Tracking toxicity in fast and complex markets

Abstract

We propose a new method for tracking toxic order flow in high-frequency equity markets. This method consists of two steps. First, a toxicity indicator is calculated for a single asset. Then, the comovement of the indicators of all of the assets is quantified. These steps are based on the (proportional) order imbalances used in the volume probability of informed trading (VPIN), and on the thick pen measure of association (TPMA) methods, adapted from a discrete-time to a continuous-time setup, respectively. Through careful implementation of indicators such as order imbalance and volume-bucket-wise standard deviation, the new metric operates in real time, is void of look-ahead bias and does not rely on long-outdated historical data. When applied to individual Dow Jones Industrial Average stocks and several flash crashes, it tracks the flash crashes adequately. The new metric can aid market makers and market regulators in tracking toxicity based on intraday data, but it does not anticipate the flash crashes.