Oil Futures Trading In Troubled Waters

A recent working paper by Pradeep Yadav, Michel Robe and Vikas Raman tackles a very interesting issue: do electronic market makers (EMMs, typically HFT firms) supply liquidity differently than locals on the floor during its heyday? The paper has attracted a good deal of attention, including this article in Bloomberg.

The most important finding is that EMMs in crude oil futures do tend to reduce liquidity supply during high volatility/stressed periods, whereas crude futures floor locals did not. They explain this by invoking an argument I did 20 years ago in my research comparing the liquidity of floor-based LIFFE to the electronic DTB: the anonymity of electronic markets makes market makers there more vulnerable to adverse selection. From this, the authors conclude that an obligation to supply liquidity may be desirable.

These empirical conclusions seem supported by the data, although as I describe below the scant description of the methodology and some reservations based on my knowledge of the data make me somewhat circumspect in my evaluation.

But my biggest problem with the paper is that it seems to miss the forest for the trees. The really interesting question is whether electronic markets are more liquid than floor markets, and whether the relative liquidity in electronic and floor markets varies between stressed and non-stressed markets. The paper provides some intriguing results that speak to that question, but then the authors ignore it altogether.

Specifically, Table 1 has data on spreads in from the electronic NYMEX crude oil market in 2011, and from the floor NYMEX crude oil market in 2006. The mean and median spreads in the electronic market: .01 percent. Given a roughly $100 price, this corresponds to one tick ($.01) in the crude oil market. The mean and median spreads in the floor market: .35 percent and .25 percent, respectively.

Think about that for a minute. Conservatively, spreads were 25 times higher in the floor market. Even adjusting for the fact that prices in 2011 were almost double than in 2006, we’re talking a 12-fold difference in absolute (rather than percentage) spreads. That is just huge.

So even if EMMs are more likely to run away during stressed market conditions, the electronic market wins hands down in the liquidity race on average. Hell, it’s not even a race. Indeed, the difference is so large I have a hard time believing it, which raises questions about the data and methodologies.

This raises another issue with the paper. The paper compares at the liquidity supply mechanism in electronic and floor markets. Specifically, it examines the behavior of market makers in the two different types of markets. What we are really interested is the outcome of these mechanisms. Therefore, given the rich data set, the authors should compare measures of liquidity in stressed and non-stressed periods, and make comparisons between the electronic and floor markets. What’s more, they should examine a variety of different liquidity measures. There are multiple measures of spreads, some of which specifically measure adverse selection costs. It would be very illuminating to see those measures across trading mechanisms and market environments. Moreover, depth and price impact are also relevant. Let’s see those comparisons too.

It is quite possible that the ratio of liquidity measures in good and bad times is worse in electronic trading than on the floor, but in any given environment, the electronic market is more liquid. That’s what we really want to know about, but the paper is utterly silent on this. I find that puzzling and rather aggravating, actually.

Insofar as the policy recommendation is concerned, as I’ve been writing since at least 2010, the fact that market makers withdraw supply during periods of market stress does not necessarily imply that imposing obligations to make markets even during stressed periods is efficiency enhancing. Such obligations force market makers to incur losses when the constraints bind. Since entry into market making is relatively free, and the market is likely to be competitive (the paper states that there are 52 active EMMS in the sample), raising costs in some state of the world, and reducing returns to market making in these states, will lead to the exit of market making capacity. This will reduce liquidity during unstressed periods, and could even lead to less liquidity supply in stressed periods: fewer firms offering more liquidity than they would otherwise choose due to an obligation may supply less liquidity in aggregate than a larger number of firms that can each reduce liquidity supply during stressed periods (because they are not obligated to supply a minimum amount of liquidity).

In other words, there is no free lunch. Even assuming that EMMs are more likely to reduce supply during stressed periods than locals, it does not follow that a market making obligation is desirable in electronic environments. The putatively higher cost of supplying liquidity in an electronic environment is a feature of that environment. Requiring EMMs to bear that cost means that they have to recoup it at other times. Higher cost is higher cost, and the piper must be paid. The finding of the paper may be necessary to justify a market maker obligation, but it is clearly not sufficient.

There are some other issues that the authors really need to address. The descriptions of the methodologies in the paper are far too scanty. I don’t believe that I could replicate their analysis based on the description in the paper. As an example, they say “Bid-Ask Spreads are calculated as in the prior literature.” Well, there are many papers, and many ways of calculating spreads. Hell, there are multiple measures of spreads. A more detailed statement of the actual calculation is required in order to know exactly what was done, and to replicate it or to explore alternatives.

Comparisons between electronic and open outcry markets are challenging because the nature of the data are very different. We can observe the order book at every instant of time in an electronic market. We can also sequence everything-quotes, cancellations and trades-with exactitude. (In futures markets, anyways. Due to the lack of clock synchronization across trading venues, this is a problem in a fragmented market like US equities.) These factors mean that it is possible to see whether EMMs take liquidity or supply it: since we can observe the quote, we know that if an EMM sells (buys) at the offer (bid) it is supplying liquidity, but if it buys (sells) at the offer (bid) it is consuming liquidity.

Things are not nearly so neat in floor trading data. I have worked quite a bit with exchange Street Books. They convey much less information than the order book and the record of executed trades in electronic markets like Globex. Street Books do not report the prevailing bids and offers, so I don’t see how it is possible to determine definitively whether a local is supplying or consuming liquidity in a particular trade. The mere fact that a local (CTI1) is trading with a customer (CTI4) does not mean the local is supplying liquidity: he could be hitting the bid/lifting the offer of a customer limit order, but since we can’t see order type, we don’t know. Moreover, even to the extent that there are some bids and offers in the time and sales record, they tend to be incomplete (especially during fast markets) and time sequencing is highly problematic. I just don’t see how it is possible to do an apples-to-apples comparison of liquidity supply (and particularly the passivity/aggressiveness of market makers) between floor and electronic markets just due to the differences in data. Nonetheless, the paper purports to do that. Another reason to see more detailed descriptions of methodology and data.

One red flag that indicates that the floor data may have some problems. The reported maximum bid-ask spread in the floor sample is 26.48 percent!!! 26.48 percent? Really? The 75th percentile spread is .47 percent. Given a $60 price, that’s almost 30 ticks. Color me skeptical. Another reason why a much more detailed description of methodologies is essential.

Another technical issue is endogeneity. Liquidity affects volatility, but the paper uses volatility as one of its measures of stressed markets in its study of how stress affects liquidity. This creates an endogeneity (circularity, if you will) problem. It would be preferable to use some instrument for stressed market conditions. Instruments are always hard to come up with, and I don’t have one off the top of my head, but Yanev et al should give some serious thought to identifying/creating such an instrument.

Moreover, the main claim of the paper is that EMMs’ liquidity supply is more sensitive to the toxicity of order flow than locals’ liquidity supply. The authors use order imbalance (CTI4 buys minus CTI4 sells, or the absolute value thereof more precisely), which is one measure of toxicity, but there are others. I would prefer a measure of customer (CTI4) alpha. Toxic (i.e., informed) order flow predicts future price movements, and hence when customer orders realize high alphas, it is likely that customers are more informed than usual and earn positive alphas. It would therefore be interesting to see the sensitivities of liquidity supply in the different trading environments to order flow toxicity as measured by CTI4 alphas.

I will note yet again that market maker actions to cut liquidity supply when adverse selection problems are severe is not necessarily a bad thing. Informed trading can be a form of rent seeking, and if EMMs are better able to detect informed trading and withdraw liquidity when informed trading is rampant, this form of rent seeking may be mitigated. Thus, greater sensitivity to toxicity could be a feature, not a bug.

All that said, I consider this paper a laudable effort that asks serious questions, and attempts to answer them in a rigorous way. The results are interesting and plausible, but the sketchy descriptions of the methodologies gives me reservations about these results. But by far the biggest issue is that of the forest and trees. What is really interesting is whether electronic markets are more or less liquid in different market environments than floor markets. Even if liquidity supply is flightier in electronic markets, they can still outperform floor based markets in both unstressed and stressed environments. The huge disparity in spreads reported in the paper suggests a vast difference in liquidity on average, which suggests a vast difference in liquidity in all different market environments, stressed and unstressed. What we really care about is liquidity outcomes, as measured by spreads, depth, price impact, etc. This is the really interesting issue, but one that the paper does not explore.

But that’s the beauty of academic research, right? Milking the same data for multiple papers. So I suggest that Pradeep, Michel and Vikas keep sitting on that milking stool and keep squeezing that . . . data Or provide the data to the rest of us out their and let us give it a tug.

• Become A Funded Futures Trader