Stock Shocks Modelling and Forecasting

Stock shocks are abrupt changes in a stock price time series. Major shocks, such as the 1929 Big Crash and, more recently, the subprime mortgage crisis, have greatly influenced the world economy in the last centuries. Nevertheless, shocks of minor intensity are frequent events that often go unnoticed but whose prediction can significantly help an investor to protect their investments. In this paper, we provide a formal definition of stock shocks and use limit order-book data to implement an algorithm to forecast stock shocks. The proposed algorithm is built upon a formal mathematical model approximating stock prices and return distributions with fat-tailed Lévy-stable models. A preliminary study of different machine-learning approaches allowed us to design an algorithm based on Random Forest, hierarchical clustering, and Bayesian optimization that outperforms other machine-learning methods that exhibit tunable and higher Precision and Recall values.