This question is related to the discussion on Model Validation Criteria However it appeard to be very high level to me and I would like to go more into detail.

Not working at a pricing desk the following has always puzzled me: How are quants able to verify whether their calculated prices are any good? (by "calculated prices" I mean the ones output by some model - e.g. B&S, Heston )

One could perhaps distingusih the two:

  • A priori
  • A posteriori

I would assume a priory mainly consists of model choice (fitting the market data etc.) - Thus the A priori part is related to this question.

Alas, I am quite clueless what an a posteriori check would look like. What would be the benchmark one would check against ?

  • 3
    $\begingroup$ In a risk-neutral setting the P&L should be the definitive a posteriori verification, right? $\endgroup$
    – user1157
    Feb 23, 2014 at 12:14
  • 1
    $\begingroup$ When you say their calculated prices, you're referring to the values that are output from models (like pricing an option with BS), right, rather than the bid/ask prices quoted in the market? $\endgroup$
    – John
    Feb 23, 2014 at 14:43
  • $\begingroup$ correct - I will edit the questipn - this is a necessary specification :) thank you $\endgroup$ Feb 23, 2014 at 14:44
  • 4
    $\begingroup$ On the sell-side, one could argue that a 'good' model is one whose a posteriori replication PNL exceeds the quoted margin, or perhaps one that allowed you to sell the deal in the first place (depending on your level of cynicism). On the buy-side, assuming you aren't hedging, then a posteriori, the models which identify options that are mis-priced under P are the good ones. There's a big difference between models which are widely used, models which match the market well, and models which capture asset dynamics accurately. papers.ssrn.com/sol3/papers.cfm?abstract_id=2365294 $\endgroup$ Feb 23, 2014 at 19:04
  • $\begingroup$ thank you for the paper. Do you have any more references on the topic. Please write your comment as an answer. It was already a great help - I would like to upvote it and for it to be visible to other. cheers $\endgroup$ Feb 24, 2014 at 8:36


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.