Matter of Stats

View Original

The MoS Tipsters, Predictors and Funds for 2022

The 2022 season, at least from a MoS viewpoint, will run little differently from how it did in 2021, the main differences being:

  • The use of re-optimised MoSHBODS and MoSSBODS models, which now

    • Have different parameter values for the three eras of scoring in V/AFL history: 1897 to 1978, 1979 to 1996, and 1997 to the current day

    • No longer use generic “home team bias” and “away team bias” assessments (which always felt a bit arbitrary)

    • Have different windows of history to consider when calculating VPVs and expected scoring shots and scores

    • Have different season splits (which are used to determine which alpha to apply when adjusting team ratings)

    • Now apply any cap to a team’s score and not, as previously, to a margin

    • Have multipliers to adjust forecast raw game totals to better align them with history

  • The new MoSSBODS is about 0.06 points per game better than the old MoSSBODS in terms of Game Margin MAE and 0.08 points per game better in terms of Game Total MAE. The new MoSHBODS is about 0.11 points per game better than the old MoSHBODS in terms of Game Margin MAE and 0.07 points per game worse in terms of Game Total MAE. The new MoSHBODS is also about 0.1 points per game better than the new MoSSBODS in terms of Game Margin MAE and 0.07 points per game worse in terms of Game Total MAE.

  • The use of a re-optimised MoSHPlay, which combines the new MoSHBODS forecasts with player-by-player SuperCoach forecasts

  • The use of a re-optimised ENS_Linear, which has been fitted to seasons 2010 to 2019 and 2021, and which now includes only a conditional inference tree forest and a MARS model in the ensemble

  • A reweighting of the Combined Portfolio to 65% Line, 30% Head-to-Head, and just 5% Over/Under

Long-time readers will know that I had determined to exit Over/Under wagering this year, but the new MoSSBODS model’s improved Totals forecasting record has made me stick with it for one more year at least, albeit with a halved Fund weighting.

Everything else will remain the same.

So, if you want to know more about the various models, or about the Funds and their wagering strategies, have a look at this blog from 2020.

Somehow we all made it here.