Insights through Data, Research, and Analysis
Relative Value
The model seeks to find an intrinsic power ranking for each team using play-by-play data, game drive statistics, and detailed box score data. The model converts the betting line into fundamental expectations for each team and then compares these expectations against the QuantCappers intrinsic ranking for each team. The difference between the public's expectation (the betting line) and the QuantCappers ranking is the relative value. Where the difference is large, you have the opportunity to profit.
The QuantCappers Model
The model analyzes play-by-play data (variables) and assigns a point value to each play. Points are based on the context of the play (down, distance, score, opponent, etc.) relative to the most likely outcome for that play. For example, a defense that allows a five yard gain on 3rd and 6 will rank higher in the model than a defense that allows a 3 yard gain on 3rd and 2. This is why standard metrics such as yards allowed and yards allowed per play can be misleading.
The goal of analyzing play-by-play, drive stats, and box scores is to develop factor scores and an intrinsic value for each team. The QuantCappers model uses the factor scores to generate Monte Carlo simulations and output density forecasts. The model is specifically designed to predict how a team will perform against an average opponent on a neutral field. For example, the Chicago Bears will outperform the average opponent on a neutral field by 4.5 points over 10,000 games.
Statistical testing was used to determine which variables to include in the model. Variables such as turnover differential, red zone performance, defensive touchdowns, 3rd down performance, and blocked kicks (for example), explain past performance but are not particularly valuable as predictors of a team's future performance. The QuantCappers model instead focuses on predictive variables. These are primarily efficiency based metrics (per carry / per attempt metrics) in low leverage situations (1st and 2nd down, first half of games, close scoring games, etc.). Beginning-of-year intrinsic valuations are derived from forecasts that reflects prior years’ on-field performance as well the NFL draft, trades and acquisitions, front office and coaching changes, injuries, scheme changes, and team continuity and other qualitative changes. Prior to the availability of play-by-play data, drive stats and box score data are used in an expected points per drive model to make projections.
Market Inefficiencies
The QuantCappers model adjusts for market inefficiencies, such as:
- Based on the explosion of fantasy sports, teams with superstar players tend to be overbet.
- Similar to the Fama-French model for stocks where small companies tend to outperform larger ones, a home-dog affect remains an under exploited opportunity in NFL games.
- Weather and injuries, especially late in the season, are typically not "priced" into games.
Model Development
The QuantCappers model was developed using a formal research and development process that began with a set of theories and assumptions. We then acquired and built a data repository, back-tested (and in some cases rejected) our theories using historical data, and constructed the working model. We continue to perform research to identify and incorporate new factors that will improve the model’s performance.