Distinguishing true and false buy/sell triggers from financial technical indicators

The objective of this study is to develop decision support system applying machine learning methods to distinguish True and False Buy/Sell recommendations. Various recommendation schemes, like 30/70 RSI (Relative Strength Index) scheme, are effectively used by many traders. However, the triggers produced by such recommendation schemes are found suspicious most of the time, and hence are non-actionable. In this study we develop a dynamic programming formulation to extract an optimal trade pattern from the price data sets. Such patterns are further augmented with several financial indicators to obtain binary classification model which is going to be consulted online. So, our system assists investors with removing uncertainties left from the primary recommender. We show that our dynamic programming formulation runs efficiently in linear time. Dynamic Programming is used to identify and label the better ones among the primary BUY and SELL recommendations. The approach is experimentally evaluated on BIST-100, DOW-30 stocks and currency pairs. The technical indicators used as predictor features are RSI, Trend Normalized RSI, Percentage Price Oscillator, Bollinger Band Percentage, Stochastic Oscillator, Rate of change (ROC), and Commodity Channel Index (CCI). We use Support Vector Machines as the binary classification algorithm. In addition to using the indicators formed at the end of each day as an attribute, better results were obtained by using the technical indicators of two consecutive days as attributes. Accuracy over %70 was achieved with data from a single day and over %73 with data from two consecutive days in this very difficult binary classification task.