A forecasting lesson on why R-squared alone led to overfitting and worse predictions
mdancho84 · x · 2026-07-21
The author says it took two years to realize why their forecasting models were underperforming, and that fixing two common mistakes made them about 50% more accurate than peers.
The two mistakes
- They relied on in-sample metrics, which encouraged overfitting.
- They optimized only for variance explained, which is not always the right objective for forecasting.
What changed
- They switched to cross-validation instead of training-set scores.
- They combined R-squared with RMSE to get a fuller picture: RMSE for stability around the mean, R-squared for variance capture.
The attached graphic explains R-squared and adjusted R-squared with formulas and visual examples, including a comparison showing how a model can explain 100%, 80%, 40%, or 0% of variance.
More from Research
- Hermes Agent rewrite proposal applies RIA and Logic Bus rules — Promptmethus · 2026-07-22
- AllTheBacteria turns 2.44 million genomes into an AI-ready resource for new antibiotics — shae_mcl · 2026-07-22
- WeirdChat catalogs strange model behaviors from more than 100 million sampled responses — JacobSteinhardt · 2026-07-22
- New agentic benchmark shows AI managers escalate to coercion and fake success — Jasmine Brazilek · 2026-07-22
- Ai2’s Asta adds one-click handoff and self-checking deep paper search — allen_ai · 2026-07-22
- NVIDIA says physical AI starts in simulation with OpenUSD and synthetic data — MonaJalal_ · 2026-07-22