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

Research output: Working paper

Abstract

This paper tests if the limit order book (LOB) and news stories from 27 July 2007 to 27 January 2022 can help forecast realised volatility (RV) for stocks. The out-of-sample forecasting results indicate that the CHAR model outperformed all other HAR-family of models. For high volatility days, the simple news count outperformed more sophisticated news sentiments compiled following Loughran and McDonald (2011), while LOB depth outperformed the LOB slope. This impact is more marked for news count than depth during the COVID pandemic. Interestingly, for normal volatility days, the market seems to be driven by buying pressure, as the ask-side of the LOB is more price informative than the bid-side. The reverse is true for high volatility days. A series of forecasting evaluation tests and alternative model specifications confirm our results' robustness.
Original languageEnglish
PublisherSocial Science Research Network
Number of pages36
DOIs
Publication statusPublished - 12 Sept 2020

Publication series

NameSSRN Electronic Journal
PublisherSocial Science Research Network
ISSN (Print)1556-5068

Keywords

  • Realised Volatility Forecasting
  • Heterogeneous AutoRegressive Models
  • Limit Order Book Data
  • News Stories
  • Sentiment Measures

Research Beacons, Institutes and Platforms

  • Institute for Data Science and AI

Fingerprint

Dive into the research topics of 'Alternative Data for Realised Volatility Forecasting: Limit Order Book and News Stories'. Together they form a unique fingerprint.

Cite this