Comparing the ACER and POT MCMC Extreme Value Statistics Methods Through Analysis of Commodities Data
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2397922Utgivelsesdato
2016Metadata
Vis full innførselSamlinger
Sammendrag
The Average Conditional Exceedance Rate and Peak Over Threshold Markov Chain Monte Carlo are two extreme value statistical methods, compared in this work. They are tested for both extrapolations and prediction intervals. The methods are compared for difference scenarios concluding that the Peak Over Threshold Markov Chain Monte generally prefered better for prediction intervals. It also seems to be preferable for extrapolation of independent and identically distributed data, and data approximately so. There are some indications that the Average Conditional Exceedance Rate method maybe favorable for capturing the data dependencies and extrapolation for correlated observations, but more work is needed for a conclusive result on that aspect.