Dial back ten years and stock markets rushed to be the fastest. The New York Stock Exchange slashed latency in 2009 from 105 milliseconds (ms) to only 5 ms. One year later, NASDAQ OMX in Europe went even further, reducing the time it takes to send a trade on the exchange to only 250 microseconds.
In 2019, we are witnessing history in reverse. Exchanges are trying to actively slow down the market by implementing so-called “speed bumps” — an intentional delay to trading orders. Speed bumps, however, come in many flavours: some are asymmetric (only liquidity takers are delayed) whereas others are symmetric (both liquidity takers and market-makers are delayed, such as on IEX). Further, delays can be either random or deterministic. A recent Wall Street Journal article notes that until 2020, at least twelve major exchanges worldwide would have introduced delay on trading orders, particularly on liquidity-taking ones.
Why the change of heart? Recent evidence (e.g., Shkilko and Sokolov, forthcoming in the Journal of Finance) suggests ultra-fast trading may encourage speculators to trade on very short-lived information, generating costs for market-makers and, ultimately, worse liquidity for everyone. Moreover, the mere possibility of lightning-fast profits encourages speculators to invest heavily in technology. Biais, Foucault, and Moinas (2015) argue that investments in technology for the sole purpose of short-lived zero-sum trading games are wasteful. The 2019 movie “The Hummingbird Effect” tells the story of two traders who built a straight fiber-optic cable line between Kansas and New Jersey, spending millions of dollars, to gain a sub-second speed advantage over the competition.
Do speed bumps reduce wasteful investments in speed? Which design is the best to achieve such an objective? On the one hand, if speculators are delayed relative to the market-maker, they might simply “give up” and invest less. On the other hand, they could double down on speed in an attempt to make up for the delay.
Obtaining data on high frequency traders’ investments is challenging, to say the least. To answer the question, my colleague Mariana Khapko and I decided to set up a lab experiment at the University of Toronto. We recruited 56 undergraduate management students and allowed them to invest in trading speed on a time-priority market. Students competed for an arbitrage opportunity both against each other and against computer-generated market-maker.
We find that the introduction of an asymmetric speed bump reduces investment in speed, but only by 20%. Since speculators do not only compete against a (un-delayed) market-maker, but also against their peers, speed bumps cannot fully eliminate the low-latency arms’ race.
A symmetric speed bump, on the other hand, has no effect on investment. If all traders are delayed by the same amount, they invest in speed as if there is no speed bump at all.
Importantly, the size of the speed bump matters. Moving from a small speed bump (20% of the unconditional exchange latency) to a large one (100% of the unconditional exchange latency) further reduces investment in speed by 17%. The reason is that a larger speed bump de-emphasizes competition between delayed speculators and reinforces the advantage of the market maker. Finally, we find no difference in investment behaviour if we implement random or deterministic speed bumps.
Our experimental results contribute to the on-going debate among traders, market operators, and regulators. The lab data suggests that only asymmetric delays have the potential to curb the low-latency arms’ race. However, competition between delayed speculators prevents speed bumps to fully mitigate inefficient investment in high-frequency trading technology.