Return value calculation from gumbel cdf for MAXIMA not minima

I am trying to calculate some return values (e.g. find x when cumulative probability is 0.99) from a series of annual maxima. In matlab it says that to fit a Gumbel to maxima (not minima) you must negate the X values, which I have done. This gives me a (negative) mu and sigma. I then use evinv and negate the result to get the final answer.

However, I get negative values, which is completely outside the range of annual maximum values that I have (these are all positive).

Here is the code:

par = evfit(-P) mu = par(1) sigma = par(2) 100yrVal = -evinv(0.99,mu,sigma)

Can anyone advise?

Answer 1

[Tom Lane]

It could be as simple as your needing to get the lower tail of the distribution of negative values, to produce the upper tail of the distribution of the original values:

Val100yr = -evinv(0.01,mu,sigma)

[Hannah]

Yes that's it, thank you!