Kevin Chen
@jiafengkevinc
Followers
4,013
Following
1,774
Media
275
Statuses
2,008
Postdoc @SIEPR '24 | Assistant Professor @StanfordEcon '25 | PhD @HarvardHBS '24 | AB/SM @Harvard '19 | Econometrics, causal inference, and stats
Joined July 2019
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#TGS2024
• 170748 Tweets
Tigers
• 103271 Tweets
Miami
• 79815 Tweets
Knicks
• 67573 Tweets
Towns
• 64649 Tweets
Minnesota
• 63816 Tweets
#SmackDown
• 53662 Tweets
Randle
• 48205 Tweets
Wolves
• 34983 Tweets
Rudy
• 34560 Tweets
Royals
• 27266 Tweets
Hart
• 22474 Tweets
#DesafioXX
• 21379 Tweets
भगत सिंह
• 21148 Tweets
Asharamji Bapu Case
• 18228 Tweets
日本破壊クソメガネ
• 16671 Tweets
Equality Exploited
• 15666 Tweets
Cam Ward
• 15401 Tweets
Celtics
• 11244 Tweets
how it started // how it's going
Recently accepted by
#QJE
, “Logs with Zeros? Some Problems and Solutions,” by Chen (
@jiafengkevinc
) and Roth (
@jondr44
):
4
195
723
12
63
1K
We've all encountered this problem: "I want to take logs, but my data have zeros => ???"
Commonly (myself included), people use an ad hoc transformation m(.) (e.g. m(x) = arcsinh(x)) that approximates log, and interprets the resulting effect as a %
🧵of paper with
@jondr44
1/n
Twitter is fun because we can promote the work of friends and people we admire!
Today I want to call your attention to a brand new paper by
@jondr44
and
@jiafengkevinc
, available at:
What is the gist?
Check the abstract!
6
78
387
8
154
785
I'm very excited to join
@StanfordEcon
as an assistant professor in 2025, after a year
@SIEPR
. I'm stressed out about moving my 🐱 across the continent, but otherwise beyond thrilled for what's to come!
(also, definitely welcome all advice re: bay area)
46
10
495
Seems like today is a good day to talk about why the bootstrap works
Here's my intuition---
5
34
397
only an economist would call the sample mean "OLS coefficient on a constant"
7
9
370
@KhoaVuUmn
🥴: "Hello, I am
(G'WG)^-1 G'WΩWG (G'WG)^-1.
I want to find the W that makes me as small as possible, but, look at me, I'm a mess"
3
12
318
👀🤩
Synthetic control can be interpreted as an online learning algorithm. Even without assumptions on outcomes, results from online learning imply that synth predictions perform almost as well as those made by the best weighted match.
@jiafengkevinc
1
60
246
4
14
238
as a professional econometrician, i think the WHO's notation choices are amateurish compared to mine
8
7
223
the people who always nod in interviews are my personal heroes
5
6
220
New short paper documenting an easy recipe for incorporating machine learning into linear IV to boost instrument strength *essentially for free*:
Joint work with Daniel Chen and
@econ_greg
#EconTwitter
#CausalTwitter
2
47
212
Excited to share a new working paper taking a closer look at causal inference in school choice settings, where students are matched to schools via algorithms. The paper derives the identified causal estimands under heterogeneous treatment effects (1/n)
2
33
172
because otherwise no one can write a paper, obviously
Seriously though, why should audits be less likely for taxpayers earning $399,999 than those earning $400,000?
6
0
13
1
6
159
Thanks
@jondr44
! A thread on this short paper 🧵
(no econometric toys are harmed in the writing of this paper)
Very cool new paper by
@jiafengchen42
that gives a justification for synthetic control methods
He shows that when treatment timing is random and you have many time periods, SC has close to optimal regret when an adversary chooses the potential outcomes
1
25
129
3
28
157
what is often missing in the statistical software debate is that the labor market value of python and r is positive
Grad students planning to go non-academic
Learn Python and R.
That is all.
Thank you for coming to my TedTalk.
12
31
388
4
8
152
I want to open a discussion about this post by Gelman (
@StatModeling
) and various aspects of RDD inference. Gelman criticizes a paper that shows a close-election victory effect of 2-10 years on "days alive after election" (thread 1/n)
#EconTwitter
1
29
151
I too drop 15 new results in "The Tortured Job Market Paper: The Appendix"
Taylor Swift drops 15 new songs on double album, 'The Tortured Poets Department: The Anthology'
25
43
168
2
8
151
New paper on statistical inference with data from (batched) adaptive experiments with Isaiah Andrews [1/n]
1
21
143
Most regression estimators in the presence of heterogeneity
0
7
128
your daily “regress two independent I(1) against each other” tweet
1
8
126
foolish: use this to instrument trade flows
wise: use this to instrument new trade papers
1
3
114
Since the Twitter exploit must be worth more than $118,000 to a profit-maximizing attacker, the fact that the realized profit is only $118,000 illustrates that the criminal market is not efficient. In this essay I will
That bitcoin scam racked up at least $118,000 in just three hours =>
7
51
105
2
10
105
i'll have to speak to my therapist about this
Here is one:
If X has the same distribution as X', Y as Y', and Z as Z', do we have 𝔼[X+Y+Z]=𝔼[X'+Y'+Z']?
31
134
783
3
5
101
I'm thrilled to announce that I'll be offering difference-in-difference offsets, where I promise to not write DID papers for a price
For a higher price I'll start working on projects like "Why TWFE is More Robust Than You Think" and "Why Constant Treatment Effect is Innocuous"
8
8
102
(while I have your attention, here's a PSA to use weak IV robust procedures)
0
11
100
Live footage of conda dependency conflicts
King cobra bites python. Python constricts cobra. Cobra gets crushed to death. Python dies from the cobra’s venom. 🤯
5K
23K
274K
1
6
88
Excited to share a new preprint with
@DRitzwoller
on semiparametric estimation (read, *machine learning* estimation) of long-term treatment effects, such as those identified by a latent unconfoundedness or a surrogacy assumption (1/n)
1
19
84
i sense a market for this
can someone make the “X as seen by Y” of
{statisticians, biostatisticians, econometricians, polmeth, machine learning}
0
1
19
3
14
79
i'm also on the job market this year 😬
website @ [(n+1)/n]
2
9
71
views about real analysis Did kinda cut the discipline in half
9
4
67
*crosses out “partial identification in causal inference” from research interests because that job is, um, automated*
Our new working paper: tell the computer what you know (and don't know!) about a causal question w/ discrete data → automatically get most precise possible answer (bounds, or a point estimate). Joint w/
@guilhermejd1
@nsfinkelstein
@dean_c_knox
Shpitser.🧵
8
125
533
2
2
68
so the good news is that github copilot does work in stata, the bad news is that it gives you R code
1
1
69
my duck confit came out so well I feel compelled to show it off
1
0
64
T̶a̶y̶l̶o̶r̶'̶s̶ ̶t̶h̶e̶o̶r̶e̶m̶ Mean-value theorem (Taylor's version)
2
0
64
possibly really dumb question: does IV "work" if the instrument Z is confounded with treatment W?
i.e. DAG is Z -> W -> Y, W <- V -> Y, Z <- U -> W,
In PO notation,
Z = Z(U, e1)
W = W(Z, U, V, e2)
Y = Y(W, V, e3)
and (e1, e2, e3, U, V) are jointly independent
10
14
64
Been learning DAGs lately and wanted to be more fluent in translating between DAGs and POs
I think this is a reasonably short proof of the front-door formula/napkin problem that uses only potential outcomes..?
(it can conceivably fit on a big napkin?)
2
2
62
i wonder if anyone has explained the sample variance as
1. regress data on a constant, take residuals
2. regress the squared residuals on a constant
3
0
57
Another way of saying this is that we can't have all three: (a) an average effect, (b) scale-invariance, and (c) point identification. In practice, then, we must decide which two properties we want. [5/n]
1
5
54
that's just econometrics
0
7
57
in short: no alternative transformation can define away the log(0) problem, but sensible alternatives do exist!
[n/n]
3
1
52
your daily macroeconomics tweet
the small angle approximation solves the dynamics of the pendulum by approximating sin(x) = x. How good an approximation is this? This shows the true dynamics (blue) and the approximation (white) for various initial angles
14
156
781
0
5
51
i've got my synthetic wisconsin ready to go
BREAKING: Wisconsin Supreme Court strikes down Gov. Tony Evers' order to stay home in a 4-3 decision.
2K
6K
25K
2
0
47
intro statistics rarely emphasizes the difference between an estimand and an estimator (usually it's fairly clear, e.g. sample mean estimates the mean), but the difference becomes super pronounced and subtle in causal inference
1
9
48
gonna be that guy in the seminar with
@economeager
but in meme form
[this is actually
@jondr44
's example and i just memeified it]
2
1
43
live view of the behavior of the implicit estimand of TWFE under treatment effect heterogeneity
@KhoaVuUmn
1
1
40
Well, a percent is invariant to units, and so a basic requirement is that differences in m(.) should be invariant to scaling.
It turns out the opposite is true: For any m(.) that approximates log, but is defined at 0, one can choose the scaling to get anything one wants 😬[2/n]
2
2
40
Super excited to be presenting this paper at the NBER SI Labor Studies Methods session tomorrow morning!
Thanks
@jondr44
! A thread on this short paper 🧵
(no econometric toys are harmed in the writing of this paper)
3
28
157
2
5
37
If you're at CODE
@MIT
, I'll be presenting some work (joint with Isaiah Andrews) at 11:15am ET today, come check it out!
2
2
38
meanwhile in econometrics, most of what I do is writing P in different fonts
Defining notation for a theory paper is halfway between making up a language and choosing an opening in chess. Small early choices affect what moves are available later.
4
17
382
1
0
37
A buddy and I passed a bunch of SEC 10-K filings through a big pre-trained neural network and examined its brain---because why not, university compute is free and we can't mine BTC anyway
Turns out the network understands which companies are close to which
#EconTwitter
⬇️⬇️⬇️
How can transformer models help us understand economic and financial information?
New workshop paper with
@jiafengchen42
develops an industry classification index from text with greater informativeness about financial metrics than existing text/expert-based classifications [1/2]
1
7
37
2
6
36
𝙉𝙤𝙬 𝙞𝙣 𝙥𝙤𝙨𝙩𝙚𝙧 𝙛𝙤𝙧𝙢
meme form coming soon...
(since this is just linear IV under the hood, it's compatible with AMIP!)
New short paper documenting an easy recipe for incorporating machine learning into linear IV to boost instrument strength *essentially for free*:
Joint work with Daniel Chen and
@econ_greg
#EconTwitter
#CausalTwitter
2
47
212
1
3
35
@otis_reid
A lot of times fancy estimators are actually just comparisons of the correct set of means
3
0
36
i can't believe no one brought up radon-nikodym derivative yet
5
0
34
for some particular reason the harvard-mit econometrics seminar seems to have 13x attendance today
4
1
33
@ben_golub
i'd argue the Rao-Blackwell theorem is better-known, based on me knowing Rao-Blackwell but not knowing the Blackwell information ordering 😂
1
1
32
Sir, this is a landscaping company
71,000,000 Legal Votes. The most EVER for a sitting President!
342K
95K
722K
0
0
33
i think that's what "large-T asymptotics" means
I am reviewing a paper and came across the word “eternal validity,” which is a typo…but its a brilliant typo. Why us economists only talk about internal and external validity? Why not eternal validity???? 😏
#EconTwitter
#AcademicChatter
23
26
476
2
2
31
it's also a good time to finally open yellow van der vaart and build some more intuition why bootstrap can fail---
0
1
32
@AnthonyLeeZhang
seems like most results in real analysis can be understood without real analysis
0
0
32
@ben_golub
@CaltechEconThry
Proof I first saw from
@stat110
's stat 210:
WLOG, assume f(X), g(X) mean zero.
Observe that
E[f(X) g(X)] + E[f(Y) g(Y) ]= E[(f(X) - f(Y)) (g(X) - g(Y))] for X, Y independent and Y ~ X
Conclude by noting that (f(x) - f(y))(g(x) - g(y)) ≥ 0 a.e.
0
0
32
this is why the pscore-as-dimension-reduction thing is a bit of mental gymnastics and word sorcery---estimating the pscore is a high-dimensional problem!
2
3
31
TIL from
@conlon_chris
that one could be using a slower BLAS on Apple silicon, here's a fix that's worked for me, so far, from our thread
@jiafengkevinc
@paulgp
Compare to this -- the point is not that that NumPy is 180x faster than R. The point is that my NumPy setup is using the correct BLAS and my R setup is not:
3
0
8
0
8
31
P̶e̶n̶a̶l̶i̶z̶e̶d̶ ̶r̶e̶g̶r̶e̶s̶s̶i̶o̶n̶ ̶e̶s̶t̶i̶m̶a̶t̶o̶r̶s̶ The Tortured Predictors Department
#TSTTPD
0
2
27
just realized supreme court rulings and the QJE have the same (i think?) typeface
2
0
26
"i have discovered a truly marvelous proof of the results in the main text, which this online appendix has too few pages to contain"
📢 **REStud Page Limit**
With effect from 1st July 2022, a page limit policy applies to all submissions. Papers should be under 45 pages. Online appendices should not exceed 30 pages. A “grace period” is in place until 15th August 2022. For more details:
2
42
272
1
0
27
@jondr44
@ShengwuLi
the easy mnemonic is that Jensen's inequality is always in the direction where you don't want it to be
0
1
26
Happy to share my new NBER working paper with Ed Glaeser and
@davidmwessel
. We find that the Opportunity Zone program, as a part of the 2017 Tax Cuts and Jobs Act, has not (so far) generated a response in residential housing prices. Link: .
#econtwitter
1
1
27
Now, I can't possibly pass on an opportunity to promote some recent work on Twitter, to Twitter, by quote-tweeting
@TwitterEng
(joint work with
@DRitzwoller
)
We built a causal estimation framework on the idea of statistical 'surrogacy' (Athey et al 2016) - when we can’t wait to observe long-run outcomes, we create a model based on intermediate data.
2
23
254
1
5
26
Suppose (Xn, Yn) converges in distribution to (X, Y), and (Yn, Zn) converges in distribution to (Y', Z). (This implies Y' ~ Y). Suppose Xn is independent of Zn given Yn.
T/F: (Xn, Yn, Zn) converges in distribution to some (X'', Y'', Z'')?
4
0
24
I will be presenting this paper on inference post adaptive experiments this Thursday (11:30 am ET) at the International Seminar on Selective Inference.
Come check it out!
1
1
24