A Machine Learning-Based Prediction of Stock Price Crash Risk Using Micro and Macro-Level Determinants

Abstract

This research study investigates the impact of macro and micro-level indicators of the stock price crash risk. We used the daily stock prices of 15 top-performing stocks listed in the S&P 500 index. We choose these specific companies based on their trading volume. We used data ranging from Jan 2010 to Dec 2022 for all the companies. In the first step, we calculated the monthly series of stock price crash risk using the negative skewness approach. In a similar pattern, we use monthly data of macro indicators, which are exchange rate, interest rate, and economic policy uncertainty. In addition, we use trading volume and short selling as micro-level determinants of individual stock price crash risk (SPCR). We deploy four different models to forecast the stock price crash risk and the impact of individual determinants on the SPCR. These models include linear regression, support vector regression, a single-layer perceptron model, and a multilayer perceptron model. The findings suggest that both the micro (firm) level and macro-level potential predictors are highly significant. The overall accuracy of machine learning models improved significantly when macro-level indicators were incorporated. Furthermore, machine learning models, especially SLP and MLP, outperform linear regression.