Simultaneous Equations Models (SEM's) played an absolutely central role in the history of Econometrics. Simultaneous systems and measurement errors went hand in glove in forcing the emergence of econometrics as a field in its own right.
It's not that long ago that courses in econometrics devoted a good deal of time to SEM's. These days we spend much less time discussing these models, which is a shame because there are lots of important insights associated with them.
It's surprising how many newly-minted Ph.D. students in economics have never estimated an SEM using FIML. It's disturbing how many researchers doing empirical micro. don't have a clue where I.V. estimation came from (long before they were born), or what "identification" is really all about.
The other day I dug out the notes for a course on SEM's that I taught at Monash University, around 1977 or 1978. Yes, I know that "dates" me! The notes are rather "dated" too, and there's much that I'd explain rather differently if I were writing them today. (The first handout on large-n asymptotics is especially in need of a major re-write.)
None the less, these notes do bring the main material on SEM's together in one place, and with a consistent set of notation. It occurred to me that they may be of some interest, so for what they're worth you can download these old notes from here.
Oh yes, if you're one of those people who has suddenly become excited about instrument validity and "weak instruments", you'll find a brief discussion of the Anderson-Rubin test (and other related tests for over-identification) in Handout 10.
It's not that long ago that courses in econometrics devoted a good deal of time to SEM's. These days we spend much less time discussing these models, which is a shame because there are lots of important insights associated with them.
It's surprising how many newly-minted Ph.D. students in economics have never estimated an SEM using FIML. It's disturbing how many researchers doing empirical micro. don't have a clue where I.V. estimation came from (long before they were born), or what "identification" is really all about.
The other day I dug out the notes for a course on SEM's that I taught at Monash University, around 1977 or 1978. Yes, I know that "dates" me! The notes are rather "dated" too, and there's much that I'd explain rather differently if I were writing them today. (The first handout on large-n asymptotics is especially in need of a major re-write.)
None the less, these notes do bring the main material on SEM's together in one place, and with a consistent set of notation. It occurred to me that they may be of some interest, so for what they're worth you can download these old notes from here.
Oh yes, if you're one of those people who has suddenly become excited about instrument validity and "weak instruments", you'll find a brief discussion of the Anderson-Rubin test (and other related tests for over-identification) in Handout 10.
© 2012, David E. Giles
I am surprised about that, Davidson-MacKinnon does a pretty good job introducing SEM in their proper place, before introducing IV estimation. Cameron-Trivedi does give proper respect to SEM by introducing them early as well.
ReplyDeleteWoolridge is one example introducing them as "special case" of IV estimation, though. Maybe a source of your annoyance.
I like the historical approach you seem to favour better as well, personally. Econometrics is difficult enough that I don't want to jump steps in its understanding!
Alexia - thanks for the comment. The books are one thing, but what students (don't) get taught is another. Judging by what I encounter, day to day, there's a gap between the two.
DeleteDear prof,
ReplyDeleteHow about Seemingly unrelated regression? What do you think about some studies (Konya, 2006; Boubtane et Dramane, 2011) who test Granger-Causality in the case of SURE regression using Wald statistics (bootstrapped)?