Quantifying the high-frequency trading "arms race"

Summary

Focus

We use stock exchange message data to quantify the negative aspect of high-frequency trading, known as "latency arbitrage". This means arbitrage opportunities that are sufficiently mechanical and obvious that capturing them is primarily a contest in speed.

Contribution

The paper makes three main contributions to the literature. The first is methodological: we utilise the entirety of exchange message data, as opposed to standard order book data, to measure latency arbitrage. The second is the set of empirical facts we document about latency arbitrage. The third contribution is that we develop two new approaches to quantifying latency arbitrage as a proportion of the overall cost of liquidity.

Findings

We show that races are very frequent and very fast, and that over 20% of trading volume takes place in races. A small number of firms win the large majority of races, disproportionately as takers of liquidity. Most races are for very small amounts of money, averaging just over half a tick. But even just half a tick, over 20% of trading volume, adds up. The latency arbitrage tax, defined as latency arbitrage profits divided by trading volume, is 0.42 basis points. This amounts to about GBP 60 million annually in the UK. Extrapolating from our UK data, our estimates imply that latency arbitrage is worth about $5 billion annually in global equity markets alone.

The new approaches we develop to quantify latency arbitrage as a proportion of the overall cost of liquidity, used in conjunction with our results, show that latency arbitrage accounts for 33% of the effective spread and 31% of all price impact, and that market designs that eliminate latency arbitrage would reduce the cost of liquidity for investors by 17%.

Abstract

We use stock exchange message data to quantify the negative aspect of high-frequency trading, known as "latency arbitrage". The key difference between message data and widely familiar limit order book data is that message data contain attempts to trade or cancel that fail. This allows the researcher to observe both winners and losers in a race, whereas in limit order book data you cannot see the losers, so you cannot directly see the races. We find that latency arbitrage races are very frequent (about one per minute per symbol for FTSE 100 stocks), extremely fast (the modal race lasts 5-10 millionths of a second), and account for a remarkably large portion of overall trading volume (about 20%). Race participation is concentrated, with the top six firms accounting for over 80% of all race wins and losses. The average race is worth just a small amount (about half a price tick), but because of the large volumes the stakes add up. Our main estimates suggest that races constitute roughly one third of price impact and the effective spread (key microstructure measures of the cost of liquidity), that latency arbitrage imposes a roughly 0.5 basis point tax on trading, that market designs that eliminate latency arbitrage would reduce the market's cost of liquidity by 17%, and that the total sums at stake are on the order of $5 billion per year in global equity markets alone.

JEL classification: D47, G10, G12, G14

Keywords: market design, high-frequency trading, financial exchanges, liquidity, latency arbitrage, trading volume, message data