One of the biggest misconceptions in sports betting is that you need to create your own point spread to find an edge against the sportsbooks.

That’s not necessarily true.

Our NBA statistical model uses a different approach, something I call the Margin Concept.

Rather than producing a projected point spread and comparing it to the sportsbook’s number, we rank all 30 NBA teams based on our statistical model and look for specific matchups where we believe the market may be mispricing the teams.

The key is understanding that we’re not simply betting on the highest-ranked team. We’re looking for situations where the difference in our rankings, combined with the sportsbook’s closing line, creates a potential betting advantage.

In this article, I’ll explain the three primary rules of the model and walk through the games from October 28 to demonstrate exactly how we identify qualifying plays.

Understanding the NBA Power Rankings

Our statistical model ranks every NBA team from 1 through 30.

But instead of treating every ranking equally, we divide teams into several groups.

• Ranks 1–7: Our highest-rated teams.

• Ranks 8–10: Strong teams that also help define our top-team exclusions.

• Ranks 11–23: The middle of the league, where we believe the greatest opportunities exist.

• Ranks 24–30: Our lowest-rated teams and the foundation of our Play Against Concept.

These rankings are not necessarily a prediction of which team will win the NBA championship. They’re a statistical evaluation designed to help identify betting opportunities.

Rule #1: Playing Our Top Seven Teams

The first rule focuses on teams ranked 1 through 7.

We will play a top-seven team when:

1. Its opponent is ranked outside the top 10.

2. The sportsbook closing spread is -8 or less.

For example, suppose our model ranks Houston No. 1 and Milwaukee No. 18.

Houston qualifies from a rankings perspective because Milwaukee is outside the top 10.

However, we still need to evaluate the sportsbook’s closing line.

If Houston closes at -8 or less, it qualifies as a play. If Houston closes at -9.5, we pass.

Why the -8 cutoff?

The best teams in the NBA are generally easy to identify. Sportsbooks know who they are, and the betting public knows who they are.

As a result, those teams can become overvalued.

During our 2025 backtest, the top-seven strategy generated only approximately three units of profit, despite the overall model producing more than 30 units.

The objective isn’t simply to bet on good teams. It’s to find good teams at prices that provide a reasonable margin.

Rule #2: The Middle of the League — Our 50/50 Concept

This is where things get interesting.

During our backtesting, the greatest profitability came from matchups involving teams ranked 11 through 23.

Why?

Because the middle of the NBA is much more difficult to price efficiently.

It’s relatively easy to identify the league’s best teams and its worst teams. But separating the teams in the middle becomes much more challenging, especially early in the season.

Our approach is to treat matchups between teams ranked 11 through 23 as essentially 50/50 games.

When both teams fall within that range, we play the underdog using these rules:

• Underdog +6 or greater: Play the point spread.

• Underdog +5.5 or less: Play the moneyline, provided the price is +115 or greater.

This doesn’t mean every game has a literal 50% probability of either team winning.

It’s a handicapping concept based on the assumption that these middle-ranked teams are more evenly matched than the sportsbook’s pricing may suggest.

And during our backtest, this category generated the majority of our profits, particularly through moneyline underdogs.

Rule #3: The Play Against Concept

The third rule is something I’ve used in other sports as well.

I call it the Play Against Concept.

Rather than focusing on how good a particular team is, we’re identifying teams that our statistical model considers especially weak.

Here’s the rule:

1. Identify a team ranked 24 through 30.

2. Its opponent must be ranked 1 through 23.

3. The opponent must close as a favorite of -8 or less.

If all three conditions are satisfied, we play against the bottom-ranked team.

The philosophy is simple.

Sometimes the greatest betting opportunity isn’t identifying a great team. It’s identifying a poor team and finding a reasonable price to play against it.

Putting the Model to Work: October 28, 2026

Now let’s walk through the actual games from October 28.

This is the best way to understand how the rules work in practice.

1. Miami vs. Atlanta — Play Against Concept

Miami was ranked 25th, placing it in our bottom-seven category.

Atlanta was ranked eighth, well within the required 1–23 range.

With Atlanta closing at -2.5, all three conditions of the Play Against Concept were satisfied.

Model Play: Atlanta -2.5

Result: Miami won 129–117. Loss.

This is an important reminder that a qualifying model play is not a guarantee. We’re identifying situations that we believe offer a long-term advantage, not predicting every individual outcome.

2. Indiana vs. Philadelphia — Play Against Concept

Here’s an interesting example.

Our model ranked Indiana 11th and Philadelphia 29th.

Despite Indiana being significantly higher in our rankings, the sportsbook had Indiana as a 7.5-point underdog.

This creates a substantial difference between our statistical evaluation and the market’s evaluation.

We take the 7.5 points with Indiana.

Model Play: Indiana +7.5

Result: Philadelphia won 114–109. Indiana covered. Win.

3. Charlotte vs. Detroit — Play Against Concept

Charlotte ranked 28th, while Detroit ranked 12th.

Detroit qualifies because Charlotte is in our bottom seven, Detroit is ranked inside the top 23, and the closing spread is less than -8.

Model Play: Detroit -7.5

Result: Detroit won 97–89. Win.

Detroit covered by half a point.

4. Toronto vs. Orlando — Play Against Concept

Orlando ranked 26th, qualifying as one of our bottom-seven teams.

Toronto was ranked 21st.

With Toronto closing at -5, the matchup satisfied our Play Against Concept.

Model Play: Toronto -5

Result: Toronto won 129–99. Win.

5. Houston vs. Milwaukee — Top Seven Concept

Houston was our No. 1-ranked team, while Milwaukee was 18th.

Because Milwaukee was outside the top 10 and Houston closed at exactly -8, the game qualified under Rule #1.

Model Play: Houston -8

Result: Houston won 132–100. Win.

This is exactly what we’re looking for when playing our highest-ranked teams: a favorable rankings matchup without laying more than eight points.

6. Denver vs. New Orleans — Top Seven Concept

This is another interesting example of the difference between our model and the sportsbook.

Our statistical model ranked New Orleans fourth and Denver 15th.

Yet New Orleans was receiving 4.5 points.

Because New Orleans was a top-seven team playing an opponent outside the top 10, this qualified under our top-team concept.

Under our model’s rules, when a qualifying underdog is receiving 5.5 points or less, we play the moneyline, provided the price is +115 or greater. If the underdog is receiving 6 points or more, we take the points.

Model Play: New Orleans Moneyline

Result: Denver won 104–91. Loss.

7. Oklahoma City vs. Dallas — The 50/50 Concept

This is probably my favorite example because it demonstrates the concept that generated the greatest profitability during our backtesting.

Both teams ranked between 11 and 23.

Under our model, we consider this a 50/50-type matchup.

Dallas was receiving five points, which falls within our moneyline rule of +5.5 or less.

The closing moneyline was +148, exceeding our minimum requirement of +115.

Model Play: Dallas ML +148

Result: Dallas won 120–96. Win.

By risking one unit on Dallas, we gained 1.48 units.

8. LA Lakers vs. LA Clippers — Top Seven Concept

The Lakers ranked third in our model, while the Clippers ranked 22nd.

The Clippers were outside the top 10, and the Lakers closed as just a one-point favorite.

That easily satisfied our top-seven rule.

Model Play: Lakers -1

Result: Lakers won 103–87. Win.

Understanding Why Some Games Are Passes

An important part of using any statistical model is knowing when not to wager.

Two matchups from October 28 demonstrate this.

New York vs. Chicago: Both teams were ranked in our bottom group. Because neither team qualified as an opponent ranked 1–23, there was no Play Against opportunity.

Golden State vs. Minnesota: Both teams were highly ranked. Under our top-seven rule, we don’t play one of our best teams against another top-10 opponent.

Neither matchup qualified.

And that’s perfectly fine.

The objective isn’t to find a wager on every NBA game. It’s to identify the situations that satisfy our predetermined rules.

October 28 Model Results

Final Thoughts

Our NBA Power Rankings are designed to identify potential betting value using three concepts: Play For, Play Against, and 50/50 Matchups.

The key is understanding how the rankings work together with the sportsbook’s closing lines to identify qualifying plays.

Our 2025 backtest finished 62–39, +30.89 units, with the majority of profits coming from matchups involving middle-ranked teams.

For those interested in learning more about how the model is built, including which statistics we use and how to calculate the NBA Power Rankings, I’ve created a separate article explaining the entire process.

(Click here to learn how to build our NBA statistical model and calculate the Power Rankings)

— MG Covers | @mgcoverz

Over the past several weeks, I’ve been experimenting with different approaches to building a profitable soccer betting model.

We’ve tested recent form, goals scored, goals allowed, home and away performance, and different methods of calculating projected outcomes.

Some of these concepts have shown promise. Others haven’t.

But the more research I do, the more I’m leaning toward a much simpler approach.

Instead of trying to predict the exact outcome of a soccer match, what if we focused on identifying situations where the sportsbook might be undervaluing an underdog?

More specifically, can we develop a model that consistently identifies underdogs capable of winning approximately 40% of their matches while offering favorable moneyline prices?

That’s the question I’m trying to answer.

This article isn’t about presenting a finished, profitable model. It’s about showing you how I’m approaching the research, what we’ve discovered, what hasn’t worked, and how we might refine the concept.

Read more

That’s the foundation of this NHL power rankings model.

Read more

Sometimes, a relatively simple statistical approach can identify opportunities that the betting market may be undervaluing.

The key is finding a statistic that has a meaningful relationship with winning, applying it consistently, and then determining whether the sportsbook is accurately pricing that information.

That’s the approach we’ve been exploring with NHL betting this season. And the early results have been encouraging.

Today, I want to walk through another application of that approach, review the early results, and explain what we’re going to investigate next.

Read more

It’s easy when every player has the same amount of history. The difficult part comes when the samples aren’t equal. Some players have years of reliable data, others have one meaningful season, and some have almost no usable history at all.

That’s where you need rules.

If you don’t establish those rules beforehand, it’s very easy to start making subjective adjustments based on what you think a number should be. And once you start doing that, you’re no longer allowing the data to drive the process.

In today’s post, I want to show you how we’re handling those different sample sizes and the rules we’re using to keep the calculations consistent.

Read more

There will be a link below to the original post where I explained the model in greater detail. For this update, I mainly want to walk through the calculation process so everyone understands where the Week 4 lines are coming from.

Read more

That doesn’t mean you can’t build an objective model. It means you need to figure out what historical information you can use until the current season produces a large enough sample.

For our early-season NHL totals model, we’re keeping things extremely simple.

We’re using one statistic:

Starting goalie goals-against average (GAA).

Before we go any further, an important disclaimer:

This is not a proven betting model.

I have not done a large historical backtest on this system, and I’m not claiming that this will be profitable going forward. Our initial 10-game test went 7-3, which is encouraging, but 10 games tells us very little about the long-term profitability of a model.

The purpose of this exercise is bigger than these 10 NHL games.

I want to show you how you can take a statistic, create an objective model from it, establish rules, track the results, and eventually determine whether you’ve found something that has value in the betting market.

Why Goalies?

When I build models, I generally like simplicity.

Rather than throwing 10 or 15 statistics into a formula, I like looking for one statistic that:

1. Has a logical relationship to what we’re trying to predict.

2. May be undervalued relative to everything else the market is considering.

For an NHL totals model, goalie performance obviously has a relationship to scoring.

The sportsbook isn’t ignoring goalies. That’s not the argument.

The sportsbook is accounting for everything.

Team strength. Offensive ability. Defensive ability. Players. Injuries. Expectations. Public perception. And, of course, the starting goalies.

But that creates an interesting possibility.

Imagine two teams that the market considers very good offensive teams. Because of the quality of those teams, the sportsbook might hang a relatively high total.

Our model doesn’t care about any of that.

We’re isolating the starting goalies.

If both starting goalies have historically allowed very few goals, our model could produce a considerably lower number than the sportsbook.

That’s the potential inefficiency we’re trying to identify.

It doesn’t mean the market is wrong.

It gives us something objective to test.

Why Use Two Seasons?

Early in the season, I don’t want to rely on the first handful of games from the current year.

At the same time, using only one previous season can create problems.

So we’re generally using the previous two seasons of goalie data.

The calculation couldn’t be simpler:

Year 1 GAA + Year 2 GAA ÷ 2 = Goalie Rating

Once we have a rating for each starting goalie:

Goalie A Rating + Goalie B Rating = Our Model Total

That’s it.

We’re deliberately keeping this simple.

The same general idea can be applied when you’re experimenting with models in other sports. Find a statistic that logically relates to the outcome you’re trying to predict, use enough historical data that you’re not completely dependent on one season, and then test whether the statistic identifies something the betting market may be undervaluing.

What If We Don’t Have Two Good Seasons of Data?

This is where we made an adjustment.

Not every goalie has two seasons containing a meaningful sample.

Akira Schmid is a good example.

Schmid posted a 2.59 GAA last season, but his earlier NHL seasons contained relatively limited playing time. Rather than treating those smaller samples exactly the same as a full season, we used last season’s 2.59 and regressed the other half of his rating toward the NHL league average of 3.08.

So:

2.59 + 3.08 = 5.67

5.67 ÷ 2 = 2.835

Schmid’s rating in our model becomes 2.835.

We can use exactly the same idea with a rookie or a goalie with only one meaningful season.

Carolina’s Brandon Bussi had a 2.47 GAA last season.

Instead of assuming 2.47 represents his true expected performance, we average it with the 3.08 league average:

2.47 + 3.08 = 5.55

5.55 ÷ 2 = 2.775

His adjusted number becomes approximately 2.77.

This gives us a simple way to handle incomplete historical samples without completely throwing the goalie out of the model.

Establishing the NHL Baseline

Next we need something to compare our model against.

The NHL league average we’re using is 3.08 goals per team.

Since two teams play in every game:

3.08 × 2 = 6.16

That gives us an average NHL game total of 6.16 goals.

Now we can compare our goalie-based projection against that baseline.

How We’re Identifying Unders

We’re currently tracking two potential methods.

Remember, this model hasn’t been sufficiently backtested yet. We’re not going to pretend we know which method is superior. We’re going to collect the data and find out.

Method 1 compares our model total against the NHL scoring average of 6.16.

We currently require a difference of at least 0.50 goals to qualify as an under.

There is one additional rule:

The sportsbook total must be 5.5 or higher.

If the sportsbook is already hanging a 5 or lower, we’re not interested in playing the under.

Method 2 compares our projection directly against the sportsbook total.

We’re going to track both approaches and see what the data tells us.

UNDER EXAMPLE: Florida vs. Carolina — September 29

This is a perfect example of how the model works.

Florida starting goalie:

Markstrom
Season 1 GAA: 2.50
Season 2 GAA: 3.07

Average:

(2.50 + 3.07) ÷ 2 = 2.785

Carolina starting goalie:

Bussi
Previous-season GAA: 2.47

Because we didn’t have two meaningful seasons to use, we averaged his 2.47 with the NHL league average of 3.08:

(2.47 + 3.08) ÷ 2 = 2.775

Our model total:

2.785 + 2.775 = 5.56

Now compare that against the NHL average:

6.16 - 5.555 = 0.605

Our requirement is 0.50.

It qualifies.

The sportsbook total was 6.5, comfortably above our minimum requirement of 5.5.

UNDER 6.5 QUALIFIES

We can also look at our second method:

Sportsbook: 6.5
Model: 5.555

6.5 - 5.555 = 0.945

Again, we have a significant gap between the goalie model and the market.

The game finished 1-0.

Overs Are Even Simpler

For now, we’re handling overs differently.

If our combined goalie projection is 6.00 or higher, it qualifies as an over provided the sportsbook total is below 7.

We’re not requiring the same 0.50 differential that we’re using for unders.

Again, these aren’t rules handed down from a mountain.

They’re starting points that we’re going to track.

The data ultimately needs to tell us whether these thresholds make sense.

OVER EXAMPLE: Florida vs. San Jose — October 1

This is also a good example because it incorporates our adjustment for a goalie without two full seasons of meaningful data.

Florida:

Schmid
Last-season GAA: 2.59
League average: 3.08

Adjusted rating:

(2.59 + 3.08) ÷ 2 = 2.835

San Jose:

Askarov
Season 1 GAA: 3.10
Season 2 GAA: 3.63

Average:

(3.10 + 3.63) ÷ 2 = 3.365

Now combine the two:

2.835 + 3.365 = 6.20

Our model projects 6.20 goals.

The sportsbook total was 6.

Our current over rule requires our model to project at least 6.00 goals, with the sportsbook total below 7.

OVER 6 QUALIFIES

Again, notice what we’re doing.

We’re not trying to handicap every aspect of Florida and San Jose.

We’re asking one very specific question:

What happens if we isolate goalie performance and use that information to create our total?

Now we track it.

The Bigger Modeling Lesson

The most important part of this exercise isn’t whether this particular NHL model ultimately makes money.

It’s the process.

Find a statistic that logically correlates with what you’re trying to predict.

Look for something that you believe could be undervalued within the broader marketplace.

Create an objective way to calculate it.

Use a reasonable historical sample rather than relying entirely on one season.

Establish rules before you know the result.

Then track everything.

That’s how you begin finding out whether you actually have something.

The mistake would be looking at a 7-3 start and immediately declaring that we’ve discovered a profitable NHL betting system.

We haven’t.

Ten games is an extremely small sample.

But 7-3 gives us a reason to continue collecting data.

And that’s exactly what we’re going to do.

What Comes Next?

The next question is probably the most interesting one:

When do we stop relying primarily on the previous two seasons and begin incorporating 2026-27 results?

We don’t want to react to three or four games because the sample is too small.

But we also don’t want to continue treating two-year-old information as equally important once we have meaningful current-season data.

For at least the first couple of weeks, I prefer using the historical numbers.

After that, we need to determine when the current season has produced enough information to start changing the model — and how heavily that new information should be weighted.

That may be Part 2 of this experiment.

For now, we’re going to keep the model simple, track every qualifying play, and let the sample grow.

The goal isn’t to prove that the model works.

The goal is to find out whether it does.

Click here to see full google sheets model.

Instead of looking at individual teams, quarterbacks or coaches, we studied every NFL regular-season game from 2022 through 2025, concentrating on Weeks 1–8 and grading games according to the closing point spread.

Read more

The Week 5 college football model is a perfect example.

I already had the idea for the model and knew exactly what statistic I wanted to use. The tedious part was taking the data, matching it to every game on the board, pulling the correct sportsbook line, and calculating everything consistently.

That’s where I used ChatGPT.

And there’s a bigger lesson here: the exact same process can be used to build databases and backtest ideas in almost any sport.

Read more

Watch a college football game today and you’ll routinely see teams move the ball between the 20s. Quarterbacks complete easy throws. Offenses operate in space. Coaches are more aggressive on fourth down. Tempo creates additional possessions. And gaining 400 yards of offense isn’t particularly shocking anymore.

The natural reaction is to assume that the team moving the football most effectively is controlling the game.

But what if we’re looking at it wrong?

Because moving the football and actually accomplishing what matters are two very different things.

A team can dominate stretches of a game, put together long drives, accumulate impressive numbers and still walk away without accomplishing enough to cover the spread.

That’s something I’ve become increasingly interested in as this season has progressed.

Not because I’ve found some magic formula. And not because one particular style of football suddenly guarantees winners.

But because there appears to be a disconnect between what looks impressive during a football game and what ultimately matters when we’re trying to beat a betting line.

And once I started looking at college football from that perspective, I found something worth investigating.

Read more

Read more

Sometimes the best approach is simply identifying a statistical property of the sport, understanding why it matters, and then consistently applying it to the betting market.

I’ve talked about statistical properties quite a bit recently. To me, a statistical property is simply a fact about a sport—something supported by a large enough sample of historical data that it should influence how we handicap and model that sport.

And with the NFL, one particular statistical property continues to stand out.

What makes it interesting is how simple it is. There’s no complicated formula, no public betting splits, and no attempt to interpret every piece of line movement. It’s simply taking something the historical NFL data has repeatedly shown us and applying it to the current season.

Through the first two weeks of 2026, the results have been impressive.

Read more

First, here are the home favorites with at least a five-point differential via the model. I’m also including the line movement beside each team. A plus number represents positive line movement in our direction, while a negative number represents negative line movement.

Read more

Read more

The goal is not to react to wins, losses, covers or final scores. The goal is to determine whether the information we received from that week’s games gives us enough evidence to believe our current power rating is wrong.

Our normal maximum adjustment after a single week is plus or minus three points. A one-point move is a small correction. Two points represents meaningful evidence. Three points is a major adjustment.

There can be exceptions, but we need a very good reason to move beyond three points. Usually that means we now have multiple games providing consistent evidence that our original rating was significantly wrong.

Read more

Bet responsibly.

Don’t chase.

Manage your bankroll.

Know your limits.

All of those things are true. Ultimately, every person who places a wager is responsible for clicking the button.

But at some point, we have to ask a much more uncomfortable question:

What responsibility belongs to the sportsbooks, media companies and networks that are influencing those decisions in the first place?

We have created an industry where the overwhelming majority of recreational sports bettors lose money long term, while some of the largest media organizations in America simultaneously employ personalities to publicly recommend wagers.

And those personalities aren’t simply talking about sports.

They’re being presented as betting authorities.

That distinction matters.

A Real Example

I recently watched a video from ESPN sports betting analyst Erin Dolan.

She introduced the video by saying these were seven picks she liked for the NFL weekend and explained that some of the picks had already been submitted for shows she would appear on, including SportsCenter and Sunday NFL Countdown.

Erin Dolan’s “NFL Week 1 Best Bets” video discussed in this article. Screenshot used for commentary and criticism.

Her seven recommendations were:

• Trevor Lawrence over 4.5 rushing attempts

• Cam Ward over 31.5 passing attempts

• Josh Allen to throw an interception

• Tee Higgins anytime touchdown

• Dallas Goedert anytime touchdown

• George Pickens anytime touchdown

• Dallas Cowboys team total over 24.5

She even discussed combining three of the touchdown scorers into a parlay paying roughly 19-to-1.

The card went 2-5.

Trevor Lawrence had two rushing attempts.

Cam Ward attempted 32 passes and cleared his prop.

Josh Allen threw zero interceptions.

Tee Higgins didn’t score.

Dallas Goedert scored twice.

George Pickens didn’t score.

Dallas scored 20 points, falling short of its team total.

And let me immediately make something clear:

Going 2-5 means almost nothing when evaluating someone’s ability as a handicapper.

I’ve done this professionally since 2015. I’ve had terrible days. I’ve had terrible weeks. Every professional bettor has.

Seven wagers are a tiny sample.

So the 2-5 record isn’t really the point.

The process is.

What Exactly Makes Someone a “Sports Betting Analyst”?

ESPN identifies Erin Dolan as a sports betting analyst and puts her across some of the most recognizable sports programming in America.

That’s where I begin to have a problem with this—not personally with Erin Dolan, but professionally with what the title represents.

What qualifies someone to be presented as an authority on sports betting?

What is the long-term betting record?

What is the ROI?

How many wagers make up the sample?

At what average odds?

Are the selections independently tracked?

Does the person consistently beat closing numbers?

What methodology is being used to determine whether a wager actually has value?

Those are not personal attacks.

Those are basic questions that should be asked of anyone being presented to the public as an expert in an activity involving people’s money.

Because being knowledgeable about football is not the same as being a professional sports bettor.

Being good on television isn’t the same thing.

Finding interesting statistics isn’t the same thing.

And picking winners isn’t even the same thing.

Professional sports betting is ultimately about identifying prices that are wrong.

There Are Really Two Ways to Find an Edge

This is where I think the distinction becomes extremely important.

There are essentially two fundamental ways I know to consistently search for an edge against a sportsbook.

The first is bottom-up.

You build a model.

The second is top-down.

You use the betting market itself to identify potential inefficiencies.

Everything else should ultimately connect back to one of those two processes.

Bottom-Up: Build Your Own Number

A bottom-up approach means independently evaluating the underlying information and producing your own number.

If I’m handicapping an NFL side, my model might produce a point spread.

If I’m handicapping a total, it might project expected points.

If I’m betting a player prop, it might project attempts, yards, receptions or touchdowns.

If I’m evaluating something binary—such as whether a player scores a touchdown—I ultimately need a probability.

Then I compare my number against the sportsbook’s number.

That’s the entire point.

Suppose my work determines that an event should occur 60% of the time.

If the sportsbook’s price implies something closer to 52%, I potentially have an edge.

I’m not betting because I think something will happen.

I’m betting because I believe the sportsbook has incorrectly priced the probability of it happening.

There is an enormous difference.

Top-Down: Let the Market Tell You Something

The second approach is top-down.

Instead of attempting to independently price every variable from scratch, you’re studying the market itself.

Where did the number open?

Where is it now?

Which direction did it move?

How much did it move?

Did the market move despite public betting appearing to favor the opposite side?

Did respected sportsbooks move first?

Is there resistance at a particular number?

Is the current market telling me something that the opening market didn’t know?

Line movement is information.

I’ve spent a lot of time studying exactly this because the market itself can become part of your model.

You don’t necessarily have to know why a number moved to recognize that the movement may contain valuable information.

That’s top-down handicapping.

Bottom-up asks:

What should the number be?

Top-down asks:

What is the market telling me about what the number should be?

Both are legitimate ways to search for an edge.

Statistics Alone Are Not an Edge

Now go back to the seven recommendations in this video.

For Trevor Lawrence over 4.5 rushing attempts, we’re told that he averaged 4.4 rushing attempts the previous season and that kneel-downs could potentially push him over the number if Jacksonville was ahead late.

For Cam Ward over 31.5 passing attempts, we’re told that he averaged 34 attempts in one-possession games and 32 attempts overall, along with the expectation that the game would remain competitive.

For Josh Allen to throw an interception, we’re given Houston’s interception numbers, Allen’s interception numbers and information about how frequently his interceptions occurred under pressure.

For Tee Higgins to score a touchdown, we’re told about Cincinnati’s implied team total, how frequently Higgins scored when Cincinnati put up 28 or more points, and his touchdown production from the previous season.

Those are statistics.

There’s nothing wrong with statistics.

I use statistics every single day.

But statistics are inputs.

They are not automatically an edge.

What’s missing is the calculation connecting those statistics to the price being offered by the sportsbook.

Where Is the Probability?

This is the question I kept asking while watching the video.

Where is the true probability?

Take an anytime touchdown wager.

If a sportsbook is offering +135, that price carries a break-even probability of approximately 42.6%, ignoring the additional issue of bookmaker hold.

So telling me that a player scored 11 touchdowns last season isn’t enough.

Telling me that his team is expected to score 27 points isn’t enough.

Telling me that he has historically performed well in high-scoring games isn’t enough.

Those pieces of information might help me calculate his probability.

But I still need to calculate it.

If my model says his probability of scoring is 48%, and the market is effectively pricing him around 42.6%, now I have something.

I potentially have value.

But if my true probability is 38%, then +135 is actually a bad wager—even if I can find five impressive statistics supporting the player.

This is one of the biggest misunderstandings in sports betting.

You aren’t trying to predict what will happen.

You’re trying to determine whether the sportsbook’s price accurately represents the probability of what will happen.

The Sportsbook Already Knows the Statistics

And here’s another problem.

The sportsbook knows Trevor Lawrence’s rushing attempts.

The sportsbook knows Cam Ward’s passing attempts.

The sportsbook knows Josh Allen’s interception history.

The sportsbook knows Tee Higgins’ touchdown rate.

The sportsbook has access to the same publicly available statistics everyone else has—and considerably more sophisticated information and modeling behind its market.

You haven’t discovered an edge simply because you found a favorable statistic.

The real question is:

What do I know—or what have I calculated—that isn’t properly reflected in the current price?

That’s sports betting.

The question isn’t:

“Can I make an argument for this bet?”

You can make an argument for almost any bet.

The question is:

“Can I demonstrate that this wager is mispriced?”

That’s a completely different standard.

The Danger of Starting With the Bet

There’s another trap here that recreational bettors fall into constantly.

They decide they like something.

Then they start looking for statistics supporting it.

That’s backwards.

If I decide Josh Allen is going to throw an interception and then search long enough, I’ll eventually find statistics explaining why he might.

If I decide Tee Higgins will score a touchdown, I can find statistics supporting that too.

That’s confirmation bias.

A model is supposed to work in the opposite direction.

The information produces the number.

The number produces the potential edge.

The edge produces the wager.

Not:

Wager → statistics → justification.

That distinction is one of the biggest differences between professional handicapping and sports betting entertainment.

The Bettor Gets a Disclaimer. The Analyst Gets a Television Show.

And this brings me back to the larger issue.

Whenever the customer loses, we’re constantly reminded about personal responsibility.

Again, I agree with that.

But where is the corresponding responsibility for the industry influencing those wagers?

Sportsbooks advertise constantly.

Networks integrate betting into their programming.

Odds appear during broadcasts.

Analysts provide best bets.

Parlays are discussed.

Sportsbook prices are displayed.

The viewer is encouraged to participate in an activity where the overwhelming majority of participants will ultimately lose money.

And then we put a responsible gambling disclaimer underneath it.

At some point, responsible gambling has to involve responsibility from both sides of the transaction.

More Than 95% Lose—So What Exactly Are We Teaching Them?

Research into actual sports betting behavior has repeatedly demonstrated how difficult sustained profitability is. One frequently cited analysis of betting-related financial transactions found that fewer than 5% of sports bettors withdrew more from betting platforms than they deposited.

Different studies and estimates produce different numbers, which is why I wouldn’t pretend we know that exactly 95% or exactly 97% of every sports bettor loses.

But the larger point isn’t seriously debatable:

The overwhelming majority lose long term.

Think about the ecosystem we’ve created around that fact.

Sportsbooks make money.

Media companies make money.

Leagues make money.

Advertisers make money.

Content creators make money.

And the bettor overwhelmingly loses.

Then when that bettor loses, the industry’s response is:

You should have gambled more responsibly.

Maybe that’s only half the conversation.

This Isn’t About Erin Dolan Going 2-5

I want to emphasize this because otherwise the entire argument can be dismissed as somebody criticizing another person’s losing week.

I don’t know Erin Dolan personally.

I’m not saying she committed fraud.

I’m not saying losing picks make someone dishonest.

I’m not saying someone should be indicted because they went 2-5.

That would be ridiculous.

Her video is simply an example of a much larger problem I see throughout mainstream sports betting media.

Someone is presented as a betting analyst.

Statistics are provided.

Picks are given.

Odds are discussed.

Parlays are discussed.

The audience is told, “let’s make some money.”

But where is the demonstrated edge?

Where is the bottom-up model producing a probability different from the sportsbook?

Where is the top-down market analysis demonstrating why the current price may be wrong?

Where is the fair price?

Where is the expected value?

Those are the things that determine whether a wager makes mathematical sense.

Not whether somebody can find three statistics supporting it.

Why This Personally Bothers Me

I’ve been a professional handicapper since 2015.

This is how I make my living.

So yes, I find the way the term “sports betting analyst” is sometimes used insulting to the profession.

Not because I believe I’m entitled to that title and someone else isn’t.

It’s because I know what it takes to win long term.

I’ve built models.

I’ve thrown models away.

I’ve changed models.

I’ve studied thousands of betting lines.

I’ve tracked results.

I’ve studied statistical properties.

I’ve watched markets move against me.

I’ve watched them move toward me.

I’ve been right and lost.

I’ve been wrong and won.

And I’ve learned one of the most important lessons in sports betting:

The result of an individual wager tells you almost nothing.

The process tells you everything.

A professional doesn’t need to win tonight.

A professional needs a repeatable process capable of identifying positive expected value over thousands of wagers.

That’s a completely different occupation from giving out picks on television.

What Should the Standard Be?

If a major media organization wants sports betting content to be entertainment, that’s fine.

Call it entertainment.

If someone is giving opinions about which players might score touchdowns, that’s fine too.

Call them opinions.

But once a network presents someone as a sports betting authority, I believe a higher standard of transparency is reasonable.

Show the record.

Show the units.

Show the sample size.

Show the average price.

Show the methodology.

Show whether the analyst beats the closing market.

And most importantly, show us the process used to determine that the sportsbook’s number is wrong.

If you’re using a bottom-up approach, show us how you arrive at your number.

If you’re using a top-down approach, explain what you’re seeing in the market.

You don’t have to reveal every proprietary input.

I certainly don’t.

But there should at least be evidence that an actual methodology exists beyond collecting statistics and constructing a narrative around a wager.

Could This Eventually Become a Regulatory Issue?

This is where I think sports betting could become very interesting over the next decade.

I’m not predicting criminal indictments against Erin Dolan, ESPN or anybody else.

Losing bets aren’t fraud.

Being bad at handicapping isn’t illegal.

And none of the material here establishes that Erin Dolan or ESPN has committed a crime.

But I do think there is a legitimate question about where consumer-protection standards eventually go as sports betting becomes increasingly integrated with mainstream media.

What happens if regulators eventually start asking networks to substantiate claims about betting expertise?

What happens if they begin distinguishing more aggressively between entertainment, advertising and financial claims surrounding wagering?

What disclosure should exist when sportsbook relationships, media programming and betting recommendations intersect?

At what point does presenting someone as an authority create a responsibility to substantiate that authority?

Those are much more interesting questions than whether somebody went 2-5 on an NFL Sunday.

And I suspect we’re eventually going to have to answer them.

Accountability Has to Work Both Ways

Sports bettors absolutely need to accept responsibility for their decisions.

I’ve said that for years.

Nobody forces you to place the bet.

Nobody forces you to chase.

Nobody forces you to bet beyond your bankroll.

But an industry cannot endlessly promote wagering, normalize wagering, advertise wagering, integrate wagering into sports broadcasts, present personalities as betting authorities and then pretend that responsibility exists exclusively with the person holding the phone.

Responsibility has to work both ways.

The bettor has responsibility.

The sportsbook has responsibility.

The advertiser has responsibility.

And the network putting sports betting advice in front of millions of people has responsibility.

Especially in an industry where the overwhelming majority of customers ultimately lose.

If you’re going to present someone as an authority on beating one of the most efficient betting markets in the world, asking for evidence of an actual edge isn’t unreasonable.

It’s the minimum standard.

Because there is a massive difference between talking about sports betting and actually beating sports betting.

And maybe it’s time the sports betting industry started acknowledging that difference.

It doesn’t.

That becomes especially important early in the NFL season when quarterbacks change. Maybe you have a rookie making his first NFL start. Maybe you have an established starter with years of data. Or, as we have this week, maybe you have an older journeyman quarterback like Cooper Rush suddenly stepping into the lineup.

The goal isn’t to force every quarterback into the same calculation. The goal is to gather all the information available, understand the sample you’re working with, and then make an educated decision that allows the model to continue functioning properly.

That is exactly what we had to do with Cooper Rush.

Read more

The sample size is still small, so I’m not treating these numbers as permanent trends. But they’re significant enough that I don’t want to ignore them either.

Through the current sample:

Home teams: 63-41 ATS — 60.6%
Favorites: 66-38 ATS — 63.5%
Home favorites: 61-36 ATS — 62.9%
Away dogs: 36-61 ATS — 37.1%

That last number is especially important because we’ve been playing some away dogs with model value, and that group has been getting crushed.

Read more

The goal is to help you understand how we blend the statistical properties of the model with market information—particularly line movement, market efficiency, and when value is most likely to exist. The model gives us the data-driven foundation; understanding the market helps us decide how and when to actually use that information.

Read more

That doesn’t mean we have to wait a month before building an edge.

Over the last several weeks, I’ve been testing a separate approach designed specifically for the opening three weeks of the NFL season. The goal was to create a model that could begin with reliable historical information, gradually incorporate what we’re learning from the new season, and then get out of the way once enough current-year data exists to use a true statistical model.

The completed backtest went:

17-7 ATS
+6 units

More importantly, this isn’t a model that spits out 12 plays every Sunday. The backtest helped us define specific situations where the numbers actually justify a wager—and situations where the correct decision is simply to pass.

Below, I’m going to walk through exactly how I built it, how every number is calculated, how I handle rookies and quarterback changes, how the model evolves by Week 3, and when it actually generates a play.

Read more