I am dealing with a project where I have to optimize the Risk Weighted Assets. The RWA currently consists of 3 components. ? Incremental Risk charge (IRC) ? Value at Risk (VaR) ? Stressed VaR (SVaR) Lets say for example that RWA [$]=[$] [$]a_1[$] VaR [$]+[$] [$]a_2[$] SVaR [$]+[$] [$]a_3[$] IRCI have a year long time series for all of them. I have to make forecast for RWA. I looked at RWA individually and any of the ARIMA models are not suitable for the simple reason that the residuals are not normal ( which is not surprising). The next step to model this time series that I am considering is Vector Autoregressive model. Given the equation (1) above, can anyone think of any cons of that approach and suggest what else can I try?

Well the idea here is to forecast a time series. This means choosing a stochastic differential equation that you use to describe your variable. The way to approach that choice is to consider the behavior of the density of that SDE. Does it have the right boundedness? (i.e. use a geometric process if your data is bounded at zero, etc.)Now, calibrate your model and project forward using the density. Now you can assign probabilities or confidence limits, etc.What does your series look like (visually)? Does it look like compounded growth (geometric), mean reverting(ornstein-uhlenbeck), or exploding-dissipating(CIR)?

Hi Orbit, Thanks for your reply. It certainly shows a growth pattern.(sorry but I don't know why wilmott was not accepting a jpeg uplaod. So I posted the screenshot here).https://www.flickr.com/photos/133994830 ... Screenshot of Time Series hereI am curious to know the thoughts on why Vector Autoregression might not be the correct approach?

I assume that this is for a capital adequacy/enterprise stress testing exercise?This seems a bit dubious to me. These are portfolio market risk measures, and if you talk to your risk department I think you will find that quite often what is driving material changes of these quantities are changes in positions and changes in market volatility (except perhaps IRC, which will depend a lot on what the model is and how it is updated...). So I believe that typically one tells a story about the event, how it will impact the business/positions, and what happens to market volatility.