Wednesday, February 20, 2013

Overfitting

By Rick Nason, PhD, CFA
Partner, RSD Solutions Inc.

In academic finance we tend to research questions where there is data, rather than where there are important questions that need some answers.  One of the problems this raises (there are many) is that of the problem of “overfitting”.  In other words, with enough data you can make almost any hypothesis have statistical significance.  In turn we tend to think of statistical significance (and correlation) as the same as causation.  Overfitting means that you can explain almost anything with enough data.  It is not knowledge.  Important difference.

No comments: