Who is the Best at Predicting Oscar Winners?
I love Oscar predictions. It makes the whole Oscar race more fun when you’ve got some personal investment. Here’s the thing, though. My strategy has always been just to copy what other, more informed people have already guessed. But who to copy?
In this article I check the receipts from a couple of the top Oscar pool pundits and see who has the clearest crystal ball.
The Leaderboard
I compiled 970 individual predictions from eight major Oscar pundits across the last five ceremonies (2021-2025). The purpose of this article is purely selfish. If I’m going to win, I’ve got to copy the best.
After compiling all the predictions and checking them against the result, Scott Feinberg emerges as the leader, with predictions better than anyone else. So, I’ll just copy him. Problem solved!
Not so fast.
Wait. Is That Gap Even Real?
Eight points sounds meaningful. Over 100+ predictions, Feinberg gets seven more right than Davis. But could this ranking be entirely random?
The problem is that all eight pundits are predicting the same categories. When Chadwick Boseman loses to Anthony Hopkins, every single pundit gets that wrong together. When Best Director goes to the obvious favorite, every pundit gets it right together. The outcomes are deeply correlated; the average pairwise correlation between any two pundits’ results is 0.46. That’s enormous. It means that the 102 shared predictions they’re being scored on behave more like 24 independent trials than 102. And with only 24 effective coin flips, an 8-point spread between your luckiest and unluckiest is completely expected.
To test this formally, I ran a permutation test under the null hypothesis that all eight pundits have identical underlying skill. Under this null, I shuffled which pundit “owns” which outcome within each category-year 100,000 times, preserving the correlation structure, and measured the max-min spread each time. The p-value is 0.48, meaning 48% of random shuffles produce a gap this large or larger. The leaderboard is statistically indistinguishable from what you’d get by randomly assigning the same outcomes to different names.
Year by year, the gap evaporates completely. Feinberg only outright beats Davis in 1 of 5 years (2025). They tie in three years. Davis wins one. Feinberg’s entire edge comes from his perfect 2025. Remove that single year and the ranking reshuffles.
The Consensus Bot
Wait, so if no one pundit is meaningfully more skillful at picking winners, is the best strategy to just go with the crowd, guess on the rest? I built a bot to find out.
The bot averages 81.8% accuracy across 50,000 simulations. That’s higher than every single pundit on the leaderboard. It beats the #1 ranked pundit 93% of the time.
This doesn’t mean the pundits are bad. It means the opposite. They’re so good at reading the same signals that 95% of categories have a clear majority consensus. The consensus itself is the valuable product. Once you have the consensus, individual pundit “skill” is just who happens to guess right on the 8% of categories where the field splits. Feinberg isn’t measurably better than Davis. Davis isn’t measurably worse than Feinberg. They’re all reading the same information, arriving at the same answers, and occasionally diverging on dice-roll categories where neither has a real edge.
The Categories That Are Already Decided
So, which categories are already decided and which ones are coin flips?
Best Director and Best Supporting Actor are 100% correct across 43 predictions each. Every pundit, every year, nailed them. By the time final predictions are published, these categories have been decided by guild wins, precursor awards, and critical momentum. The Oscar ceremony is a formality.
Best Actress is the opposite. Across five years, the collective got it right 48.8% of the time. Worse than flipping a coin. Best Documentary Short is at 47.4%.
In 2021, four of eight pundits picked Carey Mulligan. Frances McDormand won (though three pundits got it right). In 2024, seven of nine picked Lily Gladstone. Emma Stone won. In 2025, seven of nine picked Demi Moore. Mikey Madison won. The consensus got it right in 2022 (Jessica Chastain) and 2023 (Michelle Yeoh), but those two correct years don’t make up for the chaos of the other three.
The Year That Broke Everyone
The same pundit can look like a genius one year and a coin-flipper the next.
In 2022, the 94th ceremony was a layup. Most pundits cruised to 80%+ accuracy. The only real surprise was CODA over The Power of the Dog for Best Picture, and even that was well within the range of plausible outcomes.
In 2025, the floor fell out. Clayton Davis hit 56.5%; his worst year by nearly 9 points, missing 10 of 23 categories. Pete Hammond dropped to 60.9%. It was the year of upsets. Best Actress (Mikey Madison), Best Animated Feature (Flow over The Wild Robot), Best Picture (Anora over Conclave), multiple short-film categories.
And then there’s Feinberg, who went 20 for 20 on his scored categories in 2025. Perfect. While everyone else was drowning, he called every single one. That’s either extraordinary skill or extraordinary luck, and with a sample size of one perfect year, I can’t tell you which.
The Herd Effect
Here’s the pattern that jumped out when I scanned the misses. Pundits don’t fail independently. They fail together.
In 2021, all eight pundits picked Chadwick Boseman for Best Actor. Anthony Hopkins won. That’s not eight independent failures; it’s one collective failure repeated eight times. The same thing happened with Best Actress 2024, Best Animated Feature 2024, and Best Cinematography 2021. When the consensus signal is wrong, everybody follows it off the cliff.
This has a practical implication. The marginal value of reading a second pundit is low. If Feinberg and Hammond agree on a pick, Thompson and Davis almost certainly agree too. And if they’re all wrong, they’re all wrong because the precursor signals pointed one direction and the Academy voted another.
The pundits are a herd, not a panel. Their errors are correlated because their information is shared.
What This Means
The best Oscar forecasters in the business are right about 80% of the time. That’s genuinely impressive for a 23-category forecast, but the ranking among individual pundits is not statistically meaningful. A simple consensus bot that takes the majority pick and randomly guesses on the rest beats every single pundit 93%+ of the time. Copying any pundit’s picks is a solid strategy. Copying the consensus is better.
About 70% of the ceremony is effectively decided before the envelopes are opened. Best Director, Best Supporting Actor, Best International Feature, and Best Supporting Actress are these are locked in by the time the pundits file their columns. The remaining 30% is where the drama lives.
Best Actress is the chaos category. Three upsets in five years. The precursor signals that work everywhere else break down here. If you want to win your pool, don’t waste contrarian energy on Best Director. Spend it on Best Actress and the shorts.
Year matters more than pundit. The same forecaster can go from 91% to 61% across two ceremonies. The Academy’s predictability varies more than any individual pundit’s skill. Hard years humble everyone.
Reading one pundit is enough. Their predictions are so highly correlated that the ninth pundit you follow adds almost no information beyond the first. When the consensus is right, everyone is right. When it’s wrong, everyone is wrong together.
Appendices
What Does the Herd Say About 2026?
The 98th Academy Awards are tomorrow night. I collected 2026 predictions from the same pundits and ran them through an accuracy-weighted consensus model. Each pundit’s vote is weighted by their 5-year historical accuracy, so Feinberg’s picks count ever so marginally more than Davis’s.
19 of 24 categories are unanimous or near-unanimous. The pundits agree on almost everything this year. The consensus ballot looks like a two-film sweep. One Battle After Another takes the prestige haul (Picture, Director, Adapted Screenplay, Supporting Actor, Editing, Cinematography) while Sinners takes the craft and auteur prizes (Original Screenplay, Score, Casting, Actor). Frankenstein runs the table on Production Design, Costume Design, and Makeup.
Best Animated Short is a genuine toss-up. Feinberg and Hood pick The Girl Who Cried Pearls, Hammond and Anderson pick Butterfly. This is exactly the kind of category that historically breaks everyone’s bracket. The rest of the variance categories lean moderately in one direction but have a credible dissenter.
Notable: Clayton Davis is the only pundit picking Sinners for Best Picture and Ryan Coogler for Best Director. Feinberg is the only one picking Delroy Lindo over Sean Penn for Best Supporting Actor. Zauzmer is the only one picking Teyana Taylor over Amy Madigan for Best Supporting Actress. If any of these contrarians are right, they’ll gain ground in the all-time rankings. If they’re wrong, the herd wins again.
Methodology
Data. 970 individual category predictions across 9 pundits and 5 ceremony years (2021-2025). Every prediction sourced from a published, dated, pre-ceremony article with a verifiable URL. Source links for all 45 pundit-year articles are documented in the repository.
Coverage. Most pundits predicted all 23 categories across all 5 years (115 total). Some have partial years where short-film categories could not be extracted from the source article. Totals reflect only scored predictions.
The Consensus Bot’s 2026 Ballot
The bot’s methodology is simple: take the majority pick for each category. For the two categories where pundits split (Best Animated Short, Best Live Action Short), randomly pick among the options. Here’s the ballot it would submit tonight:
Best Picture: One Battle After Another · Best Director: Paul Thomas Anderson · Best Actor: Michael B. Jordan · Best Actress: Jessie Buckley · Best Supporting Actor: Sean Penn · Best Supporting Actress: Amy Madigan · Best Original Screenplay: Sinners · Best Adapted Screenplay: One Battle After Another · Best Animated Feature: KPop Demon Hunters · Best International Feature: Sentimental Value · Best Documentary Feature: The Perfect Neighbor · Best Casting: Sinners · Best Cinematography: One Battle After Another · Best Film Editing: One Battle After Another · Best Production Design: Frankenstein · Best Costume Design: Frankenstein · Best Makeup and Hairstyling: Frankenstein · Best Original Score: Sinners · Best Original Song: Golden (KPop Demon Hunters) · Best Sound: F1 · Best Visual Effects: Avatar: Fire and Ash · Best Animated Short: The Girl Who Cried Pearls* · Best Documentary Short: All the Empty Rooms · Best Live Action Short: Two People Exchanging Saliva*
*Dice roll — no clear majority.
After tomorrow night, we’ll have one data point. To know whether the bot is as good as this analysis says it is we’ll need roughly 10 more years of ceremonies. With 23 categories per year and the high correlation structure we’ve documented, it takes about 230+ additional predictions before a ~2 percentage point edge clears significance. Check back in 2036.
Vibe caveat
This article came together really fast, and it is based on much more vibe coding than usual. I’ll double check everything and edit, but this has to come out before the ceremony or else we won’t know if the bot works until 2037!