2016 Round 6 : Tending the Wounds
/This feels like an important week for the MoS Funds, even if it is still very early in the season. With, in footballing terms, a 4 and 1 start for Overall Portfolio profitability, a move to 5 and 1 would provide more evidence that the various algorithms are spotting value, while a slide to 4 and 2 would raise a small query about the relative contributions of luck and talent to those initial successes.
That said, a loss won't be at all catastrophic and certainly won't signal a need to head for the exits. As with many things in life, the trick is to appropriately weight new evidence in the light of previous experience to estimate the probability of plausible futures. The medium- to long-term future could still be one of profitability even if the short-term reality isn't.
Anyway, enough of the amateur hour Bayesian statistics and philosophy, here are this week's wagers.
The Head-to-Head Fund, which missed with its three wagers last weekend, has peeled off another three this week, one on a slight favourite (Richmond), another on a moderate underdog (GWS), and the third on a longshot (Brisbane Lions).
As is usually the case, the Line Fund has opted for the same teams as the Head-to-Head Fund, in those games giving 2.5, receiving 12.5 and receiving 38.5 points respectively. The Line Fund has also, however, added a very small bet of its own on the Eagles giving the Pies 39.5 points start.
The Overs/Unders Fund has made five selections this week, four on the Unders and just one on the Overs.
Combined, a best case weekend would yield almost 16c of profit, while a worst case weekend would see a lose of just over 8c.
(By the way, on Sunday I'll have a riotous or tragic - I'll only which by then - story about the follies of placing wagers hurriedly before having adequately checked your data inputs. Make a mental note to look out for that.)
TIPS AND PREDICTIONS
As noted in the previous blog post about MoS' Team Rating Systems, MARS, ChiPS and MoSSBODS all rate GWS quite highly. ChiPS, in fact, ranks them as the 4th-best team and MoSSBODS ranks them 3rd.
Those high rankings, coupled with the fact that the Giants are playing at home this weekend, has led to the MoSSBODS- and ChiPS-based Head-to-Head Tipsters selecting an "upset" Giants win over the Hawks. Home Sweet Home, of course, has done the same thing, leaving the overall tally for that game at just 6 to 5 in favour of the Hawks.
In other games, the largest minority groups amount to no more than three of the 11 Tipsters, a minority of this size arising for the Western Bulldogs (playing the Kangaroos), St Kilda (playing Melbourne), and Port Adelaide (playing Richmond).
Overall, disagreement is at relatively high levels this week, partly because the smallish amounts of dissension come from different Tipsters in different games. The final Disagreement Index of 23% is the 3rd-highest Index for a round this year, the major contributors to its elevated level being the three MoSSBODS-related Tipsters and the single-minded, form-ignoring Home Sweet Home.
The Margin Predictors have, in historical terms, even higher levels of dissension this week, their overall mean absolute deviation (MAD) of 6.0 points per game per Predictor a season record by some considerable margin.
Driving that high MAD is dissension about the likely victory margin in the GWS v Hawthorn game, where the MAD is a massive 13.0 points per Predictor, and about the likely victory margins in the Roos v Dogs matchup (MAD of 9.2), Dees v Saints matchup (MAD of 7.1), Crows v Dockers matchup (MAD of 6.8), and Lions v Swans matchup (MAD of 6.6). In the first three of those games we have the unusual situation of three or four Predictors tipping wins by the underdog, in some cases by as much as four goals.
From a Predictor-by-Predictor perspective, a substantially disproportionate share of the dissension can be attributed to the three MoSSBODS-based Predictors, Bookie_3 being the only other Predictor exhibiting above-average levels of dissension.
The MoSSBODS' Predictors, it seems, have made a habit this year of standing a bit apart. One way of summarising the level of disagreement displayed by MoS' Margin Predictors across the season so far is by employing a statistical technique known as Principal Components Analysis, which allows us to extract the main sources of variability amongst a set of variables, here the margin predictions of our Predictors.
By charting the results of such an analysis (technically, by plotting the Predictors in the space of the first two Principal Components, which are the two that explain most of the variability), roughly speaking, the closer two Predictors are in the chart the more consistently similar are their predictions. And, conversely, the further apart they are on both dimensions, the more they've tended to disagree.
We see in the chart what we might already have intuited, which is that the Predictors form three natural groups - one in the lower left that includes all the Predictors that use Bookmaker and MARS rating information, another in the upper right that uses only MoSSBODS Ratings, and the last, comprising only C_Marg, forming a third group but one much closer to the MoSSBODS group than to the Bookmaker/MARS group.
Next, let's look at the Head-to-Head Probability Predictors, which this week, like the Head-to-Head Tipsters, are displaying moderate levels of disagreement. That level has been elevated substantially by the probability assessments of MoSSBODS_Prob, which, on average, differs in its assessments from the all-Predictor average by over 9% points per game.
Taking a game-by-game view, it's the GWS v Hawthorn game where dissension is greatest, as the Giants' chances are rated as anywhere between 36% and 71% depending on which Predictor you choose, and the MAD for the game coming in at a whopping 15.5% points per Predictor. Two other games carry above-average MADs: the Roos v Dogs game where the MAD is 8% and assessments of the Roos' chances range from 28% to 55%, and the Dees v Saints game where the MAD is 7.1% and assessments of the Dees' chances range from 38% to 62%.
This week we get a very obvious assessment of the efficacy of excluding bookmaker data from our predictive models.
MoSSBODS DETAILS
You might well be curious to know a bit more about how the various MoSSBODS Tipsters and Predictors arrived at their opinions. Below is the usual table showing the breakdown of MoSSBODS inputs.
In it we can see that the contrarian MoSSBODS opinions in the Roos, Dees and Giants game are at least partly attributable to the assessed Net Venue Effects in those games. Both the Roos and the Dees are assessed as performing slightly worse than their normal abilities would suggest when playing at Docklands, while GWS is assessed as enjoying a 3 SS advantage because they're facing a team from interstate.
Five other teams also enjoy that interstate bonus this week: Adelaide, Geelong, Richmond, Brisbane Lions and West Coast.
The round's highest-scoring game is, narrowly, expected to be the Crows v Freo game for which 191 points is the most-likely total. That's just one point higher than the tally expected in the Roos v Dogs game, however.
Lowest scoring is tipped for the Blues v Dons game, in which only about 42 Scoring Shots are predicted.
The TAB also thinks the Blues v Dons game will be the low-scoring game of the round, but (implicitly) forecasts about two more goals will be scored than does MoSSBODS.
Generally, MoSSBODS is predicting smaller totals than the TAB, this week by amounts ranging from just a couple of points, which is the case in the Lions v Swans game, to as much as 20 points, which is the case in the Giants v Hawks game. In only one game is MoSSBODS prophesying a larger total, the Crows v Dockers game where its projected total of 191 points is about a goal more than the TAB's. (And so we have an Overs bet.)
On an individual team basis, 12 teams are predicted by MoSSBODS to score less than the TAB expectation, and six are expected to score more. In absolute terms, the difference only exceeds 12 points for three teams however:
- The Kangaroos, expected to score 86 by MoSSBODS and 101 by the TAB
- Melbourne, expected to score 81 by MoSSBODS and 98 by the TAB
- Hawthorn, expected to score 74 by MoSSBODS and 97 by the TAB
Overall this season, MoSSBODS' and the TAB's mean absolute error (MAE) records are as follows:
- Home Team Score: MoSSBODS 25.0 points per game; TAB 21.3 points per game
- Away Team Score: MoSSBODS 18.6 points per game; TAB 19.9 points per game
- Total Team Score: MoSSBODS 31.6 points per game; TAB 28.0 points per game
