Does Machine Learning Algorithms Improve Forecasting Accuracy? Predicting Stock Market Index Using Ensemble Model

Abstract

Forecasting market performance and understanding the mechanism of price discovery is inherent to develop trading strategies. This paper examines the predictive power of machine learning algorithms in forecasting stock index. The study applies various machine learning algorithms and suggests the best model for forecasting stock index. We have developed an ensemble model to predict the daily closing index of Nifty 50 based on open, high, low and previous day’s close. The ensemble model includes a mix of simple linear regression, gradient boosted tree, decision tree and random forest. The parameters of the models are tested for its accuracy train-test split under supervised learning. The predicting accuracy of machine learning algorithms is further refined using cross-validation techniques that include leaving one out cross-validation and k-fold cross-validation. We found that the ensemble model provides an accurate forecast of the stock market index for the short term. The outcome of the study would facilitate the investors and portfolio managers to use the appropriate model for forecasting and take an informed decision by considering the nature of stock market volatility.