A Course in Econometrics

I would give this book 6 stars if I could. I think James Heckman has called this book a masterpiece, and I would fully agree. This is a unique text that takes a distinctive approach -- one which, in my opinion, is essential for really understanding econometrics. I am a PhD in economics, and I explored a great many econometrics texts in my quest to get a better handle on the subject. I found that while I can follow all the proofs in standard texts like Greene, I didn't really get the intuition behind how things worked. All that changed once I picked up Goldberger. Goldberger takes what I would call the "identification" approach (an approach emphasized by other well regarded econometricians such as Heckman and Manski). The identification part of econometrics is the link between a model and the probability distribution function of observed variables. If you had an infinitely large sample, so you knew the joint probability distribution exactly, how does that help you identify some interesting parameter in your model? Secondary to this is the issue that in real life, you have only finite samples, and you estimate parameters of the joint pdf only with uncertainty. This is where standard errors and confidence intervals come in. But the identification part is really the core part of econometrics, and is very simple. Most econometrics texts mix identification and estimation, and so unnecessarily confuse the issue. For instance, in the standard approach, the fact that OLS estimates are biased when there is measurement error in the independent variable is usually directly proved by algebraically manipulating the OLS estimator. But this can be seen in the identification part alone, without any reference directly to the OLS estimator. Goldberger makes clear that the OLS estimator is still a great estimator of the best linear predictor (BLP) of the distribution. But the BLP no longer tells you what you need to know given that there is measurement error. So your attention is rightfully directed to why the BLP no longer tells you what you want when there is measurement error, rather than why the OLS estimator is biased. This really simplifies and clarifies everything for me. (Note: I don't recall whether Goldberger directly discusses measurement error; this was just an example to highlight the difference in appraoch). While it is true that this book was published nearly 20 years ago and may not be up to date with all the latest techniques, it is still the best way to learn econometrics, in my opinion. Once you really understand the fundamentals, everything else becomes a clearer.

This book offers a rare bridge between undergraduate and graduate econometrics. The book is well written, with consistent notation, clear exposition, and provides coverage of topics too advanced for undergrad curricula and often not covered by graduate instructors. I highly recommend this book. It will provide you with the necessary foundation to tackle Green, Cameron & Trivedi, etc. This book should be mandatory reading for all first year graduate students in the social sciences.

This book does not have the best explanations and there is little to no examples but the basic material is there.

I used this textbook in an upper-division undergraduate mathematical statistics course. It is well-written, concise, and clear. It made the material significantly easier to understand than it would have been otherwise.

This is an exceptionally well-written text on introductory econometrics suitable for self-study or for use in an advanced undergraduate or a first-year graduate-level course in econometrics. The only prerequisites for reading this text are a good understanding of freshman calculus and a working knowledge of basic matrix algebra. The necessary probability theory and statistics knowledge are developed in the text. This book is less than 400 pages long and a motivated reader can read this text from cover to cover in a few weeks. This is not a very high price to pay to gain a solid understanding of the most fundamental tools of cross-sectional econometrics. The emphasis is on developing the core tools used in econometrics. This text is very readable but at the same time fairly concise. Compare this to many other texts that are padded with hundreds of pages of empirical examples and other verbal detours from the core material. The author is never too verbose, but at the same time offers helpful explanations and examples in cases where the reader is likely to be confused. These days, most graduate econometrics courses are taught from other, more modern, and supposedly more advanced econometrics texts. While many of those popular graduate-level econometrics texts cover significantly more material, they also read like a terrible train-wreck of badly assembled subjects that are extremely difficult to digest on the first (and sometimes second) reading and specially on your own. Therefore, even an advanced graduate student who was once confused by those text may benefit from reading Goldberger's text. Approximately one third of the text is devoted to the background knowledge in statistics and probability. The second part of the text develops the classical normal linear model. The last third is devoted to various kinds of departures from the standard classical assumptions and to models such as GLS, nonlinear models, simultaneous equations, 2SLS, and 3SLS. Only this last part of the text can honestly be called "econometrics". The rest of the text is the standard material on statistical inference and linear modeling. However, this background material is at the core of most econometric tools, and Goldberger nails all issues of this background material "from A-Z". A full proof or at least a sketch of the proof is given pretty much to every result in the text. The first 13 chapters of this textbook cover standard probability theory and statistical inference. This sets Goldberger's text aside from the rest of graduate-level introductory texts in econometrics because most of them relegate the necessary probability and statistics background into tersely written appendices. Goldberger uses some of the ideas and notation developed in those chapters later in the text, so it is useful to review the first 13 chapters even if you have studied statistics before. Chapters 7 and 18 serve as a good introduction to the bivariate and multivariate normal random variables (again, some other texts do not spend as much effort here). Chapters 14 through 25 are devoted to meticulous development of the classical normal regression model. This is where this text truly shines. Everything is proved and explained very well. Chapter 22 and 24 are devoted to issues and strategies for empirical work. Unfortunately, most of the material in this text is developed under the assumption of non-random regressors. Chapter 25 lifts this assumption and shows that nothing really changed (except for notation). Nonetheless, I feel that it would be more in line with the spirit of econometrics to assume random regressors from the beginning. The large sample results of the least squares are stated but not proved, which is unfortunate. Given the asymptotics machinery already developed in the text, presenting a sketch of large sample proofs would not take too much space. The rest of the text talks about GLS, nonlinear models, and simultaneous equations. The presentation of the simultaneous equations model in the subsequent chapters is very thorough with many examples. Most emphasis is on the 2-equation supply and demand type of models. Finally, yet another interesting feature that sets this text apart is that the author emphasizes throughout it the link between OLS, conditional expectation, and best linear predictors. Many other texts barely mention this simple insight. Unfortunately, the material on maximum likelihood is very brief and sketchy. Therefore, it is best to use some other text for MLE theory and models. There is also nothing on panel data models or GMM. I will give this text four stars. It is hard to give five stars to an basic econometrics text that does not have a chapter on standard panel data models. To recap, the best features of this text are: - Short, concise, yet very readable and suited for self-study. - A brief, reasonably rigorous, but intuitive development of the necessary probability and statistics material. - A very good analysis of the classical normal linear model. - Good introduction to analysis of stationary time series, GLS, and SEM. - Emphasizes the link between OLS, conditional expectation functions, and best linear predictors. The weak points are: - No panel data models. - No GMM. - MLE sections are brief and sketchy. - Large sample theory for OLS.