Variable selection of yearly high dimension stock market price using ordered homogenous pursuit lasso

Abstract

It is noting that the response variable and the explanatory variables are highly correlated in high dimension data. Hence, the selection of informative variables is important in order to achieve a better model interpretation and concomitantly improve the accuracy of the prediction. In this study, the variable selection in stock market price using statistical approach was carried out. It is pertinent since most of the previous study only concerns on the financial interests of the stock market. Therefore, this study considers the homogeneity structure in the highly correlated data on yearly stock market price by applying ordered homogenous pursuit lasso (OHPL) method. The performance results of OHPL were compared with lasso and elastic net. As a result, OHPL a had higher number of selected variables and a better prediction power than of lasso and elastic net. In conclusion, OHPL shows its capability to enhance variable selection while increasing the prediction power of the selected variables than its counterpart.