making sense of 100's of funds

John, Sylvain and colleagues,

I find both your remarks very interesting and to contain some truth, I
have included a working paper from Angela, ea (2003). Where they use a
time varying two factor model to test for contagion. They find that in
the Asian Crisis the correlations are indeed increased however no strong
evidence is found for contagion in the Mexican crisis. Now my question
is do you think that the use of time-varying models or stochastic models
(such as a Markov VAR) can reduce risk in a crisis by switching to safer
portfolios and by switching to high risk (and hopefully high yielding)
portfolios in better times?

I'm very interested to hear your opinions,

  

I broadly agree with what John and others have said on this but here is 
my two centavos and this is related to the other thread on Copula 
functions. We repeatedly see that the marginal distributions are 
non-normal and there is asymmetry in the returns. e.g. Markets tank 
together but go up in a de-correlated fashion that is not quite captured 
in a correlation estimate. {Also recall Correlation(Pearson's) is a 
linear measure of dependence}

So perhaps this is the sort of thing that is best modeled using Copula 
functions with non-Gaussian marginals. Someone with a little spare time 
could perhaps back-test the performance of risk models that use other 
measure of dependence besides correlation and see how they measure up.