MacroModelling.jl by
@kockerols
provides a unified interface for DSGE modeling. Includes dynare.
My fellow macroeconomists: No more excuses to stick to MATLAB.
#juliacon2024
It's official: I got an ERC Advanced Grant to study the macroeconomics of managers. I am really excited about this research agenda. What better way to kick off the project than the Empirical Management Conference at Stanford?
This can indeed be super confusing. Which is why I wrote a Stata command to correct the event study coefficient after xthdidregress. Pick a baseline time period (-1 or any other) and report levels relative to that, not changes.
Here's an illustration from a simple simulated DGP w/non-staggered timing and a linear violation of PT.
TWFE shows a straight line. But some of the other methods show a kink or jump at the time of treatment
Want to please your coauthors? Use our
@Stata
style guide. Better yet, ask them to use it, too. > Koren, Miklós. (2021). CEU MicroData Stata Style Guide (1.0). Zenodo.
I never quite understood why
@Stata
users are so afraid of relative paths. Among the hundreds of .do files I review regularly, only a handful do not hard code their absolute path in the code. 1/n
How to make your research reproducible from day 1? We are doing a workshop with
@AeaData
@larsvil
at the
@EEANews
congress: We will start with an empty folder and do a mini-project together. Join us in Barcelona on August 28.
There are tradeoffs in using different programming languages.
@mschoberml
Some make you happy, some make you fast.
I will also feature a Pareto frontier in my Kezdi.jl talk tomorrow.
Fő állításaink:
1. Karácsony legtöbb szavazója tiszás.
2. Vitézy legtöbb szavazója fideszes.
3. A 24e érvénytelen szavazatból kb 18ezret a Fidesz szavazói adtak le.
Our
@PLOSONE
paper with Rita Pető on social distancing is my first where we followed reproducible research practices from the start. Here are the steps we took, the tools we used, and how they helped us later. An
#econtwitter
thread on tooling. 1/n
The data editors of
@aeadata
,
@EJ_RES
,
@reveconstudies
,
@canjecon
, and
@InquiryWEAI
launch the Data and Code Availability Standard DCAS describes an appropriate replication package and helps authors comply with journals' data and code availability policies
1. Working with data requires coding skills in most jobs: academia, policy, consulting or business analytics. So the question is not whether we teach students to code, but when and how.
Results are in after the
@CEU
debate about
@Stata
,
#RStats
and
#Python
. Not many people changing camps. Most changes are from R to Python.
@Stata
fans are loyal.
The answer is 5%.
Data cleaning (replace, collapse, egen, and the like) and programming (local, if, foreach, and the like) take up 88% of commands.
The rest is exploratory data analysis (tabulate, summarize) and plotting.
#EconTwitter
I did a horse race of Python, Julia, Rust and DuckDB for some simple string matching. To my surprise, Python comes out winning.
If you can improve your favorite language’s performance, submit a PR.
All code written by ChatGPT.
Analyzing 250 Stata replication packages submitted to
@RevEconStudies
. 4k do files, 799k lines of code. What fraction of commands you think are estimation commands like regress or reghdfe?
#EconTwitter
@jaredcrubin
With my Stata package eventbaseline (on ssc), you can pick your base period for event study graphs.
Relative to -3 there is no effect.
You can also use baseline(average), to express all ATTs relative to the average of the pretreatment periods.
Make feels like a secret weapon: helps productivity enormously, yet very few economists use it. I have been meaning to to create a short free course on Make to share this secret. Give me a nudge by signing up.
#EconTwitter
Now that the Stata-R-Python debate is over, I can be non-partisan about this topic. A long thread.
Python won. Congrats to
@arieda_muco
and Gayan Gadyan. 1/
Stata allows command names to be abbreviated. Some of the longer ones are almost always shortened by users.
This is a bad practice because the short names are inconsistent and hard to read.
Some of the abbreviation rules are arcane and confusing…
Kezdi.jl lets you roll your own functions to be used in any of its Stata-like commands.
Every command can be executed on a subset of the data like ‘
@replace
y = 5
@if
y < 5”
The 10x speed is just a bonus.
Do you work in applied theory, financial economics, or applied machine learning? We have three open tenure-track positions at
@ceu
Economics in Vienna.
#EconTwitter
@betanalpha
"I cannot, for the life of me, figure out why this paper was written." Other referees also had issues with our motivation, but did not put it so eloquently.
Super helpful intro to
@JuliaLanguage
macros by
@MoseGiordano
. I'v been playing with Julia for some time, but haven't really understood :() $() and esc() until now
Learning and loving Tidier.jl by
@kdpsinghlab
. Even though I never used R, I agree with the decision to copy a well-designed and popular interface.
@mutate
(log_gdp = log(gdp)) is pretty readable.
Building something exciting in
@JuliaLanguage
for applied microeconomists. If you’d like to shape it, please make your voice heard at our initial design meeting
My colleagues say I have conservative views about big data. I love collecting new sources of data, but strongly believe in applying the old-school rules of statistics and economics.
I am too slow for science. Before I even had time to read "your favorite DSGE sucks," it is debunked. This is peak
#EconTwitter
@JoshuaBrault3
, I'll use your reanalysis in today's macro class!
Instead, what you’ll find is that the volatility of variables in the simulated data from their code is dramatically lower than what you find from DYNARE. The attached image is simulated output growth using identical calibrations.
My colleagues and I are giving a "Data Wrangling for Economists with
@Stata
" workshop at the
@EEANews
conference. This is a pilot
@datacarpentry
workshop. Apply by July 25
Which businesses suffer most from social distancing? We built a model and analyzed O*NET and CBP data to find businesses relying on F2F communication and physical presence. In the U.S., 49 million workers are in affected occupations. 1/4
@PHuenermund
I love it. No-nonsense (~MATLAB-like) syntax with multidimensional arrays in a proper programming language. Fast, too, especially in parallel. Here are some code examples we used in a 1st-year macro class:
@AeaData
Fully agree. You want to use relative paths so that your coauthors can also run the code. But `cd` wreaks havoc with relative paths (relative to where, again?). Run all your code from the root of the project folder.
1+2+3= We need to teach coding first, as part of literacy. Focusing on applications with zero statistics: web scraping, basic text processing. Only then teach statistics, when all students are comfortable power users of their computers.
Know bash, git or Stata? Come help out at the
@EEANews
@datacarpentry
workshop August 23-24. Due to strong interest, I need more helpers who assist learners in debugging problems they encounter during live coding. All online, join for as much as you can.
I got phished.
Got a message in AirBnB about an upcoming trip. Credible URL, credibly cute mixture of Slovak and English.
Asked for some details about the trip. had the basics already filled in. Like what time we are arriving.
Then asked for a credit card. 1/
🎉 The second edition of Data Science at the Command Line is out! 🎉
You can read the entire book for free at
It took a good year to rewrite and expand the first edition, so I'd like to say a few things. 🧵
An exciting paper by
@atrisovic
et al studies R code published in ~2000 replication packages. Only 25% of them run out of the box. About a third of the others can be fixed by two easy changes. Can you guess what these two changes are?
Tomorrow I will deliver my first online class at
@CEUhungary
. It will be crude and glitchy at first, but if you would like to learn about the Dornbusch-Fischer-Samuelson model, you are welcome to join at 9am Budapest time:
Early
@JuliaLanguage
memories ahead of
@JuliaConOrg
next week.
I started with 0.4. Rewrote the whole codebase for a macro paper from Matlab. Lot of work, but two orders of magnitude faster and better code quality.
When the revision came around, upgraded to 0.6.
—
Hungarian Election Office: We spent huge amounts on our great new website.
Users after polls close: Let's check the results.
Hungarian Election Office: Oops.
@VollCornHirsch
@FiveThirtyEight
No, because the uncertainty in
@FiveThirtyEight
model has been calibrated to polls, not results. Unless they retrain the model in real time, even their conditional prediction (what if Biden wins Arizona) will be way off.
Zur Frage der Wirtschaftsbeziehungen zwischen 🇩🇪 und 🇨🇳 war
@julianhinz
gestern als Sachverständiger im Bundestag. Abhängigkeit nicht so groß ist, wie oft angenommen & kommuniziert. Details👉
📺Aufzeichnung der Anhörung 👉
@katewaldock
@Noahpinion
I agree we learn/teach very little about *how* research is done. Leaving students to learn by doing can be very frustrating. I believe the how is also teachable.
@rlmcelreath
A key misunderstanding in scientific computing is that software engineering best practices are only for large-scale production systems. Wrong. These practices shine not when the exact same code runs a zillion times, but when code has to be changed often. Exactly like in science.
Salvatore Lattanzio tells an alternative angle of the story the declining gender pay gap in Italy and the US: Convergence is driven by men's fall closer to women.