Paul Goldsmith-Pinkham Profile picture
@YaleSOM professor & Bulls fan. I study consumer finance, and econometrics is a big part of my research identity. @sgil1122’s life-partner. He/him/his
4 subscribers
Jun 3 21 tweets 8 min read
Just posted a new paper that updates Currie, Kleven and @Esmee_Zwiers 's very nice AEA P&P.

I study how fields differ in their use of causal inference methods.

One punchline: a big chunk in the growth of methods over last 20 years is from diff-in-diff.

paulgp.github.io/papers/trackin…



Image
Image
Image
I try to understand how the credibility revolution (as termed by @metrics52 and @PinkseJoris) has permeated economics, especially across different fields.

In particular, are different empirical methods being adopted evenly?

In general: no -- there's significant heterogeneity Image
May 22 9 tweets 3 min read
Interesting paper that has sparked some discussion here! I think a lot of folks have focused on the headline abstract, so wanted to give my constructive feedback on interpretation of the results First, it’s quite clear that these QT give substantial visibility! Thousands of views and non-trivial increase in likes (control mean of 3 likes, so enormous increase!) Image
Mar 8 16 tweets 5 min read
Finally posting a new paper on how diffusion estimates on networks (e.g. for epidemics, information spread, tech adoption) can be highly non-robust to even tiny (vanishing!) measurement errors. 🧵

[Link to paper: ]
1/ paulgp.github.io/papers/diffusi…
Image What is a diffusion model? It's a way to study how things spread through a network. For example, how a disease spreads through a population, or how information spreads through a social network.

These models are used to do research and make policy decisions!

2/

Image
Image
Image
Aug 15, 2023 19 tweets 8 min read
2023 was a crazy year -- remember how we had a bank run at Silicon Valley Bank that caused a banking sector collapse (over 20% decline in bank equities in a week!) and prompted a Federal Reserve facility intervention?

No? Well then I have the thread and working paper for you! 🧵 Image Following the collapse of SVB, there an immediate response in the stock market. Many banks other than SVB plunged (most notably First Republic Bank, who finally failed many weeks later). Overall, the bank sector corrected sharply downward. Image
Mar 21, 2023 6 tweets 2 min read
Bank failures are a common phenomenon in the United States.

In the 30s and 40s, the FDIC had 20 bank failures a year.

In the 50s and 60s, a lull of 3 per year.

In the 70s, this picked up to 9 per year.

In the 1980s and 90s, an average of 150 banks failed per year.

🧵 What is distinctive is how much the size of the banks failing has changed, even when adjusting for inflation. The total assets is swamped by the size of the crisis in 2008, but even just looking at the average size banks that fail, it is striking:
Mar 11, 2023 10 tweets 4 min read
The market definitely thinks there are more banks that will be run on.

Here's the 7 banks with the largest decline in the market in the last two weeks.

1/🧵 We have:

SVB Financial Group (SIVB) (-60%)
PacWest Bancorp (PACW) (-54%)
Signature Bank (SBNY) (-36%)
Western Alliance Bancorp (WAL) (-32.4%)
First Republic Bank (FRC) (-31.3%)
Customers Bancorp Inc (CUBI) (-23.5%)
First Foundation Inc (FFWM) (-20.3%)

2/
Nov 3, 2022 4 tweets 2 min read
This is a really interesting blog post about DSGE models -- almost makes me want to get into the identification / estimation issues around macro structural models! (almost)

From the paper: From the blog:
Oct 3, 2022 17 tweets 7 min read
New NBER working paper with @jwswallace and @jasonlschwartz on Covid mortality: “Excess death rates for Republicans and Democrats during the COVID-19 pandemic” nber.org/papers/w30512

Ungated on arxiv here: arxiv.org/abs/2209.10751

Thread 🧵1/ Political affiliation has emerged as a potential risk factor 2/ A popular commentary on the Covid crisis has been how much higher the Covid death toll has been in Republican vs. Democratic counties in the U.S.

nytimes.com/2021/11/08/bri…
Sep 30, 2022 5 tweets 2 min read
How it started how it's going ImageImage Two children and a pandemic later
Sep 26, 2022 5 tweets 1 min read
How are these proposed classes @lukestein

1. Intro to teaching online: covers material on setting up systems (zoom, etc.), technology (microphones, cameras, boards), grading/feedback

1/n 2. Advanced teaching online: chat rooms, ways to improve student engagement, integrating advanced technology
Aug 24, 2022 17 tweets 7 min read
We now flesh out the connection of our contamination bias paper to the new diff-in-diff lit.

The two issues flagged by the metrics lit:

a) negative weights
b) contamination across periods

are consequences of broader issues in linear regression!

arxiv.org/pdf/2106.05024…

🧵 In the new appendix, we discuss four examples of DiD and how our main proposition nests these cases.

1) a single intervention
2) a staggered intervention with a single treatment
3) a dynamic event study
4) a single intervention period with multiple treatments
Jun 7, 2022 34 tweets 10 min read
Economists love using linear regression to estimate treatment effects — it turns out that there are perils to this method, but also amazing perks

Come with me in this 🧵 if you want to learn… How:
1⃣ some of the problems with linear regressions in the recent TWFE/DiD lit generalize to other strategies
2⃣ regression can be best for estimating treatment effects when these problems aren't present
3⃣ to bring back regression’s mojo when the problems exist
Dec 14, 2021 18 tweets 5 min read
Thank you for tolerating such a vague poll question.

Let me explain why I think this is a useful thing to bring front and certain, and highlight what I think is a flaw in how much of econometrics is taught, currently.

1/n @instrumenthull gave the best answer to this, and let me try to explain why.
Apr 24, 2021 5 tweets 2 min read
I was looking for an example of how to learn to use this code below from Melissa Dell and co-authors, and found a fun one parsing NBA Injury Report data.

I'm posting the code in the thread below if you're interested. 👇 It started when @EdKupfer suggested pulling this data automatically, and I figured this would be a useful case of trying to apply the layout parser approach

(as it turns out, since PDFs are structured data already, there are less complicated solutions)

Apr 23, 2021 8 tweets 4 min read
This article is a very interesting discussion on what Vietnam got right during Covid -- vox.com/22346085/covid…

Some really amazing quotes that map directly to my paper with Arun Chandrasekhar, Matt Jackson, and Sam Thau: 1/n🧵

arxiv.org/abs/2008.10745 In the paper, we highlight how the success of regional, jurisdictional lockdowns are influenced by four pieces:

1) the delay in detection of infection (how long until you can identify the infection)
2) how detectable the infection is (e.g. ebola vs. covid) Image
Mar 12, 2021 4 tweets 2 min read
If you liked the Gary Chamberlain interview, here's a collection of interviews of econometricians in econometric theory

cambridge.org/core/journals/… That list appears to be incomplete, however -- here's the search list for all interviews in ET: cambridge.org/core/journals/…
Feb 17, 2021 12 tweets 4 min read
My friend @itsafronomics wanted to know the breakdown of student debt by income, so I pulled data from the SCF.

Student debt is more concentrated amongst high earners, but disproportionately amongst Black borrowers. The plot below plots the average within Income Quartiles.
1/n But those are the averages. What's the share that's above 10k? (This is the Biden current cutoff for debt forgiveness, so a useful benchmark).

For comparison, 20% of white Americans have student debt, and 31% of Black Americans.
Dec 21, 2020 11 tweets 5 min read
New paper now out in Annals of Internal Medicine (@AnnalsofIM ):

Worked on a paper with an extraordinary team of doctors trying to deliver Covid related info to Black and Latinx communities. [My contribution was small]

Brief thoughts / takeaway below:

acpjournals.org/doi/10.7326/M2… Image The study used videos of physicians (42 of them!) to try to educate participants on the symptoms of covid and appropriate health-preserving behaviors. We found that watching these videos improved participants knowledge! ImageImage
Jul 28, 2020 14 tweets 4 min read
Now that this is finally officially out, I can give a very brief history of how this paper started -- in 2012.

paulgp.github.io/papers/bartik_… For graduate school, Henry and I were both in Cambridge (he was at MIT, I was at Harvard), while Isaac was at Michigan. All three of us were the same year at Swarthmore, and took most of the same classes together there.
Mar 26, 2020 4 tweets 2 min read
We've (@aaronsojourner and I) updated with forecasts for the week of 3/22-3/28: paulgp.github.io/GoogleTrendsUI…

Our baseline estimate for this coming week (3/22-3/28) is 4.696 million new claims, with confidence interval of 4.2 - 5.1 million. As we discussed, the advance estimates appear to be quite low, perhaps due to processing issues in swamped states’ administrative systems.

When we adjust for our numbers to get closer to "advance" numbers we get an estimate of 3.54 million new claims for 3/22-3/28.
Mar 26, 2020 4 tweets 2 min read
With the daily model, it looks like given the reported stats, we overshot.

The non-seasonally adjusted advance state numbers have 2.9m UI claims. Our two models reported 3.36m (weekly model) and 3.87m (daily).

So what how did we do? Here are the state predictions, with the advance stats on the x-axis, and our two predictions on the y-axis (level scale and log-scale). ImageImage