Time Varying Variable Selection of Time Series Regression Models with Illustration of Taiwan Stock Assets.

• Main Authors: Lin Cheng Ting, 林政廷
• Other Authors: Huang Yu Min
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/854yy2

碩士 === 東海大學 === 統計學系 === 101 === In financial market, one crucial issue for managing a portfolio is to attain particular returns as expected while the risk can be reduced. Markowitz (1959) laid down the ground-breaking work on the mean-variance analysis. Such analysis can be transformed into a regression model in which a target reference is taken as the dependent variable and the assets returns or prices of the portfolio are taken as covariates. The intercept from such fitted regression model is the estimated expected return of the final selected portfolio. The selection of the assets for the portfolio is the same issue to select the covariates inside the regression model. In this study, we utilize a regression model for a target response and a set of asset price as covariates. We employ LASSO approach to select the variables. Further, we extend the work of variable selection to be done at different local values of covariates using kernel regression. The results of selection then can be explained by different situations of the assets through the time zone. Several assets from Taiwan stock market are studied as for illustrations.