A simulation study of structural estimators of dynamic simultaneous equations models with autoregressive errors.
Date
1990
Authors
Ogwang, Tomson.
Journal Title
Journal ISSN
Volume Title
Publisher
Dalhousie University
Abstract
Description
In this study we investigated the small sample properties of six two-step estimators of dynamic simultaneous equation models with autoregressive errors using the Monte Carlo approach. The six estimators were proposed by Hatanaka.
The study focused on the relative performances of the estimators in structural estimation and prediction.
All the six estimators exhibited significant biases. The rankings of the estimators depended on the magnitude of autocorrelation, the coefficient of the lagged endogenous variables and the sample size. Furthermore, the problem of choice among estimators was relatively more important for prediction than for structural estimation. The kernel estimates of the sampling distributions of the estimators were quite similar and were almost symmetric. Significant differences among the estimators emerged only if there were large differences between the autocorrelation coefficients of the equations. The full information estimators generally performed worse than their limited information counterparts when autocorrelation was high and the reverse was true at low levels of autocorrelation. Whereas the asymptotic covariance matrices of the structural parameters were unreliable for purposes of inference in small samples, very large samples were required for the asymptotic covariance matrices of dynamic simulation forecasts to make valid inferences concerning forecasts.
Thesis (Ph.D.)--Dalhousie University (Canada), 1990.
The study focused on the relative performances of the estimators in structural estimation and prediction.
All the six estimators exhibited significant biases. The rankings of the estimators depended on the magnitude of autocorrelation, the coefficient of the lagged endogenous variables and the sample size. Furthermore, the problem of choice among estimators was relatively more important for prediction than for structural estimation. The kernel estimates of the sampling distributions of the estimators were quite similar and were almost symmetric. Significant differences among the estimators emerged only if there were large differences between the autocorrelation coefficients of the equations. The full information estimators generally performed worse than their limited information counterparts when autocorrelation was high and the reverse was true at low levels of autocorrelation. Whereas the asymptotic covariance matrices of the structural parameters were unreliable for purposes of inference in small samples, very large samples were required for the asymptotic covariance matrices of dynamic simulation forecasts to make valid inferences concerning forecasts.
Thesis (Ph.D.)--Dalhousie University (Canada), 1990.
Keywords
Statistics.