Prediction the Direction of SET50 Index Using Support Vector Machines

Chongkolnee Rungruang, Wilawan Srichaikul, Somsak Chanaim, Songsak Sriboonchitta

Abstract

Support vector machine (SVM) is a very specific type of learning algorithms characterized by the capacity control of the decision function, the use of the kernel functions and the sparsity of the solution. In this paper, we investigate the predictability of stock index movement direction with SVM by forecasting the daily movement direction of SET 50 index over the period 5 April, 2000 to 22 August, 2018.  The experiment results show that SVM with autoregressive lag $p=10$ and training data equal 37 have accuracy(ACC) $92.56\%$.