How do our predictive models work?
Each of the Stats Insider models uses a fairly similar approach to simulating a sporting event. They vary in complexity and approach: our NRL & AFL models are more sophisticated than our AFLW ones, because there is more information available. Our Cricket model differs from our ‘ball-sport’ ones, because Cricket is effectively a series of individual matchups between a bowler and a batsman (with some fielding quality impact) where as ball sports are impacted by the ability and positioning of multiple players at a time.
In essence though, we seek to predict the distribution of a players performance (e.g. how often a player will score 0, 1 or 2 tries, how often they’ll have 10, 20 or 30 disposals). Once we have those distributions, we use a Monte Carlo approach to predicting the outcome of the match; that is we take those player distributions and simulate the game 10,000 times, so as to understand how likely a team is to win, how likely there will be 40 points scored, or how likely a player is to score the first goal.
Once we’ve run those simulations, then for the punters and tippers amongst you we can compare the results to common betting markets, and the odds on offer. If a team wins 6,520 of our 10,000 simulations, then we predict them to be a 65.2% chance of winning the match. If you’re able to bet them at $2 at your local TAB or friendly online bookmaker, we might highlight them as an investment worth considering further.
How accurate are they?
If we’re predicting a team has a 55% chance of winning, or even an 80% chance of winning, that’s not saying they are going to win, or even we think they’re going to win. We’re saying we think if they play this match 100 times, they will win 55/80 of them. That’s a crucial difference, and underlies everything we publish on Stats Insider: we provide data, not tips.
We won’t publish anything on this site that we do not believe to be profitable against available odds, and similarly everything we publish we expect to be within a reasonable margin of error in predicting match or season outcomes. That said, the profitability of the models will vary from sport to sport and market to market, and we would expect to lose 45-46% of our line or total bets, and to have losing periods – and even seasons – across all sports.
Why do they change?
If you came to the site on Monday and we highlighted the Broncos as an investment worth considering, but by Saturday we’re suggesting you consider opposing them, don’t be surprised. Our models gain information all the time: on Tuesday when (NRL) team lists are announced, on Thursday when it’s announced the hooker has picked up an injury, on Friday when one of the big betting Syndicates hammers the price, and on Saturday morning when the BOM issue a weather alert for a severe thunderstorm. All of those things impact the possible outcomes of the match, and so all of them impact our models. Our suggested investments may change frequently right up until game time.
How do your in-game simulations work?
They’re very similar. 10,000 simulations take a fair bit of time – and computing power – to run, and we can’t do that every minute during the game. Depending on the sport, we may run 500 or 1000 simulations of the rest of the game. But the approach is the same: if we played the rest of this game numerous times, how many times does each team win?
How about the projected ladder?
Our seasonal projections take two primary inputs: the individual match simulations (that is, we simulate each match remaining in the season 10,000 times), and the likelihood that a team will be impacted by injury to one or more of their key players. If a player has historically missed 25% of the season through injury, or if they’re in doubt for the coming week, our season simulations will capture this and force the team to ‘play’ some of those games without that player. Our seasonal simulations are more complex, and take longer to update, so you may notice these lag a day or a few days behind the individual match projections.
What should I do with the information?
Well, that’s up to you. We bet many of our suggested investments each week. We enter them in tipping contests. We use the information to inform our fantasy decisions and pick our daily fantasy teams.
But, most of the time, we just use them to watch the games from a different angle. At kickoff, the Tigers don’t have a 50% chance of beating the Lions, it might be closer to 70%. When a referee makes a controversial decision and disallows the try, we can see that 15% increase in win probability come right off the board. It makes you appreciate every play that happens. It makes sport more fun.