When trading NBA games on Polymarket, perhaps you, like many others, have had this experience: before the game, you see one team's win probability is significantly higher than their opponent's, only for them to collapse in the fourth quarter and get swept away (like the recent Hornets and Heat game—I lost so much on that bet I started questioning my life).
Since everyone says Polymarket is a "truth machine," does that mean I can just blindly bet on the team with the higher pre-game win probability and easily make money?
To test this hypothesis, I backtested the 1,096 regular-season games of the NBA 2025-26 season. The data reveals the truth—
Blindly following the market won't make you money, but you won't lose much either; the pre-game probabilities are fully priced in.
Blindly Buying with the Market Guarantees a Loss
The backtesting strategy was very simple:
- Used the average probability from 3 minutes before the game as a benchmark
- Traded $100 per game
- Always bought the side with the "higher win probability"
Results:
- Total amount spent: $109,600; total amount returned: $107,545.2; net loss: $2,054
- ROI: -1.87%
This shows that Polymarket's prices are quite efficient; the market has fully priced in the teams' win probabilities, leaving no "arbitrage" opportunity.
The difference in ROI likely comes from other dimensions like transaction costs and emotional premiums. If you insist on "buying blindly," you might as well bet against the market; that would yield a 1.87% profit.
The Real Value: Not All Teams Are Created Equal
The above backtest was for the overall sample of a thousand games. I then broke it down from multiple angles to try and find parts that break free from the market's gravity:
- By week: Random walk
- By probability: Still a random walk. That is, betting on pre-game win probabilities of 50%, 60%, versus 70%, 80% showed no difference in returns.
- By team: Here, clear differences emerged.
Some teams live up to the market's trust—
When the market thinks they will win, they are more likely to actually win.
- POR (Trail Blazers): ROI 19%
- PHI (76ers): ROI 14%
- SAS (Spurs): ROI 12%
- LAL (Lakers): ROI 11%
- CHA (Hornets): ROI 9%
Why do these teams show such differences? As the author previously had little knowledge of NBA teams, I first had a hypothesis:
Are they the strongest or the weakest teams, thus having high expectation consistency?
But upon checking, the facts proved otherwise. Except for SAS (Spurs), the other four teams are only ranked in the middle to slightly above average positions.
What about the teams with the best records? Actually, the market has already fully priced them in. The average ROI from blindly buying their higher probability is only 2.16%; the pre-game betting probabilities contain no水分 (water).
- DET (Pistons): ROI 1%
- BOS (Celtics): ROI 4%
- NYK (Knicks): ROI 3%
- OKC (Thunder): ROI -2%
- DEN (Nuggets): ROI -5%
What about the weakest?
Here, there's extreme divergence. These teams hardly ever have games where the market favors them to win. For example, the Nets (BKN) only had 7 games with a win probability greater than 50%, won 5 of them, resulting in an ROI of 21%; whereas the Pacers (IND) had only 8 games greater than 50%, won 4, but had an ROI of -20%. The sample size is too small to serve as a trading reference.
This means, theoretically (only theoretically!), POR (Trail Blazers), PHI (76ers), SAS (Spurs), LAL (Lakers), and CHA (Hornets) are the range划定 (delineated) by the existing data for following.
