Predicting Stock Market Indices Using Classification Tools

Increasing interest has been shown in the use of classifiers to extract informative patterns from time series data generated by monitoring financial phenomena. This paper investigates data mining and pattern recognition methods in forecasting the movement of the Standard & Poor’s 500 index. We use functional forms of varying classifiers to predict financial time series data and to evaluate the performance of different classifiers. By using the time series ARIMA model, we forecast the Standard & Poor’s 500 index. Additionally, with the AdaBoost algorithm and its extensions, we compare the classifying accuracy rates of bagging and boosting models with several classifiers, such as support vector machines, k-nearest neighbor, the probabilistic neural network, and the classification and regression tree. Results indicate that the boosting classifier with real AdaBoost (exponential loss) best forecast the Standard & Poor’s 500 index movements. This result should be relevant to firms that want to predict the stock prices.