I did a quick Google search to see if anybody already had an algorithm named Confidence-based Incremental Stratified Sampler. I didn't get any hits, but I did turn up a really relevant paper from Ernst & Young of all places. The reason it's so relevant is that it's stratified sampling of financial data and they indicate that they see the Gamma distribution all the time. (I'm still not sure if my data is Gamma or Chi-squared, but both have the problem of heavy tail).
Rotz, Wendy, et al. "PPS vs stratified sampling in modern audits." Joint Statistical Meetings, 2002.
Anyway, it's a relevant hook into the financial literature on sampling, so that's a good thing in and of itself. I'll add it to my citation tree.
No comments:
Post a Comment