Predictions 2014-2015

[Crow]

3 of top5 are after production blends. The other two are based on RPM, which with a prior is doing some blending of data seasons (and maybe with SPM too?). Reduction of outliers is an advantage, even if metrics share a lot of "same stuff".

[bbstats]

Any updates?

[Mike G]

Relative to b-r.com forecast, at 77% of the season.

Code: Select all

error abs  -'14py     error abs  -'14py
AJb1  5.5   3.27      v-0   7.1   1.63
bbs   6.1   2.66      fpli  7.1   1.63
crow  6.1   2.63      itca  7.1   1.61
myst  6.1   2.61      bobb  7.3   1.43
HDon  6.5   2.23      AJb2  7.4   1.32
atc   6.6   2.16      snd1  7.5   1.29
ncs   6.7   2.03       eW   8.0    .77
DrP   7.0   1.76      14py  8.8    .00

[Crow]

Giving weight to 3 different Vegas lines pulled down my blend this season. The late emergence of player tracking pm affected readiness to capitalize on it.

[sndesai1]

through last night's games


mike, i'm coming for you lol

[Dr Positivity]

Hard to argue against RAPM after seeing some of the top results, I feel like this year is the most I could ask for in terms of predictions breaking my way, and still not that close to the green names

[EvanZ]

It's sort of like bringing a knife to a gun fight.

Heuristics are great when you don't have any better tools. But in the case of the NBA, I don't think it will be easy to find one that works as well as regression.

[permaximum]

It looks this PTPM developed by using the pure RAPM values here which is multi-year. Since there wasn't PT data before 2013/14, I think the developer of it used one season of PT data to regress into those RAPM values which makes the metric completely incorrect. However it leads atm. Any explanation about this issue would be nice.

If he indeed used the multi-year RAPM values, by using a proper RAPM, the metric can become even better than this.

[Mike G]

May be luck. Last year's winner is nearer the bottom this year.
Also, a record % of entrants are better than Vegas this year. None of us are far behind.

[willguo]

Can the compiler of this data sort by wins on top 5/10 plays rather than RMSE? For ex., if i projected ATL to win 82 games, I'd have beaten Vegas, despite having a much higher RMSE than someone who projected 39.5. Thanks.

[sndesai1]

update through last night's games:



will, i wasn't 100% sure what you meant by top plays, but i basically picked teams with the largest absolute differences from the o/u for each person. also, i'm using the AVG from above for purpose of deciding the bet

top 5

Code: Select all

╔══════════════════════╦════════╦══════════════════╦══════════════════╗
║                      ║ Placed ║ Predicted to Win ║ Predicted Return ║
╠══════════════════════╬════════╬══════════════════╬══════════════════╣
║ Crow                 ║ 7      ║ 7                ║ 90.9%            ║
║ bbstats              ║ 5      ║ 5                ║ 90.9%            ║
║ Dr Positivity        ║ 5      ║ 5                ║ 90.9%            ║
║ AcrossTheCourt       ║ 5      ║ 4                ║ 52.7%            ║
║ Bobbofitos           ║ 5      ║ 4                ║ 52.7%            ║
║ HoopDon              ║ 5      ║ 4                ║ 52.7%            ║
║ italia13calcio       ║ 5      ║ 4                ║ 52.7%            ║
║ Mike G               ║ 5      ║ 4                ║ 52.7%            ║
║ mystic               ║ 5      ║ 4                ║ 52.7%            ║
║ nbacouchside         ║ 5      ║ 4                ║ 52.7%            ║
║ AJ PT-PM             ║ 5      ║ 4                ║ 52.7%            ║
║ ESPN Fall Forecast   ║ 5      ║ 4                ║ 52.7%            ║
║ Five Thirty Eight    ║ 5      ║ 4                ║ 52.7%            ║
║ Pyth + .500          ║ 5      ║ 4                ║ 52.7%            ║
║ Average              ║ 5      ║ 4                ║ 52.7%            ║
║ ESPN Summer Forecast ║ 8      ║ 6                ║ 43.2%            ║
║ AJ ASPM              ║ 6      ║ 4                ║ 27.3%            ║
║ fpliii               ║ 5      ║ 3                ║ 14.5%            ║
║ sndesai1             ║ 5      ║ 3                ║ 14.5%            ║
║ v-zero               ║ 5      ║ 3                ║ 14.5%            ║
║ Arturo Galetti       ║ 5      ║ 3                ║ 14.5%            ║
║ jerrytizzle          ║ 5      ║ 3                ║ 14.5%            ║
╚══════════════════════╩════════╩══════════════════╩══════════════════╝

top 10

Code: Select all

╔══════════════════════╦════════╦══════════════════╦══════════════════╗
║ Entry                ║ Placed ║ Predicted to Win ║ Predicted Return ║
╠══════════════════════╬════════╬══════════════════╬══════════════════╣
║ bbstats              ║ 10     ║ 9                ║ 71.8%            ║
║ nbacouchside         ║ 13     ║ 11               ║ 61.5%            ║
║ italia13calcio       ║ 12     ║ 10               ║ 59.1%            ║
║ Crow                 ║ 11     ║ 9                ║ 56.2%            ║
║ Five Thirty Eight    ║ 11     ║ 9                ║ 56.2%            ║
║ AcrossTheCourt       ║ 10     ║ 8                ║ 52.7%            ║
║ mystic               ║ 10     ║ 8                ║ 52.7%            ║
║ Pyth + .500          ║ 10     ║ 8                ║ 52.7%            ║
║ Average              ║ 10     ║ 8                ║ 52.7%            ║
║ Dr Positivity        ║ 12     ║ 9                ║ 43.2%            ║
║ ESPN Summer Forecast ║ 12     ║ 9                ║ 43.2%            ║
║ AJ PT-PM             ║ 11     ║ 8                ║ 38.8%            ║
║ HoopDon              ║ 10     ║ 7                ║ 33.6%            ║
║ AJ ASPM              ║ 10     ║ 7                ║ 33.6%            ║
║ jerrytizzle          ║ 10     ║ 7                ║ 33.6%            ║
║ ESPN Fall Forecast   ║ 15     ║ 10               ║ 27.3%            ║
║ Bobbofitos           ║ 12     ║ 8                ║ 27.3%            ║
║ sndesai1             ║ 11     ║ 7                ║ 21.5%            ║
║ v-zero               ║ 11     ║ 7                ║ 21.5%            ║
║ fpliii               ║ 10     ║ 6                ║ 14.5%            ║
║ Arturo Galetti       ║ 10     ║ 6                ║ 14.5%            ║
║ Mike G               ║ 11     ║ 6                ║ 4.1%             ║
╚══════════════════════╩════════╩══════════════════╩══════════════════╝

[sndesai1]

It looks this PTPM developed by using the pure RAPM values here which is multi-year. Since there wasn't PT data before 2013/14, I think the developer of it used one season of PT data to regress into those RAPM values which makes the metric completely incorrect. However it leads atm. Any explanation about this issue would be nice.

If he indeed used the multi-year RAPM values, by using a proper RAPM, the metric can become even better than this.

pretty sure the pt-pm is based on one year of PT data versus one year of RAPM - no mismatch

[willguo]

Yeah - that's basically what I mean - on your biggest deviations, how many did you beat the line?

I'm also not sure if you guys are using the opener or the close - these are what I have for the open and close at the LVH, which I think is the largest limit for season win totals. Not quite fair to compare to the open, which came out on Sept 30, and most of the projections came after, say, Westbrook's injury, which shifted the line quite dramatically.

For ex., the ESPN summer forecast would go vs the closing line 15-15 on all games, 11-11 on games with >1 game deviation, 9-4 on games with >2 games deviation, and 4-3 on >3 games deviation.

Against the open, 19-11, 15-5 with >1 game deviation, 10-2 with >2 game deviation, 3-0 with >3 game deviation.






LVH Opener LVH CLose
ATL 40.5 43
BOS 26.5 28
BRK 41.5 41.5
CHA 45.5 44
CHI 55.5 55.5
CLE 58.5 58
DAL 49.5 50
DEN 40.5 42.5
DET 36.5 36.5
GSW 50.5 52
HOU 49.5 48.5
IND 32.5 31
LAC 55.5 56
LAL 31.5 29
MEM 48.5 49
MIA 43.5 44.5
MIL 24.5 24.5
MIN 25.5 29.5
NOP 41.5 43.5
NYK 40.5 39.5
OKC 57.5 53
ORL 28.5 26
PHI 15.5 15.5
PHX 42.5 43.5
POR 48.5 48.5
SAC 30.5 30
SAS 56.5 56.5
TOR 49.5 48
UTA 25.5 28
WAS 49.5 47.5

[sndesai1]

got it - then what i've shown above is pretty much what you're looking for. my lines are william hill from around october 26, so comparing to the lvh close shouldn't be drastically different.

i'll probably switch to the lvh closing lines since you posted them


one thing i don't get is how you get 15-15 for the espn summer forecast? if espn had 56 for the clippers and the closing line is also 56, shouldn't you consider it as no bet?

[permaximum]

It looks this PTPM developed by using the pure RAPM values here which is multi-year. Since there wasn't PT data before 2013/14, I think the developer of it used one season of PT data to regress into those RAPM values which makes the metric completely incorrect. However it leads atm. Any explanation about this issue would be nice.

If he indeed used the multi-year RAPM values, by using a proper RAPM, the metric can become even better than this.

pretty sure the pt-pm is based on one year of PT data versus one year of RAPM - no mismatch

I don't think so. He thought it was one year of RAPM but it isn't.

This is what he wrote abot it.

RAPM is split into offense and defense in terms of estimates of a player's impact on the floor, so I used the two scores to perform two separate regressions weighted by the number of possessions on the court for each player, using box score data and SportVu from NBA.com and RAPM from Jerry Engelmann's Stat's for the NBA.

The RAPM values he used are identical to J.E's 2013/14 Pure RAPM which is multi-year. J.E. can confirm that. Here's the RAPM values he used.



Source: http://counting-the-baskets.typepad.com ... minus.html

It's either he's there by luck or PT data is so good that it still leads despite the fact that RAPM data he used was not accurate enough (It's multi-year but 2013-14 RAPM's weight is more than previous years I think)