In the Second World War, the UK and US used “Huff-Duff”–high frequency direction finding–to determine the location of U-boats by triangulating their radio signals. In modern finance, “high frequency” is all about high frequency trading, but it is a form of Huff Duff; high frequency traders are attempting to determine the direction of prices by homing in on signals present in high frequency financial price data.
Financial Huff-Duff–or “HFT,” for “High Frequency Trading”–is currently a source of intense controversy. This type of trading allegedly accounts for billions of dollars of profits by big trading houses that invest vast sums in high power computers co-located on exchanges, and sophisticated algorithms designed by equally high power quants and programmers. There is widespread concern that HFT is fundamentally unfair, and allows the big houses to take advantage of other market participants. There is also concern that this type of trading is nothing but an arms race that throws substantial real resources for a merely relative advantage; that is, it is rent seeking that does not improve resource allocation.
HFT does provoke numerous interesting questions. A back-to-basics approach might help light on some of these issues. These thoughts are preliminary, but try to apply some basic finance concepts at a high level to get a better grasp on the costs and benefits of HFT.
Specifically, I think it is valuable to use market microstructure theory to analyze HFT as a form of automated market making. (Historical sidenote: Fischer Black raised the possibility of automated “specialists”–i.e, market makers–over 40 years ago. Another testament to how forward looking Black was, and how wide-ranging his mind was. Although the Black-Scholes model is his most well-known contribution, he made many others as well, all by applying in a very careful way basic economic and finance principles to very complex problems.)
It is well known that market makers face a fundamental problem: they are vulnerable to being “picked off” by other traders with better information. That is, market makers face an adverse selection problem that imposes losses on them when they trade with the better informed. Thus, market makers attempt to condition they prices at which they are willing to buy or sell–their quotes, or bids and offers–on the best information they have available, and in so doing incorporate a margin that compensates them for the losses they expect to incur by trading with the better informed.
One way market makers obtain information is by examining the order flow. Order flow can communicate valuable information. Indeed, one of the advantages of being an NYSE specialist was preferential access to order flow information.
One thing that HF traders do is apply algorithms to try to extract information–signals–from the vast amounts of orders flowing through electronic markets. Some information from the order flow in one stock is useful in determining the value of that stock. The information in the order flow in one stock may also be useful in determining the value of other, related stocks. For instance, information in the order flow on ExxonMobil might help predict the value of Conoco stock.
By trolling through vast amounts of order flow and quote information very quickly, and applying algorithms that have proven successful in extracting information signals from this torrent of data, HF traders enter or alter quotes into the marketplace. Some of the orders are entered in a single market. Some are entered into multiple markets.
That is, they do what human market makers do, but just a lot faster. They extract information from order flow, and use it to adjust their quotes. They are, in essence, the automated specialists that Fischer Black described in the 1960s. It just took 40+ years for computing technology and the automation of markets to catch up to Fischer’s vision.
Speed can be very important in this regard, as one of the major risks that market makers face is the “stale quote” problem. A market maker that enters a quote and doesn’t change it when new information arrives is at risk of having that order executed at a very unfavorable price. Manual market makers can be distracted, or have limited ability to collect and process information. Thus, they are vulnerable to being picked off. They adjust for this possibility by quoting wider spreads–higher offers, lower bids. The quicker they can react, the tighter the spreads they can quote.
Automating the process of updating quotes effectively eliminates the distraction/cognitive limit/stale quote problem.
To the extent that processing vast amounts of information, and applying sophisticated algorithms also allows HF traders to identify informed traders more reliably, it also reduces their vulnerability to adverse selection. I am on less certain ground here, not knowing with certainty whether HFTs can do this, but identification of informed orders and taking that into account when deciding whether to trade against those orders or the price at which to trade against those orders, would help separate the informed from the uninformed orders.
In sum, HFT allows market makers to collect more information, process it more quickly, and change quotes rapidly in response to new information. To the extent that HFT algorithms can also identify informed orders more reliably, it also allows them to avoid the adverse selection problem. All of this should tend to permit them to quote tighter markets than manual market makers who cannot process as much information as rapidly.
How does the entry of HFT affect the welfare of other traders? Well, for sure it hurts market makers who do not make the same investment in hardware and software to extract information from order flow. The marginal cost of HFTs is lower than the marginal costs of market makers who do not have the same information. The competitive process will tend to result in the elimination of old-school, LFT market makers, and the dominance of HFTs. Entry of HFTs will occur until their investments in hardware, software, and people earn a competitive return. That is a natural competitive process, analogous to automobiles leading to the drastic decline in the saddle-making business. Tough luck for saddle-makers, but not for society at large.
Privately informed traders may also suffer as a result of HFT, especially to the extent that HFTs can better distinguish between relatively informed and uninformed orders. This is not necessarily a bad thing, though, because privately informed trading can be a rent seeking activity. That is, people invest in getting information for the purpose of extracting a rent from somebody else. (Much informed trading in market microstructure models is in fact parasitic if it is costly to collect information.)
This is a hard call, though, because to the extent that HFT reduces the profit of informed trading, it would tend to reduce incentives to collect information, making prices less informative. It is nigh on to impossible, however, to determine just what the social value of that information is. Unless the information embedded in prices as the result of informed trading improves the allocation of resources (e.g., companies make better capital investment decisions because they can extract information from stock prices), informed trading is not socially valuable. The problem is, no one has, and I would argue no one can, quantify the value of that information.
This also relates to a widely-heard defense of HFT–that it makes markets more informatively efficient faster. The use of real resources to speed the incorporation of information into prices is valuable only to the extent that this better information leads to better resource allocations. It is hard to see how speeding the incorporation of information into prices from minutes into seconds into milliseconds into microseconds would really improve resource allocation. Thus, I find this defense unpersuasive. Indeed, by exploiting information that is in the public domain (nobody owns the price and quote information), and making money off it, HFT may have a rent seeking component as well. That is, that information may not be socially valuable, but it is privately valuable to the party who can find it and act on it first. Spending resources on computers, software, and people to collect that rent is wasteful. Thus, there may be a rent seeking component to HFT.
This means that from a rent-seeking perspective, it is difficult to determine whether HFT is a plus or a minus. It can discourage some sorts of rent seeking activity, but may involve a rent seeking element itself. Moreover, this rent seeking waste must be balanced against the potential efficiency gains that arise from deterring other parasitic informed trading, and reducing the costs of making markets.
Relatively uninformed investors should benefit from HFT, especially if there is substantial competition between HF traders. To the extent that HFT reduces market making costs, it narrows spreads and increases market depth, both of which redound to the benefit of uninformed “liquidity” traders, such as those trading to make portfolio adjustments, or because of cash needs.
Put differently, HFT tends to favor price takers, and harm non-HFT price makers.
Market microstructure thinking also sheds some light on one of the most controversial elements of HFT–so called “flash orders.” In a flash order, some HFTs are given an opportunity to execute against some orders before anybody else; this opportunity lasts for very small fractions of a second. That is, the HFT is given an option to trade against an order; if it declines, the order is made available to everybody.
The optionality in this might be troublesome, but it might not be. If HFTs can more reliably identify informed orders, they will pass on the option to trade when they identify an order that is likely to be informed, and let the order go into the market. They will exercise the option when they perceive that the order is unlikely to be informed, or when it is mispriced (conditional on the information the HFT has). This exposes others to a greater adverse selection problem–a larger fraction of the orders exposed to the market at large are informed. This would tend to cause spreads in these other markets to widen.
The equilibrium consequences of this are complex, as it would affect order submission strategies, including the decisions by order submitters as to where to send their orders. My hunch is that it would tend to result in liquidity traders (i.e., the relatively uninformed) sending their orders to venues that permit flash orders. Informed traders would have no incentive to direct orders to these venues, as it would be unlikely that their orders would be executed there.
This segmentation of markets has complicated efficiency implications. I’ve delved into similar types of issues in some of my research, including a paper in the JLEO in 2002 (“The Macrostructure of Securities Markets”) and a working paper called “Third Markets and the Second Best.” “Cream skimming” uninformed orders–and flash trading may represent that kind of trading–worsens the terms of trade for some, and improves it for others. Under some conditions, the net effect can be positive–the gainers gain more than the losers lose. Under other conditions, the net effect can be negative.
Thus, the debate over flash orders is just another incarnation of a very old debate about fragmentation and competition between trading venues. Cream skimming and fragmentation have been criticized for years, but as I’ve shown cream skimming and fragmentation can be welfare improving if there are other inefficiencies in the market (e.g., market power, entry barriers).
The cream skimming aspect comes into play if HFT can more reliably identify privately-informed trades, i.e., if it is a way of mitigating adverse selection by deterring informed trading. If HFT is itself a form of informed trading, i.e., HFTs process huge amounts of data and hence can determine which counterparty orders are over- or under-priced, then flash orders are more problematic. Then they allow those that have made investments in real resources–the ability to trade really fast–to make exploiting information that isn’t really socially valuable. Without flash orders, the benefit of trading against mispriced orders would be captured by somebody, but importantly, by somebody who hadn’t spent as much on hardware, software, and people. The mispricing is a zero sum game; to the extent it is possible to limit the cost of playing that game, it is efficiency enhancing to do so.
Thus, what it comes down to is whether HFT reduces the costs of making markets, most notably by mitigating adverse selection problems, or contributes to adverse selection problems. It is likely that HFT does both. The net effect, therefore, is hard to determine. Showing that HFT has reduced execution costs would tend to favor the argument that HFT has been efficiency enhancing. Isolating the effect of HFT on trading costs is quite difficult, however. As a result, I expect the debate will rage on.
The quality of the debate would be improved, IMHO, however, if those participating in it would adopt a microstructure approach, and focus on the key questions of how HFT affects the costs of market making, and whether it tends to mitigate or exacerbate information asymmetries. It might do either. It might do both. Until you know the mix, you can’t make a reasoned judgment on the subject. Unfortunately, in all the ink and pixels that have been spilled in writing about the subject, these crucial questions have been largely overlooked. Hopefully that will change, and soon.
