A tour of the nine big problems that motivate the course, multicollinearity, heteroscedasticity, autocorrelation, mis specification, panel heterogeneity, distributed lags, simultaneity, VAR feedback, and limited dependent variables, plus a short OLS and CLRM refresher to start everyone on the same footing.
Perfect vs high but imperfect multicollinearity, why OLS coefficients become imprecise even though they stay unbiased, detection (VIF, condition index, eigenvalues), and remedial measures (drop a variable, get more data, use prior information, ridge style fixes).
When the variance of the disturbance grows with X, OLS stays unbiased but its standard errors are wrong. Generalised Least Squares (GLS), weighted least squares, detection (Breusch Pagan, White, Goldfeld Quandt), and robust standard errors as the practical fix.
When today's disturbance carries over into tomorrow, common in time series, OLS again loses efficiency and produces invalid t and F statistics. AR(1) error processes, the Durbin Watson and Breusch Godfrey tests, GLS corrections, and Newey West HAC standard errors.
Types of specification errors, omitted variable bias, inclusion of irrelevant variables, errors of measurement, mis specification of the stochastic error term, plus tests (Durbin Watson, Lagrange Multiplier), non nested hypothesis testing, encompassing models, and model selection criteria (AIC, BIC, Mallows Cp).
Panel data structure and its advantages over pure cross sections, the constant coefficient model, fixed effects (LSDV and within group), random effects, assumptions and estimation, and the comparison of fixed and random effects via the Hausman test.
The nature of simultaneity, illustrative systems of equations, the bias and inconsistency of OLS in such systems, and the difference between structural form and reduced form equations.
Under identification, exact identification, and over identification, the rank and order conditions for identification, the Hausman test of simultaneity, and tests for exogeneity.
Limited information methods: OLS on recursive systems, Indirect Least Squares for just identified equations, Two Stage Least Squares (2SLS) and its IV interpretation for over identified equations, plus limited information and full information maximum likelihood.
Lag and difference operators, finite distributed lag models, the infinite lag model, the Koyck approach, adaptive expectations, partial adjustment, estimation of autoregressive models, the IV fix for autoregressive disturbance, and Granger causality.
Econometric forecasting, Vector Autoregression (VAR) forms and estimation, forecasting with VAR, and the Granger causality test in the VAR framework. Scope: Sec. 22.9 onward, the VAR portion of the chapter only.
When the dependent variable is binary, censored, or truncated. The Linear Probability Model, Logit, the grouped Logit, Probit, truncated regression, censored regression (Tobit), and Poisson regression for count data.