How to measure slippage - look for negative surprises

 

The paper by Ilija I. Zovko, "Realtime price impact detection"  provides an interesting solution to the slippage monitoring problem. 

Traditional methods for managing market impact typically rely on monitoring post-fill price slippage. This approach fails in real-time environments for two key reasons:

Statistically Slow - Estimating price slippage requires hundreds of fills before the signal can be statistically distinguished from background market volatility. By the time a trader detects impact through slippage, the order execution is usually finished.

Causally Ambiguous - Slippage measures correlation, not causality. An adverse price movement could be caused by the trader’s own actions (information leakage) or by an unrelated participant/market alpha. The required response is opposite in each case: slow down trading if caused by self-impact, or speed up trading if competing against outside alpha.

Instead of tracking price level changes over time, Zovko proposes tracking timing synchronicity.

The model tests for statistical surprise in how quickly an adverse market event occurs immediately following a trader's action (a fill or limit placement). The underlying assumption is that when a trader's actions leak information or cause market impact, the market reacts unusually fast. Unusually fast adverse market prints serve as a signature of causal impact.

To turn timing surprise into an actionable real-time metric, the research establishes a two-step statistical approach:

First, the background stream of market events is modeled as a Poisson process with a locally estimated, time-varying intensity. Each post-action event yields a $p$-value quantifying how surprising it is to observe an adverse event so rapidly.

Second, because a single fast event is insufficient proof, the p-values across successive fills are aggregated using Fisher's method. This allows statistically significant evidence to accumulate rapidly.

The resulting statistic reacts within a handful of fills (e.g., 3–5 actions) rather than requiring hundreds of fills. Because it operates in real time and detects causality, traders can dynamically throttle exposure to specific venues, counterparties, or execution styles mid-order if impact is detected.