iiinova — where your digital self keeps evolving

Reposted by Jimbo Brand

Grant McDermott @gmcd.bsky.social · 26/01/2026

Things are grim. But in more frivolous news... @jamesbrandecon.bsky.social and I have been chipping away at `dbreg`, a 📦 for running big regression models on database backends. For the right kinds of problems, the speed-ups are near magical. Website: grantmcdermott.com/dbreg/ #rstats [1/2]

grantmcdermott.com

dbreg

46815 0

Jimbo Brand @jamesbrandecon.bsky.social · 07/01/2026

yeah your collaborator on dbreg is definitely a little too willing to vibe-code to get things finished quickly.... Been interesting to see when it bites. But my thinking is that the only way I can figure out how to manage the LLM well enough is to stretch the limits, break things, then fix/iterate

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 02/01/2026

New blog post, on the impact of cheap verification on data science hiring, work, and compensation: jamesbrandecon.github.io/blog/posts_h...

jamesbrandecon.github.io

(Essentially) free verification of facts and skills

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 05/12/2025

You say MCMC but I've thought about this before and didn't immediately see the gains. GPU speeds up some bayesian stuff (variational inf) but MCMC is basically happening in serial so what's the benefit? Just that each draw requires matmul which is now gpu-fast?

120 0

Jimbo Brand @jamesbrandecon.bsky.social · 15/11/2025

+1 and because the (implicit) exclusion restriction isn't as front and center. Easier for most, I think, to wave hands and say generally "I hope unobservables aren't too bad" than to say "boy unobservables are really bad here but I found one special variable that fixes it"

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 15/07/2025

Wrote a second blog post! This time it's about using LLMs as part of an ensemble imputation method, relying on their "knowledge" of the world to provide additional prediction signal. Seems to work well, especially for bigger models! jamesbrandecon.github.io/blog/posts_h...

jamesbrandecon.github.io

2025-07-14_LLMs-for-Imputation

020 0

Jimbo Brand @jamesbrandecon.bsky.social · 01/05/2025

P.S. I also tried to make plotting easier. Here, for example, are estimates from a model with two product characteristics and correlated preferences. More to come, including docs to make it easier to dig through the "problem" objects to extract results.

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 01/05/2025

As with FRACDemand (which is now registered!), the coolest thing here is that it's fast. less than a minute to define the problem, estimate it, and calculate price elasticities for 500 markets with 20 products each

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 01/05/2025

Have a day off so I made some small updates to FKRBDemand.jl (new name!). Hopefully it's now easier to use the Fox-Kim-Ryan-Bajari method to estimate a random coefficient model with market-level data (2-step approach described here: www.jamesbrandecon.com/blog/0jxkvfr...) github.com/jamesbrandec...

github.com

GitHub - jamesbrandecon/FKRBDemand.jl

Contribute to jamesbrandecon/FKRBDemand.jl development by creating an account on GitHub.

182 0

Jimbo Brand @jamesbrandecon.bsky.social · 20/01/2025

Nice, thanks for explaining! Agreed, it seems useful to help see how I'm using memory while building something (which I still wish I understood better)

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 19/01/2025

Is the goal just that GC can catch when to run? Or is there another performance benefit I don't understand? Seems like a lot of additional code so I'm assuming there's more value

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 18/01/2025

This point you've been on has stuck with me as an industry guy. How do firms succeed while being terrible at stats reasoning? Because some big ideas are robust to that bad logic and (like VCs?) you get a lot of shots and only need a few to pay off. That changes how to think of data work for the firm

120 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

Finally, we put this all in a Julia package! Trying to take seriously that these complex methods cost a lot of time and effort to implement, we tried to make it as easy as possible to use our methods for estimation, inference, and some basic counterfactuals 6/6

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

Third, we apply our method to real demand from 12 categories of products at a large retail food chain. Our estimates agree with our simulations — constraints matter, and imposing them generates more reasonable elasticities and counterfactuals (below) 5/

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

Second, we show in simulations that our method improves upon the previous state of the art for this model, both because we can fully enforce constraints and because doing so gives us more accurate estimates of price elasticities. 4/

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

This two-step approach not only helps us impose new constraints -- it also guarantees that our constraints are satisfied *everywhere*, which we show is often not true of the best previous approach 3/

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

First, we develop a new quasi-Bayes approach to solve this (hard!) problem. We show that imposing constraints on nonparametric demand curves is better suited to MCMC sampling approaches than frequentist optimization, and develop a two-step process for sampling in our setting 2/

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 17/01/2025

New paper with @adam-n-smith.bsky.social papers.ssrn.com/sol3/papers.... Our paper develops a new approach for estimating demand nonparametrically while imposing economic constraints and comes with a new package, NPDemand.jl! Some things we do in the paper 1/

33710 0

Jimbo Brand @jamesbrandecon.bsky.social · 11/12/2024

The paper looks really nice, but Prop 2 (no ME) feels stronger than is useful in practice... do we hold ourselves to that standard for any other measurement tool? Or am I being unfair/missing something

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 08/12/2024

Ditto, my brain hurts enough switching between the languages themselves. Adding UI as another thing to switch is a pain

000 0

Jimbo Brand @jamesbrandecon.bsky.social · 08/12/2024

Wow, as an increasingly frequent R user I wish I'd known this was possible months ago -- thanks for figuring this out in public

150 0

Jimbo Brand @jamesbrandecon.bsky.social · 27/11/2024

Been (finally) reading through it in detail recently -- great paper!

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 26/11/2024

Agreed (and I get this argument from ML people a lot), but funny enough we'll never know the counterfactual there either! I could walk around in the dark and get where I'm going but it'd still be nice to have the lights on

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 26/11/2024

My experience is that the former often dominates, and is the default even in settings where academics would lean toward the latter (when estimating the value in the world is statistically hard) but folks are surprisingly open to understanding the issues with "ML everything" when demonstrated

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 26/11/2024

There's definitely some of "just do whatever works" in my experience but there are also a ton of people (often PhDs) trying to do things as right as possible under constraints, including experimentation and causal inference

120 0

Jimbo Brand @jamesbrandecon.bsky.social · 23/11/2024

Agreed, I wondered how reallocating regular up to premium is being counted. New planes with 50% premium seems high! Curious to see how premium margins decline (seems like they have to?) as lower WTP customers are forced into it just by prevalence

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 23/11/2024

Awesome find. Silly question -- what does "incremental" here really mean? Net new seats mostly come from new planes/routes (I assume) and I don't see how 85% can be premium. Any idea?

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 12/11/2024

My group at Microsoft is hiring again! This time looking for junior candidates specifically wanting to study advertising. Academic friends (econ or business school especially), please let me know if you have PhD students we should look at! Listing here: jobs.careers.microsoft.com/global/en/jo...

jobs.careers.microsoft.com

Search Jobs | Microsoft Careers

031 0

Jimbo Brand @jamesbrandecon.bsky.social · 23/10/2024

Oh yeah? Well where else can you get a "taco" with a hot dog in it?

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 18/12/2023

If y_i are just means, doesn’t going full or empirical Bayes make sense? Feels like that kind of structure

020 0

Jimbo Brand @jamesbrandecon.bsky.social · 08/12/2023

What do you mean exactly? There are a handful of studies looking at productivity of different forms. Usual finding is that lower-skill users are more likely to benefit, though not always. One paper with a few studies www.microsoft.com/en-us/resear...

microsoft.com

Early LLM-based Tools for Enterprise Information Workers Likely Provide Meaningful Boosts to Product...

This report presents the initial findings of Microsoft’s research initiative on “AI and Productivity”, which seeks to measure and accelerate the productivity gains created by LLM-powered product...

110 0

Jimbo Brand @jamesbrandecon.bsky.social · 29/11/2023

Sorry, I’m not familiar with any guides! The particulars are an odd mix of very general (simulate the DGP you care about and do your estimation many times) and DGP specific (e.g. realistic IV strength in nonlinear models)

000 0

Jimbo Brand @jamesbrandecon.bsky.social · 28/11/2023

Isn’t the norm to do it via simulation? For most non-RCTs I’ve seen its hard to do anything else

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 24/11/2023

In case non-twitterers are interested: A demand estimation approach that I've always liked but never tried is Fox-Kim-Ryan-Bajari 2011. Wanted to see how easy it'd be to package + whether it would perform better than FRAC.jl, so I wrote a quick gist. Works well! gist.github.com/jamesbrandec...

gist.github.com

A one-file implementation of Fox, Kim, Ryan, Bajari (2011)'s linear estimator for the mixed logit de...

A one-file implementation of Fox, Kim, Ryan, Bajari (2011)'s linear estimator for the mixed logit demand model, in the style of FRAC.jl and NPDemand.jl. - FKRB.jl

071 0

Jimbo Brand @jamesbrandecon.bsky.social · 06/10/2023

Agree. It seems like an interesting problem to solve. Easy to make some things worse by giving intro classes on pandas/ggplot/scikitlearn without the causal inf (preaching to the choir) but clearly lots of resources could be saved if MBAs could SELECT * from XX and summarize data programmatically

010 0

Jimbo Brand @jamesbrandecon.bsky.social · 06/10/2023

To me it feels like the common setup in industry where there are “business people” and “data people” is a pit stop toward a world where the best business folks can do some basic data analysis themselves. Is that a common view? Are there MBA programs w that view (I.e. with basic coding courses)?

101 0

Jimbo Brand @jamesbrandecon.bsky.social · 01/10/2023

This paper is one of my favorites in a long time, and also resonates as an understudied issue in IO whenever we study big firms: what do you do if your supply side consists of a forecaster, a capacity planner, a price setter,etc.? Are there other new papers on this? academic.oup.com/qje/advance-...

academic.oup.com

Organizational Structure and Pricing: Evidence from a Large U.S. Airline*

Abstract. Firms facing complex objectives often decompose the problems they face, delegating different parts of the decision to distinct subunits. Using compreh

0222 0

Jimbo Brand @jamesbrandecon.bsky.social · 23/09/2023

Couldn't get into Empire Central in the same way on a first listen but maybe should go back. Yeah love that they get the crowd to sing -- had a similar experience with vulfpeck

100 0

Jimbo Brand @jamesbrandecon.bsky.social · 23/09/2023

Man, always wanted to see them live. I account for more than my fair share of We Like it Here youtube views

110 0