The sensitivity of VPIN to the choice of trade classification algorithm

The VPIN metric (Easley et al. 2012b) aims to detect and predict the toxicity of order flow. This paper examines the sensitivity and robustness of VPIN to the choice of trade classification scheme, which is the major input used to compute VPIN. We compare deterministic trade-by-trade classification approaches with results computed using a newly proposed heuristic approach, bulk volume classification. We find substantial differences for all levels of aggregation: trade classification, order imbalance, VPIN and identifying “toxic periods”. We also find that the detection of toxic periods does not yield consistent results in more than 60% of cases. But regression analysis can identify volume and return volatility as parameters that contribute to higher levels of sensitivity.