Alternative Data for Realised Volatility Forecasting: Limit Order Book and News Stories

Abstract

This study explores the predictive power of the limit order book (LOB) and news in forecasting realised volatility. News count outperforms sentiment, and LOB depth proves more informative than slope. During the COVID-19 pandemic, news count had a greater impact than LOB depth, emphasising the importance of news in extreme conditions. Analysis of LOB data reveals that on normal volatility days, markets are driven by buying pressure, which reverses during high volatility periods. Furthermore, a consistent trade-off in forecasting accuracy between normal and high volatility days is observed. Robustness is confirmed through forecasting evaluation tests and alternative model specifications.

Keywords: Realised Volatility Forecasting, Heterogeneous Autoregressive Models, Limit Order Book, News Stories, Sentiment Analysis

JEL Classification: C22, C51, C53, C55, C58