Stress-testing with AnaliTeG

In our initial work with AnaliTeG we limited ourselves to only basic crude tests (like crisis stress-testing) not because we didn’t want to or couldn’t do anything else but because it would require endless time that we didn’t have. That said, people ask what else can be done with AnaliTeG.

What’s given below is my guess based on 6 months testing of the existing trained Generator. We didn’t try any of below but I think it’s worth trying.

1) Testing effects of shocks in any single price category or assumption or any logical combination of them. The estimated impact will include all direct and indirect impact and it will be path-dependent (same 20% currency devaluation likely to have different impact whether it’s a first time move or market is correcting previous strength, whether it’s fast or slow etc)

2) testing counterfactual history with respect to change in any price category series, assumptions or any logical combination of them. Also including indirect impact and also path dependent.

Those two things above can be done on existing pre-trained generator.

3) Testing policy response (will require smarter initial variable selection and building new generator)

The approach to testing of any of the above is as follows:

1) generate baseline scenario (B) by focusing only on recent actual history and ignoring all the assumptions (zero weight)

2) change history and assumptions series as you like (past or future, one or many) and generate stress scenario (S) with the same weights on recent history as before, weights 1 on your altered variables and 0 weights on everything else.

3) Measure impact by subtracting B from S

Leave a Reply

Your email address will not be published. Required fields are marked *