Matter of Stats

View Original

2018 - Round 18 : Overs/Unders Update

We seem now to have settled into a collective view - shared by the MoS twins and bookmakers alike - that the new normal is rounds of AFL football producing average total scores somewhere around the mid 160s.

And, the empirical evidence continues to be that this is a reasonable position to hold.

This week, neither of the bookmakers expect any game to produce a total over 180, and quite a few games are expected to produce totals around 160 points or less. The MoS twins are more optimistic about a couple of games, but otherwise broadly agree.

MOST LIKELY HIGH-SCORING GAME

  • All : Brisbane Lions v Adelaide (179.5 to 182)

MOST LIKELY LOW-SCORING GAME

  • MoSSBODS : Carlton v Hawthorn (155)
  • MoSHBODS : Port Adelaide v GWS (154)
  • TAB & Centrebet : West Coast v Western Bulldogs (148.5 to 151.5) [NB the 148.5 for the TAB is the adjusted, even-money total]

MOST LIKELY HIGH-SCORING TEAM

  • All : Sydney (104 to 108)

MOST LIKELY LOW-SCORING TEAM

  • All : Gold Coast (52.5 to 57)

MoSHBODS' performance was mixed relative to naive forecasting in selecting the extremes of team and game scoring last weekend, attaching:

  • 18.9% probability to the highest-scoring team, Melbourne (against a naive forecast of 5.6% since 18 teams were playing)
  • 2.9% probability to the lowest-scoring team, Port Adelaide (against a naive forecast of 5.6%)
  • 9.0% probability to the highest-scoring game, Adelaide v Geelong (against a naive forecast of 11.1% since 9 games were played)
  • 20.1% probability to the lowest-scoring game, Fremantle v Port Adelaide (against a naive forecast of 11.1% since 9 games were played)

During the week, I estimated the forecasting performance of the TAB and Centrebet in these four high-low markets (by taking their implied estimates of expected team scoring in each game and then using these expectations to simulate the games in each round, in exactly the same way as estimates are created from MoSHBODS' expectations).

I found that (across Rounds 1 to 16):

  • The TAB and Centrebet performed better than a naive forecaster in all four markets
  • MoSHBODS performed better than a naive forecaster in choosing the low-scoring and the high-scoring teams
  • MoSHBODS, the TAB, and Centrebet did better in forecasting the low-scoring team than in forecasting the high-scoring team
  • Centrebet did better than the TAB, and the TAB better than MoSHBODS on forecasting the low-scoring team, and in forecasting the high-scoring team
  • MoSHBODS performed worse than a naive forecaster in choosing the high-scoring game, and in choosing the low-scoring game
  • The TAB and Centrebet performed better than a naive forecaster in choosing the high-scoring game, but worse than a naive forecaster in choosing the low-scoring game
  • Centrebet did better than the TAB, and the TAB better than MoSHBODS on forecasting the low-scoring game, and in forecasting the high-scoring game

This week, MoSHBODS has

  • nine teams as more likely than 1 in 18 to be the highest-scoring: Sydney, Richmond, West Coast, Hawthorn, Geelong, Brisbane Lions, Collingwood, Essendon, and Adelaide
  • six teams as more likely than 1 in 18 to be the lowest-scoring: Gold Coast, Carlton, Western Bulldogs, St Kilda, Fremantle, and GWS.
  • four games as more likely than 1 in 9 to be the highest-scoring: Lions v Crows, Cats v Dees, Pies v Roos, and Saints v Tigers.
  • four games as more likely than 1 in 9 to be the lowest-scoring: Power v Giants, Blues v Hawks, Swans v Suns, and Dons v Dockers.

WAGERS

Investors have four wagers this week, one unders and three overs.

The Essendon v Fremantle (overs bet) and Carlton v Hawthorn (unders bet) games are being played at Docklands, so wet weather shouldn't be a feature for either of those, and the Geelong v Melbourne game (overs bet) is scheduled for Kardinia Park where the forecast is "cloudy".

That leaves only the West Coast v Western Bulldogs game to be considered, where Investors have the last of their overs bets, but where the forecast is a troubling "Showers. Windy. Possible hail". If that is the weather we get, we're going to need all of our notional 15.5 point effective overlay. (The actual overlay is 13.5 points, but we've secured a $2 price).

I'd be happy to settle for a 3 and 1 performance this week.

PERFORMANCE TO DATE

MoSSBODS finished on the correct side of the TAB's totals in four of the nine games last weekend, and on the correct side of Centrebet's in five. MoSHBODS did better, finishing on the correct side of the TAB's totals in five of the games, and on the correct side of Centrebet's in six.

That left MoSSBODS' with a season-long 49% record against the TAB and Centrebet, and left MoSHBODS with a 47% record against the TAB and a 51% record against Centrebet. 

MoSSBODS' landed 1 of 2 overs bets with Centrebet last weekend, as well as 1 of 1 unders bets. Its overall strike rate on overs bets is now 29% and on unders bets 61%.

On the mean absolute error (MAE) metric, MoSSBODS registered the lowest MAE for game margins and home team scores, while the TAB did the same for away team scores, and total scores.

Centrebet still leads on the season-long view for home team scores (narrowly, from MoSSBODS) and for game totals (narrowly, from the TAB), and now also leads on game margins (also narrowly from the TAB). The TAB still leads on away team scores.

MoSSBODS performance on game totals provides an interesting example of how a predictor can almost match the bookmakers' performance on one relevant total-related metric (MAE), but do no better than chance on being on the right side of those bookmakers' actual game-by-game estimated totals. MoSSBODS' MAE on game totals is only 0.7 points per game worse than the TAB's and Centrebet's, but it's landed on the right side of the their totals only 49% of the time.

There's a similar situation with a number of our Margin Predictors where, for example, Bookie_9 has an MAE only 0.15 points per game worse than Bookie_Hcap's, but has landed on the right side of Bookie_Hcap's line less than 49% of the time