Skip to main content

Stupid Data Mining Tricks: Overfitting the S&P 500

Posted in

I found this PDF about overfitting the S&P 500 pretty amusing. Creating a model to fit S&P 500 using Bangladeshi sheep. It's a fun anecdotal story but the ending really hits home. 95% confidence in the model means 1/20 models are bogus. That's a lot of junk models and it's somewhat scary to think that these statistical models are being used a lot for very important things.