Matter of Stats

View Original

2017 - Round 1 : Is This Thing On?

I'm flipping the order of the sections in the weekly Wagers & Tips blog this year, providing the tips and predictions before the wagers. I think that everyone who comes here is interested in the tips, but only a subset are interested in the wagers as well. And, populism is all the rage right now ...

Anyway, let's get into those tips and predictions.

TIPS AND PREDICTIONS

There have been a few changes to the suite of tipsters and predictors this year. You can read a summary of those changes in this earlier blog.

As well, the summary table has been slightly altered. The new version appears below.

The top section provides the head-to-head tips of MoS' nine Head-to-Head Tipsters, with forecasts of a win by the TAB's underdog for a game now being flagged by a jaunty canine icon. In the column at the far right we still have the Disagreement Index for each Tipster, the number there the probability that a randomly chosen Tipster in a randomly chosen game would disagree with that Tipster's tip. These Indices now carry a traffic light icon reflecting their magnitude, which will be green for values under 25%, amber for values between 25% and 50%, and red for values above 50%.

This week we see that Home Sweet Home has the highest Disagreement Index (readers from last year will know that this is completely unsurprising), and that disagreement levels are generally high as the Tipsters are split 5-4 in three of the games, and 7-2 in two more.

In the middle section are the margin predictions, which this year are being provided by 10 Predictors.

Here, the Predictor most bearish about the Home team's chances carries a red, downward-pointing triangle, and the Predictor most bullish carries a green, upward-pointing triangle. Scanning the triangles reveals that Bookie_3 is the most bearish in five games, and MoSSBODS_Marg in two, while C_Marg is the most bullish in three games, and MoSSBODS_Marg and MoSHBODS_Marg in two each.

The column on the far right summarises the level of disagreement between a Predictor's forecasts and the all-Predictor averages. These too are traffic-light coded with mean absolute difference (MAD) values of less than 4 points per game carrying green lights, of between 4 and 6 points per game carrying amber lights, and of more than 6 points per game carrying red lights. We see then that it's the three team-rating based Predictors that this week have the highest MADs.

The MAD for each game is traffic-light encoded in the same way, and these show that the Adelaide v GWS, Kangaroos v West Coast, and Carlton v Richmond games have elicited the most divergent margin predictions. These are also the three games where the Head-to-Head Tipsters are split 5-4.

At the base of the table are the head-to-head probability estimates of MoS' six Predictors. Here, the lowest estimates of the Home team's chances carry the red, downward-pointing triangles and the highest estimates carry the green, upward-pointing triangles. Bookie_RE has five of the red triangles while Bookie_OE and MoSSBODS_Prob have two each. The green triangles are more broadly distributed, C_Prob claiming three, and Bookie_LPSO, MoSHBODS_Prob and MoSSBODS_Prob two each.

On the right are the Predictors' MADs, here measured as mean absolute differences from the all-Predictor average probability estimate. The values are coded red if they're above 6% points, amber if they're between 4% and 6% points, and green if they're below 4% points. This week they're all red, the MADs being driven up especially by the divergence of opinion in the Adelaide v GWS, Kangaroos v West Coast, Carlton v Richmond, St Kiida v Melbourne, and Essendon v Hawthorn games. Estimates in these games differ by as much as 20 or 30% points.

We can expect then, quite a divergence of probability scores across the Predictors.

WAGERS

For each of the last 10 seasons, MoS wagering has started in Round 1, though in some years there have been caps on bet sizes or restrictions on which of the Funds could wager at all. In more recent years, encouraged by historical empirical evidence for the new Fund algorithms, no such restrictions have been imposed. None are in place this year either (as you'll soon see ...)

Actual wagering performance in the early rounds has been mixed. For example, looking solely at Round 1 results, half the time the Funds have made a profit, once they've abstained from wagering, and on the four other occasions they've turned a loss.

Last year was probably the best ever start to a season MoS has had, with profits being returned in the first four rounds, and then again in the sixth. After that point, the net result was a loss across the remainder of the season. So, for 2016 anyway, starting early with no restrictions was clearly a good decision.

Activity levels were moderate in Round 1 last year, with the Head-to-Head Fund risking three bets, the Line Fund doing the same, and the Over/Under Fund throwing money at six games.

It's fair to say that "moderate" would be a poor description of this season's Round 1 action - and the over/under markets haven't even been posted yet. Let's take a look at it now.

(Please click on the image to access a larger version.)

There's a lot more information in the Wagers table this season. 

The top three rows provide the team ratings of MARS, MoSSBODS and the new MoSHBODS Systems. Here's how to interpret these ratings:

  • For MARS, an average team is rated 1,000. There's no strict mapping between MARS ratings and scoring but, empirically, final margins have averaged about 75% of the difference in team ratings.
  • For MoSSBODS, a team with an average offence or defence will rate 0 on that measure. Its ratings are calibrated as scoring shots, so a team with a combined rating that is X scoring shots less than its opponent's will be expected to register X fewer scoring shots than its opponent at a neutral venue. Scoring shots are converted to points using the average conversion rate of teams during the 2016 season.
  • For MoSHBODS, a team with an average offence or defence will also rate 0 on that measure, but its ratings are calibrated in points. So, for it, a team with a combined rating that is X points less than its opponent's will be expected to score X points less than its opponent at a neutral venue

The next two blocks are TAB and Centrebet prices at the time of writing. This year, after a number of polite enquiries in-person, on Twitter and via e-mail about why I was still exclusively wagering with the TAB, I've given myself a second option in Centrebet. About 60% of the head-to-head and 65% of the line action is with Centrebet this week.

The final three blocks relay information about the week's wagers. This includes the team wagered on, the amount and the price, but also a star rating, reflecting the estimated overlay that the wager carries. Bets are only made when these estimated overlays exceed some threshold: 5% points for head-to-head, 6 points for over/under, and 8 points for line betting.

You'll see that there are a number of wagers with large estimated overlays this week. I'll be skeptical about the accuracy of these estimates in the early rounds, but it is nice to see such large differences between what the models are suggesting and what the bookmakers think. The models would have to be very wrong in some cases in order for the overlay to disappear entirely.

Looking across the nine games you'll see that we have head-to-head wagers in eight of them, including two on away teams, which would not have been possible under the Fund constraints in place last season. These bets represent a startling 30% of the initial Head-to-Head Fund but, with its diminished weighting this season, represent only 6% of the entire portfolio.

There are also six line bets, three of those on away teams and representing, in total, just under 20% of the initial Line Fund and about 8% of the entire portfolio.

The ready reckoner for these wagers is shown below.

Three games carry the largest risk and reward, with swings of as much as 6 to 7% points possible in the Carlton v Richmond, Kangaroos v West Coast, and Adelaide v GWS matchups. Only one other game could see a swing of much more than about 1% point: the Essendon v Hawthorn game, where a Hawks win by 11 points or more would add 1.4c to the Overall Portfolio, and a loss would strip 2c from that Portfolio.

Across the nine games the aggregate swing is over 28% points, from a 14c gain to a 14c loss. The magnitude of this swing and the heightened level of Fund activity that has produced it, can be attributed mostly to relatively large differences in MoSSBODS and MoSHBODS opinions about the teams in comparison to the bookmakers', but also partly to the fact that the Funds now have two bookmakers from which to choose prices.

Whether the level of risk is too high this early in the season is difficult to determine. But, we will have some empirical data come this time next week. 

***~~~***

For a lot of reasons, it's nice to finally get started this year. 

I'm particularly interested in:

  • The performance of MoSHBODS, especially relative to MoSSBODS
  • The relative attractiveness of TAB vs Centrebet prices from week-to-week
  • The improvement, in any, in the probability estimates of MoSSBODS and MoSHBODS (I've made some subtle changes to the way that the ratings differences are mapped to margins that I'll write up in a future blog)

Hope the 2017 season is kind to you.