An approach to portfolio optimization with time series forecasting algorithms and machine learning techniques

Abstract

The challenge of identifying suitable stocks for portfolio inclusion, particularly in the context of complex stock forecasting dynamics characterized by nonlinear time series and various influencing factors, is addressed in this study. To tackle this challenge, an approach combining the auto-regressive integrated moving average (ARIMA) and least-square support vector machine (LS-SVM) models is proposed for stock selection. Furthermore, the mean–variance portfolio optimization model is utilized for optimal portfolio selection. The effectiveness of this approach is demonstrated through comprehensive comparisons with alternative machine learning models, including support vector machines (SVM), LS-SVM, ARIMA, combined ARIMA+SVM models, and several benchmarking models from the existing literature. Validation of the proposed technique is conducted using historical data from the Bombay Stock Exchange (BSE), India.