I tried this. The only thing I didn't really like about this approach is the parameterization of the model that I used. The shifts in volatility can break the model, so engineering the model was 30% of the effort, and then dealing with all the various breakdowns was 70% of the effort.
I've...