A Biclustering Technique for Mining Trading Rules in Stock Markets

Technical analysis including a large variety of indicators and patterns is widely used in financial forecasting and trading. However, it is difficult to select a combination of indicators that can well capture useful trading points in a specific market. In this paper, we propose a biclustering algorithm to find a subset of indicators with different periodic parameters which produce a similar profitability for a subset of trading points from the historical time series in the stock market. The discovered trading points are grouped into two categories: buy and sell signals. These trading points are applied to both of training and testing periods and the returns are compared with the conventional buy-and-hold trading strategies. We test this algorithm by using the Dow Jones Industry Average Index and Hang Seng Index. The results demonstrate that the trading strategies based on the discovered trading rules using the biclustering algorithm outperform the conventional buy-and-hold strategy.