DescriptionCurrently, there appears to be a tradeoff between the performance of a semiparametric estimator in finite and large samples. In Chapter 1, we argue that this tradeoff occurs because of the nature of the bias reduction methods that are often employed in implementing these estimators. Accordingly, we develop a bias control mechanism that eliminates this tradeoff so as to ensure that the estimator performs well in finite samples while retaining desirable large sample properties.
Semiparametric models are commonly estimated under a single index assumption. In estimating these models, the consistency of the estimator critically depends on this assumption being correct. Therefore, in Chapter 2, we develop a test of this assumption. We formulate such a test and derive its large sample distribution under the null hypothesis of a single index. To ensure that the test statistic has good size and power properties in finite samples, we formulate a test whose form adapts to the model under the alternative hypothesis. Monte Carlo results confirm that the adaptive feature significantly improves the performance of the test statistic in finite samples.
Studying healthcare decisions poses many empirical challenges. Healthcare utilization and expenditures depend on health insurance and other health related variables. As insurance is a choice variable for the individual, there are potential endogeneity issues. Expenditures are only observed when utilization occurs and hence there is a selection problem. Furthermore, the decision to utilize healthcare and the decision about the level of treatment are determined by different decision makers. In Chapter 3, we study a system of three simultaneous equations: insurance, utilization, and expenditures. To avoid making traditional parametric distributional assumptions, we propose a semiparametric approach based on the previous two chapters. Both parametric and semiparametric approaches are employed in an empirical study using the Medical Expenditure Panel Survey (MEPS) 2005 data. We find that insurance increases the likelihood of seeking healthcare by about 15% points (from about 80% to 95%). We also find that the parametric approach predicts insurance to increase the level of expenditures by 125%; while the semiparametric method predicts an increase of 51%, a number in accord with an important experimental study in the literature.