Pb0 − P0
(7.29) = λxP0 ,
P0
7.3. The price impact of trading 203

where λ is the measure of illiquidity and xP0 the trading volume.

Suppose now a risk averse investor with absolute risk aversion z invests into
this asset for T time periods which has an expected return of µ per time period
and a variance of σ 2 :

PbT − P0
(7.30) = µT + εT ,
P0

with εT ∼ N (0, σ 2 T ). Assuming that the investment is liquidated at the end of

the investment horizon at the true value PT we can now determine the trader’s
optimal investment strategy. The investor has to decide whether to invest his
initial wealth W0 into riskless bonds, B, which are perfectly liquid or into x
shares, thus with paying Pb0 for shares worth P0 we have

(7.31) W0 = (B − xPb0 ) + xP0 .

At liquidation we have with inserting from (7.29) and (7.30) and a risk free rate
of r:

(7.32) WT = (B − xPb0 )(1 + rT ) + xPT

= B(1 + rT ) + xP0 ((µ − r)T − (1 + rT )λxP0 + εT ) ,

hence

(7.33) E[WT ] = B(1 + rT ) + xP0 ((µ − r)T − (1 + rT )λxP0 ) ,

V ar[WT ] = x2 P02 σ 2 T.

Making the usual approximation that the expected utility with absolute risk aver-
sion z is E [U (WT )] = U E[WT ] − 12 zV ar[WT ] ,we obtain the optimal investment

amount as

(µ − r)T
(7.34) xP0 = .
zσ 2 T + 2λ(1 + rT )

Inserting this result back into (7.33) we get the expected utility as

1 (µ − r)2 T 2
(7.35) E [U (WT )] = B(1 + rT ) + .
2 zσ 2 T + 2λ(1 + rT )
204 Chapter 7. Asset Pricing with Trading Frictions

Let us consider a perfectly liquid asset, i.e. an asset with λ = 0, whose expected
utility would be

1 (µ0 − r)2 T
(7.36) E [U (WT0 )] = B(1 + rT ) + .
2 zσ 2

From the derivation of the CAPM we know that in equilibrium the expected
return of such an asset would follow2

(7.37) µ0 = r + zσ 2 ,

hence (7.36) reduces to

1
(7.38) E [U (WT0 )] = B(1 + rT ) + zσ 2 T.
2

In order to compensate for the illiquidity of the asset, the return should be higher
than for the perfectly liquid asset. This compensation should be such that the
expected utility of holding them is equal, thus requiring that

(µ − r)2 T 2
(7.39) zσ 2 = ,
zσ 2 T + 2λ(1 + rT )

which solves for

r
zσ 2 (zσ 2 T + sλ(1 + rT ))
(7.40) µ=r+ .
T

The liquidity premium is then given by

r
zσ 2 (zσ 2 T + sλ(1 + rT ))
(7.41) ψ = µ − µ0 = − zσ 2 .
T

Using partial derivatives we can easily obtain that the required liquidity premium
is increasing in the illiquidity λ as should be expected. With increasing illiquidity
demand for the asset is reduced as we see from equation (7.34), costs are not
increasing linearly. This effect causes the liquidity premium to be concave in the
illiquidity, a similar argument to the clientele effect in the model by Amihud
and Mendelson (1986a)
2
As we only have a single asset in our market, the covariance of the asset with the market
of course becomes the variance of the asset.
7.4. Pricing liquidity risk 205

As with a longer time horizon the costs are spread, the liquidity premium is
actually reducing. An increasing risk aversion of the investor as well as a higher
volatility of the asset the liquidity premium increases, despite the lower demand
for the asset. However the effect on the liquidity premium is very small and can
be neglected in most cases.
Assuming that a trade of 0.1% is the average trade size, we see from (7.29)
and using the result in Breen et al. (2002) that λ = 0.0265. With this result
we can calculate that the liquidity premium for a time horizon of 1 year is about
2.2% p.a., for 3 months it is 5.9% p.a. and for 1 month even 12.2% p.a. for
an average stock. Such a liquidity premium is very much in line with empirical
evidence as shown in Pástor and Stambaugh (2003).
Other models have been developed which provide very similar results to those
presented here. Pereira and Zhang (2004) use the same framework but allow
for dynamic trading strategies. Their analysis also includes an uncertain invest-
ment horizon and investors are subject to another price impact when selling their
holdings at the end of their time horizon. Huang (2003) investigates uncertain
investment horizons but focuses his analysis on the presence of borrowing con-
straints to finance consumption rather than having to sell the asset. Despite
these differences his results are compatible with the above model.

7.4 Pricing liquidity risk

The models presented thus far all implicitly assumed that the costs investors
face remain constant over time. Empirical evidence in Chordia et al. (2000),
Huberman and Halka (2000) and Hasbrouck and Seppi (2001) suggest
that various measures of liquidity do not only vary over time but are also varying
systematically across different assets. Thus costs are varying and it is reasonable
to propose that investors require a compensation for assuming this additional
risk as shown in Acharya and Pedersen (2004).
Suppose a risk averse investor with absolute risk aversion z trading N assets
which have a return of ri , proportional liquidity costs of ci and the covariance two
206 Chapter 7. Asset Pricing with Trading Frictions

assets i and j is σij = Cov[ri − ci , rj − cj ], interpreted as the covariance of the net

return. A trader will maximize the expected utility of his holding in these assets
by choosing optimal weights for each of the assets, xi , subject to the constraint
that N
P
i=1 xi = 1.
 
1
(7.42) E [U (rP )] = U E[rP ] − zV ar[rP ] ,
2
where rP denotes the net return on the portfolio of the trader. We have obvi-
hP i P
N N
ously that E[rP ] = E i=1 (ri − ci )xi = i=1 (µi − ci )xi , with µi = E[ri ] and

ci = E[ci ]. Furthermore it is that V ar[rp ] = N

P PN
i=1 j=1 xi xj σij . The Lagrange

approach thus yields the objective function

N N N
1 XX X
(7.43) L = (µi − ci )xi − z xi xj σij + λ(1 − xi ),
2 i=1 j=1 i=1

from which we obtain the optimal solution that

(7.44) µi = λ + ci + zσiP ,
PN PN
where σiP = Cov[ri − ci , rP − cP ] = j=1 xj Cov[ri − ci , rj − cj ] = j=1 xj σij .
Suppose that there exists an asset k that is risk free in its gross return and
trades at zero costs, hence as ck = 0 and σkP = 0 we obtain that µk = λ. More
commonly this risk-free rate is denoted r, thus λ = r and (7.44) becomes

(7.45) µi = r + ci + zσiP .

With the portfolio of assets being a linear combination of the individual assets,
the expected return of the portfolio becomes

(7.46) µP = r + cP + zσP2 .

With no differences between investors, the portfolio held will be identical for all
investors. In equilibrium the portfolio held will thus have to be equal to the
market portfolio, denoted by the subscript M . Using this notation and inserting
(7.46) into (7.45) we obtain
σiM
(7.47) µi = r + ci + (µM − cM − r) 2
,
σM
7.4. Pricing liquidity risk 207

which on first sight looks very similar to the traditional CAPM, but on closer
inspection the covariance term shows a significant difference:

(7.48) σiM = Cov[ri − ci , rM − cM ]

= Cov[ri , rM ] + Cov[ci , cM ] − Cov[ri , cM ] − Cov[rM , ci ].

It is apparent that only the first term, Cov[ri , rM ], is found in the CAPM while
the three other terms are associated with the liquidity risk of the assets.
The second term reflects the commonality in liquidity as found empirically
in Chordia et al. (2000), Huberman and Halka (2000) and Hasbrouck
and Seppi (2001). The rationale for requiring an additional return for this risk
is that when the covariance is high, an investor wishing to buy or sell an asset
and is faced with high liquidity costs, he is very likely to find high costs also in
other assets that otherwise might have served as suitable substitutes to trade in
order to avoid these high costs. This liquidity risk has to be compensated.
When stocks are reacting sensitively to changes in the market liquidity, as
measured by the third term, this reduces the required returns as any illiquidity
in the market as this effect is already captured in the first term. Empirical evi-
dence shows support for this relationship as shown in Pástor and Stambaugh
(2003).
If the market is not performing well, investors are usually more willing to
invest into liquid assets whose return requirements are consequently reduced as
captured in the final term above. Here investors seem to value the opportunity to
react to developments in the market at no great costs, thus it act as an insurance
against liquidity shortfalls.
As we also find that

2
(7.49) σM = V ar[rM − cM ] = V ar[rM ] + V ar[cM ] − 2Cov[rM , cM ]

we see immediately that the first term in (7.48) does not recapture the traditional
β of the CAPM, but the total risk to investors, including the liquidity risk.
Empirical work on the pricing of liquidity risk by Hasbrouck and Seppi
(2001) suggests that it can explain about two thirds of the return differences
208 Chapter 7. Asset Pricing with Trading Frictions

across stocks, while Martı́nez et al. (2003) find no evidence of liquidity risk
to be priced in the Spanish market. Acharya and Pedersen (2004) report a
total contribution of liquidity risk to the expected return of 1.1% p.a., of which
the commonality (the second term) contributes only 0.08% p.a., the sensitivity
(the third term) also only 0.16% p.a. The most important contribution the the
returns comes from the liquidity insurance (the final term) which contributes
0.82% p.a., thus becoming the most important liquidity risk for investors. The
average liquidity costs are estimated at ci = 3.5% p.a. to give a total of 4.6%
p.a. attributable to liquidity effects. Similarly Porter (2003) reports a liquidity
premium between 2 and 5% p.a.

7.5 The equity premium puzzle

The return on equity in the United States has exceeded that of Treasury Bills,
the proxy for the risk free rate, by about 6% p.a. over the last century. Such an
equity premium has been pointed out by Mehra and Prescott (1985) to be
inconsistent in traditional models with the preferences of investors as found by
other empirical work. In order to require an equity premium of 6% p.a. investors
have to be extremely risk averse with an absolute risk aversion of about 20, while
empirical evidence gained in other settings suggests an absolute risk aversion of
no more than 10 and in many cases even below 3. Also the introduction of fixed
transaction costs for trading stocks does not alleviate this problem, as the costs
would have to be prohibitively high in order to justify the observed returns. Stan-
dard economic models seem thus not be able to explain this ”puzzle”. Realistic
parameter constellations suggest a risk premium of only 0.35% p.a. as pointed
out in Mehra and Prescott (1985).
As Kocherlakota (1996) mentions, the puzzle remains robust against mod-
ifications of the underlying theory. He shows that modifying preferences, for
example, does not resolve the puzzle. The introduction of market frictions is,
however, seen as a possible solution to this problems as

”. . . the models will be stronger if they explicitly take into account

7.5. The equity premium puzzle 209

the informational problems that lead to trading frictions.”3

As the previous sections have explicitly investigated the effects such trading
frictions have on the required return of assets, it would be reasonable to summa-
rize the evidence and see how much they may be able to contribute to resolving
the equity premium puzzle.
As we have seen that a spread of 1% increases the required return by about
2.5% p.a. and noting that the average effective spread for liquid stocks on the
NYSE is about 0.2%, we obtain that these trading frictions contribute about 0.5%
p.a. to the equity premium. Furthermore traders face adverse selection costs
in the market;4 evidence suggests that about 12% of trades are conducted by
informed traders for liquid stocks on the NYSE, thus contributing about 3% p.a.
to the equity premium. We furthermore found that the price impact contributes
about 2% p.a., but we have to take into account that a part of the price impact
will be due to informational asymmetries rather than genuine illiquidity. From
Chung and Wei (2005) we can infer this amount is approximately 1.5% p.a.,
thus leaving us with 0.5% p.a. from illiquidity. Finally we found the liquidity
risk to contribute about 1% p.a. to the equity premium.
Adding these components up we obtain that trade frictions contribute about
5% p.a. to the equity premium. Together with the 0.35% p.a. to compensate
for the risk of the asset, these data suggest a total risk premium of 5.35% p.a.,
relatively close to the observed 6% p.a. As the liquidity of market has increased
in recent years, the estimate for the liquidity risk is likely to be too low for the
more distant past and would therefore underestimate the total risk premium it
is able to explain.
This result has, however, to be taken with caution. In particular, no joint
estimation has been conducted and any correlations between factor identified can
easily overestimate the total effect. Although this analysis is no evidence that
the equity premium puzzle has been solved, it provides some strong evidence that
3
Kocherlakota (1996, p.66).
4
Although adverse selection costs are included in the spread, they are the adverse selection
costs facing market makers rather than investors.
210 Chapter 7. Asset Pricing with Trading Frictions

trading frictions can indeed explain a substantial fraction of this puzzle.

Review questions 211

Review questions
1. Why is the time horizon important for the size of the premium in the
presence of a bid-ask spread?

2. Why is the expected return concave in the bid-ask spread?

3. Why do investors need to be compensated for adverse selection?

4. Why is the precision of information important for the adverse selection

premium?

5. How is the liquidity premium determined?

6. Why is the liquidity premium not linearly increasing in the illiquidity?

7. What are the components of liquidity risk?

8. Why does the liquidity premium also compensate for a covariance between
returns and trading costs?

9. How well can trade frictions explain the equity premium of stocks?

10. Why is it not appropriate to sum up the components of the equity premium
as determined in individual investigations?

Application
You observe that the excess returns of stocks vary considerably across markets.
Five different markets have been chosen and a variety of characteristics have been
recorded for the leading index in that market. These characteristics are shown
in the table below. Can the data given explain the different excess returns in the
markets?

GOB40 SRX50 HGSE25 UGR30 PBX100

Excess returns (p.a.) 0.147 0.103 0.076 0.119 0.091
Volatility (p.a.) 0.347 0.233 0.172 0.305 0.251
Volatility of cash flows (p.a.) 0.081 0.064 0.123 0.106 0.0096
β in world portfolio 1.34 1.29 1.41 1.17 1.22
Trading volume ($ bn p.a.) 107 375 275 462 327
Market capitalization ($ bn) 51 540 1021 527 616
Average spread (%) 1.52 1.34 0.81 0.48 1.24
Institutional trading (%) 24 39 74 56 93
212 Chapter 7. Asset Pricing with Trading Frictions
Outlook

This book provided an overview of the basic models used in market microstructure
theory for auction, dealer and limit order markets. We also saw the rationale
for the different market forms as well as the relationship between the market
microstructure and asset pricing.
Despite a vast amount of literature covering these aspects alone, we have
not been able to consider many other trading rules that can have a substantial
impact on the price formation process. Although it is well beyond the scope of
this book to provide an exhaustive review of all these aspects, it is nevertheless
worth mentioning a number of these trading rules such that the reader is aware
of their relevance.

Competition between market makers When investigating dealer markets

we had assumed that market makers either behave competitive or are
granted a monopoly. In reality, however, they are in many cases in di-
rect or indirect competition with each other, giving rise to partial market
power either through monopolistic competition (market makers for different
stocks) or implicit collusion (market makers for the same stocks). Dutta
and Madhavan (1997) have shown that non-competitive spreads are likely
to be observed in this case. This work also gave rise to a widespread discus-
sion on the collusion of market makers to raise spreads by avoiding certain
ticks systematically, Christie and Schultz (1994) initiated this debate
with their findings on the avoidance of odd-eighth by market makers on the
NASDAQ.

Competition between market makers and limit orders Throughout this

book we have assumed that we operate either a pure dealer market where
214 Outlook

traders are only allowed to trade with market makers via market orders or a
pure limit order market in which no market maker was present. In contrast,
we observe in most cases that dealer markets allow for the submission of
limit orders. Consequently limit orders are competing with market makers
for order flow and it is likely that this competition will affect the resulting
behavior of market makers and limit order traders. Seppi (1997) as well as
Bondarenko and Sung (2003) model such a market structure.

Order preferencing arrangements In many markets such as the NASDAQ,

brokers reach an agreement with a specific market maker to route their
entire business towards them; the market makers in return agree to match
the best available price. Brokers receive a payment for each order thus
received. With a certain order flow guaranteed these arrangements will
easily impact in the degree of competition between market makers and
thus on the spread quoted as well as the market depth. Bloomfield and
O’Hara (1998) investigate this impact in more detail.

Market entry in dealer markets In market with multiple competing market

makers, such as the NASDAQ, the way new market makers can enter the
market will have an impact on the degree of competition between market
makers with all its consequences for the trading process as Wahal (1997)
and Krause (2005) show.

Market transparency Markets differ substantially on the amount of informa-

tion that is available to market participants. Differences include the knowl-
edge of the order book beyond the best available bid and ask prices, the
anonymity of quotes and of traders. Increased information about these
aspects would without doubt increase the informational efficiency of the
market and will have implications for the behavior of market participants
as shown by Board and Sutcliffe (2000), de Frutosa and Manzano
(2005) and Foucault et al. (2004).

Opening and closing of trading In many cases the opening and closing prices
Outlook 215

on stock exchanges are determined differently from the process employed

throughout the trading day. The way the opening prices are determined
can be of relevance for subsequent trading as it does not only set a first
benchmark for the price but also affects the holdings of investors. Similarly
does the closing procedure affect the behavior prior to this. Bacidore
and Lipson (2001) as well as Ellul et al. (2003) investigate this aspect
in more detail.

Tick size Prices in markets have to be quoted at discrete prices on a given grid.
If we allow this grid to change, e.g. by reducing the tick size this will have
direct impacts on the price setting behavior of all market participants and
traders, as already seen in chapter 4.3. More detailed models are presented
in Cordella and Foucault (1999), Harris (1994) and Bourghelle
and Declerck (2004).

Block trading In many cases stock exchanges have special trading facilities for
large orders to which different trading rules apply. Such trades obviously
provide information to all market participants and the way they conducted
and how much information becomes available will affect the normal trading
process. Models in this area include Seppi (), Booth et al. (2002) and
Saar (2001).

Market fragmentation In a similar way to block trading for large orders do

many stock exchanges provide a different trading facility for small orders,
often through electronic trading. As larger orders can be split into a num-
ber of small order to be routed via that trading facility, the presence of
fragmented markets is likely to alter the behavior of market participants
and thus affect prices as Chowdhry and Nanda (1991) and Madhavan
(1995) show.

Priority rules We usually assumed that trading was firstly conducted using
strict price priority as the first priority rule, followed by time priority. In
216 Outlook

many cases, however, stock exchanges employ more complex priority rules
on the secondary priority rule, such as elements of size priority. With the
order in which orders are executed, it is obvious that prices are affected.
Moulin (2000) as well as Angel and Weaver (1998) investigate this
aspect further.

Price limits and trading halts Many stock exchanges employ circuit breakers
which prevent stocks from changing more than a predefined amount. Once
reaching these limits the implications range from a suspension of trading
for a short time to not allowing prices to go beyond the price limit for the
remaining trading session. The presence of such price limits will not only
be relevant once the price limit is reached, but also prior to that as the
prospect of the stock reaching the price limit and the subsequent inability
to trade will naturally change the behavior and thus prices. Investigations
of this topic are found in Chan et al. (2005) and Edelen and Gervais
(2003).

Ownership of stock exchanges Stock exchanges have traditionally been mu-

tual organizations dominated by brokers and dealers, only recently have
they incorporated and are now in some cases themselves listed on the stock
exchange as a public company. Such changes in the ownership of stock ex-
changes affects the incentives they have in changing the trading rules and
are thus indirectly of relevance for the price formation. Coughenour and
Deli (2002), Pirrong (1999), Pirrong (2001) and Hart and Moore
(1996) provide some insights into these topics.

Competition between stock exchanges Many stocks are listed at several

stock exchanges and traded at the same time. Consequently there will
be competition between the different stock exchanges to attract trading
volume in these stocks, see e.g. the model in Parlour and Seppi (2003).
Furthermore stock exchanges will also be competing to attract stocks for
listing in the first place, as modeled in Foucault and Parlour (2004).
Outlook 217

But recent developments did not only see increased competition between
stock exchanges, but also increased cooperations and alliances as investi-
gated by Arnold et al. (1999). All these aspects will affect the behavior
of market participants and therefore affect price formation.

In order to fully understand the operation of a stock exchange all of the

aspects mentioned above have to be investigated. Using the models developed in
these areas as well as the empirical evidence collected will help to build a market
structure that is most suitable for the stock exchange. The details of the market
structure will be affected by a large number of factors such as the characteristics
of the companies listed as well as the characteristics of the investors. We can
expect that in most cases a compromise between different aspects will have to
be reached as it is quite unlikely that all factors will lead to the same optimal
solution.
Using the provided references the reader is actively encouraged to investigate
some of the trading rules mentioned here in more detail to deepen his under-
standing of the trading process. This is particularly valuable as in many cases
the optimal rule is not what intuitively would be expected. Many competing as-
pects often give rise to market structures that would not be thought of as being
optimal.
218 Outlook
Appendices

219
Appendix A

The NASDAQ Stock Market

In chapter 1 possible market structures have been described, this appendix will
give an overview of the structure of a specific stock market, the NASDAQ Stock
Market. Although the New York Stock Exchange (NYSE ) is the dominant ex-
change not only of the United States but of the entire world,1 it faces fierce
competition, especially from the NASDAQ Stock Market.2
Increased public attention has been paid in recent years to this market as many
companies operating in fast growing sectors like information technology, biotech-
nology or telecommunications are listed on the NASDAQ. While the number of
companies listed on the NYSE grew only slowly in the past years, the NASDAQ
was able to attract a much larger number of companies. An increasing number
of non-US companies consider to be listed on the NASDAQ rather than on the
NYSE.3 This increased importance of the NASDAQ also resulted in widened at-
tention of NASDAQ trading rules in the academic literature, especially after the
findings of Christie and Schultz (1994) on implicit collusion among NASDAQ
market makers.
The NASDAQ is a dealer market with many market makers competing for the
order flow of a specific security. For most securities there are between 3 and 15
1
When mentioning ”Wall Street” this mostly is referred to the NYSE, which is located at
11 Wall Street in New York, but it is also used as a synonym for any stock exchange in the
United States.
2
The importance can be seen from the wide dispersion of computers displaying NAS-
DAQ quotes. In 1996 300,613 such computers were operated within the United States and
37,846 in other countries. The countries with the largest number of these computers were
Canada (16007 computers), Switzerland (6731) and the United Kingdom (5984), see The NAS-
DAQ Stock Market, Inc. (1997, p. 32).
3
418 foreign companies have been listed on the NASDAQ in 1997 compared to 343 on the
NYSE.
222 Appendix A. The NASDAQ Stock Market

Fig. A.1: Distribution of the number of market makers for NASDAQ securities
in 1996 Data: NASDAQ Factbook 1997

market makers, with an average of about 10 market makers per security. Every
market maker makes the market for nearly 100 securities on average. Figure A.1
shows the distribution of the number of market makers per security.

A.1 History of the NASDAQ Stock Market4

With the Securities Exchange Act of 1934 every registered securities exchange was
allowed to issue their own rules for admission to the market, listing of securities
and trading within the framework of the Act. Exchanges were established as self-
regulatory organizations. The Act did not encompass the trading taking place in
securities not listed on a securities exchange, i.e. traded on the over-the-counter
market (OTC market).
The Maloney Act of 1938 amended the Securities Exchange Act to provide
also a framework for OTC transactions. It allowed the establishment of national
4
This section follows Smith et al. (1998, pp. 3-14) unless otherwise stated.
A.1. History of the NASDAQ Stock Market 223

securities associations that would issue guidelines for OTC trading and serve as
self-regulatory organizations for their members. The only such association ever
founded was the National Association of Securities Dealers, Inc. (NASD) in
1939. As the Act allowed members of such associations to discriminate against
non-members in trading, the number of members increased from a portion of 22%
of all firms engaged in securities trading in 1939 to 83% in 1982. Those firms
not being members of the NASD where only regulated by the Securities and
Exchange Act, while members were also subject to the rules of the association.
In order to simplify and standardize supervision, in 1983 all firms engaged in the
OTC markets were forced to join a national securities association. As the NASD
was the only such association, nearly all companies trading securities became
members, with the exception of those few only trading on a registered exchange,
the NYSE, Amex or one of the five regional exchanges.5
A characteristic of OTC markets is that market participants are not central-
ized on a trading floor like the NYSE, but that they are dispersed all over the
country or even the entire world. A transaction before the start of computeri-
zation typically occurred as follows: an investor submitted an order to a broker,
the broker then tried to find out the market maker quoting the best price.6 As
a medium of communication in most cases the telephone had been used, the
broker had to phone the market makers and ask for their quotes.7 The broker
traded with the market maker quoting the best price on his own account at the
stated price. The customers had been charged a mark-up on this price to cover
the expenses of the broker. The prices of the market makers were not published
real-time due to the lack of any technologies enabling this, quotes (typically aver-
age or closing quotes) were published the following day in printed bulletins only
5
The remaining regional exchanges in the United States are: Boston Stock Exchange,
Chicago Stock Exchange, Cincinnati Stock Exchange, Pacific Stock Exchange, and Philadelphia
Stock Exchange.
6
OTC markets are typically organized as dealer markets. By having established a market
maker it is easy to find a counterpart that otherwise would hardly be found as a result of
decentralization.
7
Often it happened that market makers and brokers were identical and the search process
therefore was simplified, but it remained difficult to determine whether there existed a more
favorable price from another market maker.
224 Appendix A. The NASDAQ Stock Market

available to brokers and market makers.

Such a trading mechanism has the disadvantage that finding the best quoted
price is very difficult, time intensive and may not be found, such that transactions
occurred at less favorable prices. Transaction costs of trading will be high due to
the lack of transparency in such a market. Furthermore the reaction of investors
to new information was difficult as the informativeness of prices has been low.
These inefficiencies of OTC markets made them not very attractive for investors
and companies, hence for a long time they were no meaningful competitors to
registered exchanges. Often for small companies it was the only possibility to
raise new equity by being traded on OTC markets, most of them applied to be
listed on an exchange when they fulfilled the listing requirements.

Improvements in telecommunications and computer technologies during the

1960’s enabled quotes to be disseminated faster. In 1966 the NASD began to
consider an automated quotation system that would allow real-time quotes to
be displayed on screens connected with a central computer. The market mak-
ers would have been able to enter their quotes and the best quotes were to be
displayed on screens, mentioning the market maker quoting it. With this infor-
mation a broker could directly address the market maker quoting the best price
and was no longer forced to call all market makers in order to get this infor-
mation. Under the name National Association of Securities Dealers Automated
Quotation System (NASDAQ) this system was put into operation on February
8, 1971 by linking about 500 market makers, a large number of brokers and even
more interested parties, like investment consultants, to a central computer. Three
different levels of service had been established: The level 1 service allowed to fol-
low the market by observing the best available quotes. This service was designed
for investment consultants and the public. With a level 2 service, available for
institutional investors and brokers, it was possible to observe not only the best
quotes, but all quotes of the market makers. The names of the market makers
quoting the prices were also displayed. With level 3 service market makers were
able to enter their quotes.
A.1. History of the NASDAQ Stock Market 225

To be listed on the NASDAQ, companies had to meet minimal requirements

with respect to size and corporate governance, otherwise they were not included
into this new system.8 Trading securities listed on the NASDAQ changed signif-
icantly. The determination of the best available price had become much easier
and the transparency of the market increased. Also the settlement between bro-
kers and their customers changed, brokers did no longer charge their customers
a mark-up on the price they received from the market maker, but charged the
same price and instead used commission fees to cover their costs.
Initially the NASDAQ had been designed only to disseminate information on
quotes, information on trades having occurred could not be obtained. Improved
computer technologies, however, allowed to provide these information for 40 of the
most active securities in 1982. In due time more securities that met requirements
more restrictive than being listed on the NASDAQ were incorporated into this
new service, called NASDAQ National Market System (NASDAQ/NMS ).9 In
1983 682 securities were listed on the NMS, 2587 in 1990 and 4371 in 1996. Those
securities not included into the NMS are mostly small and infrequently traded.
These regular NASDAQ 10 securities were traded only with quote information
until 1992, when information on trades had also been added.
By launching the Small Order Execution System (SOES ) in 1984 the NAS-
DAQ became a trading platform rather than only a tool for information dis-
semination of quotes and trades. The SOES enabled orders to be automatically
routed to the market maker quoting the best price. The market maker only had
to confirm the execution of orders by pushing a button, a confirmation of order
execution is sent to the broker electronically without further personal interfer-
ence. Originally participation in the SOES was voluntary for market makers, the
use was restricted to NMS securities and order sizes of 500 shares or below. In
1985 all securities were included and the maximum order size for NMS securities
raised to 1000 shares. During the crash of 1987 market makers were difficult to

8
These requirements are stated in appendix A.3.
9
In 1993 renamed into NASDAQ National Market (NNM ).
10
In 1993 this tier of the NASDAQ has been renamed into SmallCap Market.
226 Appendix A. The NASDAQ Stock Market

reach by phone and many orders could not executed within an acceptable time. In
reaction to this experience, participation in the SOES for NMS securities became
mandatory for all market makers in 1988.
The same experience during the crash of 1987 led to the development of the
Order Confirmation Transaction Service (OCT ) in 1988,11 where orders could be
submitted electronically to a specific market maker instead of using the phone.
By pushing a button to confirm the execution of the order, this system enables
faster execution of trades, hence larger trading volumes can be processed than
by using the phone only.
Also in 1988 the Advanced Computerized Execution System (ACES ) has been
introduced. Participation in this system is voluntary for market makers and
brokers. It allows orders to be automatically routed to the best participating
market maker and the execution is again confirmed only by pushing a button.
Unlike in the SOES, the order size is not restricted by the system. Every market
maker participating in this system has to negotiate with one or more brokers up to
which order size he is willing to execute the orders at the stated prices.12 He can
negotiate different order sizes with different brokers and for different securities. A
negotiation with all brokers is not necessary. Between large brokers and market
makers similar private systems exist, especially in cases where brokers and market
makers are employed by the same financial institution.
Since these developments the systems have continually been improved to be
easier to handle and to be able to conduct an increasing number of trades. A
new system called NASDAQ Order Delivery and Execution System (NODES ) is
currently awaiting approval by the Securities and Exchange Commission. It aims
to replace and improve the current SOES and SelectNet.13
The progress in trading transparency and increased standards in regulation

11
An improved system has been introduced in 1990 under the name SelectNet.
12
As will be presented in section A.5 the trading rules require the quotes to be valid for a
minimum order size. This system enables market makers and brokers to negotiate a higher
order size bilaterally.
13
See Research Matters 1(2), 1998, p. 4, published by the NASD Economic Research De-
partment.
A.1. History of the NASDAQ Stock Market 227

Year Event
1939 Foundation of NASD
1971 NASDAQ starts operation as a quote dissemination system
1982 Introduction of a two tier market with dissemination of trade
information for the National Market
1984 SOES launched with mandatory participation
1988 OCT introduced
ACES introduced
SOES becomes mandatory for National Market securities
1992 Dissemination of trade information for the SmallCap Market

Tab. A.1: Main historical events of the NASDAQ

has made the former OTC market comparable to a securities exchange, conse-
quently in most minds it is regarded as an exchange. Table A.1 summarizes
the main events in the history of the NASDAQ. The NASDAQ is today widely
accepted by investors and companies as a market comparable to the NYSE. It
has become the largest market of the world in dollar and share trading volume
ahead of the NYSE and the second largest in market capitalization, just behind
the NYSE. In recent years it has significantly catched up with the NYSE and in
many respects surpassed it.
Most recently the NASDAQ Composite Index outperformed the Dow Jones
Industrial Average Index, which mostly consists of stock listed on the NYSE.14
In combination with more attention being paid to internet and biotechnology
stocks, which are mostly listed on the NASDAQ, the market received more and
more interest from the general public. Figures A.2 to A.4 illustrate these recent
developments.
Many large companies, although fulfilling the requirements to be listed on
the NYSE, such as Microsoft or Intel, remain to be listed on the NASDAQ and
also many foreign companies decide to be listed on the NASDAQ rather than
on the NYSE. This gives evidence that the NYSE and NASDAQ have become
equal competitors. Recent improvements in the transparency of the markets are
14
Only in late 1999 the Dow Jones Index included large companies listed on the NASDAQ,
like Microsoft or Cisco Systems.
228 Appendix A. The NASDAQ Stock Market

Data: NYSE and NASDAQ

Fig. A.2: Daily US-Dollar trading volume on the NYSE and the NASDAQ (20
day moving average) Data: NYSE and NASDAQ
A.1. History of the NASDAQ Stock Market 229

Fig. A.3: Daily share trading volume on the NYSE and the NASDAQ (20 day
moving average) Data: NYSE and NASDAQ

Fig. A.4: Development of the Dow Jones Industrial Average and NASDAQ
Composite Index Data: Datastream
230 Appendix A. The NASDAQ Stock Market

National Association of Securities Dealers, Inc.

? ? ?
NASDAQ Stock Market, Inc. Amex NASD Regulation, Inc.

- NASDAQ Stock Market

- NASDAQ National Market
- SmallCap Market

- OTC Bulletin Board

- Fixed Income Pricing System

- PORTAL Market

- Third Market

- NASDAQ International

Fig. A.5: Organization of the NASD

attributed to the competition for trading volume and the listing of companies.

A.2 The organization of the NASD15

The NASD is a self-regulatory organization supported by its members, brokers

and market makers trading on OTC markets. The NASD itself only operates
departments that serve the organization as a whole, such as economic research,
human resources or finance. All operations are conducted by three subsidiaries,
the NASDAQ Stock Market, Inc.16 , the American Stock Exchange (Amex) and
the NASD Regulation, Inc. Figure A.5 shows the organizational structure of the
NASD.
The NASDAQ Stock Market, Inc. operates the different OTC markets. It
develops and maintains the computer and telecommunications networks used
for market operations and develops new trading systems. It further promotes
15
This section is based on Smith et al. (1998, pp. 14-20).
16
The NASD currently considers plans for a going public of the NASDAQ Stock Market, Inc.
A.2. The organization of the NASD 231

the listing of securities and investments into NASDAQ securities, e.g. through
sponsoring or seminars.

The NASDAQ Stock Market, Inc. runs several OTC markets, most promi-
nent is the NASDAQ Stock Market with its two tiers, the NNM and the SmallCap
Market. These markets are mostly referred to as NASDAQ, a convention that
will also be used in this work. The other markets are of less importance and
therefore receive only limited attention. The OTC Bulletin Board (OTCBB ) is a
pure quotation system for securities not listed on the NASDAQ or any exchange.
No trade information is displayed and no trades can be conducted or initiated
through this system. The Fixed Income Pricing System (FIPS ) is a quotation
and trade information system for about 50 of the most actively traded high yield
corporate bonds (rated BB+ or lower by Standard & Poor’s). The Private Of-
ferings, Resales and Trading Through Automated Linkages (PORTAL) Market
allows private placements of securities to be better allocated by publishing in-
formation on prices and the securities themselves. This market is restricted to
institutional investors with an investment of at least USD 100 Mio. in security
markets. It also provides a platform for trading those securities, but this possibil-
ity is rarely used. In the Third Market securities listed on a registered exchange
can be traded off the exchange by using certain facilities of the NASDAQ and ap-
plying similar rules. The attempt to offer trading in NNM securities and selected
foreign securities at European trading hours at NASDAQ International operat-
ing in exactly the same way as the NASDAQ Stock Market using its computer
facilities, has not generated much interest thus far. Currently the NASDAQ is
planning to expand their activity to Japan, Canada and Europe through build-
ing up new trading platforms and seeking cooperations with established stock
exchanges and private trading platforms.

The American Stock Exchange (Amex ) has become a subsidiary of the NASD
since their merger has come into effect on October 30, 1998. It is a registered
exchange and is operated independently of the other markets.

The NASD Regulation, Inc. (NASDR), founded in 1996, has overtaken all reg-
232 Appendix A. The NASDAQ Stock Market

ulatory affairs that have formerly been conducted directly by the NASD.17 Besides
defining rules for trading on NASDAQ markets, it also supervises the compliance
to these rules and has the authority to sanction violations. If laws have been
violated it informs the legal authorities and cooperates in investigations. The
NASDR further administers written tests to qualify securities professionals and
registers them.

A.3 Listing requirements

To be listed on the NASDAQ Stock Market, a company has to meet certain
criteria regarding corporate governance, public disclosure, and size. The aim of
these criteria is to ensure a minimum of investor protection and to enable an
orderly trading process.
According to Rule 4310 all companies have to meet the following qualitative
criteria:18

• at least two independent directors on the board,

• an independent internal audit committee,

• an independent public accountant auditing the company,

• an annual report to be distributed to all shareholders,

• an annual meeting of shareholders,

• important corporate actions have to be approved by the shareholders,

• a quorum of at least 1 /3 of all outstanding shares for all decisions of the

shareholders,

• prompt disclosure of information through the media that affect the value
of the shares.
17
As Schultz (2000) points out, it was the Christie-Schultz debate that forced the NASD
to give its regulatory division more autonomy by founding a separate subsidiary.
18
These criteria can be adapted for foreign companies to meet the regulatory framework in
their country of residence.
A.3. Listing requirements 233

Additionally to these qualitative criteria, companies have to meet certain

quantitative standards. To be listed on the SmallCap Market Rule 4310 requires
the companies to meet these minimal criteria:

• net tangible assets: USD 4 Mio.,

• market capitalization: USD 50 Mio.,

or net income: USD 750,000,
or operating history: 1 year,

• number of publicly held shares: 1 Mio.,

• round lot shareholders19 : 300,

• minimum bid price: USD 4,

• number of market makers for the security: 3.

To be listed on the NNM Rule 4420 requires that one of the following three
standards has to be met:

• Standard 1:

– net tangible assets: USD 6 Mio.,

– pretax income: USD 1 Mio.,

– number of publicly held shares: 1.1 Mio.,

– market value of publicly held shares: USD 8 Mio.,

– round lot shareholders: 400,

– minimum bid price: USD 5,

– number of market makers: 3.

• Standard 2:

– net tangible assets: USD 18 Mio.,

19
Shareholders holding at least shares of one trading lot, in most cases 500 shares.
234 Appendix A. The NASDAQ Stock Market

– operating history: 2 years,

– number of publicly held shares: 1.1 Mio.,

– market value of publicly held shares: USD 18 Mio.,

– round lot shareholders: 400,

– minimum bid price: USD 5,

– number of market makers: 3.

• Standard 3:

– market capitalization: USD 75 Mio.,

or total assets: USD 75 Mio. and total revenue: USD 75 Mio.,

– number of publicly held shares: 1.1 Mio.,

– market value of publicly held shares: USD 20 Mio.,

– round lot shareholders: 400,

– minimum bid price: USD 5,

– number of market makers: 4.

Once a company is listed, it can fall short of these quantitative criteria, but not
of the qualitative criteria. In order to maintain the listing similar, less restrictive
criteria have to be met according to Rules 4310 and 4450.
Additional to these criteria, companies have to pay an entry fee for being
listed and an annual fee to maintain the listing. These fees depend on the market
on which the company is listed and its size. The entry fee for a listing on the
SmallCap Market is between USD 5,000 and USD 10,000, for a listing on the
NNM between USD 5,000 and USD 50,000. The annual fees are USD 4,000 for
the first security of a company listed on the SmallCap Market and USD 1,000
for each additional security. In the NNM this fee varies between USD 5,250 and
USD 20,000.20
20
See Rules 4510 and 4520.
A.4. Registration as broker and market maker 235

A.4 Registration as broker and market maker

There exist two prerequisites to register as broker or market maker. The first
concerns the capital requirements to ensure those market participants to con-
duct their duties without facing the threat of bankruptcy. These prerequisites
are regulated by the Securities and Exchange Commission (SEC ). The other
prerequisites refer to their qualifications and are regulated by the NASD.
A broker has to maintain a net capital21 of at least USD 100,000. A market
maker needs a net capital of USD 2,500 for each security he makes the market
in.22 Additionally, a minimum of USD 100,000 and a maximum of USD 1 Mio.
applies. In most cases brokers and market makers are companies rather than
individuals, whose business is conducted by employees. In this case the company
as a whole has to fulfill these requirements, it has not to be fulfilled for every
single employee acting as market maker or broker.23
The Securities Exchange Act requires every broker and market maker acting
on OTC markets, like the NASDAQ Stock Market, to be member of a national
securities association, hence they have to be member of the NASD. Not only the
brokerage companies and companies acting as market makers have to be regis-
tered, but the by-laws of the NASD require every employee of those companies
who is involved in brokerage or market making activities to become a member.
While companies are registered with approval of the SEC, their personnel has
to prove their qualifications to become members. Without being registered as
member, no individual is allowed to conduct businesses related to brokerage or
market making.
The by-laws of the NASD require members to have an appropriate qualifica-
tion to conduct the business they are assigned to.24 According to Rules 1021,
1031 and 1041 these qualifications have to be shown by passing a qualification
21
Net capital is the net worth adjusted for unrealized profits and losses, subordinated loans
and similar.
22
For securities with a market value of less than USD 5 per share the requirement is USD
1,000.
23
See SEC Rule 15c3-1.
24
See Article III, Section 2 of the by-laws of the NASD.
236 Appendix A. The NASDAQ Stock Market

examination conducted by the NASDR. These examinations are designed to ex-

plore the qualifications for a specific duty. If the duties of a person change, it can
be necessary to pass another examination for his new duties. After having passed
the examination the person is registered as member and allowed to conduct the
business he has been assigned to. Furthermore, Rule 1120 requires registered
members to follow certain continuing education requirements.
When registered as broker, one is free to act as broker for all securities listed
on the NASDAQ Stock Market. Being registered as market maker also allows
to become market maker in every listed security, it is only necessary to register
for the securities one wants to make a market in. Market making can begin the
next trading day.25 With registration as market maker for a specific security it
is the obligation to quote always a price at which one is willing to buy and sell
the security.26
To withdraw the registration as market maker for a specific security follows a
similar process. A request to withdraw the registration for this security becomes
effective the next trading day. The only restriction faced upon withdrawal is that
it is not allowed to register again as market maker for the same security during
the next 20 trading days.27 This free market entry and exit in most cases leads to
more than two market makers being registered for a security. Furthermore, it is
the aim to prevent registered market makers from making extraordinary profits
through their activities by imposing the threat of new market makers entering.

A.5 Trading rules

A market maker registered for a security has the obligation to quote prices both
for buying and selling the security from the public during the trading hours from
9.30 am to 4 pm.28 The monthly average spread a market maker quotes for a
security must not exceed 150% of the average spread of all market makers for
25
See Rule 4611.
26
See Rule 4613. Section A.5 describes the trading rules in more detail.
27
See Rule 4620.
28
See Rules 4613 and 4617. Extending trading hours to 8 or 10 pm is currently considered
by the NASDAQ.
A.5. Trading rules 237

this security. Rule 2440 and Interpretation IM-2440 require the market makers
not to charge a too large spread. As a guideline a maximum spread of 5% is
mentioned in this rule, depending on the circumstances, e.g. market conditions
and characteristics of the security. Larger spreads can be justifiable, but also a
spread of 5% may be viewed as too large by supervisors of the NASDR, forcing
market makers to reduce their spread. However, no fixed rule can be applied to
determine the maximum spread, it is subject to interpretations by the NASDR.
The quotes are further restricted by tick sizes, the increments have to be
multiples of USD 1 /32 for securities with bid prices below USD 10 and USD 1 /16
for those above.29 Quotes have to be firm, i.e. upon request the market maker
has to trade at least at the stated prices, but he is free to choose a more favorable
price for the transaction.30 The obligation of a firm quote is only waived for a
short period of time to enable the market maker an update of his quotes after
having executed an order.31
Furthermore, quotes have to be valid at least for a normal trading size, a lot
of 100 shares. The number of shares a market maker is willing to trade at the
quotes are displayed on the screen next to their quote. Rule 4613 requires the
minimum trade sizes for which the quotes have to be valid to be larger than 100
shares under certain conditions. For securities listed on the SmallCap Market
the minimum trade sizes is 500 shares if the average daily non-block volume32
exceeds 1000 shares or the bid price is below USD 10.
For securities listed on the NNM these limits are:

• 1000 shares if

– the average daily non-block volume is above 3000 shares,

29
Changing to a decimal system is considered for late 2000. Through 1997 the tick size has
been USD 1 /8 for securities with a bid price above USD 10 and USD 1 /16 for securities with a
bid price of USD 10 or below.
30
When choosing a more favorable price he is not restricted to the tick sizes in determining the
price he charges. The tick sizes only apply to quotes, not to transaction prices. More favorable
prices than those quoted are frequently observed as a result of preferencing arrangements, which
are described below.
31
See Rule 3320 and interpretation IM-3320.
32
For the definition of block trades see below.
238 Appendix A. The NASDAQ Stock Market

– the bid price is below USD 100, and

– there are at least 3 market makers for this security.

• 500 shares if

– the average daily non-block volume is above 1000 shares,

– the bid price is below USD 150, and

– there are at least 2 market makers for this security.

• 200 shares if

– the average daily non-block volume is below 1000 shares,

– the bid price is below USD 250, and

– there are at least 2 market makers for this security.

Orders that are larger than the market makers are willing to accept, can be
broken into parts of at least a lot and be executed like several smaller orders.
This may result in different prices applied for each part and the parts may be
executed by different market makers. Limit orders may also be executed in parts
of at least a lot with offsetting orders. To avoid partial execution the order has
to be specially marked as a All-or-None order by the investor.
Trades of 10,000 shares and above are called block trades. Such trades are
subject to special treatment. They can either be traded through market or limit
orders as a whole or be broken into several smaller orders within the normal
trading procedure. Investors face the risk of influencing the price significantly
in an unfavorable way through the placement of such an order. For this reason
such orders are usually traded in a special market (upstairs market) for separate
negotiation with other block trades.
An order arriving on the market has to be executed at the best available
price (price priority), i.e. the market maker quoting the most favorable price has
to execute the order at the stated or an even more favorable price.33 If several
33
See Rule 2320.
A.5. Trading rules 239

market makers quote the same price, the market maker executing the order can
be chosen without restrictions by the brokers. Preferencing arrangements, as
described below, are applied in most of these cases to determine the routing of
the order flow.
Price priority and interpretation IM-2110-2 of Rule 2110 give the guidelines
for handling limit orders. Limit orders can be accepted by market makers, but
they do not have to be. By accepting a limit order, the market maker has to follow
the established rules. He must not trade ahead of a limit order he has received,
i.e. is not allowed to execute an offsetting order on his own account at the same
or a less favorable price than the limit has been set (public before dealer ). A
market maker can immediately execute limit orders on his own account or can
route them to other market makers. To enhance the transparency of the market,
SEC Rule 11Ac1-4 requires unexecuted limit orders to be displayed in the quotes
of a market maker. If the limit order has the best available price, also its size has
to be displayed. The obligation is only waived for orders below 100 and above
10,000 shares and if it must not be partially executed (All-or-None orders).
The aim of these rules on the handling of limit orders is to guarantee a max-
imum of transparency and enhance competition further by allowing limit orders
directly to compete with the quotes of market makers.34
As has been stated above, preferencing in most cases determines the market
maker who executes an incoming order in the case where several market makers
quote the best price. With preferencing a broker routes his entire order flow to
one specific market maker, provided he quotes the best available price and price
priority can be applied.35 The two main reasons for such a behavior are either
34
Through 1994 limit orders were interpreted to be offers of investors to trade with a market
maker at the stated price. Hence market makers could trade with other investors on their own
accounts at less favorable prices, i.e. higher ask and lower bid prices. Limit orders were only
executed against quotes of the market makers and not orders from other investors, consequently
they also have not been published. From 1994 onwards, market makers were still allowed to
trade ahead of limit orders, but only if they quoted the same or a more favorable price, limit
orders had not to be published. In 1997 the current regulation has been introduced. Allegations
of market makers colluding on wide spreads in 1994 lead to these changes in the handing of
limit orders to enhance competition.
35
In many cases preferencing arrangements require that the entire order flow is routed to
a specific market maker, regardless of his current quotes. To fulfill the requirement of price
240 Appendix A. The NASDAQ Stock Market

internalization or payment-for-order-flow.
In many cases companies act both as broker and market maker. In this case
vertical integration results in preferencing, what is also called internalization. To
maximize profits, the brokerage department has to route all orders to the own
market makers if they quote the best available prices.
A broker may also be willing to route his order flow to a specific market maker
because he receives a payment from him (payment-for-order-flow ). This payment
can either be in form of cash, or the market maker charges a more favorable price
to the broker than his quote. The broker can either receive this difference to the
quoted price by charging his customer the quoted price or he can forward the
whole or a part of this surplus to his customer and gain a competitive advantage
over other brokers, either by charging a more favorable price or by reducing his
commission fees. Other forms of payments can also include various services, e.g.
research reports on companies or conducting the clearing process. Payments
typically have a value between USD .01 and USD .02 for each share.
Those market makers and brokers participating in ACES or similar private
arrangements are very likely in preferencing arrangements. Preferencing adds
another source of competition between market makers, besides price competition
they also compete for trading volume.

A.6 Summary
The NASDAQ has grown out of a telecommunications network for disseminat-
ing quote information to a network having all features of an exchange. This
development has been made possible by improvements in computer and telecom-
munications technologies. Differences to registered computerized exchanges are
only minor nowadays, so that the NASDAQ is mostly referred to as an exchange,
although it misses this formal status and is ”only” an OTC market.
The NASDAQ is characterized by the competition of market makers for the
priority for investors, the market makers in turn guarantee to charge only the best available
price, i.e. if necessary they improve their quotes. Such arrangements are also called price
matching arrangements.
A.6. Summary 241

best price and for trading volume. If admitted as market maker to the NASDAQ,
there are virtually no entry barriers for market making in a specific security. This
allows for hit-and-run competition, ensuring no extraordinary profits to be made
by market makers, hence low spreads should be expected.36 Together with rules
ensuring high standards of transparency for the market, trading costs should
be low. Other stock exchanges, like the NYSE, have to establish a much more
complex set of rules to abandon the use of market power that arises as the result
of high entry barriers or a lack of competition by granting monopolies of market
making.
Listing requirements of the NASDAQ are much less restrictive compared to
those of other stock exchanges, e.g. the NYSE, as it is designed for small compa-
nies. These small, in many cases highly innovative companies operating in fast
growing industries, made the NASDAQ well known to be a market for high-tech
stocks. High market transparency and a seemingly competitive trading environ-
ment induced many companies having grown to sizes that qualify for a listing
on the NYSE to remain their listing on the NASDAQ. It has become the major
competitor of the NYSE for the listing of companies and the NASDAQ Com-
posite Index is one of the most important indices of the world, having received
increased attention in recent years.

36
This result can best be described with the theory of contestable markets. It states that the
threat of new market participants entering in case of excess profits, forces market incumbents
to charge competitive prices. The absence of entry barriers (legal restrictions, sunk costs) are
the prerequisites for a contestable market. Baumol et al. (1988) provide a detailed overview
of the theory of contestable markets. However, implicit collusion between market makers will
enable them, despite these competitive forces, to quote noncompetitive prices and receive excess
profits.
242 Appendix A. The NASDAQ Stock Market
Appendix B

Mathematical methods

The aim of this appendix is to provide a short introduction to selected topics

which are used in the main parts of this books. Necessarily the coverage is not
comprehensive and may at times be incomplete for the sake of simplicity. For
a more detailed overview the reader is referred to specialist literature on the
topics. Generally it is assumed that the reader is familiar with the basics of
analysis, algebra and statistics.

B.1 Taylor series expansion

Let f : D 7→ R be (n + 1)-times differentiable on D with D ⊂ R.1 We then can
define a Taylor polynome by

(B.1) Tn : D 7→ R
n
X f (k) (a)
x 7→ Tn ≡ (x − a)k ,
k=0
k!

where a ∈ D and n ≥ 0 and f (k) denotes the kth derivative of f .

Let there exist an open interval D0 ⊂ D, a ∈ D and two constants c > 0 and
M > 0 such that for all k ≥ 0 and x ∈ D0

(B.2) f (k) (x) ≤ ck!M k .

If this condition is fulfilled, we can expand f into a Taylor series:

∞
X f (k) (a)
(B.3) f (x) = (x − a)k
k=0
k!
1
We can derive similar results for multidimensional functions f : D 7→ RM with D ⊂ RN .
244 Appendix B. Mathematical methods

1
for all x ∈ D0 with |x − a| < M
.
As higher order terms become arbitrarily small, we can approximate f by
n
X f (k) (a)
(B.4) f (x) = (x − a)k
k=0
k!

and call this an expansion of f into a nth order Taylor series around a. In
practice it is always assumed that the conditions for a Taylor series expansion
are met.

B.2 Dynamic programming2

When individuals have not only to make decisions that are optimal at a given
point of time by choosing an optimal value of the control variables, u, but that
these values have to be chosen optimal over a certain period of time, they have
to determine an optimal time path for these variables. In many cases an ad-
ditional problem arises, that the environment determining the outcome of this
optimization problem changes through a changing state variable, x, which may
be influenced by the control variables.
We define a function I(x, u, t) which measures the payoff at a certain point
of time, t. The aim of the individual now is to maximize the payoffs he receives
over time, i.e. the control problem is given by3
Z t1
(B.5) max J = I(x, u, t)dt,
u(t) t0

where t0 and t1 denote the starting and end point of the considerations. The
state variable changes according to the differential equation

∂x
(B.6) = f (x, u, t).
∂t

We define the solution to the control problem by J ∗ (x, t) and call this the optimal
performance function.
2
This section is based on Intriligator (1971).
3
In dynamic programming it is usually assumed that future payoffs are not discounted to
their present value.
B.3. Constrained optimization 245

The principle of optimality now requires that regardless of the current state
the remaining decisions have to be optimal. Therewith at point t + ∆t with state
x + ∆x the optimal performance function has to be J ∗ (x + ∆x, t + ∆t). We can
now write the optimal performance function as

(B.7) J ∗ (x, t) = max{I(x, u, t)∆t + J ∗ (x + ∆x, t + ∆t)},

u(t)

which is known as the fundamental recurrence relation. Here I(x, u, t)∆t denotes
the payoff in the interval ]t, t + ∆t]. Approximating the second term in brackets
by a first order Taylor series around (x, t), we get
∂J ∗ ∂J ∗
(B.8) J ∗ (x + ∆x, t + ∆t) = J ∗ (x, t) + ∆x + ∆t,
∂x ∂t
which gives after inserting into (B.7):
∂J ∗ ∂J ∗
(B.9) 0 = max{I(x, u, t)∆t + ∆x + ∆t}.
u(t) ∂x ∂t
Dividing by ∆t and taking the limit ∆t → 0 we get with
∆x ∂x
(B.10) lim = = f (x, u, t)
∆t→0 ∆t ∂t

∂J ∗ ∂J ∗
 
(B.11) − = max I(u, x, t) + f (x, u, t) .
∂t u(t) ∂x
This partial differential equation is known as the Bellman equation. Solving this
equation will give the optimal performance function, given boundary conditions.

B.3 Constrained optimization

We want to find the optimal solution to a problem by maximizing (or minimizing,
which works similar to the method demonstrated here) an objective function. In
many cases however the individual conducting this optimization faces one or more
constraints, such as a limited budget to buy goods or assets. Suppose we face
the following problem:

(B.12) max f (x, y)

(x,y)∈R

s.t. g(x, y) = 0.
246 Appendix B. Mathematical methods

If the constraint is originally that h(x, y) = c, we simply define g(x, y) = h(x, y)−
c and obtain the above constraint. With this setting we can now define the
Lagrangian function as

(B.13) max f (x, y) + λg(x, y),

(x,y,λ)∈R×R

where λ is the Lagrangian multiplier. This equation is now equivalent to the

initial problem (B.12) as easily can be seen. We solve this maximization problem
by analyzing the following equations simultaneously:

(B.14) fx + λgx = 0,

fy + λgy = 0,

g(x, y) = 0,

where the indices x and y indicate the partial derivative of the function with
respect to this variable. We see that the final condition recovers our constraint.
The second order condition for this optimization problem is then derived from
the Hesse-matrix being negative definite:
 
fxx − λgxx fxy − λgxy gx
(B.15) H(x, y) =  fxy − λgxy fyy − λgyy gy  .
gx gy 0
The Lagrange multipliers also have an economic interpretation. Consider the
∂f (x,y)
initial constraint that h(x, y) = c, we then can write that ∂c
= fx ∂x
∂c
+ fy ∂y
∂c
.
Using the optimality conditions in (B.14) and noting that gx = hx and gy = hy
we obtain
 
∂f (x, y) ∂x ∂y
(B.16) = λ hx + hy
∂c ∂c ∂c
= λ,

where the final equation follows from the fact that h(x, y) = c implies hx ∂x
∂c
+
hy ∂y
∂c
= 1.
We can thus interpret the value of λ as the shadow value of the constraint, i.e.
by how much the objective function would change if we changed the constraint
marginally. In cases where the constraint is a budget constraint, we can interpret
λ as the marginal utility of this budget.
B.4. Conditional moments 247

B.4 Conditional moments

B.4.1 Truncated distributions

We have a random variable x which has a distribution f (x). Let us assume

that we know that the value of x is above a threshold t. We are interested in
the distribution of x given the information that x > t. Using Bayes’ rule we
immediately see that this distribution f (x|x > t) is given by

f (x)
(B.17) f (x|x > t) = .
P rob(x > t)

This function is also referred to as the truncated distribution. The moments of

this distribution are now simply given by

Z +∞
(B.18) E[x|x > t] = xf (x|x > t)dx,
t
Z +∞
V ar[x|x > t] = (x − E[x|x > t])2 f (x|x > t)dx.
t

A particularly simple case can be obtained for the normal distribution f (x) =
N (µ, σ 2 ). In this case we can show that

1 x−µ


σ
φ σ
(B.19) f (x|x > t) =
,
1− Φ t−µ
σ

where φ(·) denotes the standard normal distribution and Φ(·) its cumulative
distribution. The moments of this truncated normal distribution can be found as

φ t−µ


σ
(B.20) E[x|x > t] = µ + σ
,
1 − Φ t−µ
σ
!!
φ t−µ φ t−µ



2 σ σ t − µ
V ar[x|x > t] = σ 1 −
− .
1 − Φ t−µ 1 − Φ t−µ


σ σ
σ

For other forms of truncation, such as f (x|x < t), f (x|t0 < x < t) as well as
other distribution functions, see Johnson and Kotz (1970) for a comprehensive
overview.
248 Appendix B. Mathematical methods

B.4.2 Joint distributions

Suppose we have two random variables x and y who have a joint distribution
f (x, y). The marginal distributions are defined as
Z
(B.21) fx (x) = f (x, t)dt,
y
Z
fy (y) = f (t, y)dt.
x

They are thus the probability distributions in only one variable. Suppose now
that we know the value of one of the two variables, say x. We now want to obtain
the probability distribution of y conditional on the value of x. We can use Bayes’
theorem to obtain that in that case the distribution function is given by

f (x, y)
(B.22) f (y|x) = .
fx (x)

This expression serves as the new density function, the conditional mean and
variance are obviously given by
Z
(B.23) E[y|x] = yf (y|x)dy,
y
Z
V ar[y|x] = (y − E[y|x])2 f (y|x)dy.
y

This expression is in general very difficult to evaluate as the marginal distributions

can be very different from the original joint distribution. The exception to this
is the multi-variate normal distribution, whose marginal distributions are normal
and for whose conditional moments we have some very simple formulae.
Let µ = (µx , µy ) be the vector of expected values of the two variables x
and y, σx2 and σy2 their variances and σxy their covariance. If x and y have a
 2 
σx σxy
joint normal distribution with mean µ and covariance matrix Σ = ,
σxy σy2
f (x, y) = N (µ, Σ), we have as the marginal distributions fx (x) = N (µx , σx2 )
and fy (y) = N (µy , σy2 ) and the conditional distribution is given by f (y|x) =
σ2
 
N µy + σσxy2 (x − µx ), σy2 − σxy2 .
x x
B.5. Stochastic processes 249

The conditional moments are thus given by

σxy
(B.24) E[y|x] = µy + (x − µx ),
σx2
2
2
σxy
V ar[y|x] = σy − 2 .
σx
We use such distributions for example in cases where investors are not knowing
the fundamental value of an asset, y, but receive a signal x which is correlated
with the fundamental value. This signal is then used to update the information
the investor has on the fundamental value in the way shown above.
For other joint distributions or other forms of conditioning, Johnson and
Kotz (1970) provide a comprehensive overview.

B.5 Stochastic processes

B.5.1 Definition

Let us consider a variable that changes randomly over time investigate this vari-
able at fixed intervals of length ∆t. The easiest form the random variable xt can
evolve is given by

(B.25) xt+∆t = xt + µ∆t + ε0t ,

where ε0t ∼ iidN (0, σ 2 ∆t). We can now define ∆xt = xt+∆t − xt and the above
equation becomes ∆xt = µ∆t + εt . If we define a standard normally distributed
variable εt ∼ iidN (0, 1) equation (B.25) can be rewritten as
√
(B.26) ∆xt = µ∆t + εσ ∆t.

Letting ∆t becoming very small, hence investigating the variables at ever smaller
periods of time, will allow us to obtain the values in continuous time and we write
this as
√
(B.27) dx = µdt + σε dt,

which is also known as the Brown-Wiener process. With µ = 0 σ = 1 we obtain

the standard Wiener process
√
(B.28) dz = ε dt
250 Appendix B. Mathematical methods

and it is quite common to write (B.27) as

(B.29) dx = µdt + σdz.

Although this stochastic process is the most widely used, there exist a wide variety
of stochastic processes which have been widely investigated. Todorovic (1992)
provides an introduction to the topic.

B.5.2 Itô’s Lemma

One of the most fundamental results on stochastic processes is Itô’s lemma. It

essentially states how to differentiate a function F (x, t) that contains a stochastic
process, x. Using the Taylor series expansion of F we obtain

∂F ∂F
F (x + ∆x, t + ∆t) ≈ F (x, t) +
(B.30) ∆x + ∆t
∂x ∂t
1 ∂2F ∂2F ∂2F
 
2 2
+ (∆x) + 2 ∆x∆t + 2 (∆t) .
2 ∂x2 ∂x∂t ∂t

In letting ∆t becoming very small we obtain as above

1 ∂2F ∂2F ∂2F

 
∂F ∂F 2 2
(B.31) dF (x, t) ≈ dx + dt + (dx) + 2 dxdt + 2 (dt) .
∂x ∂t 2 ∂x2 ∂x∂t ∂t

With dx = µdt + σdz we obtain easily that

∂F ∂F 1 ∂2F 2
(B.32) dF (x, t) = dx + dt + σ dt,
∂x ∂t 2 ∂x2

which is known as Itô’s lemma. To obtain the last equation the important as-
sumption is made that as ∆t → 0, all terms with a higher than linear order will
vanish.
Appendix C

Economic concepts

This appendix provides a brief overview of key economic concepts used implicitly
or explicitly in this book. It does usually not provide a critique of the concepts
presented nor does it allow for alternative approaches to the problems being high-
lighted. For a more comprehensive coverage, the reader is referred to specialist
literature on the subjects. In compiling this appendix it was assumed that the
reader is familiar is basic microeconomic theories.

C.1 Utility theory

Since John von Neumann and Oskar Morgenstern introduced the expected utility
hypothesis in 1944, it has become the most popular criterion for modeling deci-
sions under risk. This appendix will give a brief introduction into the reasoning
behind expected utility and its implications for risk aversion.

C.1.1 The expected utility hypothesis

The value, and therewith the returns of assets depend on their future cash flows.
These future cash flows usually cannot be predicted with certainty by investors,
they are a random variables, hence returns are also random variables. Investment
decisions therewith have to be made under risk.1 In their work von Neumann
1
According to Knight (1921, pp. 197ff.) a decision has to be made under risk if the outcome
is not known with certainty, but the possible outcomes and the probabilities of each outcome are
known. The probabilities can either be assigned by objective or subjective functions. Keynes
(1936, p. 68) defines risk as the possibility of the actual outcome to be different from the
expected outcome. In contrast, under uncertainty the probabilities of each outcome are not
known or even not all possible outcomes are known. Cymbalista (1998) provides an approach
of asset valuation under uncertainty. In this work we only consider decisions to be made under
252 Appendix C. Economic concepts

and Morgenstern (1953) presented a criterion to make an optimal decision if

five axioms are fulfilled.2
Let A = {a1 , . . . , aN } be the set of all possible alternatives an individual
can choose between,3 S = {s1 , . . . , sM } all states that affect the outcome,4 and
C = {c11 , . . . , c1M ; . . . ; cN 1 , . . . , cN M } the outcomes, where cij is the outcome if
state sj occurs and alternative ai has been chosen.

Axiom 1. A is completely ordered.

A set is completely ordered if it is complete, i.e. we have either ai  aj or

aj  ai for all i, j = 1, . . . , N and ”” denotes the preference. The set has
further to be transitive, i.e. if ai  aj and aj  ak we then have ai  ak . Axiom
1 ensures that all alternatives can be compared with each other and are ordered
consistently.5
To state the remaining axioms we have to introduce some notations. Let any
alternative be denoted as a lottery, where each outcome cij has a probability of
pij . We can write alternative i as

M
X
ai = [pi1 ci1 , . . . , piM ciM ] , where pij = 1 for all i = 1, . . . , N.
j=1

Axiom 2 (Decomposition of compound lotteries). If the outcome of a lottery

is itself a lottery (compound lottery), the first lottery can be decomposed into its
final outcomes:
Let ai = [pi1 bi1 , . . . , piM biM ] and bij = [qij1 c1 , . . . , qijL cL ]. With p∗ik =
PM
l=1 pil qilk 6 we have [pi1 bi1 , . . . , piM biM ] ∼ [p∗i1 c1 , . . . , p∗iM cL ]

risk.
2
Many different ways to present these axioms can be found in the literature. We here follow
the version of Levy and Sarnat (1972, p. 202)
3
As for N = 1 there is no decision to make for the individual it is required that N ≥ 2.
4
As with M = 1 the outcome can be predicted with certainty we need M ≥ 2 possible states.
5
The transitivity ensures consistent decisions of individuals. It is equivalent with the usual
assumption in microeconomics that indifference curves do not cross.
6
This representation of joint probabilities assumes that the two lotteries are independent. If
the two lotteries where not independent the formula has to be changed, but the results remain
valid. It is also assumed throughout this appendix that there is no joy of gambling, i.e. that
there is no gain in utility from being exposed to uncertainty.
C.1. Utility theory 253

Axiom 3 (Composition of compound lotteries). If an individual is indifferent

between two lotteries, they can be interchanged into a compound lottery:
If ai = [pi1 bi1 , . . . , pij bij , . . . , piM biM ] and bij ∼ [qij1 c1 , . . . , pijL cL ] then
ai ∼ [pi1 bi1 , . . . , pij [qij1 c1 , . . . , pijL cL ], . . . , piM biM ].

These two axioms ensure that lotteries can be decomposed into their most
basic elements (axiom 2) and that more complex lotteries can be build up from
their basic elements (axiom 3).

Axiom 4 (Monotonicity). If two lotteries have the same two possible outcomes,
then the lottery is preferred that has the higher probability on the more preferred
outcome:
Let ai = [pi1 c1 , pi2 c2 ] and bi = [qi1 c1 , qi2 c2 ] with c1  c2 , if pi1 > qi1 then
ai  b i .

Given the same possible outcomes this axiom ensures the preference
relation”” to be a monotone transformation of the relation ”>” between prob-
abilities.

Axiom 5 (Continuity). Let ai , bi and ci be lotteries. If ai  bi and bi  ci then

there exists a lottery di such that di = [p1 ai , p2 ci ] ∼ bi .

This axiom ensures the mapping from the preference relation ”” to the
probability relation ”>” to be continuous.
The validity of these axioms is widely accepted in the literature. Other axioms
have been proposed, but the results from these axioms are identical to those to
be derived in an instant.
Given these assumptions the following theorem can be derived, where U de-
notes the utility function.

Theorem 1 (Expected utility principle). An alternative ai will be preferred to

an alternative aj if and only if the expected utility of the former is larger, i.e

ai  aj ⇔ E [U (ai )] > E [U (aj )] .

254 Appendix C. Economic concepts

Proof. Define a lottery ai = [pi1 c1 , . . . , piM cM ], where without loss of generality

c1  c2  · · ·  cM . Such an order is ensured to exist by axiom 1.
Using axiom 5 we know that there exists a lottery such that

ci ∼ [ui1 c1 , ui2 cM ] = [ui c1 , (1 − ui )cM ] ≡ c∗i .

We can now use axiom 3 to substitute ci by c∗i in ai :

ai ∼ [pi1 c∗1 , . . . , piM c∗M ].

This alternative only has two possible outcomes: c1 and cM . By applying axiom
2 we get
ai ∼ [pi c1 , (1 − pi )cM ]
PM
with pi = j=1 uij pij , what is the definition of the expected value for discrete
random variables: pi = E[ui ].7
The same manipulations as before can be made for another alternative aj ,
resulting in
aj ∼ [pj c1 , (1 − pj )cM ]

with pj = E[uj ]. If ai  aj then we find with axiom 4 that, as c1  cM :

pi > p j .

The numbers uij we call the utility of alternative ai if state sj occurs. The
interpretation as utility can be justified as follows: If ci  cj then axiom 4
implies that ui > uj , we can use ui to index the preference of the outcome, i.e. a
higher u implies preference for this alternative and vice versa.
Therewith we have shown that ai  aj is equivalent to E [U (ai )] > E [U (aj )].

The criterion to choose between two alternatives, is to take that alternative

with the highest expected utility. To apply this criterion the utility function has
to be known. As in most cases we do not know the utility function, it is necessary
to analyze this criterion further to derive a more handable criterion.
7
The extension to continuous random variables is straightforward by replacing the proba-
bilities with densities.
C.1. Utility theory 255

C.1.2 Risk aversion

”Individuals are risk averse if they always prefer to receive a fixed

payment to a random payment of equal expected value.”8

From many empirical investigations it is known that individuals are risk

averse, where the degree of risk aversion differs widely between individuals.9 The
Arrow-Pratt measure is the most widely used concept to measure this risk aver-
sion. We will derive this measure following Pratt (1964), a similar measure has
independently also been developed by Arrow (1963).
With the definition of risk aversion above, an individual prefers to receive
a fixed payment of E[x] to a random payment of x. To make the individual
indifferent between a fixed payment and a random payment, there exists a number
π, called risk premium, such that he is indifferent between receiving E[x] − π and
x. By applying the expected utility principle we see that the expected utility of
these two payments has to be equal:

(C.1) E [U (x)] = E [U (E [x] − π)] = U (E [x] − π) .

The term E [x] − π is also called the cash equivalent of x. Approximating the left
side by a second order Taylor series expansion around E [x] we get

(C.2) E [U (x)] = E U (E [x]) + U 0 (E[x])(x − E[x])

1 00 2
+ U (E[x])(x − E[x])
2

= U (E[x]) + U 0 (E[x])E[x − E[x]]

1
+ U 00 (E[x])E[(x − E[x])2 ]
2
1
= U (E[x]) + U 00 (E[x])V ar[x].
2
where U (n) (E[x]) denotes the nth derivative of U with respect to its argument
evaluated at E[x]. In a similar way we can approximate the right side by a first
8
Dumas and Allaz (1996, p. 30), emphasize added.
9
Despite this clear evidence for risk aversion, many economic theories assume that individu-
als are risk neutral. Prominent examples in finance are the information-based models of market
making (see section 3.2) and several asset pricing theories.
256 Appendix C. Economic concepts

order Taylor series around E[x] and get

(C.3) U (E[x] − π) = U (E[x]) + U 0 (E[x])π.

Inserting (C.2) and (C.3) into (C.1) and solving for the risk premium π we get

U 00 (E[x])
 
1
(C.4) π= − 0 V ar(x).
2 U (E[x])

Pratt (1964) now defines

U 00 (E[x])
(C.5) z=−
U 0 (E[x])

as the absolute local risk aversion. This can be justified by noting that the risk
premia has to be larger the more risk averse an individual is and the higher the
risk. The risk is measured by the variance of x, V ar[x],10 hence the other term
in (C.4) can be interpreted as risk aversion. Defining σ 2 = V ar[x] we get by
inserting (C.5) into (C.4):

1
(C.6) π = zσ 2 .
2

If we assume that individuals are risk averse, we need π > 0, implying z > 0. It
is reasonable to assume positive marginal utility, i.e. U 0 (E[x]) > 0, which implies
that U 00 (E[x]) < 0. This relation is also known as the first Gossen law and states
the saturation effect. The assumption of risk aversion is therefore in line with
the standard assumptions of microeconomic theory.
The conditions U 0 (E[x]) > 0 and U 00 (E[x]) < 0 imply a concave utility func-
tion. The concavity of the function (radius) is determined by the risk aversion.11
Figure C.1 visualizes this finding for the simple case of two possible outcomes,
x1 and x2 , having equal probability of occurrence.

10
A justification to use the variance as a measure of risk is given in appendix C.2.
11
For risk neutral individuals the risk premium, and hence the risk aversion, is zero, resulting
in a zero second derivative of U , the utility function has to be linear. For risk loving individuals
the risk premium and the risk aversion are negative, the second derivative of the utility function
has to be positive, hence it is convex.
C.2. Portfolio selection theory 257

U (x)
6

qqqqqqqqqqqqqqqqqqqqqqq U (x)
qqqqqqqq qqqqqqqqqq qqqqqq
qqq
U (x2 ) qqqqqqqq
qqqqq qqqqqqq
q
qqqqqq
qqqq qqqqqq
qqqqq
qqqqqqqqq
q
U (E[x]) qqqq
qqqqqqq
qqqqq
E[U (x)] = qqqqqqq
qqq
U (E[x] − π) qqq
qqq qq
qq
qqq
qqq qq
qq
qqq
qqq q
q
qqq
qqqqq
U (x1 ) qq
qq
qqq q
qq
qq
qqq q
qq  π-
-
x1 E[x] − π E[x] x2 x

Fig. C.1: The Arrow-Pratt measure of risk aversion

C.2 Portfolio selection theory

When considering to invest into asset markets, an investor has to make three
decisions:

• the amount he wants to invest into the asset market,

• determine the assets he wants to invest in,

• determine the amount he wants to invest into each selected asset.

This appendix describes a method how to make these decisions and find an
optimal portfolio.12 Such a portfolio

”. . . is more than a long list of good stocks and bonds. It is a balanced

whole, providing the investor with protections and opportunities with
respect to a wide range of contingencies. The investor should build
toward an integrated portfolio which best suites his needs.”13
12
A portfolio is the entirety of all investments of an individual.
13
Markowitz (1959, p.3).
258 Appendix C. Economic concepts

For this reason the associated theory is called portfolio selection theory or
short portfolio theory, rather than asset selection. The portfolio selection theory
has been developed by Markowitz (1959), Tobin (1958) and Tobin (1966).
Although the concepts employed in their theory have much been criticized for
capturing the reality only poorly, it has been the starting point for many asset
pricing models and up to date there has been developed no widely accepted
alternative.

C.2.1 The mean-variance criterion

Even by using the Arrow-Pratt measure of risk aversion, the utility function has
to be known to determine the first and second derivative for basing a decision on
the expected utility concept. Preferable would be a criterion that uses only ob-
servable variables instead of individual utility functions. For this purpose many
criteria have been proposed,14 the most widely used is the mean-variance crite-
rion. Although it also is not able to determine the optimal decision, it restricts
the alternatives to choose between by using the utility function.
The mean-variance criterion is the most popular criterion not only in finance.
The reason is first that it is easy to apply and has some convenient properties
in terms of moments of a distribution and secondly by the use of this criterion
in the basic works on portfolio selection by Markowitz (1959), Tobin (1958),
and Tobin (1966). Consequently, theories basing on their work, like the Capital
Asset Pricing Model, also apply the mean-variance criterion, which by this mean
became the most widely used criterion in finance.
It has the advantage that only two moments of the distribution of outcomes,
mean and variance, have to be determined, whereas other criteria make use of
the whole distribution. The outcome is characterized by its expected value, the
mean, and its risk, measured by the variance of outcomes.15
14
See Levy and Sarnat (1972, ch. VII and ch. IX) for an overview of these criteria.
15
One of the main critics of the mean-variance criterion starts with the assumption that risk
can be measured by the variance. Many empirical investigations have shown that the variance
is not an appropriate measure of risk. Many other risk measures have been proposed, see
Brachinger and Weber (1997) for an overview, but these measures have the disadvantage
C.2. Portfolio selection theory 259

Mean
6

aj  ai ?

u
ai
E[ai ]

? aj ≺ ai

-
V ar[ai ] Variance

Fig. C.2: The mean-variance criterion

The mean-variance criterion is defined as


 V ar[ai ] < V ar[aj ] and E[ai ] ≥ E[aj ]
(C.7) ai  aj ⇔ or .
V ar[ai ] ≤ V ar[aj ] and E[ai ] > E[aj ]


It is a necessary, although not sufficient, condition to prefer ai over aj that

V ar[ai ] ≤ V ar[aj ] and E[ai ] ≥ E[aj ]. An alternative is preferred over another
if it has a smaller risk (variance) and a larger mean. Nothing can in general be
said about the preferences if V ar[ai ] > V ar[aj ] and E[ai ] > E[aj ], other decision
rules have to be applied.
In figure C.2 an alternative in the upper left and lower right areas can be
compared to ai by using the mean-variance criterion, while in the areas marked
by ”?” nothing can be said about the preferences. If we assume all alternatives to
lie in a compact and convex set in the (µ, σ 2 )-plane,16 all alternatives that are not
dominated by another alternative according to the mean-variance criterion lie on
a line at the upper left of the set of alternatives. In figure C.3 this is illustrated
of being less easily computable and difficult to implement as a criterion. In more recent models
higher moments, such as skewness and kurtosis are also incorporated to cover the distribution
in more detail.
16
We will see that this condition is fulfilled in the case of portfolio selection for all relevant
portfolios.
260 Appendix C. Economic concepts

Mean
B
...r
..r
...r ...r
...r..r
...r ...r
...r
...r..r ...r..r
...r ...r
...r
...r ...r
...r..r ...r
...r..r
...r r.....................................................................
...u
...r
...r..r
...r
6
Efficient Frontier r rrr ...r
...r..r
..r ....r..r
...r
...r ...r
...r
...r ...r
...r..r rr r . ..............
....r
.
..
..
..
..r
..
..
.
...r
..
.. ...........
...r
.
...
..r
...r
.r
rrrr
..........
.........

rrrrrr
......... .......
......... ......
rrr ........ .....

rrrr
............. ....
....
....
rrrr
..
... ...
....
. ...
r .. ...
r
rr .
...... ...

rr
..
. ...
...
. ...
..
rrr
. ...
... ..
..
.... ..
..
rr
... ..
... ..
ru
...
..
...
.
A ..
..
..
... .....
...
.
.
... .
... ...
... ..
... .
...
...
.... ...
.... ...
.... ...
......
....... .
.....
........ ....
......... .....
......... .......
........... ........
.............
............... ...
..
..
.........
.
................... ..........
.......................... ..............
........................................................ ....................
................................................................................

-
Variance

Fig. C.3: The efficient frontier

where all alternatives are located in the oval. The undominated alternatives are
represented by the bold line between points A and B. All alternatives that are not
dominated by another alternative are called efficient and all efficient alternatives
form the efficient frontier. Without having additional information, e.g. the utility
function, between efficient alternatives cannot be distinguished.

The mean-variance criterion can be shown to be not optimal in general, i.e.

the true preferences are not always reflected by the results of this criterion.17 If
the utility function is quadratic, we will show that the mean-variance criterion
always reflects the true preferences.

Instead of defining the utility function by a term like y = b0 + b1 x + b2 x2

we can without loss of generality normalize the function by choosing b0 = 0 and
b1 = 1.18

17
Levy and Sarnat (1972, pp. 310 f.) also provide a generalization of the mean-variance
criterion that is always optimal. As this criterion cannot be handled so easily, it is rarely
applied and therefore not further considered here.
18
The concept of expected utility implies that the utility function is only determined up to
a positive linear transformation. This allows to apply the transformation y → y−b b1 to achieve
0

the normalization.
C.2. Portfolio selection theory 261

The utility function and its derivatives are therefore given by

(C.8) U (x) = x + bx2 ,

(C.9) U 0 (x) = 1 + 2bx,

(C.10) U 00 (x) = 2b.

According to (C.5) the Arrow-Pratt measure of risk aversion turns out to be

2b
(C.11) z=− .
1 + 2bE [x]
If we concentrate on risk averse individuals and assume positive marginal utility,
(C.11) implies that

(C.12) b < 0.

But if b < 0 we see from (C.9) that the marginal utility is only positive if
1
(C.13) E [x] < − .
2b
For large expected values the marginal utility can become negative. This unrea-
sonable result can only be ruled out if the risk aversion is sufficiently small.19
If we define E [x] = µ and V ar [x] = σ 2 we can write the expected utility as

E [U (x)] = E x + bx2 = µ + bE x2 = µ + b µ2 + σ 2 .
   

(C.14)

The indifference curves are obtained by totally differentiating both sides:

(C.15) dE [U (x)] = (1 + 2bµ)dµ + 2σdσ = 0.

The slope of the indifference curve in the (µ,σ)-plane is obtained by rearranging

(C.15):20
dµ 2bσ
(C.16) =− = zσ > 0,
dσ 1 + 2bµ
19
This restriction on the expected value is another argument often brought forward against
the use of a quadratic utility function and hence the mean-variance criterion. Another argument
∂z 4b2
is that the risk aversion increases with the expected outcome: ∂E[x] = (1+2bE[x]) 2 > 0, which

contradicts empirical findings. Moreover in many theoretical models a constant risk aversion is
assumed, which has been shown by Pratt (1964) to imply an exponential utility function. If
the expected outcome does not vary too much, constant risk aversion can be approximated by
using a quadratic utility function.
20
Instead of using the variance as a measure of risk, it is more common to use its square root,
the standard deviation. As the square root is a monotone transformation, the results are not
changed by this manipulation.
262 Appendix C. Economic concepts

i.e. for risk averse investors the indifference curves have a positive slope in the
(µ,σ)-plane.
The equation of the indifference curve is obtained by solving (C.14) for µ:

(C.17) E [U (x)] = µ + bµ2 + bσ 2

1 E [U (x)]
µ2 + µ + σ 2 =
b b
 2
1 1 1
µ+ + σ2 = E [U (x)] + 2 .
2b b 4b

Defining r∗ = − 2b
1
as the expected outcome that must not be exceeded for the
marginal utility to be positive according to equation (C.13), we can rewrite the
equation for the indifference curves as

(C.18) (µ − r∗ ) + σ 2 = −2r∗ E [U (x)] + r∗2 ≡ R2 ,

which is the equation of a circle with center µ = r∗ , σ = 0 and radius R.21 With
this indifference curve, which has as the only parameter a term linked to the
risk aversion, it is now possible to determine the optimal alternative out of the
efficient alternatives, that is located at the point where the efficient frontier is
tangential to the indifference curve. Figure C.4 shows the determination of the
optimal alternative C.
We will now show that with a quadratic utility function the mean-variance
criterion is optimal.22 We assume two alternatives with µi = E [ai ] > E [aj ] = µj .
Let further σi2 = V ar [ai ] and σj2 = V ar [aj ]. If ai  aj it has to be shown that

E [U (ai )] > E [U (aj )] .

Substituting the utility functions gives

µi + bµ2i + bσi2 > µj + bµ2j + bσj2 ,

21
The results that the indifference curves are circles gives rise to another objection against
the use of a quadratic utility function. An individual with a quadratic utility function should be
indifferent between an expected outcome of r∗ + v and r∗ − v for any given σ. From the mean-
variance criterion (C.7) we know that for a given σ the alternative with the higher expected
outcome will be preferred. In practice this problem is overcome by using only the lower right
sector of the circle.
22
Such a proof is given e.g. in Levy and Sarnat (1972, pp. 385 ff.).
C.2. Portfolio selection theory 263

Mean Indifference
.
Curves
.. . . .
.. .. ..
6 .. .. ..
..
.. ..
... .. ..
.... .... ..
...

rrrrrrrrrrrrrrrrrrrxB
.. .. ..

rrrrrrrrrrrrrr
... ... ...
... ... ...
..
rrrrrrrr .. ..
rr
rrrrrrr
. . .
..
.. ... ...

rrrrrr
...
... Efficient Frontier ..
... ...
..
.
.....
r r
rrr .
...
...
.
r
.. ...
....
r ...

rrrrr
.... .... ...
... ... ..
rrrr
........ ....
. ...
rx
.
. .. .
......
r
.
.... ...
.......
.......
rrr .... ...

rrr C
....... ... ....
........ .... ....

rr r
. ......
. ... .
. .
......... ...
.....
..
....
........... ...... ....
.............
................ ...... ....
rr
.................................... .......... .....
.. .
...... ...
rrr
....... ......
....... ......
........ ......
......... ......
rr
....... ...
. .
.......... ..
......
.......... .......
rrr
............ ......
............. .......
.................. .......
.......................................... .. .....
.
.........

.........................
..........
..........
.........
A rrx
......
....................
............................................

-
Standard Deviation

Fig. C.4: Determination of the optimal alternative

µi − µj + b(µ2i − µ2j ) + b(σi2 − σj2 ) = (µi − µj ) [1 + b(µi + µj )] + b(σi2 − σj2 ) > 0.

Dividing by −2b > 0 gives us

σi2 − σj2
 
1 µ i + µj
(C.19) (µi − µj ) − − − > 0.
2b 2 2
1 1
From (C.13) we know that − 2b > µi and − 2b > µj , hence we find that

1 µi + µj
(C.20) − >
2b 2

With the assumption that µi > µj and as b < 0 the first term in (C.19) is positive.
If now σi2 ≤ σj2 as proposed by the mean-variance criterion, (C.19) is fulfilled and
we have shown that it represents the true preferences.
If σi2 > σj2 in general nothing can be said which alternative will be preferred.
For µi = µj we need σi2 < σj2 in order to prefer ai over aj . This is exactly the
statement made by the mean-variance criterion in (C.13). Therewith it has been
shown that in the case of a quadratic utility function the mean-variance criterion
is optimal, i.e. represents the true preferences.23
23
A quadratic utility function is not only a sufficient condition for the optimality of the
mean-variance criterion, but also a necessary condition. This is known in the literature as the
Schneeweiss-Theorem, see Schneeweiss (1967).
264 Appendix C. Economic concepts

C.2.2 The Markowitz frontier

The portfolio selection theory is based on several assumptions:24

• no transaction costs and taxes,

• assets are indefinitely divisible,

• each investor can invest into every asset without restrictions,

• investors maximize expected utility by using the mean-variance criterion,

• prices are given and cannot be influenced by investors (competitive prices),

• the model is static, i.e. only a single time period is considered.

Some of these assumptions, like the absence of transaction costs and taxes have
been lifted by more recent contributions without giving fundamentally new in-
sights.
In portfolio selection theory the different alternatives to choose between are
the compositions of the portfolios, i.e. the weight each asset has.25 Assume an
investor has to choose between N > 1 assets, assigning a weight of xi to each
asset. The expected return of each asset is denoted µi and the variance of the
returns by σi2 > 0 for all i = 1, . . . , N .26 The covariances between two assets i
and j will be denoted σij .
The weights of the assets an investor holds, have to sum up to one and are
assumed to be positive as we do not allow for short sales at this stage:
N
X
(C.21) xi = 1,
i=1
xi ≥ 0, i = 1, . . . , N.
24
See Lintner (1965a, p. 15).
25
The decision which portfolio is optimal does not depend on total wealth for a given constant
risk aversion, hence it can be analyzed by dealing with weights only.
26
Instead of investigating final expected wealth and its variance after a given period of time
(the time horizon), we can use the expected return and variances of returns as they are a positive
linear transformation of the wealth. As has been noted above, the decision is not influenced by
such a transformation when using expected utility.
C.2. Portfolio selection theory 265

For the moment assume that there are only N = 2 assets. The characteristics of
each asset can be represented as a point in the (µ,σ)-plane. We then can derive
the location of any portfolio in the (µ,σ)-plane by combining these two assets.
The expected return and the variance of the return of the portfolio is given
by

(C.22) µp = x1 µ1 + x2 µ2 = µ2 + x1 (µ1 − µ2 ),

(C.23) σp2 = x21 σ12 + x22 σ22 + 2x1 x2 σ12

= σ22 + x21 (σ12 + σ22 − 2σ1 σ2 ρ12 ) + 2x1 (σ1 σ2 ρ12 − σ22 ),

σ12
where ρ12 = σ1 σ 2
denotes the correlation of the two assets.
The portfolio with the lowest risk is obtained by minimizing (C.23). The first
order condition is

∂σp2
= 2x1 (σ12 + σ22 − 2σ1 σ2 ρ12 ) + 2(σ1 σ2 ρ12 − σ22 ) = 0.
∂x1

The second order condition for a minimum is fulfilled unless σ1 = σ2 and ρ12 6= 1:

∂ 2 σp2
= 2(σ12 + σ22 − 2σ1 σ2 ρ12 ) > 2(σ1 − σ2 )2 > 0
∂x21

Solving the first order condition gives the weights in the minimum risk portfolio
(MRP ):

σ22 − σ1 σ2 ρ12
(C.24) xM
1
RP
= .
σ12 + σ22 − 2σ1 σ2 ρ12

The minimum variance can be obtained by inserting (C.24) into (C.23):

2 2 (σ22 − σ1 σ2 ρ12 )2 (σ22 − σ1 σ2 ρ12 )2

(C.25) σM RP = σ2 + − 2
σ12 + σ22 − 2σ1 σ2 ρ12 σ12 + σ22 − 2σ1 σ2 ρ12
(σ 2 − σ1 σ2 ρ12 )2
= σ22 − 2 2 2
σ1 + σ2 − 2σ1 σ2 ρ12
σ1 σ2 (1 − ρ212 )
2 2
= .
σ12 + σ22 − 2σ1 σ2 ρ12

If the returns of the two assets are uncorrelated (ρ12 = 0), then (C.25) reduces to

2 σ12 σ22
(C.26) σM RP = .
σ12 + σ22
266 Appendix C. Economic concepts

This variance is smaller than the variance of any of these two assets.27 By hold-
ing an appropriate portfolio, the variance, and hence the risk, can be reduced,
whereas the expected return lies between the expected returns of the two assets.
With perfectly negative correlated assets (ρ12 = −1) we find that

2
(C.27) σM RP = 0

and the risk can be eliminated from the portfolio.

In the case of perfectly correlated assets (ρ12 = 1) the minimum variance is
the variance of the asset with the lower variance:
 2
2 σ1 if σ12 ≤ σ22
(C.28) σM RP = .
σ22 if σ12 > σ22

We can derive a general expression for the mean-variance relation:

(C.29) σp2 − σM
2 2 2 2 2 2
RP = σ2 + x1 (σ1 + σ2 − 2σ1 σ2 ρ12 ) + 2x1 (σ1 σ2 ρ12 − σ2 )

σ12 σ22 (1 − ρ212 )

−
σ12 + σ22 − 2σ1 σ2 ρ12

= (x1 − xM
1
RP 2
) (σ12 + σ22 − 2σ1 σ2 ρ12 ).

With µM RP denoting the expected return of the minimum risk portfolio, we find
that

(C.30) µp − µM RP = (x1 − xM
1
RP
)(µ1 − µ2 ).

Solving (C.30) for x1 −xM

1
RP
and inserting into (C.29) we obtain after rearranging:
σp2 − σM
2
2
RP
(C.31) (µp − µM P R ) = 2 2
(µ1 − µ2 )2 .
σ1 + σ2 − 2σ1 σ2 ρ12
This equation represents a hyperbola with axes
µ1 − µ2
µp = µM RP + p σp ,
σ12 + σ22 − 2σ1 σ2 ρ12
µ1 − µ2
µp = µM RP − p σp .
σ12 + σ22 − 2σ1 σ2 ρ12
27 2 σ12 σ22
Suppose σM RP = σ12 +σ22
> σ22 , this would imply that σ12 > σ12 + σ22 and hence σ22 < 0,
which contradicts the assumption that σ22 > 0. A similar argument can be used to show that
2 2
σM RP < σ1 .
C.2. Portfolio selection theory 267

µ uB
......
. ................
........................ ......
........................... .......
6 .....
. .....
.. ...................................... ...........
.. . .....
............... ............ .......
............... ........... .......
............... .....................
............... . ......
........
..................... ..................... . .. . . ........
... .. .....
............... ........... .......
.............. ........... .......
............... ..........
............... .......... ......
......
................. .
........... .
ρ = −1 ..............
...............
.... ...... ....
.........
..........
...
.... .
......
.
......
.
...

............... .......... .......

............... ......... .......
. ... . ....... ............... .....
............ .
....
.......
..... .. ......
............... ......... .......
............... .........
...............
..............
........
........ ρ=0 ......
.......

............. .
. .....
..... .
................
.......
.....
.........
........
..
........ ρ=1
u MRP
....................
.......
..............
.......
.......
.......
.......
.......
.......
.......
....... .
......... ..
..........
....... .
..... ......
....... ...... ......
....... ..... .......
....... .... ......
....... .. ..... .............
....... . .....
....... ... .......
....... ... .......
.......
... .......
.......
u MRP
.......
....... ... .....
....... ... .....
....... .. .......
.......
....... .... .......
.......
....... ..
....... ... ..
...........
........... ....
......... .......
....... .......
....... ............
u ........
A MRP
-
σ

Fig. C.5: Efficient portfolios with two assets

The efficient portfolios lie on the upper branch of this hyperbola, i.e. above the
minimum risk portfolio.28 Figure C.5 shows the efficient portfolios for different
correlations. It can easily be shown that in the case of perfect positive correlation
the efficient portfolios are located on a straight line connecting the two assets,
in case of perfectly negative correlation on straight lines connecting the assets
with the minimum risk portfolio. Between efficient portfolios can only be distin-
guished by using the utility function. Figure C.6 adds the indifference curve to
the opportunity locus and determines the location of the optimal portfolio (OP ).
The location of the optimal portfolio depends on the risk aversion of the investor,
the more risk averse the investor is the more close the optimal portfolio will be
located to the minimum risk portfolio.

It is now possible to introduce a third asset. In a similar way hyperbolas can

be deducted representing all combinations of this asset with one of the other two.
Furthermore we can view any portfolio consisting of the two other assets as a
single new asset and can combine it in the same manner with the third asset.
Figure C.7 illustrates this situation. All achievable portfolio combinations are

28
The efficient frontier is also called opportunity locus.
268 Appendix C. Economic concepts

µ Indifference Curve uB
....
............
............
............
6 .
... ..
..
..
..
..
.............
....
. ...........
.. ...........
.. ...........
... ...........
....
..
..
...........
.
..
..
..
............
. Opportunity Locus
... ...........
..........
... ..........
... ..
..
..
..
...........
..
... ..........
... .........
..........
.. .........
..
... .
..
..
..
..........
.
....
. .........
.... .........
........
.... ..........
.
..... ..............
... .....
... .......
.............
..........
u .
.....
.
........
.
OP
......
..........
..........
. . ....
...............
.
........ ...
........ ....
.........
........... ...
...........
.........................................
.
.....
..................
..
..
..
..
..
.
...u MRP
...
..
...
...
...
....
.....
u
.....
.......
..
A
-
σ

Fig. C.6: Determination of the optimal portfolio with two assets

now located in the area bordered by the bold line connecting points A and C,
where the bold line encircling the different hyperbolas is the new opportunity
locus.

This concept can be generalized to N > 3 assets in the same manner. All
achievable assets will be located in an area and the efficient frontier will be a
hyperbola. Using the utility function the optimal portfolio can be determined in
a similar way as in the case of two assets as shown in figure C.8. If an asset is
added, the area of achievable portfolios is enlarged and encompasses the initial
area. This can simply be shown by stating that the new achievable portfolios
encompass also the portfolios assigning a weight of zero to the new asset. With a
weight of zero these portfolios are identical to the initially achievable portfolios.
To these portfolios those have to be added assigning a non-zero weight to the
new asset. Therefore the efficient frontier moves further outward to the upper
left. By adding new assets the utility can be increased.

Thus far it has been assumed that σi2 > 0, i.e. all assets were risky. It is also
possible to introduce a riskless asset, e.g. a government bond, with a variance of
zero. Define the return of the riskless asset by r, then in the case of two assets
C.2. Portfolio selection theory 269

µ qqq..q...q..q...q..q.....q...q..quB
q q qq q q q q q q q q qq qqqqqq.q..q...q..q...q..q...q..q...q..q...q....q.....q......q.....q.....q.....q..............................
...q..q...q..q.................
q
qq ......
qqqqqqqq...q..q..q...q....q......q....q......q....q................................
6 .........
.. ......
...

q...q..q...q....q......q....q.....q........q.....q.....q.....q.....q.....q............................................................................................................................................................................... PAB
...
..........
.........

q q q q qq q q q q q q
q ... ...
qqqqqqqqqq...q....................................................................................
....
.... .
.

qqqqqqqqq........................................................................P.....BC
.... ..
.........
.........
q .q....................................
.........
q qq qq q .........
q..q..q..q...q..... ..................
.
........ ........
.q..q....q...
.
....
. ........ . .....
.. .......
.

q..q...q..q...q..q..q.. ....................
.. ...... .........
.
..... ........... ........
...q..
........
qq ......u
... ..... ........
q q ........
qqqq..... .... C .......................
..
. . .... ... . . .. . ..

qqqq ..... ........

........
.......
........
qqq qq ....... ........
.......
. ....... ........
... .
.. . . .......
qqq ......
.. .......
... ...... .......
.... ........
. ....

MRP qqqqx ........

.... ........ . ....
.......
. ..
..... PAC ......
......
qqq ..... . .....
...
.........
......

qqq ..... ..........

. .... .
....
....
qqq ...... ..... ....
....
qqq .......... . ..
qqqq ....... ....
...
qqqqq ......... .......
qqqqq........... .....
qqq..q..q........... .....
q.q..q..q....q........
..q..q.....q..
...q...qu

A
-
σ

Fig. C.7: Portfolio selection with three assets

µ ......
......................................
uB
............................ ...................
.......................... ..........
6 ..............
...... ......................... ........
.
...... ...
.. ..................... ..........
.. ................... .........
.. .................. .........
.. ................. .........
..
... .............................. .....
..
..........
... .
. .............. .........
.. ............. .........
.. ............ .........
... ........... .........
..
. ...
..
.............
. ..
..
.............
...... .
... ......... .........
.. ......... .........
.. ..........
................ u ........
........
........
..
...
......
.. C .
........
.......
.....
.

.......
......
....... uOP ..
........
........
........
...
.. ..
..
......... .......
...... .. .......
........ .... .......
.
.......
..
..
..
..
.........
.......... .
..
.. u MRP .
..
...
.......
......
............................... .. ......
... ......
......
... .....
...
.. ....
... ..
... ....
.... ....
.... ...
.... .....
....
.... ...
.... ...
.....
...... ...
...... ..
.
...... .
...... ....
...... ..
..........
..... u
A
-
σ

Fig. C.8: The optimal portfolio with N > 2 assets

270 Appendix C. Economic concepts

we get from (C.22) and (C.23):

(C.32) µp = x1 µ1 + x2 r = r + x1 (µ1 − r),

(C.33) σp2 = x21 σ12 .

Solving (C.33) for x1 and inserting into (C.32) gives

σp µ1 − r
(C.34) µp = r + (µ1 − r) = r + σp .
σ1 σ1

The expected return of the portfolio is linear in the variance of the portfolio
return, i.e. the hyperbola reduces to a straight line from the location of the
riskless asset, (0, r), to the location of the asset. In the case of many risky
assets we can combine every portfolio of risky assets with the riskless asset and
obtain all achievable portfolios. As shown in figure C.9 all achievable portfolios
are located between the two straight lines, the upper representing the efficient
frontier. There exists a portfolio consisting only of risky assets that is located
on the efficient frontier. It is the portfolio consisting only of the risky assets at
which the efficient frontiers with and without a riskless asset are tangential.29
This portfolio is called the optimal risky portfolio (ORP ). The efficient frontier
with a riskless asset is also called the capital market line.
All efficient portfolios are located on the capital market line, consequently
they are a combination of the riskless asset and the optimal risky portfolio. The
optimal portfolio can be obtained in the usual way by introducing the indifference
curves. As the optimal portfolio always is located on the capital market line,
it consists of the risky asset and the optimal risky portfolio. Which weight is
assigned to each depends on the risk aversion of the investor, the more risk
averse he is the more weight he will put on the riskless asset. The weights of the
optimal risky portfolio do not depend on the risk aversion of the investor. The
decision process can therefore be separated into two steps, the determination of
the optimal risky portfolio and then the determination of the optimal portfolio
29
It is also possible that no tangential point exists, in this case a boundary solution exists
and the risky portfolio consists only of a single risky asset.
C.2. Portfolio selection theory 271

µ uB
..............
................................................
............................
.......................... ..........
6 ........ .
.........
... ........................ ...
...
............
....... ...
.................... ..........
.................. .........
.................. .........
................. .........
... ..
........................ ..
..............
.
... .
.........
.............. .........
............. .........
............ .........
........... .........
...
...
..
........... .....
..
.
......... .
.........
u ORP .......
.........
........
u ........
........
........
....... ........
..
... ...........
.. ....
C .......
........
.
.......
.
.. ......... .......
.............
OP u .................... ........
....
.
..
.
. ....
.........
............ . .......
................ .... .......
...................... .......
........................................... ............
...
. .
....uMRP .
.
..
...
..
.........
.......
.......
. .. .....
.......
.. ......
.......
... ......
....... ... .....
.
..
........ ...
......
.
.. ... ..
..............
r ..........
..........
..........
...
....
.... ....
....
....
.......... .... .....
.......... ....
.......... .... ...
..........
.......... .... ...
.......... ...... ..
.......... ......
. ...
.......... ...... .
.......... ...... ...
.......... ...... ..
u
................ ..
..................
...
A
-
σ

Fig. C.9: The optimal portfolio with a riskless asset

as a combination of the ORP with the riskless asset. As this result has first been
presented by Tobin (1958) it is also called the Tobin separation theorem.30
So far we have assumed that xi ≥ 0 for all i = 1, . . . , N . If we allow now some
xi to be negative, the possibilities to form portfolios is extended. An asset with
xi < 0 means that the asset is sold short, i.e. it is sold without having owned
it before. This situation can be viewed as a credit that has not been given and
has not to be repaid in money (unless money is the asset), but in the asset. The
assets can be the riskless asset or the risky assets, in the former case the short
sale is an ordinary credit. It is assumed that credits can be obtained at the same
conditions (interest rate or expected return and risk) as investing in the asset.
By allowing short sales the efficient frontier of the risky portfolios further
moves to the upper left as new possibilities to form portfolios are added by lifting
the restriction that the weights must be non-negative. Therewith the capital
market line becomes steeper and the utility of the optimal portfolio increases.

30
For investors being less risk averse it is possible that the optimal portfolio is located on
the part of the efficient frontier above the ORP, in this case the optimal portfolio does not
contain the riskless asset and assigns different weights to the risky assets compared to the
ORP. Therefore in general the Tobin separation theorem does only hold with the inclusion of
short sales, as described in the next paragraph.
272 Appendix C. Economic concepts

...........
.......................
......................
......................
.....................
Capital Market Line .
.......
.
.......
....................
.
..
..
..
..
..
..
..
..
..
....... ...................
....... ..................
µ .
.......
.
..
..
.
.......
....... .
.......
..
..
..
..
..
.
.................
................
. uB
....... ...............
....... ...............
....... ..............
..............
6
..
..
........
. ..
.........
..............
.. ...
....... ............
....... .......................
....... .
....... ...........
.................................
... ...
................
................
...............
..........
.......
u ......
..
.......
.......
.......
u
..
..
.... ...
.....
.
.
.........
.
.
.
.......
ORP C
.. .................
... ......... ....
... ....... ....
.............. .....
OP u ..
.
.
....
........
.
...
............. ...
..................... ..
...................................................... ..
..
.. . .
....... .
..
.......
........
........
.......
u MRP...
..
..

r ....... ..
..
..
...
...
...
...
...
...
...
....
....
....
....
....
....
u
A .....
......
......
......
......
......
..
-
σ

Fig. C.10: Portfolio selection with short sales

Figure C.10 illustrates this case. The Capital Market Line extends beyond the
ORP and therewith the optimal portfolio will always be a combination of the
riskless asset and the ORP. The Tobin separation theorem applies in all cases,
independent of the degree of risk aversion. If the ORP is located above the
ORP, the riskless asset is sold short and a larger fraction of the optimal portfolio
consists of the ORP.
In applying the portfolio theory to determine the optimal portfolio several
problems are faced:

• determination of the risk aversion of the investor,

• determination of the expected returns, variances and covariances of the

assets,

• computation of the efficient frontier and the optimal portfolio.

There exists no objective way to determine the risk aversion of an investor, most
investors are only able to give a qualitative measure of their risk aversion, if at
all. The transformation into a quantitative measure is an unsolved, but for the
C.2. Portfolio selection theory 273

determination of the optimal portfolio critical problem. It is important for the

allocation between the riskless asset and the optimal risky portfolio.
Expected returns, variances and covariances can be obtained from estimates
based on past data. But there is no guarantee that these results are reasonable for
the future. It is also possible to use other methods to determine these moments,
e.g. by using subjective beliefs. The determination of these moments are critical
for the determination of the optimal risky portfolio.
To determine the efficient frontier and the optimal portfolio non-trivial numer-
ical optimization routines have to be applied.31 Advances in computer facilities
and the availability of these routines do not impose a threat anymore as it has
done in former years.
When having solved the above mentioned problems, the portfolio theory does
allow to answer the questions raised at the beginning of this appendix:

• the share to be invested into risky assets is determined by the optimal

portfolio,

• the assets to invest in are those included in the optimal risky portfolio,

• the shares to invest in each selected asset are given by the weights of the
optimal risky portfolio.

The portfolio theory has developed a method how to allocate resources opti-
mal. Although mostly only financial assets are included, other assets like human
capital, real estate and others can easily be included, although it is even more
difficult to determine their characteristics.
A shortcoming of the portfolio theory is that it is a static model. It determines
the optimal portfolio at a given date. If the time horizon is longer than one
period, the prices of assets change over time, and therewith the weights of the
assets in the initial portfolio change. Even if the expected returns, variances and
covariances do not change, this requires to rebalance the portfolio every period.
31
For a detailed description of the mathematical concepts to solve these problems see
Markowitz (1959) and Aschinger (1990).
274 Appendix C. Economic concepts

As assets with a high realized return enlarge their weight, they have partially to
be sold to buy assets which had a low return (sell the winners, buy the losers). In
a dynamic model other strategies have been shown to achieve a higher expected
utility for investors, but due to the static nature of the model such strategies
cannot be included in this framework.

C.3 The Capital Asset Pricing Model

In this section we will derive the Capital Asset Pricing Model (CAPM ) that is the
most prominent model in asset pricing. Sharpe (1964) and Lintner (1965a)
developed the CAPM independent of each other by using portfolio theory to
establish a market equilibrium.
The basis of the CAPM is portfolio theory with a riskless asset and unlimited
short sales. We do not consider only the decision of a single investor, but aggre-
gate them to determine a market equilibrium. In portfolio theory the price of an
asset was exogeneously given and could not be influenced by any investor. Given
this price he formed his beliefs on the probability distribution.32 The beliefs were
allowed to vary between investors.
In the CAOM asset prices (or equivalently expected asset returns) will no
longer be exogenously given, but be an equilibrium of the market. From basic
finance theory we know that the current price affects the expected returns and vice
versa. Given future expected dividends and assuming that markets are efficient,
i.e. that the prices of assets equal their fundamental value, a high current price
results in a low expected return in the next period and a low current price in
a high expected return. In the same way in order to expect a high return, the
price has to be low and for a low expected return a high price is needed. This
equivalence of price and return allows us to concentrate either on prices or on
expected returns. Convention in the academic literature requires us to focus on
expected returns.
32
As portfolio theory makes use of the mean-variance criterion it is sufficient to form beliefs
only about means, variances and covariances instead of the entire distribution.
C.3. The Capital Asset Pricing Model 275

Additionally to the assumptions already stated for portfolio theory, we have

to add that all investors have the same beliefs on the probability distribution of
all assets, i.e. agree on the expected returns, variances and covariances.33 If all
investors agree on the characteristics of an asset the optimal risky portfolio will
be equal for all investors, even if they differ in their preferences (risk aversion).
Because all assets have to be held by the investors the share each asset has in
the optimal risky portfolio has to be equal to its share of the market value of
all assets. The optimal risky portfolio has to be the market portfolio. Moreover
all assets have to be marketable, i.e. all assets must be traded and there are no
other investment opportunities not included into the model.
Every investor j (j = 1, . . . , M ) maximizes his expected utility by choosing
an optimal portfolio, i.e. choosing optimal weights for each asset:
 
 j  j 1 2
(C.35) max E U (Rp ) = max U µp − zj σp
{xi }N
i=1 {xi }N
i=1
2
 
1 2
= max µp − zj σp
{xi }N
i=1
2
N N N
!
X 1 XX
= max xi µi − zj xi xk σik
{xi }N
i=1 i=1
2 k=1 i=1
PN
for all j = 1, . . . , M with the restriction i=1 xi = 1.
The Lagrange function for solving this problem can easily be obtained as
N N N N
!
X 1 XX X
(C.36) Lj = xi µi − zj xi xk σik + λ 1 − xi .
i=1
2 k=1 i=1 i=1

The first order conditions for a maximum are given by

N
∂Lj X
(C.37) = µi − zj xk σik − λ = 0, i = 1, . . . , N,
∂xi k=1
N
∂Lj X
(C.38) = 1− xi = 0
∂λ i=1

33
Lintner (1965b) calls this assumption idealized uncertainty. Sharpe (1970, pp. 104
ff.) also considers different beliefs. As the main line of argument does not change, these
complications are not further considered here. Several assumptions made in portfolio theory
can also be lifted without changing the results significantly. Black (1972) restricts short sales
and Sharpe (1970, pp. 110 ff.) applies different interest rates for borrowing and lending the
riskless asset.
276 Appendix C. Economic concepts

for all j = 1, . . . , M .34

Solving the above equations for µi gives
N
" N
#
X X
(C.39) µ i = λ + aj xk σik = λ + zj Cov Ri , x k Rk
k=1 k=1
= λ + zj Cov [Ri , Rp ] = λ + zj σip .

With σip = 0 we find that µi = λ, hence we can interpret λ as the expected return
of an asset which is uncorrelated with the market portfolio. As the riskless asset
is uncorrelated with any portfolio, we can interpret λ as the risk free rate of
return r:

(C.40) µi = r + zj σip .

From (C.40) we see that the expected return depends linearly on the covariance
of the asset with the market portfolio. The covariance can be interpreted as a
measure of risk for an individual asset (covariance risk ). Initially we used the
variance as a measure of risk, but as has been shown in the last section the risk
of an individual asset can be reduced by holding a portfolio. The risk that can-
not be reduced further by diversification is called systematic risk, whereas the
diversifiable risk is called unsystematic risk. The total risk of an asset consists
of the variation of the market as a whole (systematic risk) and an asset specific
risk (unsystematic risk). As the unsystematic risk can be avoided by diversifica-
tion it is not compensated by the market, efficient portfolios therefore only have
systematic and no unsystematic risk.
The covariance of an asset can be also interpreted as the part of the systematic
risk that arises from an individual asset:
N N
" N
#
X X X
xi σip = xi Cov[Ri , Rp ] = Cov x i R i , Rp
i=1 i=1 i=1
= Cov[Rp , Rp ] = V ar[Rp ] = σp2 .

Equation (C.40) is valid for all assets and hence for any portfolio, as the equation
for a portfolio can be obtained by multiplying with the appropriate weights and
34
The second order condition for a maximum can be shown to be fulfilled, due to space
limitations this proof is not presented here.
C.3. The Capital Asset Pricing Model 277

then summing them up, so that we can apply this equation also to the market
portfolio, which is also the optimal risky portfolio:

(C.41) µp = r + zj σp2 .

Solving for zj and inserting into (C.40) gives us the usual formulation of the
CAPM:
σip
(C.42) µi = r + (µp − r) .
σp2
σip
Defining βi = σp2
we can rewrite (C.42) as

(C.43) µi = r + (µp − r)βi .35

βi represents the relative risk of the asset i (σip ) to the market risk (σp2 ). The beta
for the market portfolio is easily shown to be 1. We find a linear relation between
the expected return and the relative risk of an asset. This relation is independent
of the preferences of the investors (zj ), provided that the mean-variance criterion
is applied and that the utility function is quadratic. This equilibrium line is called
the Capital Market Line (CML). Figure C.11 illustrates this relation.
For the risk an investor takes he is compensated by the amount of µp − r per
unit of risk, the total amount (µp − r)βi is called the risk premium or the market
price of risk. The risk free rate of return r may be interpreted as the price for
time. It is the compensation for not consuming the amount in the current period,
but wait until the next period.36
Equation (C.43) presents a formula for the expected return given the interest
rate, r, the beta and the expected return on the market portfolio, µp . However,
the expected return of the market portfolio is not exogenous as it is a weighted
average of the expected returns of the individual assets. In this formulation only
relative expected returns can be determined, the level of expected returns, i.e.
the market risk premium, is not determined by the CAPM. Although we can
35
It also can frequently be found that only excess returns over the risk free rate of return are
considered. by defining µ0j = µj − r (C.43) becomes µ0i = µ0p βi .
36
This in opposition to the interpretation of Keynes (1936, pp. 165) who interpreted the
interest rate as a compensation for giving up liquidity, i.e. binding resources over time.
278 Appendix C. Economic concepts

µ
6 CML


 

µp  

 6



µi 

Market risk premium
 6

Risk premium

 

r  ? ?

-
βi 1 β

Fig. C.11: The Capital Asset Pricing Model

reasonably assume µp to be given when investigating a small capitalized asset,

we have to determine µp endogenously.
Thus far we only considered the market portfolio, but an investor will in
general not hold the market portfolio (optimal risky portfolio). As we saw in
section C.2 the optimal portfolio will be a combination of the market portfolio
and the riskless asset, where the shares will vary among investors. We can now
add another restriction to our model. The return on the optimal risky portfolio
has to be such that the market for the riskless assets has to be in equilibrium.
The amounts of riskless assets lent and borrowed have to be equal:37
M
X
(C.44) xjr = 0,
j=1

where xjr denotes the demand of the jth investor for the riskless asset. With this
additional market to be in equilibrium it is possible to determine µp endogenously,
whiose value will depend on the risk aversion of investors. We herewith have found
an equilibrium in the expected returns.
Nevertheless this result remains to determine only relative prices. The risk free
37
Alternatively a fixed supply of the riskless asset can be assumed without changing the
argument.
C.4. The Rational Expectations Approach 279

rate r is not determined endogenously. Although is can reasonably be assumed

that r can substantially be influenced by monetary policy, especially for longer
time horizons it is not given.

C.4 The Rational Expectations Approach

This appendix introduces the concept of rational expectations. The literature
on this subject has grown rapidly since it was first introduced by Muth (1961).
Therefore here only the basic concept is presented, it is not the aim to give
an overview of the entire theory. The first applications of rational expectations
could be found in macroeconomics in the 1970’s, most prominently by the Nobel
laureate Robert E. Lucas. Nowadays rational expectations are used in nearly
every field of economics, especially in financial markets theory.38
Using his individual information, every investor will form his expectations on a
certain economic variable which is not known with certainty, e.g. the fundamental
value of an asset and act on this individual expectation. These expectations and
actions transform into prices, where the price realized may be different from
the expectations of all investors. In order to analyze how expectations and the
therewith associated actions transform into prices it is necessary to model the
way expectations are formed.
When introducing rational expectations Muth (1961, p. 316) proposed that
expectations

”... are essentially the same as the predictions of the relevant economic
theory.”

He explains this definition of rational expectations by stating that these expec-

tations tend to be distributed for a given information set about the distribution
of the theory, i.e. on average the mean of the expectations equals the forecast of
the theory. This implicitly states that all investors agree on the theory, i.e. on
the structure of the economy as well as on the parameters. The assumption that
38
Sheffrin (1996) presents several examples of using rational expectations in different fields
of economics.
280 Appendix C. Economic concepts

rational expectations produce no systematic errors, implies that investors do not

only have to agree on the theory, but that the theory also has to be true.
Rational expectations are represented by conditional expectations. Let Xt be
the variable to be forecasted and Ωt−1 the information available, then with Xte
representing the expectation of Xt we define rational expectations as

(C.45) Xte = E [Xt |Ωt−1 ] .

That there exists no systematic error in these expectations gives

(C.46) E [Xt − Xte |Ωt−1 ] = 0,

(C.47) E [(Xt − Xte )Xt |Ωt−1 ] = 0,

i.e. the expected forecast error is zero and uncorrelated with the true value of
the variable. Many examples show that this in general only holds if the theory
is correct.
A narrow version of the definition given by Muth (1961) has been presented
by Lucas and Prescott (1971). They require not only the mean of the ex-
pectations and the predictions of the theory to coincide, but they must have the
same probability distribution.
A weak version assumes that information is costly and therefore every indi-
vidual only acquires information until his marginal costs and benefits equal. Only
this information acquired is used to form rational expectations, not all available
information. It also allows for rational expectations that are not based on the
correct theory. Investors then form their expectations ”as if” they knew the cor-
rect model. These expectations in general will be biased, i.e. not fulfill equations
(C.46) and (C.47).39
Rational expectations have much been criticized for the very restrictive as-
sumptions, especially in the narrow version. The assumption that all individuals
know the true model of the economy has been the main target, but also the as-
sumption that all available information has to be taken into account. The weak
39
Pesaran (1987, p.23) also points out that in such situations self-fulfilling prophecies can
occur until the model is identified to be wrong and a sudden change in the model applied may
correct the situation.
C.4. The Rational Expectations Approach 281

version has also been attacked as by its definition it allows to define every form
of expectations to be called rational. Especially how investors could learn the
true model of the economy remains an unsolved problem.

A more general criticism on rational choice theory, and therewith rational ex-
pectations, in most cases addresses the fact that individuals are not only rational
but in many situations behave emotional or imitate others. These behavioral ap-
proaches have attracted increased attention in recent years, especially in finance,
where they are used to explain several of features of asset prices that cannot be
explained using rational behavior.40 In most cases models using rational choice
make use of very restrictive assumptions to be able to derive results by using
sophisticated mathematical methods.41 These assumptions make predictions of
behavior in actual markets very difficult and in many cases they fail. Here Beed
and Beed (2000) question the usefulness of rational choice theory and its con-
tribution to the advancement of economic knowledge in general. However, at
present there has been developed no more powerful tool to address economic
problems.

By allowing investors to learn some aspects of the economy, the narrow version
can be weakened partly. We can allow investors to learn some parameters of the
economy, e.g. the beta in the CAPM. If we assume that all investors know the
structure of the economy, but not the parameters, it can be shown that they
will learn the true parameters over time, i.e. their expectations converge to the
narrow version. The most widely used concept of learning is Bayesian learning
to be presented in the next section.

Despite these modifications it remains an unsolved problem how investors

learn the true structure of the economy. As no other theory on expectation for-
mation exists that provides a better explanation of economic phenomena, rational
expectations are widely used in economics.

40
See Thaler (1993), Shleifer (2000) or Hirschleifer (2001) for an overview.
41
Another objective against the current economic theory and its dominating rational choice
theory is the concentration on self defined and very abstract problems rather than on ”real
world problems” that concern the society as Frey (2000, 25 f.) points out.
282 Appendix C. Economic concepts

C.5 Bayesian learning

We assume investors to know the structure of the economy, but not the param-
eters in the model. In a first step they form beliefs about these parameters, e.g.
by assigning a random number or another reasoning. Every investor can have
different beliefs. Based on these first beliefs they form their expectations and
act accordingly. When they see the realization of the outcome, they realize that
their expectations were not correct. As they know the structure of the economy,
the only source of these deviations can be the values of the parameters assigned.
Hence they want to change their beliefs about these parameters. This process
continues until expectations and realizations coincide and the beliefs are correct.
If there are random variables in the economy, the most widely used method to
model changes in beliefs uses Bayes rule and hence is called Bayesian learning.42
We know from probability theory that for discrete random variables with Prob
denoting the probability:43

(C.48) Prob[Ωt ] = Prob[Ωt |a = at ]Prob[a = at ] + Prob[Ωt |a 6= at ]Prob[a 6= at ],

Prob[a = at , Ωt ]
(C.49) Prob[a = at |Ωt ] = ,
Prob[Ωt ]
Prob[a = at , Ωt ]
(C.50) Prob[Ωt |a = at ] = ,
Prob[a = at ]

where a is the true parameter and at the current belief of the value of this pa-
rameter. Combining relations (C.48) - (C.50) gives
(C.51)
Prob[Ωt |a = at ]Prob[a = at ]
Prob[a = at |Ωt ] = ,
Prob[Ωt |a = at ]Prob[a = at ] + Prob[Ωt |a 6= at ]Prob[a 6= at ]

which is also known as Bayes’ rule or Bayes’ theorem. The probability that
a = at in the last period, Prob[a = at ], is called the prior belief. On this belief
the investor based his decision. Given this belief for all possible values the entire
42
If no random variables are present, other models of learning have to be used. In some cases
it may be possible to solve the equations directly for the parameters and obtain the true values
in a single step. As in most models random variables are incorporated we concentrate on this
case here.
43
For continuous random variables the argument does not change. Instead of probabilities
densities have to be used.
C.5. Bayesian learning 283

distribution is known. In the next period he learns the realization of the process,
that forms part of his new information set, Ωt . Based on this new information he
changes his belief that a = at to Prob[a = at |Ωt ] according to equation (C.51),
his posterior belief. For the next period these beliefs are his prior beliefs, which
he updates on the new information received in the next period. Applying this re-
lation for all possible at we receive the distribution and can calculate the relevant
parameters, e.g. mean and variance.
By using Bayes’ rule to change beliefs it can be shown that the beliefs converge
to the true values of the parameters. Hence expectations converge to rational ex-
pectations in the narrow sense. This property makes Bayesian learning attractive
to use in rational expectation models. What remains an unsolved problem is how
to learn the true structure of the economy.
284 Appendix C. Economic concepts
Bibliography

Acharya, V. V. and Pedersen, L. H. (2004): Asset pricing with liquidity risk. Mimeo,
London Business School.
Admati, A. R. and Pfleiderer, P. (1988): A Theory of Intraday Patterns: Volume and
Price Variability. In: Review of Financial Studies, 1, 3–40.
Admati, A. R. and Pfleiderer, P. (1989): Divide and Conquer: A Theory of Intraday and
Day-of-the-Week Mean Effects. In: Review of Financial Studies, 2, 189–223.
Affleck-Graves, J., Hedge, S. P. and Miller, R. E. (1994): Trading Mechanisms and
the Components of the Bid-Ask Spread. In: Journal of Finance, 49, 1471–1488.
Amihud, Y. and Mendelson, H. (1980): Dealership Market: Market Making with Inventory.
In: Journal of Financial Economics, 8, 31–53.
Amihud, Y. and Mendelson, H. (1986a): Asset Pricing and the Bid-Ask Spread. In: Journal
of Financial Economics, 17, 223–249.
Amihud, Y. and Mendelson, H. (1986b): Liquidity and stock returns. In: Financial
Analysts Journal, 43–48.
Amihud, Y. and Mendelson, H. (1988): Liquidity and asset prices: Financial management
implications. In: Financial Management, 5–15.
Amihud, Y. and Mendelson, H. (1989): The effect of beta, bid-ask spread, residual risk
and size on stock returns. In: Journal of Finance, 44, 479–486.
Amihud, Y. and Mendelson, H. (1991): Liquidity, asset prices and financial policy. In:
Financial Analysts Journal, 56–66.
Angel, J. J. (1997): How Best to Supply Liquidity to a Small-Capitalization Securities
Market. Mimeo, Georgetown University.
Angel, J. J. and Weaver, D. G. (1998): Priority Rules !. Mimeo, Georgetown University.
Arnold, T., Hersch, P., Mulherin, J. H. and Netter, J. (1999): Merging markets. In:
Journal of Finance, 54, 1083–1107.
Arrow, K. J. (1963): The Theory of Risk Aversion, chapter 9, 147–171. In: Arrow (1984).
Arrow, K. J. (1964): The Role of Securities in the Optimal Allocation of Risk-bearing. In:
Review of Economic Studies, 31, 91–96.
Arrow, K. J. (1984): Collected Papers of Kenneth J. Arrow: Individual Choice under
Certainty and Uncertainty. Oxford: Basil Blackwell.
Aschinger, G. (1990): General Remarks on Portfolio Theory and its Application. Working
Paper Nr. 158, Insitute for Economic and Social Sciences, University of Fribourg (Switzer-
land).
Atkins, A. B. and Dyl, E. A. (1990): Price reversals, bid-ask spreads, and market efficiency.
In: Journal of Financial and Quantitative Analysis, 25, 535–547.
Bacidore, J. and Lipson, M. L. (2001): The effects of opening and closing procedures on
the nyse and nasdaq. Mimeo, University of Georgia.
286 Bibliography

Bagehot, W. (1971): The Only Game in Town. In: Financial Analysts Journal, 27, 12–14,
22.
Baker, H. K. and Edelman, R. B. (1992): Amex-tp-nyse transfers, market microstructure,
and shareholder wealth. In: Financial Management, 60–72.
Barclay, M. J., Kandel, E. and Marx, L. M. (1998): The effects of transaction costs on
stock prices and trading volume. In: Journal of Financial Intermediation, 7, 130150.
Baruch, S. and Saar, G. (2004): Asset returns and the listing choice of firms. Mimeo,
University of Utah.
Baumol, W. J., Panzar, J. C. and Willig, R. D. (1988): Contestable Markets and the
Theory of Industry Structure. Revised edition, San Diego, CA: Harcourt Brace Jovanovich
Publishers.
Beed, C. and Beed, C. (2000): Intellectual Progress and Academic Economics: Rational
Choice and Game Theory. In: Journal of Postkeynesian Economics, 23, 163–185.
Benveniste, L., Cappozza, D. R. and Seguin, P. J. (2001): The value of liquidity. In:
Real Estate Economics, 29, 633–660.
Bernhardt, D. and Hughson, E. (1997): Splitting Orders. In: Review of Financial Studies,
10, 69–101.
Biais, B. (1993): Price Formation and Equilibrium Liquidity in Fragmented and Centralized
Markets. In: Journal of Finance, 58, 157–185.
Biais, B., Foucault, T. and Salani, F. (1995): Implicit Collusion on Wide Spreads. Work-
ing Paper No. 153, Finance and Banking Discussion Paper Series, Universitat Pompeu
Fabra.
Biais, B., Hillion, P. and Spatt, C. (1997): An empirical analysis of the limit order book
and the orde flow on the paris bourse. In: Journal of Finance, 50, 1655–1689.
Black, F. (1972): Capital Market Equilibrium with Restricted Borrowing. In: Journal of
Business, 45, 444–455.
Bloomfield, R. and O’Hara, M. (1998): Does order preferencing matter?. In: Journal of
Financial Economics, 50, 3–37.
Blume, L., Easley, D. and O’Hara, M. (1994): Market Statistics and Technical Analysis:
The Role of Volume. In: Journal of Finance, 49, 153–181.
Board, J. and Sutcliffe, C. (2000): The proof of the pudding: The effects of increased
trade transparency in the london stock exchange. In: Journal of Business Finance &
Accounting, 27, 887–909.
Bondarenko, O. (1999): Competing Market Makers, Liquidity Provision, and Bid-Ask
Spread. Mimeo, University of Illinois at Chicago.
Bondarenko, O. and Sung, J. (2003): Specialist participation and limit orders. In: Journal
of Financial Markets, 6, 539–571.
Booth, G. G., Lin, J.-C., Martikainen, T. and Tse, Y. (2002): Trading and pricing in
upstairs and downstairs stock markets. In: Review of Financial Studies, (15), 1111–1135.
Bourghelle, D. and Declerck, F. (2004): Why markets should not necessarily reduce the
tick size. In: Journal of Banking and Finance, 28, 373–398.
Brachinger, H.-W. and Weber, M. (1997): Risk as a Primitive: A Survey of Measures of
Perceived Risk. In: OR Spektrum, 19, 235–250.
Breen, W. J., Hodrick, L. S. and Korajczyk, R. A. (2002): Predicting equity liquidity.
In: Management Science, 48, 470483.
Bibliography 287

Brennan, M. J. and Subrahmanyam, A. (1996): Market Microstructure and Asset Pricing:

On the Compensation for Illiquidity in Stock Returns. In: Journal of Financial Economics,
41, 441–464.
Brooks, R. and Masson, J. (1996): Performance of Stoll’s Spread Component Estima-
tor: Evidence from Simulations, Time Series, and Cross-Sectional Data. In: Journal of
Financial Research, 19, 459–476.
Brown, D. P. and Jennings, R. H. (1989): On Technical Analysis. In: Review of Financial
Studies, 2, 527–551.
Brown, D. P. and Zhang, Z. M. (1997): Market Orders and Market Efficiency. In: Journal
of Finance, 52, 277–308.
Caballe, J. and Krishnan, M. (1994): Imperfect Competition in a Multi-Security Market
with Risk Neutrality. In: Econometrica, 62, 595–604.
Calcagno, R. and Lovo, S. M. (1998): Bid-Ask Price Competition with Asymmetric Infor-
mation between Market Makers. CORE Discussion Paper 9816, Universite de Catholique
de Louvain.
Chakravarty, S. and Holden, C. W. (1995): An Integrated Model of Market and Limit
Orders. In: Journal of Financial Intermediation, 4, 213–241.
Chan, S. H., Kim, K. A. and Rhee, S. G. (2005): Price limit performance: evidence from
transactions data and the limit order book. In: Journal of Empirical Finance, 12, 269–290.
Chau, M. (1998): Dynamic Trading and Market-Making with Inventory Costs and Private
Information. Mimeo, ESSEC, France.
Cho, J.-W. and Nelling, E. (2000): The probability of limit order execution. In: Financial
Analysts Journal, 28–33.
Chordia, T., Roll, R. and Subrahmanyam, A. (2000): Commonality in Liquidity. In:
Journal of Financial Economics, 56, 3–28.
Chowdhry, B. and Nanda, V. (1991): Multimarket trading and market liquidity. In:
Review of Financial Studies, 4, 483–511.
Christie, W. G. and Schultz, P. H. (1994): Why Do NASDAQ Market Makers Avoid
Odd-eighth Quotes?. In: Journal of Finance, 49, 1813–1840.
Chung, K. H. and Li, M. (2003): A statistical model of a limit order market. In: Financial
Review, 38, 257–272.
Chung, K. H., Li, M. and McInish, T. H. (2005): Information-based trading, price impact
of trades, and trade autocorrelation. In: Journal of Banking & Finance, 29, 16451669.
Chung, S. and Wei, P. (2005): The relationship between bidask spreads and holding periods:
The case of chinese a and b shares. In: Global Finance Journal, 15, 239 249.
Cochrane, J. (2000): Asset Pricing. Chicago: Graduate School of Business, University of
Chicago.
Cohen, K. J., Maier, S. F., Schwartz, R. A. and Whitcomb, D. K. (1981): Transactions
Costs, Order Placement Strategy, and Existence of the Bid-Ask Spread. In: Journal of
Political Economy, 89, 287–305.
Cohen, K. J., Maier, S. F., Schwartz, R. A. and Whitcomb, D. K. (1986): The
Microstructure of Securities Markets. Englewood Cliffs, NJ: Prentice-Hall.
Constantinides, G. M. (1986): Capital market equilibrium with transaction costs. In:
Journal of Political Economy, 94, 842–862.
Copeland, T. E. and Galai, D. (1983): Information Effects on the Bid-Ask-Spread. In:
Journal of Finance, 38, 1457–1469.
288 Bibliography

Cordella, T. and Foucault, T. (1999): Minimum Price Variations, Time Priority, and
Quote Dynamics. In: Journal of Financial Intermediation, 8, 141–173.
Coughenour, J. F. and Deli, D. N. (2002): Liquidity provision and the organizational
form of nyse specialist firms. In: Journal of Finance, 52, 841–869.
Cournot, A. (1963 (orig. 1838)): The Mathematical Principles of the Theory of Wealth.
Homewood, Ill.: Richard D. Irwin, Inc.
Cymbalista, F. (1998): Zur Unmöglichkeit rationaler Bewertung unter Unsicherheit - Eine
monetär-keynesianische Kritik der Diskussion um die Markteffizienzthese. Studien zur
Monetären Ökonomie. Marburg: Metropolis Verlag.
de Frutosa, M. A. and Manzano, C. (2005): Trade disclosure and price dispersion. In:
Journal of Financial Markets, 8, 183–216.
Demsetz, H. (1968): The Cost of Transacting. In: Quarterly Journal of Economics, 82, 33–53.
Dennert, J. (1993): Price Competition between Market Makers. In: Review of Economic
Studies, 60, 735–751.
Diamond, D. W. and Verrecchia, R. E. (1981): Information Aggregation in a Noisy
Rational Expectations Economy. In: Journal of Financial Economics, 9, 221–235.
Domowitz, I. (1993): A Taxonomy of Automated Trade Execution Systems. In: Journal of
International Money and Finance, 12, 607–631.
Duffie, D. (1996): Dynamic Asset Pricing. 2nd edition, Princeton, NJ: Princeton University
Press.
Dumas, B. and Allaz, B. (1996): Financial Securities: Market Equilibrium and Pricing
Methods. London: Chapman & Hall.
Dutta, P. K. and Madhavan, A. (1997): Competition and Collusion in Dealer Markets.
In: Journal of Finance, 52, 245–276.
Easley, D., Engle, R. F., O’Hara, M. and Wu, L. (2001): Time-varying arrival rates of
informed and uninformed trades. Mimeo, Cornell University.
Easley, D., Hvidkjaer, S. and O’Hara, M. (2002): Is information risk a determinant of
asset returns?. In: Journal of Finance, 57, 2185–2221.
Easley, D., Kiefer, N. M., O’Hara, M. and Paperman, J. B. (1996): Liquidity, infor-
mation and infrequently traded stocks. In: Journal of Finance, 51, 1405–1436.
Easley, D. and O’Hara, M. (1995): Market Microstructure, chapter 12, 357–383. In:
Jarrow et al. (1995).
Edelen, R. and Gervais, S. (2003): The role of trading halts in monitoring a specialist
market. In: Review of Financial Studies, 16, 263–300.
Eleswarapu, V. R. (1997): Cost of transating and expected reurns in the nasdaq market.
In: Journal of Finance, 52, 2113–2127.
Ellis, K., Michaelt, R. and O’Hara, M. (2000): The accuracy of trade classification rules:
Evidence from nasdaq. In: Journal of Financial and Quantitative Analysis, 35, 529–551.
Ellul, A., Shin, H. S. and Tonks, I. (2003): How to open and close the market: Lessons
from the london stock exchange. Mimeo, Indiana University.
Fama, E. F. (1970): Efficient Capital Markets: A Review of Theory and Empirical Work. In:
Journal of Finance, 25, 383–423.
Foster, F. D. and Viswanathan, S. (1990): A Theory of Interday Variations in Volume,
Variance, and Trading Costs in Securities Markets. In: Review of Financial Studies, 3,
593–624.
Bibliography 289

Foster, F. D. and Viswanathan, S. (1993): The Effect of Public Information and Com-
petition on Trading Volume and Price Volatility. In: Review of Financial Studies, 6,
23–56.
Foucault, T., Kadan, O. and Kandel, E. (2005): Limit order book as a market for
liquidity. In: Review of Financial Studies, 18, forthcoming.
Foucault, T., Moinas, S. and Theissen, E. (2004): Does anonymity matter in electronic
limit order markets?. Mimeo, HEC, School of Management, Paris.
Foucault, T. and Parlour, C. A. (2004): Competition for listings. In: RAND Journal of
Economics, 35, 329–355.
Frey, B. S. (2000): Was bewirkt die Volkswirtschaftslehre?. In: Perspektiven der Wirtschaft-
spolitik, 1(1), 5–33.
Gârleanu, N. and Pedersen, L. H. (2004): Adverse selection and the required return. In:
Review of Financial Studies, 17, 643–665.
Garman, M. B. (1976): Market Microstructure. In: Journal of Financial Economics, 3,
257–275.
Gehrig, T. and Jackson, M. (1998): Bid-Ask Spreads with Indirect Competition Among
Specialists. In: Journal of Financial Markets, 1, 89–119.
George, T. J., Kaul, G. and Nimalendran, M. (1991): Estimation of the Bid-Ask Spread
and its Components: A New Approach. In: Review of Financial Studies, 4, 623–656.
Glosten, L. R. (1989): Insider Trading, Liquidity, and the Role of the Monopolist Specialist.
In: Journal of Business, 62, 211–235.
Glosten, L. R. and Harris, L. E. (1988): Estimating the Components of the Bid/Ask
Spread. In: Journal of Financial Economics, 21, 123–142.
Glosten, L. R. and Milgrom, P. R. (1985): Bid, Ask and Transaction Prices in a Specialist
Market with Heterogenously Informed Traders. In: Journal of Financial Economics, 14,
71–100.
Gomber, P. (2000): Elektronische Handelssysteme. Innovative Konzepte und Technologien
im Wertpapierhandel. Heidelberg: Physica-Verlag.
Grammig, J. and Theissen, E. (2002): Estimating the probability of informed trading -
does trade misclassification matter?. Mimeo, University of St Gallen.
Grossman, S. J. and Miller, M. H. (1988): Liquidity and Market Structure. In: Journal
of Finance, 43, 617–637.
Grossman, S. J. and Stiglitz, J. E. (1980): On the Impossibility of Informationally
Efficient Markets. In: American Economic Review, 70, 393–408.
Hagerty, K. (1991): Equilibrium Bid-Ask Spreads in Markets with Multiple Assets. In:
Review of Economic Studies, 58, 237–257.
Hanousek, J. and Podpiera, R. (2002): Information-driven trading at the prague stock
exchange. In: Economics of Transition, 10, 757–769.
Hansch, O., Naik, N. Y. and Viswanathan, S. (1998): Do Inventories Matter in Dealership
Markets? Evidence from the London Stock Exchange. In: Journal of Finance, 53, 1623–
1656.
Harris, L. E. (1994): Minimum Price Variations, Discrete Bid-Ask Spreads, and Quotation
Sizes. In: Review of Financial Studies, 7, 149–178.
Hart, O. and Moore, J. (1996): The governance of exchanges: Members cooperatives versus
outside ownership. In: Oxford Review of Economic Policy, 12, 53–69.
290 Bibliography

Hasbrouck, J. and Seppi, D. J. (2001): Common factors in prices, order flows, and liquidity.
In: Journal of Financial Economics, 59, 383–411.
Hayek, F. A. (1945): The Use of Knowledge in Society. In: American Economic Review, 35,
519–530.
Hirschleifer, D. (2001): Investor psychology and asset pricing. In: Journal of Finance, 56,
1533–1597.
Ho, T. and Stoll, H. R. (1980): On Dealer Markets under Competition. In: Journal of
Finance, 35, 259–268.
Ho, T. and Stoll, H. R. (1981): Optimal Dealer Pricing under Transactions and Return
Uncertainty. In: Journal of Financial Economics, 9, 47–73.
Ho, T. and Stoll, H. R. (1983): The Dynamics of Dealer Markets under Competition. In:
Journal of Finance, 38, 1053–1074.
Holden, C. W. and Subrahmanyam, A. (1992): Long-Lived Private Information and
Imperfect Competition. In: Journal of Finance, 47, 247–270.
Hollifield, B., Miller, R. A. and Sandås, P. (2004): Empirical analysis of limit order
markets. In: Review of Economic Studies, 71, 1027–1063.
Hopt, K. J. and Baum, H. (1997): Börsenrechtsreform in Deutschland, chapter 3, 287–467.
In: Hopt et al. (1997).
Hopt, K. J., Rudolph, B. and Baum, H., (eds.) (1997): Börsenreform - Eine ökonomische,
rechtsvergleichende und rechtspolitische Untersuchung. Stuttgart: Schäffer-Poeschel Ver-
lag.
Huang, M. (2003): Liquidity shocks and equilibrium liquidity premia. In: Journal of Economic
Theory, 109, 104–129.
Huang, R. D. and Stoll, H. R. (1997): The Components of the Bid-Ask Spread: A General
Approach. In: Review of Financial Studies, 10, 995–1034.
Huberman, G. and Halka, D. (2000): Systematic liquidity. In: Journal of Financial
Research, 24, 161–178.
Ingersoll, J. E. (1987): Theory of Financial Decision Making. Studies in Financial Eco-
nomics. Savage, MD: Rowman & Littlefield.
Intriligator, M. D. (1971): Mathematical Optimization and Economic Theory, chapter 13,
326–343. Englewwod Cliffs, N. J., USA: Prentice Hall.
Jain, N. and Mirman, L. J. (1999): Insider Trading with Correlated Signals. In: Economics
Letters, 65, 105–113.
Jarrow, R. A., Maksimovic, V. and Ziemba, W. T., (eds.) (1995): Finance. Handbooks
in Operations Research and Management Science. Amsterdam: Elsevier.
Jevons, W. S. (1924 (orig. 1911)): Die Theorie der Politischen Ökonomie. Jena: Verlag von
Gustav Fischer.
Johnson, N. and Kotz, S. (1970): Distributions in Statistics, 4 vols. New York: Houghton
Miffin.
Kadlec, G. B. and McConnell, J. J. (1994): The effect of ¡arket segmenttation and
illiquidity on market prices: Evidence from exchange listings. In: Journal of Finance, 49,
611–636.
Keenan, W. M. (1987): The Securities Industry: Securities Trading and Investment Banking,
chapter 9. Wiley Professional Banking and Finance Series. 6th edition, New York: John
Wiley & Sons.
Keynes, J. M. (1930): A Treatise on Money. London.
Bibliography 291

Keynes, J. M. (1936): The General Theory of Employment, Interest, and Money. London.
Knight, F. H. (1957 (Orig. 1921)): Risk, Uncertainty and Profit. New York: Kelly & Millman,
Inc.
Kocherlakota, N. R. (1996): The equity premium: Its still a puzzle. In: Journal of
Economic Literature, 34, 4271.
Krause, A. (2003): Inventory effects on daily returns in financial markets. In: International
Journal of Theoretical and Applied Finance, 6, 739–765.
Krause, A. (2005): Optimal stock allocation in specialist markets. In: Research in Economics,
59, 23–39.
Krishnan, M. (1992): An Equivalence between the Kyle (1985) and the Glosten-Milgrom
(1985) Models. In: Economics Letters, 40, 333–338.
Kyle, A. S. (1985): Continuous Auctions and Insider Trading. In: Econometrica, 53, 1315–
1335.
Laux, P. A. (1993): Trade Sizes and Theories of the Bid-Ask Spread. In: Journal of Financial
Research, 16, 237–249.
Lee, R. (1998): What is an Exchange? The Automation, Management and Regulation of
Financial Markets. Oxford: Oxford University Press.
Lei, Q. and Wu, G. (2001): The behavior of uninformed investors and time-varying informed
trading activities. Mimeo, University of Michigan Business School.
Lei, Q. and Wu, G. (2005): Time-varying informed and uninformed trading activities. In:
Journal of Financial Markets, 8, 153181.
Levy, H. and Sarnat, M. (1972): Investment and Portfolio Analysis. New York: John
Wiley & Sons, Inc.
Lin, J.-C., Sanger, G. C. and Booth, G. G. (1995): Trade Size and Components of the
Bid-Ask Spread. In: Review of Financial Studies, 8, 1153–1183.
Lintner, J. (1965a): Security Prices, Risk, and Maximal Gains from Diversification. In:
Journal of Finance, 20, 587–615.
Lintner, J. (1965b): The Valuation of Risk assets and the selection of Risky Investments in
Stock Portfolios and Capital Budgets. In: Review of Economics and Statistics, 47, 13–37.
Lo, A. W., Mamaysky, H. and Wang, J. (2004): Asset prices and trading volume under
fixed transactions costs. In: Journal of Political Economy, 112, 1054–1090.
Loderer, C. and Roth, L. (2003): The pricing discount for limited liquidity: evidence from
swx swiss exchange and the nasdaq. In: Journal of Empirical Finance, 12, 239 268.
Lucas, R. E. and Prescott, E. C. (1971): Investment under Uncertainty. In: Econometrica,
39, 659–681.
Luckock, H. (2001): A statistical model of a limit order market. Mimeo, University of Sydney.
Ma, T. and Yang, J. (2002): Measuring the probability of informed trading in a call auction
market and a comprehensive analysis on the determinants of informed trading. Mimeo,
Department of Finance, National Sun Yat-sen University, Kaohsiung, Taiwan.
Madhavan, A. (1995): Consolidation, fragmentation, and the disclosure of trading informa-
tion. In: Review of Financial Studies, 8, 579–603.
Markowitz, H. M. (1959): Portfolio Selection: Efficient Diversification of Investments. New
Haven, CT: Yale University Press.
Marshall, A. (1997, orig. 8th. ed. 1920): Principles of Economics. Amherst, N.Y.:
Prometheus Books.
292 Bibliography

Martı́nez, M. A., Nieto, B., Rubio, G. and Tapia, M. (2003): Asset pricing and sys-
tematic liquidity risk: An empirical investigation of the spanish stock market. Mimeo,
Universidad Carlos III de Madrid.
Mehra, R. and Prescott, E. C. (1985): The equity premium: A puzzle. In: Journal of
Monetary Economics, 15, 145–161.
Merkt, H. (1997): Zur Entwicklung des deutschen Börsenrechts von den Anfängen bis zum
Zweiten Finanzmarktförderungsgesetz, 17–141. In: Hopt et al. (1997).
Moulin, H. (2000): Priority Rules and Other Asymmetric Rationing Methods. In: Econo-
metrica, 68, 643–684.
Muth, J. F. (1961): Rational Expectations and the Theory of Price Movements. In: Econo-
metrica, 29, 315–335.
Neal, R. and Wheatley, S. (1995): How Reliable are Adverse Selection Models of the
Bid-Ask Spread?. Federal Reserve Bank of Kansas City Research Working Paper RWP
95-02.
Němeček, L. and Hanousek, J. (2002): Market structure, liquidity, and information based
trading at the prague stock exchange. In: Emerging Markets Review, 3, 293–305.
O’Hara, M. (1995): Market Microstructure Theory. Cambridge, Mass.: Blackwell Publishers.
O’Hara, M. (2003): Liquidity and price discovery. In: Journal of Finance, 58, 1335–1354.
Parlour, C. A. (1998): Price Dynamics in Limit Order Markets. In: Review of Financial
Studies, 11, 789–816.
Parlour, C. A. and Seppi, D. J. (2003): Liquidity-based competition for order flow. In:
Review of Financial Studies, 16, 301–343.
Pástor, L. and Stambaugh, R. F. (2003): Liquidity risk and expected stock returns. In:
Journal of Political Economy, 111, 642–685.
Pereira, J. P. and Zhang, H. H. (2004): The liquidity premium in a dynamic model with
price impact. Mimeo, University of North Carolina.
Pesaran, M. H. (1987): The Limits to Rational Expectations. Oxford: Basil Blackwell Ltd.
Pirrong, C. (1999): The Organization of Financial Exchange Markets: Theory and Evidence.
In: Journal of Financial Markets, 2, 329–357.
Pirrong, C. (2001): The macrostructure of electronic financial markets. Mimeo, Oklahoma
State University.
Porter, D. C. and Weaver, D. G. (1996): Estimating Bid-Ask Spread Components:
Specialist versus Multiple Market Maker Systems. In: Review of Quantitative Finance
and Accounting, 6, 167–180.
Porter, R. B. (2003): Measuring market liquidity. Mimeo, University of Florida.
Pratt, J. W. (1964): Risk Aversion in the Small and in the Large. In: Econometrica, 32,
122–136.
Roll, R. (1984): A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient
Market. In: Journal of Finance, 39, 1127–1139.
Rosu, I. (2003): A dynamic model of the limit order book. Mimeo, MIT.
Rubio, G. and Tapia, M. (1998): The liquidity premium in equidity pricing under a contin-
uous auction system. In: Eurpean Journal of Finance, 4, 1–28.
Rudolph, B. and Röhrl, H. (1997): Grundfragen der Börsenorganisation aus ökonomischer
Sicht, chapter 2, 143–285. In: Hopt et al. (1997).
Bibliography 293

Saar, G. (2001): Price impact asymmetry of block trades: An institutional trading explana-
tion. In: Review of Financial Studies, 14, 1153–1181.
Schiller, B. and Marek, M. (2000): DTB + SOFFEX = EUREX. In: WiSt, (4), 219–221.
Schmidt, H. and Treske, K. (1996): Komponenten der Geld-Brief-Spanne am Deutschen
Aktienmarkt. In: Zeitschrift für Betriebswirtschaft, 66, 1033–1056.
Schneeweiss, H. (1967): Entscheidungskriterien bei Risiko. Hamburg: Springer Verlag.
Schultz, P. H. (2000): Regulatory and Legal Pressures and the Costs of NASDAQ Trading.
Mimeo, University of Notre-Dame.
Schwartz, R. A. (1988): Equity Markets: Structure, Trading, and Performance. New York:
Harper & Row.
Schwartz, R. A. (1993): Reshaping the Equity Markets: A Guide For the 1990s. Homewood,
Ill.: Business One Irwin.
Seppi, D. J.. Equilibrium block trading and asymmetric information. In: Journal of Finance.
Seppi, D. J. (1997): Liquidity provision with limit orders and a strategic specialist. In: Review
of Financial Studies, 10, 103–150.
Sharpe, W. F. (1964): Capital Asset Prices: A Theory of Market equilibrium Under Condi-
tions of Risk. In: Journal of Finance, 19, 425–442.
Sharpe, W. F. (1970): Portfolio Theory and Capital Markets. New York: McGraw-Hill Book
Co.
Sheffrin, S. M. (1996): Rational Expectations. Cambridge: Cambridge University Press.
Shleifer, A. (2000): Inefficient Markets - An Introduction to Behavorial Finance. Oxford:
Oxford University Press.
Smith, J. W., Selway, J. P. and McCormick, D. T. (1998): The Nasdaq Stock Market:
Historical Background and Current Operation. NASD Working Paper 98-01.
Snell, A. and Tonks, I. (1999): Measuring Microstructure Effects for Less-Liquid Stocks in
a Dealer Market. University of Bristol, Mimeo.
Spiegel, M. and Subrahmanyam, A. (1992): Informed Speculation and Hedging in a
Noncompetitive Securities Market. In: Review of Financial Studies, 5, 307–329.
Stoll, H. R. (1978): The Supply of Dealer Services in Securities Markets. In: Journal of
Finance, 33, 1133–1151.
Stoll, H. R. (1989): Inferring the Components of the Bid-Ask spread: Theory and Empirical
Tests. In: Journal of Finance, 44, 115–134.
Subrahmanyam, A. (1991): Risk Aversion, Market Liquidity, and Price Efficiency. In: Review
of Financial Studies, 4, 417–441.
Swan, P. L. (2002): Can ”illiquidity” explain the equity premium puzzle? the value of
endogenous market trading. Mimeo, University of New South Wales.
Thaler, R. H., (ed.) (1993): Advances in Behavioral Finance. New York: Russel Sage
Foundation.
The NASDAQ Stock Market, Inc., (ed.) (1997): The NASDAQ Stock Market 1997
Factbook. New York.
Tobin, J. (1958): Liquidity Preference as a Behavior Towards Risk. In: Review of Economic
Studies, 25, 65–86.
Tobin, J. (1966): The Theory of Portfolio Selection. In: F. H. Hahn and F. P. R. Brechling:
The Theory of Interest Rates, New York, 3–51.
294 Bibliography

Todorovic, P. (1992): An Introduction to Stochastic Processes and Their Applications. New

York: Springer Verlag.
Unser, M. and Oehler, A. (1998): Das Börsenhandelssystem XETRA. In: WiSt, (9),
463–468.
Vayanos, D. (1998): Transaction costs and asset prices: A dynamic equilibrium model. In:
Review of Financial Studies, 11, 1–58.
Venter, J. H. and de Jongh, D. C. J. (2002): Extending the ekop model to estimate the
probability of informed trading. Mimeo, Centre for Business Mathematics and Informatics,
North-West University, South Africa.
von Neumann, J. and Morgenstern, O. (1953): The Theory of Games and Economic
Behavior. Princeton: Princeton University Press.
Wahal, S. (1997): Entry, Exit, Market Makers, and the Bid-Ask Spread. In: Review of
Financial Studies, 10, 871–901.
Yavas, A. (1992): Market Makers versus Match Makers. In: Journal of Financial Intermedia-
tion, 2, 33–58.
Author index

Acharya, Viral V. 205, 208 Caballe, Jordi 59, 63, 64

Admati, Anat R. 71, 105, 114, 115 Calcagno, Riccardo 102
Affleck-Graves, John 160 Cappozza, Dennis R. 197
Allaz, Blaise 255 Chakravarty, Sugato 126
Amihud, Yakov 81, 195–197, 204 Chan, Soon Huat 216
Angel, James J. 16, 216 Chau, Minh 59
Arnold, Tom 217 Cho, Jin-Wan 149
Arrow, Kenneth J. 7, 255, 285 Chordia, Tarun 205, 207
Aschinger, Gerhard 273 Chowdhry, Bhagwan 215
Atkins, Allen B. 195 Christie, William G. 213, 221
Chung, Kee H. 163
Bacidore, Jeff 215 Chung, Shifei 209
Bagehot, Walter ii, 100–102 Cochrane, John 6
Baker, H. Kent 197 Cohen, Kalman J. 127, 129, 130, 135
Barclay, Michael J. 197 Constantinides, George M. 197
Baruch, Shmuel 202 Copeland, Thomas E. ii, 102
Baum, Harald 2 Cordella, Tito 215
Baumol, William J. 241 Coughenour, Jay F. 216
Beed, Cara 281 Cournot, Augustin 2
Beed, Clive 281 Cymbalista, Flávia 251
Benveniste, Lawrence 197
de Frutosa, M. Angeles 214
Bernhardt, Dan 112
de Jongh, D. C. J. 163
Biais, Bruno 93, 141, 149
Declerck, Fany 215
Black, Fisher 275
Deli, Daniel N. 216
Bloomfield, Robert 214
Demsetz, Harold ii, 77
Blume, Lawrence 65, 67, 68
Dennert, Jürgen 112
Board, John 214
Diamond, Douglas W. 66
Bondarenko, Oleg 59, 214
Domowitz, Ian 20
Booth, G. Geoffrey 161, 215 Duffie, Darrel 6
Bourghelle, David 215 Dumas, Bernard 255
Brachinger, Hans-Wolfgang 258 Dutta, Prajit K. 213
Breen, William J. 202, 205 Dyl, Edward A. 195
Brennan, Michael J. 202
Brooks, Raymond 159 Easley, David i, 65, 67, 68, 161, 163,
Brown, David P. 66, 150 198, 202
296 Author index

Edelen, Roger 216 Hughson, Eric 112

Edelman, Richard B. 197 Hvidkjaer, Soeren 202
Eleswarapu, Venkat R. 197
Ellis, Katrina 164 Ingersoll, Jonathan E. 6
Ellul, Andrew 215 Intriligator, Michael D. 244
Engle, Robert F. 163, 198 Jackson, Matthew 117, 120
Jain, Neelam 59
Fama, Eugene F. 6
Jennings, Robert H. 66
Foster, F. Douglas 59, 70
Jevons, William Stanley 3
Foucault, Thierry 137, 141, 214–216
Johnson, N. 247, 249
Frey, Bruno S. 281
Kadan, Ohad 137
Galai, Dan ii, 102
Kadlec, Gegrory B. 197
Gârleanu, Nicolae 201
Kandel, Eugene 137, 197
Garman, Mark B. i, 77, 81
Kaul, Gautam 160
Gehrig, Thomas 117, 120
Keenan, W. Michael 19
George, Thomas J. 160
Keynes, John Maynard 6, 7, 251, 277
Gervais, Simon 216
Kiefer, Nicholas M. 161, 163
Glosten, Lawrence R. 105, 110, 113,
Kim, Kenneth A. 216
160, 161
Knight, Frank H. 251
Gomber, Peter 27
Kocherlakota, Narayana R. 208, 209
Grammig, Joachim 164
Korajczyk, Robert A. 202, 205
Grossman, Sanford J. 46, 65, 151, 171,
Kotz, S. 247, 249
178
Krause, Andreas 164–166, 214
Hagerty, Kathleen 116 Krishnan, Murugappa 59, 63, 64, 110
Halka, Dominika 205, 207 Kyle, Albert S. ii, 31, 32, 40, 41, 43,
Hanousek, Jan 163 44, 59, 60, 72, 99, 105, 110–112, 122
Hansch, Oliver 160 Laux, Paul A. 161
Harris, Lawrence E. 160, 161, 215 Lee, Ruben 2, 8
Hart, Oliver 216 Lei, Qin 163
Hasbrouck, Joel 205, 207 Levy, Haim 252, 258, 260, 262
Hayek, Friedrich August 7 Li, Mingsheng 163
Hedge, Shantaram P. 160 Lin, Ji-Chai 161, 215
Hersch, Philip 217 Lintner, John 264, 274, 275
Hillion, Pierre 149 Lipson, Marc L. 215
Hirschleifer, David 281 Lo, Andrew W. 149, 197
Ho, Thomas 87, 89, 93, 95, 98, 99 Loderer, Claudio 197
Hodrick, Laurie Simon 202, 205 Lovo, Stefano M. 102
Holden, Craig W. 41, 45, 59, 110, 113, Lucas, Robert E. 280
126 Luckock, Hugh 150
Hollifield, Burton 149
Hopt, Klaus J. 2 Ma, Tai 163
Huang, Ming 197, 205 Madhavan, Ananth 213, 215
Huang, Roger D. 160 Maier, Steven F. 127, 129, 130, 135
Huberman, Gur 205, 207 Mamaysky, Harry 149, 197
Author index 297

Manzano, Carolina 214 Pesaran, M. Hashem 280

Marek, Michael 26 Pfleiderer, Paul 71, 105, 114, 115
Markowitz, Harry M. 257, 258, 273 Pirrong, Craig 216
Marshall, Alfred 3 Podpiera, Richard 163
Martikainen, Teppo 215 Porter, David C. 160
Martı́nez, Miguel A. 207 Porter, R. Burt 208
Marx, Leslie M. 197 Pratt, John W. 255, 256, 261
Masson, Jean 159 Prescott, Edward C. 208, 280
McConnell, John J. 197
McCormick, D. Timothy 222, 230 Rhee, S. Ghon 216
McInish, Thomas H. 163 Röhrl, Heiner 8, 22
Mehra, Rajnish 208 Roll, Richard 155, 166, 205, 207
Mendelson, Haim 81, 195–197, 204 Rosu, Ioanid 150
Merkt, Hanno 2 Roth, Lukas 197
Michaelt, Roni 164 Rubio, Gonzalo 197, 207
Milgrom, Paul R. 105, 110 Rudolph, Bernd 8, 22
Miller, Merton H. 171, 178
Miller, Robert A. 149 Saar, Gideon 202, 215
Miller, Robert E. 160 Salani, François 141
Mirman, Leonard J. 59 Sandås, Patrick 149
Moinas, Sophie 214 Sanger, Gary C. 161
Moore, John 216 Sarnat, Marshall 252, 258, 260, 262
Morgenstern, Oskar 252 Schiller, Bettina 26
Moulin, Hervé 20, 216 Schmidt, Hartmut 160
Mulherin, J. Harold 217 Schneeweiss, H. 263
Muth, John F. 279, 280 Schultz, Paul H. 213, 221, 232
Schwartz, Robert A. 2, 20, 23, 127,
Naik, Narayan Y. 160 129, 130, 135
Nanda, Vikram 215 Seguin, Paul J. 197
Neal, Robert 160 Selway, James P. 222, 230
Nelling, Edward 149 Seppi, Duane J. 205, 207, 214–216
Netter, Jeffry 217 Sharpe, William F. 274, 275
Nieto, Belén 207 Sheffrin, Steven M. 279
Nimalendran, M. 160 Shin, Hyun Song 215
Němeček, Libor 163 Shleifer, Andrei 281
Oehler, Andreas 23 Smith, Jeffrey W. 222, 230
O’Hara, Maureen i, 9, 65, 67, 68, 161, Snell, Andy 160
163, 164, 198, 202, 214 Spatt, Chester 149
Spiegel, Matthew 49, 58
Panzar, John C. 241 Stambaugh, Robert F. 205, 207
Paperman, Joseph B. 161, 163 Stiglitz, Joseph E. 46, 65, 151
Parlour, Christine A. 132, 135, 216 Stoll, Hans R. ii, 79, 80, 86, 87, 89, 93,
Pástor, Luboš 205, 207 95, 98, 99, 154, 155, 160
Pedersen, Lasse Heje 201, 205, 208 Subrahmanyam, Avanidhar 49, 58,
Pereira, Joao Pedro 205 202, 205, 207
298 Index

Subrahmanyam, Avinadhar 41, 45, 59, von Neumann, John 252

110, 113
Sung, Jaeyoung 214 Wahal, Sunil 214
Sutcliffe, Charles 214 Wang, Jiang 149, 197
Swan, Peter L. 197 Weaver, Daniel G. 160, 216
Weber, Marianne 258
Tapia, Mikel 197, 207 Wei, Peihwang 209
Theissen, Erik 164, 214 Wheatley, Simon 160
Tobin, James 258, 271 Whitcomb, David K. 127, 129, 130,
Todorovic, Petar 250 135
Tonks, Ian 160, 215
Willig, Robert D. 241
Treske, Kai 160
Wu, Guojun 163
Tse, Yiuman 215
Wu, Liuren 163, 198
Unser, Matthias 23
Yang, Josephine 163
Vayanos, Dimitri 197 Yavas, Abdullah 179, 185, 187, 188
Venter, J. H. 163
Verrecchia, Robert E. 66 Zhang, Harold H. 205
Viswanathan, S. 59, 70, 160 Zhang, Zhi Ming 150
Index

hhh, i
jjj, i

299
300 Index
This book explains how the way assets are traded in markets affect their prices.
It explores how asymmetric information between market participants cause
informed traders to behave strategically in order to maximize the profits they
make from their information. This behavior causes information to be incorpo-
rated into the price gradually. Also in the absence of asymmetric information
will some traders behave strategically in order to minimize any trading costs, e.g.
through the submission of optimal limit orders. These are only a few examples
of the way trading rules in markets affect behavior and thus the prices emerging.

Andreas Krause is Lecturer in Finance at the School of Management of the

University of Bath, Great Britain. He obtained a PhD in Finance from the
University of Fribourg (Switzerland) with his dissertation on strategic behavior
of market makers in dealer markets. He has published several papers on market
microstructure theory and extensive teaching experience at all undergraduate and
postgraduate levels.

Price £25