Daniel is one half of the duo man behind our extremely successful AFL betting tips package.
With the new season finally here, we get his insights on their results last year and any further improvements they’ve made for 2017.
- The enormous impact of rookies and injuries on a team’s fortunes
- How a team’s age profile can help predict their improvement
- Adjusting the model to include team defense metrics
Get More Betting 360
Make sure you don’t miss our punting tips to come! Subscribe on:
What Do You Think of the Show?
Mark: My guest on today’s podcast is Daniel who’s one half of our AFL tipping team. Thanks for joining us Daniel and can you start off by giving everyone a bit of a background of your own history and your partner in crime Steven as well.
Dan: Steven and I met at university back when we were 18, so about eight years ago now. We were both studying actuarial studies and finance. So we were both kind of interested in sport in terms of just an interest stand point and like most young Aussie blokes we enjoyed a recreational punt throughout university.
It was more towards the end of university where we became more interested in the betting side of things a bit more seriously. We both got a job just as casual employees at a bookmaker and so that started giving us an insight in to the betting industry and how it all works behind the scenes.
While I was working there we started playing around with just scraping AFL data and trying to model games based on historic data. So that was where it all started, from there we graduated university and both spent about two years working in finance and during that time we were developing the betting side of things, just in our spare time after work. We kind of got more and more detailed and confident at it and at the end of 2014 we both decided to give the betting a crack full time.
So yeah we’ve been doing the betting thing since then so this will be our third year on the full time betting. AFL is definitely our main sport so it’s over the course of five years we’ve developed a model that started out as a primitive team based statistics model ’till today we’ve got it down to a pretty fine art where it’s quite a complex model that’s largely based on player ratings.
Mark: So you’re just saying that you’ve refined your model and it’s now player based, do you maybe want to explain a little bit about how that works?
Dan: Yeah sure. So basically our pricing each week is contingent on the teams that are named so we actually only price up our games on Thursday night when the teams get released on AFL.com. So it’s based on who takes the park for each team so our opinions on a team can change a lot from week to week with injuries, changes in player form. Then year to year we can have pretty drastic opinion changes on teams with trades, players retiring, players coming back from a year away with injury, stuff like that.
So obviously this year with all the Essendon players coming back, our model will have a pretty significant upgrade on the Dons. So in terms of how we rate players, we won’t really give too much away obviously but we rate every player in the competition including rookies. It’s all done mathematically based on statistics so it’s not really opinion based it’s all backed up by data. Coming from the actuarial background it’s kind of in our nature.
We look at the significant statistics for players in different positions. So it’s not as simple as having one equation to rate a player for the competition. We actually have six different player rating formulas for each of the main positions being Key Defenders, General Defenders, Midfields, Rucks, Key Forwards and General Forwards. So that’s the high level of how it works.
Mark: And you just mentioned rookies who haven’t played senior footy before. How do you go about giving them a rating before their careers start?
Dan: So initially when we developed our player ratings a few years back, we just gave rookies an entry level rating and then changed that rating as they played games. We refined things at the start of last year a lot more so we did a big analysis on rookies from the past six or seven years and we looked at how they actually performed and then we looked at their draft pick number, their age, and we also looked at how Champion Data rated them, because the prospectus Champion Data releases each year gives a write-up of each player and they actually give some pretty good insight into rookies. Whether it be some commentary on their junior form or how they played in the SANFL or the WAFL.
So we use a combination of those things to form our initial opinion on rookies. We can have quite significant differences in two different rookies for example nowadays that we go into that level of detail.
Then obviously once rookies actually get a game we then price them, well we then look at, you know, their actual rating based on their data from the game and we’ll update our pre-season opinion on them pretty aggressively towards their actual game data. So basically if a rookie has played one game that’ll give us more information about him than say one additional game for a player who’s already played a hundred games.
So last year, that was the first time we’d actually done that big detailed rookie analysis and it seemed to work really well at the start of the season. We got off to a bit of a flyer, I think it was like seven wins and one loss on line betting for the first two weeks, so obviously it’s the early rounds where you do have a fair few rookies getting a game. There’s a lot of kind of unknowns about those players so we think our approach handles it probably a little better than the bookmakers, so we find a fair bit of value in those early weeks.
Mark: Looking at the other end of the spectrum, for experienced players who might have played 50, 100, 200 games even, do their ratings keep developing throughout the season as well?
Dan: Yeah definitely. A players rating will change every time they play a game. The speed at which it changes will differ depending on what level of their career they’re at. So for established players who have played one or two or three hundred games we’re not going to adjust their rating heavily based on one additional game kind of data coming in. We’re still going to put some pretty significant weight to how they’ve played notably in the past two years is where we mainly look.
And then on the other end of the spectrum, if you have young player who’s played, say 10 games last year, and then he starts playing this year and he’s playing at a level much higher than the prior year, we’ll be updating his rating much more aggressively. Because you see with young players they can really develop year to year quite rapidly in their first five seasons generally, and that’s another thing we do with the younger players we actually predict expected improvement based on another historical data analysis we did. The data showed that first year players generally improved in their second year, and improve again in their third year up until around their fifth year. So we make an allowance for that so young teams with a lot of players returning for their third and fourth seasons we’ll actually anticipate that team to have some improvement at the start of the year. On the other side of things, if a team has a lot of older players we generally won’t expect them to improve as much as teams with those younger players who have more upside to potential improvement with their rating.
So basically using that player approach for the first time last year we saw great results with the line betting. We went at 61.8%, so 55 wins 34 losses. We also managed to find some pretty good value betting on some big underdogs head-to-head through that player based approach when the strong favourite might’ve been missing a fair few players but the market probably didn’t catch it as well as it should’ve. So we’ll be kind of following that same approach this year.
Mark: Well anybody who bets on any sport with line betting will tell you that’ going at 60% across a season is a really great achievement. Can you just give everyone a bit of context from 2016, maybe a team or two you guys did particularly well on against the market?
Dan: Yeah so in terms of backing teams Gold Coast is a team we did pretty well on. We had six wins and two losses backing them at the line often as a decent sized underdog obviously because they weren’t starting favourite in too many games. So we made 18 and a half units profit betting on them. They were a team with quite a lot of injuries but they did have injured players coming in and out of the team throughout the year so the model was kind of picking and choosing weeks to back them when they had a relatively strong line-up in. So there was a few situations where you know, they may have lost five games on the trot but we were actually quite bullish on their prospects for the coming week due to having some good players coming back. So we found some good value on them.
On the flip side, a team we struggled backing were Brisbane, so we had one win on them and seven losses when betting on Brisbane. So that was our worst result by far, we lost 23.5 units on Brisbane. So they were one of three teams that we actually lost betting on, so the other 15 teams we made a profit betting on so that was pretty good to only have three teams that we had a negative return on. And then for a bit of a different look at it if we talk about laying teams, so betting against them, Fremantle was an interesting one. Obviously in 2015 they were minor premier and made a prelim final. Last year straight out of the gate we were pretty bearish on Fremantle so we had a big bet against them round one, against the Bulldogs, and we pretty much backed them most weeks, so we bet against them 15 times for 10 wins and five losses, at an overall profit of 21.6 units.
Obviously Fremantle had a lot of injuries and our model really picked up on that and even though the market knew Fyfe and a fair few other guys were out we still seemed to rate them lower than the market week in week out.
Then again on the flip side a team we bet against without success was Geelong, so we bet against them seven times for just one win, six losses and we lost about 18.5 units on them. I’ve kind of touched on Geelong, and laying Geelong and backing Brisbane and that kind of prompted some off-season research which we’ll talk about shortly. I won’t say any more on that for the moment, but that’s a bit of context on some teams we went well on a didn’t go so well on last year.
Mark: So with your success last year, I guess as you’ve touched on with your new rookie ratings, and used your new player based model for the first time, have you made any further developments in the off season this year for season 2017?
Dan: So obviously we haven’t wanted to change too much given things went so well last year. But the main thing we’ve been working on is obviously we pore through all the teams lists and in particular those rookies, and make sure we’re comfortable with the rating of every player as at the start of the season. So we’ve been doing that for the past month or two and then we’ve also been keeping an eye on rookie form in JLT series just to kind of get comfortable with our rookie ratings.
But the one big thing we did look in to was how team defence could be incorporated into our model. So obviously player ratings, they’re a great way to measure the skill of each player but they are largely attacking measures, most metrics are offensive in that you have the ball in hand, kicking goals, winning contested possessions, disposals, clearances, they’re all measures you accumulate with the football. Whereas there’s only a few measures that are defensive such as tackles and one percenters.
So what we wanted to look at was – are there any defensive variables we could measure at a team level that could improve our model? So we looked at things for the past three years of data and we did manage to find a defensive variable that came out statistically significant in our analysis. And basically by improving that variable in our model, the model’s error was reduced. The model error is basically looking at how different our prediction is to the actual game result. The better your model, the lower on average that error will be and by including this defensive variable we found to be significant we were able to reduce our model error. We calibrated that variable on 2014 and 2015 data and then we ran an out-of-sample test on 2016 data and it did improve our model’s results on the betting side of things particularly.
We weren’t finding value on Brisbane as much because their defence was so bad that the new variable really punished them. On the flip side the Cats had really good defence last year so they were getting a bit of an upgrade with the inclusion of that variable and we weren’t betting against them as much. Other teams also were changing around a bit with the variable but Brisbane and Geelong were at opposite ends of the spectrum and as a whole, the model error went down which is the main thing. The out-of-sample back tests in terms of betting results improved as well so that was the main inclusion to our model that which we’ll be using in 2017.
Mark: Great stuff, so it sounds like the model is in good shape. Just the service that you’re going to be running this year, do you just want to touch on maybe what the guys who are subscribing can expect? So when the bets will be coming out, the bookmakers that you use etc.
Dan: We basically based the service around Pinnacle just because it’s the only book where you have guaranteed limits for all punters, regardless of whether you’re a winner or not. And we do that just because it’s no good quoting corporate bookmakers where no one can really get a bet on if they’ve got their account limited like I’m sure many of the listeners have.
So we always quote Pinnacle. Last year we did start out quoting our initial line bets on Thursday night, this is when Pinnacle had $6000 limits, and then we’d also release any later plays on game day when Pinnacle had $27,000 limits which is obviously huge and means all the punters can get a big bet on whether you know they’re a smaller or a bigger subscriber.
We did see Pinnacle actually reduce all their limits on Australian sports by about 60% though, around round eight or nine last year. So for this reason we didn’t release on Thursday any more just because the market didn’t have big enough limits, and we changed to a game day release. So generally we released around 11:00 a.m. or midday at the latest, and we did this when Pinnacle limits hit their maximum, which was $11,000 for line betting and $5,000 for head-to-head betting. So the downside of this betting on game day is you might miss out on a bit of value from the markets moving earlier in the week, but at the end of the day we want to operate a service where smaller recreational punters and bigger professional punters can follow along and have a good bet if they want to.
For someone who follows with a $5000 bankroll or a $100,000 bankroll, we’re going to be quoting when Pinnacle is at maximum limits of 11K. So that means punters should be able to get their desired bet on. Obviously there will be a bit of competition at Pinnacle probably from our different subscribers to get the exact same price we quote, but last year we always watched the market when we send our tips out and we’ll generally see Pinnacle trim maybe five cents in that first few minutes of release and they might move it one point. But we didn’t see Pinnacle going nuts moving four points at our release or anything which was a positive sign which means most people would have been able to match what we were quoting and punters who can bet at corporate bookmakers in a lot of cases could have probably got a better price than what we were quoting. So we’ll be sticking to that this year with game day release and hopefully that means it’ll give everyone the best chance of getting their desired bet size on at the right price.
Mark: Yeah it’s unfortunate these days with all the bookies, it’s almost half the battle just trying to get a bet on at the price you’re after so I think as you said that approach worked pretty well last year and we’ll continue to monitor and hopefully it does the same this year. And just with your other main bet type that you guys do is the total points under-over line for a match. Do you just want to take us through how that kind of works, and how that went last year?
Dan: Yeah sure, with the total over-under points we’ve always done pretty well on these around the 58 to 60% mark. Last year we went 23 wins 15 losses, so that’s on that 60% mark. We generally have a smaller volume on total points bets than line bets and that’s purely because the market is a lot less liquid, so the limits on totals for Pinnacle are only about 10% of the line betting limit.
Basically on game day Pinnacle used to be betting to win about $2500, but they’ve dropped that back to around $1,000 which happened last year along with the decrease on the line bets. So what that means is if someone say had two max bets of $1000 they could be moving the total 20 cents or two or three points. It’s just a much less liquid market, a bit more of a niche. So by game day often a lot of the value has been kind of sucked out of the market with people betting earlier at the low limits. That results in a lower volume so this year we’ll be doing our best on the totals to find any value on game day, we’ll definitely need to wait until those Pinnacle limits hit their maximum though to make sure at least a few punters can hopefully get on to Pinnacle and at the corporates.
So generally we might find one or two a week, I would kind of approximate. So maybe around that 40 bet mark on the totals whereas the line betting will be closer to 100 if the past years are anything to go by.
Mark: Good stuff. Well we might leave it there for today Daniel. As we’ve seen last year was a really great year for the AFL model and we’re looking forward to another one this year. So we’ve got a lot of guys on board and still room for a few more so, if you just want to head over to the website and we’ll link to your results from last year and your season review. We’re looking forward to another big season.
Dan: Yep we’re definitely ready to go. Been putting in plenty of hours the past few months on it so can’t wait ’till things kick off.