Building a Winning MLB DFS Process
Part 4: Adjusting your projections
In this 5-part series, you'll learn everything you need to know to build a winning MLB DFS process.In this lesson, I'll show you how to add value by adjusting your projections.
Now let’s talk about what we want to do with that information that we just got from the test build. And ultimately our goal is to make sure that the assumptions that SaberSim uses, the blueprints that SaberSim uses to build that foundation of lineups for us is as accurate as possible. And so what we are looking for is just, are there any spots where it’s possible that SaberSim is off? Because again, we invest significantly in constantly improving our models, but no model is ever going to be perfect and there’s always going to be things that you miss. And that’s why it’s important to always look at other sources of data. And that test build process will help so that you can focus your limited amount of time on the outliers, on those spots that need the most focus, rather than just trying to look at everything.
So what you see here on the left is our game projections. And this shows you, the average runs scored by each team across all of our simulations and on the right you’ll see the player projections. And let’s start with the games though, this is something that other tools aren’t able to do because they don’t have that simulation data. What we can do though, is if you adjust the projected score of a team, we will then go through all of our simulations and pick a new set of those that match that total you just put in and then recalculate all of the players’ projected from within that game based on that new total. And so if you have a opinion on an overall game, this is going to be the best way to get those adjustments down to the individual player level. You don’t have to guess how Minnesota scoring more runs would impact all the other players in the game, including the hitters and pitchers and all of that. Change the score total, we’ll do that work for you.
And so this is where looking at Vegas can be really helpful, this applies to us as well as Vegas. You don’t want to put all of your faith in one system alone. Vegas is obviously very good at creating odds, but just like there are things we’re going to miss, there’s going to be things that they miss. And on average, we put our baseball numbers up against theirs any day. But again, especially in the start of the season, you want to be very cautious about going all in on one system. And so what can be helpful is just finding spots where we’re way off Vegas and seeing what you can do to get it back in line. And so this is where you can just open up a sports book yourself and just compare our totals to theirs. But if you have the sports betting add on SaberSim, it gets really easy.
So over here, what we’ll do is pull in the up to the minute odds and you can just see, okay, if a bet is highlighted in green, that means we think that bet is profitable. Meaning we disagree with Vegas and think there’s edge on this side of the bet, the larger the number, the bigger of an edge we think the bet has. Really anything over two is pretty big and something I would be cautious of. Whereas below one I wouldn’t be that concerned. And one is just give it a look, but Vegas can definitely be off. But when it gets to two and above, that’s where it’s obvious that someone, either us or them, is missing something. And those are the spots you really want to look for.
So what I want to look at is first that Dodgers Rockies game, because we were getting a ton of them. My guess was going to be that we were over, we were projecting more runs than Vegas was but weirdly we’re projecting fewer runs than Vegas, but looking at the totals. This is nine and a half, but everything else is like seven and a half, eight, eight and a half and then this game is 11. So this is the highest projected run game on the slate by a significant margin. Vegas actually thinks this going to be more runs than we do, but I’m not going to change the projection because I don’t think the projection is wrong. If anything is low. What will most likely happened in this case, I think they’re just priced too cheap on DraftKings. Or conversely, there are really good value pitchers, so you don’t have to spend as much in the pitchers and can spend more on the hitters. I think it is a pricing issue rather than a projection issue. So I’m leaving this alone because if anything, I would increase the projection, not lower it. And I’ll show how we can still make some exposure adjustments to keep things in line without completely changing the projections.
But I did also notice that we are strongly on the under for the first five bet for the Minnesota Milwaukee game, and even full game those are pretty big bets on those ones. So I don’t think we need to make massive changes there, but I am going to try to change that, to get it more in line with Vegas. So let me just show you how to do that. A common question we get, whether it’s for games or for players is how much should I adjust the projection? And there’s not a clear cut answer for this, but I just want to say don’t obsess over getting it perfect, you’re not going to get it perfect. What you’re trying to do is just go in the right direction. You’re not trying to make massive, massive changes because while it’s possible, neither Vegas nor SaberSim is going to be just way, way off where you need to double a teams runs or cut them in half even. You don’t want to making massive changes, you’re just trying to nudge it in the right direction.
So for this, all right, we were significantly on the under let’s add half a run to… That might even be a bit much, let’s just do half a run to Minnesota, they were the underdog and then do 0.7 runs to Milwaukee as the favorite. So I’m just going to give the favorite a little bit more of a boost. This isn’t a science, this is just again, pushing it in the right direction to try to get it more in line with Vegas. So then when I apply this, again, what we do is we go through all the simulation data and say, okay, if these are the average run totals, how does that impact all the players in that game? And so what you see is it lowers a projection of both pitchers because the hitters are getting more runs. And then we can look at Minnesota, it’s going to boost all of the hitters in that game, Milwaukee same thing. So that’s a very quick way to make really accurate adjustments in very little time, just by looking at some other sources of data.
But now let’s talk at the players and the kind of research that I’m going to do. So there were three that stood out at least quickly. It was Cedric Mullins, Austin Hays and David Dahl. Let’s bring up Cedric Mullins. So what I’m looking at here is the value in the order, we don’t have official lineups yet, but we’re projecting that Cedric Mullins will lead off. If that’s not the case, what we’re constantly doing is we’re pulling in new information and always updating your projections so that you have the most up-to-date information you can. And so if the weather changes, when they announced their empires who’s going to be pumping the games, but most importantly, when the lineups are confirmed, we’re going to run new simulations and update the projections. These are based on him batting top of the order. If that’s the case, he probably should get a bit of a boost, but this is still really high value. Let’s just look at the values for other players, low to mid twos is pretty normal. I guess those are pitchers but let’s go to batters. Yeah. So it’s high one’s low twos, is really what we’re looking at.
So for these ones that are high threes, again, it can be an issue of DraftKings or FanDuel just dramatically underpricing them, which does happen. Like maybe this is someone who doesn’t play that often, they put up the salaries early for opening day and then they’re going to play. And so like that is a spot where they are going to be a good value play. So if this sort of changes again, then the projections will automatically update. So I’m going to go into this with the assumption that he is leading off. So let’s just go to FanGraphs, which is a great free resource. What I’m really looking at here is seeing, okay how young of a player is this? How much experience do they have at the majors? How many games have they been playing? Because if they haven’t been playing much, it’s possible we just don’t have a big enough sample for our model to warrant extreme confidence in that individual player. The guy was been in the minors for a while, did play some of the majors in 2018, some 2019, but played more in 2020 for the partial season. So it does seem like he’s here, he played okay. So this person I’m going to be cautious on, but I still am going to lower it to get it closer just in line in terms of value.
So this is from hitters, this is a good way to determine how much to adjust to projection by, look at the value. So if this is someone who hasn’t had an insane amount of major league experience and truly does bat first at this price point at 2100, that’s a great value play. I think this might be a little bit high, so let’s give them a three times value. So to do that, you just take the how many thousands of dollars does it cost, the 2.1 times the value and then his projection should be 6.3.
Now let’s jump over to Austin Hays. So he’s six in the order, cheap, big value. Let’s check them out. Yeah. Similar story didn’t play as much in 2020, but similar performance. He’s not batting first. So this is somewhere where, again, I’m not trying to create entirely new projections for them, I let our team of data scientists do that. What I’m trying to do is massages these in the right direction, based on more qualitative analysis. And so this is someone who it’s clear that we like the Baltimore Boston matchup, it’s clear that these guys are cheaply priced. So I still think these are good plays, but probably not quite at that rate. And so we had given Cedric Mullins a bit of a relative boosting though we lowered his projection because he was leading off. But at six I don’t think we need to go that high, so let’s go a little bit lower, but we still want to say, okay I think SaberSim might be on to something with this matchup. So let’s go 2.75, which is still above most of the other players. So let’s call it 5.78.
And the last one is David Dahl. So he is projected to be near the top of the order so that is something to keep in mind. And again, all this is just trying to figure out what factors make this projection go up? It’s a cheap salary, near top of the order, good matchup, but it’s like are we over-projecting these players? So, I mean, this guy has been in the majors for a while. Games dropped off significantly from 19 to 20. And this is where like, I love baseball and I’ll follow the Sox and I’ll watch as much as I can, but I’m not following this where I’m paying attention to all the teams, all the rosters, all the injury histories, all of that. So nothing jumps out to me when I’m just like racking my brain for this player. What this looks like to me is I think he got injured.
So just did a quick Google, David Dahl injury. Yeah, in December is was recovering from surgery, his latest setback, his spleen removed, rib cage injury. Oft injured, David Dahl, he’s coming back from injury, it’s not as first one. Injuries are tough for models to account for and so this is one where I will lower. I mean, if he is batting second, again, this is with the assumption these batting orders are accurate because if they’re not the projections will automatically change. But if he is batting second, that’s the team saying we think he’s recovered enough at least to be near the top of the order. So I’ll lower him, but not go nuts, let’s go 2.5.
Sorry for the bad mental math and needing the calculator. So we’ve got those adjustments in there and now we’re good to go.