Point Spread Space
The Algorithm
In this article we will discuss the point spread algorithm. Let's begin by talking about the two main parts of the code: The ratings estimator and the point spread bet selector. The ratings estimator is none other than the plain old ordinary ELO rating system that is used in chess matches, and with this we can go from zero knowledge to about 52% accuracy in predicting the games. This gives us a good starting point but it will not win money at Vegas, so we need an edge. That's where the point spread bet selector comes in, looking at all the possible bets and telling us which ones to actually wager.
The bet selector has three confidence factors and if all three are above their separate threshold values, then the game is declared a "bet" game meaning that we should bet on it, otherwise the game is declared a "nobet" game meaning that we should not bet on it. The overall results after wagering appear to be about 55% to 60%, enough to win money at Vegas.
In every game played, each team begins with an ELO rating. New teams are given a rating of 0 to start. At the end of the game the ratings are adjusted based on the outcome of the game. This adjustment depends on the rating difference between the two teams. If both teams are about the same strength, then the winner gets around 20 points and the loser loses around 20 points. If the teams are greatly different in strength, then the points are awarded according to the difficulty of the win. For example, if a stronger team defeats a much weaker team, it may only receive 3 rating points, and the weaker team will only lose 3 rating points. If, however a much weaker team defeats a much stronger team, then the weaker team might receive as many as 40 points while the stronger team would lose 40 points. This way expected wins don't change things much, underdog upsets reap strong rewards, and equal games are treated somewhere in between. The actual points awarded is an exponential function of the rating difference.
The point spread bet selector works on a very simple concept. It may sound unlikely or strange, but it works and I have made and continue to make money in pretend wagers from it. Vegas sets the line, or the predicted point spread. The program has made its own prediction of what the point spread will be. The secret is that by examining the history of these two predictions for accuracy, we can determine a good bet from a bad one. This is done by three confidence factors, each of which is an estimate of the local successfulness of each proposed region in point spread space. So by looking at the success of past bets we can predict the success of future bets.