Model risk provisions February 18, 2010 at 9:12 am

Zero hedge picks up on some interesting information at the end of an Economist article:

JPMorgan Chase holds $3 billion of “model-uncertainty reserves”

That number feels reasonable in the context of a bank with nearly a hundred billion of capital. But I’d love to know how they got it past their auditors…

2 Responses to “Model risk provisions”

  1. I happen to work on model-uncertainty calculations at one of JP’s competitors. Typically, these reserves have a model all of their own. However, the reserve model is much simpler vis-a-vis the “underlying” model. Further, it is not just the model that is bound up in the “model-uncertainty reserves”, but a number to account for the uncertainty around unobservable inputs to the underlying model.

    Depending on the complexities, market commonality, etc, of specific products in question, the reserves held can be rather pessimistic or something more optimised. From my limited experience, auditors take the same approach with reserve models as they do the underlying models.

    So the number isn’t just “3 billion”, but it can be decomposed into its parts. I imagine there are strong competitive incentives to not expose which business-lines are consider to have weak models.

    Cheers.

    PS: Thanks for the work you put into the blog; glad to finally be able to contribute.

  2. Thanks for the comment – much appreciated.

    You are of course correct: that $3B is made up of many individual valuation adjustments, model risk reserves, and parameter risk reserves I am sure. Moreover paradoxically it may well be that the businesses with the highest reserves are the safest. In my experiences those business heads that think carefully about model risk and are prudently reserved for it aren’t often the ones you have to worry about – it’s those who argue that (say) base correlation is known and stable that are more concerning.

    The fundamental issue of auditing remains, though. It is very hard for external auditors to have either the expertise necessary to audit these numbers or be paid enough audit fees to spend the time necessary even if they did understand what was going on. So the sense in which this is a truly audited number rather than (perhaps, on a bad day) a honey pot is interesting.