This is the second post in a series responding to FCA Insight “Big bucks from small change,” and Occasional Paper No. 50 Quantifying the High Frequency Trading ‘Arms Race’: A new methodology and estimates (Aquilina, O’Neill and Budish, Jan 2020) and related papers (collectively, the “FBA Papers”), in which their authors allege that Central Limit Order Books (“CLOBs”) lead to a $5b “latency arbitrage tax,” increased costs to investors, destroyed social welfare, among other things.

The FBA Papers also suggest matching buyers and sellers of financial instruments solely via discrete time Frequent Batch Auctions (“FBAs”) as a viable and superior alternative to continuous matching market design.

If you haven’t seen Part I of this Series, you may find it on Highbury Associates’ website or on LinkedIn.

Let’s get to the core of the inflammatory assertions from the FBA Papers:

Flawed market design significantly increases the trading costs of large investors, and generates billions of dollars a year in profits for a small number of HFT firms, who then have significant incentive to preserve the status quo.
- Aquilina, Budish and O’Neill Jan 2020

None of this is true. Literally none of it.

Flawed hypotheses and assumptions lead to crazy conclusions

If not wholly explicit, the paper postulates that arbitrage, and specifically “latency arbitrage”, is:

Damaging to markets in general
Materially taxing to the end investor and trading firm liquidity providers
Avoidable in some way
More injurious than the implied solutions they propose

Arbitrage is none of these things. On the contrary, arbitrage is a necessary component of competitive, dispersed and interconnected markets. We live in a 24/7, global and interconnected world, and competitive markets deliver real value (more later).

While arbitrage exists as a result of market inefficiencies, it ensures markets are efficient

Arbitrage ensures fair value in the same instrument across multiple market centers
Arbitrage ensures fair value for fungible instruments, such as Currencies, Equities, ETFs, Index Futures, Commodities and their underlyings within and across market centers
In doing the two above, it also keeps Exchanges and other trading venues competitive and thus responsive to their constituents: cheaper, more innovative and more resilient

Rather than view arbitrage rents as a tax, they should be considered a small and occasional expense paid to those who absorb the costs and risks associated with providing ubiquitous liquidity across the global markets for the convenience of investors and other trading firms. As the FBA Papers point out, these costs rise occasionally and proportionately with volatility, which is a key measure of risk absorption. Competing arbitrageurs ensure momentary shocks are just that… momentary. Market making firms compete to be at the top of the book on price, and in many CLOBs (but not all) are rewarded for arriving there first and remaining there to maintain their priority. This is to the direct benefit of end investors. In order to do this, market makers incur several fixed costs, including exchange memberships, market data, order entry capacity, networking and colocation, highly-qualified personnel, record keeping and compliance and constant development of sophisticated risk management systems. None of those costs are passed along to the end investor. They are the ante to sit at the table for the market maker.

Discussion points: Are market makers extracting a tax, or are they providing a service? If this tax exists, who is paying it? Is it cumulative, or netted out among market makers while they provide their service? Could market makers provide this service if there was no spread to collect, the implicit assumption of the authors?

You won’t find the answers in the papers. I’ll try to address this.

Sniper.. (no Sniping)!

Swiper no swiping!
*Credit: Nickelodeon Animation*

After the Volcker Rule in the United States and other balance sheet restricting regulations were implemented after the 2009 financial crisis, many banks exited their market making businesses. Consequently, today’s market makers/liquidity providers are proprietary trading firms with finite capital to deploy. Many, but not all, deploy low inventory and/or delta neutral trading strategies that rely upon the contractual fungibility of various inter-related securities, derivatives and exchange trade products. Other prop trading firms are able to deploy greater balance sheet, as they mode, absorb and tolerate risk differently. While also providing liquidity, they are performing the act of market making and risk transfer using different edge. This doesn’t make them better, just different. It may make one or another a more compelling counterparty in different use-cases, but that is why agency brokerages exist. In trading, it literally takes all types. Onward...

Meanwhile, in Europe, MiFID II regulation has encouraged trading to migrate to pre-trade transparent, public markets. Liquidity Providers (“LPs”) hold themselves out visibly, giving liquidity takers a free-option on either side of their resting orders. As the authors point out, LPs earn their living collecting part of the spread. If that spread moves relative to other markets or instruments in a multi-legged trade or a position, LPs must be quick in order to “hedge out” risk. LPs of all types, fast and slow, use Immediate-or-Cancel orders as risk management tools, and do so in all conditions. In a fast moving market, what the FBA Authors refer to as a “stale quote snipe” is often not a profit generator, but a grappling hook to prevent a fall as the spread moves.

In any case, this goes a far way to explain why the many active LPs with resting orders require speed to establish or vacate price levels, and are also often the most likely to use Immediate or Cancel (“IOC”) orders, the “sniper’s” weapon of choice.

Discussion points: If spreads are in the 2-4 bps range, to whom is the average 0.42 bps “tax” material (referred to as not “small beer” in the paper)? Is one minute reversion more material to the low margin market maker/LP who is trying to collect the spread, or the long-term investor who is looking to buy and hold for months or years?

I’ll hold your “small beer” while you do the easy math.

Still, it’s not clear that the end investor pays any of this “tax”, should it even exist. Nor do the trading firms, on a net-basis, either. Let’s unpack that now.

“Ohh man!” Then Who is paying the latency arbitrage tax, if it exists?

“These arbitrage rents increase the cost of liquidity provision. In a competitive market, trading firms providing liquidity incorporate the cost of getting sniped into the bid-ask spread that they charge; so there is a positive bid-ask spread even without asymmetric information about fundamentals.”
- Budish, Crampton, Shim 2015

The assertions herein are dubious and confused. Positive bid-ask spreads are required for liquidity provision (on this we agree), and are the premium commanded to reflect risk-transfer service. Period. A race condition snipe is one of many risks LPs manage all day and everyday.

Despite all these risks, the LP is still willing to compete to be top of book, all day and every day, because the LP doesn’t earn a living without:

a positive bid-ask spread and
a trade on both sides to capture that spread

The authors argue that LPs provide less depth and wider bid-asks for fear of getting caught out by snipers. LPs will always command a positive spread, and face the most risk in doing so in anonymous markets. LPs' resting orders act as highly visible “free options” for takers who can come in at any time, with any size, on either side of the book. LPs price the risk of taking on greater inventory from unknown counterparties who can “clear out the book”, and therefore the spread widens with increased risk. This is why LPs, acting as Systematic Internalisers, can offer better pricing in bilateral environments including private price feeds and RFQs.

To be clear, the LP’s incentive is to compete to earn every trade within the risk parameters of their trading strategy. It is not my experience that firms “incorporate the cost of getting sniped in the bid-ask spread they charge”; more likely, firms assume, again, that this is the cost of doing business. You win some, you lose some. Laughlin’s study of Virtu’s pre-IPO disclosures supports this assertion, where he estimates that Virtu’s US Equities trading performs at a 51% win rate, a 24% loss rate, and 25% at scratch. In that study, Laughlin shows how important it is to trade a lot with narrowly positive win rates, and why speed is such an important risk management tool for some in this business:

Probability of profitability based on number of trades per day

Source: Insights into high frequency trading trading from the Virtu Initial Public Offering, Laughlin, 2014

The glass just needs to be 51% full to be buzzin’

Furthermore, even the FCA Insight’s Occasional Paper’s findings also hint strongly that the winners of the “tax” are even more often also the payers of that “tax”, whether sending an IOC to profit, hedge or attempt to cancel their stale quotes:

Race participation is concentrated. The top 3 firms win about 55% of races, and also lose about 66% of races. For the top 6 firms, the figures are 82% and 87%.
- Aquilina, Eric Budish and Peter O’Neill Jan 2020

The assertion that liquidity providers back away and require wider bid-ask spreads is inconsistent with these observations. More likely, firms understand that they need to be “in it to win it”, and use whatever tools available and economically feasible to them to ensure they manage the risk required to a >51% overall win rate all day and everyday, while also ensuring that the magnitude of losses that fall into the losing 49% aren’t severe. In the end, the $5b cumulative latency arbitrage tax estimate (assuming it is accurate, which we shouldn’t do) is shared as profit and loss amongst these firms. The net total is likely far closer to zero.

Not good: a solution in search of a problem, supported by selectively incomplete analysis; the result is some overfit

“Although an analysis of the effect of frequent batch auctions on market stability is beyond the scope of the present article, here we discuss several computational simplicity advantages of discrete-time trading over continuous-time trading. As we note in the conclusion, we think that market stability is an important topic for future research.”
- Budish, Crampton, Shim 2015

I laugh and then get frustrated every time I read that.

It is unclear how eliminating continuous trading in CLOBs and replacing with FBA/anti-CLOBs would work, or in all likelihood, would impair our interconnected, multi-asset, international markets. While acknowledging the relationships between assets and markets, the FBA Papers still make no attempt to investigate their proposal’s impact on market stability. That is to say, how could they operate outside the theoretical academic world and in the real world.

At the same time, they eagerly choose to continue to oversimplify, selectively analyze and extrapolate their analysis, ultimately concluding a global $5b tax (their mid-range estimate) is imposed on the system and its participants. This grabs headlines and provides fodder for those who wish to vilify the markets for political gain, but it doesn’t make markets better. It is untrue. It’s misleading. It’s inflammatory. And it is irresponsible.

Near beer analysis is complicated, lightweight and unsatisfying. we need “stout” analysis

Furthermore, the academic approach applied in Occasional Paper 50 of the FCA Papers is both complicated and lightweight. It’s “near beer” when we deserve something more stout.

It uses data from a single asset class, sourced from 9 weeks of activity from over 4 years ago on the London Stock Exchange in a now-obsolesced tick regime which exists no-where on the planet today.

My mother might assume that the “stock market” is the national exchange; in reality, at the time of the study, the LSE represented roughly 50% of the volume in the FTSE 100 and 250 indices (source: Cboe), while related derivatives were traded in other order books and trading systems near and far from the LSE. We should be cautious about drawing conclusions about the UK market, and even more so to the entire world. Occasional Paper 50 draws a trendline from a single data point, and it’s not even the right data point.

The FBA Papers’ authors assume that concurrent arrival times of IOCs and cancels equate to a single shared strategy, “latency arbitrage”. This is a large leap. While interesting to note that the same firms show up at these times, it’s not a surprise. Without knowing who (and how many) sit behind those orders, and what their trading strategies are, we’ll never know. As mentioned earlier, a quick price move will trigger hedging activity in multi-legged trades. Such concurrent behavior may also reflect the arrival of signals or risk management parameters common to multiple market participants. If trying to visualize this, the same thing happens in a turnover in football; immediately the offense switches to defense, and vice versa. Agility is important. We understand the motivations of the wholesalers in the markets is to maintain a 51% win rate, whether they absorb inventory, they hedge in correlated or contractually fungible instruments or otherwise. Replacing continuous trading on one or “n” markets removes risk management tools and further complicates LPs’ ability to maintain that 51% win-rate.

We shouldn’t be hasty to create more problems when its not clear we understand the problem we are trying to solve and the implications to all players , in well-functioning, complex markets. The conclusions made from this dated, anonymized and limited data set are a stretch given the over-simplified assumptions required to make them, notwithstanding the complexity of math and computing power that the authors used. While the authors show a strong understanding of bottom-up microstructure and the technology at exchanges, they equally lack some of the fundamentals, and common sense, one would expect of their pedigree in performing academic analysis and applied research.

Bottoms-up… but how what about some Top-down (analysis)?

Discussion point: Is the $5b global “tax” on equities trading a reasonable estimate?

We wouldn’t know by reading the FBA Papers. The authors make no attempt to sanity-check the profitability of HFTs in order to triangulate their figures. Thankfully, others have done the research:

"... the bonanza has now ended. Trading firms are struggling to wring profits from the incremental millisecond. Subdued volumes and reduced volatility have shrunken the size of the pie. Exchanges have ratcheted up market data and technology costs for customers. In 2017, aggregate revenues for HFT companies from trading US stocks was set to fall below $1bn for the first time since at least the financial crisis, down from $7.2bn in 2009, according to estimates from Tabb Group, a consultancy. At the same time, the costs of doing business have gone up, including market data, technology infrastructure, etc."
- Tabb Group in the Financial Times, Jan 2018

That's less than $1bn of net total trading revenue in the US equities markets across all trading strategies, not just "latency arbitrage" strategy alleged by the authors. The FBA Papers’ estimates are $2.7b in the US Equities alone. So, no, it’s not adding up.

Discussion point: So where does all that “tax” go?

You win some, you lose some. Maybe it's not a tax after all, and it’s just P&L shared among those firms competing for the top of the book.

Maybe those 6 firms in Europe spend a bit of their own “socially wasteful money” in order to ensure that they can provide liquidity all day for our benefit, contributing to competition among market centers, ensuring fair value in instruments and their derivatives across all market centers.

Do they like spending their profits on this technology, preserving the status quo? Definitely no. But this insurance means they only lose most of the time in race conditions, and when they do lose, they don’t lose everything. At 51% win-rate, and if they do it enough, they have a pretty sound business that works for them and investors who benefit from their liquidity . Win-Win from not losing big. How’s that for “computational simplicity”?

That’s it for now. But I’ll be back again and I’ll try to illustrate the potential market stability trade-offs of abandoning CLOBs and migrating to FBAs.

Cheers!

The $5 billion heist that doesn’t exist - Frequent Batch Auctions, Part II