TWAP: how an algorithm slices a large order, and why you can see it

2026-09-18

If you need a million dollars filled and the coin does three million a day, a market order is not an option: price runs away and your own entry eats half the edge. So the size gets sliced.

What TWAP does

TWAP stands for Time-Weighted Average Price. The algorithm takes an order, splits it into equal pieces and sends them at even intervals over a set window. A million dollars over six hours is roughly $2,800 every minute. The goal is to land on the period's average price without advertising the size.

Next to it lives VWAP — same idea, but the slices scale with market volume: more turnover, bigger slice. And POV (percentage of volume) — "take N percent of whatever trades." From the outside all three look similar: a long run of uniform prints on one side.

The paradox: the camouflage is the fingerprint

The algorithm hides its size, and in exchange it acquires a regularity no human has. People trade raggedly: $300, then $12,000, then a minute of nothing. TWAP keeps two numbers constant:

• Slice size. Consecutive prints land within a few percent of each other. The amounts are often round in dollars rather than in coins — the algorithm counts money.
• Beat. The average gap between hits. Half a second of jitter is normal, the algorithm adapts to the book. Ten seconds of jitter is something else.

What the speed tells you

The beat is not only a tell, it is information about intent.

• Beat of tens of seconds, small slices — calm, stretched-out execution. Somebody is building a position and is in no hurry. These runs go on for hours.
• Sub-second beat, large slices — urgency. News, a position being closed, or a liquidation being worked. Short runs, and they almost always move price.
• The beat accelerates mid-run — the algorithm has fallen behind schedule and is catching up. A sign the market is walking away from it while the size still has to be done.

Where it runs and what price does

The useful observation is not the TWAP itself but how price reacts to it.

• The algorithm buys, price stays flat — somebody above is feeding it, most likely out of an iceberg. No rally until that seller is finished.
• The algorithm buys and price follows — no resistance, the position is being built into thin air. The interesting case: the buyer must complete the size and nobody is supplying it.
• The algorithm goes quiet — it either finished or got switched off. Price frequently continues in the direction it was pushing, simply because the pressure that was containing it is gone.

Limits of the method

TWAP says nothing about being right. A fund may be rebalancing, a market maker hedging an option, an exchange moving reserves between wallets. You will never learn the motive. What you do learn is that a large participant is present, which side they are on, and how fast they work. That alone is enough to avoid trading into them and to avoid placing a stop where they will still be operating.

Second limit: sophisticated execution deliberately adds randomness, varying size and timing to blend in. Those are harder to catch, but not impossible — their spread is still tighter than a human's, because the schedule constraint never goes away.

How machines count it

Manual counting works on one coin. To watch TWAP across a thousand pairs at once, software listens to every exchange's trade stream, groups prints by coin and side, computes average slice size and average beat, and the moment a window fits the tolerances it emits a card: exchange, coin, side, speed, how much has gone through so far. That is precisely what botscan does.

← All articles