Working With Greene's Econometric Analysis: A Practical Guide
Most people treat William H Greene Econometric Analysis Fifth Edition Prentice Hall as a reference book they dip into when things go wrong. That is technically correct but completely misses how useful it actually is if you use it right. The book is dense. It covers everything from basic OLS through panel data models, limited dependent variables, and time series in exhaustive detail. When I was first learning econometrics, I went through it cover to cover in three weeks, took nothing away except a headache, and stopped trusting anything I did not independently verify. The fifth edition predates a lot of what is now standard in applied work. It focuses heavily on Gauss programming examples rather than R or Stata, which immediately limits its practical utility for most students entering the field today. That said, the theoretical content holds up remarkably well. Chapter 5 on matrix algebra is still the clearest review of the subject I have seen in any single text. The treatment of maximum likelihood estimation in Chapter 7 is also where the book earns its reputation — Greene derives everything from first principles rather than handwaving through proofs, which matters more than you might expect when you eventually need to code a custom likelihood yourself. The sections on limited dependent variables — specifically chapters 21 through 24 covering logit, probit, Tobit, and censored models — remain among the best technical explanations available. Greene does not shy away from the computational difficulties that arise with these models. He also does not pretend that standard errors from MLE are always reliable without qualification. That honesty is rare in textbooks from this era.
One thing beginners consistently miss: the book assumes you already understand calculus at an intermediate level. It is not a self-contained introduction. If you have not taken real analysis or multivariable calculus recently, you will struggle through the first two-thirds before finding yourself lost. I learned this the hard way. I tried to start with Chapter 3 on distribution theory and spent six hours on notations that should have taken twenty minutes once you know what they mean.
A Specific Problem I Ran Into and How I Fixed It
There is a section in the time series chapter on cointegration where Greene presents the Engle-Granger two-step method using Gauss code. I tried to replicate the results in Stata for a project on housing prices. The coefficients matched roughly, but the standard errors were systematically smaller in my Stata output. After tracing through the code line by line, I discovered that Greene's Gauss implementation uses a bias-corrected estimator for the second stage that most software packages do not apply by default. The fix was straightforward once I knew where to look — I manually adjusted the variance-covariance matrix in Stata using the correction factor he derives on page 871. This kind of detail is exactly why the book stays relevant despite its age. The book is not without serious limitations. The Gauss programming examples are essentially unusable for anyone who does not already know Gauss. This means students in 2024 and beyond must translate every example by hand, which takes time and introduces opportunities for mistakes. There is no official source code repository or companion website for the fifth edition, so you are on your own if a code snippet does not run correctly. Another issue: the book does not cover bootstrapping in any meaningful depth. For a text this comprehensive, that gap is surprising and problematic. Bootstrap methods are now standard in applied econometrics for dealing with small samples and complex estimators. If your work involves finite samples or non-linear models where asymptotic approximations are shaky, this book will not help you there. You should pair it with a more modern resource for that purpose.
Get the Full Details
The section on simultaneous equations models is also thinner than it should be. Greene covers the identification problem adequately but moves too quickly past 2SLS and LIML into less common alternatives without explaining when one might be preferable to the other in practice. For applied work on structural estimation, I would recommend supplementing with a text like Hayashi or Wooldridge rather than relying on Greene alone.
How I Use This Book Now
I keep a copy on my desk primarily for theoretical reference. When I need to verify a derivation or check the assumptions behind a particular estimator, Greene is usually the first place I look. The chapters on panel data methods — fixed effects, random effects, and Hausman testing — are particularly useful for quick recall. The derivations are complete enough that I rarely need a second source for the core material. For hands-on implementation, though, I rely entirely on R packages and Stata documentation. The conceptual understanding from Greene translates directly, but the actual coding requires a different toolkit. If you are a graduate student working through this book, spend most of your time on the theory chapters and treat the programming sections as illustrative rather than instructional. The mathematical framework is what makes the book worth reading. The code is what makes it frustrating. One last note: if you can find the sixth edition, it adds more material on panel data and microeconometric methods but retains the same structural weaknesses. The fifth edition remains the tighter read. The later editions expand the page count without always expanding the clarity.
