On 15 Sep 2026 the near month Nifty future closed 105.3 points above the index, and over the 300 sessions in the school's files that gap ran from −124.9 to 221.3 points. A tick, the smallest price step, is taken as 0.05 of a rupee, the step every Nifty settlement price in the files sits on. On one lot it is worth 3.25 rupees. A race to earn one tick is a race for a few rupees. Firms run it thousands of times a day and spend heavily to win. This lesson asks who pays for that and who keeps the prize. It builds a small model of a winner takes all race, shows how entry eats the profit, and reads what SEBI's circulars did and did not do about it. The model is made up, the basis figures are real, and none of it is a way to earn.
3By the end you can
- Work out how many firms enter a winner takes all race and what they spend in total.
- Explain the arms race as a prisoner's dilemma and say what a batch auction changes.
- Place SEBI's 2018 transparency measures and the 2026 closing auction on that map.
The idea, plainly
What is the latency race?
Two prices that should move together sometimes do not, for an instant. The Nifty future may be quoted a tick away from where the index says it should be. The first firm whose order reaches the exchange takes the tick, and the others find their orders arriving to a price that has gone. The prize is tiny and indivisible. Whoever is first wins it all; second place wins nothing, and has paid for the attempt. In HFT, MFT and LFT compared I showed how small a millisecond of Nifty movement is. A race for the gap is where firms spend to be a few microseconds ahead.
Why is the race wasteful?
Because speed is bought and the prize is shared by who wins, not by who spends. Suppose the opportunities in a year are worth a fixed pool. Every firm that joins pays the cost of being fast enough to compete, and the pool is split, on average, among all of them. Firms keep joining while the expected share exceeds the cost. They stop when it equals the cost. At that point each firm earns nothing above its costs and the whole pool has gone on speed. This is the finding of the research on high frequency trading and the arms race: the gain from being first is private, the cost of the race is social, and a faster wire does not make the market a fairer place to trade. Whether the race also brings benefits, such as tighter spreads, is a separate question on which studies differ, and the model does not try to settle it.
Can the race be changed?
The market design answer is to stop rewarding time at the microsecond scale. A frequent batch auction collects orders over a short interval and matches them together at one price, ranking by price first, so that being a few microseconds earlier within the interval gains nothing. It removes the prize for speed without banning speed. A random end to an auction does the same job at one moment of the day, because nobody can win by being the last order in before a close they cannot predict.
What has India done?
SEBI's discussion paper of 5 Aug 2016 floated minimum resting times, frequent batch auctions, random speed bumps and others. None of those was adopted. The circular of 9 Apr 2018 chose transparency: a free tick by tick feed for all members, published latency statistics, a managed co-location service for smaller members and a unique identifier on every algo order. Co-location itself is legal and regulated, and the lesson on NSE's co-location case reads the orders. The newest step is different in kind. From 3 Aug 2026 the cash market has a closing auction, and SEBI's circular of 15 Oct 2026 sets it to run from 3:30 to 3:40 pm from 7 Dec 2026, closing at a random moment between 3:38 and 3:40. A call auction with a random close is the mechanism the theory describes, applied at one moment of the day. From 3 Aug 2026 its closing prices also set the expiry day settlement of index options; from 7 Dec 2026 that settlement moves to the index computed from the last 30 minutes of continuous trading, while the auction keeps setting the close of the shares.
3 terms this section turns on
India chose transparency in 2018 and an auction with a random close in 2026
- Co-location guidelines
Fair, transparent and equitable access, similar latency, quarterly latency reports
- Discussion paper
Resting time, batch auctions and speed bumps floated, none adopted
- SEBI circular
Free tick feed, latency statistics, managed co-location, an identifier for every algo
- NSE co-location order
Disgorgement and a six month bar, later set aside in part on appeal
- Closing auction circular
Closing price from an auction, in the cash market from 3 Aug 2026
- Auction timings circular
Auction 3:30 to 3:40 pm with a random close from 7 Dec 2026
Where do the races come from?
The same index is traded through several doors at once: the cash basket, the future, the options and, for some, an exchange traded fund. A price in one door moves first and the others follow, and the gap between a price and its no-arbitrage value is what a fast firm collects. The fair value of the future is worked out in the lesson on Nifty futures fair value, and an option's link to the future by put call parity. Each link is a possible race. Where a link is slow or costly to trade, such as the cash basket, the gap stays open longer and slower firms can reach it. Where it is quick, the gap lasts for microseconds and only the fastest see it.
Two terms to hold on to from 19.12
Is the profit really zero?
Only in the simplest version. Firms are not equal: some have better code, better sites or better data, and they earn a rent that entry does not remove because others cannot copy them. The model's point is about the part of the profit that comes from being first by a hair. That part is competed away. The part that comes from knowing something others do not, or from carrying risk others will not, is not. It is also not the same as the market maker's spread in the lesson on market making, which is paid for holding inventory. A firm can do both, and its accounts do not say which part of its profit came from which.
Does the race help anyone?
Possibly. Fast firms that quote both sides keep spreads tight, and prices adjust sooner to news across doors. Against that, the spending is a cost to the system, and slow quoters can be picked off by fast ones, so they quote wider to protect themselves. The two effects pull in opposite directions and their sizes depend on the market and the period, so studies do not agree. This lesson does not claim a net answer. It claims only that spending which exists to be first by a margin that does not change any trade is a cost that rules can lower.
Worked example: a race for one tick
Start with how small the prize is next to the thing being chased. The line shows the gap between the near month future and the index in points. A tick of 0.05 is a rounding error against it, so the race is not for the gap. It is for the small leftover that appears when one price moves and the other has not caught up.
The basis ranged over 346.3 points while a tick is 0.05
Entry continues until the share falls to the cost, near 18 firms
Maths: free entry and the arms race
N* = pool ÷ cost
With N equally fast firms each earns pool ÷ N. Profit is zero when pool ÷ N equals the cost C, so N* = pool ÷ C and total spending on speed, N* × C, equals the pool. This is the zero profit result of free entry into a winner takes all contest. If one firm buys a faster link for an extra cost u, it wins every race and takes the whole pool for the price of u. Every other firm then has a reason to buy the same upgrade, and when all have it the pool is shared again while each pays u more. The upgrade is a good move for one firm and a loss for all of them together, the shape of a prisoner's dilemma.
The prize goes to the first order and the cost is paid by all
Read 3 more paragraphs, 259 words
Now the model, with every input a choice. Say a firm can win one tick on a lot in a race and that there are 20,000 such races a day over 250 days. One tick on one lot is 3.25 rupees, so the pool of prizes is 20,000 times 250 times 3.25, or 1,62,50,000 rupees a year. The cost of being fast enough to compete is set at one full rack at the exchange, 9,00,000 rupees a year at the rate in NSE's circular of 29 Apr 2026. A real firm's cost is far higher, since the rack is only one line, so this understates the waste.
With free entry, firms join until the pool divided by the number of firms falls to the cost. That gives 18 firms. Each wins about 9,02,778 rupees a year and pays 9,00,000, so it earns nothing above its costs. All of the pool, 1,62,50,000 rupees, has gone on racks.
Now the upgrade. A firm pays a further 2,00,000 rupees for a faster link. If it alone buys, it wins every race and takes the whole pool of 1,62,50,000 rupees, for a profit that dwarfs the upgrade. But the others see this and buy too. Once all 18 have the faster link, each is back to a share of about 9,02,778 rupees, and each has paid 9,00,000 for the rack and 2,00,000 for the upgrade, so each now loses 2,00,000 rupees a year. The pool did not grow and nobody trades better. The wire is faster and the profit is gone.
What does this mean for a retail algo?
Nothing to race for. A retail API algo routes through a broker, is held to ten orders a second per exchange before it must be registered, and has no rack at the exchange. Its cost of speed is not the issue; it is out of the contest. The real use of this lesson is to read headlines. When a firm is said to earn a fortune from speed, ask what it spends on speed, and whether the profit is a rent that new entrants will compete away. Where profit stays high for years, the question is what stops entry: a barrier of cost, a licence, a skill or an unfairness that rules were written to remove.
Try it yourself
Change two inputs in the model and redo the arithmetic: double the number of races a day, then halve the cost of a rack. Write down the number of firms that enter in each case and the total spent. Notice that the answer depends on the ratio of pool to cost, and that cutting the cost of speed does not make the firms richer. It only brings more of them in.
The common mistake
The belief: Fast firms earn a fortune from speed alone.
Speed is the entry fee and not the profit. In a winner takes all race the prize is shared out among entrants until the average return is the cost, and the margin that survives belongs to whatever others cannot copy. Reading the other way is just as wrong: The race is all waste, so speed adds nothing.
Speed is also behind tight quotes and quick price discovery, and the evidence on the net effect is mixed. What the theory says is narrower. Speed that exists only to be first by a hair earns a private prize and has a social price, and market design can change that price.
Where this section sits in module 19.12
Key takeaways
- In a winner takes all race, entry spends the prize on speed.
- Free entry stops where the share of the pool equals the cost.
- A faster link is good for one firm and a loss for all.
- A batch auction or a random close removes the prize for timing alone.
- SEBI chose transparency in 2018 and a closing auction in 2026.
- A retail API algo is not in the race.
Before you trade. Derivatives trading carries a high risk of loss. SEBI's study "Profitability of Individual Traders in the Equity Derivatives Segment (FY25 to FY26)" (August 2026) found that 87.7% of individual traders in equity F&O made net losses in FY26, with aggregate net losses of Rs 91,685 crore. Read the SEBI study