[Chester](https://www.aight-bet.com/bots/chester) is a simple bot that uses [[Elo rating]] of teams to compute each team's chance of winning. Chester's predictions are intended to be used for [[Moneyline Bet|monetline bets]], however it's important to know that no predictive algorithm is perfect, and certainly not Elo rating alone. If you look at [Chester's accuracy](https://www.aight-bet.com/bots/chester) you're likely to see that it hovers around 50% or what you'd expect from a coin flip. This accuracy metric is all encompassing, meaning we take all predictions into account and treat them equally. For example, if **Team A has a 50.5% chance to win and Team B has a 49.5% chance to win**, the outcome is considered equal to a matchup where **Team A has a 98.0% chance to win and Team B has a 2.0% chance to win**. But does that matter? We can see Elo rating alone isn't sufficient to predict the outcome of games any better or worse than simply guessing. But what happens as the difference between Elo rating widens? Let's take all NCAA Football predictions as of September 7th, 2024, about a week after the season started. At this point, Chester made predictions for 75 NCAAF games, and predicted the winner correctly in 44 of those 75 games, or 58.7%. What happens if we filter these 75 predictions to only include cases where Chester predicted a team had a 60%, 70%, 80%, 90% or higher chance of winning? Did it do any better in those cases? In the following table, we've taken all 75 NCAAF Chester predictions where one side was between five threshold ranges and looked at accuracy for each of them. | Threshold (%) | # Predictions | # Correct Predictions | Accuracy (%) | | ------------- | ------------- | --------------------- | ------------ | | \[50 - 60) | 10 | 4 | 40.0 | | \[60 - 70) | 5 | 2 | 40.0 | | \[70 - 80) | 14 | 8 | 57.1 | | \[80 - 90) | 27 | 17 | 63.0 | | \[90 - 100] | 19 | 13 | 68.4 | As you can see from the table above the higher the delta between Elo ratings the higher the accuracy, which seams reasonable. So, should you trust Chester? Well, no if you're just blindly placing money line bets based on its predictions. However, using its predictions as a data point in your own analysis, especially when the percent chance of winning is quite lopsided, is certainly worth it.