Model Risk


New Member
Hi David

I dont have core readings for this chapter.i am unable to see clear picture of this concept from your notes.

Model risk does not arise from a discrepancy between the model value and the ‘true’ or intrinsic value of an instrument but rather due to a discrepancy between the model value and the value is recorded for accounting purposes. Model risk is linked to trading-book products that are marked-to-market on a daily basis.

Plz help me out in understanding the will be great if u give some real life example.


David Harper CFA FRM

David Harper CFA FRM
Staff member
Hi Anil,

Observant as always, you will be keeping me on my toes, for which i am grateful :)

But please first note, GARP retained this AIM from a discontinued reading. Even if you had the assigned core reading (K. Dowd), you would not find this AIM. Last year, the FRM shifted from the previous, philosophical reading (Rebanato) to the currently assigned, more practical, Kevin Dowd reading. Yet this AIM was retained from the old Rebanato (probably mistakenly); his reading on this, frankly, is quite disputable (I only mean debatable, I don't mean wrong) in regard to model risk definition. The currently assigned Dowd is "safer" because on this point in particular, the AIM could give rise to a long argument instead of a conclusive answer.

Therefore, your mention above is not a particularly important sub-point to this AIM. The test question will almost surely be along the "practical lines" related to EMH: i.e., if you believe in EMH, you want to find a better model because under an EMH belief, such a Platonic ideal-model exists to describe the single intrinsic value of an instrument. If you do not follow EMH, you do not need to think an instrument has an intrinsic value and therefore, your attention naturally shifts to, how do participants (even behaviorally) use their models?

But, okay, to your question. A great example is the current raging debate over fair value; e.g., FAS 157. Here is a nice summary and the 6-page PDF is, imo, a great example of what Rebanato means by the above. Specifically, note how Micheal Young (and maybe fair value contrarians) are not really saying "CDO models can't really find the intrinsic value" but rather, here is the Rebanato point, "models need inputs that are 'unknowable' or 'imponderable'" and therefore, different model users will predictably produce different values - there must be a difference between "volatile" model values and recorded (market to market values). Model risk here is not that a user will abuse a model, but rather than a correct model cannot be found and therefore different (even expert) users will necessarily find different model answers. In such a world, what can be the mark to market (recorded) value of an opaque instrument like a CDO tranche?

The reason in particular, IMO, this AIM is dubious on this aspect is arguably we have commingled the liquidity risk topic. I hope that's not too much.