I know that McElroy R^2 is a measure of goodness of fit for Seemingly Unrelated Regressions (SUR models), but how can one judge that the estimated equations are well fit by using the McElroy R^2?


This kind of $R^2$ is a monotonic transformation of the F statistic from the the test that all the coefficients in the system are zero. It is bounded on $[0,1]$, so it has the same kind of interpretation as the ordinary $R^2$:

$$ \text{McElroy }R^2 = 1-\frac{U'W*U}{Y'W*Y},$$ where

  • U is the stacked vector of residuals
  • Y is the stacked vector of dependent variables
  • W is the variance-covariance matrix of residuals

If your model is the cat's pajamas, the vector of residuals will be close to zero, so $R^2 \approx 1$ since the second term drops out. If the model has no explanatory power, then $U\approx Y$, so $R^2 \approx 1-1=0$.

The full citation is McElroy, Marjorie B. (1977) "Goodness of Fit for Seemingly Unrelated Regressions", Journal of Econometrics, 6(3), November; 381-387. I can't seem to find an ungated link.


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.