Smartgambler
Pro-Punter

Go Back   OZmium Sports Betting and Horse Racing Forums > Public Forums > Sports and Gambling
User Name
Password
Register FAQ Search Today's Posts Mark all topics as read

To advertise on these
forums, e-mail us.

Reply
 
Thread Tools Search this Thread Display Modes
  #1  
Old 13th March 2015, 01:00 AM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default NRL Team Ratings -> Value Bets

Decided to make my own thread. Recapping from my posts made in the Rugby League thread: I've made a spreadsheet implementing Massey-style ratings, with Colley-style touches. It can be used for any sport where two final scores are compared to each other (i.e. not something like tennis). I've an interest in NRL so have chosen to live-test it with that sport (though I'm also going to use it for a roller derby tipping competition that's soon to begin).

As an aside, I had a bit of an experiment using it with fantasy sports. Well, in selecting my inital team, anyway. Instead of using the scoreboard results, I used the team totals of supercoach points attained last year. For each player I determined their average percentage of the team's total they earnt each week (ignoring rounds where that were obviously injury-affected). Looking at the draw I came to an estimate of the total points players would earn over the first 10 rounds. I selected the ones whose value was likely to go up the most/earn the most points.

The first round of supercoach, probably better than anything else, demonstrated a major problem with my - and many other - ratings methodologies: teams (and player roles) can change from season to season quite substantially.

ARGH Rankings account for this somewhat by doing some pre-season seeding of teams. According to wikipedia the seeding is "based on several stated criteria, including records from the previous two seasons, the previous year's strength of schedule, the number of returning starters, whether the head coach is returning, whether the starting quarterback is returning, and published recruiting rankings from several preseason publications". Presumably they've done regression testing to quantify the effect of each, using past data; which is something I don't have.

My ratings for last week were created without using time degradation, i.e. the results from Round 1, 2014, were as pertinent to the final ratings derived as were those from Round 26. In the past week I fixed my code so that wasn't the case. To determine the time degradation I'm using a power law; simply because they seem to crop up all over the place. My implementation is flexible enough to apply any weighting to the past 20 rounds, so there is likely a better solution. I'm not quite sure how I would go about finding it. In any case, last season's results will stop affecting ratings as soon as there is enough current data (i.e. rounds played this season) to negate its influence.

Regarding other adjustments I've heard of: I account somewhat for score blowouts. I've read about weighting games, but I'm just taking into account regular season matches, so won't worry about that. I'm unsure about implementing something regarding strength of schedule. From what I understand, if a bad team beats a good team the bad team gets an extra boost (they played extra-well and/or are a team likely on the way up); whereas if the reverse occurs the good team doesn't get such a boost as it was an expected result. The NRL's pretty evenly matched, so I'm not sure it's worth the effort.

Anyway, I develop the ratings, I look at this week's match-ups, calculate the odds based on my ratings, and compare them to the ratings offered by Centrebet. If Centrebet's prices are a certain percentage higher than what I have determined, it's a bet (I'm not being cagey; I'm not sure what that percentage is - I'm monitoring as I go and may bump it up or down in the future depending on results).

The thread is regarding value bets. I don't expect all bets to win. The intent is that they win more often than Centrebet's price indicates.

I haven't really thought much about staking. Last week I stated that I would bet 2% of my (imaginary) starting bank.

Speaking of last week, there were 2 bets for Gold Coast to win H2H (sooo close) and Canberra to win H2H (they did). Post-implementation of time degradation, there was an additional (successful) bet highlighted for Canberra +6.5. That won't count in the tally.
Reply With Quote
  #2  
Old 13th March 2015, 01:08 AM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

...and on to this week...

My tips for NRL Round 2:
Canterbury-Bankstown Bulldogs
Brisbane Broncos
Penrith Panthers
Manly-Warringah Sea Eagles
North Queensland Cowboys
South Sydney Rabbitohs
New Zealand Warriors
St. George Illawarra Dragons

There are plenty of "value" bets this round (i.e. it could be make or break this week! Already!). Stakes are up to $20.32. 1 stake on each, at the specified odds.

Brisbane Broncos - WIN @ 1.91
Gold Coast Titans - WIN @ 4.50
Manly-Warringah Sea Eagles - WIN @ 2.20
Newcastle Knights - WIN @ 3.10
St. George Illawarra Dragons - WIN @ 2.50
South Sydney Rabbitohs : -1.5 @ 1.91
Brisbane Broncos : -2 @ 2.10
Reply With Quote
  #3  
Old 16th March 2015, 11:24 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Another change I made prior to the round just past that I forgot to mention: side-specific Home Advantage. The NRL-wide home ground advantage is worth 3-4 points to the home team. I figured this was likely to vary quite alot between the teams though, rationalising that sides like the Warriors and Cowboys were many hours travel for their opposition and would be most advantaged; vs the Rabbitohs and Bulldogs, for example, who play most home games out of ANZ Stadium: a bus ride for the majority of teams and a ground that everybody would likely be familiar with.

During the TV coverage I heard that the Sharks had lost their last 9 home games; my calculations do indeed have them as the side least advantaged by playing at home. In fact, it's a *disadvantage* for them, being an average 6 points worse off than if they played anywhere else! The only other team who are worse off at home rather than away are the Tigers, maybe because their home games are scattered between Leichardt, Campbelltown and ANZ Stadium.

Rabbits, Roosters, Cowboys, Melbourne and Manly appear to be most advantaged by playing on their home field. Some of them possibly are positively affected by travel distance, one perhaps by playing field/facilities "quirks", and perhaps the other 2 by....good teams being even better after sleeping in their own bed?

_
I'm still not happy with how I handle games at the start of the season. Obviously initial ratings should be seeded by results from the previous year, but I haven't quite figured out how best to implement it. At present I'm using the same time degradation as I do for current season results but that has the effect of putting more stock in end-of-last year results: fine if a side is still challenging for the top 8; perhaps not so representational if they'd already put the cue in the rack by that stage. I've a sneaking suspicion this methodology has over-penalised Wests Tigers - who had a horrid end to season 2014 - and have been responsible for 2 of my 3 failed bets thus far.

Speaking of, results for this week:

Tipping was 6/8, for a running total of 10/16 (62.5%).

Bets with stakes of $20.32.

Brisbane Broncos - WIN @ 1.91
Gold Coast Titans - WIN @ 4.50
Manly-Warringah Sea Eagles - WIN @ 2.20
Newcastle Knights - WIN @ 3.10

St. George Illawarra Dragons - WIN @ 2.50
South Sydney Rabbitohs : -1.5 @ 1.91
Brisbane Broncos : -2 @ 2.10


This Week
Out: $142.24
In: $227.99
60.3%POT

Overall
Out: $182.24
In: $283.99
55.8%POT

New Bank: $1101.75 (stakes for next round: $22)

H2H S/R: 4/7 (57.1%)
Line S/R: 2/2 (100%)

_
I also noted this week that my methodology for picking a "value" bet was somewhat flawed. You can Pick Your Own Line in these games and I discovered that for one of the teams, *every* offered margin (and they're only 1 point apart) below a certain value would have qualified as a value bet. I reasoned that I need to start also taking into account, likelihood. Perhaps it's a reflection of my own temperament regarding loss (after all, if you're getting 100-1 on a horse that is actually a 20-1 chance, it's a good thing and you should come out ahead; eventually) but I'm thinking I should set a dollar ceiling, perhaps omitting to bet on prices longer than $4 in these kinds of markets.

This week I intend to solve the pre-season seeding problem (to an extent: I won't look at accounting for changes in roster strength, etc - though I believe this would also be quite helpful during Origin, in response to player's injuries, and so on; but I just don't have the data). I also want to run a few season's worth of estimates so I can determine the stdev for the error. With this I should be able to develop prices for game totals, margins, etc.
Reply With Quote
  #4  
Old 20th March 2015, 01:43 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Last post I said I was intending to work on a number of things during the week:

- Pre-season seeding problem: I've come to the conclusion that I'm handling it (almost) as best as I can. The theory was the previous season/s rating/s would influence this season's ratings BUT you obviously can't get sensible ratings solely as a function of this season's results until there are more than a handful of matches to study. My current modus operandi - treating matches as a continuous "blob" despite a season break in the middle - seems a good compromise. The games of this season have more of a bearing on the ratings by virtue of their timeliness - just as I would expect during season 2015 that Round 9 says more about Round 10 results, than does Round 1. I do have an idea regarding how to cope with team changes/injury/representative duty that may assist further - I'll discuss it later.

- Analysis of past results: lengthy text below

I have a couple of season's -worth of scores which I was able to test against my predictions. As mentioned previously I can specify how much weight I give to previous game results, using a model of diminishing returns based on how long ago the game was played (I refer to how many rounds ago, but - had I the data - it could very well be number of weeks ago; applying even less relevance to previous season's results; wish I'd thought of that earlier...). In determining past performance I thought I would trial a few different weighting methods, using a curve like: 1 - a normal distribution, 2 - a power law, 3 - a sigmoid. Of the three the sigmoid curve (a stretched, backward "s") gave slightly better results. I may do further experimentation with different variations of this curve and see if there are further improvements.

In comparing my team score predictions vs the actual results I discovered that my mean error is -0.4. Which sounds pretty good. Unfortunately, the stdev of the error is in the order of 13.4 points. Over the course of the season the estimates are quite accurate, but on any one individual match, the predictions may be way out: 2 converted tries either way, for both teams! As a consequence of this discovery I've lost a lot of confidence in my existing methodology of finding value bets. While the results have been impressive thus far, it's likely down to a fair measure of luck (along with outsiders being quoted longer odds than they warrant, perhaps). ...but I have come up with something a little more mathematically sensible - described in a follow-up post.

Another interesting discovery, made by accident: on average, my predicted score for home teams was 4.1 points greater than it should be. That's pretty much the same as the league-wide home ground advantage figure I came up with. Away teams scores were on average 2.7 points less than they should be. So should the "home ground advantage" (or a portion there-of) be added to the visiting team instead?! Looking at individual teams it wasn't quite so cut and dried, but only a small number of home games (per team) were available for comparision and undoubtedly any results would've been skewed by strength of opposition. Following this discovery I'm at a bit of a loss regarding what to do: either add 0.7 * the "home ground advantage" to visiting teams (as bizarre as it sounds it's what the data suggests) vs keep logical consistency despite the poorer results (!)

These picks are using the same methodology (the "logical" rules) as was applied in past weeks

Tips:
Manly-Warringah Sea Eagles
Brisbane Broncos
New Zealand Warriors
Canberra Raiders
Melbourne Storm
South Sydney Rabbitohs
Newcastle Knights
Sydney Roosters

"Value Bets":
Parramatta Eels WIN @ $3.10
Cronulla-Sutherland Sharks WIN @ $3.80
Wests Tigers WIN @ $4.65
Melbourne Storm -10 @ $1.91
South Sydney - 12.5 @ $1.91
Reply With Quote
  #5  
Old 20th March 2015, 02:01 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Quote:
Originally Posted by walkermac
These picks are using the same methodology (the "logical" rules) as was applied in past weeks


Nope. Stuffed up. Used the wrong file.

Tips shoud be:
Canterbury-Bankstown Bulldogs
North Queensland Cowboys
Parramatta Eels
St. George Illawarra Dragons
Melbourne Storm
South Sydney Rabbitohs
Newcastle Knights
Sydney Roosters


"Value Bets" should be:
Parramatta Eels WIN @ $3.10
St-George Illawarra Dragons WIN @ $2.25
Cronulla-Sutherland Sharks WIN @ $3.80
Wests Tigers WIN @ $4.65
Penrith Panthers WIN @ $2.95


Stakes still at 2% of bank
Reply With Quote
  #6  
Old 20th March 2015, 03:19 PM
Rinconpaul Rinconpaul is offline
Banned
 
Join Date: Feb 2013
Posts: 1,645
Default

Don't punt sports at all, but an observation if I may?

You're missing a few factors possibly, like star players OUT while they front court on any range of drug, alcohol, public indecency or assault charges. In hospital having surgery, recovering from injury, sick of the coach or club and working on a transfer, clubs having financial dramas etc, etc

And then the team ranks can be bolstered when someone is released early from prison or their ban is over. So much extra curricular activity going on, the mind boggles at how the hell you can just apply stats of past home & away scores and hope to get a correct outcome? Like all sports, too many of the wrong sort of variables.
Reply With Quote
  #7  
Old 20th March 2015, 03:23 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Quote:
Originally Posted by walkermac
...but I have come up with something a little more mathematically sensible - described in a follow-up post.


Determination of a value bet - as above - depends on the accuracy of my predicted score. I use the predicted score to determine the probability and in turn the price. During my data analysis I discovered that while my prediction was close to spot-on (taking the averages over the season), the standard deviation (on any one game) was quite large. The best approach then, seems to be a Monte Carlo Simulation.

Given my predicted score for home team A and presuming the past data is predictive, I know that the actual score should be -4.1 points less + some random number of points that I'm 95% certain lies between + or - 1.96 * 13.4. I also know the range of values that away team B lies within, with the same certainty. Though A and B only play once on the field, in the computer they can play thousands of times. Restricting their scores to the range and probability that past data suggests, I can determine the number of times Team A wins, or wins by a certain margin, or the score is above a certain amount, etc. This, in turn, I can change into a price.

Using this method, tips are:
Canterbury-Bankstown Bulldogs
North Queensland Cowboys
New Zealand Warriors
St. George Illawarra Dragons
Melbourne Storm
South Sydney Rabbitohs
Newcastle Knights
Sydney Roosters

Using this approach I get the following market (value bets - taking Centrebet's current prices - are marked):

Manly vs Canterbury
Manly: $2.06
Canterbury: $1.95
Manly +2.5: $1.85
Canterbury -2.5: $2.17
Manly 1-12: $4.10
Manly 13+: $4.13
Canterbury 1-12: $4.01
Canterbury 13+: $3.78
Game Total Over 38.5: $1.78
Game Total Under 38.5: $2.29*


Brisbane vs North Queensland
Brisbane: $2.82
North Queensland: $1.55*
Brisbane -2: $3.17
North Queensland -2: $1.65*
Brisbane 1-12: $4.81
Brisbane 13+: $6.84
North Queensland 1-12: $3.82
North Queensland 13+: $2.61*
Game Total Over 38.5: $1.67
Game Total Under 38.5: $2.49


New Zealand vs Parramatta
New Zealand: $1.65
Parramatta: $2.53*
New Zealand -8: $2.29
Parramatta +8: $1.78
New Zealand 1-12: $3.85
New Zealand 13+: $2.90
Parramatta 1-12: $4.54
Parramatta 13+: $5.72*
Game Total Over 42.5: $1.91
Game Total Under 42.5: $2.10


Canberra vs St George-Illawarra
Canberra:$2.90
St George-Illawarra: $1.50*
Canberra -3.5: $3.72
St George-Illawarra +3.5: $1.37*
Canberra 1-12: $5.08
Canberra 13+: $7.29
St George-Illawarra 1-12: $3.87
St George-Illawarra 13+: $2.45*
Game Total Over 40.5: $1.90
Game Total Under 40.5: $2.11


Melbourne vs Cronulla
Melbourne: $1.22
Cronulla: $5.63
Melbourne -10: $1.54
Cronulla +10: $2.86
Melbourne 1-12: $4.49
Melbourne 13+: $1.67
Cronulla 1-12: $8.17
Cronulla 13+: $21.68
Game Total Over 36.5: $1.94
Game Total Under 36.5: $2.06


South Sydney vs Wests
South Sydney: $1.09
Wests Tigers: $12.37
South Sydney -12.5: $1.30*
Wests Tigers +12.5: $4.29
South Sydney 1-12: $6.56
South Sydney 13+: $1.30*
Wests Tigers 1-12: $16.20
Wests Tigers 13+: $52.27
Game Total Over 44.5: $1.77
Game Total Under 44.5: $2.30


Gold Coast vs Newcastle
Gold Coast: $7.20
Newcastle: $1.16*
Gold Coast +2: $6.11
Newcastle -2: $1.20*
Gold Coast 1-12: $10.07
Gold Coast 13+: $25.28
Newcastle 1-12: $5.16
Newcastle 13+: $1.50*
Game Total over 40.5: $2.09
Game Total Under 40.5: $1.92


Sydney vs Penrith
Sydney: $1.38
Penrith: $3.61
Sydney -7.5: $1.73
Penrith +7.5: $2.36
Sydney 1-12: $3.95
Sydney 13+: $2.13
Penrith 1-12: $5.80
Penrith 13+: $9.57
Game Total Over 37.5: $1.33
Game Total Under 37.5: $4.06


Cross-referencing with the "old style" value bets...and there must be some doubt about most of them! Perhaps it lends slightly more confidence to Parramatta Eels WIN @ $3.10 and St-George Illawarra Dragons WIN @ $2.25
Reply With Quote
  #8  
Old 20th March 2015, 03:27 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Quote:
Originally Posted by Rinconpaul
So much extra curricular activity going on, the mind boggles at how the hell you can just apply stats of past home & away scores and hope to get a correct outcome? Like all sports, too many of the wrong sort of variables.


I do have a bit of an idea about that, as alluded to briefly in the first post of the day. Paraphrasing Fermat, however: "I have to go out to dinner and don't have time to write it down". I'm sure it would sound better in Latin...
Reply With Quote
  #9  
Old 20th March 2015, 04:48 PM
Rinconpaul Rinconpaul is offline
Banned
 
Join Date: Feb 2013
Posts: 1,645
Default

Quote:
Originally Posted by walkermac
I do have a bit of an idea about that, as alluded to briefly in the first post of the day. Paraphrasing Fermat, however: "I have to go out to dinner and don't have time to write it down". I'm sure it would sound better in Latin...


Perhaps from the Latin verb "tangere", means to touch. Intangibles are something that's not physical and of the mind. In other words what's in the mind of the players can make all the difference between winning and losing.

I guess that's where your standard deviation comes into play to cover the constant turmoil in the minds of players from game to game?
Reply With Quote
  #10  
Old 21st March 2015, 08:58 PM
walkermac walkermac is offline
Member
 
Join Date: Nov 2013
Posts: 1,036
Default

Quote:
Originally Posted by Rinconpaul
I guess that's where your standard deviation comes into play to cover the constant turmoil in the minds of players from game to game?


It definitely would be part of it. Psyche is a known unknown, as it is for horseracing. I presume the majority of the standard deviation though is accounting for "game momentum". Two teams can be perfectly matched, but something as innocuous as a 20m restart versus a forced line drop out can have a big effect on the scoreline. In what's a matter of inches, the latter case sees the attacking team get the ball back against a tiring opposition, in the other the defenders get a 20m headstart downfield and a bonus tackle (I think I heard the stat that something like 20% of these restart sets end in a try).


Quote:
Originally Posted by Rinconpaul
Don't punt sports at all, but an observation if I may?

You're missing a few factors possibly, like star players OUT while they front court on any range of drug, alcohol, public indecency or assault charges.


Quote:
Originally Posted by walkermac
I do have an idea regarding how to cope with team changes/injury/representative duty that may assist further - I'll discuss it later.


Massey Ratings - and a bunch of other methods that have featured in far too many Phd theses - are borne from a problem regarding US College sport: there are hundreds of colleges and each team only plays a couple of dozen matches; how can it be determined which teams should be playing in the end-of-season championships? From this perspective, no-one cares who was on the field or not, results are all that matter. Beginning with this retrodictive approach, people added tweaks to predict a margin if two teams would theoretically meet, and finally comparisons of offensive and defensive ratings which would in turn indicate a hypothetical scoreline.

As we've both noted, in the practical realm it doesn't translate very well given the effect missing players (be it through swapping teams, injury, suspension or representative duty) can have on a team's fortunes.

To account for this, players need to have a value attached to their presence. If this were real life, the people in the best position to determine a player's worth are their club. And the best way to measure their relative worth, is their wage. This has problems of its own: 1. we're not privy to player's salary packages (let alone the brown paper bags...), 2. it's not very timely (contracts span seasons and form may drop year to year, let alone week to week), 3. they're not uniform (it's said it costs 10% extra to entice a player to switch clubs and probably an even greater premium to get cattle to less-successful teams), 4. they're not entirely aligned with on-field performance (e.g. while Todd Carney was a pretty good player, his off-field indiscretions likely saw him as a bargain - as far as the accountants were concerned).

Players are part of another economy however that is removed from much of these problems. It's transparent, reactionary, uniform across the clubs and blind to anything that goes on off-field: it's fantasy sports.

A brief primer for NRL Supercoach: armchair players have a salary cap from which they must select a squad of 25 NRL players. At the start of the year they are priced as a function of their average score from the previous season. Players score by performing positive feats on the field in real matches: tries, hit-ups, tackles, offloads, etc. They are penalised for negative feats on the field: errors, missed tackles, penalties, etc. At the completion of each round their $-value varies by virtue of their performance: armchair players can buy and sell players with respect to this value. The players who perform the most positive feats on the field (be it from making a heap of tackles and/or scoring a bunch of tries) become the most expensive - and arguably the most relevant to an actual game's result.

With this information we can look at a past match, study the teamlist and say it was achieved by a team worth $2.5mil. In the upcoming round, if the star player earning $500,000 a season is replaced by a rookie (at $100,000), we might expect a team to perform in the order of 2.1/2.5 * the rating indicated by the previous result (i.e. upcoming team value divided by past result team value).

It seems like a pretty good solution - ignoring the variation that can come from mid-game injuries, the coach's whims regarding bench time, sin-binning or a change in role when switching clubs (all of which can only be accounted for in retrospect).
Reply With Quote
Reply


Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is On
Forum Jump



All times are GMT +10. The time now is 12:03 PM.


Powered by: vBulletin Version 3.0.3
Copyright ©2000 - 2018, Jelsoft Enterprises Ltd.
2008 OZmium Pty. Ltd. All rights reserved . ACN 091184655