The average monthly transitions from each state of the labor market in the US are well documented in the Current Population Survey (CPS) conducted by the Bureau of Labor Statistics. It is convenient to use for its consistency, high frequency and large sample size and is summarized in table 2.2.
Fitting the Model to Data
To map flows from the model to the data, I choose an arrival rate for low and high search cost workers to employment to match the U E and N E flows. I then choose Ỏ to match the total flow probability out of employment, so that ỏ = EU +EN. To choose parameters for the transition matrix X, I use the flow rates of workers between nonparticipation and unemployment. Since flows between U and N in steady
This transition matrix yields a steady state population of types where the low search cost types who choose unemployment are about one-twelvth the population of inactive types. This is slightly higher, but similar to the data in the CPS over this time period, where average unemployment is about 6% of the Labor Force, or roughly 3.8% of the population in my sample, while the percentage of the population not in the labor force is on average 35.5%.
Recall that to match the flows out of employment, there is only one parameter, 8. The exogenous separation probability dictates the minimum rate of flows from employment to non-employment. It is the ratio of types in employment that controls the relative difference between the EU and EN flows. I fail to get variation in the relative magnitude of the EU and EN flows – they are always different by a magnitude of about nine, reflecting the relative sizes of the stocks of each type of agent in employment based on the stationary distribution generated by X. It is this tension between the relative sizes of the EU and EN flows versus the stationary distribution of types from X that makes matching flows in a 3 state model so difficult. The results of the initial calibration in Table 2.4 show that the calibrated model does not produce enough separations of employed workers to unemployment, and produces too many separations of workers to nonparticipation. This discrepancy in flow rates causes stocks to be off by a large factor as well. The fraction of the population in nonparticipation that is generated by the model is too high by about 8 percentage points.The result that the N E and U E flows vary by a factor of about nine in my calibration comes directly from the fact that the employed population is converging to the same stationary distribution of 1 low cost type to 12 high cost types. The separation rate to unemployment and nonparticipation is dependent on only one parameter, Ỏ. In section 2.4, I show that a type-independent separation rate is consistent with data on outflows from employment. If the data is consistent with the same separation rate for both types of agents, it is the evolution of types in the employment pool that must be counterfactual. I now propose that I allow the persistence of agents’ types to vary depending on whether they are employed or not employed, and estimate this transition process along with the separation rate for employed workers using the 1996 panel of the Survey of Income and Program Participation.