Global Stock Price Forecasting during a Period of Market Stress using LightGBM

Abstract

Accurately predicting future stock prices is crucial for investors, particularly during market stress, enabling informed decisions to mitigate losses and reduce financial exposure. While machine learning techniques have shown promise in this field, most studies have focused on local models. Global forecasting models, which are trained on a variety of time series, have demonstrated encouraging outcomes in outperforming local models. This article presents a global forecasting approach utilizing LightGBM to predict stock prices in the Moroccan market during highly volatile period. The study includes a sector-wise analysis, with the most volatile sector being evaluated, and correlated stocks from other sectors were considered to enhance the data. Compared to other deep learning models that use a local approach, our findings demonstrate the effectiveness of utilizing a global forecasting with optimized LightGBM for predicting stock prices. Additionally, incorporating data from multiple stocks improves accuracy of stock price prediction.