The HFT Arms Race Argument

Every microsecond a firm shaves off its market-data-to-order pipeline requires real spending — better hardware, microwave links, colocation, engineering talent — and firms have collectively spent enormous sums chasing single-digit-microsecond advantages. The economic question this raises is whether all that spending actually improves markets, or whether it's closer to a treadmill: every firm has to keep running just to stay in the same relative place, because the moment one firm gets faster, the others must match it or lose every race, and the eventual winner of any given race captures the whole prize while the losers get nothing for their investment. This is the arms race argument: past some point, speed spending is privately rational for each firm but produces no corresponding social benefit, purely redistributing a fixed prize while burning real resources to compete for it.

Why continuous trading makes this happen

The mechanism is specific to how modern exchanges match orders: continuous limit order books, where orders are matched the instant a marketable order arrives, turn every piece of new information into a discrete race — whoever reacts fastest wins the entire prize. Because the race is winner-take-all and repeats constantly, there's no natural stopping point to speed investment: a firm 10 microseconds behind wins nothing, so competitors keep spending until further gains simply cost more than they're worth.

Researchers such as Budish, Cramton and Shim have argued that switching from continuous trading to very frequent batch auctions — collecting orders over a short window (say, every 10 or 100 milliseconds) and matching them all at once at a single clearing price — would defuse the race entirely. Being 5 microseconds faster than a rival provides no advantage if both orders land in the same 100-millisecond batch and are matched together; the winner-take-all race collapses into a tie, removing the incentive to keep spending on speed for its own sake while preserving trading itself.

Worked example: a symmetric race with no social value

Two market makers each spend $5 million a year on infrastructure to shave microseconds off their reaction time to a common signal, and in equilibrium each wins the race to react first roughly half the time, earning a comparable share of the available sniping profit either way. If both firms instead spent nothing extra, the relative race outcome would be unchanged — still roughly 50/50 — but combined they'd have saved $10 million a year with the exact same trading outcome for the market as a whole. The spending doesn't change who wins on average; it only raises the entry cost of competing, which is the textbook signature of an arms race rather than genuine value creation.

What this means in practice

The arms race argument doesn't claim all speed investment is wasteful — faster systems genuinely do reduce stale-quote risk and can tighten spreads up to a point — only that competitive speed spending has diminishing and eventually negative social returns once it becomes purely about winning symmetric races rather than serving customers better. It's a live policy debate: a handful of venues (notably IEX, with its speed bump) have implemented mechanisms specifically designed to blunt the value of a raw speed edge, while most major exchanges still run continuous, race-based matching.