Holiday Haze

Happy holidays! 

Your blogger is about to vanish, returning in the new year. Many thanks for your past, present, and future support. 

If you're at ASSA Atlanta, I hope you'll come to the Penn Economics and Finance parties.

Causality as Robust Prediction

I like thinking about causal estimation as a type of prediction (e.g., here). Here's a very nice slide deck from Peter Buhlmann at ETH Zurich detailing his group's recent and ongoing work in that tradition.

More on Google Dataset Search

Some months ago I blogged on Google's new development of a dataset search tool.  Evidently it's coming along.  Check out the beta version here. Also, on dataset supply as opposed to demand, see here for how to maximize visibility of your datasets to the search engine.

[With thanks to the IIF's Oracle newsletter for alerting me.]

Greater New York Area Econometrics Colloquium

Last week's 13th annual Greater New York Area Econometrics Colloquium, generously hosted by Princeton, was a great success, with strong papers throughout. The program is below. I found two papers especially interesting. I already blogged on Spady and Stouli's “Simultaneous Mean-Variance Regression”. The other was "Nonparametric Sample Splitting", by Lee and Wang.

Think of a nonlinear classification problem. In general the decision boundary is of course a highly nonlinear surface, but it's a supervised learning situation, so it's "easy" to learn the surface using standard nonlinear regression methods. Lee and Wang, in contrast, study an unsupervised learning situation, effectively a threshold regression model, where the threshold is determined by an unknown nonparametric relation. And they have very cool applications to things like estimating effective economic borders, gerrymandering, etc. 

The 13th Greater New York Metropolitan Area Econometrics Colloquium

Princeton University, Saturday, December 1, 2018

9.00am-10.30am: Session 1

“Simple Inference for Projections and Linear Programs” by Hiroaki Kaido (BU), Francesca Molinari (Cornell), and Jörg Stoye (Cornell)

“Clustering for multi-dimensional heterogeneity with application to production function estimation” by Xu Cheng (UPenn), Peng Shao (UPenn), and Frank Schorfheide (UPenn)

“Adaptive Bayesian Estimation of Mixed Discrete-Continuous Distributions Under Smoothness and Sparsity” by Andriy Norets (Brown) and Justinas Pelenis (Vienna IAS)

11.00am-12.30pm: Session 2

“Factor-Driven Two-Regime Regression” by Sokbae Lee (Columbia), Yuan Liao (Rutgers), Myung Hwan Seo (Cowles), and Youngki Shin (McMaster)

“Semiparametric Estimation in Continuous-Time: Asymptotics for Integrated Volatility Functionals with Small and Large Bandwidths” by Xiye Yang (Rutgers)

“Nonparametric Sample Splitting” by Yoonseok Lee (Syracuse) and Yulong Wang (Syracuse)

2.00pm-3.30pm: Session 3

“Counterfactual Sensitivity and Robustness” by Timothy Christensen (NYU) and Benjamin Connault (IEX Group)

“Dynamically Optimal Treatment Allocation Using Reinforcement Learning” by Karun Adusumilli (UPenn), Friedrich Geiecke (LSE), and Claudio Schilter (LSE)

“Simultaneous Mean-Variance Regression” by Richard Spady (Johns Hopkins) and Sami Stouli (Bristol)

4.00pm-5.30pm: Session 4

“Semi-parametric instrument-free demand estimation: relaxing optimality and equilibrium assumptions” by Sungjin Cho (Seoul National), Gong Lee (Georgetown), John Rust (Georgetown), and Mengkai Yu (Georgetown)

“Nonparametric analysis of monotone choice” by Natalia Lazzati (UCSC), John Quah (Johns Hopkins), and Koji Shirai (Kwansei Gakuin)

“Discrete Choice under Risk with Limited Consideration” by Levon Barseghyan (Cornell), Francesca Molinari (Cornell), and Matthew Thirkettle (Cornell)

Organizing Committee

Bo Honoré, Michal Kolesár, Ulrich Müller, and Mikkel Plagborg-Møller

Participants Adusumilli  Karun UPenn
Althoff	Lukas	Princeton
Anderson	Rachel	Princeton
Bai	Jushan	Columbia
Beresteanu	Arie	Pitt
Callaway	Brantly	Temple
Chao	John	Maryland
Cheng	Xu	UPenn
Choi	Jungjun	Rutgers
Choi	Sung Hoon	Rutgers
Cox	Gregory	Columbia
Christensen	Timothy	NYU
Diebold	Frank	UPenn
Dou	Liyu	Princeton
Gao	Wayne	Yale
Gaurav	Abhishek	Princeton
Henry	Marc	Penn State
Ho	Paul	Princeton
Honoré	Bo	Princeton
Hu	Yingyao	Johns Hopkins
Kolesar	Michal	Princeton
Lazzati	Natalia	UCSC
Lee	Simon	Columbia
Li	Dake	Princeton
Li	Lixiong	Penn State
Liao	Yuan	Rutgers
Menzel	Konrad	NYU
Molinari	Francesca	Cornell
Montiel Olea	José Luis	Columbia
Müller	Ulrich	Princeton
Norets	Andriy	Brown
Plagborg-Møller	Mikkel	Princeton
Poirier	Alexandre	Georgetown
Quah	John	Johns Hopkins
Rust	John	Georgetown
Schorfheide	Frank	UPenn
Seo	Myung	SNU & Cowles
Shin	Youngki	McMaster
Sims	Christopher	Princeton
Spady	Richard	Johns Hopkins
Stoye	Jörg	Cornell
Taylor	Larry	Lehigh
Vinod	Hrishikesh	Fordham
Wang	Yulong	Syracuse
Yang	Xiye	Rutgers
Zeleneev	Andrei	Princeton

Dual Regression and Prediction

Richard Spady and Sami Stouli have an interesting new paper, “Dual Regression". They change the usual OLS loss function from quadratic to something related but different, as per their equation (2.2), and they get impressive properties for estimation under correct specification. They also have some results under misspecification.

I'd like to understand more regarding dual regression's properties for prediction under misspecification. Generally we're comfortable with quadratic loss, in which case OLS delivers the goods (the conditional mean or linear projection) in large samples under great generality (e.g., see here). The dual regression estimator, in contrast, has a different probability limit under misspecification -- it's not providing a KLIC-optimal approximation.

If the above sounds negative, note well that the issue raised may be an opportunity, not a pitfall! Certainly there is nothing sacred about quadratic loss, even if the conditional mean is usually a natural predictor. We sometimes move to absolute-error loss (conditional median predictor), check-function loss (conditional quantile predictor), or all sorts of other predictive loss functions depending on the situation. But movements away from conditional mean or median prediction generally require some justification and interpretation. Equivalently, movements away from quadratic or absolute predictive loss generally require some justification and interpretation. I look forward to seeing that for the loss function that drives dual regression.