AIC after non-parametric regressions

[Antonio Tintore Vicent]

Hello everyone,
Can the AIC method be used after non-parametric regressions? I've tried the 'estat ic' command but it's not allowed after ´npregress.´ Is there any alternative command?

Much appreciated!

[Clyde Schechter]

No. The AIC (and BIC, for that matter) are calculated from the log-likelihood and number of parameters. But the non-parametric regressions don't do maximum likelihood estimation, so there is no likelihood statistic available for these models.

[Antonio Tintore Vicent]

Thank you for the response Clyde! Would you recommend any other criteria or tool to choose among different non-parametric specifications?

[Clyde Schechter]

I'm afraid my experience with and understanding of these models is insufficient to give you good advice here. I hope some other forum member can and will respond.