Hi,This is my first post and was wondering if I might get some pointers as to my following questions.I am currently studying for a Masters in Finance and I have yet to start my dissertation / thesis. I originally had the idea of doing something on financial instability by maybe looking at VIX and TED Spreads in a structural model but I am now shying away from that subject area as I think maybe it is too complex and broad a topic. With that in mind I need to come up with some alternative idea, preferably a bit more specific. Therefore I was thinking of doing something like....Take stock returns of a major index (S&P 500 for example) then model the volatility (GARCH) (or possibly returns (ARMA)) over the time period by splitting the data into regimes and create sub-models that fit the data. So when that is done I have 3 or 4 GARCH models that are specific to particular periods of volatility, so we has a GARCH for low volatility, one for high volatilty, etc.Then when that is done I create a multi-regime switching model that will randomly assign probabilities of a regime entering a particular threshold, when a threshold is breached a particular GARCH model, based on the previously specified models will activate.This model could then be tested on an out of sample period to see how it performs.So my basic questions are....Would this be a viable subject matter for a masters dissertation?Is it of the right order of complexity?Has this type of model been done before in the literature? If so are there any papers you are aware of that I could look at as I might be able to expand on it or apply it to a different country?Any other feedback or ideas would be most welcome.Thanks,Robert.

http://research.stlouisfed.org/wp/1994/94-015.pdf

Here is a variation that perhaps has not been done and would be interesting.Create a regime switching model with just two regimes: expansion and recession.In general, index return volatility is higher in a recession, so there is some hope for a GARCH-stylemodel, perhaps with some other exogenous explanatories (I strongly suggest VIX)Demonstrate that your model can 'call' a US recession well before the official NBER declaration,which shouldn't be too hard because they wait quite a while. With any luck, you should be able to call the end of the current recession before you graduate andbefore the NBER puts in their two cents.

Thanks bilbo1408 for the link to the paper. I will have a proper read of it once I get my exams out of the way.Also Alan thanks for the interesting idea you proposed. I will certainly have a think about how I might construct a model based on your input.Thanks.