Your ETF quotes are derived from a futures feed you receive 500 microseconds late. The future moves one tick worth 2.5 basis points about 80 times a day, and on each move your resting quotes are exposed with a 40% chance of being hit on 200 lots of $10,000 notional each. What is the daily cost?

Your ETF quotes are derived from a futures feed you receive 500 microseconds late. The future moves one tick worth 2.5 basis points about 80 times a day, and on each move your resting quotes are exposed with a 40% chance of being hit on 200 lots of $10,000 notional each. What is the daily cost?

Approach: Count the exposed events per day, multiply by the probability of being hit and the size at risk, then value each hit at the tick move converted into dollars per lot.

$16,000. A 2.5 basis point move on $10,000 of notional is $2.50 a lot. Of the 80 moves a day, 40% end in a hit, so 32 events; at 200 lots each that is 6,400 lots and 6,400 * $2.50 = $16,000 a day, about $4m a year. That figure is the budget for removing the latency, since any feed, hardware or colocation package costing less than $16,000 a day and eliminating the exposure pays for itself. The same arithmetic prices the alternatives, which are to widen the ETF quote enough to cover $16,000 a day of stale quote losses or to cut resting size on the side the future is more likely to move toward. Adverse selection from a latency gap scales linearly in size, so halving quoted size halves the daily cost.

Follow-up: The exposure only exists on the side the future moves toward. How does that change the estimate and what does it imply about quoting asymmetrically?

Key concepts: stale quote, latency, adverse selection, daily cost.