Paretofit - threshold values as input?

Hey guys,

i have monthly return data and essentially i want to do exactly what the function paretotails does (a gpdfit in the left and in the right tail and a kernel in the center). Now i defined thresholds for my left and right tails using Mean-Excess Plots, (picked the threshold for which the ME function is appr. linear).

so lets say for my right tail the threshold is 5.324% and for my left tail my treshold is -4.327%

So i have my data and i have the thresholds but Matlab wants pl and pu as input (which are not the quantiles of my empirical data but from the fitted GPD). So what can i do?

my solution then should look like:

gpd fit for -infinite to -0.04327 kernel smoot for -0.04327 to 0.05324 gpd fit for 0.05324 to infinite

Is this possible with the paretotails function? And if not, how can i achieve it then?

thx in advance

[Bruno Pop-Stefanov]

Can you plot the cdf of the return data and see what lower and upper cumulative probabilities the thresholds correspond to?

[Markus]

The Problem is these cum. probabilites would then be the pl and pu of my empirical data, but matlab wants pl and pu for the fitted gpd (these do not match because the pareto does not fit the empirical data 100%)

[Markus]

can nobody help me? pls its very important

[Kostas]

Possible you could take a look in evim package, it contains tools for POT analysis and fitting the GPD.

[Markus]

thx. i can of course fit the gdp tails separately and the kernel separately but is there any possibility to "link" these 3 segments to one distribution where i can simulate from?

Answer 1

[andrea lorenzini]

hello! how did you find the mean excess plot?