pith. sign in

arxiv: 2606.08359 · v1 · pith:MV2WUCP6new · submitted 2026-06-06 · 💰 econ.EM

Adaptive Estimation of Aggregated Values of Conditional Linear Programs

Gevorg Khandamiryan , Vira Semenova This is my paper

Pith reviewed 2026-06-27 18:30 UTC · model grok-4.3

classification 💰 econ.EM
keywords partial identificationsupport functionregression aggregationconditional linear programsasymptotic theorytreatment effect boundsunder-identified systems
0
0 comments X

The pith

The support function of the identified set equals an average of intersections of regression functions over the covariate distribution.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a covariate-assisted estimator for partially identified parameters that solve under-identified linear equations with known fixed coefficients. It shows that the boundary of the identified set, expressed as a support function, takes the form of an average of intersections of regression functions aggregated over the covariate distribution. This representation establishes the boundary as a regular parameter. Standard asymptotic theory then applies, covering applications such as treatment effect bounds and choice models.

Core claim

The boundary (support function) of the proposed identified set is represented as an average of intersections of regression functions, aggregated over the covariate distribution. The boundary is a regular parameter, and asymptotic theory is proposed for its estimation.

What carries the argument

Average of intersections of regression functions aggregated over the covariate distribution, which equals the support function of the identified set.

If this is right

  • Asymptotic normality holds for the estimator of the support function.
  • Confidence intervals can be constructed for the boundaries of the identified set.
  • The method applies directly to bounding treatment effects and models with state dependence.
  • Inference is available for parameters in IV choice models and random utility models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same aggregation step may simplify computation of conditional linear programs in other partially identified settings.
  • Pairing the regression intersections with flexible nonparametric estimators could extend the approach to high-dimensional covariates.
  • The representation suggests testable restrictions on how covariate variation affects the width of identified sets.

Load-bearing premise

The parameters of interest are solutions to an under-identified system of linear equations whose coefficients are known and fixed.

What would settle it

A simulation study with data generated from an under-identified linear system where the estimator of the support function fails to converge at the root-n rate would falsify the regularity claim.

Figures

Figures reproduced from arXiv: 2606.08359 by Gevorg Khandamiryan, Vira Semenova.

Figure 1
Figure 1. Figure 1: Expansion of the boxed coarse entries into granular blocks         β2n,1r β0r,2n −1 +1 0 0 0 0 0 0 0 0 0 0 0 0 +1 −1 0 0 0 0 0 0 0 0 0 0 −1 0 −1 −1 −1 0 −1 0 0 0 0 0 0 0 0 +1 −1 2n −1 +1         ⇓          · · · βb6n,b1r βb6n,b2r βb6n,b3r βb6n,b4r βb6n,b5r βb7n,b1r βb7n,b2r βb7n,b3r βb7n,b4r βb7n,b5r βb8n,b1r βb8n,b2r βb8n,b3r βb8n,b4r βb8n,b5r · · · β0r,b6n β0r,b7n β0r,b8n · · · .… view at source ↗
read the original abstract

We develop a covariate-assisted approach to partially identified parameters that are solutions to an under-identified system of linear equations with known coefficients. Examples include bounds on treatment effects, models of unemployment with state dependence, choice-theoretic models of IV, and random utility models. The boundary (i.e., support function) of the proposed identified set is represented as an average of intersections of regression functions, aggregated over the covariate distribution. We show that the boundary is a regular parameter, propose asymptotic theory, and demonstrate using an empirical application to Jobs First.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper develops a covariate-assisted estimator for the boundary (support function) of partially identified parameters that solve under-identified linear systems with known coefficients. It represents this boundary as an average, over the covariate distribution, of pointwise intersections (min or max) of regression functions, claims that the resulting functional is a regular parameter, derives asymptotic theory for its estimator, and illustrates the method with an application to the Jobs First experiment.

Significance. If the regularity claim holds under explicit conditions, the representation would allow standard semiparametric tools to be applied to a class of partial-identification problems (treatment-effect bounds, state-dependence models, IV choice models, random-utility models) that are otherwise difficult to estimate sharply with covariates. The empirical illustration shows the approach is implementable, but the overall contribution hinges on whether the asymptotic theory survives the non-smoothness inherent in the intersection operator.

major comments (3)
  1. [§3] §3 (Representation of the support function): the boundary is defined as an average of pointwise min/max intersections of regression functions. The min/max operator is not Hadamard differentiable at crossing points where the underlying regressions have distinct slopes. No conditions are stated ensuring that the set of crossings has measure zero under the covariate distribution or that the crossing locations are independent of the estimators in a way that restores differentiability. This directly affects the claim that the boundary is a regular parameter.
  2. [§4] §4 (Asymptotic theory): the derivation of asymptotic normality and the influence function appears to rely on the regularity established in §3. Because the differentiability step is not secured when crossings occur on a positive-measure set, the validity of the asymptotic distribution and the proposed inference procedures cannot be assessed from the given argument.
  3. [§2.1] Abstract and §2.1 (model setup): the target is defined as the boundary of the identified set for solutions to an under-identified linear system whose coefficients are known and fixed. The paper does not supply an explicit verification that the proposed representation recovers exactly this boundary for the listed examples (e.g., treatment-effect bounds or random-utility models) without additional restrictions on the support of the covariates.
minor comments (2)
  1. [§3] Notation for the intersection operator (min versus max) is used interchangeably in the abstract and §3 without a uniform definition; a single consistent symbol would improve readability.
  2. [§5] The empirical application section would benefit from a table reporting the estimated bounds together with the associated standard errors obtained from the proposed asymptotic theory.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. We address each major comment below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [§3] §3 (Representation of the support function): the boundary is defined as an average of pointwise min/max intersections of regression functions. The min/max operator is not Hadamard differentiable at crossing points where the underlying regressions have distinct slopes. No conditions are stated ensuring that the set of crossings has measure zero under the covariate distribution or that the crossing locations are independent of the estimators in a way that restores differentiability. This directly affects the claim that the boundary is a regular parameter.

    Authors: We agree that the Hadamard differentiability of the min/max operator requires an explicit condition to handle crossing points. In the revised manuscript we will add an assumption that the covariate distribution places measure zero on the set of points at which any pair of the relevant regression functions intersect. Under this condition the intersection functional is Hadamard differentiable almost everywhere with respect to the covariate measure, which restores regularity of the aggregated support function. This is a standard technical requirement in the literature on nonsmooth functionals and does not require additional independence between crossing locations and the estimators. revision: yes

  2. Referee: [§4] §4 (Asymptotic theory): the derivation of asymptotic normality and the influence function appears to rely on the regularity established in §3. Because the differentiability step is not secured when crossings occur on a positive-measure set, the validity of the asymptotic distribution and the proposed inference procedures cannot be assessed from the given argument.

    Authors: The asymptotic normality and influence-function results in §4 are obtained from the representation and regularity properties derived in §3. Once the measure-zero crossing assumption is added, the Hadamard differentiability holds and the existing derivations remain valid. In the revision we will explicitly link the new assumption to the statements in §4 so that the conditions for the asymptotic theory are stated completely. revision: yes

  3. Referee: [§2.1] Abstract and §2.1 (model setup): the target is defined as the boundary of the identified set for solutions to an under-identified linear system whose coefficients are known and fixed. The paper does not supply an explicit verification that the proposed representation recovers exactly this boundary for the listed examples (e.g., treatment-effect bounds or random-utility models) without additional restrictions on the support of the covariates.

    Authors: We will add a short appendix that supplies the explicit verification for the main examples (treatment-effect bounds and random-utility models). The verification will confirm that the average-of-intersections representation recovers the support function of the identified set under the model assumptions already stated in §2.1, without imposing further restrictions on the support of the covariates. revision: yes

Circularity Check

0 steps flagged

No circularity; representation and regularity claim are independent of fitted inputs

full rationale

The abstract introduces a representation of the support function as an average of regression intersections and then claims to establish regularity for asymptotic theory. No quoted step reduces the target parameter to a fitted quantity defined by the same data, nor does any self-citation serve as the sole justification for the central regularity result. The derivation chain therefore remains self-contained against external semiparametric benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that the linear system coefficients are known and that regression functions can be estimated from observable data; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Coefficients of the linear system are known and fixed
    Stated directly in the abstract as part of the problem setup for under-identified systems.

pith-pipeline@v0.9.1-grok · 5608 in / 1172 out tokens · 31400 ms · 2026-06-27T18:30:19.711620+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

299 extracted references · 60 canonical work pages · 1 internal anchor

  1. [1]

    Review of Economic Studies , year =

    Rambachan, Ashesh and Roth, Jonathan , title =. Review of Economic Studies , year =

  2. [2]

    Efficient and Convergent Sequential Pseudo-Likelihood Estimation of Dynamic Discrete Games , volume=

    Dearing, Adam and Blevins, Jason R , year=. Efficient and Convergent Sequential Pseudo-Likelihood Estimation of Dynamic Discrete Games , volume=. Review of Economic Studies , publisher=. doi:10.1093/restud/rdae050 , number=

    work page doi:10.1093/restud/rdae050

  3. [3]

    and Lorentz, George G

    DeVore, Ronald A. and Lorentz, George G. , title =

  4. [4]

    Econometrica , volume =

    Magnac, Thierry and Thesmar, David , title =. Econometrica , volume =

  5. [5]

    and Souza-Rodrigues, Eduardo , title =

    Kalouptsidi, Myrto and Scott, Paul T. and Souza-Rodrigues, Eduardo , title =. Quantitative Economics , volume =

  6. [6]

    American Economic Review , year =

    Kalouptsidi, Myrto , title =. American Economic Review , year =

  7. [7]

    Review of Economic Studies , year =

    Kalouptsidi, Myrto , title =. Review of Economic Studies , year =

  8. [8]

    Review of Economic Studies , year =

    Kalouptsidi, Myrto and Kitamura, Yuichi and Lima, Lucas and Souza-Rodrigues, Eduardo , title =. Review of Economic Studies , year =

  9. [9]

    and Souza-Rodrigues, Eduardo , title =

    Kalouptsidi, Myrto and Scott, Paul T. and Souza-Rodrigues, Eduardo , title =. International Journal of Industrial Organization , volume =

  10. [10]

    Greenwood, P. E. and Wefelmeyer, W. , title =. The Annals of Statistics , year =

  11. [11]

    On Bunching and Identification of the Taxable Income Elasticity , journal =

    Blomquist, S. On Bunching and Identification of the Taxable Income Elasticity , journal =. 2021 , doi =

    2021

  12. [12]

    American Economic Journal: Economic Policy , year =

    Saez, Emmanuel , title =. American Economic Journal: Economic Policy , year =

  13. [13]

    Confidence Intervals for Partially Identified Parameters , Volume =

    Imbens, Guido and Manski, Charles , Journal =. Confidence Intervals for Partially Identified Parameters , Volume =

  14. [14]

    2025 , eprint=

    Method-of-Moments Inference for GLMs and Doubly Robust Functionals under Proportional Asymptotics , author=. 2025 , eprint=

    2025

  15. [15]

    Econometric Theory , volume =

    Song, Kyungchul , title =. Econometric Theory , volume =

  16. [16]

    Quantitative Economics , year =

    Kline, Patrick and Santos, Andres , title =. Quantitative Economics , year =

  17. [17]

    , title =

    Manski, Charles F. , title =

  18. [18]

    Minimax Regret Treatment Choice with Covariates or with Limited Validity of Experiments , journal =

    Stoye, J. Minimax Regret Treatment Choice with Covariates or with Limited Validity of Experiments , journal =. 2012 , volume =

    2012

  19. [19]

    and McFadden, Daniel , title =

    Newey, Whitney K. and McFadden, Daniel , title =. Handbook of Econometrics , editor =. 1994 , pages =

    1994

  20. [20]

    and Poirier, Alexandre , title =

    Masten, Matthew A. and Poirier, Alexandre , title =. Econometrica , year =

  21. [21]

    Masten and Alexandre Poirier , title =

    Matthew A. Masten and Alexandre Poirier , title =. Journal of Econometrics , year =

  22. [22]

    Donoho and Richard C

    David L. Donoho and Richard C. Liu , title =. The Annals of Statistics , year =

  23. [23]

    and Nagin, Daniel S

    Manski, Charles F. and Nagin, Daniel S. , title =. Sociological Methodology , year =

  24. [24]

    , title =

    Stancu-Minasian, Ioan M. , title =

  25. [25]

    , title =

    Andrews, Isaiah and Shapiro, Jesse M. , title =. Econometrica , year =

  26. [26]

    and Evans, William N

    Angrist, Joshua D. and Evans, William N. , title =. American Economic Review , volume =

  27. [27]

    Quantitative Economics , volume =

    Tolerating Defiance?. Quantitative Economics , volume =

  28. [28]

    and Wolpin, Kenneth I

    Rosenzweig, Mark R. and Wolpin, Kenneth I. , title =. Journal of Economic Literature , volume =

  29. [29]

    Conference on Learning Theory , pages=

    Sparse sketches with small inversion bias , author=. Conference on Learning Theory , pages=. 2021 , organization=

    2021

  30. [30]

    Journal of Machine Learning Research , volume=

    Unbiased estimators for random design regression , author=. Journal of Machine Learning Research , volume=

  31. [31]

    Murphy , title =

    Susan A. Murphy , title =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , year =

  32. [32]

    Handbook of the Economics of Education , editor =

    Bruce Sacerdote , title =. Handbook of the Economics of Education , editor =. 2011 , volume =. doi:10.1016/B978-0-444-53429-3.00004-1 , url =

    work page doi:10.1016/b978-0-444-53429-3.00004-1 2011

  33. [33]

    Peer Effects in the Workplace , journal =

    Thomas Cornelissen and Christian Dustmann and Uta Sch. Peer Effects in the Workplace , journal =. 2017 , volume =. doi:10.1257/aer.20141300 , url =

    work page doi:10.1257/aer.20141300 2017

  34. [34]

    The Economic Journal , year =

    Giacomo De Giorgi and Michele Pellizzari , title =. The Economic Journal , year =. doi:10.1111/ecoj.12165 , volume =

    work page doi:10.1111/ecoj.12165

  35. [35]

    The Review of Economic Studies , year =

    Leonardo Bursztyn and Georgy Egorov and Robert Jensen , title =. The Review of Economic Studies , year =. doi:10.1093/restud/rdy026 , url =

    work page doi:10.1093/restud/rdy026

  36. [36]

    Donoho and Jiashun Jin , title =

    David L. Donoho and Jiashun Jin , title =. Annals of Statistics , volume =. 2004 , doi =

    2004

  37. [37]

    Tsybakov and Nicolas Verzelen , title =

    Yuri Ingster and Alexandre B. Tsybakov and Nicolas Verzelen , title =. Electronic Journal of Statistics , volume =. 2010 , doi =

    2010

  38. [38]

    Ingster , title =

    Yuri I. Ingster , title =. Mathematical Methods of Statistics , volume =

  39. [39]

    Ingster and Irina A

    Yuri I. Ingster and Irina A. Suslina , title =. 2002 , series =

    2002

  40. [40]

    Annals of Statistics , volume =

    Emmanuel Arias-Castro and Emmanuel Candès and Yaniv Plan , title =. Annals of Statistics , volume =. 2011 , doi =

    2011

  41. [41]

    Review of Economics and Statistics , volume=

    Short and simple confidence intervals when the directions of some effects are known , author=. Review of Economics and Statistics , volume=. 2025 , publisher=

    2025

  42. [42]

    arXiv preprint arXiv:2409.09962 , year=

    A Simple and Adaptive Confidence Interval when Nuisance Parameters Satisfy an Inequality , author=. arXiv preprint arXiv:2409.09962 , year=

    arXiv

  43. [43]

    2017 , booktitle =

    Deep Sets , author =. 2017 , booktitle =

    2017

  44. [44]

    Journal of the American Statistical Association , year =

    Han, Sukjin , title =. Journal of the American Statistical Association , year =

  45. [45]

    Soumya Ganguly and Khoa Tran and Rahul Sarkar , year=. Uniform. 2403.01339 , archivePrefix=

    arXiv

  46. [46]

    Journal of Econometrics , year =

    Han, Sukjin and Yang, Shenshen , title =. Journal of Econometrics , year =

  47. [47]

    Robins , title =

    James M. Robins , title =. Proceedings of the Second Seattle Symposium in Biostatistics , editor =. 2004 , doi =

    2004

  48. [48]

    The Annals of Statistics , year =

    Chengchun Shi and Alin Fan and Rui Song and Wenbin Lu , title =. The Annals of Statistics , year =

  49. [49]

    2025 , eprint=

    Robust Identification in Randomized Experiments with Noncompliance , author=. 2025 , eprint=

    2025

  50. [50]

    Lehmann and Joseph P

    Erich L. Lehmann and Joseph P. Romano , title =. 2005 , series =

    2005

  51. [51]

    2025 , eprint=

    Policy Learning with Confidence , author=. 2025 , eprint=

    2025

  52. [52]

    2024 , journal =

    Heterogeneous Treatment Effect Bounds under Sample Selection with an Application to the Effects of Social Media on Political Polarization , author =. 2024 , journal =

    2024

  53. [53]

    2024 , eprint=

    Treatment Evaluation at the Intensive and Extensive Margins , author=. 2024 , eprint=

    2024

  54. [54]

    Annals of Operations Research , year =

    Alexander Shapiro , title =. Annals of Operations Research , year =

  55. [55]

    On Nondifferentiable Functions and the Bootstrap , journal =

    Lutz D. On Nondifferentiable Functions and the Bootstrap , journal =. 1993 , volume =

    1993

  56. [56]

    2025 , eprint=

    Lee Bounds with a Continuous Treatment in Sample Selection , author=. 2025 , eprint=

    2025

  57. [57]

    Econometrica , volume =

    Identifying Effects of Multivalued Treatments , author =. Econometrica , volume =. 2018 , month =

    2018

  58. [58]

    Econometrica , volume =

    Pietro Tebaldi and Alexander Torgovitsky and Hanbin Yang , title =. Econometrica , volume =. doi:https://doi.org/10.3982/ECTA17215 , url =

    work page doi:10.3982/ecta17215

  59. [59]

    Econometrica , volume =

    Alexander Torgovitsky , title =. Econometrica , volume =. doi:https://doi.org/10.3982/ECTA14138 , url =

    work page doi:10.3982/ecta14138

  60. [60]

    2014 , month = sep, journal =

    Identification and Counterfactuals in Dynamic Models of Market Entry and Exit , author =. 2014 , month = sep, journal =

    2014

  61. [61]

    1997 , month = jun, journal =

    The Career Decisions of Young Men , author =. 1997 , month = jun, journal =. doi:10.1086/262082 , url =

    work page doi:10.1086/262082 1997

  62. [62]

    1998 , month = may, journal =

    Paying for Health Insurance: The Tradeoff between Competition and Adverse Selection , author =. 1998 , month = may, journal =. doi:10.1162/003355398555649 , url =

    work page doi:10.1162/003355398555649 1998

  63. [63]

    2010 , month = sep, journal =

    Estimating Welfare in Insurance Markets Using Variation in Prices , author =. 2010 , month = sep, journal =. doi:10.1257/aer.100.4.193 , url =

    work page doi:10.1257/aer.100.4.193 2010

  64. [64]

    2009 , journal =

    Sufficient Statistics for Welfare Analysis: A Bridge Between Structural and Reduced-Form Methods , author =. 2009 , journal =. doi:10.1146/annurev.economics.050708.143323 , url =

    work page doi:10.1146/annurev.economics.050708.143323 2009

  65. [65]

    1984 , month = dec, journal =

    Job Matching and Occupational Choice , author =. 1984 , month = dec, journal =. doi:10.1086/261272 , url =

    work page doi:10.1086/261272 1984

  66. [66]

    2019 , month = jul, journal =

    Nonstationary Dynamic Models with Finite Dependence , author =. 2019 , month = jul, journal =. doi:10.3982/QE1021 , url =

    work page doi:10.3982/qe1021 2019

  67. [67]

    Duality in dynamic discrete-choice models: Duality in dynamic discrete-choice models , volume=

    Chiong, Khai Xiang and Galichon, Alfred and Shum, Matt , year=. Duality in dynamic discrete-choice models: Duality in dynamic discrete-choice models , volume=. Quantitative Economics , publisher=. doi:10.3982/qe436 , number=

    work page doi:10.3982/qe436

  68. [68]

    2021 , journal =

    Multiclass classification by sparse multinomial logistic regression , author =. 2021 , journal =

    2021

  69. [69]

    2020 , month = dec, journal =

    Screening and Selection: The Case of Mammograms , author =. 2020 , month = dec, journal =. doi:10.1257/aer.20190632 , url =

    work page doi:10.1257/aer.20190632 2020

  70. [70]

    Tax Incentives and the Supply of Low-Income Housing , author =

  71. [71]

    2016 , month = oct, journal =

    Solving the Problem of Overdiagnosis , author =. 2016 , month = oct, journal =. doi:10.1056/NEJMp1610811 , url =

    work page doi:10.1056/nejmp1610811 2016

  72. [72]

    2009 , month = oct, journal =

    Rethinking Screening for Breast Cancer and Prostate Cancer , author =. 2009 , month = oct, journal =. doi:10.1001/jama.2009.1498 , url =

    work page doi:10.1001/jama.2009.1498 2009

  73. [73]

    2024 , journal =

    Identification of Average Marginal Effects in Fixed Effects Dynamic Discrete Choice Models , author =. 2024 , journal =

    2024

  74. [74]

    2022 , month = apr, journal =

    Equilibrium Trade in Automobiles , author =. 2022 , month = apr, journal =. doi:10.1086/720463 , url =

    work page doi:10.1086/720463 2022

  75. [75]

    2015 , month = mar, journal =

    Nonparametric Welfare Analysis for Discrete Choice , author =. 2015 , month = mar, journal =. doi:10.3982/ECTA12574 , url =

    work page doi:10.3982/ecta12574 2015

  76. [76]

    2024 , month = jun, journal =

    Nonparametric Approaches to Empirical Welfare Analysis , author =. 2024 , month = jun, journal =. doi:10.1257/jel.20221534 , url =

    work page doi:10.1257/jel.20221534 2024

  77. [77]

    2024 , month = apr, journal =

    Demand and Welfare Analysis in Discrete Choice Models with Social Interactions , author =. 2024 , month = apr, journal =. doi:10.1093/restud/rdad053 , url =

    work page doi:10.1093/restud/rdad053 2024

  78. [78]

    Journal of the American Statistical Association , volume =

    The Generalized Oaxaca-Blinder Estimator , author =. 2021 , month = dec, journal =. doi:10.1080/01621459.2021.1941053 , url =

    work page doi:10.1080/01621459.2021.1941053 2021

  79. [79]

    2023 , journal =

    A Simple and General Debiased Machine Learning Theorem with Finite Sample Guarantees , author =. 2023 , journal =

    2023

  80. [80]

    2022 , eprint=

    Finite-Sample Guarantees for High-Dimensional DML , author=. 2022 , eprint=

    2022

Showing first 80 references.