Just what are we optimizing, anyway?

The best techniques for optimizing financial market prediction models are generally very different from the best techniques for optimizing models of physical processes. Nevertheless, educational tradition and work experience lead many or most developers of automated trading systems to fall back on engineering solutions rather than to implement methods better suited to the task at hand. One of the most common yet most easily corrected errors is choosing an unsuitable optimization criterion. This paper explores the weaknesses of some of the most frequently used optimization criteria and proposes some alternatives that can be more appropriate when the goal is to develop a profitable algorithm for automated trading of financial markets.