Probability Based NRL

[goty0405]

I'm a bit of a nerd for data, stats an anything math related so I often get involved in building tools to automatically retrieve data, process it in some way and then spit out various information.

My latest project has been to build a probability simulator which simulates upcoming matches for sports based on the past performance of teams. It's been fun building it and now it's time to test it!

Backtesting on the NRL if I had backed the winner from my simulator in each match over the past 3 years it would have had 62-65% success (depending on various settings). So it seems viable but there's only one way to find out.

I'll post the results of the simulation for this weeks games soon and then talk about how I will be using the information to pick my bets.

[goty0405]

After running 100,000 match simulations (with settings to weight the last 5 games as more important than previous games) the results are:

Round 18

New Zealand vs Gold Coast
NZ = 70.03%
GC = 24.56%
Draw = 5.41%
Expected Points = 34.05

Parramatta vs Wests
p = 29.70%
W = 64.49%
D = 5.81%
Expected Points = 34.82

Cronulla vs Canberra
Cro = 44.57%
Can = 49.25%
D = 6.18%
Expected Points = 37.47

Canterbury vs Melbourne
C = 21.82%
M = 73.04%
D = 5.14%
Expected Points = 33.61

Sydney vs Penrith
S = 59.24%
P = 34.70%
D = 6.05%
Expected Points = 35.57

Manly vs Souths
M = 72.50%
S = 22.57%
D = 4.93%
Expected Points = 37.60

Newcastle vs North Queensland
New = 53.14%
NQ = 40.65%
D = 6.21%
Expected Points = 37.05

[Beniphone]

Hmmm, is this for the spread or ML? 65% for ML is pretty weak

[MrXYZ]

Good luck with your project.

Your New Zealand/Gold Coast total at 34.05 is interesting. Like most punters, when I first saw 36.5 being offered by the books, the over seemed like the play but might be worth fading the public on this one. Gold Coast will struggle to get over the line IMO.

[goty0405]

Hmmm, is this for the spread or ML? 65% for ML is pretty weak

That figure is just the ML and I disagree that it's weak. That is the strike-rate for betting very match, even those where the favourite is only slightly ahead (e.g. see the Cronulla vs Canberra game listed above).

However I do think that it can be greatly improved with some tweaking to the simulation algorithm and also including other factors, like the "strength" of the favourite, home ground advantage, and so on. That's what I'm going to work through with this thread and hopefully reach a conclusion by the end that enables me to have a solid algorithm ready for 2012 season.

[goty0405]

Good luck with your project.

Your New Zealand/Gold Coast total at 34.05 is interesting. Like most punters, when I first saw 36.5 being offered by the books, the over seemed like the play but might be worth fading the public on this one. Gold Coast will struggle to get over the line IMO.

Yeah it's very interesting. A lot of people tend to overrate the overs...I'm thinking the simulation errs slightly (1-2 points) on the under side of points scored simply because of the standard binomial distribution it uses to determine each team's points for each trial. So factoring that in the 36.5 may be a pretty accurate line. Definitely worth looking at the under though if everyone jumps on the over

[goty0405]

New Zealand vs Gold Coast
NZ = 70.03% (72.73% adj = 1.37)
GC = 24.56% (27.26% adj = 3.67)
Draw = 5.41%
Expected Points = 34.05

Parramatta vs Wests
p = 29.70% (32.6% adj = 3.07)
W = 64.49% (67.39% adj = 1.48)
D = 5.81%
Expected Points = 34.82

I've put my "adjusted" figures and their decimal odds equivalent in the above quote. All I did was just spread the draw chance around in order to get estimated odds for ML markets that don't offer the draw option.

So make of that what you will. There are many ways this sort of information could be used but as this is a new theory I am going to play it a little cautious and only bet where I can get good value on my predicted winner.

New Zealand vs Gold Coast - None (NZ currently ~ 1.34)
Parramatta vs Wests - Wests @ 1.75

I will also keep an eye on the total points markets. Both games are predicted to come under the lines (36.5 and 37.5 respectively) so I am hoping to see a few people smash the overs and blow the unders odds out for me.

[goty0405]

Not too bad last night. I lost on the only H2H bet I made:

Parramatta vs Wests - Wests @ 1.75 - Lost

But I managed to get late bets on the Unders for both games (1.99 and 2.08 respectively) and both came off so that was nice.

Overall Prediction Stats
H2H: 1/2 at -0.66 units
Total: 2/2 at +2.07 units

My Value Bet Stats
H2H: 0/1 at -1.0 unit
Total: 2/2 at +2.07 units

[goty0405]

Cronulla vs Canberra
Cro = 44.57% (47.66% adj = 2.10)
Can = 49.25% (52.34% adj = 1.91)
D = 6.18%
Expected Points = 37.47

Canterbury vs Melbourne
C = 21.82% (24.39% adj = 4.10)
M = 73.04% (75.61% adj = 1.32)
D = 5.14%
Expected Points = 33.61

Cronulla vs Canberra - Canberra @ 2.31
Canterbury vs Melbourne - Melbourne @ 1.48

[goty0405]

The total points lines are up for tonight's matches and they are at 40.5 and 34.5 respectively. Even factoring in the estimated 1-2 points that my simulator seems to lean towards the under that 40.5 line for Cro v Canb seems high. In fact I believe that odds 1.90+ will be good value for Under 40.5 and will bet it accordingly.

The line for Cant v Melb seems pretty accurate compared to my simulation with maybe a slight preference towards over. I will keep an eye on the odds but it seems likely that no betting will occur.

[Duff85]

I am interested to see how this goes. As far as winning percentages go - it means nothing if your betting varying mls.

I'm hitting 50.63% (40-38) this year at nrl, but i'm ahead 3.85 units. Conversely someone who loved favourites would be down money with a higher win percentage than that.

Units won and lost will be the true test.

I should say too i'm especially interested in how your totals do, because i'm yet to come up with anything that gives me an edge in these.

[goty0405]

I am interested to see how this goes. As far as winning percentages go - it means nothing if your betting varying mls.

I'm hitting 50.63% (40-38) this year at nrl, but i'm ahead 3.85 units. Conversely someone who loved favourites would be down money with a higher win percentage than that.

Units won and lost will be the true test.

Agreed. The 62-65% mark was for betting every game (after round 1) in past 3 years so it was just to give a rough idea of accuracy. Worst case scenario I could use it to enter tipping comps and finish close to the top

But yeah the real test is comparing the picks to actual odds and measuring ROI. No good getting 60% if they are all 1.20 favs. I'm thinking of obtaining some historical odds data just to satisfy my interest.

I should say too i'm especially interested in how your totals do, because i'm yet to come up with anything that gives me an edge in these.

I think its one of those markets that often the most likely outcome can seem a little counter intuitive, which is why I like using stats. I've already got an idea on how to make the total points predictions more accurate too so when I make that change it will be very interesting.

Good luck.

[goty0405]

Got some money on Under 40.5 @ 1.94. Could have gotten better too but I missed the market move.

[YorkHunt]

Ehh not lookin good.. 16 pts already

[Duff85]

Got some money on Under 40.5 @ 1.94. Could have gotten better too but I missed the market move.

Cheering for this one to hold for you.

[MrXYZ]

Cronulla defense holds! Well done.

[goty0405]

Phew! Saved by the bell..thanks for the support guys
Now let's hope Canterbury and Storm play out a low scoring affair (on Under 34.5 @ 2.11) in a wet Adelaide so I can go 4/4 on the totals for the weekend.

[goty0405]

Cronulla vs Canberra
- Canberra @ 2.31 Lost
- Under 40.5 @ 1.94 Won

Canterbury vs Melbourne
- Melbourne @ 1.48Won
- Under 34.5 @ 2.11 Lost


Overall Prediction Stats
H2H: 2/4 at -1.18 units
Total: 3/4 at +2.01 units

My Value Bet Stats
H2H: 1/3 at -1.52 unit
Total: 3/4 at +2.01 units

[goty0405]

Sydney vs Penrith
S = 59.24% (62.27% adj = 1.61)
P = 34.70% ( 37.73% adj = 2.65)
D = 6.05%
Expected Points = 35.57

Manly vs Souths
M = 72.50% (74.96% adj = 1.33)
S = 22.57% (25.04 adj = 3.99)
D = 4.93%
Expected Points = 37.60

I was looking at those values from the Sydney/Penrith game and I couldn't figure out why Sydney are favoured so much. I went back through the results and it still didn't make sense...So I checked out my database and found an error. The last 3 games from last round had failed to update which would throw a massive spanner in the predictions for this game and a few others for this round.

So I am re-running my data retrieval and then will resimulate this round. Hopefully in the next 20 minutes before kick-off

[goty0405]

Oh dear. I fixed that little error but found out that it was only masking a bigger error which was skewing the simulation results significantly. Serves me right for rushing to finish it on Friday so I can start test bets. Luckily I was using small stakes.

It wont take me long to fix but it means that this week will essentially be written off as erroneous and I will have to start again. Back to work...

[goty0405]

Been "working form home" the past few days so I've spent most of that time tweaking my simulator program. I fixed a couple of errors that were skewing results and am now re-running it with various settings to find the best. Going to go crazy looking at all these numbers

[Meyhem]

Hmm, just can get how you can get a effective simulator for NRL....

Best of luck for it

[Duff85]

I'm keen to see what your numbers are for the previous week when yourreadjust them. Looking forward to seeing the totals for this week.

[binomial]

do you consider whether it is a day or night game?

[goty0405]

Hmm, just can get how you can get a effective simulator for NRL....

Best of luck for it

Is that because of the up and down nature of the competition? I agree. In reality I don't think an NRL simulator will pick any more games correct than I could do myself. I think the real value will come from combining it with some extra level (like only betting if the odds provide value) or using the simulator to tackle other markets like total points, margins, etc.

I'm keen to see what your numbers are for the previous week when yourreadjust them. Looking forward to seeing the totals for this week.

Coming soon...I am re-testing now to find a good starting point for settings for things like weighting home vs away games, weighting recent games more, and how many games to consider (all? last 5? last 10?). Once I find a starting point that I am happy with I will put last week's up.

do you consider whether it is a day or night game?

Not yet but including extra variables like that is on my to do list. I'm curious though - how do you think the time of day influences a game? I was going to add things like that so it applies different weights to past performances based on time of day.


If there is enough interest in the near future I can do a post with the variables I currently use (and what I plan to add in the future) and we can brainstorm how much they affect a match and what sort of values could be used.

[benrama]

I'm a senior analyst who's done plenty of number crunching over the years, and I find this type of thing interesting - keen to see how you progress with it. I've wanted to do something similar myself but never been willing to put in the hours - plus, I think being successful at sports betting still comes down to situational analysis.

I'm interested to hear what variables you are currently using and to give some feedback. Off the top of my head I'd probably be looking to play around with:

Broken down for home and away (where applicable):
- W/L record in last 10, season
- W/L record vs Top 8, vs non-Top 8, also average margin of W/L
- Previous H2H record
- W/L record @ ground
- Effective tackle %
- Metres gained
- Pts for and against
- Line breaks
- Penalties
- Completion rate %

Would also be looking at giving each player in the whole comp a rating from 1-10, and I think you could look at using existing Fantasy leagues for this. I think this alone could form a system (e.g. if a team is playing at home, and has a higher total of fantasy points and they are getting a points head start then bet - perhaps even in proportion to how much they are ahead on points)

[goty0405]

Time for an update on the probability simulator. I'll try to keep it brief but I can get carried away when talking numbers, statistics, probability and anything else nerdy.

Firstly, I've found the "best" settings for the 2011 NRL season. I've done things through a roughly trial and error method but on my todo list for future version is to incorporate some machine learning (where I back-propogate the results and it changes it's settings to increase performance).

I have specifically focussed on the 2011 season which has turned out to be very different to the 2008-2010 season data I have. For example compare the results for the current settings on just 2011 season and then all data:

2011

• Overall: 65%
• Home wins when favoured: 76.5%
• Away wins when favoured: 52.5%
• Under:Over for the estimated total = 50:50%
• Overall picking favoured team to beat the spread: 58.65%
• Home team (if favoured) beat the spread: 63.79%
• Away team (if favoured) beat the spread: 52.17%

All data (2008-now)

• Overall: 59%
• Home wins when favoured: 63.5%
• Away wins when favoured: 51.5%
• Under:Over for the estimated total = 37.6:62.4%
• Overall picking favoured team to beat the spread: 51.97%
• Home team (if favoured) beat the spread: 54.96%
• Away team (if favoured) beat the spread: 46.40%

As you can see the current settings are much better for 2011 than the rest of the data. This makes sense but also means that a generic formula is probably unachievable and some level of customisation will be required each season. I'm ok with that though as the majority of the data and settings will remain the same.

So looking at the stats just for this year there are a few interesting things to note which could be used for betting systems:

Interesting Points for 2011

• 65% overall is not too bad. It's probably not much better than most professional NRL analysts could pick and may not yield profit straight up but could serve as a basis for other theories.
• Predicted favourites (not actual favourites) playing at home get the points 76% of the time. That's pretty big and while a strong team will be shorter odds at home I think there is the possibility for profit there.
• The estimated total points has been spot on so far this year with a 50:50 split of match results. I am surprised by this as I thought it was significantly lower than expected. Interestingly when applied to all past seasons the estimated points was under-estimating by an average of 4 points.
• The favoured team beats the spread 58.65% (63% if just considering home favourites) which seems expected. A note to remember is that this is based off historical line data I have from an Australian TAB so obviously some places would run different lines which would change the results. But it still gives a good indication of what to look for.

There are a few more obvious points and ideas to draw from those stats, and lots of betting strategies that could arise from all the permutations. I might do up a shortlist in one of my next posts as something for me to track over the next few rounds.

[s2230011]

I may be misunderstanding but when a home team is favoured they win 76.5% of the time and cover the spread 83%? How?

[goty0405]

So with the above post in mind it's time to re-run the simulator for last week...

Round 18

New Zealand vs Gold Coast
NZ = 74.14%
GC = 20.85%
Draw = 5.00%
Expected Points = 33.73

Parramatta vs Wests
p = 37.26%
W = 56.57%
D = 6.17%
Expected Points = 35.76

Cronulla vs Canberra
Cro = 55.98%
Can = 37.87%
D = 6.15%
Expected Points = 35.94

Canterbury vs Melbourne
C = 23.38%
M = 71.17%
D = 5.45%
Expected Points = 31.91

Sydney vs Penrith
S = 29.42%
P = 64.56%
D = 6.02%
Expected Points = 31.79

Manly vs Souths
M = 92.77%
S = 5.20%
D = 2.02%
Expected Points = 34.00

Newcastle vs North Queensland
New = 45.40%
NQ = 48.08%
D = 6.51%
Expected Points = 33.64


Stats:
Head 2 Head: 5/7 (71.4%)
U/O Expected Points: 2:5 (28.6:71.4%)

A few big scoring matches blows the U/O out but the good news is that most bookies had their lines much higher than mine which would have brought it back closer to even.

[goty0405]

I may be misunderstanding but when a home team is favoured they win 76.5% of the time and cover the spread 83%? How?

Those values are for betting those markets only when the simulation has the home team as predicted winner. I think the 83% is incorrect (as it is just a recent addition to my simulator) but I wanted to put it up and see if anyone else thought it looked high. As you were instant-reply it makes me think there is something fishy in my data.

The 76.5% I have tested multiple times. It would have resulted in 46/59 bets so far this which is actually 77.97%. The original stats didn't include last week's results.

[goty0405]

Found the problem with the spread stats already.

The bookie odds format changed half way through the season (who does that?) where the flipped who the spread was referring to. So some of those games when the home team have a -ve points start my stats considered it a +ve points start.

All fixed and I've updated the stats in my original post. It looks much saner now. Thanks

[s2230011]

That looks better ... I guess if you were taking an away favourite to win you might as well take them to cover.

So based on this we would 58% of the time if we just take favourites to cover ! interesting.

I know this is a hassle ... but take out the 3 weeks worth of games where origin stars are out and I have a feeling the numbers will change a bit. I believe more underish games and also teams not covering. ... just a hunch

[goty0405]

That looks better ... I guess if you were taking an away favourite to win you might as well take them to cover.

So based on this we would 58% of the time if we just take favourites to cover ! interesting.

I know this is a hassle ... but take out the 3 weeks worth of games where origin stars are out and I have a feeling the numbers will change a bit. I believe more underish games and also teams not covering. ... just a hunch

Good point on the origin but do you only consider the games where the stars are out? What about the following games where they backup? And what about a team with no origin players versus a team with 6 guys backing up?

I think origin is definitely something to factor into betting but it's also a bit of a grey area as to how much effect it actually has. I will ponder how I could go about getting stats for these weeks...

[s2230011]

Its rare you will find a game not missing an origin player ... obviously each player is different to each team .. I just have a gut feeling on this trend, try and see, if you can come with something better go for it. But its a starting point then we can refine

[Duff85]

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!