THREE ESSAYS IN DYNAMIC ECONOMETRICS

This thesis is a collection of three essays in dynamic econometrics. These essays cover a wide range of topics from model selection in distributed lag models to the problems in large sample estimation in some dynamic models of discrete choice, such as those studied recently by Heckman. The common theme that runs through these essays is dynamics in economic behavior. In all of the models analysed in the following chapters, intertemporal considerations play a central role.;In Chapter 2, we develop an asymptotically optimal procedure for determining the exact degree of the denominator lag polynomial in the general rational distributed lag model, defining carefully the notion of asymptotic optimality of our tests. We formulate the problem as one of testing composite hypotheses in a sequence on the lag coefficient until one of the hypotheses is rejected or all of the hypotheses are accepted. In testing these hypotheses, we consider a sequence of local alternatives rather than a fixed alternative.;Chapter 3 develops large sample properties of maximum likelihood estimators in the following classes of dynamic models of discrete choice: (1) Habit Persistence models, (2) State Dependence models, (3) Serial Correlation models. Inconsistency of joint maximum likelihood estimators of structural parameters in the presence of incidental parameters (group specific effects) is a serious problem in the fixed effects formulations of these models. We devote special attention to the following two problems: (1) the problem of initial conditions that arises in estimating a discrete time-discrete data stochastic process, (2) the problem of incidental parameters that besets an otherwise attractive solution to the problem of initial conditions.;Chapter 1 deals with the problem of choosing between non-nested models--serial correlation models and partial adjustment models. We apply the tests developed by Cox (1961) for separate families of hypotheses to these models. We derive the exact test statistics under both hypotheses and present some Monte Carlo evidence on the behavior of the power function of these tests.