extreme value modeling questions

["Praprut Songchitruksa" <praprut@purdue.edu>]

I have two questions about extreme value modeling using S-Plus 6.1. I'm using the codes written by Coles (2001).

 

1) The mle estimation procedure doesn't converge sometimes due to singular hessian. I tried to work around this by modifying the function "hess" to use "ginverse" instead of "solve". I've found that it works fine but I don't know if it has any impact on the results? Is there a better way to work around this?

 

2) I'm using a bootstrap to get a standard error of a value which is a function of mle from a GEV model. I block the data over time and then pick the block maxima to fit the GEV. My data is time series. Can I bootstrap a selected set of block maxima directly? If not, how can I bootstrap the entire sample without destroying the serial correlation?

 

Thank you very much.

 

Regards,

-Praprut

 

Transportation and Infrastructure Systems

School of Civil Engineering

Purdue University