A novel framework for stock trading signals forecasting

Abstract

This study investigates stock trading points prediction that is an attractive yet challenging research topic in the financial investment area, as the stock market is an unstable and complex system. A small improvement in the predictive performance can make profit. To realize trading points detection, we propose a novel method which integrates piecewise linear representation (PLR) and feature weighted support vector machine (FW-WSVM) to forecast the stock trading points (PLR–FW-WSVM). Firstly, we generate numerous trading points (valley or peak) from the trading data by PLR and formulate the stock trading points prediction as a weighted four-class classification problem. Then, we estimate the importance of each input feature by computing the information gain and apply FW-WSVM to learn the prediction model between the trading points and the input features from the historical data. Afterward, the model is used to forecast the future turning points from the input features. Lastly, we conduct a series of experiments among PLR–FW-WSVM, PLR–WSVM and PLR–ANN over 30 stocks with different investment strategies. The results show that our proposed method generates the highest accuracy and profits in average, which indicates PLR–FW-WSVM is effective and can be applied to forecast the future trading points in the real-world application.