The battle for speed: a race to the bottom or a race to the top?

NAISS 2023/22-851


NAISS Small Compute

Principal Investigator:

Zhiheng Xu


Stockholms universitet

Start Date:


End Date:


Primary Classification:

50202: Business Administration




The emergence of high frequency trading (HFT) has garnered a lot of attention both in the industry and in academia. The increasingly more sophisticated high-speed technology enables market participants to detect arbitrage opportunities on an ever finer scale, down to milliseconds, and to extract rent from the market. Despite the research that has been conducted, however, we contend that there are still several issues that have yet to be addressed. First, in Aquilina, Budish and O’Neill (2022), they argue that they have successfully identified the arbitrage trades (the first one in literature) and therefore used the identified trades as the basis for further analysis of the cost of arbitrage. But from their definition of the “race”: “either a mix of take attempts and cancel attempts, or all take attempts” and the result of the race, they did not check if the other side of the arbitrage actually took place. Therefore, one cannot be absolutely certain that the transactions they identified are in fact, arbitrage. It could well be that they were, in effect, market makers. Second, one cannot be sure that if it is merely an attempt to make a trade when consolidated liquidity is high. There is a possibility that these races can also be found in near-arbitrage regions. Then the identification of the arbitrage in their analysis would be flawed. Furthermore, although Foucault, Kozhan and Tham (2017) have identified that from the adverse selection the races are consuming market liquidity, the opposing force, the inadvertent market making is simultaneously providing market liquidity. Although Aquilina, Budish and O’Neill (2022) concluded from their model that the net effect of the races is negative for market liquidity, it remains obscure in empirical evidence. Our study aims to address the problems described above. Namely, we aim to test if arbitrage is present in a race and how prominently it effects the possibility of a trade taking place when arbitrage profit is positive (net of transaction costs). Using the methods of Aquilina, Budish and O’Neill (2022) will not serve the purpose due to the unavailability of the data on the other side of the arbitrage. Furthermore, although it is detectable whether the restoration of the prices is the result of a quote revision or a trade, there is a problem distinguishing a market maker and an arbitrageur: the profit of an arbitrage is almost perfectly (negatively) correlated with consolidated market liquidity. Therefore, we propose a new framework in identifying the presence of an arbitrage, with our framework, it is not necessary to identify the traders, circumventing the difficulty of the availability of the traders’ data.