

1999  DiplomVolkswirt, University of Cologne  2000  M.Sc., Economics and Philosophy (with Distinction), LSE, London, England, UK  2001  M.A., Economics, Northwestern University, Evanston, IL  2005  PhD, Economics, Northwestern University, Evanston, IL  2005  2010  Assistant Professor, New York University, NY, USA  2009  Visiting Research Fellow, Cowles Foundation, Yale University, New Haven, CT, USA  Since 2010  Associate Professor, Cornell University, Ithaca, NY, USA (on leave, July 2016  present)  Since 2016  Professor, University of Bonn 


In past research, Stoye explored connections between statistical decision theory and applied economic analysis, with special attention to treatment/policy choice problems and to minimax regret as optimality criterion. In axiomatic analyses, he provided normative foundations for minimax regret and related criteria [6,7,8]. Parallel work in econometrics undertook finite sample (nonapproximate) analysis of econometric treatment choice problems, e.g. the use of covariates in treatment assignment [9,10,11]. Stoye also advanced the literature on partial (set valued) identifiability of parameters and on inference in such settings [12,13]. More recently (with Stefan Hoderlein, Boston College), he proposed statistical tests of revealed preference analysis from microeconomic theory, asking whether the homo oeconomicus model is testable under otherwise weak assumptions on realworld data [1,2].
Some of Stoye’s current research extends this last project. With Yuichi Kitamura (Yale), he identifies the precise empirical content of Random Utility models in repeated crosssection data, assuming unrestricted unobservable heterogeneity among consumers and therefore an infinite dimensional nuisance parameter. Their statistical test overcomes both computational and theoretical (in the form of nonstandard asymptotic behavior) hurdles. It is extended to other, less standard economics models in work with Kitamura, Rahul Deb (Toronto), and John Quah (Johns Hopkins).The longterm vision is to fundamentally rethink the large economics literature on nonparametric demand, complementing its current “specific to general” approach (i.e., imposing an extremely tight structure and maybe gradually relaxing it) with a “general to specific” approach that initially tests whether data are consistent with minimal economic assumptions and, in future research, gradually relaxes the generality to obtain tighter conclusions. In other early stage research, Stoye (with Hiroaki Kaido, Boston University, and Francesca Molinari, Cornell) develops confidence sets for the optimal values of programs with estimated objective function as well as constraints. These sets will be valid uniformly over a large class of sampling processes and without socalled constraint qualifications. Notable applications are to policy counterfactuals in economic models as well as to projections of partially identified parameter vectors and are being explored empirically.


Research Area H My recent focus has been on two issues: (i) What discipline do core economic assumptions, e.g. homo oeconomicus, directly impose on data that are realistically observed if one drops auxiliary assumptions of convenience? For example, in the empirically relevant context of repeated crosssectional data, this means to assume that all individuals in an underlying population are rational in the sense of maximizing some criterion function, but that the population distribution of criterion functions is unrestricted. This leads to nonparametric testing of extremely high dimensional models, though important dimension reductions turn out to be available in practice [1,2,3]. Extensions to less standard models of consumer behavior are in progress. (ii) How can we perform inference on lowdimensional functions of moderate to high dimensional parameters that are only partially statistically identified. Applications include separate inference on components of partially identified vectors but also value functions of optimization problems (e.g., maximization of social welfare in macroeconomic models) with estimated objective function and constraints. The inference problem is highly irregular due to the presence of nonidentifiable nuisance parameters. The solution involves powerful new regularization techniques as well as novel blackbox optimization algorithms whose convergence is established [4,5]. 


[ 1] Stefan Hoderlein, Jörg Stoye
Testing stochastic rationality and predicting stochastic demand: the case of two goods Econ. Theory Bull. , 3: (2): 313328 2015[ 2] Stefan Hoderlein, Jörg Stoye
Revealed Preferences in a Heterogeneous Population The Review of Economics and Statistics , 96: (2): 197213 2014[ 3] Yuichi Kitamura, Jörg Stoye
Nonparametric Analysis of Random Utility Models eprint arXiv:1606.04819 Revise and Resubmit at Econometrica 2016[4] Hiroaki Kaido, Francesca Molinari, Jörg Stoye
Confidence Intervals for Projections of Partially Identified Parameters eprint arXiv:1601.00934 Revise and Resubmit at Econometrica 2016 [5] Hiroaki Kaido, Francesca Molinari, Jörg Stoye, Matthew Thirkettle
Calibrated Projection in MATLAB: Users' Manual eprint arXiv:1710.09707 2017 [6] Jörg Stoye
Choice theory when agents can randomize J. Econom. Theory , 155: : 131151 2015 [ 7] Jörg Stoye
Axioms for minimax regret choice correspondences J. Econom. Theory , 146: (6): 22262251 2011[ 8] Jörg Stoye
Statistical decisions under ambiguity Theory and Decision , 70: (2): 129148 2011[ 9] Jörg Stoye
Minimax regret treatment choice with covariates or with limited validity of experiments J. Econometrics , 166: (1): 138156 2012[ 10] Jörg Stoye
Minimax regret treatment choice with finite samples J. Econometrics , 151: (1): 7081 2009[ 11] Jörg Stoye
Minimax regret treatment choice with incomplete data and many treatments Econometric Theory , 23: (1): 190199 2007[ 12] Jörg Stoye
Partial identification of spread parameters Quant. Econ. , 1: (2): 323357 2010[ 13] Jörg Stoye
More on confidence intervals for partially identified parameters Econometrica , 77: (4): 12991315 2009





• Review of Economics and Statistics (since 2014)



Download Profile 