Pairs Trading: Some Resources

Pairs Trading: Performance of a Relative-Value Arbitrage Rule (Evan Gatev, William N. Goetzmann, K. Geert Rouwenhorst), Review of Financial Studies, 2006.

This is a useful paper for anyone who is interested in the idea of pairs trading. It has a nice historical discussion of the method, and it gives a semi-reasonable --- and purely quantitative --- basis for studying pairs trading. Most pleasing of all, it offers the prospect of some genuine cheese. There are some puzzles that remain with this paper, but read it and see what you think.

Pairs trading (Elliott, van der Hoek, and Malcolm) Quantitative Finance, 2005.

This is a theoretical paper that posits a class of models where mean reversion is "baked into the cake." The main problem I have with the paper is that the proposed models do not confirm to our favorite stylized facts. For example, if you look at the returns on an equity spread (say the returns on a portfolio that is long Dell and short HP in equal dollar amounts) you will handily reject the normality of returns --- yet normality is implicit in the posited model. Equity spreads also exhibit conditional heteroskedasticity and Black's leverage effect which are not reflected in the model.

In email correspondence, John van der Hoek offers a response to this critique (to which I have added bold font at one point to aid the eye):

"Pairs trading is a short term speculation strategy used by hedge funds [see Gatev et al 2006, SSRN=141615], and our paper provides a short term model which is regularly to be updated with new data. With each calibration the mean-reverting level is estimated  and the size of the mean-reversion. It goes without saying that if the calibration indicates weak mean reversion then the pair under consideration would not be engaged in pairs trading.  Thus our procedure is also to be used to identify strong mean reversion if it occurs.  If a pair does not display desired properties from calibration, then trading will not proceed with that pair. The issue of stylized facts is not relevant in the application of what we propose. It would be interesting for your students to apply our algorithm in the way I have indicated and to estimate annualized returns observed."

Thus, we have a bit of a debate. Among other things, this conversation may motivate the general question:

"What do we lose (or gain) by limiting our attention to only those hypotheses or models that conform to the stylized facts of the associated empirical series?"

Navigation: Back to 434 Home, Back to Steele's Home Page.