Using our Elo ratings, which serve as a measure of each team’s strength and update after each game, we simulated the remaining college football season 10,000 times. This was done for every matchup for every team across the sport, and we’ve separated out the result for each Power 5 conference below. Using our Elo ratings, we simulate the outcome of each game over and over as we progress through 10,000 hypothetical seasons. At the end of these 10,000 seasons, we take some averages and calculate some stats to see how many games each team is likely to win on average, their probability of going undefeated, and more.
One particularly interesting thing about our season simulations is that they are run “hot.” This means that as teams win and lose games in the simulated seasons, their Elo ratings change just as they would in real life. The reason for doing this is because it adds in some much needed variation to the simulations and allows teams to go on a run (or keep losing), increasing their Elo rating (or decreasing it) and thereby increasing (or decreasing) their probability of winning future games. It’s also a better reflection of real life. Nebraska does not play at the same level each and every game, and we want to adjust their Elo rating in the simulated seasons just as we do when the real games are played.
We ran this model preseason, and you can find those results here. Now that the season has started, we can continue to run our season simulations and only simulate the remaining games, taking the ones that have already been played into consideration with their actual result, and starting the simulation of all remaining games with every team’s current Elo rating.
One more note about the simulation before we get into the results. Our program just simulates the regular season games, not the playoffs or conference championship games. Simulating conference championship games is something that I hope to add in the future, but the playoff will likely not be added for a long time as picking the top four is not governed by a set of rules. We can (and hopefully will soon!) write the rules into code for how the two teams get chosen to play in each conference championship game. Once the top four is set at the end of the season, though, we can simulate out those games too.
Here are the results as of 9/7/2021.