With human capital. This result is consistent with findings in DeJong and Ingram (2001). Thus, the education sector acts as a buffer sector. It allows agents to compensate for the initial reduction in labor income by increasing the human capital stock. The next section builds on these results and analyzes for the first time the behavior of agents separately by ability and age.
3.1 differences by ability type and age
Fig. 4 shows the impulse response functions for education by age and productivity type. The top left graph shows the behavior of the average high and low type. The other graphs show the impulse response functions for three age groups: 18-24, 25-34, 35-64 years of age. Time spent studying responds more to the shock for low types than high types. This is due to the fact that, on average, high-productivity agents have already accumulated a large amount of human capital before the crisis. They are more efficient at work and they earn a higher labor income. Therefore, it is more expensive for them to reduce hours worked and forgo labor income in order to study and accumulate more human capital. Further, the marginal product of human capital is relatively low for high types. They benefit less by substituting physical capital with human capital. However, there is no significant difference between ability types when individuals are in the age group 35-64. This is due to the fact that the payoff to acquiring additional human capital for this age group is small for both typs.
3.2 Implications for hours worked volatility
Table 2 shows the average business cycle statistics computed from 500 simulations of four versions of the model, along with annual business cycle statistics from US data. The data about labor supply are from CPS, March Supplement (1962-2012). Hours worked are obtained using the answer to the question “how many hours did you actually work last week?”. Data for output, consumption and investment are from US Bureau of Economic Analysis (1962-2012). Output is measured by real GDP, consumption by personal consumption expenditures, and investment by gross private domestic investment. Both the actual and the simulated series are transformed by taking natural logarithems and de-trendse using the Hodrick-Prescott filter. Following Ravn and Uhlig(2002), the smoothing parameter is set to 6.25. Since ATUS is only available from 2003 until 2013, we are not able to produce reliable business cycle statistics for the time spent studying, E. However, the ability of the model to match the empirical evidence regarding the empirical evidence regarding the education sector is analyzed in section 5.
To better understand how human capital accumulation and heterogeneity in learning ability affect business cycle properties, and in particular time allocation, the version of the model outlined in section 2 will be compared to several simplified versions of the life-cycle RBC model. Model1 is the model presented in section 2. This is our main specification in which agents are heterogeneous in age and productivity in learning. In the second version of the model, Model2, agents are heterogeneous in age only. Thus, individuals of a given age are equally productive in learning. This is the specification that most resembles the model by Hansen (2009). However, while they look at learning-by-doing and on-the-job training, we focus on formal education. Model is a life-cycle RBC model with exogenous human capital. In this case, agents are heterogeneous by age and productivity at work. However, there is no education section. Therefore, differences in productivity at work (i.e. differences in the human capital stock) are exogenously given and determined by the calibrated human capital life-cycle profile from Fig.1(first panel). Model4 refers to a version of the model without human capital in which agents are equally productive over the life cycle.
In summary, Model1 and Model3 include both heterogeneity by age and productivity at work. However, in Model1 differences in productivity at work are endogenously determined; while in Model3 they are exogenously given. Further, Model2 and Model4 include heterogeneity by age with and without human capital accumulation, respectively.
All parameters in the alternative specifications have been re-calibrated following the procedure discussed in section 2.2, except for the relative risk aversion parameter. This parameter is set to match US output volatility in Model1. Since the purpose of this section is to analyze how different version of the model are able to match business cycle facts, the comparison is this section is to analyze how different versions of the model are able to match business cycle facts, the comparison is possible only if we consider the same value of relative risk aversion. Table A4 in Appendix reports the calibrated values for the alternative versions of the model.
As shown by Table 2, the volatility of hours worked is underestimated by all models. However, models with heterogeneity in productivity at school or at work (i.e. Model1 and Model3) can explain a higher percentage of the volatility empirically estimated. The performance of the model with one ability type (i.e. Model2) is very similar to the performance of the model by Hansen (2009) and the model without human capital (i.e. Model4). Introducing heterogeneity within ages generates differences among agents in terms of the cost of reducing hours worked. Reducing hours worked is cheaper for agents with a lower human capital stock because they give up a lower labor income. Therefore, when the shock hits the economy, these agents reduce hours worked more and their volatility increases. In fact, Model1 and Model3 are consistent with the data in predicting a higher volatility for low types compared to high types. This empirical regularity is also documented in Rios-Rull(1993). Since low types have accumulated a lower human capital stock in the steady state, it is less expensive for them to reduce hours worked and give up labor income. Their volatility of hours worked is higher compared to high types. At the aggregate level, hours worked volatility increases when we model this type of heterogeneity. Instead, with one ability type (i.e. Model2) the productivity profile is more similar to the profile of high types. Aggregate volatility is lower and remains close to that of high types from Model1. This result suggest that heterogeneity by age is not enough in order to increase the ability of the ability of the model to match hours worked volatility. It is important to include heterogeneity by productivity as well. The presence of low types increases aggregate volatility of hours worked.
Maliar and Maliar (2001) showed that hours worked become more volatile by incorporating heterogeneity in physical capital and skills into an otherwise standard RBC model. Our findings confirm their result in a finite-horizon setting. However, contrary to their results, our model is able to replicate the empirical fact that hours worked are increasing in skills even for low values of the intertemporal elasticity of substitution for consumption. Moreover, since skill differences are endogenous in our model, we are able to provide an explanation for why we observe this type of heterogeneity. Specifically, hight-skilled agents are those individuals who have high productivity in learning and spend more time studying. Finally, by introducing schooling in the model, we can also quantify the impact of education on hours worked volatility. We investigate this aspect in the next section.
3.3 The role of schooling
By comparing Model1 and Model3 we are able to answer another important question: can schooling increase volatility? Since schooling provides an alternative to work, introducing the education section in the model could increase hours worked volatility. This is certainly true in an RBC model with identical agents(e.g. Dejong ang Ingram.2001: Einarsson and Marquis.1998). With heterogeneity, instead, the benefits from education differ among agents. As a result, hours worked volatility may not increase at the aggregate level. Table 2 shows that, by introducing schooling in the model, the volatility increases for low types while it decreases for high types. Overall, Model1 produces a lower aggregate volatility compared to Model3. Since low types are more likely to use the education sector as alternative to work, their volatility increases. Instead, the volatility decreases for high types for two reasons. First, they are less likely to substitute work with schooling compared to low types. Second, high types have a stronger incentive to substitute work with leisure in Model3 when education is absent from the model. When schooling is not an option, high types borrow less capital since they do not need to finance education. Their savings are higher later in life. Thus, when the shock hits the economy, they reduce hours worked more to increase leisure time. In other words, leisure becomes a better alternative to work when education is absent. For these two reasons, aggregate hours worked volatility of high types is lower in Model1 compared to Model3.
These results suggest that hours worked volatility is mainly affected by differences in the human capital stock. Having education in the model increases volatility for certain groups, but it decreases the volatility for other groups. Overall, the impact is negative. This result is in contrast with previous findings on the impact of schooling on hours worked volatility within representative-agent RBC models. In these models, the introduction of education can increase the volatility of hours worked (see Einarsson and Marquis,1998, for example). Within a life-cycle framework this result no longer holds because some agents are less likely to substitute work with schooling (e.g. high types). Nevertheless, Model1 remains our preferred specification because it is able to explain why we observe