On 2009-04-22, F George McDuffee wrote:
On Tue, 07 Apr 2009 19:05:18 -0500, Ignoramus7501
wrote:

My own programming involves mathematical modeling of financial
derivatives prices, which is in many ways akin to mathematical physics
models. That's heavy duty math involving partial derivatives, and
such.

i
-------------
Any comments that you can make about why Monte Carlo simulation
is not being used for the bank "stress" tests?

It would appear that the range of outcomes and probabilities this
generates would be far more useful that the expected pass/fail
rankings.

But the distribution of such outcomes depends on assumptions of your
monte carlo simulation. And no one knows what assumptions to make
(what is the standard deviation of housing prices, etc). But
your idea is a good idea in the right direction. The simulation would
involve not only purely random variables, but rather, their likely
effect on other variables.

Example, let's say that a variable "housing default rate" would rise
to a certain amount. That would have a direct effect on the banks'
capital levels, but it would also have an effect on housing prices,
further influencing the default rate of mortgages that would be deeper
underwater.

Then it would affect how much more would be the losses of the likes of
AIG and other mortgage insurers.

In other words, any realistic simulation would go beyond "pick a
number" game and would model at least the basic dynamics of the
process.

The distributions of most of the inputs/metrics such as prime
interest rates, unemployment, money market interest rates, etc.
should be the same across all institutions.

Any thoughts on the use of Cauchy v normal distributions? Also
how do you rank the Excel add-ins for MC simulation/analysis?
Any good?

I do not use excel add-ins, personally.

i

On Wed, 22 Apr 2009 19:02:11 -0500, Ignoramus1265
wrote:
snip
In other words, any realistic simulation would go beyond "pick a
number" game and would model at least the basic dynamics of the
process.
snip
Indeed. In many areas the unemployment has blown past the worst
case unemployment rate used for the stress test. The "black
swan" seems alive and well...


Unka' George [George McDuffee]
-------------------------------------------
He that will not apply new remedies,
must expect new evils:
for Time is the greatest innovator: and
if Time, of course, alter things to the worse,
and wisdom and counsel shall not alter them to the better,
what shall be the end?

Francis Bacon (1561-1626), English philosopher, essayist, statesman.
Essays, "Of Innovations" (1597-1625).

Bob Engelhardt writes:

I could be wrong, of course - it has been 17 years since I retired.
If so, I'd be very interested in seeing how software is modeled. Just
hearing what the parameters are would be interesting. Maybe I should
find a software NG to ask.

Not all models are purely numbers/math based.

Some work is being done in ontologies and the Semantic Web. I just
attended an interesting talk on the subject. One of the problems is
that rules for an expert systems are based on the model of the system
underneath.

There can be a large gap there in these two models. 10 years ago, I
was a domain expert for a new expert system. I knew when a system was
behaving properly, and how to diagnose a system that didn't have
maximum performance. The challenge was how to owrute rules that
expressed my knowledge so automated systems can duplicate my
skill. Expressing rules that a human "knows" in a form a computer can
use is a challenge.

So now - new languages are being created to describe the system
underneath. These languages are fully integrated into the Integrated
Development Environment (IDE). Try to describe a rule for a system,
and the editor lets you know there's a semantic error, forcing you to
correct the rule as you type it in. So you can either define a new
rule, or extend the model underneath, using the same IDE.

Don Foreman writes:

I understand the distinction. My understanding of how well-designed
software is designed and modelled is very sketchy but I worked with
people who did it very well indeed. The software was every bit as
robust as the hardware that executed it.


The real issue - is how much pressure is the company to produce a
profit? Companies that operate on the edge of profitablility will cut
corners and slap together software chucks, and hope fot the best.

Consider Carnegie Mellon's Capability Maturity Model.

Some companies operate at a very low level, especially start-ups.

On Apr 25, 8:11*am, Maxwell Lol wrote:

Some companies operate at a very low level, especially start-ups.

How do you incorporate the response to random opportunity or
inspiration into a model? From what I've seen in R&D that depends on
the judgement, previous experience and risk tolerance of the decision
makers.

jsw

Jim Wilkins writes:

On Apr 25, 8:11Â*am, Maxwell Lol wrote:

Some companies operate at a very low level, especially start-ups.

How do you incorporate the response to random opportunity or
inspiration into a model?

I'm not sure that applies. CMU's CMM is not about response to
opportunities, or about research per se..

The model MEASURES how mature the development process is. For
instance, if a new product is created, are lessons learned
from the last project applied to the new project?

A start-up might struggle to get products out before they are forced
to close their doors. The schedule is uncertain. Quality is uncertain.
No one knows if the product will work, or be successful. Therefore the
maturity model score would be very low.

On the other hand, a company with a well established development
system, and an experienced staff can deliver a new product on time,
and on budget, without cutting corners. That would be a higher score.

This is an overview:

http://www.sei.cmu.edu/cmmi/adoption...overview07.pdf


From what I've seen in R&D that depends on
the judgement, previous experience and risk tolerance of the decision
makers.

Sure. But not all R&D environments are the same.