Rachel Leah Childers

For Spring semester, I'm bringing back free weekly open office hours for anyone in the world with Econometrics questions. Tuesdays 4-6 PM Central European Time (US EST 10AM-12PM) or by appointment; sign up and drop by!

@causalinf Lecture 1 & 2 of my computational econ notes, based on Judd's textbook, explain floating point and its propagation. For regression, differences arise due to multicollinearity and mixed scaling, but different algorithms matter a lot in high dimesnions! github.com/donskerclass/C…

I am embarrassed to say I didn’t know jack about “big numbers” being handled differently by different programming environments on the same computer. I have avoided learning all this float stuff for so long and now I have to. “Claude make a beautiful deck abt floating numbers”

@jmwooldridge @pedrohcgs Surely that goes into the Intro to Probability and Statistics class that's a prerequisite for Intro to Econometrics, right? x.com/DonskerClass/s…

@pedrohcgs Of course! But when do you cover Donsker classes?

What does it say that I feel comfortable teaching extended TWFE for staggered interventions to my senior seminar undergraduates at MSU?

For Spring semester, I'm bringing back free weekly open office hours for anyone in the world with Econometrics questions. Tuesdays 4-6 PM Central European Time (US EST 10AM-12PM) or by appointment; sign up and drop by!

@cis_female That's a pretty modest amount relative to contemporary persistent current account surpluses? There are good reasons to dislike commodity-based currencies, but this is not a feature unique to them. On the Chinese example, see Jin Xu's "Empire of Silver": goodreads.com/review/show/65…

perils of a commodity based economy

the chinese spent very roughly 1% of gdp per year importing silver from the new world 1600-1800

Sign up for free open Econometrics office hours at donskerclass.github.io/OfficeHours.ht…

Go see Emily to learn about principled ways to use synthetic data!

@omaclaren Oh, right, because all minimizers correspond to perturbations in the null space of the sample Xs. Fair.

@DonskerClass Fair point! I'm OK with more sophisticated arguments along those lines, but feel the Gaussian random field picture/proof sketch isn't really capturing the core ideas/geometry involved. Eg in toy over-param linear regression, every critical point is a global minimiser & no saddles

But I feel like this picture leaves out the very important fact that modern neural networks are massively overparameterised and so eg the local Hessian would almost always be singular (zero eigenvalues) with non-isolated critical points...

We are very proud to present our outstanding 2025/26 cohort of job market candidates from the @UZH_en Department of Finance/Swiss Finance Institute! ✨✨✨✨✨ df.uzh.ch/en/studies/phd… Please find their research interests and job market papers below👇👇👇

@xuanalogue economics.mit.edu/sites/default/…

watching friends queue up for a hawker center stall and now wondering when "got queue means good" reasoning can sustain an equilibrium where everyone falsely thinks the stall has great food before actually receiving the food 🤔

I will be at University of Tübingen on Tuesday Dec 2, and at NeuRIPS in San Diego Dec 3-8 spreading the good word of the Generalized Method of Moments in the age of AI. Econometrics lovers come say hi!

@t_holden I was in a seminar the other day where the speaker really did forget to cluster their standard errors. Sometimes you just gotta play the hits.

Obviously people shouldn't do this ⬇️, but I do understand why they do. If papers are refereed by people from the same sub-field, the referees have no incentive to point out field-wide issues. Think, for example, about applied finance papers with time fixed effects but which seek to draw conclusions on the aggregate effects of shocks. So as an audience member outside the referee bubble it can be frustrating: seeing a string of papers with the same problem. At what point do you give up and shut up? Sometimes the problem is not so clear cut, but there are still genuine disagreements in the mainstream profession which are not addressed by refereeing due to these referee bubbles. Some are sceptical about the informativeness of surveys, for example. Having a discussion about the general informativeness of surveys is clearly not very helpful in a paper using one specific survey. But we do need some forum as a profession for these discussions to take place (other than just grumbling over post-seminar coffee). For various reasons, journals are unlikely to fulfil that role.

Updated Fall time for free weekly open office hours for anyone in the world with Econometrics questions. Wednesdays 3:30-5:30PM Central European Time (US EDT 10:30/EST 9:30AM) or by appointment; sign up and drop by!

Econometrics Free Open Office Hours details and signup at: donskerclass.github.io/OfficeHours.ht…

Working on this project to understand how LLM simulation data could contribute to drawing valid scientific conclusions from real world data, led by @yewonbyun_ with @shantanug7, was a great learning experience. Happy to see it forthcoming at #NeurIPS2025

Valid inference w/ LLM-simulated data: 1. Take subsample of texts, extract vars 2. Construct moments identifying param on those vars 3. Ask LLM to simulate vars on sample & remaining texts 4. Use same moments w/ simulated vars 5. Combine moments, estimate jointly w/ 2-step GMM

For today's meeting, I read about Forward-Backward Stochastic Differential Equations. Among other things, these allow for a stochastic version of Pontryagin's maximum principle for optimal control along a path. Some notes I made explaining how they work: donskerclass.github.io/misc/DeepBSDE.…

@Apoorva__Lal Thoughts: without costs ads are always preferred. What they do to engagement depends on whether dɑ/da or dβ/da<0. Your model is static, but you could add dynamics with customer inertia, cf PK paulkrugman.substack.com/p/the-general-… Correlation of timing and value could get an evolving userbase.

Eqm existence and uniqueness proof is vibes-based atm but people do higher stakes things on a vibes basis in 2025 so i figured the bones were interesting enough to scratch out anyway

economic models are surprisingly fun devices for expressing strident opinions with no empirical basis