Supply and demand:
without a good instrument,
not identified.
Panel data hint—
for OVB use FE,
or use CRE.
T-stat looks too good.
Use robust standard errors—
significance gone.
Emptiness of mind,
Like a blank sheet of paper.
General exams.
Method of moments:
replace population mean
with sample version.
Complete class theorem:
admissible rules are Bayes;
the converse also.
Testing restrictions:
Wald, Lagrange Multiplier,
Likelihood Ratio.
From negation comes
growth, progress; not unlike a
referee report.
Consumption Smoothing (after L. Peter Deutsch)
Use value functions:
to iterate is human;
to recurse, divine.
Note: these were written as a sort of thank-you to my dissertation committee in 1998, as evidence that I had learned some things in graduate school. . .
A revised version of the t-stat haiku appears in Angrist and Pischke's Mostly Harmless Econometrics.
Here is one written at the request of Guido Imbens:
Red bus or blue bus?
Multinomial logit
may lead you astray.
Another one written at the request of Victor Chernozhukov:
Without Philip Wright
would there have been causal DAGs?
Who can really say?
Here is a physics haiku I wrote for Peter Steinberg:
Antimatter
we are puzzled by
baryon asymmetry:
where did it all go?
And here is my friend Cyd Harrell's UX Poetry.