any one how to apply large scale optimization theory in financial modeling? In pratice, do you have lots of problems assoicated with large scale optimization?

there are a number, not necessarily large scale.Finding an optimal portfolio for example (risk vs return)Finding the free boundary for american options is another

finding parameters for your pricing model that match the observed market prices is an optimization problem i am frequently faced with. sometimes involves calibrating 20+ parameters. use differential evolution and downhill simplex, mainly. i'm open to any better ideas.

There are a couple of commercial software systems which apply stochastic dynamic programming to ALM (e.g. pension A/L mismatch).

Hi DavidJN,Very interesting. Could you provide some examples of such commercial software systems? Have you used any of them? Do you have any opinions on how well they work and in which circumstances? Thanks for the info in advance.-------

There are a couple of commercial software systems which apply stochastic dynamic programming to ALM (e.g. pension A/L mismatch). DavidJN,Dynamic programming is governed by the 'Bellman' equation. The implication is that the variables cannot be stochastic (well known over 30 years ago) in this type of optimization problem. This is the main reason many OR guys like me eshewed the stochastic calculus fad.

Quote...commercial software systems which apply stochastic dynamic programming to ALM Quote...Dynamic programming is governed by the 'Bellman' equation.So N, what makes you think "stochastic dynamic programming" is dynamic programming? If we were to go purely by names, then is an Equity Default Swap a swap? and how do equities default, anyhow? if you want a more general analogy, take the Holy Roman Empire. It was neither Roman, nor Holy nor even an Empire!

QuoteOriginally posted by: NDynamic programming is governed by the 'Bellman' equation. The implication is that the variables cannot be stochastic (well known over 30 years ago) in this type of optimization problem. This is the main reason many OR guys like me eshewed the stochastic calculus fad.So if current payoff and next state in dynamic programming are known only by their probability function then we do we have?

Hey dude, dynamic programming is only deterministic. As we say in Maine, you can't get there from ear.If you want to use probability, stick with the Feynman path integral. Technically, the issue is the constraint manifold for probability which is the same problem that exists with the Feynman path integeral.

QuoteOriginally posted by: DavidJNThere are a couple of commercial software systems which apply stochastic dynamic programming to ALM (e.g. pension A/L mismatch).AFAIK, these software systems are mostly stochastic linear/quadratic programs misnamed as stochastic dynamic programs. I recall some stuff Frank Russell company did in this years ago.

I don't write their ad copy or research papers, so I couldn't tell you whether they are misnamed or otherwise.Commercial software includes (I have no affoliation with any of them): 1. Towers Perrin-Tillinghast (see Mulvey, Gould and Morgan, 2000)2. Russel-Yasuda Kasai model (see Carino, Meyers and Ziemba, 1998) 3. www.profitstar.com/ 4. www.surya-soft.com/ 5. www.fernbach.comThe last company is affiliated with the University of St Galen, where they seem to be doing some really interesting work. There is alsoa software company affiliated with Cambridge University, I think.Overall impression? A lot of technical details but light on practical finance content (much like how this post has evolved!).

using Matlab's Optimization Toolbox I managed to optimize across 65 parameters- it was to estimate a multi-factor Vasicek model with the maximum likelihood-based Kalman Filter method (see Babbs & Nowman '99 for this kind of approach)The available constrained optimization function fmincon() frequently broke down, so instead I imposed constraints with transformation functions and applied unconstrained optimization fminunc() (BFGS Quasi-Newton), with good results.

suggested reading:

QuoteOriginally posted by: Jany one how to apply large scale optimization theory in financial modeling? In pratice, do you have lots of problems assoicated with large scale optimization?out of curiosity, why are you asking.I can think of 2 reasons, probably many more:(1) you're looking for a project for a thesis(2) you're an expert on optimization trying to scam a job at a bank

criskyQuote instead I imposed constraints with transformation functionsDo you mean penalty functions?