Matter of Stats

View Original

2018 - Round 8 : The Triumph of Hope Over Experience

You'd think after a decade I'd be far more immune to the slings and arrows of outrageous forecasting, but the truth is it still feels a lot easier to write blogs and discuss MoS models' opinions in weeks following a non-embarrassing set of results.

And so we come to Round 8 where we find the TAB Bookmaker foreshadowing another challenging week of picking winners.

The all-game average expected margin is just under 16 points a game, which is highish in the context of 2018 but very low in the context of the last half-dozen seasons or so. In all but two of the games this week, the expected margin is under 4 goals, and in all nine it's under 4 and a half goals.

So, in summary, good luck with your tipping.

Before launching into the week's forecasts I want to explore the notion that this season has somehow been an unusually unpredictable one. Sure there have been some upset victories this year - take last weekend's Roos victory as an example - but, in a probabilistic forecasting sense, the question is whether or not upsets have occurred at roughly the expected rate.

This view of predictability draws on the notion of the calibration of a probabilistic forecaster, which, simply put, has it that events estimated as having an X% chance of occurring by a well-calibrated forecaster should happen about X% of the time. So, for example, even 90% favourites should lose 10% of the time if the forecaster is well-calibrated - and, we should predict that this will occur 10% of the time.

This season, the average TAB-estimated victory probability of favourites (using the risk-equalising method for unwinding overround) has been 68,8% for games where there was a clear favourite and that didn't end in a draw. Those favourites have won 70% of the games. I'd call that (a) well-calibrated, and (b) highly predictable.

Anyway, onto the MoS forecasters.

TIPS AND PREDICTIONS

Despite the average closeness of the contests, it's another week of broad agreement across the Head-to-Head Tipsters, punctuated only by the usual sound and fury signifying not much from Home Sweet Home, and a few token dissenting voices from Consult The Ladder, the RSMP twins, and C_Marg in one game each.

At the end of the round then, depending on the result in the Hawthorn v Sydney game, BKB and ENS_Linear will either hold a two tip lead over the RSMP twins or be tied with them. 

Amongst the Margin Predictors, it's C_Marg back in the role of Predictor most extreme, its forecasts being the most or least enthusiastic about the home team in six of the nine contests. Its mean absolute deviation (MAD) for the week is 6.7 points per game, its fourth-highest of the season. Bookie_3 has the week's next-most deviant forecasts, slightly ahead of the MoS twins'.

At the game level, there are again no huge MADs this week, with only the North Melbourne v Richmond, Carlton v Essendon, and Gold Coast v Melbourne games generating MADs above 4 points per Predictor and forecasts spanning roughly 20-point ranges.

The mean expected margin across the nine games and all Margin Predictors is 14.3 points per game, which is, as is now custom, a little lower than the TAB's average, which this week stands at 15.9. 

Turning lastly to the Head-to-Head Probability Predictors we find C_Prob again registering the all-Predictor high MAD, this week at 5.9% points per game ahead of MoSHBODS_Prob (4.7% points) and MoSSBODS_Prob (4.0% points). To make most ground on the leading trio of bookmaker-based Predictors, MoSHBODS needs a Carlton upset, and Melbourne and Carlton wins. MoSSBODS also needs Carlton and Collingwod wins, but would prefer a Port Adelaide win to a Melbourne one.

The Carlton v Essendon game carries this week's largest MAD at 6.7% points per Predictor, driven up by the relatively higher probability estimates offered by C_Prob and the MoS twins.

WAGERS

The Head-to-Head and Line Funds are offering a little less action this week, with the former opting for five wagers totalling 12.5% of the original Fund, and the latter three wagers totalling only just over 2%.

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

The head-to-head bets are on a favourite, an equal-favourite, and three underdogs, while the line bets are all on underdogs.

The ready reckoner, which quantifies those wagers' collective risk, appears below.

Four games carry no risk at all and, clearly, the Carlton v Essendon game carries the greatest swing, with a Blues win promising a 3.2c jump in the Combined Portfolio, and a Blues loss by four goals or more threatening a 1.5c drop. Across the other games, Investors face a 2.3c swing in the Suns v Dees game, and a 2.4c swing in the Pies v Cats game.

In total, just over 5% of the original Combined Portfolio is at risk and the maximum upside is just over 6%.

To finish, here are MoSHBODS' and MoSSBODS' opinions about likely team scoring.

Commentary about these forecasts will come in the week's Overs/Unders blog.