Forecasting movement direction of ISE 100 index using support vector machines method

This study aims to forecast the movement direction of Istanbul Stock Exchange National 100 Index (ISE 100) using Support Vector Machines (SVM), one of the classification methods in data mining. Open source programming language "R" was used to analyse statistical data, and SVMs' classification performance was compared with Logistic Regression (LR), the other method used in this study, in order to forecast the movement direction of ISE 100 Index. Technical indicators that are among the devices useful for technical analysis in stock prediction were used. These indicators included in models were analysed with LR analysis and then, significant ones were used as independent variables. The analysis includes the data from 03.04.1995 to 19.03.2012. 4226 data were established as daily, weekly and monthly data sets. 4 models were organized for each dataset and index movement direction forecasting performance of these methods was evaluated by applying different criteria for each model.The results of this study show that SVMs estimate the movement of ISE 100 Index best with Weekly Model-1 (70.0%) among 12 models. Additionally, it is observed that this model has a high level of estimation and the closest increase (82.89%) and decrease (54.68%) direction of ISE 100 Index.In consideration of the results obtained with this study, it is obvious that in estimating the movement direction of ISE 100 Index, Weekly Model-1 will be useful for investors and academic studies.Key Words: Support Vector Machines, Index Movement Direction Forecast, Technical Indicators, Logistic Regression, Classification, ISE 100.