It's been a difficult season for tipping game margins, by which I mean that Mean Absolute Errors (MAEs) for most of the forecasters I follow have been elevated relative to last year.

Perhaps most prominent of those forecasters is the bookmaker at TAB Sportsbet, whose margin predictions can be inferred from the handicap he sets in the Line Market. By my calculations, his MAE for the season to date is at 32.4 points per game, which is about 3.6 points per game higher than his average last season, and over 6 points per game higher than his average for 2013.

The table at right provides the average MAE for each team across all of their games, including Finals, from the start of 2012 to the current round. You can find the numbers I've just alluded to at the foot of this table.

To calculate the MAE for any single game I've simply taken the absolute value of the difference between the (negative of the) TAB Bookmaker's pre-game line market handicap and the actual game margin. For a handful of games where the handicap I have is not priced at even-money for both teams (which can happen for a variety of reasons), I've inferred what the true handicap would have been given the head-to-head prices for that same game.

I've sorted the table rows on the basis of each team's overall average MAE across the three-and-a-bit seasons, which results in Fremantle rising to the top. They've maybe been, therefore, in a margin sense, the most predictable team across the period of time spanned by the data.

That predictability has extended to the current season in which they've produced game margins that have been, on average less than 3 goals different in absolute terms from what the TAB Bookmaker predicted. Port Adelaide, who sit second across the entire time period, also lie in second for the current season.

Fremantle and Port Adelaide are not, therefore, the teams we should be blaming for the blowout in MAEs this season. That blame should be sheeted home to the Crows, Suns, Lions, Tigers and Dogs, all of whom have allowed game margins to be, on average, more than seven goals different from expectations.

The first three of the teams just named have a long history of bloated MAEs and fill three of the bottom five places on the table all having average MAEs above 30 for the period from 2012 to now.

Looking more closely at those bottom five teams, in particular splitting their MAEs into those for games played at home (left) and away (right) we can see from the table at left that the Suns', Dons' and Giants' inflated MAEs have been mostly caused by unexpected margins in home games, while the Crows' has been mostly from away games. The Lions have produced relatively large MAEs at home and away.

At the top of the table amongst the teams with the five lowest MAEs, Fremantle, Geelong and the Western Bulldogs have had lower home game MAEs while Port Adelaide and Hawthorn have had lower away game MAEs.

IMPLICATIONS

Some of the differences between MAEs recorded above are large enough to be statistically significant. Practically that means there are some teams for which the TAB Bookmaker's lines have been more in error, on average and in absolute terms, than for some other teams. It's interesting to speculate why that might be the case.

We might think about decomposing the differences we see into two causes

1. The extent to which there are inherent differences in the predictability of any team relative to any other
2. The extent to which the TAB Bookmaker is either poorer at estimating the week-to-week abilities of particular teams, or chooses to frame line markets for them that do not exactly reflect his opinions about the true final game margin but which include a (probably small) handicap premium or discount for particular teams that allow him to exploit market imperfections.

On the latter point, research has been done using US betting data that suggests bookmakers there could manipulate lines (which they call "spreads") to their own advantage by capitalising on reliable and exploitable bettor behaviours. I wonder if the same thing might be in evidence in the data above, notwithstanding that some teams might also just be inherently more unpredictable.

More on this topic, I think, in the future.