This is the first post in a new series of blogs about betting and statistical modelling in golf. These are going to be shorter posts that dig into the details of how we think about modelling golf and how to find value in betting markets. Each week we will point out specific bets where we think our model has an edge. We are both inexperienced gamblers, with this being the first year we are taking the betting process seriously. This blog will document a learning experience for us, and hopefully its readers will learn something of value too.
For those that have followed our website closely in the past you will know that we did in fact bet with our model over the past two years. We mainly bet on top 20s, for no reason other than that those were the easiest given what our model outputs each week. (You can see our 2018 results here
, and our 2017 results here
.) Despite the vast difference in performance, the model methodology was largely identical in these two seasons (it improved, in our opinion, in the 2018 season). Further, our model appears to do a pretty good job when you evaluate every prediction it makes (model evaluation
) (as opposed to just the select few we bet on). That last graph is perhaps not so helpful for explaining away our 2018 betting results as just bad luck because first, there is no benchmark to compare it to (i.e. what would a “bad” model look like on this graph?), and second, maybe the bookmakers we are betting against are using a model that would look just as good in the plot.
What is undoubtedly true is that betting on top 20 finishes in golf has an incredible variance in possible outcomes, and I don’t think the number of bets we made in either season (268 in 2017 and 508 in 2018) were close to large enough to come to any strong conclusions on the model’s edge (or lack thereof). To get a sense of just how random sports betting can be, I’d recommend checking out some of Christian Baier’s
plots of betting performance over time for a few of his betting projects.
Anyways, enough excuses about our abysmal 2018 betting performance, on to 2019! This year our goal is to place at least a few thousand bets across several different markets in golf. This will include top 5s, top20s, and several variations on tournament matchups and round matchups. The latter two bet types are much lower variance and also have much less vigorish applied by the bookmakers, and we are excited about (hopefully) beating those markets.
Our model has also gotten a bit of a makeover for 2019. You can read a detailed description of the updated methodology here
, but very briefly the two biggest changes are 1) the model is a bit more responsive to short-term movements in form than it was in the past, and 2) we are now incorporating the strokes-gained categories, making use of the fact that ball-striking performance is more predictive of future performance than short-game or putting performance.
This series of blogs will serve as a place where we can go into detail on specific questions or ideas behind our statistical model, and also explore why the model finds value against bookmakers where it does. That’s all for now. We hope this blog provides an interesting read for 2019!