2014 - Round 16: Staring At The Suns
/You're a portfolio manager basking in the warmth of six consecutive periods of sweet profit that have served to lift your original investment by over 20%. What do you do? Pull back a touch, take stock, mentally bank a little of that profit and adopt a more cautious approach to for a while? Maybe even stay out of the market entirely except for a few "just maintaining a level of interest" investments?
That's not necessarily what you do if you're guided by a set of risk-ignorant models with no concept of prudence, regret or elation, because then, if those models' mindless (in the literal sense) and objective assessments compel it, you might end up investing even more heavily. Like we are.
It's the Head-to-Head Fund that's the source of the incremental activity relative to last week, largely because of its strong attraction to the Suns at $2.15 taking on the Pies. Its 8.7% wager on them is easily its largest single wager of the season, more than doubling the 4% it wagered, successfully, on the Crows in Round 13.
The Fund's also opted for two other wagers, one of about 1% on the Dees at $7.50, and another for 0.7% on West Coast priced at $3.75.
The Line Fund has made six wagers in the round, the same number that it's made in the two previous rounds, though this week on four home team underdogs and two home team favourites. The three Head-to-Head and six Line wagers are accompanied by eight SuperMargin bets from the Margin Fund spanning four games. That's two wagers fewer than the Margin Fund ventured last week but it includes four that book-end two Chasms of Despair in different games. In the Port Adelaide v Essendon game one side of that Chasm spans only 3 points (27 to 29) and is bounded by an even more precipitous drop on the low side.
In aggregate, the week's 17 wagers represent almost 20% of initial Recommended Portfolio funds and take the total turn of the Portfolio to almost 1.5 for the season to date. In other words, every dollar in the Recommended Portfolio has now been in the TAB Bookmaker's hands about one-and-a-half times. So far, fortunately, we've made him give it back to us with considerable interest.
Those 17 bets cover all nine contests and have left Investors with a minimum upside of 1% in any single contest and a minimum downside of 0.4%.
Maximum upside rests with the Dees who, should they upset the Dockers, will raise the Recommended Portfolio's price by almost 4c all on their own. The Gold Coast, Port Adelaide, West Coast and Geelong are all also capable of adding about 3c or a little more to the Portfolio price.
Seven of the contests carry maximum downside of 2.5c or more, with the largest downsides attributable to the Cats and Port Adelaide. Total downside this week is just under 20c and upside is 23c.
If we're to make it seven profitable weeks in succession we'll probably need to land four of the six Line bets, maybe only three if results are reasonably kind to us in the other games. It is mathematically possible for us to make a profit winning just two of the Line bets but we'd require the Head-to-Head Fund to land all three of its wagers and for Port Adelaide and Carlton to deliver a SuperMargin success.
TIPS AND PREDICTIONS
This week it's the Suns v Pies game that has the Head-to-Head Tipsters arguing most, with 12 siding with the home team Suns and 16 with the narrow bookie-favourite Pies. In other games there are five Tipsters suggesting that GWS can beat the Crows, four that the Saints can toppled the Blues, three that the Dogs can do the same to the Cats, two that the Dons might prevail over Port, and just one - who else but Home Sweet Home? - alone believing that the Roos can beat the Hawks, the Dees can beat the Dockers, and the Eagles can surprise the Swans.
Only in the Richmond v Brisbane game is there unanimous support for the home team favourites.
Again it's Home Sweet Home that differs most from the all-Tipster norm, while this week it's Easily Impressed II, Short Term Memory II and Shadow who are next-most and considerably less divergent.
The Margin Predictors, with no version of their own Home Sweet Home, have reached unanimity in their opinions about the winners of every game except in the Suns v Pies contest where eight of them come down mostly narrowly on the side of the Suns and nine others come down even more narrowly on the side of the Pies.
In the Carlton v St Kilda game, though the entire suite of Predictors are suggesting that the Blues will win, eight tip a victory margin of less than a goal. Combo_NN_1, however, tips a Blues win by 50 points, which has made this game the one with the largest range in margin predictions this week.
The only other game with a range exceeding seven goals is the Port v Dons game where the low prediction comes from C_Marg tipping a Port win by 21 points and the high prediction comes from ProPred_7 tipping a 64-point victory. In four games - Roos v Hawks, Giants v Crows, Suns v Pies, and Cats v Dogs - the range of predicted victory margins spans less than four goals.
Bookie_3, who is extreme Predictor in three games has the round's largest Mean Absolute Deviation (MAD) from the all-Predictor average at almost 2 goals per game. C_Marg, who is also extreme Predictor in three games, has the third-highest MAD of 9.0 points per game, behind Combo_NN_2's 9.9 points per game despite being extreme Predictor in only two games. Interestingly, the Predictors with the five smallest MADs fill five of the top six places on the Margin Predictor MAPE-based Leaderboard.
Taking a broader view of MAD versus MAPE, the correlation between the Predictors' weekly MADs and their historical MAPEs stands at +0.48 this week.
For the season as a whole the correlation is +0.644 reflecting that Predictors usually nearer the all-Predictor average have done better than those more usually further away.
You can clearly see that tendency in the chart at right where you can also see that Bookie_3 and C_Marg deviate most from this tendency, in Bookie_3's case because its MAPE is very low given its MAD, and in C_Marg's case because its MAPE is very high given its MAD.
(By the way, that's an Excel scatter chart above where the labels have been added using the fabulous and free utility from here. Absent that utility doing this sort of labelling is painfully difficult.)
Like the Margin Predictors, the Head-to-Head Probability Predictors are also split only on the Suns v Pies game, with the Top 4 Predictors on the MatterOfStats Leaderboard opting for the Pies, and the Bottom 4 opting for the Suns. Still, the range of probability predictions in that game is only 26% points.
H2H_Unadj defines the upper bound of that range and is this week, for the first time this season, making a different probability assessment than H2H_Adj. The adjustment has been triggered by the fact that H2H_Unadj's probability estimate of the Suns' chances exceeds the bookmaker's implicit assessment by a shade over 25% points. H2H_Adj has therefore been set to be exactly 25% points higher than the TAB Bookmaker. It's a very small adjustment however - of the order of 0.64% points.
Much greater are the ranges for the Blues v Saints, and the Eagles v Swans games, in both of which the extreme Predictors have gone very close to predicting a win for the underdogs. In the Blues v Saints game, the H2H Predictors rate the Blues as only 51% chances, and in the Eagles v Swans game, C_Prob rates the Eagles 48% chances.
(You might notice in the table above that the Probability Scores I'm showing for each Predictor are higher than previously. That's because I discovered I'd been dividing each Predictor's Score by the total number of games in the season, 207, not the number played so far. The correction doesn't change the ordering of the Predictors of course, since it affected all of them equally, but it does elevate all of their averages.)
C_Prob has the round's largest MAD at 8.5% points per game, ahead of ProPred with 6.4% and WinPred with 6.1%.
In a similar manner to the analysis we performed comparing Margin Predictor MAPEs to their MADs across the season we can for the Head-to-Head Probability Predictors compare their Average Log Probability Scores to their MADs.
For the week, the correlation between ALP and MAD is near zero (it's -0.04) but for the season it stands at -0.63. (Note that, whereas lower MAPEs signify superior predictive ability for a Margin Predictor, higher ALPs signify the same for a Probability Predictor.) Here too then it seems that conformity to the average is generally rewarded.
Returning to the earlier chart to review the Line Fund algorithm's probability assessments we find another set of sharp (ie near 0 and 1) predictions, seven of them rating one team's chances as 65% or better. The Saints are assessed as being 85% chances, the Dees, Giants and Roos as about 75% chances, and the Eagles, Cats and Port as about 65% chances or a little higher. They seem bold, but who am I to argue with an algorithm that sports a 30% season ROI to Kelly-staking.
ChiPS PREDICTIONS
Yet again ChiPS' margin predictions reflect most strongly its Ratings of the relevant teams in each contest, the correlation between its margin predictions and the difference in these Ratings coming in at +0.95 this week.
By way of comparison, the correlation of the margin predictions with the other components are:
- With difference in Form: +0.08
- With HGA: -0.18
- With Interstate Status: +0.18
Just one team, West Coast, is assessed as being deserving of a wager given ChiPS' probability assessments and prevailing TAB prices. So far this season, wagering on ChiPS has not been a profitable endeavour - following its advice would have turned $1 at the start of the season into about 81.6c at the end of Round 15. A win by West Coast this week would reduce that loss by about 12.5c however.
