Model Performance Overview

How the Model Works

This college football prediction model is built on math, statistics, and play-by-play analytics. For every game, it runs thousands of Monte Carlo simulations using only its own database. It has no knowledge of Vegas lines, expert picks, human perception, injuries, weather, or late roster changes. All of which a human or professional may have, which should provide an advantage over the model.

Performance Against the Spread (ATS)

The model consistently beats the 50% critical threshold for picking games against the spread (ATS). Here are some highlights:

Beating the “coin flip” line (50% ATS) over hundreds of games, without any market or human input, shows real predictive power.

Winner Accuracy

The model also does well at picking outright winners:

Win Probability Calibration

The model's confidence is insightful, especially for high-probability predictions. Here's how often the model was correct in each win probability bracket:

Cover Probability Calibration

The model's confidence in covering the spread does not strongly correlate with actual results. Although it still hits over 50%, increased confidence does not appear to be meaningful.