Pairs trading is a very popular species of stat arb, and works like this. You look for various correlated stocks, like Dell and IBM, Coke and Pepsi, or Exxon and BP. When one goes up while the other goes down, or vice versa, you bet on a reversal. So lets say Dell is up 4% over the past 2 weeks, while IBM is up only 1%, you buy IBM and short Dell. Stat arb in general looks at mean reversion on the sub one month horizon.
I've heard many stat arb players are having a bad month. I think the problem is, when volatilities and correlations change, its difficult to adjust one's models appropriately, so you either under-adjust over over-adjust. I was on a pairs desk for a while, and must say it's a tough way to make money. I think the only way to make money is A) not use too much, so you are focusing on only 20-40 pairs or B) use the strategy in concert with an insanely fast electronic Direct Market Access system. I suppose the latter, a really fast off-the-floor marked making system, combined with a simple stat arb algorithm, is a big part of how Renaissance prints money.
This incredibly simple strategy made lots of money for traders until 2003, when it started to get crowded. I think its a great example of an arbitrage: they exist, with hindsight they are insanely simple, and they are ephemeral. Other strategies I know of that have vanished are equally simple--not go long calendar spreads simple, but simple enough--so I'm generally skeptical of really complicated strategies, and make no apologies for the simplicity of my best ideas. Good ideas are generally simple once you know them, so anything extra fancy that's meant to be practical tends to reflect deliberate obfuscation of an idea whose simplicity belies its stupidity.
Strangely, there was an academic paper on this in 1998 by Goetzmann, Rouwenhorst, and Gatev that never made it into an academic Journal. You could have read this paper and made money for 4 years on this idea. I wonder why it was never published?
Anyway, I hear many Risk Arb players at big shops are getting creamed. It seemed like you make money for 3 years, then give it all back in a couple weeks. Classic mode-mean trade: mode is positive, mean is zero.
I've heard many stat arb players are having a bad month. I think the problem is, when volatilities and correlations change, its difficult to adjust one's models appropriately, so you either under-adjust over over-adjust. I was on a pairs desk for a while, and must say it's a tough way to make money. I think the only way to make money is A) not use too much, so you are focusing on only 20-40 pairs or B) use the strategy in concert with an insanely fast electronic Direct Market Access system. I suppose the latter, a really fast off-the-floor marked making system, combined with a simple stat arb algorithm, is a big part of how Renaissance prints money.
This incredibly simple strategy made lots of money for traders until 2003, when it started to get crowded. I think its a great example of an arbitrage: they exist, with hindsight they are insanely simple, and they are ephemeral. Other strategies I know of that have vanished are equally simple--not go long calendar spreads simple, but simple enough--so I'm generally skeptical of really complicated strategies, and make no apologies for the simplicity of my best ideas. Good ideas are generally simple once you know them, so anything extra fancy that's meant to be practical tends to reflect deliberate obfuscation of an idea whose simplicity belies its stupidity.
Strangely, there was an academic paper on this in 1998 by Goetzmann, Rouwenhorst, and Gatev that never made it into an academic Journal. You could have read this paper and made money for 4 years on this idea. I wonder why it was never published?
Anyway, I hear many Risk Arb players at big shops are getting creamed. It seemed like you make money for 3 years, then give it all back in a couple weeks. Classic mode-mean trade: mode is positive, mean is zero.
Eric Falkenstein - am 2007-08-09 03:27
