USING EX POST FORECASTING AND PANEL DATA TO EVALUATE LABOR SUPPLY ELASTICITIES (EX POST FORECASTING, INCOME TAX)

The tax structure can impact both individual and aggregate labor supply and may have far reaching effects on the economy as a whole. Without a clear understanding of the labor supply response (LSR), it is impossible to measure the economic impact of either the current tax structure or proposed changes to that structure. A consistent empirical estimate of LSR has not been generated. Empirical results are dependent on sample selection, choices of proxies for independent variables, measurement error, and choice of estimation techniques. Therefore, a basic assumption of this research is that LSR can not be fully specified and all empirical estimates must contain some unknown amount of bias.; A simultaneous specification search of sixteen alternative models of LSR is conducted to examine the effects of choice of proxy for the marginal wage rate, theory formulation, functional form and adjustments for selectivity bias on empirical estimates of LSR. The models are evaluated based on the size of the adjusted R{dollar}sp2{dollar}'s. Signs and significance levels of parameter estimates are compared to a priori expectations derived from theory. A "preferred" model is then selected.; Since it is explicitly recognized that the parameter estimates cannot be held to be the best unbiased linear estimate of the underlying LSR, alternative descriptive tests must be developed to address the issues of potential overfitting of the model to the data and experimenter induced bias. Ex post predictions of hours worked (for both a holdout sample from the estimation year and the regression and holdout samples using the following year of panel data) are calculated. Ex post prediction errors are generated by comparing the forecasted value for the dependent variable with the observed value. Root mean squared errors and inequality coefficients are calculated and used to evaluate the predictive ability of each of the models.; For both men and women, after-tax wage proxies generate highest adjusted R{dollar}sp2{dollar}'s. There is no descriptive evidence indicating that these results are due to an overfitting of the models to the data. Uncompensated wage coefficients for men and women are relatively close in size and carry negative signs.