[00:02]
games to project a prop, you're doing it backwards. And it's the single biggest reason most prop bettors are losing money right now. You've seen this before. You've seen it on Twitter, on YouTube, in paid Discords. It looks like
[00:17]
math, it feels like an edge, but it's neither. In the next 15 minutes, I'm going to show you why this approach produces the wrong answer. What you should actually be looking at instead. And later on in this video, I'm going to
[00:31]
walk you through a real example in the NBA where the hit rate tells you to bet one side and the actual process says to bet the other. Stick around to the end because there's a piece of advice in here for every better. You need to hear
[00:46]
it and you need to remember it. >> Here we go. >> First, let me show you what I'm talking about. An NBA player has an alternate total of over 20 and a half points at minus 400. You pull up his last 20
[01:02]
games, you see he went over 20 and a half in 18 of them. That's 90%. The half in 18 of them. That's 90%. The implied probability at minus 400 is 80%. So, you say there's a 10% edge. There is some real value here. And it sounds
[01:16]
right. But it isn't. It's wrong. And here is the most important thing I want you to understand. Your hit rate is not a probability. A probability is forward-looking. It accounts for the
[01:30]
conditions of today's game. The matchup, the minutes, who's in the lineup, who's out of the lineup. A hit rate is backward-looking. It tells you what already happened under a completely different set of conditions. Now,
[01:45]
treating a hit rate as a true probability is not a shortcut. It is a logical error and it is the most common logical error that I see in prop betting today. If you're new to this channel, I'm Rob Pizzola. I bet for a living.
[02:00]
Circles Off is where I show the actual processes that I use for betting. Not picks, not parlays, the the work behind the bets. If that's what you're looking for, this video is going to be useful. Now, let me give you four reasons why
[02:15]
Now, let me give you four reasons why hit rate breaks down. First is endpoint hit rate breaks down. First is endpoint selection. Why 20 games? Why not 15? Why not 30? The number is arbitrary and small changes to it produce wildly
[02:28]
different results. If your edge depends on where you draw the line, you don't actually have an edge. You have a result of where you decided to start counting. of where you decided to start counting. Second, there is no context inside those
[02:42]
games. Were teammates out for part of that stretch and a player's usage was that stretch and a player's usage was inflated? Was there a run of bottom-five defenses? Was there a couple overtime games that added minutes? The hit rate
[02:55]
treats every game as identical, but they are not. Third, and this is the one that gets people heated when I bring it up. A few months ago, we did a reaction video to a LineMaker Sports video with Isaac Rose
[03:10]
same thing I'm going to tell you right now. The comments lit up. One commenter in particular, this one stuck with me because it is such a clean example of this misconception. He said this, and I quote, "If you have 10 cups and eight of
[03:26]
them have a ball in them, there is an 80% true probability of picking a cup with a ball." On the surface, that sounds right. But sports outcomes are sounds right. But sports outcomes are not cups with balls in them. Cups have a
[03:41]
fixed distribution you can count right now. Sports outcomes are uncertain. They change every night. They depend on hundreds of variables that were not present in your sample. The cups analogy is the exact mistake.
[03:56]
Past results are not a fixed probability that you can read off the page. Now, if this is starting to click for you, hit the like button down below. It tells YouTube to push this content to other people who actually want to learn the
[04:09]
process instead of just getting fed pick of the day type of content. It genuinely helps us. Now, I want you to remember this cups analogy because I'm going to come back to it when we look at the actual projection in a few minutes.
[04:24]
actual projection in a few minutes. Fourth and final, this same flawed logic gets applied to alternate lines and the math is even worse there. Same player, the main line over 20 and 1/2,
[04:39]
no edge by hit rate. So, you scroll the alternates. Over 17 and 1/2 at minus 250. Well, he went over that in 14 of 20. Over 14 and 1/2 at minus 400. 17 of
[04:51]
20. Over 11 and 1/2 at minus 50. 19 of 20. Each step down the alternates ladder, the hit rate looks better and better and the edge magically appears.
[05:03]
But, it's not appearing. Alternate lines are derivatives of the main number. They are mathematically priced off of the same projection. So, if you have no edge on the main line, you almost certainly have no edge on the alternates. Now,
[05:17]
there are some exceptions here where you might have a unique distribution for a particular game. I'll probably do a video on that in the future as well. But, what you are doing is sliding around to find the threshold where the
[05:30]
backwards-looking hit rate looks the most impressive. That is the same endpoint selection problem from before just applied to the line instead of the games. Coming up, the three-layer framework I actually use and a worked
[05:46]
NBA example with the math on screen. Okay, so we've spent some time here on what is actually broken. Now, let me get to what actually works. For every prop you bet, there is an underlying process that generates the stat.
[06:01]
That process is more stable, it is more predictive, and it is more useful than the stat itself. Your job is not to project the outcome. Your job is to project the process and let it inform the outcome. I think about
[06:17]
this in three layers. Layer one is volume. getting? The metric depends on the prop. So, if targets. Goal scorers, shot attempts. MLB hits,
[06:34]
plate appearances. NBA points, you have minutes, you have field goal attempts. Whatever the prop is, there is a volume metric underneath it that is more stable than the box score number. Most people skip this step entirely. Layer two is
[06:49]
quality and context. Not all volume is equal. A target on a go route is different than a target on a screen pass. A shot from the slot is different than a shot from the point. What kind of opportunities is the player getting, and
[07:05]
does the match up say that that continues? Layer three is conversion. This is where the actual stat lives. Receptions, goals, hits, assists. It is
[07:17]
the output of volume and quality. It is the noisiest, and it is the only layer that most people tend to look at. Now, this framework is a starting point, not the finish line. Each layer has more depth underneath it. But, going even one
[07:33]
layer deeper than the box score puts you ahead of a lot of the market. All right, this is the moment I promised you in the opening of the video. I'm going to take that cups analogy, that 80% true probability the guy was talking about,
[07:46]
I'm going to show you what real projection looks like instead. Let's say you have two players. Both are averaging 22 points per game in the NBA over their last 10 games. The hit rate method sees them as
[08:00]
They are not. Player A is playing 38 minutes a game, taking 18 shots, shooting 43% from the field, 34% from three, and 58% of his
[08:14]
Player B, playing 33 minutes, 14 shots, 52% playing 33 minutes, 14 shots, 52% shooting, 42% from three, and 35% of his shots are at the rim. Now, look at what is going on here.
[08:29]
Player A has bigger minutes. That might be inflated because of a teammate injury, could be many reasons. If those teammates come back, his him. But, the 18 shots per game on 43%
[08:45]
shooting, 58% of them at the rim, that is sustainable. The volume is real, and the shot quality is real. Player B looks great on conversion. 52%
[08:57]
from the field, 42% from three, but he is shooting 14 times a game, mostly from 52% is almost certainly not sustainable.
[09:10]
When that comes back down to earth, the 22 points come down with it. Now, does that mean player B is definitely going to regress? Well, not necessarily. Maybe he's a genuinely elite shooter. Maybe his role generates higher quality
[09:26]
looks than the surface stats show. But, those are the questions you have to dig into. Is he getting lucky? Is the shot quality real? Is the volume real? The quality real? Is the volume real? The hit rate approach does not even know to
[09:40]
ask those questions. And this is the difference between a backward-looking count and a forward-looking projection. The hit rate is gives you a number. The process tells you whether or not that number is going to hold up. If you want
[09:55]
more of this, the actual process, please subscribe down below. One of the many things this channel does is break down how I think about this stuff. About 60% subscribed. We'd love to have you as a subscriber.
[10:09]
We'd love to have you as a subscriber to Circles Off. Quick second NBA example here because this one is so clean. This comes from Wade Andrew on Twitter. He assists. A potential assist is when a player makes a pass that directly leads
[10:25]
to a shot attempt. It is the assist equivalent of a shot attempt. Whether the teammate makes the shot or not has nothing to do with the passer. So, in this example, we have two players, both of them with two assists.
[10:40]
Player one had nine potential assists. but the teammates did not knock down those looks. Player two had three potentials, converted two of them. The hit rate sees these players as the same.
[10:56]
The process sees a massive gap. Player one is generating way more opportunity, and you'd expect some positive regression towards his normal conversion rate going forwards. Now, here's where I'm going to lose some
[11:09]
of you. If your entire prop process is is pulling up a filtering tool and looking at hit rates over the last 20 games, you're not really betting. You're paying for entertainment at that point. This is not a serious way to bet.
[11:24]
Tell me in the comments down below, last 90 days, what's your actual ROI on player props? Not your win rate, your ROI. If you don't even have a way of something wrong. I want to see the
[11:37]
are going to find out that the answer is not as glamorous as they thought. Now, real quick because the framework is the same in every single sport, right? MLB Stop looking at hits per game. Look at other stats. BABIP, batting
[11:53]
average on balls in play. When I was betting baseball seriously long ago, league average was around 300. Not entirely sure where it sits right now, but the concept is still the same. A guy running a 380 BABIP is getting lucky on
[12:07]
balls in play. That comes down. A guy at 220 with elite contact quality is getting unlucky. That comes up. Then you can go deeper with comes up. Then you can go deeper with Statcast. Exit velocity, hard hit rate,
[12:20]
barrel rate. That data's all free on Baseball Savant right now. NHL goal scorer props. This is my world. I almost never look at past goals as a primary input. I look at individual expected goals. How many
[12:35]
high-quality chances is this player generating regardless of whether or not they went in the net? Shooting percentage relative to their career average. Is this player running hot or cold? Opposing goal tenders recently.
[12:49]
Ice time. Power play deployment. A player with zero goals in five games but really good individual expected goals against tough goalies, well, depending on the price, that could be interesting to me. The hit rate would
[13:03]
tell you to fade this person. The process tells you to back him. NFL receptions. Targets and target share are way more predictive than receptions. Catch rates fluctuate wildly in small samples, and
[13:17]
fluctuate wildly in small samples, and the real edge predicting usage changes before they show up in the box score. Injuries, scheme changes, game plans against specific teams. The last 10 or 15 games are useless for projecting
[13:31]
something that has not happened yet. It's the same framework everywhere. It's the same framework everywhere. Volume, quality, conversion. Look underneath the stat that you're projecting. Now, here is a piece of
[13:44]
advice I promised you at the top. When you bet a player prop, vacuum. You are betting into a market. And that market has other participants. Some of those participants are betting
[14:01]
groups with real resources, teams of people building models, they're tracking process metrics, potentially watching film, running automated pipelines, all
[14:13]
sorts of stuff. This is who you are competing with. The market opens and it moves based on the action it takes from both sides. If other competitors in the market are doing serious analytical work and
[14:27]
your process is looking at the last 15 or 20 games, I hate to break it to you, you are at a major disadvantage. That is just how these markets function, plain and simple. Now, does that mean don't bet? No. But it means be realistic about
[14:44]
what you're up against. If you're not going to do deeper analysis, you can find an edge by being faster. You can find an edge by betting early before the market moves, but you need something else, some kind of advantage.
[14:59]
Whether that's analytical depth or speed or information, you need something that is beyond what anyone can do with a basic data The people making money in props long-term are doing real work. This
[15:15]
video is not meant to discourage you. It's the exact opposite. It is meant to show you what that work actually looks like and yourself. The stat you are betting
[15:29]
is almost never the best stat to use when projecting it. I'll say that again. The stat you're betting is almost never the best stat to use when projecting it. There's almost always something underneath. Each layer you add makes
[15:43]
your projection better. You just have to do more than finding a hit rate. If you want to see this framework taken to its full conclusion, an actual model that I built from scratch, every tab, every formula, check
[15:56]
walk-through is one of the hottest videos we've done in the past couple of months, and the process is the same there.