High-Frequency Runs and Flash-Crash Predictability

Abstract

This article describes research into the short-term nature of movements in price data. The study’s key finding is that asset returns do not evolve at the Gaussian increments commonly assumed by continuous pricing models. Instead, prices exhibit strong autocorrelation, often resulting in predictable one-directional sequences, or runs. These runs are more pronounced ahead of market crashes. Identifying these runs can help predict impending flash crashes as much as a day before a crash. The research further contributes to asset pricing and derivatives literature by deriving discreet and continuous closed-form expressions for the probability of flash crashes.