This section briefly describes our calibration procedure, but for further detail we refer the
reader to AGH (2011). There are a total of 25 parameters to be calibrated. This is less
than the 33 parameters in AGH (2011) because 7 of those pertained to the behavior and
regulation of banks, which are absent from the present setup.11 These parameters can be
categorized as shop parameters, personal parameters, and government policy parameters.
They are listed in Table 1 along with their assigned values, their six-letter codes used in
Figure 3, and their levels of calibration as explained in the rest of this section.
Our calibration took place at three different levels. At the first level, one subset of
parameter values was chosen to match empirical counterparts in the U.S. data and/or to
match values used in previous studies. The 14 parameters with a "1" in the fifth column of
Table 1 were calibrated this way, and they take on the same values as in AGH (2011) but
with one exception. The exception is the markup, which was uniformly distributed across
shops in the earlier paper where the mean of this distribution was calibrated at the second
level. Here, in our baseline calibration, we keep the markup fixed across shops at 13 percent,
which falls within the 10 to 20 percent range of markup estimates cited by Golosov and
Lucas (2007), but we relax this assumption in Section 6 below.
At the second level, the values of other parameters were chosen to be internally consistent
with the central tendency of simulation outcomes. The two parameters with a "2" in the
fifth column of Table 1 were calibrated this way, namely the initial values of the government
targets for log potential output and the real interest rate. We used an iterative procedure to
select each value so that it would equal the median outcome of the targeted variable across
simulations.
The remaining 9 parameters, indicated with a "3" in the fifth column of Table 1,
were calibrated at the third level. The values of these parameters, for which we could
find no convenient empirical counterparts, were chosen so as to make median outcomes
across simulations (loosely) match certain properties of the U.S. data. More specifically,
we ran 10,000 simulations of 60 years. Each 60-year simulation began near the no-shock
equilibrium and continued for 20 years before we started calculating the average value of
each outcome variable across the remaining 40 years of that simulation. This was done to
eliminate transient efiects due to initial conditions. For each variable, we then computed the