Sports Betting Model
In the 2000s and 2010s, it seemed like every high school or college guy who had ever listened to a rap song wanted to start making music once the internet afforded them easy platforms to do so. “Check out my mixtape” was a very common phrase during this era on social media. Now, in the 2020s with the rise of AI and the legalization of sports betting across most states, it seems everyone who has ever placed a wager is creating a sports betting model.
This is not a criticism of those people, because obviously, I’m one of them (though I started in the 2010s so I’d like to think I beat the majority of the crowd to the punch). Having had conversations with several of these people, and often providing advice and/or sharing my own early pitfalls in what is an ongoing evolution of the T Shoe Index, I thought writing an article covering some of the most common Q&As would be beneficial to the masses of people looking to build their own models to try to become a sharper bettor.
So, here are some common topics I have discussed with up-and-coming originators in the space and advice I have given them. Disclaimer: there is no right or wrong way to model. I have found several ways that “work”, some better than others, but there are almost infinite amount of ways you can make sports projections, so digest this article accordingly.
I want to build a sports betting model but I don’t know where to start:
When starting out, I’d recommend sticking to one sport to allow for more trial and error, and to not overwhelm yourself. Building and maintaining a betting model are not for the faint of heart, as it requires an immense amount of time and/or skills with automation. In my conversations and in my own experience, the automation skills are generally something people don’t already possess, so there’s a learning curve with that part as well.
Anyway, take the sport you decide on and begin to assess what stats you may want to incorporate. Spoiler: based on my testing over the years, the simpler the better here. I know it sounds sexier to use every advanced metric you can possibly corral on the internet, but I have not found that “sexy” = “more successful”.
I have my stats selected, but I have no clue how to turn that into a projection:
I think this is where a lot of people fall off the wagon. Remember, stats are great, but at the end of the day, we are trying to project POINTS (or goals, or runs, or whatever scoring metric is used in your sport), so your first real critical thinking step is “how do the stats you chose correlate into points”.
For example: if you’re doing football (college or NFL), choosing a stat like QBR might be great if you’re trying to determine who the best quarterback is, but not so great if you’re trying to project a game because there’s not really a quantifiable correlation to points scored. You might consider trying something like points per play or points per drive (hell, even points per game could be useful if can opponent-adjust effectively. More on that later.)
Now you have a correlated stat, so then it would just be a matter of projecting how many plays or possessions a team is expected to have against the upcoming opponent and you’re on your way to a full-blown game projection.
I turned my correlated stats into a points projection but it is nowhere near the market line:
I get this kind of question a lot, and 99% of the time the root cause is that you’ve taken the raw stats to make your projection, but any good model is going to opponent-adjust the data. In my opinion, this is the “secret sauce” of making a betting model. Most people with an iPhone calculator and Notes app can plug in some raw stats to make a quick projection; opponent-adjustment is what separates the TSI, FPI, SP+, FEI, and KFord Ratings of the world from your nephew who just turned legal age to bet and created a “model” using nothing but raw points per game.
Now, let me preface the rest of this by saying I don’t always care if my projection is off market on any given game. In fact, some of my more notable betting wins have come when the market strongly disagreed with my projection and even moved the opposite way. However, until you get your feet wet and are totally comfortable with your process, it can be a useful benchmark to let you know you’re on the right track. That being said, please don’t just: A. Bake the market number into your projection or B. Try to just replicate the market number. If you’re going to do those things, why even bother making a model in the first place?
The biggest piece of opponent-adjusting, in my opinion, is getting yourself organized for some iterative calculating, which requires excellent organization of your data. More advanced data scientists may scoff at the fact I operate out of Google Sheets, but it keeps me organized and I can integrate it with all the tools I need to keep TSI up and running at a high level.
How do I get organized enough to opponent-adjust my data?
This is where the road forks into about 10,000 different paths. There is no right or wrong way to organize, only what makes sense to YOU, so I won’t try to dictate to you how to do this (or give away my structure, necessarily).
I would say there are two primary ways I’ve experimented with over the years; one is to opponent-adjust every game once you’ve established your baseline for how good each team is (i.e. based on the raw performances). This method is preferred but more nuanced and requires much more time and organizational setup. The other method is to simply take a team’s raw average stats and adjust it for the overall schedule to date (until you figure out how to do this using your own rating system, leveraging a reputable SOS metric would suffice here).
The first method likely requires utilizing full sheets (on Excel or Google Sheets) to organize everything, while the second method could be accomplished with just a few columns on a sheet for simplicity. This is the perfect transition into one of our final topics…
How do I get all of the data I need for this without manually inputting it into my Excel/Sheets file?
I believe this to be the single most important aspect of maintaining a model. Putting the structure in place to make a single projection is relatively easy; having the structure, tools, and skills to maintain it is the single hardest part, no doubt about it. This is where a lot of the people I’ve talked to just can’t get over the hump early on, and something I managed to overcome after YEARS of manually-intensive work maintaining TSI.
Remember, I have no background in math or data science, I simply had a goal and determination to make it work. If you’re like me and have no prior experience using Python or R (look them up if you have no idea what I’m talking about) for data scraping, you could begin by utilizing more simple tools like IMPORTHTML formulas in Excel/Sheets to pull in data from your website of choice. These will get you started, but you’ll eventually run into problems once you accumulate a certain level of data and your file starts crashing because it can’t handle the constant pulling of that much data. Once that happens, you’ll likely have to end up in Python or R anyway. Again, I had never used these platforms a day in my life, but studied extensively using YouTube and ChatGPT to help me navigate and learn what was needed to execute the basic tasks I needed to seamlessly scrape and paste data without crashing my files.
I hope these tips were helpful as you embark on your model-making journey. These are things I wish someone would have shared with me circa 2017 when I was hand-writing projections on actual loose leaf paper. Although I’ve given you these tips, remember, there is no “easy” button for creating and (most importantly) maintaining a sports projection model.
Another nickel’s worth of advice based on an observation I’ve made in the betting space recently: I would not attempt to monetize your model until you’ve accumulated a solid enough following that you can earn difference-making money (whatever that means to you). Early on, building a following is your currency. Provide great free value, build a following, and then you can try to cash in on your hard work. I began making projections in 2017 and didn’t make a single penny (outside of my actual bets) until the 2022 college football season. It’s a marathon, not a sprint. Don’t sell your work now for what is likely pennies on the dollar compared to what it will turn into if you stick with it and build your follower base.
The post Building a Sports Betting Model: Tips, Tricks and Advice appeared first on VSiN.
