Forecasting Emerging Stock Market Crashes via Machine Learning

Abstract

Stock markets indicate the overall health of an economy as they play a vital role in providing a way for companies to raise capital, create new opportunities and stimulate economic growth. However, stock markets are prone to crashes and the aftermath of such an event can cause far-reaching and long-lasting effects on the economy depending on the severity which induces a need to study stock market crashes. This work explores the idea of crashes in emerging stock markets leveraging a diverse array of machine learning models, while utilizing a comprehensive dataset comprising stock market data from 32 emerging market countries, with features derived from market data, along with several engineered liquidity features. A variation of the Artificial Neural Network model is identified as the top performer displaying high accuracy, about 96.66%, with high true positive rate and low false positive rate, outperforming existing models in the literature. In industry-specific analysis, the model consistently achieved strong true positive and false positive rates, indicating acceptable outcomes for the specific industries under consideration. Furthermore, it is found, using the SHapley Additive exPlanations framework, that return along with the attributes reflecting lag, mean, and standard deviation of liquidity indicators over the past week and month significantly contribute to the prediction of crashes suggesting that stock market crashes are typically gradual processes rather than abrupt occurrences. These findings hold profound implications for risk management and investment decision-making in emerging markets, offering valuable insights for both academia and industry practitioners.