Neural Network and PLS Regression Models for Predicting Stock Prices

Stock Index prediction has gained importance in recent times due to its commercial application and the nature of the data explored. The noisy environment of the data attracts many researchers to explore it. Common statistical models for time series data require the normality assumptions. But for the highly volatile stock data the normality conditions are violated and hence nonparametric models are essential for the analysis. In this study Multiple Linear Regression, Partial Least Square Regression and computational Neural Network models are proposed to explore stock market tendency and predict the daily closing prices using thirteen variables. Prediction ability of the models are measured using standard error values. Sum of Square error and Relative error are used to find the best Multilayer Perceptron Model. The models are compared based on MAPE and RMSE values. The results revealed that Nonparametric Partial Least Square Regression Model is better in predicting the daily closing prices than the classical Multiple Linear Regression and computational Multilayer Perceptron.