Be Cautious Counting on Correlations

Updated on

Be Cautious Counting on Correlations by Attain Capital

May 15, 2014

Great site pointed out by @reformedbroker over the weekend… looking at ‘spurious correlations’

Here’s a few of our favorites:

Number of people who died by becoming tangled in their Bedsheets
correlation with
Total revenue generated by skiing facilities (US)

= 0.96

Number of people who died falling off a Cliff
correlation with
Number of lawyers in Ohio

= -0.85

These stats would have us believe that the more margarine people eat, the more people get divorced in Maine; that the less people who die getting tangled in their sheets means the lower the national ski industry revenue is, and that every time a person goes off a cliff – there’s one less lawyer in Ohio (maybe it’s the lawyers jumping off the cliffs).

But surely those aren’t the case, and assuming such would be making the classic error of confusing causation with correlation. What’s more, we would argue these things aren’t even really correlated, despite the statistics. We would bet that going back 100 years would show correlations of closer to 0.00 for all of the above, because we know intuitively that margarine consumption and divorce have nothing to do with one another.

Which leads us to the cautionary correlation corollary, which is to be extra careful when trusting correlations, especially ones on annual data with few data points as appears to be the norm on the spurious site. The investment world is filled with classic examples such as stock returns being correlated with women’s skirt lengths; but the real lesson here may be a but more nuanced.

The real lesson for us is twofold – 1. Anything can become correlated over a set period of time. Anything. Just peek back at 2008 for a real world example when stocks and bonds, stocks and gold, stocks and commodities, stocks and hedge funds, and even stocks and money market funds moved towards  a correlation of 1.00 (went down at same time and near the same magnitude).  2. Do a fundamental double-check of your portfolio before blindly trusting the correlation metrics you’ve come up with in analyzing the components. You may be a professional trader and found your Cocoa trading model has zero correlation to your Sugar trading model, and look to lever them both up at the same time knowing they aren’t likely to lose at the same time…. And that is right when you’ll find just how non spurious two related commodities being traded by two models developed by the same brain (yours). Same goes for you, Mr. Investor – that credit arbitrage hedge fund throwing off income currently with a zero correlation to the Nasdaq may become very non-spurious also, if credit tightening causes losses in stocks and the hedge fund.

Now, if we could just get the daily number of readers of our blog to correlate with the moves in the Japanese Yen, we would be onto something!

Leave a Comment