Lag Identification?

…without careful arguments on substantive grounds, lagged explanatory variables should never be used for identification
purposes.

This is from a new paper with Marc Bellemare, from Minnesota’s Department of Applied Economics, and Cornell PhD candidate Taka Masaki, on “lag identification.” This is a very common strategy to avoid problems of endogeneity or reverse causality, especially in political science, although table 1 in our paper shows that this is not exactly uncommon in economics.

We expose the assumption that underlies this strategy, which we term “no dynamics among unobservables.” We also argue that this assumption is almost impossible to defend on substantive grounds, because it requires knowledge about the time series properties of a variable which is unobserved.

We are fairly certain that the germ of the idea for our paper was this post by Phil Arena.

(This post is being simultaneously published today on Marc’s blog.)

Comments 2

  1. Sam February 27, 2015

    Very interesting paper, Tom. Really useful for applied researchers. You may be interested in my paper with Michael Clemens, “Blunt Instruments” (http://tinyurl.com/po9he2r), part of which explores instrument strength (putting validity aside) in the popular dynamic panel GMM application of ‘lag identification’. These two papers seem like a nice complement.

    • tompepinsky February 27, 2015

      Thanks for reading, Sam, and I had forgotten about your paper with Michael Clemens. This is important because for our next paper, we want to make a similar point about lagged instruments, aiming for a short econ journal note. Let’s talk more about that next week.

Comments are closed.