A Short Term Co-Movement Detection of Financial Data

This paper aims to apply statistical and the tree-based data mining techniques to build a model for predicting the movement of Thailand stock price index (SET). Predictors are stock price indexes of Hong Kong Hang Seng (HSI) and Nikkei 225. We also incorporate the index difference from the previous day closing price of HSI and Nikkei as additional predictors. The positive or negative movement of SET compared to the earlier closing index is a binary signal that is constructed and used as a target for our model. The co-movement among SET, HSI, and Nikkei is a short term detection in that it captures intra-day association. The original built model is manipulated to be concise and comprehensible through the application of feature subset selection techniques. We finally obtain a concise tree model to forecast index movement with accuracy as high as 70%.