A market maker's quotes update 200 microseconds after the index future moves. Faster traders pick off 30% of its stale quotes over 50 moves a day, each costing 3 ticks of 5 paise on 500 shares. What does latency cost a year, and is a Rs 2 crore co-location upgrade worth it?

Latency costs about Rs 2.8 lakh a year, so a Rs 2 crore upgrade does not pay at this scale. Fifteen hits a day at Rs 75 each is Rs 1,125 a day. Cutting hits from 30% to 10% would save about Rs 1.9 lakh a year against about Rs 67 lakh a year of upgrade cost over three years. It needs roughly 36 times today's exposure, across many more products or much larger size, to break even.

Think of a vegetable seller still charging yesterday's price on a morning when the wholesale price has jumped. A sharp buyer who has seen the new price buys the lot before the seller changes the board. A stale quote is the same: the future has moved, the fair price of the stock has moved with it, and anyone faster than you can trade at your old price. This is adverse selectionTrading mostly with counterparties who know more than you at that moment, so your fills are systematically on the wrong side of the next price move. measured in microseconds. Here each hit is 3 ticks of 5 paise on 500 shares, Rs 75.

Multiply it out. 50 moves a day times 30% is 15 hits; 15 times Rs 75 is Rs 1,125 a day; 250 trading days make about Rs 2,81,250 a year, Rs 2.81 lakh. Now the upgrade. Rs 2 crore over three years is about Rs 66.7 lakh a year before any running costs. It does not remove pick-offs, it reduces them: from 30% to 10% saves two thirds of the cost, about Rs 1.88 lakh a year. The upgrade costs about 36 times what it saves, and the simple payback is about 107 years.

Scale. Latency cost grows with the number of products, the size quoted and the number of moves, while the upgrade is a fixed cost. A firm quoting 50 related stocks off the same future, at larger size, could face 36 times this exposure, and then the same upgrade pays. So the right first question back to the interviewer is how many instruments share the latency. The second change is less visible: a slow market maker protects itself by quoting wider or smaller, which loses volume. If faster updates let Tvaritam quote a tick tighter and win more fills, the gain in spread captured can dwarf the saving on pick-offs.

Then state the limits of the estimate. The 30% hit rate and 3-tick cost are averages; the expensive moves are the large ones, so the cost is concentrated in a few volatile days and a quiet sample understates it. The vendor's 10% is a claim, not a measurement, and the competition will also get faster, so the saving can erode within the three years. A careful answer prices the upgrade on a range, then asks for a trial before paying.