Overview — Regression Diagnostics and Generalized Least Squares

Many factor betas, hedge ratios, and event-study coefficients begin with an OLS fit; other estimators may be needed when its assumptions fail. Classical inference — the $t$-stats and confidence intervals a desk reports — rests on the Gauss–Markov assumptions that asset returns systematically violate: variances cluster, residuals autocorrelate, functional form breaks at regime shifts, and correlated factors inflate coefficient variance. This topic is the motivation stage of the econometrics unit: restate the classical linear model as the baseline, catalogue how financial data breaks each assumption, and build the diagnostic tests and robust/GLS fixes that restore valid inference before later topics re-engineer the model for serial dependence, unit roots, and endogeneity.