Re: OT: interesting global warming quote found elsewhwere

John Larkin wrote:
On Sun, 03 Aug 2008 22:18:18 -0400, John Popelish <jpopelish@xxxxxxxx>
wrote:

John Larkin wrote:
On Sun, 03 Aug 2008 21:57:15 -0400, John Popelish <jpopelish@xxxxxxxx>
wrote:
I'll try. The book is all about how we pretend to be able to predict the future based on inductive logic (extrapolating from past experience), and how this falls down in the presence of rare exceptions or "black swans" (with rare being defined as outside recent memory or education). It doesn't explain how to predict the unpredictable, but how to avoid being a sucker by mistakenly thinking you have predicted (quantified) risk but have not.

The black swan relies on the hypothesis that "all swans are white" being accepted at face value because you have never seen a black swan.

One of the common ways that people fool themselves into thinking they have a grasp of future risk is to model the past as a Gaussian process. But many processes, including the stock market and business in general (the big exception being gambling casinos), are more chaotic, or Mandelbrotian rather than Gaussian. But you can perform Gaussian modeling on long stretches of Mandelbrotian data between the explosive exception parts and it looks like a good fit... until the next blow up happens.

I don't think there are any serious practitioners still using Gaussian financial models in places where it would really matter. They would be crucified very quickly by the guys with better (although still incomplete models). A basic introduction in the book
http://www.springer.com/math/quantitative+finance/book/978-1-84628-419-9

Experience is the worst teacher.

Not at all. Experience is exactly how we (and now computers) learn about the external world. the most powerful methods available at present are based on Bayesian statistics and do everything these days from classify spam to guess about how to balance a portfolio.

Exactly the point of the book. Our brains are generally wired up for simple things and short terms, not long and complicated patterns. Modern life has left common sense behind. Ask any bank manager who has, in the last year, lost more money than his bank has made since the day he was born, estimating his risk on Gaussian models.

Most of the senior suits in charge of the trading desks grew up in an era when they had no clue at all what they were doing other than seat of the pants and crude calculations. That is why major banks can be broken by a single rogue trader who understands how to beat the back office checks and balances (usually someone who has worked in both roles).

Yup, people tend to not believe in catastrophes.

Speak for yourself. Catastrophes can and do happen. Tormenting a farming computer model to extremes I once had my crops trampled by a herd of unicorns. I was trying to break it and took it into uncharted territory.

That was in the 80's so I am pretty sure that the better models have included very low probability high damage events for at least that long. Some of them were very humourous scenarios (to make the point that no-one knew what the real ones would be or pretended to be able to enumerate them all - rusts mutating that would take down wheat monoculture was/is a real worry).

In your wilful ignorance in the next breath you will deny that there is any possibilty of an AGW induced catastrophe. Do you see the contradiction?

A lot of engineering has to do with keeping our primitive emotions
from screwing up our ability to think. If experience makes you think a
pattern is general, it's dangerous. If it open up the appreciation of
possibilities, it's good.

Bayesians consider a priori that all possible outcomes are equally likely until there is evidence to the contrary. There is a small but serious technical difficulty in haggling over "all possible outcomes" but for important classes of problems the uninformative priors are fairly well known. The initial distribution is modified by the observed data, the integrals are tedious, but possible with modern computers.

Now that the book of Jaynes life works is published most of the online versions of his treatment of Wolf's dice experiment have disappeared, but there is a paper still on arXiv section 1.5 if you are interested.

http://arxiv.org/PS_cache/physics/pdf/9805/9805024v1.pdf

This is where statistical mechanics crosses over into information theory and inference problems. Jaynes original paper is much better but AFAICT not freely available any more.

Regards,
Martin Brown
** Posted from http://www.teranews.com **
.

Relevant Pages

  • Re: OT: interesting global warming quote found elsewhwere
    ... (extrapolating from past experience), and how this falls ... thinking you have predicted risk but have not. ... past as a Gaussian process. ... explosive exception parts and it looks like a good fit... ...
    (sci.electronics.design)
  • Re: OT: interesting global warming quote found elsewhwere
    ... (extrapolating from past experience), and how this falls ... thinking you have predicted risk but have not. ... past as a Gaussian process. ... explosive exception parts and it looks like a good fit... ...
    (sci.electronics.design)
  • Re: OT: interesting global warming quote found elsewhwere
    ... (extrapolating from past experience), and how this falls ... thinking you have predicted risk but have not. ... past as a Gaussian process. ... explosive exception parts and it looks like a good fit... ...
    (sci.electronics.design)