Validity of the GRS Test with Non-Normal, Autocorrelated Residuals

Summary

The document asks whether the Gibbons, Ross, and Shanken test can be applied when regression residuals are non-normal and autocorrelated. The setup uses ten test assets and compares residuals from regressions on a three-factor model and a Carhart model. It seeks guidance both on the test’s validity under these conditions and on possible alternatives if its assumptions are not met.

No answer, test results, or proposed adjustment is included, so the document does not establish whether the standard GRS procedure remains valid in this setting. It frames an empirical asset-pricing question about joint alpha testing and the effect of distributional and serial-dependence assumptions. Any practical conclusion would require examining the test’s assumptions and selecting an inference method suited to the residual properties; those steps are not supplied here.

Key ideas

• The question concerns using the GRS test to jointly evaluate alphas from factor-model regressions.
• The stated residual concerns are non-normality and autocorrelation.
• The setup compares three-factor and Carhart regressions across ten test assets.
• The document gives no answer or alternative procedure, so it provides no conclusion about validity.

Tags

# Applying GRS-test on non-normal residuals with autocorrelation


# Applying GRS-test on non-normal residuals with autocorrelation












Is it valid to apply GRS-test (Gibbons, Ross and Shanken 1989) on non-normal and autocorrelated residuals?

I got residuals using 10 test-assets regressed on 3-factor and carhart.

If it is valid, how should I do it?

And if it is not valid, what should we do?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.