0
$\begingroup$

I'm using a Johansen cointegration test to check for cointegration among a large number of time series. I've found that while the eigenvectors look great in-sample, the cointegrating relationships often don't hold up out of sample. However, if I manually "regularize" the eigenvectors by setting the smaller weights to zero, then the OOS performance is much better.

Is there a more robust or principled way to perform this regularization?

$\endgroup$
1
$\begingroup$

The only work I know about "regularising" cointegration is the paper of Liao and Philips (2012) who use a Lasso shrinkage. But it is a very theoretical paper and their method is far from straightforward... good luck for the implementation!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.