The Risk Takers Podcast
Willing to bet this is the sharpest podcast in sports betting
https://www.goldenpants.com/
The Risk Takers Podcast
Talking to Models (in Sports Betting) | Ep 27
Today I try to give you a 20-minute primer on how to not epically screw up your model
I discuss getting crushed by sharp books, where I think you should plan your first attack on the sports book, and of course...me pumping Billy Walter's tires more.
Welcome to The Risk Takers Podcast, hosted by professional sports bettor John Shilling (GoldenPants13) and SportsProjections. This podcast is the best betting education available - PERIOD. And it's free - please share and subscribe if you like it.
My website: https://www.goldenpants.com/
Follow SportsProjections on Twitter: https://x.com/Sports__Proj
Want to work with my betting group?: john@goldenpants.com
Want 100s of +EV picks a day?: https://www.goldenpants.com/gp-picks
But even in that realm of imperfection, you want to set yourself up where you can be imperfect within a game of imperfection and still win. Hey, what's up everybody? Gp13 here Today I want to talk about modeling. Now I'm going to do this. It's hard to explain modeling in any extremely helpful way over a podcast, but I get a lot of questions about it and I want to do what I could to, without actually going into the math behind making a model or walking you through building the code or building an Excel sheet that would spit out something useful, just how I would approach building your first model, or not even that, how I would approach modeling, some common pitfalls and hopefully that'll give you a place to start, where then you can go into the weeds on your own, watch some videos out there or whatnot. That will be what you need to actually press your model forward, use it to make money. But I just wanted to talk about the mindset around it and help you avoid some common pitfalls that can come when you start modeling, because I'll tell you a story about my first model. So my first model was well, I built a golf model a while ago this was probably 2019. And for a while it took a little bit to get to the point where it's spitting out something that was resembling the market and after a couple of tweaks and whatnot, I thought it was ready to bet. So we put it to use in soft books at Bovada and we did really really well against Bovada just using this model, using no market input, which is a mistake. But I was stupid and young and I didn't care about the market. I thought my model was be all end all. That's mistake number one, but anyway. So we take the model, we take it to some sharper books and it's usually showing no value. No value, but sometimes it would show massive value and so we would pile money in and, lo and behold, it was basically just showing value against situations that didn't understand the full amount of information that the Sharp Books app used. They had my model plus something and I ended up losing a fair chunk of money with that and it was really a hindrance in my sports betting career. So I want you to avoid the mistakes I made that first go around. And recently I've been updating our model. I'm not going to talk really in depth at all about those updates, because this is really kind of the one area where I don't want to give away any edge, but I will talk to you in general about what you should be looking out for when you're you're building your models. Okay, so, first thing. First thing is you have to pick the right thing to model. You have to pick the right thing to model. Well, what's the right thing? I've talked about this a little bit and basically you want to pick something. You want to build a model somewhere where you're going to have the easiest time winning with an imperfect model, because, first of all, no model is perfect. Right, our model, every model, is just the best guess, it's just the best attempt at quantifying the correct percentage, right? But even in that realm of imperfection, you want to set yourself up where you can be imperfect within a game of imperfection. And so what Now?
Speaker 1:Where does this not work? Nfl sides If you be hard pressed, tell me that you've built a model trained on widely available data, last 20 games from two teams, or like all the games that were played in the NFL, and the model was some kind of machine learning model that stacked up, you know that, graded each matchup or whatever, and did some kind of relative strength and that was gonna spit out a winning NFL model. It won't. It won't because that's just like a. That's where all the smartest people are and all the most the biggest money is. So the people who are doing that NFL modeling have that basic relative strength, functionality in their model, plus subjective research, plus injury news, plus everything like coaching tendencies, everything built in right. So you're you're going into a market where the sports books lines are being set by these incredibly complex, sharp people, tons of money flowing in. Don't try and model that right. That's like. That's the classic, like learn to walk before you can run. Don't model NFL sides, just forget about it right. Don't model NBA money lines. Don't model, honestly, any of that.
Speaker 1:And for the people who are out there thinking about the DFS the DFS pick them offerings. Now I I said this on earlier podcast and it's worth repeating because we had extremely good success modeling fairways and winning at price picks and it's because there is no other market besides price picks that showed fairways. So what we had to do was basically just build a better model than price picks out, plus we get to be selective about it. And then when we bet into them, they have nothing to lean back on. They can't go look at Fandall for fairways. They couldn't, they still can't. They can't go look at underdog for fairways. They're flying blind right. So they don't know. If they don't understand the stat as well as we do, if they don't understand the projection as well as we do, there's nothing they can do to defend against us. Those are the markets you want to be it.
Speaker 1:Find something that's not going to be a massive focus of the platform of the book. That's worth your time, right. So make sure that you're gonna be able to bet enough money that it's worth your hourly to start building the model. But, that being said, even if you build a model that like breaks, even it's, that's such a massive win for you because then you get to pass. You know, go past that. That's step one. You know your first model probably not gonna be great, probably not gonna be great, but I'm trying to make it so it like could make you money, right.
Speaker 1:So first is, find a niche sports, preferably something that you know about so that when you find something, the data that might be predictive, you can just sanity check it against your knowledge of the sport. Right, try and have some domain knowledge, but try and make sure that there's not a big market of betting information that's backing up the sports books numbers. Make sure that that sports book, if possible, is just out there on their own with their model that they don't really care too much about, like prize picks, doesn't care too much about golf fairways, right? So us caring about it a little and probably having a better understanding about how golf prediction embedding works, we were just able to, you know, absolutely rip them apart. That's where you want to be and they don't know what to do. So they actually ended up taking fairways down multiple times because they just couldn't deal, right. So those are good spots.
Speaker 1:So, number one, make sure you're pointing your model at the right thing. Don't start out with a big market. I would probably not try and model picture strikeouts. That being said, I could be convinced I'm wrong. There's a lot of interesting matchup data in baseball that you could probably use and line up. If you're doing something with substitution, like line up injury news or whatever for picture strikeouts, that could be okay, but I would try and find something that I think is kind of interesting this season. Try and model hits in hockey. That's not a huge market. There's not a lot of data out there for prospects to fall back on. So if you had a hockey model that can predict hits relatively well, that could be a good spot.
Speaker 1:Number two is you're going to have to basically quantify inputs into the same output. That sounds super obvious. What I mean is, let's say you're trying to do something like project player points in the NBA. You're going to have to whatever goes into the model, whatever you think are the important factors, one super obvious one is going to be minutes played. You would have to know one minute played equals x points. But another thing could be home or away. Home equals y and points played. So it would be like is player home, is player away? Yes, no, yes, equals something in points. If you were trying to predict assists, yes, would equal something in assists. So every variable you're quantifying in terms of the output, football let's say it's windy, say it's cold. Let's say you have a team this is something from Billy Walters book, right? So I'm not going to this isn't me, but Billy was talking about, and actually this is a great chapter.
Speaker 1:If you want to learn about essentially like waiting things in terms of the points read, which is what Billy did then you have to take all the different variables that affect the game or affect whatever you're trying to predict, and have them return a number in terms of a point spread. So when I say that, billy would say a team coming off a 20 point loss or something was given an extra like four points, something like that, and I'm going to quote me, it's probably not close to that, but every factor was like a cold weather team traveling to warm weather would be given, you know, x adjustment. A warm weather team traveling to a cold weather team and it's actually cold would be given y adjustment. And those adjustments were always in terms of the points read plus one points, minus one points. Plus point two points, minus point two points, right.
Speaker 1:And whatever you're trying to predict, you have to have the variables spit out a number that is expressed in terms of what you're predicting. How much does LeBron James being out, or in effect this is where everyone's going to make fun of me Austin Reeves assists or something like that, right. And if you're a teammate being in or out, that's a, that's a box, yes, no, and that yes and that no are going to have point values or assist values or whatever. So you have to make sure that everything that you're using is going to the same type of output, right? You don't. You know you. You remember you're trying to predict one thing at first you know the model that you're going to be building. You want you to be focused with it because it's hard, right? So focus on one category and try and express all the inputs in terms of that category.
Speaker 1:And this is what got me, and I alluded to this earlier but this is that you have to. You have to, you have to respect the market, no matter how good you are modeling. Every single sports betting modeler out there I know who wins a lot of money takes the number that they get that their, their model spits out. So, let's say, lebron James 28 points. And if the rest of the market, if there's a big market out there and whoever they've decided that they think is good, a good indicator, who is sharp, who's a sharp book, they're gonna factor their price into that 28. So let's say they're at 28, they see pinnacle circa at 24 and a half, and then they see fan duel at 30. Now, this is not this. In today's day and age, this would be a very rare situation, but they would actually probably think, wow, this isn't even better bet than the 28. They're gonna regress that 28 towards the sharper books that you know how you want to do that, that's up to you and that's.
Speaker 1:There's no actual perfect way to do this because the way the books model and price things changes so you can even retroactively rank the book sharpness based on you know previous success, but it's very rarely. It's very rarely actionable in the future. It's just good to keep your ears to the ground, to keep a lookout to see who's following who, understanding where the books rank in the sharpness rankings for what sport? It's very much a being plugged in and paying attention type of it's exercise. It's not really a mathematical exercise because every it's changes. It changes some, but some. What if you know what? If circa has a great golf trader but then MGM decides they're getting killed too hard at golf and they get the circle golf trader and the GM is gonna be better at golf and this happens. So don't get too locked in. Don't just see pinnacle and think sharp. Just see bet Chris, think sharp, right, be plugged in.
Speaker 1:That being said, you have to be plugged in so you can understand which books you think that you respect and use their information to kind of mesh with your number, mesh with your model, to create a synthetic number that's a combination of your number and the sharpest books out there, and then you take that new number and you can go bet that at the most off-market spots you can find. Let's say we have the same example. Right, we have. We have Chris, whatever I said circa 24 and a half from LeBron pinnacle. You know 25 and a half. You have 28, and then you see DraftKings sitting there at 21 and you see better is at 22. You're gonna, you're gonna definitely take this 28, go bet it at DraftKings, but you're not gonna bet it like 28. It's a true number. You're still gonna take that 28, combine it with the 24 and a half 25 from the sharp books. However, you're gonna wait everything that's gonna be changing and that's gonna be up to you. You're gonna spit out a number that's gonna look somewhat like 26 and then you're gonna use that to go bet at DraftKings and you could even use that to go bet at that rivers, because you're not gonna get enough down, but you can be pretty confident that it's a good bet at 21 and 22.
Speaker 1:Right, the main thing here is respect the market. This is where I went wrong. Right, I thought that I was gonna be betting into these sharp books with a pretty basic model. Right, to go after the the really the toughest books. The people who work there they're great sports, better stew, right, like I think they could. At some of these books they could easily flip sides of the counter and they could make it. That's not the case at I don't believe that to be the case. At DraftKings, right, but I'm jam. But at Betchris, I believe that to be the case. Circa, you know, it seems like that. That's the case. So you have to. You have to have something different to go after those books, right, it's not gonna be like a basic model.
Speaker 1:So, first things first. What I would do is take a number, use the sharp book information and bet against the soft books. When you start to develop a better sense of the market or whatever it is, add some features to your model. Feel free to bet Openers at the sharp book, see if the line is moving your favor, build some confidence and then go from there. But don't blindly bet into smart people with your model Ever. If you're getting resistance, if you're getting information, no ego, use it, decide. Do I still have the best of it, maybe really back a little bit. Don't go broke. Always be willing to admit you're wrong, and you won't take too much damage, just when you can't admit you're wrong and you're betting into super smart people where you can get really hurt here. So that's of the utmost importance. So, anyway, this was just a very basic overview of what you need to first think about before you even open Excel, before you learn Python. The three kind of three boxes I would be checking your mind when it comes to your model is one are you picking the right stat A, and by picking the right stat, is this something that you're going to be able to realistically get an edge on through modeling against the sportsbooks? Two, are you getting all of your variables to express in terms of the thing that you're predicting? And then three, respect the market. Make sure that you're using sharp books, information and combining them with the line that you spit out to get some kind of composite number that you can then take to the most off-market places. That's 101, really, just sports betting, but anyway, I hope that was interesting.
Speaker 1:Again, podcast is not the real modeling format, but I've been tweaking our model so I had a lot of this on my mind and I get asked a lot about modeling and I can't really there's no good way really for me to, unless I was to maybe make a video or something to actually kind of take you through some of the steps. But there's a ton of great resources out there. Actually, I think Unabated has some good modeling videos. Correct me if I'm wrong, but I know there's some YouTube modeling videos that could be pretty good. So keep these in mind, go find some good secondary content that will go more into depth, and I'm always around DMs are open if you have any questions. So thanks everybody for listening and I'll see you on the next episode.