Resumen de Essays on duration and count data models
Pedro Henrique Costa Gomes de Sant'anna
This thesis is formed by three chapters related to duration and count data models. In the first chapter, "Testing for Uncorrelated Residuals in Dynamic Count Models with an Application to Corporate Bankruptcy", I propose new model checks for dynamic count models. Both portmanteau and omnibus-type tests for lack of residual autocorrelation are considered, and the resulting test statistics are asymptotically pivotal when innovations are uncorrelated, but possibly exhibiting higher order serial dependence. Moreover, the tests are able to detect local alternatives converging to the null at the parametric rate T ?1/2, with T the sample size. I examine the finite sample performance of the test statistics by means of a Monte Carlo experiment. Finally, using a dataset on U.S. corporate bankruptcies, I use the new goodness-of- t tests to check if different risk models are correctly specified. In the second chapter, "Nonparametric Tests for Conditional Treatment Effects with Duration Outcomes", I propose new nonparametric tests for treatment effects when the outcome of interest, typically a duration, is subjected to right censoring. The new tests are based on Kaplan-Meier integrals, and do not rely on distributional assumptions, shape restrictions, nor on restricting the potential treatment effect heterogeneity across different subpopulations. The proposed tests are consistent against fixed alternatives and can detect nonparametric alternatives converging to the null at the parametric n?1=2-rate, n being the sample size. The finite sample properties of the proposed tests are examined by means of a Monte Carlo study. I illustrate the use of the proposed policy evaluation tools by studying the effect of labor market programs on unemployment duration based on experimental and observational datasets. The third chapter, "A Simple GMM for Randomly Censored Data", is a joint work with Miguel A. Delgado. This paper proposes a simple yet powerful GMM setup to estimate parametric regression models when the outcome of interest is subjected to right censoring. The estimation procedure is based on Kaplan-Meier integrals, and is suitable for both linear and nonlinear models, with possible non-smooth moment conditions. We derive general conditions for consistency and asymptotic normality of the parameters of interest. Finally, a small scale simulation study demonstrate satisfactory finite sample properties.
