단위근을 가진 예측회귀모형에서 U-통계량을 이용한 비모수 검정

Abstract

In order to efficiently estimate and test of the parameters in a predictive regression model with a unit root or near unit root, there are several ways for conducting nonparametric tests. In this thesis, we set the model that has a unit root or near unit root. Then, we employ the assumptions of the model based on other researches. And we test the null hypothesis that the function takes the value of zero using the kernel density function with a variety of bandwidth for estimating and testing the model. Especially, we show the result that the case of bandwidth that is based on the standard deviation of the differences of is more powerful than other band- width parameters through Monte Carlo experiment. After gains from using -statistics on unit root test in the predictive regression model, we estimate the function at 24 points by using real data from NYSE/AMEX, this is T-bill rate, earning price and dividend -price ratio. The result shows how the estimation can improve to estimate about predictability in earning price ratio and T-bill rate than other papers.