Industry-Funded High-Frequency Trading Study Falls Short

I recently published an article in response to a study of high-frequency trading (“HFT”) by Professor Charles M. Jones of Columbia Business School and an opinion piece he published simultaneously in Politico. My article focused on the funding of the research by Citadel LLP, a major HFT user. It also pointed out broad concerns about the study, which asserts that computer-based algorithmic trading provides substantial net value to the economy.

Keeping in mind the Professor Jones’ funding source, it is useful to look into the studies on which the professor relies (his independent work was limited to interpretation). These should be compared with other academic work that draws alternative conclusions.

The studies that he cites as supportive are generally based on conventional views of the efficiency of markets. These studies identify lower trading costs that have been experienced during the years that HFT has emerged as a dominant force in the equities and commodities markets. He concludes that HFT provides liquidity (buying and selling interest) by generating large numbers of quotes to buy or to sell on which investors can rely for execution of desired trades, thereby reducing transaction costs.

Of course, exchanges and other trading venues have adopted innovations during the same period that also reduced transaction costs. And many institutional investors fled to “Dark Pools,” trading venues designed to allow them to hide from predatory HFT, suggesting that they did not value HFT very highly. But we will ignore these factors for now.

The fundamental problem with the supportive studies is that traditional methods of measuring market efficiency are not very useful for evaluating the effects of HFT, a decidedly untraditional business practice. Traders that use HFT deploy large numbers of price quotes, the vast majority of which are cancelled well before an investor executes a trade against them. The primary reason is to seek information on the motivations of others who are active in the markets and sometimes to create the impression that market prices are moving in a direction that suits the tactics being used by the HFT trader. Large numbers of quotes can also be used to slow down the processes of a trading venue so that price differentials among venues can be induced and exploited, a tactic called “quote stuffing.”

As a result, HFT produces quotes that investors can and do execute against, but very often not for the benign purpose of providing liquidity. And even quotes that are available for investors can disappear instantaneously as dictated by the algorithms driving the high-speed computers. This creates an appearance of liquidity that is, in fact, illusory. As I point out in my recent research paper, this is certainly not the kind of liquidity that is reliable and stabilizing to the markets.

One study (Jarrow/Protter) not referenced by Professor Jones created a model of a trading market. In the model, there are two types of participants: ordinary traders and high-frequency traders. Ordinary traders behave predictably, pursuing common goals as in a conventional market model. The high-frequency traders’ behavior is based on optionality, observing and reacting to information rapidly, before ordinary traders can act. The HFT traders act in concert in response to new information, creating price abnormalities that they then exploit. The abnormalities and exploitation are not planned, but are consequences of rational use of the HFT technique. Thus, the model does not assume the use of HFT predatory tactics, though the study acknowledges that they exist. If it had, its conclusions would be more extreme. The study concludes that HFT “can create increased volatility and mispricings (deviations from fundamental value) that they exploit to their advantage… [and] to the disadvantage of ordinary investors.” 

Another study (Zhang) that Professor Jones does consider in his review of the literature concludes that volatility and mispricing (as predicted by the Jarrow/Protter model) can be observed in market data. Moreover, it cites evidence that HFT impedes the dissemination in the market of information relevant to the fundamental (intrinsically legitimate) price of securities. This study is extremely inconvenient for Professor Jones’ conclusions.

There is a saying among social scientists that, if you can’t credibly challenge the conclusions or the mathematics, challenge the data. Professor Jones makes the case that the 14-year period (1995-2009) used is an overly broad time frame, and therefore a blunt measure. More importantly, he criticizes Zhang’s measure of the intensity of HFT activity. Zhang excluded conventional investor activity (which, unlike HFT, could be identified from reliable sources) from aggregate trading, assuming that the balance is a reasonable approximation of HFT activity.

I contacted Professor Zhang at Yale for his response. He points out that he employed a “triple-differences” approach that uses two analyses in addition to the 14-year time period and that the results are in concert. He also points out that his measurement of the level of HFT is a legitimate methodology and is the only practicable way to estimate HFT activity for the period. “I agree with him that my measure captures short-term trading that include non-HFT, but it should be driven by HFT given its huge volume. In addition, this is just a measurement error issue, and measurement errors tend to add noise to the regression and attenuate the coefficients. They cannot explain the results.” Zhang’s point is that the trading activity remaining after screening out ordinary investor activity is overwhelmingly HFT and that the non-HFT activity in the data set could not have changed the conclusions. His conclusion is logical: he used the data to show when HFT was more intense and heightened trading activity overall should correlate with HFT intensity.

A third, very recent study (Filimonov/Bichetti/Maystre/Sornette) that was not considered by Professor Jones examines HFT in the commodities markets. It is based on intensely complex mathematical proofs that reinforce the Jarrow/Protter and Zhang conclusions. It describes how HFT creates a condition in which one trade leads to a series of trades in a process described as “branching.” It further describes how branching has two important direct effects. First, it increases price volatility. Second, it makes the pricing mechanism less efficient. Prices are affected by market activity itself rather than information that is relevant to the fundamental value of commodities. The study estimates that 60-70 percent of commodity price changes over the period from the mid-2000’s to 2012 are “due to self-generated activities rather than novel information.”

Further, as this internal force affecting prices increases, “[t]he susceptibility to external shocks diverges similarly. All these singular behaviors (in the mathematical sense) point to a growing instability of the system as the branching ratio increases.” In other words, markets are more susceptible to price bubbles and crashes and the risk of financial crisis increases.

The study of HFT calls for far more imaginative analysis than a comparison of the spread between quotes to buy and quotes to sell. For one thing, as pointed out in my recent paper, the spread between bid and ask quotes is a measure of transaction costs for individual trades and tells us little about the effect on the economy of HFT and similar activities. Policy should be focused on that economic effect, not on the narrow concept of transaction costs.

You may also like