Clearly, certain market participants are willing to go to great lengths and considerable cost to gain the ultimate speed advantage. But there are occasions when latency arbitrage activity has a negative effect on outcomes for long-term investors, and, as such, it may be beneficial to curb this activity through implementation of thoughtfully designed anti-latency arbitrage mechanisms. XTX Markets’ Matt Clarke looks at the types of ALA mechanisms in the market today and the arguments for and against them.
Recently, there has been an increasing preoccupation with “speed traders” and the lengths to which some are going to establish and protect their relative speed advantages. Stories like the below from Bloomberg capture the imagination:
In fact, a major motion picture was recently released that has as its plot a race to shave a couple of milliseconds (thousandths of a second) off the time it takes to send an order message from a fictional exchange data center in Kansas City to the NYSE: www.thehummingbirdproject.film.
Clearly, certain market participants are willing to go to great lengths and considerable cost to gain the ultimate speed advantage. How valuable is the advantage to them, and what is the impact on and cost to other market participants?
These stories beg the question: Isn’t there room in market structure for venues that do not benefit those willing or able to achieve latency advantages? And might marketplaces and market participants be better off with a speedbump or any other mechanism that seeks to reduce latency advantages across participants (an anti-latency arbitrage mechanism)?
Most market structure arguments – like this one – are generally informed by the author’s frame of reference.
We believe the purpose of markets is ultimately to serve end users – that is, long-term investors and their agents. It is from that perspective that we will approach this topic. Does a given market design feature holistically benefit or harm long-term investors?
Market makers and liquidity providers are important but only inasmuch as they provide a service to end users – i.e., providing immediacy of risk transfer and taking on time/fluctuation risk in return for a spread.
Arbitrageurs – who may or may not also make markets – can be helpful, as they perform the unexciting but useful service of aligning multiple venues. However, if they introduce negative externalities for end users while doing so, the needs of arbitrageurs should be considered as subordinate to those of long-term investors.
As we will explore below, there are occasions when latency arbitrage activity has a negative effect on outcomes for end users, and, as such, it may be beneficial to curb this activity through implementation of thoughtfully designed ALA mechanisms.
The State of Latency in 2019
There is an enduringly popular legend that the Rothschild family made a fortune by utilizing its private pigeon network to learn the outcome of the Battle of Waterloo and trading on this information. Although financial markets have become much more sophisticated and electronic since then, the competition among participants to receive information first remains just as fierce. There are, however, two important things to realize.
First, as Donald MacKenzie of the University of Edinburgh notes in his excellent TabbFORUM piece, “How Fragile Is Competition in High-Frequency Trading?” the exchange groups’ considerable and successful focus on reducing “jitter” (quasi-random fluctuations in processing times) on their exchanges means “even tiny speed advantages” have become incredibly important. Given that exchange jitter is now measured in single-digit nanoseconds (a nanosecond being a thousand-millionth of a second) it follows that “in a particular market … one HFT firm – or a small number of firms – may achieve an advantage in speed that’s very hard and very costly for their rivals to overcome.”
Second, this is not about open competition. In order to obtain that last half-turn of the screw of latency advantage, the latest technologies being considered include such offerings as low-earth orbit satellite constellations, some of which propose to use free space optics – yes, space lasers – and autonomous solar-powered drones that could repeat signals from one data center to another over water. It is crucial to note that the aim of these efforts to reduce latency is not the promotion of open and free competition but rather to exploit scarcity for relative advantage.
This pattern – latency decreasing, cost increasing, scarcity increasing – is observed repeatedly. When ICE released roof space in its Basildon data center, there were a total of 32 slots – of which a small fraction were optimal for relaying signals to the LSE and trading venues with servers in Slough. Once these advantageous slots have been reserved, the market participants that have access to these resources enjoy an insuperable latency advantage over everyone else. Similarly, there are “optimal” frequencies for microwave channels and, once these are licensed, they are unavailable to other participants. This theme is echoed along the path of the microwave circuits, where access to specific rooftops and frequencies are fiercely protected and exclusively maintained assets.
The aim is to find, obtain and protect a systemic advantage: to find the rooftop that you can lease so no one else can; the intellectual property that you can exclusively license; to register the frequency that prevents others taking the same path; and so on. The costs are no longer the same as training pigeons. Furthermore, the cost of this private infrastructure is ultimately paid for by real-world end users.
What Kinds of ALA Mechanisms Exist Today?
The purpose of ALA mechanisms is to prevent latency arbitrage by levelling latency across all participants so everyone can trade and compete with equal access to timely market data. There are two kinds of ALA mechanisms: technology-based and policy-based.
Technology-based ALA mechanisms
These diverse implementations include “latency floors” or “speed bumps.” The precise implementation will differ across venues to reflect differences in products, rule books, regulatory regimes and proxy markets. Implementation details must take all these factors into account and are crucial in ensuring a well-designed ALA mechanism.
One can consider an illustrative symmetrical speed bump implementation. Incoming orders are subject to a speed bump of typically several milliseconds before being eligible for a match. In some designs the length of the speed bump may be randomized.
How does this prevent latency arbitrage?
Imagine that a related instrument jumps in price in a different market center; both the market maker (passive order) and latency arbitrageur (aggressive order) observe this at the same time, but the participant seeking to latency arbitrage has a two-millisecond speed advantage in sending this information over to the speed-bumped venue due to its private microwave network. Instead of being able to pick off the stale offer immediately, it must traverse the three-millisecond speed bump, which affords liquidity providers a level playing field, as they can incorporate the same information into their pricing and cancel the stale offer before it matches and is picked off. Both passive and aggressive incoming orders are subject to the speed bump and latency has been floored.
Other examples of technology-based ALA mechanisms include those that impose a few millisecond delay on incoming orders to remove liquidity, thereby giving market makers the opportunity to react to new information and cancel stale orders.
Policy-based ALA mechanisms
A good example of a policy-based ALA mechanism can be found in Aquis, a pan-European equities exchange. Aquis is publicly traded, and in full disclosure, XTX Markets owns a non-controlling minority stake, an investment that was made because we believed it was a positive example of market structure that is good for end users and would therefore prove popular over time.
As stated on its website, Aquis does not permit “aggressive non-client proprietary trading.” Only order flow deriving from natural buy-side exposures is eligible to remove liquidity from the platform. High-frequency trading firms may supply liquidity to the platform, but they cannot take liquidity from this market. As a result, the venue’s market makers (and end users leaving pegged orders) may be able to offer tighter spreads and/or larger size to this natural buy-side platform because they know they will not be latency arbitraged by participants with a systematic speed advantage.
If market makers are instead forced to quote blindly into an orderbook whose incoming orders might originate from end users but might equally be latency arbitrage, it follows that market makers will quote wider and in smaller size. After all, they must quote to their average experience on the venue – one participant may be subsidizing the activity of another.
This theory is intuitive; but has it proved successful in practice? Based on analysis by Liquidmetrix, again on its website, Aquis believes its ALA mechanism policy has resulted in “lower toxicity and signalling risk than other trading venues in Europe.” It has certainly proved popular with the buy side, as the venue’s rapid and sustained market share growth demonstrates. It has grown from 1% to 4.5% of pan-European market share over the past three years.
What Are the Arguments in Favor of ALA Mechanisms?
We have now examined the role of latency in modern markets and several methods by which latency can be levelled across participants. Now we must ask: Is this a desirable aim? Is the effect of latency arbitrage positive or negative for long-term investors? In this section we’ll examine the core arguments in favor of ALA mechanisms.
1. They reduce the indirect operational tax on end users of markets.
If raw speed is the determining factor, any liquidity provider that is systematically outpaced will consistently get picked off, as the fastest arbitrageurs observe quotes moving on one venue and race to hit quotes on another venue a few milliseconds before the liquidity provider receives the same market data and can react. The end result is that liquidity providers may be forced into an expensive arms race.
This is a classic prisoner’s dilemma wherein participants are commercially obliged to participate in a negative-sum activity due to the participation of others. Liquidity providers are not charities and the significant operational expenditure incurred in becoming or remaining low latency – always relative to other participants and therefore relevant even at increasingly diminishing timescales – is ultimately passed on to long-term investors. The transmission mechanism for this is typically as follows:
2. They will lead to tighter pricing and deeper books for end users.
End users are hedging genuine exposures or making long-term investments and not reacting to millisecond-level external events, unlike latency arbitrageurs. Any market maker on an all-to-all exchange has no idea with whom it will trade; it gets a mix of latency arbitrageur flow and regular end-user flow.
The end-user flow is thus subsidizing the latency arbitrageur flow because the spreads charged on a venue are determined by the average quality of flow on the venue. An ALA mechanism normalizes market data transport across all participants: If a market ticks in Chicago and a latency arbitrageur is able to ship that data over to New York (before you can), the speed bump will give a liquidity provider the opportunity to see and incorporate that tick before the latency arbitrageur can pick off its stale price.
ALA mechanisms make it harder for latency arbitrageur taking strategies to perform latency arbitrage on liquidity providers and thus encourages market makers to quote tighter and in larger size to compete for and attract more flow from end users whose orders stem from genuine economic exposures rather than intermarket races.
3. They reduce barriers to entry and encourage competition.
If raw speed is a prerequisite for success in liquidity provision, any participants – including new entrants, which cannot afford such expensive infrastructure – cannot compete and will logically withdraw. We’ve known this since the 1970s.
This is detrimental, as such liquidity providers may well have risk absorption appetite, as well as unique pricing and time horizons. Removing these resting limit orders from the market entirely (because they systematically get picked off each time a related market moves) reduces valuable liquidity.
Regional banks in FX are a good example: They are an important source of liquidity to the interbank market because they perform hedging on behalf of corporate end users that have natural resting interest at certain price levels.
4. They increase diversity and reduce systemic risk.
Latency-sensitive markets tend to have heavily concentrated market share among a small group of extremely fast participants.
This leads to systemic risk, as a small number of HFT firms have limited risk absorption capabilities in relation to their outsized market share, and the failure or operational interruption, even if brief, of such an entity would have a disproportionate adverse impact on the market and liquidity relative to its size.
Reducing the focus on minor speed advantages encourages more competition and a wider group of participants which will deepen the risk absorption capacity of the overall market.
What Are the Arguments Against ALA Mechanisms?
Not everyone is in favor of ALA mechanisms, so we should review the counter-arguments – again, through the prism of their effect on long-term investors. We have tried to compile an exhaustive list, engaging with the strongest form of each argument.
1. This is just like the ‘last look window’ in FX.
This is an erroneous and disingenuous argument, typically made by certain participants who conflate two entirely different topics.
With last look, the liquidity provider knows about an incoming order, even if it is not ultimately filled, and can themselves choose whether to accept this order. This is highly problematic since it leaks information, and “last look” has a deserved bad reputation. For example, bad actors may engage in the practice of “pre-hedging,” which is performed as follows:
With a speed bump, on the other hand, a neutral venue determines the match – the liquidity provider has no choice at all since its quotes are firm – and the liquidity provider of course has no knowledge of any orders that miss. Self-evidently, the information leakage associated with last look does not occur and harmful practices such as pre-hedging would remain impossible.
2. Any additional complexity is bad.
All being equal, simpler is better since end users and their agents tend to react to complexity and change less efficiently than specialized high-frequency traders.
The proliferation of order types in US equities is a good example of complexity harming long-term investors. End users simply cannot devote whole teams to study each order type and are therefore disadvantaged when placing orders, relative to HFTs, some of which may even support the increased complexity as they are able to exploit more edge-case scenarios. It would be ironic for participants that have contributed to the proliferation of US equity orders to object to speed bumps on the grounds of complexity!
As a principle it is therefore entirely reasonable to aim for simplicity, but this must be considered alongside the benefits of innovation to long-term investors. It is worth noting that the existing effort of trying to measure latencies and jitter and optimally splitting orders across multiple venues is far more complex than any proposed ALA mechanism and that long-term investors appear extremely comfortable trading on venues with ALA mechanisms today.