HFT Wolves and HFT Sheep
A paper on HFT by Baron, Brogaard and Kirilenko is attracting a lot of attention. They provide shocking evidence: HFT traders make money. Who knew?
What is garnering the most attention is their analysis of profit by counterparty. For instance, Bart Chilton was shocked to find that passive HFT traders earned a $5 per contract profit on average when trading with small traders.
The numbers are interesting and all very nice, but the interpretation is far more equivocal than the authors-and HFT critics like Chilton-would have you believe.
One issue is “how big is big?” or “big compared to what?” Is the per contract profit supposed to be zero?
The paper reports eye-popping Sharp ratios, which suggests that HFT earns abnormal profits. But great caution is required. This Sharp ratio only takes into account the authors’ estimate of the capital required to support the HFT positions: given the rapid mean reversion of positions (especially for “passive” HFTs), this capital can be quite modest. But HFT also requires a substantial investment in systems, software and human capital. And data! A huge cost. Arguably the trading capital is the smallest portion of the capital required to setup an HFT and keep it running. Leaving those investments out of a Sharp ratio calculation renders the number utterly meaningless. The fact that these are numbers from a single market and HFT traders usually trade in many markets also makes it impossible to know what the profitability of HFT firms is.
Indeed, given that there is free entry into HFT, why wouldn’t we expect the marginal entrant to just earn the market rate of return on all capital employed? If a study that takes all the relevant costs and investments into account finds that the marginal HFT trader earns abnormal returns, the appropriate response would be to identify the source of the entry barrier that gives rise to this rent, and attack that.
Another issue of interpretation relates to defining the but-for. That is, what is the alternative to which HFT is being compared? An electronic market without HFT? Floor markets? Consider the latter. There is no doubt that floor traders-locals-reaped profits in the same way as the HFT traders in the Baron et al analysis. Some were “passive”-quoting bids and asks, and supplying liquidity and trading against order flow. Others were “active”-speculating on price movements, and always alive to picking off stale limit customer limit orders. Some did both. Given that the per contract profits reported in the Baron et al paper are fractions of a tick, it’s almost certain that floor traders earned profits as large or larger than they report for HFT, particularly if you leave out important cost/capital components (e.g., the opportunity cost of a seat).
When it comes to floor trading, we know that the marginal trader was earning a rent. Know as in metaphysically certain. How do we know that: seat prices were positive, and frequently very large. (A paper I published in 1999 shows that Q-ratios for exchanges calculated using seat prices were astronomical. This provides clear evidence of rents.) The seat price is the capitalized value of the abnormal profits earned by the marginal floor trader. And we know the entry barrier that gave rise to these abnormal profits: exchange limitations on the number of memberships.
So that would be an interesting comparison. The (appropriately calculated) profit of the marginal HFT firm vs. the marginal profit of floor traders (as measured using seat prices).
The paper also reports substantial skewness in the profitability in HFT. But there was substantial skewness in profit on the floor too. The Tom Baldwins and Harris Brumfelds made huge money, a lot of other traders did quite well, and a very large fraction-arguably a majority-were on breaking even and at risk of getting blown out. And this isn’t a recent phenomenon. In the 19th and early 20th centuries, titans like Old Hutch made substantially more money than the run of the mill pit trader.
Which brings me back to my constant refrain: what’s really all that different about HFT? The functions and behaviors of market participants are pretty much static. We have uninformed traders and informed traders. We have liquidity suppliers and liquidity demanders. We have opportunistic traders looking to take advantage of a time-and-space advantage, and the ability to respond sooner than others (thereby profiting at the expense of the slow). The functions haven’t changed: just the technology for carrying out those functions has changed.
I’d wager that an analysis based on exchange street books from the floor days (which report price, volume, and trader type by CTI indicator) would produce very similar results to the HFT study’s, although the inability to observe the bid-ask on the floor makes it difficult or impossible to break down members trading on their own account (CTI1s) into “active” and “passive” categories in the same way Baron et al do. I am highly confident that such a study would find that: CTI1s made per contract profits as large or larger than HFT traders; CTI1s profited primarily in trades with CTI4s (customers), and that there was variation in the per contract profit depending on whether the CTI4 was big or small; that floor trader Sharp ratios based only on the capital required to support their positions would be very large; and that there was substantial skewness in the profitability of CTI1s.
Perhaps the most interesting aspect of the study is the distinction between aggressive and passive HFT traders. It is very hard to criticize passive HFT, because they are clearly providing liquidity (although no doubt Bart will find bad things to say). The activities of aggressive HFT-HFT firms that are almost always hitting bids and lifting offers-could well be less salutary and perhaps inefficient.
Aggressive firms could be creating adverse selection. One aggressive strategy could be to invest substantial real resources into scooping up huge amounts of public data, analyzing it faster than anybody else, and trading faster on it than anybody else when they recognize that quotes don’t reflect the information. Some of this information could be obtained by scraping news and social media sites. Some of it could be generated by statistical arbitrage: analyzing vast quantities of data to identify pricing relationships across multiple instruments; identifying times when prevailing quotes across these markets deviate from these relationships; and buying cheap and selling rich when such deviations are found. To the extent quote submitters-including passive HFT firms-do not have this information, the aggressive firms will be better informed, profit at the expense of quote setters, and thereby cause liquidity suppliers to widen their quotes.
If that’s all there is to it, aggressive HFT would be inefficient. Real resources are devoted to earning a rent, and this rent seeking distorts prices (notably the price of liquidity), thereby inducing others trading for risk-related reasons to trade too little.
But price discovery is the flip side of adverse selection. Informed trading drives the price discovery process. The question becomes what is the value of the price discovery provided by aggressive HFT. Impossible to answer quantitatively. I would surmise that the conventional wisdom is that the value is small. Why do we need to have prices adjust by a fraction of a cent to reflect information a fraction of a second sooner? What real decision (e.g., investment decision, planting decision) is going to change if a price embeds slightly more information a blink of an eye sooner? I am similarly skeptical about the value of speeding up price discovery, and therefore am quite willing to accept that some aggressive HFT is opportunistic rent seeking, and hence inefficient.
But even granting that, what are you going to do about it? Is there a way of restricting rent seeking HFT that does not also burden beneficial HFT, for instance by imposing costs on passive HFTs that supply liquidity?
It’s not obvious that there are discriminating ways of pricing trading services in ways that reduce opportunistic trading without also reducing liquidity supply and trading for risk shifting reasons. Tinkering with maker-taker fee structures, or tinkering with price schedules more generally seems to be the most sensible way to do it. The question becomes: who should do the tinkering? Exchanges presumably have the strongest incentives to drive out the opportunistic wolves that feed on the sheep, and the best information to do it. I therefore don’t see a strong case for external regulation intended to reduce rent seeking HFT. Let the exchanges handle it.
It is obvious that some of the regulations that have been proposed would make things far worse, rather than better. Time-in-force rules are probably the best-or worst-example of that. Requiring quote submitters-including passive HFT firms-to keep quotes in force for some minimum time period makes them incredibly vulnerable to the opportunistic HFT firms. The aggressive HFT firms feed off of quotes that aren’t adjusted to reflect the information they produce by stat arb or whatever. Time-in-force rules make it harder to adjust quotes, giving more targets of opportunity to aggressive HFT traders. Prediction: TIF rules will increase aggressive HFT trading; reduce passive HFT trading; lead to aggressive HFT representing a higher proportion of HFT overall; and lead to wider spreads.
And ironically, these effects will increase the profits HFT firms earn at the expense of non-HFT traders. So, Bart: if you think the profits that HFT earns off the little guys are unconscionable now, be careful what you ask for. Some of the restrictions on the “cheetah traders” that you advocate will make those profits even more unconscionable.
Put differently: Time-in-force rules don’t constrain aggressive traders who are taking quotes instead of making them. They only constrain those making quotes and thereby make them vulnerable to the takers. This makes sense how, exactly?
In almost any human endeavor, there is opportunism and rent seeking and inefficient behavior that goes on side-by-side with beneficial, wealth increasing conduct. We don’t have the information or sufficiently discriminating and cheap deterrents to eliminate bad conduct altogether.
When it comes to any financial trading-old school floor trading or HFT-dominated electronic trading-rent seeking trading will occur. When evaluating HFT generally, and restrictions on HFT that do not discriminate between the virtuous and vicious forms of the activity, you have to take the good with the bad and see which predominates. The vast bulk of the existing empirical evidence shows that HFT is associated with better market quality in terms of spreads and depth. So even though some HFT is almost certainly predatory, the effects of the predatory trading are more than offset by the efficient, wealth-increasing kinds. Indiscriminate regulations that constrain both types of HFT are therefore highly objectionable. Not a good idea to shoot a prize bull in the head because it has a parasite that available drugs won’t kill. But that’s what those who get all hot and bothered about aggressive-and arguably parasitical-HFT threaten to do with indiscriminate regulations that would impede all HFT.
What about minimum spread rules? Once compensation for the liquidity providers is earning the bid-offer spread. By having spreads be so narrow, I can create liquidity but not reap the benefit because some guy jumped in front for a fraction of a penny. If we want to help the “good” HFT, wouldn’t making their expected returns higher through bigger spreads help?
Comment by Highgamma — December 8, 2012 @ 4:44 pm
@Highgamma-IOW, you think tick sizes are too small. This is actually an area of active regulatory inquiry. And interestingly, ICAP increased tick size precisely to address perceived problems with opportunistic HFT. Another illustration of how operators of the execution venues might have the right incentives to get it right.
I can see some offsetting effects. On the one hand, having to cross a bigger tick will mean that opportunistic HF traders would find it costlier to take liquidity, and would trade less often holding trade size constant. That’s the upside. But you can’t hold trade size constant. A wider tick will increase depth at the inside quote, perhaps allowing the aggressive trader to trade in larger size, earning a higher profit when he does trade.
My guess is that these effects would offset, making it a wash.
Thanks, I was waiting for you to hit this. Kirilenko paints a picture of HFTs taking money from unprotected market participants. I always thought that the market makers were selling immediacy and not getting paid very much for their efforts. Isn’t that a fair trade?
Also, If a bid offer is sitting at a tick wide with little or no action, why don’t we regard the HFT who lifts the bids as providing liquidity. Doesn’t real liquidity include the velocity of trade as well as the depth of book?
Comment by Jerry Salzman — December 10, 2012 @ 10:34 am
I’m puzzled by the easy conclusion that the HFT traders and the floor traders of old were earning rents. To quote your point at the top of the post,
“the appropriate response would be to identify the source of the entry barrier that gives rise to this rent, and attack that.”
Why would rents have persisted in the absence of entry barriers in the floor trader days, and why would there be any today? What were the barriers to entry then, and what are they now?
Comment by emerich — December 11, 2012 @ 8:09 am
OK, I see that you say the limited number of seats was the source of the rents in the old days; still not clear why any would persist today.
Comment by emerich — December 11, 2012 @ 9:36 am
@emerich-open question whether the marginal HFT is earning rents. The Baron et al analysis doesn’t answer that question b/c they don’t observe the entire investment of HFT traders.
I don’t believe that there is an entry barrier at the trader level today. Back in the day, when exchanges were member-owned cooperatives, they rationally limited membership in order to enhance their wealth. The relevant entry barrier was the liquidity network effect that made it virtually impossible for a new exchange to enter to compete away the rents. I formalize this argument in several papers, including one in JLEO in 2002. The basic idea is that due to the network effect, a suboptimally small coalition of traders can form an exchange with no fear of competitive entry. Because the coalition is suboptimally small, its members earn rents. The seat price-the cost of joining the exchange-is the capitalized profit of the marginal member. In the formal model, that is the most risk averse member of the exchange. Inframarginal traders earn rents
Now, for-profit exchanges have no incentive to limit entry of traders. Quite the reverse. The network effect still works for futures exchanges, because unlike in securities markets regulators haven’t socialized order flow (as the SEC did with RegNMS). So CME or ICE don’t really have to worry about competitive entry. They now earn the rents, by charging supercompetitive trading fees.
I would therefore predict that when all costs are accounted for, the marginal HFT trader will not earn a rent. Inframarginal ones will. But in the floor days, the marginal trader earned a rent.
Put differently, the liquidity network effect gives exchanges market power, but old style and new style exchanges exercise it in different ways. Old school floor exchanges exercised it by limiting membership. New school electronic exchanges exercise it by charging supercompetitive fees.
One Question, One Answer
Question: You mention, “The vast bulk of the existing empirical evidence shows that HFT is associated with better market quality in terms of spreads and depth.” Can I trouble you for a couple of citation of pieces you like in this area–particularly studies that measure market depth?
Answer: You begin your sixth paragraph with, “Indeed, given that there is free entry into HFT, why wouldn’t we expect the marginal entrant to just earn the market rate of return on all capital employed?” In part, the answer to your question is informed by the answer to this question: In any competitive trading market, how many market agents can be fastest?
Comment by Charlie Wolin — December 11, 2012 @ 5:28 pm
Thanks, very interesting reply.
Comment by emerich — December 11, 2012 @ 6:00 pm
Interesting. but far more why should traders not make money ?
Comment by Sara — May 28, 2015 @ 9:47 am