ESTIMATION OF RANDOM COEFFICIENT DYNAMIC MODELS: TWO APPLICATIONS
Abstract
Error components models are extensively used in analyzing panel data. Properties of the estimators are investigated by several authors. Recently T. W. Anderson and Cheng Hsiao have proposed consistent estimators for parameters in error components models with lagged endogenous variables. However, in their paper they assume that the parameters are stationary over large individuals. In this research, a model with parameters varying randomly across units is proposed and consistent estimators are developed. Aitken's generalized least squares estimator and maximum likelihood estimator are shown to be consistent, first-order efficient, and asymptotically normally distributed. In order to examine the specification of fixed parameters is appropriate or not to a given situation, two asymptotic test procedures are developed: one for testing the hypothesis of equality of coefficients over units; the other for testing the hypothesis of randomness of coefficients. Finally, we present two empirical investigations. First, we apply our statistical procedures in estimating a firm's dividend policy equation using panel data on 40 firms in the public utility industry in the U.S.A. Second, we use time series data on each of 50 states of U.S. to estimate a dynamic residential electricity demand function with coefficients varying randomly across states.
Degree
Ph.D.
Subject Area
Economics
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.