383
Step 7: Simulate the Model Economy
The equilibrium stochastic process is used to generate a time series realiza-
tion of the model economy. If the number of observations in the period
being considered is N, a time series of length significantly greater than N is
generated and the last N observations considered. A longer time series is
generated because we wanted a draw from the invariant distribution for the
state of the economy as a starting point of the model’s sample path.
Step 8: Examine the Key Business Cycle Statistics and Draw Scientific Inference
The last step is to compare the key business cycle statistics for the model and
the actual economy. I emphasize that the identical statistics for the model
and the actual economy are compared.
One important statistic is the standard deviation of the cyclical component
of output. What we defined to be the cyclical component of output is first
computed for the actual economy and the standard deviation determined.
The identical procedure is followed for realizations of the equilibrium process
of the model economy. This means the model is simulated to generate the
time series of output and other series. Next the cyclical component of output
is computed and its standard deviation determined. This is done many times
so that the first two moments of the sampling distribution of the standard
deviation of the model’s cyclical output statistic can be determined.
If the sampling distribution of this statistic in question is concentrated
about some number, this number relative to the statistic specifies how variable
the economy would have been if TFP shocks were the only shocks. If the
sampling distribution of this statistic is not concentrated, theory does not
provide a precise accounting. But the sampling distribution is highly concen-
trated provided the number of quarterly observations is at least 100.
I emphasize that this is not a test in the sense of Neyman–Pearson statistical
hypothesis tests, which are useful in the search for a model or law through
induction. The way theory is tested is through successful use. The neoclassical
growth model is tested theory.
4. USING THE METHODOLOGY IN BUSINESS CYCLE RESEARCH
Kydland and Prescott (1982) found, as reported in our paper “Time to Build
and Aggregate Fluctuations,” that if elasticity of substitution of labor supply is
3 and TFP shocks are highly persistent and of the right magnitude, then
business cycles are what the neoclassical growth model predicts. This includes
the amplitude of fluctuation of output, the serial correlation properties of
cyclical output, the relative variability of consumption and investment, the
fact that capital stock peaks and bottoms out later than does output, the
cyclical behavior of leisure, and the cyclical output accounting facts.
Subsequently, Prescott (1986) found that the shocks were highly persistent
and the TFP shocks of the right magnitude. Conditional on a labor supply
elasticity close to 3, TFP shocks are the major contributor to fluctuations in
the period 1954–1981 in the United States.
K4_40319_Prescott_358-395 05-08-18 11.41 Sida 383
This finding turned out to be highly robust. Greenwood, Hercowitz, and
Huffman (1988) find that if, on average, TFP shocks are non-neutral with
regard to consumption and investment, the conclusions hold. Rotemberg
and Woodford (1995) introduce imperfect competition and show that the
finding is overthrown only if monopoly rents are far in excess of what they
could be. With imperfect competition restricted to be consistent with labor
cost share, Hornstein (1993) and Devereux, Head, and Lapham (1996) show
that the importance of TFP shocks for business cycle fluctuations hardly
changes. With the introduction of monopolistic competition, model-TFP
shock variance is picked so that the model Solow-TFP variance matches the
actual economy’s Solow-TFP variance. In these monopolistic competitive
worlds, Solow-TFP is a complex statistic and is not total factor productivity.
Investment in the model economy varies smoothly, as does aggregate
investment in the actual economy. Investment at the plant level, however, is
not smooth, and a natural question is whether this has consequences for
modeling business cycles. Fisher and Hornstein (2000) find that having
plants that make lumpy inventory investment in equilibrium does not change
the estimates of the contribution of TFP shocks to fluctuations. For invest-
ment in plant and equipment, Thomas (2002) develops an economy that
displays lumpy investment at the plant level. When calibrated to the growth
facts and establishment investment statistics, the findings for business cycles
using her abstraction are virtually the same as those using the neoclassical
growth model.
Ríos-Rull (1995) uses a carefully calibrated overlapping generations model
and finds that the estimated importance of TFP shocks for business cycle fluc-
tuations does not change. Within this framework, Ríos-Rull (1994) then shuts
down financial markets, so physical capital holdings is the only way to save.
This extreme version of market incompleteness does not affect the estimate
of the importance of TFP shocks. Introducing uninsurable idiosyncratic risk
(see Krusell and Smith, 1998) does not affect the estimate either. Hansen and
Prescott (2005) deal with capacity utilization constraints that are occasionally
binding. With their introduction, the nature of the predictions for business
cycles changes a little, but in a way that results in observations being in even
closer conformity with theory.
Using this methodology, Danthine and Donaldson (1981) and Gomme
and Greenwood (1995) investigate the consequences of various non-Walrasian
features for business cycle fluctuations. There are interesting implications for
relative variability of consumption for those with large capital ownership and
those with no capital ownership.
Freeman and Kydland (2000) and Cooley and Hansen (1995) find that
introducing money and a transaction technology does not alter the conclusion
as to the importance of TFP shocks. Chari, Kehoe, and McGrattan (2000)
find that nominal contracting does not either. They introduce staggered
nominal contracting into the basic business cycle model and find that in such
worlds, monetary shocks have effects that are persistent but too small to be an
important contributor to business cycle fluctuations. In summary, introducing
384
K4_40319_Prescott_358-395 05-08-18 11.41 Sida 384