2
$\begingroup$

I am making an OLS model. Firstly, I want to confirm that checking heteroskedasticity. How to check the presence of heteroskedasticity? Is it just plot resid against independent variables, right?

Secondly, If the model does have heteroskedasticity, does it impact this OLS model prediction?

$\endgroup$
4
$\begingroup$
  1. You can also statistically test heteroskedasticity (e.g., White's test: regress squared residuals from OLS on $\hat{y}$ and $\hat{y}^2$, or if you have a simple model, regress $\hat{u}^2$ on $x$ and $x^2$).

  2. It's unclear what you mean by your "model having heteroskedasticity" (do you mean the error term is heteroskedastic?) and "impacting the prediction". You can always choose to ignore heteroskedasticity if you want. (You seem to do so, as you mentioned OLS.) Or you can have a different estimator (e.g., WLS) for a better (sometimes only slightly better) predictor using the same model. Or you can even have a model that incorporates heteroskedasticity explicitly as in, for example, GARCH models in time series. Please be more specific if you want more help. You might want to read this post for time series: https://stats.stackexchange.com/questions/175360/forecasting-ar-arch-garch-models

$\endgroup$
  • $\begingroup$ It is OLS regression. I have the result of residuals. However, how do I detect the heteroskedasticity graphically? $\endgroup$ – Tom Apr 10 '18 at 2:23
  • $\begingroup$ Tom, this is an interesting question in its own right. Post it as a separate question. $\endgroup$ – BKay Apr 11 '18 at 13:20
  • $\begingroup$ If you have a single explanatory variable, plotting the residuals against the explanatory variable would work. If you have multiple regressors, you can plot the residuals against the fitted values. $\endgroup$ – chan1142 Apr 12 '18 at 13:53

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.