# Estimating Frechet distribution parameter in R

I have a data set with 925 members that using Easyfit, I've determined is best fit using a Frechet distribution. Now, I'm trying to replicate this functionality in R to determine the optimum fit parameters using the VGAM package. Using this code:

vglm(V1 ~1, frechet2(location = 0), data, trace = FALSE)


I get the following error:

Error in eval(expr, envir, enclos) : initial values for 'location' are out of range


As I'm trying to estimate the 2-parameter distribution the location should be 0 (correct?)

In the VGAM documentation it mentions that the package may fail for low values of shape parameter (near 2 or lower); however, in EasyFit the estimation shows the shape parameter at .2646. Any thoughts on how I should handle this? I've done a search on but haven't found any other packages that provide the same functionality.

Any help would be welcome!

-
What is the minimum value in the dataset? If it is negative, then a location of $0$ is invalid. – whuber Apr 14 '12 at 1:22
I get a general sense of unease and doom when people talk about choosing a distribution simply because it "best fits the data." – Colin K Apr 14 '12 at 2:41
What's "EasyFit"? Sounds like it should be called "FarTooEasyFit"... – Spacedman Apr 14 '12 at 7:01
@whuber The minimum value is 0, there are no negative values. – nickrobison Apr 14 '12 at 10:11
Sorry if you're stuck with a problem description that you can't change, but ... most statisticians (I don't quite count, but there are lots here on SE) would be horrified (see comments above) at this approach. Rather than matching to a specified (if large) set of parametric distributions, I would recommend a nonparametric approach like log-spline density estimation ... you might get some interesting answers if you asked the question "what's the best statistical method for making robust, accurate, automated predictions of future event probabilities given 500-1000 samples of event times?" – Ben Bolker Apr 14 '12 at 22:11