I know that Mincer-Zarnowitz type of regressions are typically applied in a time series setting to evaluate forecasts. My question is whether the same type of regressions can be used in a cross sectional setting?

Say, my forecast of a variable for entity $i$ is $f_i$ and the realization of the variable is $a_i$. Does it make sense to run the following regression and conduct the joint test $\alpha=0$ and $\beta=1$?

\begin{equation} a_i = \alpha + \beta f_i + e_i \end{equation}


I don't see anything wrong about this. However, note that you typically want the forecasts $f_i$ to be out-of-sample. Hence when estimating the forecasting models you should not use the observations $\{a_i\}$ which are later used to run the Mincer-Zarnowitz regression.

  • $\begingroup$ Welcome! I've been wondering myself why out-of-sample evaluation methodology (such as MZ regressions, Diebold-Mariano tests, etc.) is common in time series applications (e.g. inflation or volatility forecasting) but rarely if ever applied in cross-section contexts (e.g. wage regressions or asset pricing). If this is only due to ``cultural'' reasons, then looking at these tools in cross-sections might be an interesting direction for further study. $\endgroup$ – Fabian Nov 28 '13 at 10:10

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.