CPL 2026: what the market saw, what our models missed

The season

CPL 2026 had a clear shape: Antigua & Barbuda Falcons and Guyana Amazon Warriors were the steadiest winners (both 8 wins, from 11 and 12 matches respectively), while Trinbago Knight Riders kept finding themselves on the wrong side of the key moments (3 wins from 10). Barbados Tridents (6 wins from 11) and Jamaica Kingsmen (6 wins from 13) lived in the middle—dangerous enough to beat anyone on a night, inconsistent enough to keep the table bunched.

The cricket itself was defined by extremes: brutal hitting at one end, and a handful of match-turning five-fors at the other. Dasun Shanaka produced the most violent innings in the dataset—121 off 43 balls at a strike rate of 281.4 with 12 sixes (St Kitts and Nevis Patriots v Saint Lucia Kings). Maaz Sadaqat did it twice, first with 115 off 58 (SR 198.3, 9 sixes) in Antigua & Barbuda Falcons v Jamaica Kingsmen, then 112 off 49 (SR 228.6, 8 sixes) in Barbados Tridents v Jamaica Kingsmen. Shimron Hetmyer’s 111 off 43 (SR 258.1, 9 sixes) for Trinbago Knight Riders v Guyana Amazon Warriors was a reminder that even in a season where TKR under-delivered as a team, their ceiling in a single innings was still terrifying.

The tournament’s best spells were just as decisive. Saint Lucia Kings had two of the standout bowling cards: Matthew Forde’s 5 for 23 (economy 5.75) against Jamaica Kingsmen, and his 4 for 29 (7.25) against the Falcons. Imran Tahir’s 5 for 17 (4.25) in Guyana Amazon Warriors v Saint Lucia Kings was another classic T20 heist spell—wickets in clumps, a chase or a platform suddenly feeling miles away. And Jeremiah Louis’ 5 for 10 (3.53) in St Kitts and Nevis Patriots v Saint Lucia Kings was the kind of burst that makes a match feel over early, even in a format that rarely allows certainty. Add in Daniel Sams’ 4 for 20 (5.0) for Jamaica Kingsmen v Barbados Tridents and Mohammad Nabi’s 4 for 22 (5.5) for Jamaica Kingsmen v Guyana Amazon Warriors, and you get a season where the biggest batting nights still had to survive elite spells.

Who the market misread

The most useful post-season question isn’t “who won?”—it’s “who did the market keep pricing wrong?” In CPL 2026, the clearest market underrate was Antigua & Barbuda Falcons. They played 11, won 8, and—crucially—were underdogs 7 times, winning 5 of those 7. That’s not a couple of coin-flips: it’s a meaningful sample (7 matches) of the Falcons repeatedly outperforming what the odds expected. They were favourites only 3 times (losing 1), so the story is less “dominant frontrunner” and more “team the market took too long to trust.”

Jamaica Kingsmen were the other big “priced as outsiders” side, but with a slightly different texture. They were underdogs in 11 matches and won 5 of them (11 is a strong sample), and they were never listed as favourites in this dataset (0 favourite matches). That says the market treated Jamaica as a persistent long shot, even though they finished with 6 wins from 13. They weren’t secretly the best team—but they were good enough, often enough, that the constant underdog price looked too harsh.

Saint Lucia Kings (10 played, 4 won) are the tricky case that shows why this exercise matters. They were underdogs 6 times and won 2 (sample size 6 meets the threshold), and favourites 3 times and lost 1. Overall, that’s not a “market disaster” record—but Saint Lucia’s wins tended to come in the kind of games the market least expected (see the upsets section), which is exactly how a team can feel ‘misread’ even with a modest win total.

On the flip side, Trinbago Knight Riders were the season’s clearest market overrate. They were favourites 8 times and lost 5 of those 8. That’s enough matches to call it a pattern: the market kept installing TKR as the likelier winner, and TKR kept not cashing that expectation. They won only 3 matches all season (10 played), and they didn’t win a single game as an underdog (0 wins from 2), so there was no hidden “value identity” to balance it out.

Two other teams sit in the ‘overrated’ list but with more nuance. Barbados Tridents were favourites 6 times and lost 2 (a 6-match sample), and they won 0 of 3 as underdogs (but 3 is below the minimum sample of 4 for team splits, so that underdog note should be treated cautiously). Guyana Amazon Warriors won 8 of 12, so they weren’t a bad team at all—but the market still overrated them in the sense that they were favourites in 10 matches and lost 2, and—more importantly for our own work—our models made them favourites 11 times and got burned 4 times in those spots. That “model overrate” tag matters because it shows where our numbers leaned too hard into Guyana’s baseline strength and not enough into matchup/context variance.

The upsets

The biggest market-beaters clustered around two themes: Saint Lucia Kings landing clean punches as outsiders, and the Falcons doing the same with bigger margins. On 9 Aug, Saint Lucia Kings beat Antigua & Barbuda Falcons by 4 runs when the market gave them just a 32% chance (odds 2.9). That’s the classic T20 upset profile: a tight finish, a small margin, and a pre-match price that said Saint Lucia needed plenty to go right—then they found it under pressure.

On 26 Aug, Saint Lucia did it again, this time with far more daylight: they beat Trinbago Knight Riders by 36 runs with a market probability of 33% (odds 2.819). A 36-run win isn’t a scramble; it’s a statement, and it fed directly into the broader season-long theme that TKR were being priced like a powerhouse while bleeding losses as favourites.

Then came the most emphatic of the lot: on 8 Sep, Antigua & Barbuda Falcons crushed Guyana Amazon Warriors by 65 runs as a 34% chance (odds 2.775). In a season where both teams finished on 8 wins, that margin is the sort of result that resets perceptions. It also underlines how the Falcons’ “underdog wins” weren’t all last-ball chaos—some were dominant, match-shaping performances that the market simply didn’t bake in strongly enough.

How the methods did

This is the blunt part of the review: the bookmaker market was better than all four of our methods on both accuracy and returns in CPL 2026. The books went 21 wins from 34 (62% accuracy) and finished +3.9% ROI (+1,311). That’s the benchmark: not perfect, but profitable and clearly ahead of the pack.

Our best-performing method by ROI was the Maths model at -15.7% (profit -5,176) with 15/33 correct (45%). AI 1 was 16/34 (47%) with -16.5% ROI (-5,595). AI 2 was 15/33 (45%) with -22.1% ROI (-7,295). The ML ensemble struggled most: 12/32 (38%) and -36.7% ROI (-11,750). Across the board, the story isn’t “we were close but unlucky”—the accuracy gap to the market (roughly 15–24 percentage points depending on method) is too large to hand-wave away in a 30+ match sample.

The split that really explains the pain is what happened when our models tried to take a stand against the market. In the 15 matches where our four methods aligned with the market consensus, the record was 8 wins (53%) but still -17.1% ROI (-2,561). In the five matches where the models went against the market, the result was 0 wins from 5, -100% ROI (-5,000). Five matches is exactly the minimum sample threshold for a split here, so it’s just big enough to note—but still small enough that it should be read as a warning sign rather than a law of nature. What it does show, clearly, is that our strongest disagreements with the market were poorly calibrated this season.

What it suggests

One season is one sample, and CPL is a high-variance format by design—so the right takeaway isn’t a sweeping verdict, it’s a careful one. Still, CPL 2026 offers two clean lessons from the numbers we have. First, the market’s team priors looked sticky: it kept leaning into Trinbago Knight Riders (8 times favourites, 5 favourite losses) and kept discounting Antigua & Barbuda Falcons (7 times underdogs, 5 underdog wins) for long enough that the mispricing became a season-long pattern.

Second, our models didn’t just lose to the market overall; they lost worst when they tried to be contrarian (0/5 when opposing the market). That points to a calibration problem in the “confidence to disagree” layer—either the inputs didn’t capture what the market knew, or we overweighted signals that didn’t translate into match wins in this season’s conditions. The honest read is that the market set a high bar in 2026, and our methods didn’t clear it.