| Summary: | 碩士 === 國立臺北大學 === 經濟學系 === 95 === The main purpose of my thesis is to investigate the finite sample properties of nonstationary truncated cointegration regression. The issues we want to discuss are classified into two parts. First, we are interested in the cointegrated parameters under different estimation methods. We figure out the fitting estimation method by means of calculating mean square error. Second, whether there exists cointegrated relationship between truncated tobit model. We simulate the critical value of the cointegration statistics with Monte-Carlo methods and compare our simulation outcomes to that of censored tobit cointegration model.
According to simulations, we observe that the mean squares of ordinary least square estimator and maximum likelihood estimator are similar, thus the estimator under two methods have almost the same estimation biases. Estimation biases will converge to zero when T for two estimators. As a result, both of two methods can apply to estimate the truncated tobit cointegration model. Besides, we consider the cointegration relationship between truncated tobit model. There are two cases：the cointegrating regression without drift and with drift. Being less sensitive to the nonlinear transformations, we choose Shin (1994) test to examine the cointegration relationship of the truncated tobit model. Under the same significant level, the critical values of truncated tobit cointegration model are lower than that of censored tobit cointegration model. We attribute the phenomenon to that the likelihood function of truncated tobit cointegration model are more simplified. Consequently, the estimated sample residuals are smaller, and the critical values of Shin test of truncated tobit model are apparently lower.
|
|---|
