Re:Algorithmic Trading Using Logistic Regression
by fractionalhare on Friday September 04, 2020 @09:47PM
(#24380753)
That's right. At minimum there should be Bonferroni correction to adjust for the number of (implicit) hypothesis tests being executed. If we pursue this methodology with a significance level of p = 0.05 and test 20 independent signals against some financial metric, we can expect to find one of them to exhibit correlation simply by chance. Considering the OP tested 32 technical indicators against the dependent variable, this likely happened.
Reminds me of https://xkcd.com/882/