Excel LADZ - Weekly Update: Thursday, 7 September

G’day lads, I’m planning on releasing an NFL individual match simulator, as well as a full-season NFL model soon. Have a read lads!

Using ELO Ratings

A common approach to model NFL games is to use the ELO method. ELO Ratings are essentially an overall power rating of a team, relative to a league average. For example, a team with an ELO of 1700 is considered “stronger” than a team with an ELO of 1500. Using the below formula, it is possible to derive the Win % of a team in a matchup. Of course, you must have both team’s ELO Ratings to use the formula:

I recommend reading the following article: https://fivethirtyeight.com/methodology/how-our-nfl-predictions-work/

It’s a comprehensive explanation of 538’s approach to modelling NFL games & seasons, based off of their NFL ELO Rating System. While a great model for finding the win probability for each team in a matchup, it has a particular drawback.

That is, it doesn’t actually have the ability to simulate the points scored in games themselves. Consider a matchup between the Dolphins and the Cowboys. All the model is capable of doing is assigning the Dolphins and the Cowboys a win probability for that match, e.g. Dolphins 57% and the Cowboys 43%.

This is a common theme among all ELO models. The ELO Ratings are only fit for zero-sum, two player games. As a result, an ELO model can never come back to the user and return:

Simulation 1: 27-13 Dolphins Win

Simulation 2: 20-26 Cowboys Win

…

Simulation 10,000: 31-18 Dolphins Win.

I love to use ELO Ratings for knockout tournaments such as the World Cup, as the points/goals scored in an individual game don’t actually matter. All that matters is who progresses through to the next round/stage. However, in regular season games it’s useful for a model to be able to produce simulated points scored. This makes the model as realistic as possible.

Solution: ATT & DEF Ratings.

The solution to this problem is by creating a separate ATT & DEF Rating for each NFL team, relative to the league average. For example, the Dolphins may have an Attack Rating of 1.07, meaning they score 7% more points than the average team. On the other hand, they may have a Defence Rating of 1.09, meaning they concede 9% more points than the league average team.

In short, it involves collecting all the stats directly relating to a team’s offensive and defensive scoring capabilities. Through comparing these figures to the league average, you can come up with an Expected Points figure for two teams playing each other. For example, to see how many touchdown points would be expected, see the formula below:

Expected Touchdowns = ATT Rating * Opp's DEF Rating * League Average Touchdowns

To get access to these stats, have a look at the website ‘Pro Football Reference’ here: https://www.pro-football-reference.com/

To find the win percentage of a team, simply distribute the Expected Points along a probability distribution (i.e. using the BINOM.INV function to simulate Touchdowns, Field Goals, etc) where you’ll be able to grab a team’s probability of scoring more than their opposition, and vice versa.

Finally, you would then add in adjustments to the model. For example, defining a home-field adjustment, including a team’s schedule difficulty, and a quarterback factor (e.g. weakening the Jets if Rodgers is injured).

EPL Models

The ATT & DEF process explained above is similar to the EPL model. In the next few weeks I’ll be building off of the original EPL match model, and using stats to explore markets such as under/over goals, corners, cards, etc.

AFl & NRL

I’ve just released the AFL model onto Patreon, which simulates the entire 2023 Finals Series. The NRL model will be coming very shortly as well, and this will do the exact same thing as the AFL model.

Thanks for reading lads, if you have any questions let me know here on Patreon or on Discord.