0% found this document useful (0 votes)
345 views27 pagesTime To Build and Aggregate Fluctuations
POST KEYNESIAN
Uploaded by
Rick AntonyCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
345 views27 pagesTime To Build and Aggregate Fluctuations
POST KEYNESIAN
Uploaded by
Rick AntonyCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
Time to Build and Aggregate Fluctuations
Finn B, Kydland, Edward C. Prescott
Econometrica, Volume 50, Issue 6 (Nov., 1982), 1345-1370.
Stable URL:
hutp:/inks,[Link]/sici?sici=0012-9682%28 1982 1 1%2950%3A6%3C1345%3ATTBAAP%[Link]%3B2-E
‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at
hhup:/[Link]/about/[Link]. JSTOR’s Terms and Conditions of Use provides, in part, that unless you
have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and
you may use content in the JSTOR archive only for your personal, non-commercial use.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or
printed page of such transmission,
Econometrica is published by The Beonometric Society. Please contact the publisher for further permissions
regarding the use of this work, Publisher contact information may be obtained at
hup:/www jstor-org/journalsfeconosoe. hm
Econometrica
(©1982 The Econometric Society
ISTOR and the ISTOR logo are trademarks of ISTOR, and are Registered in the U.S. Patent and Trademark Office.
For more information on ISTOR contact jstor-info@[Link].
©2002 JSTOR
bupulwww [Link]/
Tue Mar $ [Link] 2002 ECONOMETRICA
‘Votume 50 ‘Novenper, 1982 ‘Numer 6
TIME TO BUILD AND AGGREGATE FLUCTUATIONS
By Fwy E. KyDLAND AND Epwarp C. Prescott!
‘The equilibrium growth mode! i mosifid and used to explain the yclcal variances of
‘set of economic time series, the covariances between real output and the other series, and
the autocovariance of ouput. The model i ited to quarterly data forthe post-war US.
‘conomy. Crucial features of the model ae the assumption that more than one time period
is required for the construction of new productive capital, and the non-time-separable
tly funetion that admits greater intertemporal substtation of leisure. The ft is surpes-
ingly good in ight ofthe mode!’ simplicity and the small numberof free parameters
1. INTRODUCTION
‘THAT WINE 18 NoT MADE in a day has long been recognized by economists (e.g.,
‘Bohm-Bawerk (6). But, neither are ships nor factories built in a day. A thesis of
this essay is that the assumption of multiple-period construction is crucial for
explaining aggregate fluctuations. A general equilibrium model is developed and
fitted to U.S. quarterly data for the post-war period. The co-movements of the
fluctuations for the fitted model are quantitatively consistent with the corre-
sponding co-movements for U.S. data. In addition, the serial correlations of
cyclical output for the model match well with those observed.
Our approach integrates growth and business cycle theory. Like standard
‘growth theory, a representative infinitely-lived household is assumed. As fluctua
tions in employment are central to the business cycle, the stand-in consumer
values not only consumption but also leisure. One very important modification to
the standard growth model is that multiple periods are required to build new
capital goods and only finished capital goods are part of the productive capital
stock. Each stage of production requires a period and utilizes resources. Half-
finished ships and factories are not part of the productive capital stock. Section 2
contains a short critique of the commonly used investment technologies, and
presents evidence that single-period production, even with adjustment costs, is
inadequate. The preference-technology-information structure of the model is
presented in Section 3. A crucial feature of preferences is the non-time-separable
utility function that admits greater intertemporal substitution of leisure. The
exogenous stochastic components in the model are shocks to technology and
imperfect indicators of productivity. The two technology shocks differ in their
persistence.
‘The steady state for the model is determined in Section 4, and quadratic
approximations are made which result in an “indirect” quadratic utility function
that values leisure, the capital goods, and the negative of investments. Most of
"The research was supported by the National Science Foundation. We are grateful to Sean
Beckett, Fischer Black, Robert S. Chirinko, Mark Gersovit, Christopher A. Sims, and John B.
‘Tayler for helpful comments, to Sumru Alt for research assistance, and tothe participants in the
seminars at the several universities at which earler drafts were presented
14s 1346 FE. KYDLAND AND E. C. PRESCOTT
the relatively small number of parameters are estimated using steady state
considerations. Findings in other applied areas of economics are also used to
calibrate the model. For example, the assumed number of periods required to
build new productive capital is of the magnitude reported by business, and
findings in labor economics are used to restrict the utility function. The small set
of free parameters imposes considerable discipline upon the inquiry. The esti-
‘mated model and the comparison of its predictions with the empirical regularities
Of interest are in Section 5. The final section contains concluding comments.
2. ACRITIQUE OF CONVENTIONAL AGGREGATE INVESTMENT.
‘TECHNOLOGIES
There are two basic technologies that have been adopted in empirical studies
of aggregate investment behavior. The first assumes a constant-returns-to-scale
neoclassical production function F with labor L and capital K as the inputs
Total output F(K,L) constrains the sum of investment and consumption, oF
CH+I< FKL), where C,1,K,L>0. The rate of change of capital, K, is
investment less depreciation, and depreciation is proportional with factor 6 to the
capital stock, that is, K = I — 8K. This is the technology underlying the work of
Jorgenson [19] on investment behavior.
‘An implication of this technology is that the relative price of the investment
‘and consumption goods will be a constant independent of the relative outputs of
the two goods. It also implies that the shadow price of existing capital will be the
same as the price of the investment good.” There is a sizable empirical literature
that has found a strong association between the level of investment and a shadow
price of capital obtained from stock market data (see [26)). This finding is
inconsistent with this assumed technology as is the fact that this shadow price
varies considerably over the business cycle.
The alternative technology, which is consistent with these findings, is the single
capital good adjustment cost technology.‘ Much of that literature is based upon
the problem facing the firm and the aggregation problem receives litle attention.
This has led some to distinguish between internal and external adjustment costs.
For aggregate investment theory this is not an issue (see (29) though for other
questions it will be. Labor resources are needed to install capital whether the
acquiring or supplying firm installs the equipment. With competitive equilibrium
it is the aggregate production possibility set that matters. That is, ifthe Y, are the
production possibility sets of the firms associated with a given industrial orga
2 This, of course, assumes neither C nor / is zero. Sargent [32, within a growth context with shocks
to both preferences and technology, as ata theoretical level analyzed the equibrium with corners
Only when investment was zero did the price of the investment good relative to that of the
consumption good become different from one and then it was less than one. This Was aot an
‘empirical study and Sargent states that there currently are no computationally practical econometic
‘methods for conducting an empirical investigation within that theoretical framework.
"The shadow price of capital has been emphasized by Brunner and Mele: [7] and Tobin [36] in
their agaregate models,
“See [1,17] for recent empirical studies based on this technology. AGGREGATE FLUCTUATIONS 1347
zation and ¥/ for some other industrial organization, the same aggregate supply
behavior results if S¥, = SY)
The adjustment cost model, rather than assuming a linear product transforma-
tion curve between the investment and consumption goods, imposes curvature.
This can be represented by the following technology:
GCS FKL), K=1~8K,
where G like F is increasing, concave, and homogeneous of degree one. Letting
the price of the consumption good be one, the price of the investment good 4,,
the rental price of capital r,, and the wage rate w,, the firm's problem is to
maximize real profits, C,+ q,J,~w,L, ~ r,K;, subject to the production con-
straint. As constant returns to scale are assumed, the distribution of capital does
not matter, and one can proceed as if there were a single price-taking firm.
‘Assuming an interior solution, given that this technology displays constant
returns to scale and that the technology is separable between inputs and outputs,
it follows that 1, = F(K,,L,)h(q,) = Z,h(q,), where Z, is defined to be aggregate
output. The function h is increasing, so high investment-output ratios are
associated with a high price of the investment good relative to the consumption
‘good. Figure I depicts the investment-consumption product transformation curve
and Figure 2 the function ’(q). For any 1/Z, the negative of the slope of the
transformation curve in Figure 1 is the height of the curve in Figure 2. This
establishes that a higher q will be associated with higher investment for this
technology. This restriction of the theory is consistent with the empirical findings
previously cited.
There are other predictions of this theory, however, which are questionable. If
‘we think of the g-investment curve h depicted in Figure 2 as a supply curve, the
short- and the long-run supply elasticities will be equal. Typically, economists
argue that there are specialized resources which cannot be instantaneously and
costlessly transferred between industries and that even though short-run elasti
ties may be low, in the long run supply elasticities are high. As there are no
specialized resources for the adjustment cost technology, such considerations are
absent and there are no penalties resulting from rapid adjustment in the relative
outputs of the consumption and investment good.
cuz 4 1
10 \
1
It
1
1
1.0 17z 10 172
Frou | Fours 2 1348 FE. KYDLAND AND E. C. PRESCOTT
To test whether the theory is a reasonable approximation, we examined
cross-section state data. The correlations between the ratios of commercial
construction to either state personal income or state employment and price per
square foot’ are both ~0.35. With perfectly elastic supply and uncorrelated
supply and demand errors, this correlation cannot be positive. To explain this
large negative correlation, one needs a combination of high variability in the
cross-sectional supply relative to cross-sectional demand plus a positive slope for
the supply curve. Our view is that, given mobility of resources, it seems more
plausible that the demand is the more variable. Admitting potential data prob-
Jems, this cross-sectional result casts some doubt upon the adequacy of the single
capital good adjustment cost model,
‘At the aggregate level, an implication of the single capital good adjustment
cost model is that when the investment-output ratio is regressed on current and
lagged q, only current q should matter.° The findings in [26] are counter to this
prediction.
In summary, our view is that neither the neoclassical nor the adjustment cost
technologies are adequate. The neoclassical structure is inconsistent with the
positive association between the shadow price of capital and investment activity.
‘The adjustment cost technology is consistent with this observation, but inconsis-
tent with cross-sectional data and the association of investment with the lagged
fas well as the current capital shadow prices. In addition, the implication that
Jong- and short-run supply elasticities are equal is one which we think a
technology should not have.
Most destructive of all to the adjustment-cost technology, however, is the
finding that the time required to complete investment projects is not short
relative to the business cycle. Mayer [27], on the basis of a survey, found that the
average time (weighted by the size of the project) between the decision to
undertake an investment project and the completion of it was twenty-one
‘months. Similarly, Hall [13] found the average lag between the design of a
project and when it becomes productive to be about two years. It is a thesis of
this essay that periods this long or even half that long have important effects
upon the serial correlation properties of the cyclical components of investment
and total output as well as on certain co-movements of aggregate variables.
‘The technological requirement that there are multiple stages of production is
not the delivery lag problem considered by Jorgenson (19]."He theorized at the
firm level and imposed no consistency of behavior requirement for suppliers and
demanders of the investment good. His was not a market equilibrium analysis
and there was no theory accounting for the delivery lag. Developing such a
market theory with information asymmetries, queues, rationing, and the like is a
challenging problem confronting students of industrial organization.
The data on commercial constuction and price per square foot were for 1978 and were obtained
from F. W. Dodge Division of McGraw-Hill,
“TThis observation is due to Fumio Hayashi AGGREGATE FLUCTUATIONS 1349
ur technology assumes that a single period is required for each stage of
construction or that the time required to build new capital is « constant. This is
not to argue that there are not alternative technologies with different construe-
ion periods, patterns of resource use, and total costs. We have found no
evidence that the capital goods are built significantly more rapidly when total
investment activity is higher or lower. Lengthening delivery lags (see (9)
periods of high activity may be a matter of longer queues and actual construction
times may be shorter. Premiums paid for earlier delivery could very well be for a
more advanced position in the queue than for a more rapidly constructed
factory. These are, of course, empirical questions, and important cyclical varia-
tion in the construction period would necessitate an alternative technology.
Our time-to-build technology is consistent with short-run fluctuations in the
shadow price of capital because in the short run capital is supplied inelastically.
It also implies that the long-run supply is infinitely elastic, so on average the
relative price of the investment good is independent of the investment-output
ratio.
3. THE MODEL
Technology
‘The technology assumes time is required to build new productive capital. Let
5, be the number of projects j stages or j periods from completion for j
;J— 1, where J periods are required to build new productive capacity.
‘New investment projects initiated in period ¢ are s,,. The recursive representation
Of the laws of motion of these capital stocks is
BN) kare (= 8) +5,
G2) Susi
Here, kis the capital stock at the beginning of period r, and 8 is the depreciation
rate. The element s,, is a decision variable for period 1.
‘The final capital good is the inventory stock y, inherited from the previous
period.” Thus, in this economy, there are J + 1 iypes of capital: inventories y,,
productive capital k,, and the capital stocks j stages from completion for
J=,...,J— 1. These variables summarize the effects of past decisions upon
‘current and future production possibilities.
Let g, for j=l,...,J be the fraction of the resources allocated to the
investment project inthe /th stage from the last. Total non-inventory investment
in period 1 is S¥_.gj5,. Total investment, i, is this amount plus inventory
"All socks are bepinsing-ofthe-period stocks 1350 FE. KYDLAND AND E. C. PRESCOTT.
investment y,,,, — y,. and consequently
BI =D et Her
A
Total output, that is, the sum of consumption ¢, and investment, is constrained as
follows:
BA HU S fOrckiots Ys
where 1, is labor input, A, a shock to technology, and f is a constant-returns-to-
scale production function to be parameterized subsequently.
Treating inventories as a factor of production warrants some discussion.
larger inventories, stores can economize on labor resources allocated to restock-
ing. Firms, by making larger production runs, reduce equipment down time
associated with shifting from producing one type of good to another. Besides
considerations such as these, analytic considerations necessitated this approach.
If inventories were not a factor of production, it would be impossible to locally
approximate the economy using a quadratic objective and linear constraints.
Without such an approximation no practical computational method currently
cexists for computing the equilibrium process of the model.
‘The production function is assumed to have the form
]
where 0< <1, 0 1, where 0 Pox) for k < t. Using the conditional probability
Jaws for the multivariate normal distribution (see [28, p. 208)) and letting m,, and
, be the mean and covariance of x, conditional upon p,, as well, we obtain
B14) my, = mo, + (Bi20) (208i + Vi) (pir Br), and
BIS) By =~ (Bi) (B28) + Vi)" 'B.Z
Similarly, the mean vector m,, and covariance matrix , conditional upon p2, as
well are
B16) tay = my + (B2Es)'(B2E\B5 + Va) "(pre Bam)» and
GIT) 3,= 3, — (B31) (B22 18s + Vo) 'BEs. 1354 FE. KYDLAND AND E. C. PRESCOTT
Finally, from (3.11),
G.18)_ mo, 4) = Amy, and
G19) Hy= ABA'+ Vo
The covariances Ep, 2), and 3, are defined recursively by G.15), (3.17), and
3.19). The matrix V, being of full rank along with the stability of A are sufficient
to insure that the method of successive approximations converges exponentially
fast to a unique solution.
‘The covariance elements Ep, 2), and Z, do not change over time and are
therefore not part of the information set. The mo,» my,» and ms, do change but
are sufficient relative to the relevant histories for forecasting future values of
both the unobserved state and the observable p,, 7 >t, and for estimating the
current unobserved state,
Equilibrium
To determine the equilibrium process for this model, we exploit the well-
Known result that, in the absence of externalities, competitive equilibria are
Pareto optima. With homogeneous individuals, the relevant Pareto optimum is
the one which maximizes the welfare of the stand-in consumer subject to the
technology constraints and the information structure. Thus, the problem is to
maximize ES) Buf. 1 ~ ayn, ~ n(1 ~ a9)¢,]
subject to constraints (3.1)-(3.4), (3.6), and (3.11)-(3.13), given kg,
So. + + 4)-19y Qy and that xy~W¥ (mg, The decision variables at time ¢ are
re Siw Gy and yy. Further, n, and s,, cannot be contingent upon p, for itis
observed subsequent to these decisions.
This is a standard discounted dynamic programming problem. There are
‘optimal time-invariant or stationary rules of the form
= OK Sue 8209 Sr—14 Jody Muy
SI = (Ki Sty S200 6 9 Sat Fa Be May
(ho Sa Se Jd see
Ves VCR Sts S200 0+ 9 Spe» Yoo BoM 9 May)
It is important to note that the second pair of decisions are contingent upon m,
rather than m,, and that they are contingent also upon the first set of decisions sy,
and 7,
The existence of such decision rules and the connection with the competitive
allocation is established in [31]. But, approximations are necessary before equilib-
rium decision rules ean be computed. Our approach is to determine the steady AGGREGATE FLUCTUATIONS 1355
state for the model with no shocks to technology. Next, quadratic approxima
tions are made in the neighborhood of the steady state. Equilibrium decision
rules for the resulting approximate economy are then computed. These rules are
linear, so in equilibrium the approximate economy is generated by a system of
stochastic difference equations for which covariances are easily determined.
4, STEADY STATE, APPROXIMATION, AND COMPUTATION OF EQUILIBRIUM
Variables without subscript denote steady state values. The steady state
interest rate is r= (I~ 8)/, and the steady state price of (non-inventory)
capital q= ¥}._(1 + 17> "yy. The latter is obtained by observing that units of
consumption must be foregone in the current period, @) units the period before,
etc., in order to obtain one additional unit of capital for use next period.
Two steady state conditions are obtained by equating marginal products to
rental rates, namely f=, and f, = q(r + 8). These imply f,/, = g(r-+ 8)/r.
For production function (3.5), this reduces to
ke
1) »-[4?
vos
; | 2
Differentiating the production function with respect to capital, substituting for y
from (4.1), and equating to the steady-state rental price, one obtains
(1 = 8) = yy “8-9/0
where 6 = 1 — 0 + 0b;*. Solving for k as a function of n yields
=4(r+8),
(= 8)= 0) ota
qty
4k -| an
Steady-state output as a function of m is f= by !-/"6)-"\"n = bAW%n, In the
steady state, net investment is zero, so
43) C= DAY Sn — Bk = (by — Bb,)AM%n.
‘The steady-state values of c, k, and y are all proportional to A'/%n, We also note
that the capital-output ratio is 6;/6,, and that consumption’s share to total
steady-state output is I~ (8,/b,).
‘Turning now to the consumer’s problem and letting be the Lagrange
rmultiplier for the budget constraint and w, the real wage, first-order conditions
eMa(LyI, "= py and
1
3
3S paertNatty..)™ 1356 FE. KYDLAND AND E. C. PRESCOTT
In the steady state, ¢ I, and w, = w for all r. Making these substitutions
and using the fact that the a, sum to one, these expressions simplify to
+h
owl.
$ererfmpe, and 2,020" $ a
Eliminating 4 from these equations yields 2cD*.of'aj = wl. Since Doe,
= + (1 = a)n/(r+ n) and f= I~ n, this in turn implies
(44) 2e(ag + (1 = ag)n/(r + 1)) = w(1 = 0),
Returning to the production side, the marginal product of labor equals the real
wage:
45) = Ob,A'”,
Using (4.3) and (4.5), we can solve (4.4) for n:
+200
Hew (1 hy/b9)
That m does not depend upon average A matches well with the American ex-
perience over the last thirty years. During this period, output per man-hour has
increased by a few hundred per cent, yet man-hours per person in the 16-65 age
group has changed but a few per cent.
Approximation About the Steady State
If the utility function w were quadratic and the production function / linear,
there would be no need for approximations. In equilibrium, consumption must
be equal to output minus investment. We exploit this fact to eliminate the
nonlinearity in the constraint set by substituting /(\,k,n, ») ~ i for c in the utility
function to obtain u(f,sm, y) — ina). The next step is to approximate this
function by a quadratic in the neighborhood of the model’s steady state. As
investment is linear in the decision and state variables, it can be eliminated
subsequent to the approximation and still preserve a quadratic objective.
Consider the general problem of approximating function u(x) near ¥. The
approximate quadratic function is
U(x) = u(R) + B(x — 3) + (x- RV Q(X-B),
where x, b © Rt and Q is ann x n symmetric matrix. We want an approximation
that is good not only at ¥ but also at other x in the range experienced during the
sample period. Let 2' be a vector, all of whose components are zero except for
2{ > 0. Our approach is to select the elements 6, and q,, so that the approxima- AGGREGATE FLUCTUATIONS 1357
tion error is zero at the ¥ +2! and ¥~z', where the z/ selected correspond to the
approximate average deviations of the x, from their steady state values ¥,. The
values of z//F, used for A, ky ys m, i, and a were 3, 1, 2, 3, 8 and 0.5 per cent,
respectively."!
The approximation errors being zero at the ¥ +2! and ¥~z! requires that
bole +e!) —u(@—24)]/22,, and
ay [ue +2!) — u(R) + u(e=2') — u(z)]/222.
The element gj, je slected 1 minimize the sum of the squared approxi
mation errors at ¥+2' + 2/, ¥+2'—2/, ¥—2'+ z2/, and ¥—z'—z/. The ap-
proximation emor atte ft point i
We) = WR) — be, bs — ge — 32928)
Samming over the square of this eror andthe thee other dfereniatng with
respect to g,, setting the resulting expression equal to zero and solving for 4, we
obtain
a(R +2! 2/)—u(R +2! ~ 2!) u(F 2! + 2)
+u(E—2!— 2/)]/82,
for ij
Computation of Equilibrium
‘The equilibrium process for the approximate economy maximizes the welfare
of the representative household subject to the technological and informational
constraints as there are no externalities. This simplifies the determination of the
equilibrium process by reducing it to solving a linear-quadratic maximization
problem. For such mathematical structure there is a separation of estimation and
control. Consequently, the first step in determining the equilibrium decision rules
for the approximate economy is to solve the following deterministic problem:
max 3 BUC ams yeodstiva)
"We experimented a litle and found thatthe results were esentilly the same when the second
‘order Taylor series approximation was used rather than this function, Larry Christiano [10] has found
that the quadratic approximation method that we employed yields approximate solutions that are
very accurate, even with large variably, fora structure that, like ous, sof the constant elasticity
variety. 1358 F.E, KYDLAND AND E. C. PRESCOTT
subject to
46) he 0-8) His
OD Serr= Seu Ue bee I~ Ds
8) y= AR,
49) a= (1 ma, +
410) = D Ot Ie Yor
A
GID = xy ty
‘At this stage, the fact that there is an additive stochastic term in the equation
determining x,,., is ignored as is the fact that x, is not observed for our economy.
Constraints (4.6)-(4.9) are the laws of motion for the state variables. The free
decision variables are n,,s,,, and y,,. It was convenient to use inventories taken
into the subsequent period, y,,., as a petiod decision variable rather than i,
because the decisions on inventory carry-over and consumption are made subse-
quent to the labor supply and new project decisions n, and s,,.
For notational simplicity we let the set of state variables other than the
unobserved x, be 2, = (ks JisdiSig + +++$y-1,) and the set of decision variables
4, = (Sy Joey)» The unobserved state variables x,=(x),.2,) are the perma-
nent and transitory shocks to technology. Finally, o(x,z) is the value of the
deterministic problem if the initial state is (x,z). It differs from the value
funetion for the stochastic problem by a constant.
Using constraints (4.10) and (4.11) to substitute for j, and A, in the utility
function, an indirect utility function U(x,2,d) is obtained. The value function,
v(2x,2), was computed by the method of successive approximations or value
iteration. If v,(x,2) is the jth approximation, then
pan(o2) = max[ UC 204.) + BO (1 z041)]
subject to constraints (4.6)-(4.9). The initial approximation, oy(x,z), is that
function which is identically zero.
‘The function U is quadratic and the constraints are linear. Then, if v, is
quadratic, vj, must be quadratic. As vy is trivially quadratic, all the v, are
quadratie and therefore easily computable. We found that the sequence of
quadratic functions converged reasonably quickly.!?
"= The limit of the sequence of value Functions existed in every case and, a8 a function of 2, was
bounded from above, given. This, along withthe stability ofthe matrix 4, is sufiient to
that thi iti the optimal value function and thatthe associated policy function the optimal one
(see (30). AGGREGATE FLUCTUATIONS 1359
The next step is to determine the optimal inventory carry-over decision rule. It
is the linear function y,. 4 = y(X,+4,+M,5,) Which solves
4.12) max [ UC 52,51 8060 Foon) + BOC 41 Zr41)]
subject to (4.6)-(49) and both n, and s,, given. Finally, the solution to the
program
man en(% 0214 Sn)>
where 0, is the value of maximization of (4.12), is determined. The linear
functions s,, = s(x,,2,) and n,= n(x;,2,) which solve the above program are the
‘optimal decision rules for new projects and labor supply.
Because of the separation of estimation and control in our model, these
decision rules can be used to determine the motion of the stochastic economy. In
each period r, a conditional expectation, my, is formed on the basis of observa-
tions in previous periods. An indicator of the technology shock is observed,
which is the sum of a permanent and a transitory component as well as an
indicator shock. The conditional expectation, my,, of the unobserved x, is
computed according to equation (3.14), and s,, and n, are determined from
(4.13) = s(m,,,2),
414) n= mC, 42),
where x, has been replaced by m,,. Then the technology shock, 2,, is observed,
which changes the conditional expectation of x,.. From (3.16), this expectation is
‘my, and the inventory carry-over is determined from
GIS) Yee 1 = PCMag 2408s).
To summarize, the equilibrium process governing the evolution of our economy
is given by (3.1)-G33), (3.6), (3.11)-G.14), (3.16), (3.18), and (4.13)-(4.15).
5. TEST OF THE THEORY
‘The test of the theory is whether there is a set of parameters for which the
‘model's co-movements for both the smoothed series and the deviations from the
smoothed series are quantitatively consistent with the observed behavior of the
corresponding series for the U.S. post-war economy. An added requirement is
that the parameters selected not be inconsistent with relevant micro observations,
including reported construction periods for new plants and cross-sectional obser-
vations on consumption and labor supply. The closeness of our specification of
preferences and technology to those used in many applied studies facilitates such
‘comparisons.
The model has been rigged to yield the observations that smoothed output,
investment, consumption, labor productivity, and capital stocks all vary roughly 1360 FE. KYDLAND AND E. C. PRESCOTT
proportionately while there is little change in employment (all variables are in
per-household terms) when the technology parameter \ grows smoothly over
time. These are just the steady state properties of the growth model with which
wwe began,
Quantitatively explaining the co-movements of the deviations is the test of the
underlying theory. For want of better terminology, the deviations will be referred
to as the cyclical components even though, with our integrated approach, there is
zo separation between factors determining a secular path and factors determin-
ing deviations from that path. The statistics to be explained are the covariations
‘of the cyclical components. They are of interest because their behavior is stable
and is so different from the corresponding covariations of the smoothed series.
This is probably why many have sought separate explanations of the secular and
cyclical movements.
‘One cyclical observation is that, in percentage terms, investment varies three
times as much as output does and consumption only half as much. In sharp
contrast to the secular observations, variations in cyclical output are principally
the result of variations in hours of employment per household and not in capital
stocks or labor productivity.
‘The latter observation is a difficult one to explain. Why does the consumption
of market produced goods and the consumption of leisure move in opposite
directions in the absence of any apparent large movement in the real wage over
the so-called cycle? For our model, the real wage is proportional to labor's
productivity, so the crucial testis whether most of the variation in eyelical output
arises from variations in employment rather than from variations in labor's
productivity.
We chose not to test our model versus the less restrictive vector autoregressive
model.!? This most likely would have resulted in the model being rejected, given
the measurement problems and the abstract nature of the model. Our approach
is to focus on certain statistics for which the noise introduced by approximations
and measurement errors is likely to be small relative to the statistic. Failure of the
theory to mimic the behavior of the post-war U.S. economy with respect to these
stable statistics with high signal-noise ratios would be grounds for its rejection.
Model Calibration
There are two advantages of formulating the model as we did and then
constructing an approximate model for which the equilibrium decision rules are
linear. First, the specifications of preferences and technology are close to those
used in many applied studies. This facilitates checks of reasonableness of many
parameter values. Second, our approach facilitates the selection of parameter
values for which the model steady-state values are near average values for the
‘American economy during the period being explained. These two considerations
"Sims [34] has estimated unrestricted agarepate vector autoregressive model. AGGREGATE FLUCTUATIONS 1361
reduce dramatically the number of free parameters that will be varied when
searching for a set that results in cyclical covariances near those observed. In
explaining the covariances of the cyclical components, there are only seven free
parameters, with the range of two of them being severely constrained a priori.
Capital for our model reflects all tangible capital, including stocks of plant and
equipment, consumer durables and housing. Consumption does not include the
purchase of durables but does include the services from the stock of consumer
durables. Different types of capital have different construction periods and
patterns of resource requirements. The findings summarized in Section 2 suggest
fan average construction period of nearly two years for plants. Consumer
durables, however, have much shorter average construction periods. Having but
‘one type of capital, we assume, as a compromise, that four quarters are required,
with one-fourth of the value put in place each quarter. Thus J =4 and 9, = 9,
= 95 = 947 025
Approximately ten per cent of national income account GNP is the capital
consumption allowance and another ten per cent excise tax. To GNP should be
added the depreciation of consumer durables which has the effect of increasing,
the share of output going to owners of capital. In 1976, compensation to
employees plus proprietary income was approximately 64 per cent of GNP plus
consumer durables depreciation less indirect business tax, while owners of capital
received about 36 per cent. As labor share is 8, we set 0 = 0.64.
Different types of capital depreciate more rapidly than others, with durables
depreciating more rapidly than plant and housing, and land not depreciating at
all. As a compromise, we set the depreciation rate equal to 10 per cent per year.
‘We assume a subjective time discount rate of four per cent and abstract from
growth. This implies a steady-state capital to annual output ratio of 2.4. Of total
‘output 64 per cent is wages, 24 per cent depreciation, and 12 per cent return on
capital which includes consumer durables.
The remaining parameters of technology are average A, which we normalize to
‘one by measuring output in the appropriate units, and parameters o and », which
determine the shares of and substitution between inventories and capital. Inven-
tories are about one-fourth of annual GNP so we require » and o to be such that
k/y=110. A priori reasoning indicates the substitution opportunities between
‘capital and inventory are small, suggesting that r should be considerably larger
than zero. We restricted it to be no less than two, but it is otherwise a free
parameter in our search for a model to explain the cyclical covariances and
autocovariances of aggregate variables. Given » and the value of b; = y/k, @ is
implied. From (4.1) itis o = [1 + g(r + 8)/(r5;*)]-'. For purposes of explain-
ing the covariances of the percentage deviation from steady state values,» is the
only free parameter associated with technology.
The steady state real interest rate r is related to the subjective time discount
rate, p= ~'~1, and the risk aversion parameter, y, by the equation r= p+
(1 = 1(2/c), where é/c is the growth rate of per capita consumption. We have
assumed p is four per cent per year (one per cent per quarter). As the growth rate 1362 F. E. KYDLAND AND E, C. PRESCOTT
of per capita consumption has been about two per cent and the real return on
physical capital six to eight per cent, the risk aversion parameter, y, is con-
strained to be between minus one and zer0.!*
The parameters ay and 7 which affect intertemporal substitutability of leisure
will be treated as free parameters for we could find no estimate for them in the
labor economics literature. As stated previously, the steady-state labor supply is
independent of the productivity parameter 1. The remaining parameters are
those specifying the process on 2, and the variance of the indicator. These three
parameters are var(t,), var(S,), and var(f,). Only two of these are free parame-
ters, however. We restricted the sum of the three variances to be such that the
estimate of the variance of cyclical output for the model equalled that of eyclical
output for the U.S. economy during the sample period.
In summary, the parameters that are estimated from the variance-covariance
properties of the model are these variances plus the parameter » determining
substitutability of inventories and capital, the parameters ay and 7 determining
intertemporal substitutability of leisure, and the risk aversion parameter y. For
cach set of parameter values, means and standard deviations were computed for
several statistics which summarize the serial correlation and covariance proper-
ties of the model. These numbers are compared with those of the actual U.S.
data for the period 1950: | to 1979:2 as reported in Hodrick and Prescott [18]. A
set of parameter values is sought which fits the actual data well. Having only six
degrees of freedom to explain the observed covariances imposes considerable
discipline upon the analysis.
‘The statistics reported in [18] are not the only way to quantitatively capture the
‘co-movements of the deviations.'® This approach is simple, involves a minimum
of judgment, and is robust to slowly changing demographic factors which affect
growth, but are not the concern of this theory." In addition, these statistics are
robust to most measurement errors, in contrast to, say, the correlations between
the first differences of two series. It is important to compute the same statistics
for the U.S. economy as for the model, that is, to use the same function of the
data. This is what we do.
‘A key part of our procedure is the computation of dynamic competitive
‘equilibrium for each combination of parameter values. Because the conditional
forecasting can be separated from control in this model, the dynamic equilibrium
