|
NFL Prediction Software
Software Documentation
How Predictions Are Made
Latest Predictions
send email to: admin@nfl-forecast.com
|
How the software makes predictions
Here is a little background that describes how the predictions are made. The probability of the home team winning all unplayed games are estimated.
These estimated probabilities are displayed for every unplayed game on the "Advanced Analysis" panel. The probabilities are estimated based on the relative
strength of the two teams (as reflected in the power ratings discussed below) and where the game is played (home field advantage). When the season is forecasted,
all of the unplayed games remaining in the season are simulated. The outcome of each game is randomly determined based on the probability shown on the
sliders on the "Advanced Analysis" panel. After all the remaining uplayed games in the season have been simulated, I determine the standings in each division, and all of the playoff seedings
for each conference using the NFL tiebreaker rules. I also predict the out come of playoff games using a slightly different procedure, and
determine the order of the NFL’s annual selection meeting (i.e. draft pick order). I then repeat the simulation of the unplayed games in
the season 5000 times. Nearly every simulated season has a different outcome because the outcome of every game is random. However, the 5000 simulations considered together provide a good
prediction of the probability of any outcome of interest during the NFL season.
How to interpret the predictions
A little discussion about probability is in order. Suppose the weather forecaster predicted an 80% chance of rain tomorrow.
Would you expect it to rain? Probably so. If tomorrow it actually did rain, we would think that the forecaster did a good job.
Now suppose the weather forecaster predicted an 80% chance of rain for 30 days in a row and it rained each and every day.
Should we consider this to be a good forecast? No! The forecaster should have predicted a much higher chance of rain (maybe
approaching 100%). The opposite would hold true as well. We should occasionally expect it to rain when the forecast calls for a
20% chance of rain. If it never rains under these conditions, then the forecast is not very accurate. The same view point
should be taken with these NFL forecasts. Events predicted with a low probability should occur occasionally, but not often.
Events predicted with a high probability should usually occur, but not always.
How the NFL power ratings are calculated
The power ratings are a quantitative estimate of the relative strength of any two teams in the league. There are two types of power ratings available in the software.
Before the season starts, preseason power ratings are the only ones available. These are calculated from published over/under betting lines for regular season wins. These betting lines are
the most unbiased estimate of team strenght available before the season starts. Once the season starts the outcome of actual games influences the power ratings as well.
The current power rating for a given team reflects the number of games won and lost, the strength of opponents played, and home field advantage. I use these power ratings to
determine the probability of victory for the home team for all the unplayed games remaining in the NFL season.
How accurate are the predictions?
Here are links to a couple of pdf files that summarize predictions from the 2002
and 2003 seasons. I haven't done detailed analysis of the 2004 or 2005 seasons.
A final word
With this background, you will find the predictions to be quite informative. The reason I created
this software was because I was intensely interested in knowing how my team was doing from a fan’s perspective.
Please remember that the forecasts made by this software are meant for entertainment purposes only. I don’t use any of this information to
gamble, and I don’t recommend that you do either.
|