The Speed Race Is a Design Flaw: Budish, Cramton and Shim on Frequent Batch Auctions

Most commentary on high frequency trading is about whether HFTs are good people. Budish, Cramton and Shim's paper is refreshing because it declines to care.

Their argument is that the arms race is not a story about villains. It is a story about a design flaw in the market's most basic rule, a flaw so fundamental that nobody had thought to question it. And once you see the flaw, you realize that no amount of regulating traders will fix it, because rational traders in that design have no choice but to race.

The problem: two prices for the same thing, and a footrace to exploit it

Start with a fact they document carefully. The S&P 500 futures contract in Chicago and the SPY exchange-traded fund in New York track the same underlying thing. Their prices are almost perfectly correlated.

But almost perfectly correlated at human timescales is not correlated at all at machine timescales. Zoom in far enough, and the correlation between them collapses toward zero. Over very short intervals, one moves and the other has not yet.

That gap is a pure, riskless arbitrage. The futures price jumps in Chicago. For a brief moment, everyone quoting SPY in New York is quoting a stale, now-wrong price. Whoever gets to New York first grabs the free money.

Here is the observation that turns this into a devastating argument. This arbitrage opportunity never goes away. The authors show that the frequency and profitability of these opportunities has stayed remarkably stable over years, even as speeds improved from milliseconds to microseconds to nanoseconds. Everyone got a thousand times faster, and the arbitrage is still there, worth about the same.

That is a stunning result, and it demolishes the standard defence of the arms race. The usual claim is that competition among fast traders will compete away the profits. It has not. Competition made everyone faster, and the profits remained, because the opportunity is not created by slowness. It is created by the market's rules.

The key idea via analogy: a race decided by a photo finish that keeps getting sharper

Why does this happen? Because of what the authors identify as the culprit: the continuous limit order book treats time as continuous, and processes orders one at a time, in the order they arrive.

That rule sounds innocuous. It is not. It means that being one nanosecond faster than the next person is worth as much as being one hour faster. The market processes your message first, and you get the whole prize. Second place gets nothing.

This is a winner-take-all race with an infinitely sharp finish line, which is precisely the structure that generates unbounded investment in speed. There is no natural stopping point. However fast everyone is, being slightly faster still wins everything, so it is always worth spending more.

And now, the deeper cost, which is the part that should worry you far more than HFT profits.

Market makers cannot escape this, and so they widen their spreads. Put yourself in the shoes of someone posting a quote in SPY. You know that when the Chicago futures move, you will be picked off before you can cancel, unless you are the very fastest. You will lose money on those trades, every single time, and there is nothing you can do about it. So what do you do? You charge everyone else more. You widen your quote to cover the losses you know you will suffer to the snipers.

That widened spread is paid by ordinary investors. The arms race is a tax, and the tax is baked into the price you pay to trade. This is the real cost, and it dwarfs the direct profits of the fast traders.

The key idea, part two: change the clock

Their proposed fix follows directly, and its elegance is the reason the paper is famous.

Stop processing orders continuously and one at a time. Process them in discrete batches.

Divide the trading day into short intervals, say a fraction of a second. All orders arriving during an interval are collected. At the end of the interval, they are all processed at once, in a single uniform-price auction. Everyone in the batch gets the same price. Arrival time within the batch does not matter at all.

Now look at what happens to the arms race. Chicago's futures price jumps. Ten sniping firms all race to New York to hit the stale quote. Under the old rules, the fastest one wins everything. Under batch auctions, all ten arrive in the same batch. They now have to compete on price, not on speed. And competing on price means the profit gets bid away to nothing, which is exactly what competition is supposed to do.

The change is subtle and total. It converts competition on speed, which is socially wasteful, into competition on price, which is socially useful. The prize for being a nanosecond faster becomes zero, so nobody spends anything to get it. And because market makers are no longer certain to be sniped, they can quote tighter, and ordinary investors pay less.

Notice what the fix does not do. It does not ban anything. It does not slow anyone down by decree. It does not require identifying who the bad actors are. It simply changes the rule so that the wasteful behaviour is no longer profitable. That is what good mechanism design looks like.

Why it mattered

• It reframed the entire HFT debate from morality to design. The question stopped being "are HFTs parasites?" and became "why does our market reward nanoseconds?" That is a much better question, and the shift is largely this paper's doing.
• It identified the real cost. The public focuses on HFT profits. Budish, Cramton and Shim showed the bigger cost is the spread widening that sniping forces on every liquidity provider, which is paid by everyone who trades. That is a far more important number than anyone's P&L.
• It provided a concrete, implementable alternative. Frequent batch auctions are not a thought experiment. Several venues have adopted batch auction mechanisms or speed bumps in the years since, and the design has been seriously debated by exchanges and regulators worldwide. Whether or not any given implementation succeeds, the paper put a real option on the table.
• The stability of arbitrage profits over time is a killer empirical fact. It is the single most persuasive piece of evidence that the race is structural rather than a transitional inefficiency, and it is very hard to argue with.

The honest limitations

• Somebody has to go first, and nobody wants to. An exchange that unilaterally switches to batch auctions may lose volume to continuous competitors, because traders who profit from the current design will simply go elsewhere. This coordination problem is the main reason adoption has been slow, and the paper cannot solve it.
• How long should a batch be? Too short and the race persists within the batch's granularity. Too long and you introduce real delay costs for people who genuinely need to trade now, and you may create new gaming around batch boundaries. The right interval is an empirical question with no clean answer.
• The arms race may just move. Batching removes the incentive to be fast within a batch, but there may be new incentives around predicting batch outcomes, or timing orders relative to batch boundaries, or racing across venues with different batch schedules. Clever people find new races.
• The model of the arbitrage is stylized. The sniping story is cleanest for the near-riskless futures-versus-ETF case. Much of what fast traders do is messier and less clearly wasteful, and the argument for batching is correspondingly less airtight in those cases.
• Existing evidence says the current market is quite good. Hendershott, Jones and Menkveld found algorithms improved liquidity. Brogaard and co-authors found HFTs improve price discovery. Batch auctions might sacrifice some of that efficiency. The authors argue the trade is worth it. Reasonable people disagree.

The one-line takeaway

Budish, Cramton and Shim argued that the speed arms race is not caused by greedy traders but by the rule that markets process orders continuously, one at a time, which makes a nanosecond edge worth everything, and that batching orders into frequent auctions where arrival time within a batch does not matter would convert the wasteful race on speed into a useful competition on price.