The Big Pairs Debate
Thursday, December 8, 2011 at 12:53PMMichael and Jennifer have different views on a critical issue regarding statistical pairs/spreads. The question is whether the two companies need to be similar (in the same sector or industry) or not.
The basic pillar of spread trading is that the two stocks are tied together with some sort of force, and a deviation from historical levels is an opportunity to bet on a return to those levels. Without such a force, the spread may not ever return to past levels.
Michael’s view is that all stocks are subject to the Law of One Price (LOP), this is the force uniting all securities. “Ingersoll (1987) defines the LOP as the “proposition…that two investments with the same payoff in every state of nature must have the same current value””(Gatev, 2006), regardless of whether the two companies are in the same or very different businesses. The profit generated by the arbitrageur is a compensation for enforcing the LOP.
Jennifer’s view is that current price is expectation of future cash flows which are impacted by many outside forces. Companies that have similar businesses will be impacted similarly by small changes to inputs like interest rates or commodities prices or consumer sentiment. A deviation in the prices of a bank versus an oil company may be an anomaly, or it may be due to fluctuations in inputs which could take a long time to correct; for similar companies it is more like to be an anomaly we can exploit in the short- to medium-term. Given a relatively short time horizon, it is essential that the two companies have similar businesses.
This is an important issue because requiring the pair to be similar companies means fewer potential trading candidates and therefore less diversification across positions. However, if we open up to pairs of different companies we may introduce more data mining errors.
What do you think? Do you think statistical spreads need to be pairs of similar companies, or can you make money trading statistical spreads with companies in different industries?
Written by Jennifer Galperin. Follow me on Twitter and StockTwits.
In collaboration with Michael Bigger. Follow me on Twitter and StockTwits.
You can learn about trading pairs here.
Reader Comments (1)
I might be overly concerned with datamining errors and time horizon is critical to me, so I might tend to side with Jennifer. I feel like if time horizon and and liquidity is not a concern then the arbitrager is compensated by LOP. But, the same could be said for any pair. Side note; I remember reading about royal shell and how long this "pure" arb deviated from "fair". (http://tiny.cc/1wrxl).
I do think its interesting that if you had a solid thesis and could trade everything (some large number) LOP might come more into play. I'd rather take a million bets with tiny +EV and higher variance than fewer with fatter edge and low variance which might seem counter intuitive. Isn't this kinda what getco does? I'm never going to win speed battles but always looking for more products I can trade and how the risk profile fits in with my portfolio.