How do quants manage model risk of complex exotics on the OTC market ?That's maybe one of the most pernicious risk and I would like to know how people handle that.I have found this paper Theory and practice of model risk by Rebonatobut I would like to hear other comments.Some academic people talk about "model averaging" but it seems very naive in my view. Piterbarg wrote a bit about thatMixture of Models: A Simple Recipe for a ... Hangover?I think that the worst case Uncertain Volatility Model is a very nice way of giving a upper bound to your model risk.Any comment is welcome as usual !

Just a comment on the Uncertain Voaltility Model as a bound. Since it is a diffusion model,I suppose it must satisfy the general diffusion model bounds for barrier options discussedin "Robust hedging of barrier options" by Brown et al. But ... these bounds are easilyviolated by jumps. So, the UVM bounds will also be violated.

Well you can easily check the risk of a model by trading with it.

That's what Leland, O'Brien and Rubinstein thought!

Concerning "model risk" there are one or two interesting papers by Rama Cont (check his web page).Regards

thanks The BridgeI have found this paper which indeed sounds very interesting !MODEL UNCERTAINTY AND ITS IMPACT ON THE PRICING OF DERIVATIVE INSTRUMENTSAnother interesting papers quoted this one. Morel risk adjusted hedge ratiosModel Uncertainty and derivatives pricingAnd how quants do in the banks ?

Another nice paper on model uncertainty (more theoretical).The impact of uncertainties on the pricing of contingentclaimsWe study the effect of parameters uncertainties on a stochastic diffusion model, inparticular the impact on the pricing of contingent claims, thanks to Dirichlet Formsmethods. We apply recent techniques, developed by Bouleau, to hedging proceduresin order to compute the sensitivities of SDE trajectories with respect to parameterperturbations. We show that this model can reproduce a bid-ask spread. We also provethat, if the stochastic differential equation admits a closed form representation, alsothe sensitivities have closed form representations.We exhibit the case of log-normal diffusion and we show that this framework foreseesa smiled implied volatility surface coherent with historical data.

"How do quants manage model risk of complex exotics on the OTC market ?That's maybe one of the most pernicious risk and I would like to know how people handle that."It is some very basic. So first one need to define zero-risk (perfect) model. Next step is to define less or equal model risk or find it quantitative measure. Third, check how given definition work: say define model risk for the popular models for example, geometric Brown motion or binomial scheme. Fail to perform such steps to certain degree suggests fail of the idea.

The main problem is because there is NO perfect zero risk model That's why model risk is very hard to quantify . We don't know the how the market will evolves. We can make studies of the impact of the uncertainty on paramters of the model but what if the parametric form of the model is false by itself?

QuoteOriginally posted by: frenchXThe main problem is because there is NO perfect zero risk model That's why model risk is very hard to quantify . We don't know the how the market will evolves. We can make studies of the impact of the uncertainty on paramters of the model but what if the parametric form of the model is false by itself?It might be better to think about" risk model" as a quite emotional feeling.

And it's a circular never ending problem I use a model to model my model risk but what if my controller model is false ? That's a very hard problem I think.