In this section we consider two alternative values for δIn the baseline mode δ = 0.58, which is the fraction of high-school graduates who enrolled in college during the period 1962-2012. Alternatively, one could δ calibrate this parameter to match some other targets. For example, we could use scores from cognitive tests, such as the Armed Forces Qualification Test (AFQT). By using data from the National Longitudinal Survey of Youth 1979 (NLSY79), we reset ; to 0.42 to match the fraction of individuals with an AFQT score above the mean score in the year 1980, when the test was administered. IS Alternatively, using U.S. Census Bureau data, δ; could be set to 0.225 to match the fraction of Americans with a college degree calculated over the period 1962-2012. In both cases,δ; = 0.42 and δ; = 0.225, the remaining parameters of the model are recalibrated following the procedure described in Section 2.2, except for n .
The business cycle statistics are reported in the last two columns of Table 3. Fig. AS in Appendix shows the impulse response functions for education. The model's predictions are robust to δCompared to Modell, the volatility of hours worked is slightly higher under the alternative specifications. Since the volatility is lower for high types. it is not surprising that aggregate volatility increases as the fraction of high types decreases. However. the change is very small. Nevertheless. our preferred specification remains the version with δ = 0.58. Since we observe differences in the time spent studying between who enroll in college and who does not. Setting δ to match the fraction of high-school graduates enrolling in college seems a more sensible and less arbitrary choice.