Order-flow reversals: absorption, exhaustion, and sweep-reclaim

Order-flow reversal setups compare aggressive trading with price response

An order-flow reversal strategy looks for a situation in which urgent trading fails to produce the expected continuation. The strategy does not reverse merely because volume or delta is large. It asks what happened to price while that activity occurred and what happened next at a level that mattered before the event.

Three labels are common:

• Absorption: aggressive orders repeatedly trade against available or replenishing liquidity without moving price much farther.
• Exhaustion: aggressive participation diminishes near the end of a move.
• Sweep and reclaim: price trades rapidly through a known reference, then returns to the other side under a declared sequence.

These are descriptions, not proof of hidden intention. A later reversal can make almost any busy area look like absorption in hindsight. A valid rule defines the location, event, confirmation, and failure before the later path is known.

Why a reversal might follow

Large participants often divide orders over time, while passive liquidity can replenish at or near a price. Stops may create a burst of market orders that temporarily moves price beyond a visible boundary. If the urgent flow ends and the market cannot attract new trading at the displaced price, late positions may unwind.

Persistent order flow can coexist with adaptive liquidity. That relationship is useful context for an absorption hypothesis, but it does not reveal hidden intent in any single footprint.

The competing case is informed continuation. What looks like absorption can be a pause while a large buyer or seller consumes liquidity. Persistent flow can overwhelm the level and continue. That is why invalidation is not an accessory; it separates “temporary failure to move” from “the trade is still being executed.”

Location must exist before the event

Possible references include a prior session high or low, a completed value-area boundary, an earlier VWAP band, or a structure level confirmed before the candidate flow.

Selecting the largest footprint cluster after the reversal and calling it the setup leaks the outcome into the location. The study should inventory every eligible interaction with the frozen reference, including those that continue through it.

Observable components of the rule

“Large delta with no movement” is not enough. Large relative to what, movement over which interval, and how much resting liquidity was visible or inferred?

A sweep-reclaim walkthrough

Illustrative example. Assume the prior session low at 100 was frozen before today’s trade:

1. Tick data records trades through 100 down to 99.80 over a short declared interval.
2. The rule does not immediately buy. It records a sweep candidate.
3. Price then trades back above 100 and a completed decision event satisfies the defined reclaim condition.
4. A long entry occurs at the next modeled event.
5. The position invalidates on a renewed break below 99.80, times out after the declared window, or exits at its separate target.

If only a five-minute bar with high 100.40 and low 99.80 is available, the bar does not reveal whether price swept down and reclaimed or rallied first and failed later. The same OHLC values can contain opposite sequences. The study must either use event data or reduce its claim to a bar-level interaction.

Designing the experiment

Start with the price-level setup alone. Add one flow feature at a time:

• signed-volume threshold;
• limited price progress per unit of flow;
• repeated execution at the level;
• reclaim confirmation; and
• depth replenishment, only when depth data and reconstruction support it.

Preserve all candidate levels and events. Report:

• candidate, confirmed, and traded counts;
• continuation versus reversal after every candidate;
• decision latency and distance from the level;
• sensitivity to trade signing and aggregation;
• markout at fixed horizons;
• spread, impact, and fill assumptions; and
• results under nearby thresholds.

The price-only baseline shows whether the footprint condition adds information or simply redescribes a level that already tended to reverse.

How to interpret negative results

If the signal works only with a specific aggregation setting, the footprint may be an artifact. If a later confirmation captures most of the reversal before entry, the setup may be informative but not tradable at the modeled cost.

A frequent “failure” can also be informative: continuation after aggressive flow reveals that the level did not absorb enough demand. But the rule may not change roles from reversal to continuation after observing the outcome unless a separate continuation strategy was defined.

Important failure modes

• Bar data cannot establish trade order, queue, or replenishment.
• Aggressor classification changes with feed and algorithm.
• Bundled exchange messages distort apparent sweep size.
• Reference levels chosen after the reversal contain hindsight.
• Persistent informed flow consumes the apparent absorber.
• The signal arrives after most of the tradable reversal has occurred.
• High-resolution results are sensitive to latency and realistic fill logic.

Try it in Arizmic

Further reading

• Lee and Ready, “Inferring Trade Direction from Intraday Data” (1991) — Explains the classic inference behind aggressor-side labels, so an apparent sweep or imbalance can be evaluated with the classification method in view.
• Bouchaud et al., “Fluctuations and Response in Financial Markets: The Subtle Nature of ‘Random’ Price Changes” (2004) — Studies persistent order signs and horizon-dependent price response, helping separate urgent flow from the effect that flow actually produced.
• Farmer et al., “Market Efficiency and the Long Memory of Supply and Demand” (2006) — Shows how changing liquidity can offset persistent demand, supplying a testable foundation for absorption without identifying hidden intent in one footprint.
• Cont, Kukanov, and Stoikov, “The Price Impact of Order Book Events” (2014) — Connects order-book event imbalance with short-horizon price changes using event data, clarifying the fidelity needed to claim sequence and response.
• Lo and Hall, “Resiliency of the Limit Order Book” (2015) — Measures how spread, depth, and intensity recover after liquidity shocks in Australian equities; it supports studying replenishment as a dynamic process rather than labeling it from a static snapshot.