# Model fit is High but Ramsey RESET Test suggests omitted variables. What to do?

I'm trying to figure out what next steps to take. I created a model and ran OLS on a very large sample of data (over 400000 observations) and got an R-squared value of 0.80. So the model fit seems really good. However, I ran a Ramsey RESET test and its test statistic strongly suggested that there were omitted variables. I'm not sure what do do next. Do I just throw away the model, saying that the estimates are biased. Do I keep adding terms until the RESET test no longer suggests omitted variables?

The model is intended to be predictive within and out of the sample and given the model fit to the observed data, is it still truly a good fit, despite there being omitted variables?

Thanks!

• $R^2$ says exceptionally little about fit and (in vacuo) nothing about prediction quality. – whuber May 9 '14 at 19:19

## 1 Answer

Ramsey RESET test is not about omitted variables, but about functional form. If your model is $y=\alpha+\beta x+\varepsilon$, the RESET test can say that $y=\alpha+\beta_1 x+\beta_2 x^2 +\varepsilon$ is a better model.

It is not unlikely that a RESET test rejects because of the omission of relevant variables, i.e. the inclusion of an additional variable may capture the nonlinearities, but you should try a different functional form at first.

• Ok, the way it was presented in STATA seemed to suggest that it was more so testing for omitted variables. So then how do I really interpret the results of this test? Isn't there always going to be a "better" model out there somewhere? Should I just discard my current model? – TSP May 9 '14 at 21:16
• The Ramsay RESET test cannot pick up the influence of omitted variables, see here. You should try higher order powers of your independent variables at first. A scatter plot matrix can help a lot: if you can't notice any nonlinear patterns, then you could try adding a variable. - – Sergio May 9 '14 at 21:42