Monday, June 15, 2009

How Not to Cheat

Don't do it too well. I once interviewed with a hedge fund, and they said they only take strategies with Sharpe ratios above 2. I've worked at a couple hedge funds, I never saw one with a Sharpe above 2 that didn't have access to retail flow. As the average fund had a Sharpe below 1, and these generally were diversified pools of several strategies, the Sharpe>2 criteria merely ensured you were either getting frauds or fools. They had over $5B under management at one time, now a couple hundred million. If investors merely talked to these guys, they should have known better.

Results that are too good are simply implausible. But while these guys were stupid, they weren't evil.

In the recent Iranian elections, former jailed reporter Amir Taheri noted:
Mr. Ahmadinejad was credited with more votes than anyone in Iran's history. If the results are to be believed, he won in all 30 provinces, and among all social and age categories. His three rivals, all dignitaries of the regime, were humiliated by losing even in their own hometowns.

The statistical probability the sample variance is this small is effectively nil.

I don't want my own fatwa, but I'm skeptical. Perhaps this is a good marketing pitch for statistics professors. Even if you are an evil dictatorship, it pays to know some statistics!


Anonymous said...

Regarding the election, it seems to me that a 70% landslide would almost guarantee a win in all 30 provinces. It depends on how much each province looks like the overall country. The larger and more diverse a province is, the more it will go like the country.

A US presidential candidate that won 70% of the vote would win every state.

Eric Falkenstein said...

ugh. you're right.

Lord T said...

Thats what you get when you hire incompetents on the cheap. The team who sorted the US in 2000 must have been available and well worth the money.

Anonymous said...

Entirely possible for a small fund to have a Sharpe of >= 2, as market making and arbitrage strategies will have very high Sharpe ratios. Of course, they're typically capacity-constrained, not scalable, and therefore not of interest to large institutional investors.