On balance expert systems trump human experts, hence the drive to make trading more systematic and mechanical. The problem is that Building Winning Algorithmic Trading Systems, the title of Kevin J. Davey’s new book (Wiley, 2014), can be tough. Davey recounts his sometimes gut-wrenching journey “from data mining to Monte Carlo simulation to live trading” and provides traders with useful information that will help them avoid his mistakes.

The author joins a rather small fraternity of systems developers who have shared their thoughts, for better or worse, with the reading public. I think here—and this list is in no way meant to be exhaustive—of Howard Bandy (Quantitative Trading Systems), Tushar Chande (Beyond Technical Analysis), Urban Jaekle and Emilio Tomasini (Trading Systems), Perry Kaufman (Trading Systems and Methods), Robert Pardo (The Evaluation and Optimization of Trading Strategies), and Thomas Stridsman (Trading Systems That Work).

The strength of Davey’s book is that it covers the entire process of designing, developing, testing, trading, and monitoring a system. It also includes Easy Language code for three sample strategies, and on the password-protected companion website (the password is given in the book) there are five helpful spreadsheets.

I was particularly taken with Davey’s methods for reviewing the performance of his systems. One “simple,” short-term method assumes that trade results have a normal distribution, which admittedly seldom happens. “To alleviate this concern, we can simply take the Monte Carlo results from numerous runs and use percentiles based on them. This will provide a more accurate representation of the expected bounds of the trading system.” (p. 214) But, whichever approach and whatever parameters the trader uses (the author prefers the Monte Carlo version), the graph will look roughly like a probability cone. With the Monte Carlo version, a trader can include “boundary condition” effects, “like quitting after a certain percentage drawdown.” (p. 216) “For example, if the real-time performance of your strategy falls below the lower 10 percent line, it could mean that your system is no longer working. After all, the odds were 90 percent that your strategy should be working better than this.” (p. 215)