This question is most relevant to the evaluation of embedded options, such as the refinancing option granted to borrowers in the mortgage and bank loan markets, or the call option present in some corporate bonds, than to exchange traded options.
Much more so than market participants in exchange traded options, which have purely market-based incentives, the borrowers and issuing entities which are granted embedded options often employ sub-optimal (irrational) exercise strategies. Sometimes, these only appear irrational to the lender/investor, when, in fact, they are optimal when considering other factors such as the borrower's personal financial condition or other factors (such as a homeowner's desire to move to a new house). In those cases, the difficulty from the modeler's perspective is that these factors are unobservable, and, hence, must be modeled as random. Other times, decision makers make truly sub-optimal decisions based on a private evaluation of the value of exercise that differs from a "correct" market-implied decision rule.
Which modeling approach leads to better predictions and better relative value measures?
- Rational: Option-holder follows a fully rational and optimal exercise strategy when deciding whether or not to exercise his option. The modeler attempts to reproduce the holder's payoff function as faithfully as possible.
- Behavioral: Option-holder follows a simple reduced-form exercise strategy which may lead to sub-optimal decisions. The modeler attempts to estimate the parameters of the exercise strategy, either from market data or from historical experience.
Under the rational approach, how do you treat unobserved characteristics? Does the presence of unobserved characteristics make the rational approach ultimately equivalent to a market-implied behavioral approach? Is it even possible to fit the unobserved information directly to market data?
Open-ended Bounty Offer: We may not yet have a broad enough user base familiar with the pricing and modeling of embedded options to adequately answer this question. As such, I pledge to offer a bounty of 100 points to any user who can adequately answer this question. If you are a new user and you have come to this question long after activity has died down, then so long as I am still active on this site my offer remains in effect.