Many professional sports gamblers avoid hockey because it can be a notoriously difficult to forecast winners and losers reliably. There is too much randomness and hot goalies frequently swing games. I’ve had some success using 2019/20 goal differential data to build a betting algorithm based on that season’s games, but the success was not transferable to the unique 2021 scheduling parameters. I’m not precisely sure why, but did write a blog post describing the paradox. Other traditional hockey betting strategies like “zig zagging” has shown some promise (read more here) but the profits have fluctuated.
Another popular NHL betting strategy is putting your money on hot goalies. Since the position has such a dramatic impact on the outcome of games, why not build a betting algorithm based on goaltending? I decided to take an hour to download all goalie game logs from 2019/20 to calculate the cumulative save percentage of every starting goalie in every NHL game, then copied it into my database of games and betting lines.
My preference when embarking on statistical modelling is to start with the simplest approach. My initial hypothesis in this case was; what happens when I simply bet on the team with the better starting goalie? There were 1193 games in the shortened 2019/20 schedule. I took only those where each starting goalie had played at least 5 games, then calculated the difference in starter SV%. There were 867 of those games with at least a 0.001 difference in SV%. My bet size ranged from $100 to a maximum of $500 for a 0.05 difference. Of those 867 bets, 536 were winners (62%). I bet $194,874 and won $237,638 over a very large sample of games.
None of my previous modelling attempts have come close to this level of success, not even half. My algorithm based on goal differential was able to win roughly $1.10 for each $1 bet. The “bet the best goalie” strategy won $1.22 for each dollar bet, which is remarkably good. The algorithm doesn’t care how much value you might be getting on the odds offering, or if the implied probability matches the actual probability of victory. All it does is bet on the best goalie, and it made a substantial profit.
Often times in statistical modelling, the simplest approach turns out to be the best. Adding a whole bunch of variables into complex equations rarely leads to better accuracy. My next project will likely be to start copying and pasting 2021 goalie game logs into my spreadsheet and seeing if the algorithm which had substantial success using 2019/20 data is transferable to the pandemic schedule. That’s where this all fell apart in my goal differential algorithm. What worked in a normal season was not transferable to 2021. Hopefully, “bet the best goalie” strategy will produce positive results going forward.
One logistical problem is that in many cases, the official starting goalie is not known until closer to puck drop. The betting lines that I’m using in my model were often recorded before the official starting goalie was known. Generally, when it’s announced that an inferior back-up goalie will be starting a game, the line moves after that information is released. Betting against back-ups is easier said than done. Even betting on the better goalie may require you to wait until pre-game warm-ups to log your wager. You can go to daily faceoff for a projection of who is going to start the game, but even they aren’t 100% reliable.
And remember folks, always bet responsibly and never
wager with money that you can’t afford to lose. We don’t have legal single game
sports betting in Canada, so I make all my bets in a spreadsheet with fake currency.
Also, see my February NHL Gambling Report for more observations on hockey betting in 2021.
No comments:
Post a Comment