Specification and estimation of the censored regression model in the presence of stochastic parameter variatio

Censored regression analysis is used in regression models when the range of the dependent variable is constrained in some way. In its frequent appearance in the econometrics literature, the censored regression (or Tobit) model assumes that the coefficients of the regressors remain constant throughout the sample, not allowing for structural change.;The purpose of this dissertation is to extend the framework of the censored regression model by incorporating stochastic variation in its parameters. The fixed parameter assumption is relaxed by specifying a random structure for the parameter vector. The randomness of the parameter vector is specified alternatively with nonstationary and stationary multivariate normal probability distributions. As a result, two stochastic parameter censored regression models are derived. The nonstationary model utilizes the former distributional specification while the stationary model the latter. By using such a specification regarding the coefficients, the censored regression model can be applicable to situations where observations are generated by a process with changing structural parameters.;Estimates of the parameters and the variance-covariance matrix are obtained using maximum likelihood procedures. For this, the likelihood function is derived. Since the normal equations are nonlinear in the parameters, their solution is obtained by an iterative process. For completeness, the variance-covariance matrix of the model is derived analytically. Furthermore, the statistical properties of the maximum likelihood estimator are discussed along with several other theoretical and computational issues.;To evaluate the performance of the stochastic parameter approach to the censored regression analysis, an empirical application is provided. It concerns the multinational expansion of the U.S. companies in the nonbanking financial sector. The empirical study utilizes a multicountry cross-sectional data set for 1981. The evaluation applies nested and nonnested model selection criteria. The statistical testing provides support for the stochastic parameter approach.;Apparently, this validates the argument of this dissertation. It suggests that stochastic parameter specifications increase substantially the explanatory power of the censored regression model. This method of analysis is indeed required when the situation under consideration does not allow the constancy of the structural parameters.