I have developed two
stochastic simulation models using
Excel for a 19 team
backgammon league with
3 player teams playing in
position by position order in a
round robin format .. the first model made a pre-season prediction using average team rating at the start of the season .. as play progressed the match results were input into the position by position model yeilding new predictions based on up to date individual players W-L records and associated team points.
The
pre-season model is based on an
average team rating derived from the team members ratings at the start of the season .. toward the end of last season, I developed a more sophisticated model using individual
team players by position and
included results from matches played up to that point and then almost daily thereafter.
The
position by position model was most helpful in giving players and non-players alike an accurate insight into the team standings .. the standing at the team league homepage, due to the subtle nature of how the league is scored, are useful but have proved unreliable for really "seeing" the standings of the race until the last day or two of the season.
This season for the first time, we are introducing a 6 team single elimination championship
playoff at the end of the regular season .. the model, using player ratings for unplayed matches to determine p(U), the probability the underdog wins from each players
Elo FIBS rating, and actual results reported throughout the season to calculate p(W) and p(P), probability for each team to win or make the playoffs , by simulating and recording the results (order of finish) for each team in thousands of simulated seasons in each simulation run.
We are about two weeks into a new season with over 200 of 513 matches played .. since opening day, I have been attempting to adapt and improve the model .. last season the model had a few funky fixes to keep it accurate and runing smoothly .. as in any programming project, trying to clean up old hastily written patches is often as complex as making the patch in the first place - sometimes even more so.
While I have confidence the model will work effectively again this season and expect to have its first results out over the weekend, I am encountering some unexpected problems .. I would like to use this thread as a place to discuss both problems and improvements to the model and subsequent reporting as the season progresses and perhaps beyond .. I hope you may also have some ideas I can incorprate into the model.
Once stablized and made more generic for opening day initialization and data input during the season, the model would be a useful and interesting tool for almost any individual or team league, particularly one using player ratings such as the Elo ratings used in chess and backgammon.
If interested, you can see the current season and reports from last season at the links below.
sixty_something @ FIBS
FIBS Team League current standings
FIBS TL IV final standings from TL season IV
TL4 Pre-season Odds TL4 opening day predictions
TL4 Team League Predicitons predictions from season 4 with commentary on the model and sample output
FIBS Rating system