R-Squared

R-squared is the share of the variation in the outcome that a regression explains, from 0 to 1. It is routinely over-read: a high R-squared says the model fits this sample, not that it is correct, causal or predictive. Adding any variable raises it mechanically, which is why adjusted R-squared penalises extra terms, and why a 0.99 fit on financial data usually means the model has memorised noise.