Overview — Multivariate and causal econometrics

A portfolio forecast often needs the joint history of several series. A causal estimate needs more than a predictive regression. This topic separates those tasks: VAR models use joint histories, IV supplies an identification argument, panels separate within-entity variation from persistent differences, and forecast evaluation checks performance on unseen dates.

By the end of this note you should be able to:

• specify a stable reduced-form VAR, select lags, and interpret Granger tests, IRFs, and FEVDs at statement level for SVAR;
• diagnose endogeneity and apply IV/2SLS with relevance and exclusion stated honestly;
• formulate pooled OLS, FE, and RE; use Hausman; recognise Arellano–Bond as a frontier pointer;
• evaluate forecasts with MSE/MAE, honest OOS design, and a textbook-level Diebold–Mariano comparison.

1. Contents
2. The big picture