Jeffrey S. Racine in Nonparametric Econometrics: A Primer apparently names R-squared as one of the options for model evaluation.

I am talking here in particular about Local Constant Kernel Regression

The dataset is quite small - 500 observations. I just curious how we can compare these very different models so I can select the "better" one - to somehow back up my preference. I couldn't find any R package that would calculate AIC or BIC for nonparametric models. I suppose cross validation could bet the choice then.

Normal for both models.