A dynamic threshold decision system for stock trading signal detection

Trading signal detection has become a very popular research topic in financial investment area. This paper develops a model using the Piecewise Linear Representations (PLR) and Artificial Neural Networks (ANNs) to analyze the nonlinear relationships between the stock closed price and various technical indexes, and uncovering the knowledge of trading signals hidden in historical data. Piecewise Linear Representation tools are applied to find the best stock turning points (trading signals) based on the historical data. These turning points represent short-term trading signals for selling or buying stocks from the market. This study further applies an Artificial Neural Network model to learn the connection weights from these historical turning points, and afterwards an exponential smoothing based dynamic threshold model is used to forecast the future trading signals. The stock trading signal is predicted using the neural network on a daily basis. The dynamic threshold bounds generated provide a guide for triggering a buy or sell decision when the ANN-predicted trading signal goes above or under the threshold bounds. Through a series of experiments, this research shows superior results than our previous research (Chang et al., 2009 [1]) and other benchmark researches.