How the Model Works
A logistic regression model trained on precursor award results from the past 30 years. Every number on this page was produced by testing the model on years it never trained on.
Historical accuracy
Loading accuracy metrics…
Why these numbers hold up: I use leave-one-year-out cross-validation. To produce a prediction for, say, 2015, the model trains on every other year from 2014 back to 1997, with 2015 withheld entirely. That predicted winner is then compared to the actual Oscar winner. This process is repeated for all 30 years in the modern era. The hit rate you see above is the fraction of those held-out years where the model's top pick won. No year ever influences its own prediction.
What these metrics mean
- Hit rate
- The fraction of Oscar ceremonies where the model's top-ranked nominee actually won. Across Best Picture, Best Director, and the four acting categories in the modern era, the model picks correctly — of the time. Human expert consensus (sites like GoldDerby) typically lands in the 80–88% range on these same categories. The two screenplay categories are excluded from this figure — precursor signal for writing is weaker, and their hit rate runs meaningfully lower (see the per-category breakdown below).
- AUC-ROC
- Measures how well the model ranks the true winner above every other nominee, across all predictions, all 8 categories. 1.0 = perfect ranking every time; 0.5 = random. The modern-era AUC of — means the model places the actual winner above a randomly drawn non-winner well above chance.
- Brier score
- The mean squared error between each nominee's predicted probability and their actual outcome (win = 1, loss = 0), across all 8 categories. Lower is better; 0 = perfect. The modern-era Brier of — indicates well-calibrated probabilities. When the model says a nominee has a 70% chance, that nominee wins roughly 70% of the time.
Hit rate by category
The headline numbers above are averages, and averages hide variance. Some categories are genuinely harder to call than others — the writing categories in particular have much weaker precursor coverage than acting or directing, and it shows in the numbers. This is the full breakdown, all years each category has been tracked, no cutoff.
| Category | Hit rate | Years tested |
|---|---|---|
| Loading… | ||
The core insight: precursor awards dominate
Oscar forecasting is well-studied, and the single most predictive signal is precursor awards, especially the guild results. Box office, campaign spend and critic scores are all weaker. The Producers Guild (PGA) is the strongest single predictor of Best Picture. The Directors Guild (DGA) for Best Director. SAG for acting. BAFTA, the Golden Globes, NYFCC, and LAFCA add further calibration.
This is the well-established "Oscarmetrics" finding, replicated consistently over 30 years. My model leans into it. Precursor wins are the dominant features in the Stage B winner model. The reason guilds carry this much weight is structural: their voting bodies overlap heavily with the Academy branches that hand out the equivalent Oscar, which How the Academy Votes covers in detail. Earlier-season features (critic scores, festival premieres, career pedigree) provide signal when precursor results don't exist yet.
Three-stage model for year-round coverage
A single model cannot serve the full calendar, because the signals available in August are completely different from the ones available in February. I use three stages, each adding signal to the one before it.
What is being predicted? Stages 0 and A both show nomination odds, the probability a film gets nominated at all. Stage B switches to win odds among confirmed nominees. There is no toggle. The switch happens automatically on January 20, which sits right on top of nominations day (January 21 for the 99th Oscars). Win odds before the nominees are known would be meaningless, so the page does not show them.
Pre-season
Apr–Sep
Features: career pedigree (prior Oscar noms/wins per director/actor/writer), festival premiere (Cannes/Venice/Telluride/TIFF) including whether the film won a festival prize, release-window timing (Oct–Dec films have a strong historical advantage), genre, early critic scores, box office where a film has opened, original language, and the distributor's own awards track record.
Uncertainty: a dot-meter (● ○ ○) shows low confidence, with six or more months of awards season still to come. Confidence rises in Stage A as guilds announce.
Nominations
Oct–Jan 19
Added features: precursor nominations (Globe noms, Critics' Choice noms, SAG noms, guild eligibility decisions, NBR selections) on top of all Stage 0 features.
Uncertainty: dot-meter shows medium confidence (● ● ○). Advances to high confidence (● ● ●) in Stage B once Oscar nominations are known.
Winner
Jan 20–Mar
Features: precursor wins from PGA (Best Picture), DGA (Director), BAFTA, SAG, Golden Globes, Critics' Choice, NYFCC and LAFCA, plus total Oscar nominations for the film and all earlier features. The model scores only confirmed nominees; probabilities sum to 100% per category.
Uncertainty: ±10pp. Confidence is high once the major guilds have voted, though single-digit bands would overstate the model's precision. The band is validated against how often these probabilities have actually landed in past ceremonies.
Model structure
Each category uses an L2-regularized logistic regression
(LogisticRegressionCV from scikit-learn, optimized for log-loss calibration).
I train on log-loss instead of AUC because AUC only rewards ranking. Optimizing for it
collapses probabilities toward the base rate and produces overconfident outputs in a
five-nominee field. Log-loss penalizes miscalibration, which keeps the percentages honest.
Training data: Oscar nominees and winners from 1997 to the most recent ceremony (~30 years, ~150 nominees per category). The feature set is deliberately small and strongly regularized to avoid overfitting a small sample.
Era weighting: the model applies exponential recency decay when training, so recent years carry more weight than older ones. A ceremony from 15 years ago gets 50% weight; one from 30 years ago gets 25%; the most recent year gets 100%. This reflects a real pattern. The Oscar race today, shaped by SAG, DGA, PGA and a dozen critics' guilds, looks quite different from 1997, and fitting the modern structure honestly requires prioritizing modern data.
Features by stage
| Feature | Stage | Notes |
|---|---|---|
| PGA win | B | Strongest Best Picture predictor · data 1995–present |
| DGA win | B | Strongest Director predictor · data 1997–present |
| SAG win | B | Strongest acting predictor · data 1995–present |
| BAFTA win/nom | B / A | Best Picture, Director, Actor, Actress · data 1997–present |
| Golden Globe win/nom | B / A | Drama + Comedy categories · data 1944–present |
| Critics' Choice win/nom | B / A | Best Picture, Actor, Actress · data 1997–present |
| NYFCC win | B | New York Film Critics Circle · Actor, Actress · data 1941–present |
| LAFCA win | B | LA Film Critics Association · Actor, Actress · data 1976–present |
| Total Oscar nominations | B | Best Picture only; films with 10+ noms win most often |
| Rotten Tomatoes / Metascore | 0 / A / B | From OMDb API (free tier) |
| Person Oscar pedigree | 0 / A / B | Prior noms/wins per director/actor/writer |
| Festival premiere tier | 0 / A | Telluride (tier 3) > Cannes/Venice/TIFF (tier 2) > Sundance/Berlin (tier 1) |
| Late-year release | 0 / A / B | Oct–Dec release indicator |
| Distributor track record | 0 | A24/Neon/Searchlight historical Oscar rate |
| Genre | 0 | Drama indicator |
Data sources
- Oscar nominees/winners (1997–present): hand-curated from Wikipedia per-ceremony pages into an internal CSV.
- Precursor awards (SAG, PGA, DGA, BAFTA, Golden Globes, Critics' Choice, NYFCC, LAFCA): hand-curated from Wikipedia per-award-year pages. No public API exists for any of these.
- Film metadata (cast, crew, genre, runtime, budget, revenue, production companies): TMDB API (free). Attribution required; see below.
- Critic scores + domestic box office: OMDb API free tier (Rotten Tomatoes %, Metacritic, US domestic gross).
- Festival premieres: hand-curated from Wikipedia per-festival-year pages.
IMDb data is not used. IMDb's free datasets are explicitly non-commercial; their commercial license is enterprise-priced. On a Stripe-monetized site, using IMDb data is a licensing risk. TMDB + OMDb cover the same metadata legally.
Pedigree as an early-season anchor
When no precursors exist (May–September), the strongest signal available is career pedigree, meaning how many Oscar nominations and wins the director, actors and writers attached to a film have accumulated over their careers. Pedigree counts nominations in any category: an actor who has a writing nomination in their history is a more established industry figure than someone without.
Festival premieres add a second early anchor. A Palme d'Or or Grand Prix winner at Cannes doesn't reliably predict a Best Picture win (the Academy and Cannes have notoriously different tastes), but a Cannes Competition premiere does correlate with Oscar consideration, and the model learns that distinction from the historical data. Venice and Telluride have historically been more predictive of Oscar performance than Cannes.
Every signal here is correlational. Nothing in the model claims a festival slot causes a nomination. A film with a first-time director, no festival premiere and a November release date can absolutely get nominated, and the pre-season model will simply give it low confidence and a modest probability.
Caveats
- Training N is small, roughly 30 winners per category. I lean heavily on regularization and pool related categories to compensate.
- Pre-season (Stage 0) nomination odds are inherently wide, landing six to eight months before any precursor exists. The dot-meter confidence indicator sits next to every probability to be honest about that.
- Precursor results are treated as three-way rather than binary. A guild that has not announced yet is handled differently from one that announced and passed a contender over. This keeps early-season frontrunners from reading as artificially underconfident.
- Pedigree features (prior Oscar noms/wins) are computed as of each training year, so a nominee's future nominations never leak into how their pedigree looked at the time being predicted.
- Best Actress has shown genuine volatility recently, with three consecutive upsets in 2023, 2024 and 2025 where heavy precursor favorites lost. No statistical model would have caught those.
- Oscar voting is human and idiosyncratic, so any model carries an irreducible error floor.
This product uses the TMDB API but is not endorsed or certified by TMDB.