Generating Buy/Sell Signals for an Equity Share Using Machine Learning

Abstract

This study proposes a novel model for predicting 5 days’ ahead share price direction of GARAN (Garanti Bankasi A.Ş.), an equity share that is the top traded stock in BIST100, Istanbul Stock Exchange -Turkey. The first model includes global macroeconomic indicators as well as local inputs whereas the second model is focused more on local inputs. The performances of the two models are tested using Support Vector Machines (SVM), Neural Network with Back-Propagation (BPN), and Decision Tree (DT) algorithms. Though BPN and SVM have previously been used to predict BIST100 Index movement, DT has not been utilized before with this purpose. Forecasting is carried out tested for a time span of about 6 months on a rolling horizon basis, that is, algorithms are re-run weekly with updated data to generate daily buy/sell signals for the next week. A simple trading strategy is implemented based on buy/sell signals to calculate the rate of return on investment during the testing period. The results illustrate that DT having 80% prediction accuracy outperforms BPN and SVM that achieve 60% accuracy. Consequently, DT achieves a higher rate of return.

Keywords

• Machine learning algorithms
• backward-propagation networks
• support vector machines
• decision tree learning
• stock movement forecasting