Order Flow Toxicity and How Desks Measure It

Toxic flow predicts price movement against the liquidity provider that filled it, and desks measure it with markouts and imbalance metrics.

Order flow toxicity describes how likely incoming trading activity is to be informed about near-term price direction, and desks measure it because toxic flow is expensive to trade against even when the spread captured looks attractive on paper. A market maker facing more toxic flow needs a wider spread, smaller size, or faster reaction to stay profitable, which makes measuring toxicity a direct input into quoting decisions rather than an academic exercise.

The most direct measure of toxicity is markout analysis, comparing a fill price against the market some interval later, the same technique used to quantify adverse selection cost across order book liquidity. A counterparty, venue, or time window with consistently negative markouts is, by definition, toxic to trade against, because trading against it predictably costs money beyond the spread captured.

Desks build markout distributions segmented by as many dimensions as the data supports, by counterparty type where identifiable, by order size, by which venue the fill occurred on, and by proximity to scheduled news. A single aggregate markout number hides most of the useful signal, since toxicity concentrates in specific segments rather than spreading evenly across all flow.

This segmentation exercise is often the most labor-intensive part of running a market making desk's research function, since raw markout data has to be joined against counterparty identifiers, venue codes, and a calendar of scheduled news events before any pattern becomes visible.

Volume-synchronized probability of informed trading, generally referred to by its acronym, is a measure that buckets trading activity by volume rather than by clock time, then estimates the probability that a given volume bucket is dominated by one side, buy-initiated or sell-initiated. The underlying idea is that informed trading tends to arrive as a persistent imbalance in one direction, rather than as balanced two-sided activity.