Time to celebrate! “DID estimators of intertemporal effects” accepted at @restatjournal, and very very fast did_multiplegt_dyn Stata & R commands available from SSC & CRAN, thanks to Mélitine Malézieux @fe_knau @diegociccia1 & @DSDoulo! Celebratory thread! https://t.co/i0lwPE6mtV

Sciences Po has junior positions in econometrics and in health/environmental this year, please apply, and please RT! https://t.co/3xiPoVjEcX @ScPoEcon

For the month of September at least, I am moving to where the skies are nicer than in Paris. Follow me there, if you want to hear, say, about a cool event I'm organizing at the next EEA-ESEM meeting, or about our packages updates!

Hi #EconTwitter! 📈 Looking for a book on #econometrics through the lens of optimization techniques? Check out this comprehensive textbook by @guillaume_econ (@UChicago)! 📚 It covers a wide range of topics, including MCMC methods, optimal transport, nonconvex econometric

Sad to see that SpringHill Suites Roanoke @Marriott are still running their entire breakfast with disposable plates, utensils, and cups. Buy a few dishwashers, hire a few more workers, and stop sending tons of plastic trash every year to landfills and the oceans!

Kamala Harris says she will bring back Two-Way Fixed Effects, if elected.

Hoy va hilo sobre la Reforma Laboral 2021 que trataba de reducir la temporalidad, porque: 1. La temporalidad en España ha doblado siempre la de la UE (no importa el sector!!). 2. Tiene efectos MUY negativos en el empleado y empresa. 3. Es posiblemente la reforma más ambiciosa.

Big congrats to @ArkhangelskyD who was just granted tenure at @CEMFInews and promoted to tenured associate professor! Great news for Dmitry and for CEMFI!

This is still a beta version of the command, please reach out to chaisemartin.packages@gmail.com if you find bugs or have suggestions for improvement. Thanks for reading, and please share with us the interesting analyses we hope you will be able to conduct with this tool!

We use the command to estimate effects of taxes on gasoline consumption and prices, and price-elasticity of gasoline consumption.

But resulting estimator still restricts effects’ heterogeneity across units. On the bright side, the reduced-form parallel trends assumption can be placebo tested.

We show that reduced-form parallel-trends assumption implicitly restricts treatment-effect heterogeneity. Such restrictions can be alleviated by controlling for groups baseline treatment in the IV specification, which we recommend doing.

In such cases, the command can compute the IV-WAS estimator. IV-WAS= WAS estimator of the instrument's reduced-form effect on the outcome, divided by WAS estimator of the instrument's first-stage effect on the treatment.

Then, counterfactual consumption evolution of products experiencing and not experiencing a price change may not be the same. On the other hand, taxes may not respond to demand shocks and may satisfy a parallel-trends assumption.

Finally, there may be instances where parallel-trends assumption fails, but one has at hand an instrument satisfying parallel-trends. For instance, one may be interested in estimating the price-elasticity of a good's consumption, but prices respond to demand shocks.

some units may have experienced treatment changes before t=1, those changes could still affect them over study period, but no way to account for them because unobserved. Then, ruling out dynamics or allowing for restricted dynamics not ideal might be the "least-bad" we can do.

See also did_multiplegt_dyn for estimators allowing for unrestricted dynamic effects of treatments you observe in your data. However, in the designs we consider, treatment continuous even at period one, and allowing for unrestricted dynamics opens up initial-conditions problem:

Command also assumes a static model: units' outcome at period t only depends on their period-t treatment, not on their lagged treatments. Still, command can be tweaked to obtain estimators robust to dynamic effects up to a pre-specified treatment lag, see Section 6.3 of paper.

Statistically, WAS dominates AS: often more precise (standard error of WAS ~ 3 times smaller than that of AS in our application), and amenable to doubly-robust estimation. => tuning parameters can be chosen in data-driven manner via cross-validation, see command's CV options.

WAS can not serve that purpose, but can be used to do a cost-benefit analysis of treatment changes that took place.

Now, which parameter should we choose between the AS and the WAS? Economically, serve different purposes: under functional-form assumptions, AS can be used for counterfactual analysis: point or partially identify effects of other treatment changes than those that took place.