NFL Season Simulations

In the last post (that you can find by clicking here), I discussed simulating individual games. The idea is that by coming up with individual conversion statistics on various down and distances or expectation of yardage distributions per play you can simulate an entire drive, and combine basic programming formulas to combine drives into consecutive drives until the time runs out for a half, and then simulate two halfs to simulate the game. You can press a button and run this game simulation thousands of times in less than a second based upon those assumptions and estimate win rates based upon expectations (expected strategies of each team). You can then optimize the strategy based upon your assumptions to maximize the chances of winning the game if you are creating a gameplan.

But there are various ramifications. If you used a particularly effective play one game, the other team is going to see it on tape and make adjustments, which will effect your success rates. It may be even more accurate to model an entire season and look at the season as "one game" with teams adjusting at halftime of their game as well as before their game starts based upon what they saw. You also will have an edge from adjusting to your opponent.

Based upon some assumptions we can estimate the probability of winning each individual game and simulate a win or a loss (I prefer not to include ties, but you could). We can go down the entire schedule and run the simulation. But you have to also understand that the playoffs IF you are lucky enough to make them carry with it disproportional reward for each win. As such, having a suboptimal gameplan early on (in exchange for the element of surprise or minimizing information leakage of playoff gameplans) may actually be best. However, as games become increasingly more like playoff games in that if you don't win, you fail to make the playoffs or effect the odds more significantly (divisional opponents), there is an increased value in going for it.

You can come up with complex simulations, but simple ones work too.

Then you can set up a formula to measure when conditions are made to make the playoffs. In this case I set it up so that 10-6 or better makes the wildcard and being 1 win better than opponents does as well. I set it up for after starting 3-0 with opponents 1-1,1-1 and 0-2 with the 3rd game yet to play. I set up 50% chance of beating divisional opponents and 55% chance of every other game for the team I'm tracking.

You can simulate the win total distribution or else you can simulate wehther or not a condition is meant. In the image below I set up a formula that gives me a 1 if my team is either 10-6 or better OR has a better record than ALL opponents (I didn't include ties)

 You could set up a seperate set of data and formulas IF you make the playoffs.
Basically IF the criteria set up in the formulas (10 or more wins or higher win total than every divisional opponent), you proceed to a playoff calculation where you have additional games.
IF you have 13 or more wins you might award yourself a bye and automatically operate as if you won the first game (since you get a bye).
IF you produce wins in every playoff game including the superbowl, you can reward yourself a 1 to signify a superbowl win and then run a simulation to calculate the odds of winning the superbowl (or just making the superbowl).

I assumed a 50% chance of each divisional opponent as well as playoff opponents but 55% chance of winning all other games.

THe idea here is not to set up an accurate simulator, but lay the groundwork and explain how you might go about coming up with an accurate simulation. The data from the simulation is only as good as your assumptions of your probability of winning each game, but that's why I laid the groundwork for coming up with more detailed individual assumptions within a game that perhaps varies more play to play but on average are approximately correct. If you can break an NFL season down to it's smallest unit (a single play) and breakdown all things that could happen on this play and all resulting information, you can simulate every play of each game thousands of times to come up with your projected odds of winning and then in turn apply that to the model and come up with a season total.

The tiebreaker logic and determining more accurately who of all 32 teams makes the playoffs (12 teams total) may require more details, more data, and complex tiebreaker logic, but it allows for a more detailed projection/simulation.

You can test different assumptions that saving your best plays for the playoffs is better or exactly how much saving plays for the playoffs is ideal as long as you are able to make accurate assumptions about how it influences future plays.

This same strategy may work on fantasy football teams as well, but you probably have to set up different rules and certainly a different method for individual games.