Job Market Paper

Nonparametric identification for fixed-T panel data models has received a great deal of attention recently. My job market paper contributes to this literature. Starting from a general data-generating process that allows for both individual and time heterogeneity, I give conditions for the identification of average partial effects of a discrete regressor. These conditions are then used to characterize the trade-off between restrictions on individual and time heterogeneity as well as assumptions on the structural function. The conditions also suggest a class of models for identification, which include existing models in the literature, such as Chernozhukov, Fernandez-Val, Hahn and Newey (2010), as a special case. Fortunately, these models have clear testable implications on the conditional distribution of the outcome variable. I hence propose bootstrap-adjusted Kolmogorov-Smirnov and Cramer-von-Mises statistics to test these implications. The tests proposed here include a nonparametric test for the fixed effects assumption in the presence of generalized time effects. Finally, I implement the test of the fixed-effects assumption on the returns to schooling equation using a subsample of the national longitudinal survey of youth (NLSY) previously used by Angrist and Newey (1991).

Working Papers

Presented at the African Econometric Society, 2010, and Joint Statistical Meetings, 2010

In parametric nonlinear models, the nonseparability of unobservable heterogeneity and regressors leads to the well-known incidental parameters problem. The resulting maximum likelihood estimator suffers from an asymptotic bias. Hahn and Newey (2004), among others, propose bias-correction methods for this class of models. Due to higher-order effects of the uncertainty from bias estimation, the bias-corrected estimators may not perform well in finite samples. This paper proposes the use of shrinkage to reduce the impact of bias estimation on the performance of the resulting estimator and shows that it leads to a higher-order reduction in mean square error.

Presented at the American Environmental and Resource Economics Meetings, 2012 (by Junjie Zhang)

In this paper, we test for the presence of manipulation of air pollution data by local governments in China. Our expectation, that there should be manipulation, stems from the presence of a particular cutoff for good-weather days. In addition, the number of good-weather days is used as part of the performance evaluation of local government officials. We use two different strategies to test our claim. The first uses an existing test for regression discontinuity design. The second is a variant of the Kolmogorov-Smirnov tests proposed in Ghanem (2012). Both methods support our claims and provide interesting insights on issues relating to incentive-based pollution monitoring.