Inside the Black Box: The Simple Truth About Quantitative Trading

3.3. DATA-DRIVEN ALPHA MODELS

We now turn our attention to data-driven strategies, which were not included in the taxonomy shown in Exhibit 3.1. These strategies are far less widely practiced for a variety of reasons, one of which is that they are significantly more difficult to understand and the mathematics are far more complicated. Data mining, when done well, is based on the premise that the data tell you what is likely to happen next, based on some patterns that are recognizable using certain analytical techniques. When used as alpha models, the inputs are usually sourced from exchanges (mostly prices), and these strategies typically seek to identify patterns that have some explanatory power about the future.

There are two advantages to these approaches. First, compared with theory-driven strategies, data mining is considerably more technically challenging and far less widely practiced. This means that there are fewer competitors, which is helpful. Because theory-driven strategies are usually easy to understand and the math involved in building the relevant models is usually not very advanced, the barriers to entry are naturally lower. Neither condition exists in the case of data-driven strategies, which discourages entry into this space. Second, data-driven strategies are able to discern behaviors whether they have been already named under the banner of some theory or not, which allows them to discover that something happens without having to understand why. By contrast, ...