A Guide to Econometrics

Kennedy wrote this as a companion to standard econometrics textbooks rather than a replacement. It explains what each technique is for, what assumptions it requires, what goes wrong when they fail, and how practitioners actually handle the failures, all in words rather than in proofs, with the technical material confined to separate sections.

The coverage is standard: ordinary least squares and its assumptions, violations including heteroskedasticity, autocorrelation, multicollinearity and measurement error, endogeneity and instrumental variables, simultaneous equations, limited dependent variable models, time series including unit roots and cointegration, panel data, Bayesian methods and forecasting. What distinguishes it is the treatment of practice: Kennedy is candid about specification search, about the difference between what textbooks prescribe and what published papers actually do, and about the temptation to keep running regressions until one produces the expected result. His chapter on the ten commandments of applied econometrics, which includes using common sense and economic theory, avoiding type three error by asking the right question, inspecting the data, and being willing to report negative results, is widely assigned.

For investors and analysts the value is in learning to read empirical work critically: knowing what an instrumental variable must satisfy, why a high R-squared means little, and what a specification test does and does not establish. The 1998 edition predates the modern causal inference literature, including difference-in-differences developments, regression discontinuity and the machine learning methods now common. Later editions cover some of this. The intuitive explanation of the core techniques has not been superseded.