[00:00] My model makes this game minus four. My

[00:02] model loves this side. My model has this

[00:06] total three points higher. You hear that

[00:08] stuff everywhere now. It's in every NFL

[00:11] video, every Tik Tok, every Twitter

[00:14] thread. Everyone suddenly has a model.

[00:16] The guy betting $10 parlays has a model.

[00:19] This dude recording pics in his car, he

[00:22] has a model. Content creator model.

[00:24] Discord tout definitely a model. And

[00:27] look, I'm not anti-model. I use models.

[00:30] I like models. But at some point, you

[00:32] start asking yourself, do these models

[00:35] actually mean anything? What is a model

[00:38] really? And when does my model just turn

[00:41] into this magic word people throw out

[00:44] there so you don't question anything

[00:45] that they're saying? That's what I want

[00:47] to talk about today. Not in a super

[00:50] technical way, not here's how to code a

[00:53] model in Python, just what models

[00:56] actually are, what they aren't, and why

[00:59] my model makes it this has kind of

[01:02] become the new trust me bro in sports

[01:04] betting.

[01:11] [music]

[01:16] [music]

[01:17] So quick background. So this doesn't

[01:19] feel like some random rant. I'm Rob

[01:21] Pizzola. I run the Hammer betting

[01:22] network, a sports betting media company,

[01:25] which which includes this channel,

[01:27] circles off, and many others. I spend

[01:30] way too much of my life consuming

[01:32] betting content. Not just the stuff that

[01:34] we make here at the hammer, but

[01:36] everybody else's, too. YouTube,

[01:38] podcasts, Twitter, Instagram, Tik Tok.

[01:41] If someone is talking about sports

[01:43] betting, there's a decent chance and it

[01:45] ended up in my feed at some point. And

[01:48] over the last couple years, my model has

[01:51] kind of become the backbone of a lot of

[01:54] that content. My model makes this, my

[01:56] model loves that. It it's just the

[01:58] default justification. Now, you don't

[02:00] have to explain your process. You just

[02:02] say my model and you move on. Now, I

[02:07] want to be really clear. Again, I'm not

[02:10] anti-model at all. I love this stuff. I

[02:13] use models for my own personal betting.

[02:15] I'm what's often referred to as an

[02:17] originator. I use multiple models. I

[02:19] spend a stupid amount of time trying to

[02:22] make them better. Models are a big part

[02:24] of how I bet. So, this isn't a models

[02:27] are bad video. It's not a coding

[02:29] tutorial. It's not me flexing like, "Oh,

[02:31] my numbers are better than everyone

[02:33] else's." Think of this more as a state

[02:35] of the union on model culture. what

[02:39] models actually do, how they're supposed

[02:42] to be used, and how a lot of people are

[02:44] using them in ways that frankly don't

[02:47] really make sense.

[02:49] So, let's strip away the mystery for a

[02:51] second and just talk about what a model

[02:54] actually is. At the most basic level, a

[02:58] model is just a structured way of

[03:00] turning information into a prediction.

[03:03] Numbers go in, number comes out, that's

[03:06] it. That number might be I think this

[03:08] team should be minus three and a half or

[03:10] I think this total should be 46 and a

[03:13] half or this team scores X points on

[03:16] average. But underneath it that number

[03:18] is always calculated the same way. Same

[03:21] inputs, same rules, same process every

[03:24] single time. And that can be really

[03:26] simple. The classic example is a power

[03:28] ratings model. Every team has a number.

[03:30] You take team A minus team B. You add in

[03:33] some home field advantage and boom, you

[03:36] have a spread. Or it can be way more

[03:39] complex than that. You can use

[03:40] play-by-play data. EPA per play, success

[03:42] rate, drive efficiency, player grades,

[03:45] injuries, weather, travel, all these

[03:47] things rolled together with some math

[03:50] and you end up with a prediction. The

[03:52] way I think about it is like a recipe.

[03:54] Your ingredients are things like yards

[03:57] per play, success rate, pressure rate,

[03:59] injuries, rest, travel, whatever. You

[04:02] decide how much each ingredient matters.

[04:04] You mix them together in a consistent

[04:06] way. And what comes out of the oven is

[04:08] your number on the game. The spread, the

[04:11] total, or whatever you're modeling. And

[04:13] here's the key part. A model is not a

[04:16] crystal ball. It's just an approximation

[04:19] of reality. Even the best models in the

[04:22] world are just slightly better

[04:24] approximations than everyone else's. The

[04:27] whole goal is to be a bit less wrong

[04:29] than the market on average, not to

[04:31] magically land 10 points off on every

[04:33] game, and definitely not to be perfect.

[04:36] When you hear my model makes this,

[04:39] you're not listening to some infallible

[04:41] truth. You're listening to someone's

[04:43] recipe for the game. Now, before we talk

[04:46] about different types of real models, I

[04:50] want to hit something that confuses a

[04:51] lot of people. What a model is not.

[04:55] There are a bunch of products out there

[04:57] that will let you query a database and

[04:59] do what I'd call trend mining. This team

[05:02] is an underdog in a prime time game

[05:04] coming off of a win. Uh home dogs

[05:06] between three and seven points after

[05:08] scoring 24 plus last week are 18 and

[05:11] seven against the spread. People run a

[05:13] query like that. They say, "Hey, this

[05:15] angle is 72% against the spread over the

[05:18] last 20 years." Suddenly they think

[05:20] they've have a betting model. That's not

[05:23] modeling. That's just slicing and dicing

[05:25] past betting results. You're not

[05:27] actually predicting how good the team is

[05:29] going forwards. You're not telling us

[05:31] what that information means moving

[05:33] forwards. You're just filtering history

[05:36] until you find a pattern that looks

[05:38] cool. And the problem is against the

[05:41] spread records are incredibly noisy. If

[05:43] you cut the data enough ways, you can

[05:46] find a trend for literally anything.

[05:48] Teams wearing blue jerseys on short

[05:50] rest. uh coaches named Mike coming off a

[05:53] loss. Uh how a team fares against the

[05:56] spread when they flip from a favorite to

[05:58] an underdog. If your model is basically

[06:01] step one, type in situational angle.

[06:04] Step two, look at past against the

[06:07] spread record. Step three, bet on it

[06:09] because it's 61% historically. You don't

[06:12] have a bottle. You have a search engine

[06:15] for randomness. A real model is

[06:18] forward-looking. It tries to capture

[06:20] things that cause teams to score and

[06:22] allow points. Play efficiency,

[06:26] explosiveness, turnovers, injuries,

[06:28] matchups, coaching, whatever you can

[06:31] quantify, and then it turns into a

[06:33] prediction before the game is played. If

[06:36] you're not producing a predicted score,

[06:38] spread, or total with a consistent

[06:41] process, it's not really a model, no

[06:43] matter how fancy it looks.

[06:46] So, with that out of the way, what kind

[06:48] of models are people actually using when

[06:50] they're trying to do this? Seriously? A

[06:53] lot of people start with what I'd call

[06:54] an entry-level power rating model. Every

[06:58] team has a number. You subtract one from

[07:00] the other. You add in home field, and

[07:02] that's your spread on the game. Now,

[07:03] there's nothing wrong with this as a

[07:05] learning tool. It's a great way to

[07:06] understand how markets move to track

[07:09] teams over time. maybe find a bit of

[07:11] value in smaller markets or on very

[07:14] early openers if you're sharp and you're

[07:16] disciplined, but in a market like the

[07:18] NFL with big limits, a ton of sharp

[07:21] money involved, that by itself is not

[07:23] going to cut it long term. If your plan

[07:26] is, I have 32 numbers and home field for

[07:28] each team, I'm going to beat a room full

[07:30] of people that are running massive in

[07:33] infrastructure and years of data. That's

[07:36] just not realistic. Then you move up the

[07:39] ladder into a more statisticalbased

[07:41] model. Here you use things like expected

[07:44] points added per play, success rate, red

[07:47] zone efficiency, maybe drive level

[07:49] metrics, and ideally you're adjusting

[07:51] for strength of schedule. You're trying

[07:54] to translate how teams have actually

[07:56] played into how many points they should

[07:59] score going forwards. That's closer to

[08:01] what serious betters are doing. But even

[08:03] that is just one layer. On top of that,

[08:07] the pro stuff usually includes player

[08:09] level and injury concept concepts.

[08:12] You're not just saying team X has a

[08:14] plus4 rating. You're saying what is this

[08:16] offense worth with this quarterback,

[08:19] with this offensive line, these

[08:21] receivers? What happens when the left

[08:23] tackle is out? What happens when the

[08:24] quarterback room is down to backups?

[08:27] What if the quarterback is limited and

[08:28] he can't run? At a truly professional

[08:31] level, the model is extremely detailed.

[08:34] baseline team ratings, play-by-play or

[08:37] drive level efficiency models, player

[08:40] level adjustments for injuries and

[08:42] usage, situational tweaks for things

[08:45] like rest, travel, weather, maybe even

[08:49] how teams change play calling when

[08:50] they're leading or trailing. Constant

[08:53] back testing and updating so it doesn't

[08:55] fall apart when the league changes.

[08:58] That's the reality of trying to bet into

[09:00] big limits in the NFL or the NBA or any

[09:02] major sport. It's extremely competitive

[09:05] and the people you're up against are

[09:06] trying to account for as many predictive

[09:09] factors as possible. They're not just

[09:11] saying team A is a plus4, team B is a

[09:13] plus one. Even with all that, the edges

[09:16] you're fighting for are still small.

[09:19] You're not walking around with my model

[09:21] makes this game minus 21 when the market

[09:24] is minus 7 in the most liquid markets in

[09:26] the world. You're hunting for maybe a

[09:28] point of value here or there, maybe a

[09:30] little bit more in the right spots. and

[09:32] you're trying to do that over and over

[09:34] and over without the thing breaking.

[09:37] There's also this fantasy that I see now

[09:39] of the overnight supermodel. You open up

[09:42] your Excel on Friday, you grab a couple

[09:45] stats from a website, you plug them into

[09:47] a formula, by Sunday, you've built the

[09:49] thing that's going to crush NFL sides

[09:51] for the rest of the time. That's just

[09:53] not how this works in markets like the

[09:55] NFL. Look at who you're competing

[09:58] against. You've got people betting six,

[10:00] seven figures a weekend. You've got

[10:03] syndicates with full-time quants,

[10:05] developers, data feeds that cost more

[10:08] than a car. Models they've been

[10:10] iterating on for years. You've got

[10:12] longtime pros who have seen every trend,

[10:16] every fake edge, every new angle come

[10:18] and go. And even those people with all

[10:21] that infrastructure are still constantly

[10:24] testing, breaking, and rebuilding their

[10:27] models. They run into edges that

[10:30] disappear. They hit stretches where

[10:32] their stuff stops working and they have

[10:34] to adapt. They have losing weeks, losing

[10:36] months, maybe even a losing season. So,

[10:39] if you spin up a model in a weekend and

[10:41] suddenly it says, "Well, every favorite

[10:43] on the board should be 10 points higher

[10:44] than the market." The most likely

[10:47] explanation is not that you've just

[10:49] unlocked a secret that nobody else on

[10:51] Earth has found. The most likely

[10:53] explanation is that something in your

[10:55] process is off. Maybe you double counted

[10:58] something. Maybe you didn't adjust for

[11:01] strength of schedule. Maybe you're using

[11:03] noisy stats that don't actually predict

[11:05] the future. There are a million ways to

[11:08] accidentally create fake edges. The

[11:11] healthier mindset is if your brand new

[11:14] model is screaming that the entire board

[11:17] is wrong, assume your model is wrong

[11:20] first, not the world. Use that as a cue

[11:22] to dig in. Figure out what's broken.

[11:24] Treat it as a learning step, not a

[11:27] signal to fire into every single game.

[11:30] As a general rule of thumb, the more

[11:32] efficient the market, the more likely

[11:34] you are to be wrong. One of my biggest

[11:36] pet peeves with model culture is the

[11:38] this line that you hear all the time, my

[11:41] model makes this minus3, the market's

[11:44] minus one, but I'm actually going to bet

[11:46] the dog. Okay, then why do you have a

[11:48] model? Like genuinely, what what are we

[11:51] doing here? The entire point of building

[11:53] a model is to take everything you

[11:55] believe is predictive and lock it into a

[11:57] process so that when you get to a game,

[12:00] you're not just going off of vibes.

[12:02] You're going off of something

[12:03] consistent. If every time your number

[12:05] disagrees with the market, you just

[12:07] said, "Yeah, but I'm not feeling it."

[12:09] And you bet the other side, you don't

[12:11] actually believe in your model. You're

[12:13] just using it as a prop in the

[12:15] conversation.

[12:16] Now, it's totally fine to think your

[12:18] model is missing something. Maybe you

[12:20] think there's a motivational angle, a

[12:22] look ahead, revenge game, or whatever

[12:25] narrative you like. That's great. But

[12:27] the next step shouldn't be, I'm going to

[12:30] ignore my number and bet against it. The

[12:33] next step should be, let me test whether

[12:36] this thing I believe in actually

[12:38] matters. Can I define what a look ahead

[12:42] is in a way that I can go back and I can

[12:44] study it? Can I define what a revenge

[12:47] game is in a way that I can go back and

[12:50] study it? Can I measure how many teams

[12:53] in those spots have actually performed

[12:56] in meaningful stats like EPA, success

[12:59] rate, scoring margin, not just against

[13:02] the spread noise. If you find something

[13:04] real, you bake it into the model. You

[13:07] upgrade the process so that next time

[13:10] your number already reflects that angle.

[13:12] You're not guessing in the moment. Rufus

[13:15] Peabody, one of the best professional

[13:17] betterers that I know, talked about this

[13:19] years ago on Bet the Process, his

[13:22] podcast, and it really stuck with me.

[13:23] His basic point is, if you think

[13:25] something is important, figure out how

[13:28] to capture it in your model. Don't just

[13:30] talk about it. Don't just override the

[13:32] number on gut feeling. Either prove it

[13:35] and include it or admit that it's

[13:37] probably just a story you like telling

[13:39] yourself.

[13:41] Another version of this which kind of

[13:42] lives in the same neighborhood is the

[13:44] classic uh I make this game minus four

[13:47] but it's a look ahead spot so I'm

[13:50] passing on it. I see this all the time.

[13:52] And again if you're actually a modeler

[13:54] this is an opportunity to look into the

[13:56] data further. Every spot like this in a

[13:59] sport where you can test and improve the

[14:01] assumptions of your model. If you really

[14:04] like, if you really believe look ahead

[14:07] matters, your brain should immediately

[14:09] go to, okay, what is a look ahead spot

[14:12] exactly? Is it when a team is a big

[14:15] favorite this week and they play a

[14:17] divisional rival next week? Is it a

[14:20] quote unquote sandwich game between two

[14:22] tougher opponents? Is it certain ranges

[14:26] of point spread or certain parts of the

[14:28] season? You define it first, then you go

[14:30] back and you test it. Do teams in that

[14:33] spot actually underperform in anything

[14:36] that matters? Do they score less? Do

[14:38] they give up more points? Is their EPA

[14:39] worse? Does their success rate drop? Not

[14:42] just, oh, they're 46% against the spread

[14:45] on this angle because that's mostly

[14:47] noise. If you go back and find that,

[14:50] yeah, teams in this situation really do

[14:53] underperform in meaningful ways, then

[14:55] perfect. Bake it into the model. Maybe

[14:57] that's worth a small downgrade to the

[14:59] team's rating in those specific

[15:00] situations. Maybe it's worth a fraction

[15:03] of a point off the point spread.

[15:05] Whatever the number is, now your model

[15:07] knows about look aheads. So when you get

[15:10] to week 10 and you say, "I make this

[15:12] minus4," you're not saying, "I make this

[15:15] minus4, but uh look ahead spot. I'm I'm

[15:18] going to pass." You're saying, "I'm

[15:20] making this minus4," including the look

[15:22] ahead spot. You're not guessing on the

[15:23] fly anymore. The work is already done.

[15:25] It's in the number. Another one that

[15:27] makes me shake my head is this. Market

[15:29] has team A at minus 7. Content guy comes

[15:32] along and says, "My model makes this

[15:34] game minus 21." Okay, let's just walk

[15:38] through what that actually implies. If

[15:40] that were true, the real number on this

[15:42] game or minus 21 and the entire world is

[15:45] hanging minus 7, you are sitting on one

[15:48] of the biggest edges you will ever see

[15:51] in one of the most liquid markets on the

[15:54] entire planet. You could in theory keep

[15:56] betting that edge all the way up and

[15:59] every sharp book on earth would be

[16:01] begging you to take more action. That

[16:04] almost never exists in the NFL or the

[16:06] NBA or any mature market. Now you might

[16:09] be directionally correct. Maybe minus 7

[16:12] is cheap. Maybe the true number is - 7

[16:15] and a half or - 8, even - 9. But a

[16:18] 14-point disagreement between your model

[16:21] and a heavily bet market is not usually

[16:23] evidence that you've broken the code. It

[16:26] is almost always evidence that something

[16:29] in your model is broken. And this is

[16:31] where humility matters. Treat those

[16:33] giant edges as a bug that you need to

[16:35] investigate, not proof that you're a

[16:37] genius and everyone else is an idiot. So

[16:40] with all that said, how should you

[16:42] actually use a model? To me, there are

[16:45] three big things. Number one, it's an

[16:47] anchor. It's your starting point for

[16:50] what a game should be. Not a dictator,

[16:52] not something you blindly follow no

[16:54] matter what, but the baseline you work

[16:56] from. Okay, my number is minus3. Where

[16:59] is the market? Why are we different? It

[17:02] frames the entire conversation.

[17:04] Second, it gives you consistency.

[17:06] Without a model, it's really easy for

[17:08] your opinions to swing all over the

[17:10] place from week to week. One week you

[17:12] love a team, next week you hate the

[17:14] team. You can't really explain why a

[17:17] model forces you to treat similar

[17:18] situations similarly. If two teams have

[17:22] played basically the same, they

[17:24] shouldn't be priced 10 points apart just

[17:26] because you're mad that one of them

[17:27] burned your teaser last week. And third,

[17:30] and and what is the main thing I'd love

[17:32] for you to take away from this video?

[17:34] It's an experiment lab. You have a

[17:37] theory about rest or travel or penalties

[17:39] or weather, whatever. Great. Plug it in,

[17:42] test it, see if it moves the needle on

[17:44] your prediction. If it does, keep it. If

[17:47] it doesn't, trash it. You're constantly

[17:50] refining instead of just stacking

[17:52] narratives. At the end of the day, your

[17:54] model shouldn't be a buzzword or a

[17:56] shield you hide behind. It's just a tool

[17:58] to help you make slightly better, more

[18:01] consistent decisions over hundreds and

[18:04] thousands of bets. That's where it

[18:07] really matters. So, just to wrap this

[18:09] up, I am very pro model. I think they're

[18:13] one of the best tools you can have as a

[18:14] better. What I'm against is the model

[18:18] cosplay that's taken over the space.

[18:20] Throwing around my model as a magic word

[18:24] with nothing behind it. Real models are

[18:26] powerful, but they're also hard. They're

[18:29] fragile. They're always wrong to some

[18:32] degree. The name of the game is trying

[18:34] to make them a little bit less wrong

[18:37] than everybody else's. If you're a

[18:40] recreational better and you're watching

[18:41] this, it is completely fine if you don't

[18:43] have a model. You don't need one to

[18:46] enjoy betting sports. You don't need to

[18:48] pretend you have one just because

[18:50] everyone on YouTube is saying, "My model

[18:52] makes it this." Bet small, have fun, try

[18:55] to make good decisions. That's enough.

[18:58] If you're an aspiring modeler, my advice

[19:01] is this. Take it slow. Test things.

[19:05] Break your own ideas. Be honest about

[19:07] what your numbers can and cannot do. If

[19:10] your model disagrees with the market by

[19:11] a little, great. Dig in. If it disagrees

[19:14] by a ton, assume you mess something up

[19:16] and figure out why. For all of you

[19:19] betting out there, I'd actually love to

[19:21] hear what you're using. Do you have your

[19:23] own numbers? Are you thinking about

[19:26] building something? or do you think

[19:28] model has just become a buzzword people

[19:30] are hiding behind right now? Let me know

[19:32] down in the comments below. I do love

[19:34] reading all the comments on this channel

[19:36] every single week. And if you found this

[19:37] video helpful, hit the like button,

[19:39] subscribe to the channel, share it with

[19:41] someone who's always talking about their

[19:43] model, and we'll do more videos like

[19:45] this, perhaps diving a little bit deeper

[19:48] into modeling in the future. Thanks for

[19:50] watching.