.png){kind=spacer}
A model hitting 78% of its picks should be printing money. This one isn’t.
One of our NBA moneyline models — Closeout Rater — was hitting at a 77.6% clip across its first 49 settled picks. It went 8-for-8 over the last two days. And its return on investment? Just 1.2%.
Not a typo. Not a bug.
It's actually a perfect illustration of why NBA moneyline is one of the hardest markets in sports betting to beat, and why win rate alone tells you almost nothing about profitability.
High win rate doesn’t mean high profit
Most betting content trains you to think in outcomes:
- Did the pick win?
- How often are they right?
But that’s not how betting actually works.
Winning picks ≠ valuable picks.
The goal isn’t to pick winners. The goal is to find bets where the odds are wrong.
The moneyline math problem
Here’s the part most bettors never think about:
When a model picks an NBA favorite, those odds are heavily juiced. The sportsbook has baked in that team’s probability, and added their cut (the vig) on top.
That means even when you’re right, the payout is small — sometimes very small.
Here's what it looks like:
Look at that bottom row.
At -400 odds, you need to win 80% of the time just to break even.
Even if your model is genuinely better than the market — say 82.5% — your edge per bet is tiny. One loss wipes out multiple wins.
This is the trap for NBA moneyline:
The more “certain” the pick, the less it pays.
Where the real edge comes from
At Moddy, models don’t just predict winners. They estimate probabilities, then compare those to the implied probabilities in the odds.
A pick only fires when there’s a gap. That gap is the edge.
Not “who wins,” but where the market is underpricing a team.
The problem? In NBA moneyline, those gaps are razor-thin.
The NBA is one of the most liquid, most analyzed, most efficiently priced markets in all of sports. Oddsmakers, syndicates, and sharp money are all working with massive datasets and sophisticated models.
By the time a line is posted, most of the value is already gone.
So when Closeout Rater finds a 2-3% edge, that's a real accomplishment. But a 2-3% edge on heavily juiced favorites translates to pennies on the dollar.
The model is doing its job. The market just isn’t leaving much on the table.
The metric that separates skill from luck
So how do you know if a model is actually good, or just running hot? This is where Moddy’s Performance Ratio becomes essential.
Performance Ratio compares actual profit to expected profit:
- ~1.0 → the model is calibrated (results match expectations)
- Above 1.0 → possible luck
- Below 1.0 → possible variance working against it
Before each pick, the model calculates an expected value based on its predicted edge. After the pick settles, we compare what actually happened to what the model thought would happen.
Closeout Rater currently sits at -0.12x.
On the surface, that looks concerning. But zoom out:
- Sample size: 49 picks
- Estimated edge: +2.1%
- ROI: positive
- Win rate: strong
At this sample size, variance dominates. One bad loss at heavy odds can swing results significantly.
49 picks isn’t enough to judge a model. But it’s enough to see if it behaves the right way.
Where the model actually shines: bet strength
One of the most telling signals is how the model performs at different confidence levels.
Closeout Rater categorizes picks as Strong or Normal based on the size of the predicted edge:
This is exactly what you want to see.
When the model identifies a strong edge, it crushes:
- It shows a 91.7% win rate
- It generates real profit
- Its performance matches expectations
When the edge is weaker:
- It still wins more often than it loses
- But the payouts aren’t enough to overcome losses
(That's not a flaw in the model. It's the market doing what efficient markets do.)
This is the signal: when the edge is real, the model doesn’t just win — it gets paid.
49 picks is a snapshot, not a verdict
We want to be upfront about this: it takes hundreds — often thousands — of bets to fully evaluate a model.
What we can say now:
- The model is identifying edges
- It’s correctly separating high vs. low value spots
- Its strongest signals are performing exactly as expected
- Its highest-conviction picks are delivering real returns
Where the long-term ROI settles will become clearer over time.
We'll keep showing the numbers either way.
The takeaway
A 78% win rate with a 1.2% ROI isn't a failure.
It’s reality in a sharp market.
The goal isn’t to pick winners. It’s to find value the market missed.
Most tout services would package this model as "nearly 80% winners!" and sell it as a slam dunk. We'd rather show you the full picture — win rate, ROI, performance ratio, edge, bet strength — and let you decide what it means.
Because at Moddy, we believe the sports betting space needs less hype and more transparency. Models are tools, not crystal balls. And understanding what they're actually telling you is the difference between being a bettor and being an informed one.
Want to see which models are finding real edge right now? Check out moddy.ai and see the numbers for yourself.
