Confidence interval question about pyextremes

[georgebv]

Monte Carlo idea is really simple - you sample from your data (this way you have a "different" dataset), fit model to it, then repeat this process many times. In the end you have a large number of curves - one for each sample. To get 95% confidence interval you then take those curves and for each return period calculate return values and remove top and bottom 2.5% and then draw curves through those points.

[georgebv]

The reason is in the choice of distribution - you use Gumbel distribution which is exponential and appears linear when plotted at logarithmic scale. Also, if it's so tight at 99.9% confidence, this means that your choice of model was correct and that you have good data.

[TanguyRiou]

Ok thanks for your answer. I think I am misunderstanding the confidence interval. In the book "An introduction to statistical modeling of extreme values" of S.Coles they give examples where confidence interval is computed based on a normal distribution. I saw in #36 that you are using a monte carlo method. If you have any details of this computation it would be amazing. I tried to dig in your source code but I admit that I did not understand everything.