René Núñez
Data Analyst. I build probabilistic forecasting models for sports and benchmark them against the market in public.
Get in touch with feedback, ideas, or anything else: renenunezgalaviz@gmail.com
Models
Hierarchical Bayesian model simulating every game one plate appearance at a time.
Brier
0.247
Coin flip
0.250
Runs MAE
2.50
2352 games graded
Weekly spread, total, and moneyline predictions with frozen lines, built on power ratings for every Division 1 program.
Margin MAE
12.61
Market MAE
11.99
Gap to market
+0.62
748 games graded
Bayesian power ratings for all 32 teams from drive-level EPA, with weekly spread and total projections priced against the market.
Margin MAE
9.76
Market MAE
9.78
Gap to market
-0.02
79 games graded
Poisson goal model from last-25-game shot-quality windows, priced daily against the NHL partner sportsbooks with frozen moneyline and total picks.
Log loss
0.662
Coin flip
0.693
Total MAE
1.89
107 games graded
Today's predictions
Thursday, Oct 15Times shown in Eastern Daylight Time (EDT).
Latest writing
All posts →2026-05-27
The Totals Problem
v2 beat v1 on moneyline and run line, but totals kept bleeding. Three fixes from inside the same modeling frame, none of them moved the number, and the question I probably should have been asking earlier.
2026-05-03
V2 Is Coming and It's Bringing Bayes with It. Our Findings on XGBoost.
One month of live predictions with a 14-feature XGBoost regressor. What held up, what didn't, and why I'm going hierarchical Bayesian for v2.