Two essays in labor economics.

Abstract

The first essay uses the Continuous Wage and Benefit History (CWBH) data base to evaluate the effects of the 1982 change in tax treatment of unemployment insurance benefits on compensated duration of unemployment and post-unemployment earnings. The 1982 episode is a particularly useful natural experiment because the treatment group (those newly subject to benefit taxation) is the middle income category and the two control groups (those whose benefits already were taxed and those whose benefits still were not taxed) are the high and low income categories. The empirical results suggest that taxing unemployment benefits reduced the affected workers' mean compensated duration by more than a week, but did not have a statistically significant effect on their post-unemployment earnings. The second essay investigates the extent of finite sample bias in instrumental variables (IV) estimation of intertemporal labor supply models. The two stage least squares estimator (2SLS), which has been customarily used to fix endogeneity in estimating the intertemporal substitution elasticity, is biased towards the OLS estimator and this bias grows as instruments become weak. Principal findings are summarized as follows: (a) the intertemporal substitution elasticity is much larger when estimated by LIML (a median-unbiased estimator) than 2SLS, and (b) there is a strong tendency of convergence between 2SLS and LIML estimates for both low degrees of over-identification and large samples. The findings establish that finite sample bias is a major reason why conventional estimates of the elasticity of intertemporal substitution of labor supply are so small. However, empirical evidence also shows that complete convergence is not possible, because of both limitations in increasing sample size and inconsistency potentially arising from the dubious exogeneity of the instruments used in conventional specifications. The precise magnitude of the elasticity is not clear yet and awaits future research.