Inside the Black Box: The Simple Truth About Quantitative Trading

10.1. MODEL RISK

Model risk is the most basic form of risk any quant system brings to an investor. Models are approximations of the real world. If the researcher does a poor job of modeling a particular phenomenon—for example, momentum—the strategy might not be profitable, even in a benign environment for momentum in general. In other words, model risk is the risk that the strategy does not accurately describe, match, or predict the real-world phenomenon it is attempting to exploit. Worse still, model risk need not be evident right away. Sometimes small errors in specification or software engineering lead to problems that accumulate very slowly over time, then suddenly explode on a busy trading day. Additionally, model risk can come from several sources. The most common are the inapplicability of modeling, model misspecification, and implementation errors. It bears mentioning that all types of model risk can occur not only in the alpha model but also from errors in any of the other parts of the strategy. Back-testing software, data feed handlers, alpha models, risk models, transaction cost models, portfolio construction models, and execution algorithms can all have model risk in them.

10.1.1. Inapplicability of Modeling

Inapplicability of modeling is a fundamental error that comes in two forms. The first is the mistaken use of quantitative modeling to a given problem. For example, trying to model the quality of a musician is simply the wrong idea from the start. One could conceive ...