UNONPARM.G Was written by Cameron Rookley Please direct all comments, questions and suggestions to This code is intended to automate the process of running a univariate non-parametric regression while being rather flexible. It is by no means all-inclusive in terms of capturing all aspects of the literature. For example, it is currently written for the Nadaraya-Watson weighting scheme and has not been written with Priestly-Chao weights. It also does not account for boundary effects nor attempts to modify the Kernel weights accordingly. If you desire these features you'll have to find a way to integrate them into the code yourself. Any nice modifications or debugs would greatly be appreciated by the author. (please e-mail me at ) Although it is possible to select bandwidths via cross-validation, the process might be much faster if you decide to use the eye-ball metric. Instead of setting h0=0 (or h0d=0, h0d2=0) try supplying a vector of reasonable bandwidths and see what you get by graphing the resulting fitted values. Then fine-tune from there. This is particularly true if you have a large equidistant data set which allows for the use of the fast fourier transform, and you only desire to estimate the function in levels. Currently I have not programmed any automatic bandwidth selection mechanisms which incorporate the FFT. Therefore if your data set is large (and equi-distant or approximately equi-distant) I strongly recommend the "eye-ball" metric. Other than that, have fun and I hope this code is useful for all who encounter it.