Purpose of modelling - i'm serious
Posted: Wed Nov 07, 2007 7:41 am
[b][i]how do we determine whether a given model is good or bad ?[/i][/b]
the property of goodness is the norm (maybe not the usuaul norm) of a vector whose components represent numerical measures of "Tractable/Easy to Use", "Intuitive", "Stability of Parameters*", "1/Replication Friction"
Tractability is important because, for example, all else equal, one would rather have a formula than an algorithm.
Intuitive is important because it is hard to trust models in which intuition is lacking. It's also hard to explain such models to CEOs and other people like Paris Hilton.
Stability of Parameters means that the day to day or tick to tick movement of implied parameters isn't wildly unstable. Ok, implied vol isn't constant, but it can be reasonably well studied/modeled by other relatively simple stochastic processes.
Replication Friction is the cumulative loss due to hedging errors associated with contract replication. The smaller this number, the better. So, I take the reciprocal of that number.
the property of goodness is the norm (maybe not the usuaul norm) of a vector whose components represent numerical measures of "Tractable/Easy to Use", "Intuitive", "Stability of Parameters*", "1/Replication Friction"
Tractability is important because, for example, all else equal, one would rather have a formula than an algorithm.
Intuitive is important because it is hard to trust models in which intuition is lacking. It's also hard to explain such models to CEOs and other people like Paris Hilton.
Stability of Parameters means that the day to day or tick to tick movement of implied parameters isn't wildly unstable. Ok, implied vol isn't constant, but it can be reasonably well studied/modeled by other relatively simple stochastic processes.
Replication Friction is the cumulative loss due to hedging errors associated with contract replication. The smaller this number, the better. So, I take the reciprocal of that number.