The desire for warning signals for market turbulence, and “flash crashes” in particular, has risen greatly in recent years following several spectacular episodes where market failures have threatened to undermine the integrity of the financial system.
The only measure to claim broad empirical success at this task is the VPIN, or Volume-synchronized Probability of INformed trading, metric of Easley, Lopez de Prado and O’Hara (EL) proposed as a real-time indicator of order flow toxicity. In particular, they find it to outperform the VIX index in predicting future short run volatility. VPIN is patented, it has been covered in major media outlets, proposals for futures contracts trading on the metric have been forwarded, it is under consideration by regulators as a trigger for a circuit breaker.
Unfortunately, in a careful empirical study, we find VPIN to fail as predictor of future market volatility.
VPIN relies on measures of order flow imbalance. We use near ideal data to construct accurate trade classification, and document that the schemes exploited by ELO are inferior to a standard tick rule. We find that VPIN forecasts short-term volatility solely because it generates systematic classification errors that are correlated with trading volume and return volatility. When controlling for trading intensity and volatility, the VPIN metric has no incremental predictive power for future volatility. For example, VIX subsumes the information of VPIN in this regard. We conclude that warning signals should monitor the conditions of the limit order book directly rather than infer order flow imbalances from the more limited information in trade and volume patterns.