Supply bottlenecks and machine learning forecasting of international stock market volatility

Abstract

This study explores the information value of the daily Supply Bottlenecks Index (SBI) – derived from newspaper articles – to forecast daily return volatilities of seven major developed stock markets: China, France, Germany, Italy, Spain, the UK, and the US. Volatility is measured using the interquantile range, obtained through an asymmetric slope autoregressive quantile regression model applied to stock returns to estimate conditional quantiles. From this, we derive key distributional moments including skewness, kurtosis, and lower- and upper-tail risks, which are then incorporated into a linear forecasting framework alongside leverage effects. Using Lasso shrinkage techniques to address potential overfitting, we find that the model incorporating higher-order moments outperforms a benchmark model based solely on own- and cross-country volatilities. Notably, the predictive accuracy improves further when supply constraint indicators from all seven countries are included. These results hold important implications for investors as we later highlighted.