The discovery of financial market behavior integrated data mining on ETF in Taiwan

Abstract

In practice, many physics principles have been employed to derive various models of financial engineering. However, few studies have been done on the feature selection of finance on time series data. The purpose of this paper is to determine if the behavior of market participant can be detected from historical price. For this purpose, the proposed algorithm utilizes back propagation neural network (BPNN) and works with new feature selection approach in data mining, which is used to generate more information of market behavior. This study is design for exchange-traded fund (ETF) to develop the day-trade strategy with high profit. The results show that BPNN hybridized with financial physical feature, as compared with the traditional approaches such as random walk, typically result in better performance.