Probability Based NRL

[goty0405]

Can only transfer 2 points a day as a non-pro, but very much a fan of the stuff that you have put together above. Find it interesting to follow along and get a background on how to put a model together. I don't use one at all at the moment, but am looking to develop some stuff that will allow me to play in a variety of inefficient markets.

Keep these posts coming!

Thanks Duff!

I have a bit of a passion for mathematical models and (quasi) artificial intelligence so this is something I love doing. Currently I would say this model is very crude but my train of thought is to use NRL as my basis as it is the sport I know the most about. Then when I get to a point where I am confident in the basics of the model I am going to expand to other sports (AFL, Soccer, NBA, MLB etc) and see how it performs.

I'm glad other people enjoy this kind of stuff. I'll make a mental note to create a thread where I outline the basics steps and rules for building a model. But for now, it's all about working on improving the model. Every little edge I can find will help

[goty0405]

Here we go - back in action for round 19. The simulation results are below and I will post my ideas on what I think are some good betting opportunities based on this information shortly.

Round 19

Brisbane vs Gold Coast
B = 85.76%
GC = 10.84%
Draw = 3.40%
Expected Points = 33.10

New Zealand vs Canterbury
NZ = 59.39%
C = 34.23%%
D = 6.38%
Expected Points = 31.68

Penrith vs Parramatta
Pen = 76.12%
Par = 19.18%
D = 4.69%
Expected Points = 35.60

North Queensland vs Wests Tigers
NQ = 77.47%
WT = 18.16%
D = 4.37%
Expected Points = 37.37

Canberra vs Melbourne
C = 16.82%
M = 78.81%
D = 4.37%
Expected Points = 34.89

Souths vs Roosters
S = 75.67%
R = 19.43%
D = 4.89%
Expected Points = 32.23

Newcastle vs Manly
N = 29.68%
M = 64.24%
D = 6.08%
Expected Points = 31.72

St. George vs Cronulla
S = 77.28%
C = 18.03%
D = 4.69%
Expected Points = 33.21

[s2230011]

Sp based on these the plays should be Panthers/Cowboys/Storm and Souths

Obviously JT out throws a spanner in the works, plus they played away on a monday night. I wonder what effect that has on travelling teams

[benrama]

I agree with s22, that it would be good to look at the data without the games immediately before/after Origin.

I also think a star player should be factored into the modeling. In the NBA there is a "star player sitting out" theory that says that teams more often than not cover when their star player is out. It's arguable, but has some merit as punters over-estimate the impact of the star being out and the team impacted lifts their game to try to minimise the impact. See e.g. Monday with the cowboys.

You might also want to look at results broken down by the level by which one team is favoured, e.g. 50-60%, 60-70%, 70-80%

Good work so far!

[goty0405]

Sp based on these the plays should be Panthers/Cowboys/Storm and Souths

Obviously JT out throws a spanner in the works, plus they played away on a monday night. I wonder what effect that has on travelling teams

Well the plays really depend on how you want to use the simulation. I'm thinking any of these could be useful:

1. Backing the simulated winner
2. Backing the simulated winner only if home team
3. 1 & 2 only when odds are "value" when compared to the simulated chances

And then of course those ideas could repeated for multiple permutations such as getting on the favoured team to beat the spread.

I am thinking the Cowboys game will be the most likely to blow out the simulation because it doesnt consider the players or other factors like time since last game. I'm wary about getting on the Cowboys even though it seems like good value but then again the whole point of a mathematical model is to attempt to remove any human interpretation that could be blatantly wrong

I agree with s22, that it would be good to look at the data without the games immediately before/after Origin.

I also think a star player should be factored into the modeling. In the NBA there is a "star player sitting out" theory that says that teams more often than not cover when their star player is out. It's arguable, but has some merit as punters over-estimate the impact of the star being out and the team impacted lifts their game to try to minimise the impact. See e.g. Monday with the cowboys.

You might also want to look at results broken down by the level by which one team is favoured, e.g. 50-60%, 60-70%, 70-80%

Good work so far!

All good points. I definitely want to drill down to this level of analysis, and my ideal model includes factoring in input of every player in every team...but that may be a bit too hardcore for now.

[goty0405]

Some plays I made as tests:

Warriors -6.5pts @ 2.23
Under 35.5 (NZ v Cant) @ 2.02

Brisbane @ 1.27
Brisbane -12.5 @ 2.15

In the long run I probably wouldnt be both ML and Spread markets for Brisbane but I'm just mucking around with small stakes to test the theory and having a little dinky bet on it will be fun

I'm also looking at going Under 36.5pts depending on what the late odds are. I'm hoping to get above 2.0 but I think I will settle for over 1.90.

[goty0405]

I will save a detailed recap for the end of the weekend's matches, but last night went pretty well with both predicted teams winning and covering too. The Under predictions were blown out of the water though.

Tonight...
Souths (vs Roosters) @ 1.82
Penrith (vs Parramatta) @ 1.79
Cowboys (vs Tigers) @ 2.30

All of those are considered value bets where the odds given were significantly better than expected based on the simulation results. Obviously things like JT being out for the Cowboys are not considered which may sway things. Then again if using the simulation last week the Cowboys were predicted to win over Newcastle and they did so without JT and at a very nice price.

Also looking at going Under in the Penrith/Parra (39.5) game and Souths/Roosters games (40.5) but that will be dependent on prices later on.

[goty0405]

A slightly positive night which could have been ugly if not for those golden points wins.

Today...
Melbourne (@ Canberra) @ 1.63

I've jumped on Melb as I've got them as distinct favourites and according to my model they should be much shorter than that price so good value is good value

The odds for the New v Man game are spot on with my predictions so I wont bet the H2H market there (unless Manly drift out past 1.60 before kick-off).

I'll have a look at the Under option for both games (again) if they get up near 2.0 but I'm starting to doubt this part of the simulation. I know my simulation tends towards lower scores which means I will pretty much get on the Under every match. The simulation was 50:50 so far this year which should mean it's pretty accurate estimate of the line but recent results have blown it out with many Over games.

[Duff85]

Yeah I liked Melbourne today. On the under bias - during one of the games last night Andy Raymond said that this has been the lowest scoring nrl season in many years. I think he said going back to the 80s. So possibly the reason your getting a lot of unders out of your data?

[goty0405]

Yeah I liked Melbourne today. On the under bias - during one of the games last night Andy Raymond said that this has been the lowest scoring nrl season in many years. I think he said going back to the 80s. So possibly the reason your getting a lot of unders out of your data?

Good point. It has been a very low scoring season so that would lead to lower totals being predicted. I guess the thing to consider is whether that trend will continue for the rest of the season or not. It definitely hasn't this round!

I've managed to make a few units so far this weekend but if I had ignored the Total Points markets it would have been much better. I've gotten on 5 Unders, all pretty good value outsiders, but only 1 came off.

[Duff85]

Edit: Grrr posted in wrong thread

[s2230011]

Here we go - back in action for round 19. The simulation results are below and I will post my ideas on what I think are some good betting opportunities based on this information shortly.

Round 19

Brisbane vs Gold Coast
B = 85.76%
GC = 10.84%
Draw = 3.40%
Expected Points = 33.10

New Zealand vs Canterbury
NZ = 59.39%
C = 34.23%%
D = 6.38%
Expected Points = 31.68

Penrith vs Parramatta
Pen = 76.12%
Par = 19.18%
D = 4.69%
Expected Points = 35.60

North Queensland vs Wests Tigers
NQ = 77.47%
WT = 18.16%
D = 4.37%
Expected Points = 37.37

Canberra vs Melbourne
C = 16.82%
M = 78.81%
D = 4.37%
Expected Points = 34.89

Souths vs Roosters
S = 75.67%
R = 19.43%
D = 4.89%
Expected Points = 32.23

Newcastle vs Manly
N = 29.68%
M = 64.24%
D = 6.08%
Expected Points = 31.72

St. George vs Cronulla
S = 77.28%
C = 18.03%
D = 4.69%
Expected Points = 33.21

Keep this up I check this thread all the time to see if you have anything new up.

BTW the average number of points per game at the moment is around the 38 points ... so I think your over and under numbers seem pretty low. I know there were some over games lately but have a look

Where do you get past lines from ?

Also, I have a hunch that the last 6 or 8 rounds are much more swayed to favourites that earlier rounds. It would be interesting to see if that was the case in previous years.

[goty0405]

Keep this up I check this thread all the time to see if you have anything new up.

BTW the average number of points per game at the moment is around the 38 points ... so I think your over and under numbers seem pretty low. I know there were some over games lately but have a look

Where do you get past lines from ?

Also, I have a hunch that the last 6 or 8 rounds are much more swayed to favourites that earlier rounds. It would be interesting to see if that was the case in previous years.

Yep the current average points for 2011 is at 38 and yet the simulation model is pretty much always significantly under that amount. The reason for this is to do with how the simulation runs and is actually a common occurrence in statistics. When considering a sample of a population the variance is often lower than the expected variance of the entire population. This is demonstrated by the difference in variance formulas for a population versus a sample. Whilst my simulation model is not specifically relying on the variance formula it does provide a good reason for why it would be consistently under-estimating.

To counter this I have been adding a "threshold" of 3 points when considering my bets. For example, I went Under on the NZ game this round as the estimation said 31.68 but the market was actually at 35.5. I think this would be the best way to use the estimated totals but will probably need to tweak that threshold amount.

[goty0405]

Round 19 is over and it's been a good one for my simulator with most of the predictions coming true. The stats are

Some Stats:
Predicted Team = 7/8 (87.5%)
Predicted Team beat the spread = 5/8 (62.5%)
Under:Over ratio = 1:7 (12.5:87.5 %)

Excluding the significant trend towards overs (see the above few posts as for why they happens) the rest of the results were pretty good.

The only predicted team to not win was the NQ Cowboys and this is particularly interesting because of Thurston being out. I'd prefer to keep things simple and systematic but a possibility in the future is to do some work around factoring "Star" players into the mix.

Or maybe even just ignore a team if they are predicted winners when their star player is not playing. This would have put the results at 7/7 and 5/7 on those first two betting ideas.

The predicted team to beat the spread is also looking good in general. I managed to make a bit of money on that simply because I didn't get on any close spreads ( < 3 points). My reason for this was that it was simply better value to back the team to just win and that theory held up this round with both Souths and Penrith snatching golden point wins but not beating the spread.

So all in all I'm pretty happy with that start.

[goty0405]

Round 20 of the NRL looks like a good one in terms of the match-ups so lets hope my simulator comes through for me again.

I've included the originally predicted total points, which have been significantly under the actual line and result (see above post by me explaining why this happens) and I have also included a manually adjusted figure. The adjustment is not an exact science but is me tweaking the values based on some comparison between past predictions and actual results. I doubt I will be betting on the total match points markets unless I see a line differ significantly from my adjusted figure.

Canterbury vs Parramatta
C: 66.80%
P: 27.56%
D: 5.65%
Expected Points: 35.2 (38.0 adjusted)

Melbourne vs Brisbane
M: 77.83%
B: 17.45%
D: 4.72%
Expected Points: 31.45 (34.5 adjusted)

Gold Coast vs North Queensland
GC: 18.97%
NQ: 76.34%
D: 4.69%
Expected Points: 35.05 (38.0 adjusted)

Souths vs New Zealand
S: 46.91%
N: 46.60%
D: 6.49%
Expected Points: 33.58 (36.5 adjusted)

Wests vs Roosters
W: 76.66%
R: 18.42%
D: 4.91%
Expected Points: 31.52 (34.5 adjusted)

Cronulla vs Newcastle
C: 46.07%
N: 47.31%
D: 6.62%
Expected Points: 32.87 (36.0 adjusted)

Penrith vs Manly
P: 28.40%
M: 66.01%
D: 5.58%
Expected Points: 35.81 (38.5 adjusted)

Canberra vs St George
C: 13.21%
S: 82.85%
D: 3.94%
Expected Points: 32.63 (35.5 adjusted)


Very interesting. A few surprises there, with the biggest being how little Brisbane are rated for their match against Melbourne. I would have given them much more of a chance then that but maybe the simulation knows something I don't.

[s2230011]

To me the bigger surprise is how even the Bunnies and Warriors are ... I think the Storm will win comfortable. Its in Melbourne and they will be up for it, I think after a couple of big games and playing away against a mid tier opposition is when you will see the storm dropping their next points

[s2230011]

ok it seems that the TAB is agreeing with the model on this one ... maybe its me who is off !!

I dont like the Doggies chances either .. Dymock has dropped a couple of players which I dont think is their problem at the moment. Also, when was the last time the Doggies won without Ennis?

[goty0405]

To me the bigger surprise is how even the Bunnies and Warriors are ... I think the Storm will win comfortable. Its in Melbourne and they will be up for it, I think after a couple of big games and playing away against a mid tier opposition is when you will see the storm dropping their next points

Yeah I think the Storm should win but I would have given Brisbane more like a 30-35% chance on the game.

I'm curious...what did you expect for the Souths/NZ game? Who did you think would be favourite?

ok it seems that the TAB is agreeing with the model on this one ... maybe its me who is off !!

I dont like the Doggies chances either .. Dymock has dropped a couple of players which I dont think is their problem at the moment. Also, when was the last time the Doggies won without Ennis?

Yeah all good points and things that the current model isn't able to consider. I was a little surprised at how much the model favoured the dogs too but then because I am a Parramatta supporter I thought maybe I was just being biased.

[s2230011]

I think NZ are heads and shoulders above Souths at the moment, I know its away from home but still its finals time and Warriors know they cant drop games like these, I didnt really see anything to like much from the bunnies against the roosters, I actually thought they were a little lucky to be still in the game late - credit to them they took the opportunity, but the Roosters should have put them away well before then

The thing with the Dogs is they are actually not playing terrible .. everyone is doing an ok job, I cannot see where their improvement will come from, Idris is not interested which leaves Barba their only attacking weapon.

[goty0405]

I didn't get to see the Souths/Roosters game so it's interesting to hear they were a bit lucky to get away with the win. Obviously all golden point winners can consider themselves a little lucky but some teams more than others. NZ haven't been doing that well though. They beat Canterbury and Gold Coast in last 2 matches but that isn't say too much and before that they got beaten by Storm and Cowboys.

I agree about the Bulldogs and think its a similar tale with the Eels. They are trying hard and most players are putting in ok performances but there is just that mystery "something extra" missing.

[s2230011]

Well they beat the Dogs by 24 and Titans by 16 ... They lost to the Storm by 8 and and Cowboys at NQ by 20 but JT was playing. I think the Bunnies belong to the Dogs and Titans bracket of teams more so than a full strength Storm and Cowboys sides.

I agree with you about Parra too. Will have to think about this one

[MrXYZ]

Newcastle/Cronulla's interesting, very even. I think the Sharks can take this one by at least 4.

I've been cashing on the Storm's winning streak for a month but I'm wary of taking them in this spot. Smith starting, their current form & playing at home all point to another easy win, but the Broncos are very capable of an upset.

[s2230011]

I dont rate Newcastle much at all at the moment ... since mullen got back they have declined quite badly. Just beware Gallen might be out and Gidley I think is in. I am not taking Newcastle, but if Gallen is out, I am not sure I can take cronulla either.

I think Melbourne in Melbourne will be too good. I think it would have been an even contest in Suncorp. Its been pouring cats and dogs in Sydney for a couple of days now, so watch those totals

[dRavidC]

hey goty0405 great work with these simulations. just out of sheer curiosity just wondering what numbers you are putting into the system to come up with these simulations (the % chance of victory and the total).

[goty0405]

Some good comments above about the Cronulla v Newcastle game. I'm considering it a no-bet game because the simulation couldn't pick a clear winner. Unknown factors like Gallen possibly out and Gidley possibly in just cloud the issue even further for me.

hey goty0405 great work with these simulations. just out of sheer curiosity just wondering what numbers you are putting into the system to come up with these simulations (the % chance of victory and the total).

The simulation considers a few key stats for each team but the most significant is their points (for and against) in each of their past matches weighted against the past performances (for and against) of their current opposition.

So when a big scoring team comes up against a strong defensive team the "expected" values generally settle around some point in the middle (not factoring in the extra weighting I do for home/away matches and the strength of past opposition).

How I arrive at those percentages is that I run my own version of a monte-carlo simulation. I simulate for at least 1 million trials, each time taking a random sample from within the "probable population" of results for each team and comparing it to the other team. At the end of the 1 million trials I get an output of how many games each team "won" in my simulation and how many points they scored on average.

[goty0405]

Tonights Plays
Bulldogs (vs Eels) @ 2.10
Storm (vs Brisbane) @ 1.55

Another consideration is getting on Storm -4.5 too because that looks like very good value.

The Bulldogs/Eels game should go over the line (28.5 seems very low) but the Storm/Brisbane line looks about right. I wont be betting either though as I'm not confident in using this simulation in those markets.

As a side note that Bulldogs pick is a freeroll for me. I'm using Betfair and they have a promotion going where your first $50 bet on any team to win for the round is refunded if they lose.

[s2230011]

Well its soo low because its been pouring in Sydney for 3 days straight ! I dont know how Homebush will hold up by Sunday's NRL game, I am pretty sure the wallabies are playing there tommorow.
I like the Storm, but think the Eels has more to play for. Good luck mate !

[goty0405]

You were right...the rain made it a very low scoring match. This just highlights that while my model may work for estimating which team should win it is not good enough for estimating total points as it doesn't factor in things like weather, surface, and time of day. But now my interested is piqued. I might run a query on my data and see exactly how these influence the total points result.

Tonights Plays
Bulldogs (vs Eels) @ 2.10
Storm (vs Brisbane) @ 1.55

Another ok night for me. Only "ok" because my beloved Eels lost in golden point for the 2nd week running! Not so bad though as both times I've been on the opposition to win

[goty0405]

Today is an interesting day. My simulations puts NQ as strong favourites but once again it doesnt factor them playing without Thurston. Their current odds are about 2.0 which says they are a 50% chance, yet my simulation rated them a 76-78.5% chance. I guess the question I need to ask is whether JT is worth a 25% drop in their expected chance.

The Tigers look under-valued based on my predictions so I will be getting on there. I also think the Tigers to beat the spread is a good bet so I will toss up exactly which bet I follow through on.

The Cronulla/Newcastle game is too close to call and has a few too many "other factors" that could influence the game - weather, gallen, gidley - so I'm sitting it out. The point of my simulation is to try to find places to put my money with confidence not try to bet every game so I'm happy to be a bystander for this one.

[goty0405]

I did decide to have a sneaky bet on the Cowboys last night but I decided to reduce the stake to 1/2 a unit for the JT factor. I decided that he wasn't worth a 25% shift in winning chance but was definitely enough to make it questionable...hence the 1/2 unit.

I ended up only getting on Tigers over the Roosters in the H2H instead of the them beating their spread.

There are no plays so far today as there is no value in the Pen/Manly game and the Souths/NZ game is a no-bet as its too close to call.

[goty0405]

Playing the Dragons @ 1.39 tonight. Nothing special, just good value when the simulation indicated they should be around 1.20.

[goty0405]

A great game of footy tonight and even though the play didn't come off it was a lot of fun to watch.

At half time, with the Dragons up 18-6 I briefly considered hedging my bet if Canberra were the first to score next as I felt that the there was something there for the Raiders to have a chance at winning. I decided I was being a pussy and needed to follow through on my original bet. I'm still happy with that decision as in the end it was a bit of luck from a short kick-off that really lost it for the Dragons.

[goty0405]

Quick update on those basic stats after Round 20:

This Round Stats:
Predicted Team = 6/8 (75%)
Predicted Team beat the spread = 4/8 (50%)
Under:Over ratio = 5:3 (62.5:37.5 %)

Overall Stats:
Predicted Team = 13/16 (81.25%)
Predicted Team beat the spread = 9/16 (56.25%)
Under:Over ratio = 6:10 (37.5:62.5 %)

[MrXYZ]

keep it up! that's a great record.

[goty0405]

Round 21

Manly vs Wests
M: 81.94%
W: 18.06%

Brisbane vs Cronulla
B: 76.69%
C: 23.32%

Sydney vs Canterbury
S: 51.89%
C: 48.11%

North Queensland vs Penrith
NQ: 70.43%
P: 29.57%

New Zealand vs Canberra
NZ: 76.78%
C: 23.22%

Newcastle vs Gold Coast
N: 78.66%
GC: 21.34%

St George vs South Sydney
StG: 93.55%
SS: 6.46%

Parramatta vs Melbourne
P: 18.81%
M: 81.20%

A few strange predictions in there but mainly around the strength of predicted winners. I think I should get some good "value bets" based on these figures