I Added Park Factor Correction to My NPB Bayesian Prediction Model — Backtest Validation & 2026 Forecast

Introduction In a previous project, I built an NPB (Nippon Professional Baseball) standings prediction system combining the Marcel method with Stan Bayesian regression . GitHub : npb-bayes-projection Previous article : Beyond Marcel: Adding Bayesian Regression to NPB Baseball Predictions This time, I added park factor (PF) correction to the team simulation component, validated it against 8 years of historical data (2018-2025), and updated the 2026 forecast. Why Park Factor Correction? Marcel projections are based on each player's past 3-year stats — but those stats include home park effects. This creates a bias: Vantelin Dome (Chunichi): PF_5yr = 0.844 — extreme pitcher's park, suppresses scoring ES CON Field (Nippon-Ham): PF_5yr = 1.147 — hitter-friendly park, inflates scoring If Chunichi pitchers' ERA looks great because of their park, using that ERA directly to project team runs allowed (RA) would underestimate their "true" RA when the park effect is separated out. The correction re