Equity markets and computational intelligence

Abstract

I propose a new characterization of the types of problems for which computational intelligence (CI) tends to be used, namely the identification of approximate abstractions. I then suggest that equity markets provide a challenging example for CI. Because markets are inherently adaptive, they pose a more difficult problem than traditional CI domains. I discuss my experience teaching a CI class that took the development of stock trading systems as a theme. A simple genetic algorithm to generate a trading strategy was developed as a class example. Although the astonishingly good results it achieved were due at least in part to data snooping, a simple unevolved version of the same strategy was almost as profitable. Yet it too had subtle data snooping problems—showing how difficult it is to avoid data snooping entirely, especially in adaptive domains.