We start by constructing simple systems that show consistent in and out-of-sample results. We require our models to make logical sense and not simply fit past data. We add refinements and then revalidate our systems with real-time performance validation before using them.
Our systems are a blend of fundamental, situational, and psychological factors. We look for teams having strong motivations to perform well. Mean reversion is also an important aspect of sports analysis.
We consider the motivations of the betting public as well, such as overconfidence, recency bias, and anchoring on past winners. Betting lines reflect public opinion. Much of our edge comes from exploiting behavioral biases of the public incorporated in the lines.