My fantasy team uses a playoff system that Bill Simmons suggested a few years ago. In order to study it, I'd like to run Monte Carlo simulations of last year's post-season. To do so, I need to be able to compute the win probability for a pair of teams given the site of the game. Note this is not a prediction against the spread, just the orders of one team winning. This format also means that I can't use Vegas's point spreads, money lines, etc. because I need to be able to make predictions for pairs of teams that didn't meet in real life.

I thought that it would be easy to find a suitable model online, but I haven't been able to. Here's what I've tried:

  1. AdvancedNFLStats: In principle they supply all the needed information but I've been unable to reproduce their results. You can see a long post from me at the end of the week 7 team efficiency page that has gotten no replies. I would love to be able to get this model working.
  2. NFL Predictions uses stats from AdvancedNFLStats to run Monte Carlo paths through the post-season in some way that's slightly different from what AdvancedNFLStats does, but I can't find any info on the methodology.
  3. Simple Ranking System a.k.a. simple rating system: is focused on predicting margin of victory, not probability. More info here. The creator gives a formula for computing win probabilities here, but it's not clear to me that the same constants can still be used.
  4. Jeff Sagarin's ELO ratings: also aimed at point spreads, not probabilities

This page seemed promising at first but I didn't get anywhere following up on all the links.


The comments section on this page from Advanced NFL stats describes how to convert that website's weekly GWP values into win probabilities. A commenter named James writes:

The equation is: (H*(1-A)*F) / (H*(1-A)*F+(1-H)*A*(1-F))

Where H is the GWP of the home team, A the GWP of the away team, and F the home field advantage constant of 0.575.

  • There is no methodology I have seen for the NFL that makes any sort of sense. The problem with predicting the NFL games is the fact that an individual's performance impact on a team cannot be predicted accurately and to make it even harder, an impact as to how it relates to playing another specific team. For instance losing a starting tackle is a big deal, a bigger deal if backup stinks, even bigger if going against a great DE. The problem is unlike baseball or basketball the loss value of a player is rather abstract statically, which makes your equation seem like nonsense. – Coach-D Nov 17 '16 at 19:04
  • Looking back on this as it came up in community you should really not accept your own answer when it is a poor answer. – Coach-D Dec 29 '16 at 22:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.