First stage 2SLS: F-stat well above 10 but insignificant p-values

[Owen Wallbanks]

Dear all

I know when testing the relevance of our instruments, we look at joint significance of our predictors in the first stage 2SLS, judging that if we have an F-stat above 10 we reject the null of irrelevance. But are individual p-values important? Say if the p-values on all predictors were larger than any conventional significance level, is this a problem? Could it just be that predictors are highly correlated?

Many thanks

Owen

[Maxence Morlet]

Well the p-value on your instrument is relatively important. If it reveals that your instrument insignificantly affects your endogenous variable, you'll get a critique over its relevance...

The p-values on other predictors in the 1st stage regression are less important, although noteworthy.

[Owen Wallbanks]

Thank you, Maxence!