Monday, 21 April 2014

Bayesian Bunching of Black Swans ...

Continuing (after an admittedly long pause) with our swan-related programming, here's a new preprint from Tim Graves, Bobby Gramacy, Christian Franzke and myself on a Bayesian approach to measuring long range dependence.

During my final few years at the British Antarctic Survey in Cambridge, UK, working with Tim (first a Master's and then PhD student in Cambridge Statslab), Bobby (then his supervisor), and Christian (then my colleague at BAS) was a highlight of my research life.

The scalpel of a statistician's approach to LRD which he employed with impressive dexterity, precision, and originality, helped me to cut out quite a bit of the fat in my own thinking. The experience has left me confused on a much higher level about LRD, of which more quite soon.

Tim is now back with the URS corporation in London, and we are collaborating with him to submit the papers from his thesis, starting with this one. I am presenting a poster based on the thesis next week at the EGU meeting in Vienna, which will include some of the results of his new approach. This method allows estimation of the parameters for heavy tails and long range dependence in a modified (alpha-stable innovations) ARFIMA model, and thus both "black swans" (more properly, "grey" ones in Taleb-speak), and "bunching".