Research in Microeconometric Methods and Their Applications

Northwestern University, Evanston IL

Investigators

Abstract

The project proposes research on three related subjects in theoretical and applied microeconometrics. The first study deals with estimation of nonlinear dynamic panel data models. Panel data track individuals over time and therefore offer us an opportunity to study an individual's dynamic behavior while controlling for his/her (observed and unobserved) characteristics. Econometric techniques for linear panel data models are well developed, but, in general, our knowledge of nonlinear panel data models is still very limited. In most cases, the key assumption is that the explanatory variables are strictly exogenous, which essentially requires no dynamic feedback from the outcome variable in one period to the explanatory variables in the next period (e.g. it rules out the lagged dependent variable as an explanatory variable). This assumption is not plausible in many empirical applications. For example, in an earnings model, this assumption amounts to requiring that an individual's earnings today do not affect his/her earnings tomorrow. One theme of the proposed research is to relax this assumption for a particular nonlinear model, namely the censored regression model, which is the appropriate model in the presence of data top-coding. The proposed method is applied to the matched data from the Current Population Survey and Social Security Administration (CPS-SSA) Earnings Record to study the racial difference in earnings stability. In particular, I empirically investigate whether the relative earnings stability between black and white had changed before and after the Civil Rights Act of 1964. Although a step forward, the first study considers only one special form of endogeneity, namely, the presence of the lagged dependent variable. As a result, the proposed method heavily relies on this special structure. The goal of the second element of the proposed research is to extend the basic theoretical idea to construct alternative estimators for models with more general forms of endogeneity (e.g. predetermined regressors in the panel data model and endogenous regressors in the cross-sectional model). The proposed method is relevant in many empirical applications. For example, if we are interested in estimating a life-cycle female labor supply model using panel data, then the fertility variables, such as the number of children, are predetermined instead of strict exogenous since a woman's labor supply decision for today can affect her fertility decisions for tomorrow. The third part of the project proposes a method for estimating duration models using repeated cross-section data. More specifically, we estimate the conditional probability of an individual leaving unemployment at a particular point in time using repeated cross-section data on uncompleted unemployment spells. The proposed method improves upon the existing estimators by relaxing a stationarity assumption (i.e. we allow the compositions of the inflow of individuals into unemployment to vary over time). We compare estimates between using single and repeated cross-section data (on uncompleted unemployment spells) as well as those using panel data to gauge the performance of our estimator. We also extend the theoretical results to the competing risk models that allow for multi-state exit at the end of an unemployment spell. Finally we apply the proposed method to the Spanish Labor Force Survey data to investigate, among other things, the effects of the introduction of the temporary contracts in the mid 1980s on women's labor market experiences. In sum, the proposed research is mainly on the econometric methodology, but the problems it tackles are relevant to many empirical applications. Therefore the results from the proposed project will be important for our understanding of many issues in economics and other social sciences.