Nerds on Wall Street: Math, Machines, and Wired Markets

8.6. Maximizing Predictability

Much of published material on the use of genetic and evolutionary methods in trading focuses on using the past history of prices to predict the future, as in technical analysis. This is an age-old approach. For much of the twentieth century, it was considered a legitimate area of academic investigation. Strategies based on daily stock prices alone were shown to be futile in the 1960s, and they fell into academic disrepute. This hasn't stopped people from using all sorts of tools and techniques for ever more complex forms of technical analysis. Neural nets are currently popular in this regard. This is simple in terms of the data needed, and popular among active individual investors and traders. This has resulted in a somewhat skewed perspective on the topic of financial prediction in the EC and GA literature.

In 1992, I had joined First Quadrant, an institutional investment management firm^[] that applied quantitative forecasting techniques to a variety of investment strategies. Figure 8.3 revisits the idea of maximizing predictability introduced in Chapter 5 and is a high-level perspective on maximizing predictability in finance.^[]

To expand on the previous discussion, the key point is that there are three central decisions to make in financial prediction:

1. What to predict. You can choose to predict the returns to an asset class, such as a broad market or industry group, an exchange rate, interest rates, or returns to individual securities of many ...