You rest 100 lots at the best bid of 20.00 behind 500 lots. In one millisecond 400 of the lots ahead of you are cancelled and the offer size doubles. Your fill probability has risen. Argue for pulling the order anyway and put numbers on it.

You rest 100 lots at the best bid of 20.00 behind 500 lots. In one millisecond 400 of the lots ahead of you are cancelled and the offer size doubles. Your fill probability has risen. Argue for pulling the order anyway and put numbers on it.

Approach: Separate the unconditional fill probability from the distribution of the price conditional on the fill, then ask what the cancelling makers had learned that made them give up their place.

Pull the order, because the fill that just became more likely is the one you do not want. The 400 cancelled lots belong to makers who revised fair value down and paid away queue position they had already earned, which is expensive and therefore informative. If fair was 20.005 before the cancels and the same information moves it to 19.995, a fill at 20.00 turns from 0.5 cents of edge into 0.5 cents of loss before any further move. The doubled offer adds size that wants to sell into you, so the adverse selection on the remaining fill has jumped. Conditional fill economics dominate: the fill probability may have gone from 20% to 60%, so the expected loss roughly triples while the edge per fill has already turned negative. Queue position has value only while the price it queues at is still the right price.

Follow-up: How would you separate a cancel driven by information from one driven by a maker hitting a position limit?

Key concepts: queue position, cancellation, adverse selection, conditional fill.