381
Previously, others had effectively endogenized the savings decision by ana-
lyzing the optimal growth path because, by the second welfare theorem, the
optimal path is the competitive equilibrium path for this model.
5
But in order
for the model to be used to study business cycle fluctuations, the labor supply
decision must be endogenized as well.
6
Step 2: Modify the National Accounts to Be Consistent with the Theory
Prior to our work, macroeconomics was concerned with developing a theory
of the national accounts statistics. Preferences and technology are the given,
not the national accounts statistics. This means that we had to modify the
national accounts to be consistent with the theoretical abstraction or model
we used. The most important modification when studying business cycles is to
treat consumer durable expenditures as an investment in the same way that
expenditures on new housing and home improvement are treated as invest-
ments in the national accounts. Once this is done, services of consumer
durables and consumer durable rental income must be imputed, in much the
same way as is currently done for owner-occupied housing. This increases
investment share of output and has consequences for the cyclical behavior of the
economy. What led us to think about this issue is that consumer durable expen-
ditures are highly variable, behaving very similarly to producer durable invest-
ments and not like consumer expenditures on nondurable goods and services.
Step 3: Restrict the Model to Be Consistent with the Growth Facts
The growth facts are that consumption and investment shares of output are
roughly constant, as are labor and capital cost shares. All the variables and
the real wage grow over time except for labor supply and the return on capi-
tal, which are roughly constant. This leads to a Cobb–Douglas production
function. These facts also imply the constancy of the capital-output ratio and
of the rental price of capital.
Two key growth facts are that the real wage and consumption grow at the
same secular rate as does real output per capita, while labor supply displays
no secular trend. This restricts the period utility function to be of the form
(4)
u(c,1–h) = .
5
Cass (1965) and Koopmans (1965) in deterministic situations establish the existence of an
optimal path and characterize properties of this optimal path. Diamond (1965) studies the com-
petitive equilibrium path in an economy with capital accumulation. In his economy, people live two
periods. Brock and Mirman (1972) deal with the problem of optimal growth when there are
stochastic shocks to the technology. These studies are in the nonquantitative theory tradition.
Danthine and Donaldson (1981) compute the equilibrium process for the Brock and Mirman
(1972) stochastic growth model.
6
Auerbach, Kotlikoff, and Skinner (1983) carry out a deterministic dynamic applied general
equilibrium analysis with endogenous labor supply in which they evaluate tax policies.
(c g (1 – l ))
1–
–1
1 –
K4_40319_Prescott_358-395 05-08-18 11.41 Sida 381
We set
= 1. This parameter was not calibrated to growth observations. We
made this choice based on a variety of evidence. The principal evidence used
is comparisons of the return on capital for fast- and slow-growing economies.
Fortunately, it turned out that our findings are not sensitive to this parameter,
because at the time of our work, this key economic parameter had not been
tightly tied down.
With
= 1, the above utility function is
(5)
log c + g(1– l ).
The nature of the
g function matters. The growth facts do not tie down the
elasticity of substitution between consumption and leisure, and this parameter
turned out to be key for deriving the predictions of the growth model for
business cycle fluctuations. Subsequently, this key parameter has been tied
down.
Step 4: Introduce a Markovian Shock Process
We wanted something in our model economy that led to labor supply errors
and something to propagate these errors. Here, by “labor supply errors” I mean
the difference between the optimal labor supply decision given individuals’
information set and the decision that they would make if they observed the
state of the economy without observation error. We introduced a total factor
productivity shock that is independent over time and assumes that agents see
the value of total factor productivity (TFP) with noise prior to making their
labor supply decisions. We also introduced a second highly persistent autore-
gressive TFP shock. We introduced this shock because it is simple to do and
we were curious to see what its consequences are. In order to use the Kalman
filter, the two shocks and measurement errors are all normally distributed.
Step 5: Make a Linear Quadratic Approximation
The next step is to determine the steady state of the economy when the
variances of the TFP shocks are zero. Then a linear quadratic economy is
constructed, which has the same first two derivatives at the steady state. This
linear quadratic economy displays the growth facts, and its equilibrium is easily
computed. The behavior of this economy will be arbitrarily close to the
economy we began with for sufficiently small variances of the two TFP shocks
and the measurement errors. It turned out that it was extremely close even
for variances far bigger than the ones we introduced.
7
Step 6: Compute the Competitive Equilibrium Process
The next step is to compute the recursive competitive equilibrium stochastic
process.
382
7
Danthine and Donaldson (1981), who computed the exact equilibrium for the stochastic model
using computationally intensive techniques, found this to be the case.
K4_40319_Prescott_358-395 05-08-18 11.41 Sida 382