# Can the futures market's open interest predict commodity, treasury, and equity returns?

I came across this article and became curious. Can the futures market's open interest really predict market action?

-
I think that you misread the article. It says "open interest" (i.e. the number of contracts outstanding), which is totally different than the market opening. –  Shane Mar 20 '11 at 0:14
Thank you for your comment. I rephrased my question. –  Ralph Winters Mar 20 '11 at 0:28
@Ralph Your question is still horribly misguided. Open interest (as mentioned in the cited article) is the number of unsettled contracts. It does not refer to the opening price. If you're curious about pre-open indicators, remove the cited article. –  chrisaycock Mar 20 '11 at 2:05
@chrisaycock - I think "horribly misguided" is a bit strong. But obviously the group is not with me on this one, so I'll vote to close. –  Ralph Winters Mar 20 '11 at 3:11
@Ralph - It's neat research that gets the EMH fans a bit angry. So there's a good question here, it just takes a few reads of the paper's intro and conclusion. –  Richard Herron Mar 20 '11 at 3:42

CXO-Advisory investigate the claim of this paper and conclude that evidence indicates that changes in open interest in futures markets are strong predictors of returns for associated asset classes, even after controlling for a number of conventional predictors.

They state that investors may be able to exploit these predictive powers via tactical asset class allocation but nevertheless warn that the study does not explicitly translate this evidence into a realistic trading strategy.

-

Maybe a better question title is "Can futures market open interest predict commodity, treasury, and equity returns"?

I saw this paper in an earlier form and it still baffles me. Superficially, it makes sense that price*quantity holds more information than just price when quantity can change quickly (i.e., outstanding futures contracts changes more quickly that outstanding equity shares). Open interest is the dollar value of outstanding futures contracts, so there are two sides to these contracts, and of course each side thinks that they're getting the better deal. So it's not really clear that this should be a bull or bear indicator. If the price component dominates, then this could just be a CPI precursor, which would drive the returns in treasuries and equities.

Also, the open interest term is a rolling average, which makes it highly autocorrelated. On this they regress treasury returns, which are also autocorrelated, so there could be some spurious regression in there. But Hong and Yogo are top of the field, so there's most likely something I'm overlooking and I'm interested in your thoughts.

After I saw this paper, I tried the same thing in equities. I got data from the OneChicago single stock futures market and tried to use open interest and open interest growth to predict equity returns. I couldn't find anything more powerful than short term reversal. The OneChicago single stock futures market has only been around since the mid-2000s and still has relatively thin trading. Maybe in a few more years there will be enough volume to get better tests?

So, I don't think open interest predicts returns in a profitable way. It can predict returns, but not any better than short-term reversal or other documented mechanisms.

-
I'm thinking that some of the assumptions in the paper need to be tested more, since the hypothesis of open interest does seem counter intuitive. Possible the excess returns were due to the (very) higher open interest setups signaling sharp selloffs in the correct directions, and their may also be more of a sampling bias involved, but it's hard to tell –  Ralph Winters Mar 20 '11 at 22:41

My main concern when I read this paper was that they appear to not be lagging the CFTC data appropriately. The open interest data is reported with almost a week delay. They are using 1yr smoothing of 1mth changes, so that will mitigate some of the look ahead bias, but I think the paper is untradeable.

-

I just skimmed the paper, but it looks like Tables 6 through 10 hold the "predictors". Notice the R^2 values. I assume that their R^2=3% is my R^2=0.03. If that's true, I would guess that trading frictions would dominate any of their predictions.

-
Your contention that low $R^2$ implies high trading frictions is simply untrue. The size of the effect (which, when they say it is "economically large," they mean it is greater than typical trading costs) is more important. Furthermore, 3% is actually fairly high for a regression involving a single instrument on the LHS. –  Tal Fishman Jul 22 '11 at 20:29
@sheegaon, I didn't say low R^2 implies high trading frictions. The level of trading friction exists regardless of R^2. What I said was that an R^2 of 3% was so low that if it has any use at all (which is doubtful), trading frictions will likely overpower any profits. I've tested models with much higher R^2 that turned out to be worthless. You don't have to believe me. Test it yourself. –  bill_080 Jul 23 '11 at 1:38
I've traded models with $R^2$ of under 3% (on individual instruments) and they can make quite compelling strategies if you have enough independent instruments. It is not clear whether Hong and Yogo's paper would lead to a viable quant trading strategy or not, but the $R^2$ alone does not present much evidence either way. Regardless, $R^2$ has virtually no bearing on transaction costs (i.e. trading frictions). –  Tal Fishman Jul 24 '11 at 4:45