Roy Mendelssohn pointed me to this heartwarming story of Jay Vadiveloo, an actuary who got a patent for the idea of statistical sampling. Vadiveloo writes, “the results were astounding: statistical sampling worked.”

You may laugh, but wait till Albedo Man buys the patent and makes everybody do his bidding. They’re gonna dig up Laplace and make him pay retroactive royalties. And somehow Clippy will get involved in all this.

P.S. Mendelssohn writes: “Yes, I felt it was a heartwarming story also. Perhaps we can get a patent for regression.”

I say, forget a patent for regression. I want a patent for the sample mean. That’s where the real money is. You can’t charge a lot for each use, but consider the volume!

I’m taking the day off work to go and patent Maxwell’s equations.

wow digging deep into the inside jokes. it’s probably a sign that I read this blog too much that I’m like “oh yeah, damn Myhrvold…”

Yes, if you get all the jokes you’re reading this blog too much!

Even worse, I was able to predict from just the text of the post (which I read in my reader, without visible labels) that it would be labeled “zombies”

I wonder how narrowly the patent is written. If it’s as broad as “all uses of sampling in the insurance industry” that should have run into prior art issues. If it’s as narrow as “uses of multistage cluster sampling with systematic selection in property and casualty situations where there are more than 1 million policies to be selected from” then it’s so narrow it won’t be hard to work around it.

As an actuary with broad interest i always find my field pretty limited. too much money going around for people to share and discuss applications. the result is many people doing pretty basic stuff while the statistik fun is deemed too farm away from mainstream. The perks are good though.

here is claim 1:

1. A computer-implemented method for performing analysis of financial data, comprising the steps of:

storing, in a computer readable memory,

financial data related to a population of financial data records and segmented into a number (x) of classes, wherein the class segments are mutually exclusive and collectively exhaustive of the financial data, and

scenario data for a set of scenarios, wherein each of the scenarios is defined at least in part by a set of variables, and wherein the scenario data for each of the scenarios comprise at least some parameter values for the set of variables;

providing a computer processor associated with the computer readable memory with a model of a system defined at least in part by the set of variables; and

processing, with the computer processor, the financial data and the scenario data using the model to obtain an estimated model outcome distribution comprising a distribution of estimated model outcomes relating to the system and based on the set of scenarios, the processing further comprising:

selecting a first subset of the financial data as a first sample, the first subset drawn without replacement from each of the class segments of the population and having a sample size (z);

performing, with the model and the first subset, a first test of the set of scenarios to obtain a first set of sample outcomes for each of the scenarios; and

repeating the selecting and performing steps using additional subsets of the financial data to perform additional tests of the set of scenarios and to obtain additional sets of sample outcomes for each of the scenarios;

wherein the additional subsets are drawn without replacement from each of the class segments;

wherein the first sample outcomes and the additional sample outcomes are combined to create a cumulative estimated model outcome distribution;

wherein the selecting and performing steps are repeated until the cumulative model outcome distribution is within a pre-determined acceptable tolerance limit from a distribution of fully assessed model outcomes for the set of scenarios obtainable by performing, with the model, a single test of the set of scenarios using all of the data; and

wherein the cumulative model outcome distribution is identified as the estimated model outcome distribution;

wherein the number (x) of classes, the sample size (z), and a number (y) of tests comprising a count of the number of times that the performing step is conducted ensure that the cumulative model outcome distribution is within the pre-determined acceptable tolerance limit from the distribution of fully assessed model outcomes.

from here:

http://www.google.com/patents?id=avgEAgAAEBAJ&printsec=frontcover&dq=VADIVELOO&hl=en&sa=X&ei=CQrtT8KsDcH42QWho9WvCg&ved=0CDMQ6AEwAA

[…] was recently reading Gelman’s blog, and he wrote a post about this: Patents Aren’t Only for Engineers. Apparently, some actuary received a patent […]

The author’s discussion in The Times article is pathetic.

Right out of the box he seems astonished that statistical sampling actually works!

Did he never take Stat101 where this astonishing fact is shown again and again?

Has he never heard of bootstrapping and re-sampling – apparently not from the article.

Yet he’s an actuary and runs a “research” center – good grief.

This “patent” is “patently” and thoroughly nothing but applied mathematics/statistics

which is not supposed to be patentable at all.