Dealing with micro-bursts: A congestion fee for high-speed markets

A message congestion fee that targets micro-bursts can simultaneously improve liquidity, as well as reduce the risk of a congested market.

High-frequency trading algorithms breed congested markets. One-fifth of all trades on Nasdaq and the London Stock Exchange are executed in so-called “micro-bursts,’’ each lasting a few microseconds and on aggregate taking up less than 0.05 seconds of an average trading day. This is, perhaps, not so surprising as it could seem at first sight: Trading algorithms react simultaneously to the same profitable signals, and submit buy or sell orders that are likely to reach the exchange within the same millisecond.

Why worry about surges in trading activity?

“Micro-bursts” are problematic for at least two reasons. First, they are strongly associated with a trading strategy called latency arbitrage, which research has shown that may harm liquidity and increase transaction costs for long-term investors.

Second, beyond liquidity concerns, micro-bursts also lead to congested markets, as the surge in trading messages temporarily approaches or even exceeds the exchange’s order processing capacity. On February 27, 2020 (a particularly high-volatility day), the Canadian exchange TMX was forced to close early due to a shortage in message capacity’. Presumably to avoid exactly such disruptions, Nasdaq imposes fines on traders that violate their “Excessive Messaging Policy.’’

To avoid congestion, exchanges invest in excess processing capacity, faster computer chips, and bigger data buffers. However, it is not clear whether such investments are efficient: As much as 90% of the high-throughput exchange infrastructure remains idle for 90% of a typical trading day.

A solution: “Congestion fees” during trading surges

My co-author Michael Brolley and I propose a single market design solution to tackle both concerns. We argue that a message congestion fee that targets micro-bursts can simultaneously improve liquidity, as well as reduce the risk of a congested market.

The rationale for a congestion fee is to effectively enable dynamic “surge pricing’’ of orders. If trading during micro-bursts becomes more expensive, high-frequency traders (HFTs) face reduced incentives to engage in low-latency arbitrage. Importantly, exchanges are no worse-off by charging a congestion fee (which would be levied subsequent to HFT races) as they can generate at least as much revenue as from levying co-location subscription fees on a monthly basis. Finally, congestion fees are competition-proof: in a fragmented environment where trading venues charge ex ante co-location fees, any exchange switching to ex post congestion fees would gain a competitive edge.

Our proposal is to apply the congestion fee on all liquidity-taking orders, regardless of whether the order is executed or not. The level of the fee increases in the number of messages sent to the exchange within the same narrow time window (e.g., 500 microseconds). Within this environment, HFTs weigh the trade-off between latency arbitrage profits on stale quotes and the congestion costs of competing with others. The net effect is a reduction in speed race competition, and consequently an improvement to liquidity. Such improvement in liquidity flows through as a net benefit to liquidity investors, who, because they do not arrive during micro-bursts, do not face congestion fees.

Moreover, exchanges may be able to reduce their investments in capacity, as the effective surge-pricing mechanism reduces the likelihood that an exchange hits its capacity limit during a micro-burst.

How do congestion fees compare to alternative solutions?

Congestion fees are not the only way to tackle latency arbitrage. Some exchanges have implemented speed bumps: that is, intentional delays on trading orders designed to “slow down” arbitrageurs. Another solution that attracted significant attention in academic and regulatory circles is to chop down the trading day in a large but discrete number of batch auctions, essentially watering down time priority and discouraging fast traders.

A common feature of the existing proposals is that they aim to eliminate latency arbitrage entirely. It follows that exchanges have a stake in latency arbitrage activity as they are able to extract part of the rents from high-frequency trading: On aggregate, in 2015, Nasdaq and NYSE earned between $251.6 and $281.5M from providing co-location services, a necessary pre-condition for an HFT to successfully participate in low-latency races. Not surprisingly, Budish, Lee, and Shim (2020) note that stock exchanges

have vigorously opposed […] attempts to address the negative aspects of high-frequency trading.

It turns out that a message congestion fee successfully aligns the objectives of exchanges and long-term investors. First, a congestion fee reduces the expected magnitude of latency arbitrage rents, leading to better liquidity. Second, the congestion fee redistributes latency arbitrage rents from high-frequency traders to exchanges, who consequently earn a larger share of a smaller prize.

In contrast to both frequent batch auctions or speed bumps, in a fragmented market congestion fees can lead to higher exchange revenues than those generated by co-location fees. The rationale is that, since co-location fees are charged via monthly subscription, an HFT can discipline any exchange charging high fees by “locking themselves” out of that venue and providing cheaper liquidity elsewhere. Since congestion fees are charged after a micro-burst takes place, HFTs lack the ability to credibly commit to stay out of the race.

How large should the congestion fee be?

We use aggregate data pertaining to low-latency races on the London Stock Exchange (LSE) from Aquilina, Budish, and O’Neill (2020) to obtain a ball-park for congestion fee.

A simple back-of-the-envelope calculation suggests that a marginal fee between £0.10 and £0.50 for each liquidity-taking order in a 500 microsecond interval containing three messages or more allows exchanges to fully extract any latency arbitrage rents from HFTs. The fee corresponds to approximately 15% to 35% of the typical size of an individual arbitrage opportunity, estimated at just below £2 across LSE securities in 2015.

Link to the research paper: [link].