What Drives Crash Prediction? Machine Learning Ablation Evidence from Market Microstructure

Bo Liu and Ke Xu
Additional contact information
Bo Liu: Huron University College

No 2604, Department Discussion Papers from Department of Economics, University of Victoria

Abstract: We investigate what makes mini flash crashes predictable using a feature ablation framework that decomposes the necessity and sufficiency of 253 market microstructure features. Across more than 3,600 out-of-sample experiments on SP 100 stocks, we find that crash prediction is driven primarily by liquidity and information asymmetry and relies on minutes, rather than milliseconds, of prior information. Conventional feature importance measures overstate the role of microstructure noise because its predictive value arises only through interactions with other features. Cross-sectionally, VPIN is the only significant determinant of stock-level crash predictability.

Keywords: Mini flash crash; market microstructure; machine learning; feature ablation; VPIN; high-frequency trading; market fragility (search for similar items in EconPapers)
Pages: 64 pages
Date: 2026-10
Note: ISSN 1914-2838 JEL Classification:G12, G14, G17, C53, C55
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.uvic.ca/socialsciences/economics/_assets/docs/discussion/ddp2604.pdf (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:vic:vicddp:2604

Access Statistics for this paper

More papers in Department Discussion Papers from Department of Economics, University of Victoria PO Box 1700, STN CSC, Victoria, BC, Canada, V8W 2Y2. Contact information at EDIRC.
Bibliographic data for series maintained by Administrative Assistant ().