Predicting market crashes using systemic risk and volatility spillovers : a deep learning approach

Abstract

We develop a model for predicting market crashes in the Stock Exchange of Thailand using a deep learning-based anomaly detection approach (LSTM-VAE). The model aims to detect market behavior before each market crash. Apart from the common stock market variables, we feed the model with the indices of systemic risk, and of volatility spillovers. With these two indices, the model takes into account the influences from both inside and outside the particular stock market. We find that in large crashes our model gives the crash warning signals shortly after the SET index reaches its peaks and long before the index reaches its troughs. And our model outperforms the existing models in the literature. Besides, when compared with a buy-and-hold strategy, our strategy incorporated signal from the model also leads to a higher return, because it helps us evade from large crashes.