Hi,Just want to know in other banks, when ur qaunts provide a new pricing model, how could they make sure the model is "correct" ? I lately encounter this problem as we provided a new model for FX exotic options. Our audits and mid-office asked us to proof "the model is correct". However I have no idea about what the "correct" means. So I ask them, they said we can compare our price to market quote. But I doubt that. For liquid markets,say G3, maybe they are right. Since there are lot of quotes in market. However for some emerging markets, there are rare quotes in markets. How can we take market quotes as a judegement for model ? Also, how can we make sure the market quote is "correct". In my opinion, there is no "correct model ", the only thing u can say is you running books with the mdoel. And u didn't hurted by it after a long run.So I am interested in other company's procedure. When u have to replace a model which may make the position value changed. What should a quant do? Should he provide any simulations ? market comparisons ? or something to convince the mid office or aduit (that's my second question. Who will be the umpire?) Your advise is highly appreciated.dinner

If you derived your pricing model from an existing one, or a combination of existing theories, you might want to present to them the supporting papers.

If the market is that difficult to model, there should be P&L recognised anyway and it should be held out as EITF reserve. Perhaps you should ask why they are uncomfortable.

QuoteSo I ask them, they said we can compare our price to market quoteIf the model agrees with the market then it is a model of the market not the "correct" price. If that is all the model is intended to do, why would you trade it? Why not just take the market as "correct"?I would think that the purpose of trading is to make a profit. In essence that implies you have a better idea of the "correct" price than the market. A consistent profit is the best evidence of that, but it isn't a proof, since it could come from chance.To my mind, that means that a consistent profit may be necessary but is not sufficient. Just as in regular science, the model needs a theory behind it that makes testable predictions (and in more areas than just price). Those predictions can be verified or falsified. If it doesn't make such predictions, (or there is no rational theory) then it isn't a model that can even be tested. Proofs are something for mathematics and logic only.

You should understand yourself why the p&l has changed and, if appropriate, be able to make that go away - is it that the old parameters have a slightly different meaning in the new model (so they need to be re-marked), or have you introduced new ones that are affecting the p&l (so they need to be marked), or something else?

Hi Fermion I agrees that a consistent profit maybe a best evidence of "correct model". However as u noted, it's hard to said whether the profit comes form chance or model. Also, it need a long run to see it. So I thinked only a quant can do is to provide proof on mathematics and logic.However,I don't think a math proof can make them(mid office or audit) comfortable. Because they may not understand the math theory. They are afriad the wrong price may not include all the risks in it thus a trader may be hurt. So from a banks' points of view, how to set up a procedure to make everyone comfortable?

QuoteOriginally posted by: dinnerHowever,I don't think a math proof can make them(mid office or audit) comfortable. Because they may not understand the math theory. They are afriad the wrong price may not include all the risks in it thus a trader may be hurt. Risks can be knowable or unknowable. There's nothing any model can say about unknowable risks, other than not to trade (which has its own risk). So we need to confine ourselves to knowable risks. What the model/theory says about these, if anything, is a matter of math (and the labor required to do it).So, I would suggest the obvious: whatever doubts are expressed about particular risks, supply the math that covers it, or admit that it hasn't been analysed. If they don't understand the math then you have to explain it or convince them by the force of your personality or by usng office politics. That is how things get done isn't it? As far as the objective approach of reason is concerned, I don't think there are any short cuts to a proper mathematical analysis at whatever level of rigor makes them comfortable. But reason may not be what they are looking for, particularly if they have already confessed unwillingness to understand the math. (In which case it is probable that they have already adopted an irrational course and are probably hoping for decisions to come to them by magic or divine intervention .)

Dinner,I have faced exactly the same situation from a mid-office point of view or an accounting pov.If you are facing a market that is illiquid or exists only on the client side but not as a interbank market, you have to recognise it as such and make adjustments to pnl (what on calls fair value adjustments).In the absence of these, if you terminate or back out a trade done sometime back and your model pnl is very different from the termination pnl, you get a jump which is difficult to justify and account for.Hence in a illiquid/unobservable market one tries to obtain quotes from other market participants once in a while to make sure the model after adjustments does not remain very out of tune with the market in terms of portflio liquidation value.I guess with IAS 39 these concerns are all the more vital now.

ultimately, the judge is the market. if what the model tells you to do doesn't make money in the long run, then the model is not right.

Yup the market's the judge.If you have model that always shows that you're in the money, no matter how theoretically correct it may be, if it is unrealizable (you won't be able to unwinde in the desired price), your model is useless even if the market pricing is not correct. If your model shows a price of 101 but everyone in the market uses a wrong model that gives a price of 90... you'll never going to realize your 101. Unless you convince someone to make a deal at the said price.As a practitioner, I make it a point to calibrate my models to reflect market prices at times.

QuoteOriginally posted by: jomniYup the market's the judge.If you have model that always shows that you're in the money, no matter how theoretically correct it may be, if it is unrealizable (you won't be able to unwinde in the desired price), your model is useless even if the market pricing is not correct. If your model shows a price of 101 but everyone in the market uses a wrong model that gives a price of 90... you'll never going to realize your 101. Unless you convince someone to make a deal at the said price.As a practitioner, I make it a point to calibrate my models to reflect market prices at times.If you truly believe your model is correct and it tells you the option is worth 101, and the market says its worth 90. Why dont you go and and buy every option you can at 90 and use it as a "cheap" hedge for your book??

Guess traders always do that.And accountants take away using FVA !!!

Are you sure your mid-office guys are asking you to show your model is correct?They are usually more interested in marking the books to market rather than proving the internal consistencies of a model.

QuoteOriginally posted by: FermionQuoteSo I ask them, they said we can compare our price to market quoteIf the model agrees with the market then it is a model of the market not the "correct" price. If that is all the model is intended to do, why would you trade it? Why not just take the market as "correct"?I would think that the purpose of trading is to make a profit. In essence that implies you have a better idea of the "correct" price than the market. A consistent profit is the best evidence of that, but it isn't a proof, since it could come from chance.Fermion if a model agrees with the market then it gives you some confidence that it might be giving reasonable values for your off-market positions. In general only a tiny part of a derivatives book will be directly comparable with a market traded instrument. For example if you had a better way to model the vol smile in an option market where all the observable prices were at-the-money strikes your model should still agree with the market. If certain deals have radically different valuations under the old and new models I think some investigation is needed to give some supporting arguments as to why the new model is better. You should also be able to show that your new model gives the correct result in simplified cases when, for example, the deals are simplified in some way, or some parameter takes on an extreme value.

QuoteOriginally posted by: achillesQuoteOriginally posted by: jomniYup the market's the judge.If you have model that always shows that you're in the money, no matter how theoretically correct it may be, if it is unrealizable (you won't be able to unwinde in the desired price), your model is useless even if the market pricing is not correct. If your model shows a price of 101 but everyone in the market uses a wrong model that gives a price of 90... you'll never going to realize your 101. Unless you convince someone to make a deal at the said price.As a practitioner, I make it a point to calibrate my models to reflect market prices at times.If you truly believe your model is correct and it tells you the option is worth 101, and the market says its worth 90. Why dont you go and and buy every option you can at 90 and use it as a "cheap" hedge for your book??If the purpose of a model is just to model the market, you wouldn't need the model in the first place. However, a model is used in order to make statements about sensitivities and risk. The model is to be judged on these grounds, namely whether model sensitivity predictions can be reconciliated when compared with actual market moves. And the theoretical justification is as follows. Assume a single underlying - volatility - evolve in a continuous manner and that the book is delta hedged with respect to the risk factor but assuming a constant volatility (i.e. your hedging model is captured by this parameter). Then the value of the book evolves according to a (deterministic) ODE with time-varying parameter that is proportional to gamma times the difference between realized volatility and the constant volatility parameter. This is a well-known result that can be generalized to a multifactor setting, but the main lesson is that misspecification of your model (i.e. the volatility parameter) will result in daily losses proportional to the timestep, and not stochastic. I.e. wrong models cause bleeding, not f**k-ups. So what you can and should do is to make sure that your model is able to explain realized price changes over a sufficient period of time - and this will also be your way to make sure that using the model will lead to a positive pnl.Of course assumptions can and should be considered. Especially if the market in question calls for jump diffusions in your specification (i.e. the assumption of continuous evolution is misspecified), then the value of the book will have an additional element stemming from the jump process. However, it is important to stress that this element comes not from "the market believing the product should trade at 90, not 101", but rather from the fact that jumps is a proper description of the underlying!