Prediction based mean-value-at-risk portfolio optimization using machine learning regression algorithms for multi-national stock markets

The future performance of stock markets is the most crucial factor in portfolio creation. As machine learning technique is advancing, new possibilities have opened up for incorporating prediction concepts into portfolio selection. A hybrid approach that constitutes machine learning algorithms for stock return prediction and a mean–VaR (value-at-risk) model for portfolio selection is illustrated in this paper as a unique portfolio construction technique. Machine learning regression models such as Random Forest, Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Support Vector Machine Regression (SVR), k-Nearest Neighbors (KNN), and Artificial Neural Network (ANN) are adopted to forecast stock values for the next period. The stocks with greater prospective returns are chosen in the first stage of this affection. Further, the mean–VaR portfolio optimization model is employed for portfolio selection in the second stage. The monthly datasets of the Bombay Stock Exchange (BSE), India, Tokyo Stock Exchange, Japan, and Shanghai Stock Exchange, China, are used as the research sample, and the findings show that the mean–VaR model with AdaBoost prediction outperforms other models.