The best practices of risk models–and model building–boil down to one thing: “we can’t forget the story behind it,” said
Went, a trained econometrist, described three main types of models. Fundamental models are based on rules relating basic variables such as interest rates and prices. Both the Black-Scholes model of options pricing and the Heath-Jarrow-Morton model of interest rates are fundamental models. Phenomenological models describe empirical observations of observed phenomena without knowing the deep-seated causes of such phenomena, but build on observed effects of direct causation. The option-adjusted spread (OAS) models used to price various fixed income instruments are phenomenological models. Statistical models use various statistical approaches to describe relationships across various variables, where the degrees of inferred correlation among variables determine the final components and the empirical specification of the model. Various mortgage prepayment models are statistical models that assume correlation but not causation.
When designing a valuation or risk model, it is important to look at variables that may not be immediately apparent, such as liquidity, trading characteristics, and dispersion spread. The modeller must formulate the type of mathematical relationship, such as whether it should be a stochastic or a deterministic model. And, the modellers should remember that correlation is not causation.
Modeling best practices suggest that those entrusted with building various models–from the simplest valuation models to the most complex models to that integrate various functions–should continually ask questions throughout the preliminary design phase: “Could we use another model?” and “is the model appropriate as intended?” Overall, a healthy dose of skepticism is preferable. And, quants often fall into the complexity trap – using a more complex model than necessary. While this may not always be a poorly executed strategy or lead to questionable outcomes “falling in love with the model” removes the objectivity of the modeller.
Similarly, once a model has been specified and tested, but before full implementation, the model, its components and its structure need to be verified, stressing the model parameters in the model to find its limitations. “And once you find a good model, test, test and re-test.” In order to embrace the best practices in model risk governance [more about this in the other speakers’ presentations], an organization must clearly recognize where the responsibility for models, and for model risk, occur.”
Creating a good understanding where the model lies within the business processes of the institution. It makes a good model a great model. Since many models may only end up warehoused in a model library, their direct usability must be a guiding yardstick.
Went urged webinar participants to read Derman and Wilmott’s Modelers’ Hippocratic Oath. The manifesto of the oath reads in part: “We do need models and mathematics–you cannot think about finance and economics without them–but one must never forget that models are not the world. Whenever we make a model of something involving human beings, we are trying to force the ugly stepsister’s foot into Cinderella’s pretty glass slipper.”
Modeling is an intellectually challenging and highly rewarding exercise. “Sometimes it’s easy to lose perspective,” Went said, speaking from experience. For this reason he recommended application of Occam’s razor. Simplicity and economy should prevail. “You may be measuring a complex phenomenon but the model should not aim to be complex.” ª
The webinar presentation slides can be found at: http://event.on24.com/r.htm?e=616631&s=1&k=3703B9EBC7B794F759F83FD5D47C34ED>