Ground Hog Day, Stock Trading, and Probability

Today is Ground Hog Day, and ABC News had an article on Punxsutawney Phil (the famous groundhog from Pennsvlvania).

The legend is that, if the groundhog sees his shadow, then winter will be 6 weeks longer. If it is cloudy, and he doesn't see his shadow, then spring will arrive early.

The ABC New article states that "It turns out the ultimate prognosticator- and his copycat counterparts- are wrong more often than they are right."

It then also states that "An analysis by the National Climatic Data Center found there is no correlation between Phil's predictions and the actual weather."

What's interesting is that, technically, these 2 statements contradict themselves. Being "wrong more often than right" is not the same as "no correlation" - and this has implications for developing a trading system.

If you develop a system for predicting the weather or for trading stocks, if your system does terribly, then you haven't failed. You would simply reverse the interpretation.

The worst thing that can happen when testing a trading system is not losing all your money - it's getting random results.

For example, if you tested a system of buying stocks on the first Tuesday of the month and selling 3 days later (this is a made up example), and you lost money 84% of the time, then this might not be a failure.

Why? Because this means you could short stocks on the first Tuesday of the month, and buy to cover 3 days later - which would win 84% of the time. (Of course, you would have to confirm that the total amount of money made on the winning trades exceeds the total amount lost over the 16% of the time where the rule failed).

The result you would not want would be to find that the rule broke even and gave no meaningful advantage. This would mean that, no matter which side you took, you would not expect to make enough to cover your trading costs (commissions, slippage, etc.)