no solution to this?
------------
If one is doing any stress test analysis how do he incorporate current situation in his scenarios analysis
------------
For stress testing, you can try a variety of covariance matrices, e.g. we have used constant correlations of rho = 0.0, 0.3, 0.5, 0.7, and 0.9, then also vary the volatilities (10%, 20%, 30%, 40%, 50%, e.g.) on each item. Record each VaR with its assumptions to get a grid of possible losses. Stress testing isn't supposed to be a one best guess (these are very unlikely situations anyway), but generate a surface of conditional losses: "if x goes to this, and y goes to that, we lose z", etc.
You just have to map out the whole sample space well: e.g. maybe map out all the correlations from -1 to 1 in increments of 0.1 (this may lead to some negative definite matrices, btw - you can ignore those possibilities), and let vols range from 0% to 100% (or more) in increments of, say, 5%. Use all those combinations to get a new covariance matrix on your items and calculate a different portfolio VaR from each. That should probably cover a good range of possible situations, including the current mess? Maybe not, but you just have to define your sample space well, in any case.
------------
what do Quantitative Finance teach us in this situation for portfolio management
-------------
Day to day portfolio management is different from stress testing, but maybe look at this:
Optimal Portfolios in Good Times and Bad
George Chow, Eric Jacquier, Mark Kritzman, and Kenneth Lowry
Financial Analysts Journal
May/June 1999
It discusses a way to blend "dull day" and "stressed day" covariance matrices in a weighted average, according to your forward looking view of their probabilities.
