In many financial models we are interested in measuring the correlation between variables, returns etc. However, research shows that during crises times we observe "Correlation Breaks" where previously un-correlated variables become correlated.
What is the best way to quantify "stability of correlation" by examining the historic time series of the variables on which I am interested?
Would some kind of bootstrapping be a good way to start examining the sample distribution of the correlation? Is there any other method that I could apply?
You might want to compare constant conditional correlation with dynamic conditional correlation. In R, the ccgarch package will be helpful. In Matlab, Kevin Sheppard has an implementation of DCC.