THE ET INTERVIEW: PROFESSOR JOEL L. HOROWITZ

Sokbae LeeEconometric 2025 · DOI

TL;DR

This interview with Professor Joel L. Horowitz offers a fascinating journey through his career, highlighting his pivotal transition from physics to econometrics and his profound contributions to areas like semiparametric and nonparametric estimation. It reveals a deep appreciation for the rigor and empirical focus of physics, contrasting it with the evolution of economic theory, and showcases his collaborative spirit and intellectual curiosity that have shaped modern econometrics.

What Problem Does This Paper Address?

This document isn't a single research paper but rather an interview with Professor Joel L. Horowitz, a distinguished econometrician. Therefore, it doesn't address a single problem in the traditional sense. Instead, it explores the evolution of econometric theory and practice through the lens of Professor Horowitz's career. It touches upon the challenges of:

  • Bridging the gap between theory and empirical work: The interview contrasts the physicist's approach to theory as a "target" to be tested against experimental results with the more flexible nature of economic theory.
  • Developing robust estimation methods: Horowitz's work has focused on creating sophisticated statistical tools (like smoothed maximum score estimators and nonparametric IV) to handle complex economic phenomena where standard methods fall short.
  • The increasing complexity of econometric models: The discussion highlights the shift towards high-dimensional models and the need for new techniques to manage them.
  • The role of collaboration and mentorship: The interview emphasizes the importance of working with colleagues and the impact of mentorship on shaping research directions.

Methods

The interview details Professor Horowitz's extensive work on various econometric and statistical methods. Key techniques and approaches discussed include:

  • Bootstrap Methods: Horowitz has extensively researched the application and refinement of bootstrap methods for statistical inference, particularly in situations where traditional methods are problematic.
  • Semiparametric and Nonparametric Estimation: A significant portion of his work involves developing and applying methods that make fewer assumptions about the underlying data distributions compared to traditional parametric models. This includes:
  • Smoothed Maximum Score Estimator: A breakthrough that improved upon Manski's maximum score estimator by smoothing the objective function, leading to faster convergence rates and more tractable asymptotic distributions.
  • Nonparametric Instrumental Variables (IV) Estimation: Developing methods to handle endogeneity in models, especially when dealing with complex functional forms and discrete variables.
  • Functional Data Analysis: Applying statistical techniques to data that are functions themselves, such as time series of consumption.
  • Shape Restrictions: Imposing constraints (like monotonicity or convexity) on estimated functions to align with economic theory, often using linear programming.
  • High-Dimensional Models: Research into penalized estimation methods (like adaptive LASSO and bridge penalties) for models with a large number of potential predictors.
  • Specification Testing: Developing tests to assess the validity of econometric models against nonparametric alternatives.

Datasets

Not explicitly detailed as "datasets" in the context of training and evaluation for a specific model, but the interview mentions various applications and data sources that informed his research:

  • Income maintenance experiments: Used for duration modeling.
  • Wage data: Employed in studies of labor economics.
  • Gas pricing experiments: Utilized for functional data analysis.
  • Supermarket data (UK): Used for random coefficients logit models in marketing.
  • Individual gasoline purchase data: Used for estimating demand functions with shape restrictions.
  • German data on offered and chosen prices: Used for analyzing demand functions with Berkson errors and nested logit models.

Key Insights and Ideas

  • The "Physics Mindset" in Economics: Horowitz's transition from physics instilled a deep respect for empirical validation and the idea of theories as testable hypotheses, a perspective he found sometimes lacking in economics.
  • The Power of Smoothing: The development of the smoothed maximum score estimator was a crucial insight, demonstrating how smoothing can overcome the analytical difficulties posed by discontinuous functions in estimators.
  • Bootstrap's Versatility and Limitations: While powerful, the bootstrap's effectiveness can depend on the smoothness of the underlying objective function, leading to research into alternatives like subsampling and explicit bias reduction.
  • Shape Restrictions as a Bridge: Imposing theoretically motivated shape restrictions on nonparametric estimators can lead to partial identification and more robust inference, especially when dealing with limited data or complex models.
  • The "Crypto Statistician" Identity: Horowitz's self-description highlights his comfort in working at the intersection of econometrics and mathematical statistics, often tackling problems that are mathematically rigorous but perhaps less fashionable in mainstream econometrics.
  • The Importance of Collaboration: The interview repeatedly emphasizes the value of working with diverse colleagues, leading to significant research breakthroughs.

Key Results

  • Smoothed Maximum Score Estimator: Achieved nearly an $n^{-1/2}$ rate of convergence, a significant improvement over the original maximum score estimator's $n^{-1/3}$ rate, and provided asymptotically normal distributions.
  • Nonparametric IV Estimation: Developed methods that, while theoretically complex, offer ways to address endogeneity in situations where standard parametric IV is insufficient.
  • Shape Restrictions: Enabled partial identification of linear functionals of structural functions, leading to the use of linear programming for estimation and inference.
  • Nonparametric Additive Models: Developed methods for estimating additive components with oracle efficiency, even with unknown link functions.
  • High-Dimensional Models: Proposed penalized estimation methods (adaptive group LASSO) for nonparametric additive models that are model selection consistent and oracle efficient.

Limitations and Future Work

  • Resistance to Nonparametric Methods: Horowitz notes that applied researchers sometimes resist nonparametric methods due to their complexity and the difficulty in applying them to complicated structural models.
  • Complexity of Nonparametric IV: The theoretical work in nonparametric IV can be slow to develop, and its application is hindered by the fact that many real-world problems involve discrete variables, which are not always well-handled by existing nonparametric IV literature.
  • "Crypto Statistician" Status: While a source of pride, this also implies that some of his work might be outside the mainstream of econometric theory, potentially limiting its immediate impact.
  • Future of Nonparametric IV: Horowitz is unsure about the future impact of nonparametric IV on applied research, acknowledging the difficulty of predicting specific applications.
  • Ongoing Research: The interview mentions ongoing work on nested logit models for price choices and the development of new methods for high-dimensional models.

Summary updated

BibTeX
@inproceedings{lee2025et,
  author = {Sokbae Lee},
  title = {{THE ET INTERVIEW: PROFESSOR JOEL L. HOROWITZ}},
  booktitle = {Econometric},
  year = {2025},
  doi = {10.1017/s0266466624000380},
}
Figure