Pith. sign in

REVIEW 2 major objections 4 minor 30 references

A new profiled sieve minimum-distance estimator achieves uniform consistency for semi-nonparametric demand systems estimated from micro-level choice data.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 07:47 UTC pith:7UA6VOKK

load-bearing objection A real estimation counterpart to Berry-Haile micro-data identification, with a plausible main theorem that currently has a gap in the sampling-rate assumptions; deserves a serious referee, not a desk reject. the 2 major comments →

arxiv 2607.21323 v2 pith:7UA6VOKK submitted 2026-07-23 econ.EM stat.ME

Uniformly Consistent Semi-nonparametric Demand Estimation with Micro-Data

Richard Grigorian This is my paper
classification econ.EM stat.ME
keywords differentiated-products demandmicro datasieve minimum distancenonparametric IVuniform consistencyincidental parametersBerry-Haile modeldemand estimation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that a flexible, semi-nonparametric demand system can be estimated from micro-level choice data in a way that is uniformly consistent: as the number of markets and the number of consumers per market both grow, the estimated consumer-heterogeneity function and price-side function converge to their true counterparts across the entire support, not just at single points. The author constructs a profiled sieve minimum-distance estimator that combines within-market variation in consumer covariates (which identifies heterogeneity and a composite market intercept) with cross-market price-instrument variation (which separates the intercept into a flexible price function and structural demand shocks). If the main theorem holds, estimated demand surfaces can be used for plug-in counterfactual objects — shares, elasticities, diversion ratios, markups, welfare — that are built on a consistently estimated structural demand surface. The Monte Carlo evidence supports the consistency claim and shows the value of flexible price-side estimation in counterfactual exercises.

Core claim

The central discovery is a new asymptotic result for demand estimation: under the semi-nonparametric logit model with a flexible consumer-heterogeneity function g0 and a flexible price-side function ψ0, the profiled SMD estimator satisfies ||ĝ - g0||_∞ + ||ψ̂ - ψ0||_∞ = o_p(1), with max_t ||â_t - a0t|| = o_p(1) and max_t ||ĥ_t - h0t|| = o_p(1). The composite market intercept a0t = ψ0(Pt,Xt)+h0t is profiled market-by-market via a convex share-matching problem, and then the price-side conditional moment E[a0t - ψ0(Pt,Xt)|Wt,Xt]=0 is used to decompose it. The proof handles the incidental-parameters problem by letting the within-market sample size grow, so each intercept is estimated with increa

What carries the argument

The central mechanism is the composite market intercept a0t := ψ0(Pt,Xt)+h0t, which isolates the part of demand that is constant within a market. The estimator profiles this intercept for each candidate heterogeneity function by matching predicted and observed within-market shares (a convex log-sum-exp minimization), then uses the resulting profiled residual to form micro moments for g0. The price-side block applies a sieve minimum-distance (nonparametric IV) step to the conditional moment E[a0t - ψ0(Pt,Xt)|Wt,Xt]=0, projecting the implied shock onto a growing basis of instruments. The uniformity of the consistency result comes from controlling the profile error uniformly over candidate func

Load-bearing premise

The load-bearing premise is the completeness condition (Assumption 4.3(ii)): once prices and other market characteristics are controlled for, no nonzero price-side function can have zero conditional mean given the excluded instruments; if this fails, the price function ψ0 is not identified and the uniform-consistency target is not well-defined.

What would settle it

A simulation that satisfies all other assumptions but sets the excluded instruments to be independent of prices (so E[F(Pt,Xt)|Wt,Xt]=0 for many F, violating completeness) should make the price-side SMD criterion flat and the estimated ψ̂ fail to converge to ψ0; observing such non-convergence would show that completeness is doing the identification work.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the main theorem is correct, plug-in estimates of counterfactual market shares, own/cross elasticities, diversion ratios, markups, and welfare changes are based on a consistently estimated demand surface, improving reliability of IO counterfactuals.
  • The profiling step shows that the incidental-parameters problem disappears when each market intercept is estimated from a growing number of consumers, so the number of markets can grow freely.
  • The flexible price-side function ψ0 can capture nonlinear price effects that linear random-coefficients specifications miss, which the Monte Carlo shows improves counterfactual accuracy even when heterogeneity is simple.
  • The result provides an estimation counterpart to the recent micro-data identification theorem, turning nonparametric identification into uniform consistency under the logit link.

Where Pith is reading between the lines

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

  • The profiling-plus-SMD architecture is not logically tied to the logit link; a similar two-step logic (profile market intercepts, then decompose with instruments) could be applied to other tractable discrete-choice links, provided identification and completeness hold.
  • Because the completeness assumption is strong and not stress-tested in the Monte Carlo, the practical reliability of the price-side decomposition depends on instrument strength; future applications should report diagnostics for completeness.
  • The paper stops at consistency; my inference is that a bias-corrected version of the profiled estimator will be needed for valid inference on counterfactual objects, since the noise from estimating market intercepts is unlikely to be harmless for confidence sets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. This paper develops a profiled sieve minimum-distance estimator for a semi-nonparametric multinomial-logit demand model with micro data, building on Berry and Haile (2024). Within each market, the composite intercept a0t = ψ0(Pt, Xt) + h0t is profiled by matching observed and predicted shares; the price-side function ψ0 is then estimated by a conditional-moment (NPIV/SMD) step using excluded instruments. The central result (Theorem 6.7) states that, under high-level identification, sieve, and uniform-convergence assumptions, the estimator is uniformly consistent for g0 and ψ0, and the profiled intercepts and structural shocks are uniformly consistent over markets, when T → ∞ and n_min := min_t n_t → ∞. Appendices provide identification proofs, primitive routes, Monte Carlo evidence, and IO counterfactual formulas.

Significance. If the result holds, the paper provides a practical estimation counterpart to Berry and Haile's nonparametric identification theorem for differentiated-product demand with micro data. The profiling argument is a clean way to handle the incidental parameters generated by growing market effects, and the paper is unusually explicit about the moving-sieve structure of the criterion and about separating identification, profiling, and SMD convergence. The Monte Carlo counterfactual comparisons usefully isolate the value of flexible price-side estimation. The main weakness is that the advertised double-asymptotic profiling result rests on a uniform-in-t uniform law of large numbers that is not derived from the sampling assumption; this is fixable but central.

major comments (2)
  1. [Assumptions 3.3 and 6.2 / D.1(iii); Proposition 6.3; Theorem 6.7] The double-asymptotic framework only imposes n_min → ∞, but Proposition 6.3--and hence the max_t conclusions of Theorem 6.7--uses Assumption 6.2/D.1(iii), a uniform-in-t uniform LLN for the inner criterion. For independent markets with n_t observations, sup_{g,a} |Lhat_t(g,a)-L_t(g,a)| is typically of order sqrt(log T / n_min) after maximization over t. If n_min = log log T, this is not o_p(1), so no sieve dimension sequence can make the E.8/E.13 profile error e_T times a growing basis norm vanish. The theorem therefore needs either an explicit rate condition such as n_min/log T → ∞ with a proof of D.1(iii) from primitives, or the statements concerning max_t ||â_t-a0t|| and max_t ||ĥ_t-h0t|| must be weakened to average L2 versions. As written, the advertised result is conditional on a high-level profiling assumption that is not shown to be compatible with Assumption 3.3.
  2. [Assumption 6.5 / Appendix E] Assumption 6.5 is stated as a paragraph rather than as a formal condition, and its primitive components (E.6, E.8, E.10, E.13) are themselves high-level uniform laws and stochastic-equicontinuity conditions. The paper does not give explicit growth-rate restrictions on (K_g, K_ψ, K_q, T, n_min) under which all of these components hold simultaneously. In particular, E.8 requires e_T times the empirical second moment of the growing basis to vanish, and E.10(v,vii) require strong uniform projection and uniform-law properties that are not derived from more primitive entropy or moment conditions. The main theorem would be considerably more convincing if it stated a set of explicit, verifiable rate conditions, or at least a theorem showing the existence of admissible sieve sequences under the stated sampling scheme.
minor comments (4)
  1. [Section 7.1-7.2 / Appendix H.2] The baseline consistency simulation fixes K_g, K_ψ, K_q and uses a correctly specified finite sieve. It therefore does not exercise the sieve-growth, approximation-error, or moving-criterion components of Theorem 6.7. The growing-sieve diagnostic in H.6 supplies oracle intercepts, so it isolates only the price-side sieve. I suggest describing Figure 1 as a finite-sample illustration of the profiling logic rather than as a verification of the paper's uniform consistency theorem.
  2. [Table 2 / Section 7.3] The full-regime table shows that the sieve estimator has the largest elasticity error (5.1783) and does not win on pass-through or merger. The main text highlights share changes, diversion, and welfare, which is appropriate, but it should explicitly acknowledge that the flexible estimator is not uniformly dominant once generated-object and extrapolation error are present. The appendix's discussion is honest; the main text should be equally explicit.
  3. [Section 5.4 / Appendix H.3] The theoretical criterion (5.19) uses a uniformly positive definite weighting matrix Σhat(R_t), but the reported baseline implementation uses identity weighting in the price-side block. The relationship between the theory and the implementation should be clarified, since the finite-sample behavior of the estimator may depend on this choice.
  4. [Section 6.5] Assumption 6.5 is a compound high-level condition. Even if the consistency proof is valid under it, the paper should at least state explicit sufficient rate constraints on the sieve dimensions and sample sizes, and explain why an admissible sequence exists. As written, the reader cannot verify that the assumptions are consistent with one another.

Circularity Check

0 steps flagged

No significant circularity: the consistency result is conditional on explicit identification and stochastic-equicontinuity assumptions, with primitive sufficient conditions supplied in the appendix; the core identification theorem is borrowed from external prior work, not from the author's own results.

full rationale

The paper's derivation chain is not circular in the sense of defining the target by the estimator or relabeling a fitted object as a prediction. The maintained demand model (3.7) is an assumption, and the price-side identification argument in Theorem 4.4 rests on the external Berry–Haile (2024) instrument/completeness conditions (Assumption 4.3), not on a self-citation by the present author. The profiled composite intercept a0t is defined as the sum ψ0(Pt,Xt)+h0t, and the price-side moment E[a0t−ψ0(Pt,Xt)|Wt,Xt]=0 is derived from that definition plus the normalization E[h0t|Xt]=0 and the instrument assumption; this is a definitional moment at the truth, but uniqueness of ψ0 is then obtained through the completeness assumption, so no free identification is smuggled in. The main consistency theorem (Theorem 6.7) is proved by a standard extremum-estimation argument: uniform criterion convergence (Proposition 6.6) plus moving population separation (Assumption 6.4). Assumptions 6.4 and 6.5 are high-level, and Assumption 6.4 is indeed close in form to the conclusion, but the paper supplies primitive sufficient conditions in the appendix (Propositions E.3 and E.5, and Assumptions E.6–E.13 with Propositions E.7–E.15). Proposition 6.3, on the uniform consistency of the profiled intercept, is derived from the explicit uniform convergence and separation conditions in Assumption 6.2 and Appendix D, not merely restated. The true-profile identity a*_t(g0)=a0t is proved in Proposition D.5 from a stated market-share compatibility condition rather than imposed by definition. The Monte Carlo section is explicitly a finite-sample diagnostic and is not used as the proof of the theorem; the correctly-specified finite-sieve design is acknowledged as a design choice, not hidden as a general result. There are no self-citations by the author that are load-bearing. The skeptic's concern about n_min needing to grow faster than log T is a regularity/rate gap between Assumption 3.3 and the high-level uniformity assumptions; that is a potential correctness issue, not a circularity issue, because the proof does not claim this rate follows from Assumption 3.3 alone. Overall, no circular step satisfying the required quote-and-reduction standard is present.

Axiom & Free-Parameter Ledger

1 free parameters · 6 axioms · 0 invented entities

No new particles, forces, mediators, or economic entities are introduced. The composite market intercept a0t is a definitional reparameterization (eq. 3.8), not an additional primitive. The central claim rests on strong identification and regularity assumptions: logit correctness, completeness, instrument validity, compact Holder parameter spaces, and high-level uniform-convergence conditions.

free parameters (1)
  • Sieve tuning dimensions (Kg, Kpsi, Kq) = Not fixed; grow with T and n_min; chosen by cross-validation in the Monte Carlo
    The estimator and the asymptotic theorem depend on researcher-chosen sieve and conditional-moment dimensions. They are tuning sequences rather than fitted constants, but they are chosen by hand and affect finite-sample performance.
axioms (6)
  • domain assumption The multinomial-logit link with additive separable index is correctly specified (Assumption 3.2, eq. 3.5).
    The entire consistency argument is relative to this semi-nonparametric model; if the true link is not logit, the target functions cannot be recovered as stated.
  • domain assumption Completeness of the conditional distribution of prices given instruments: E[F(Pt,Xt)|Wt,Xt]=0 implies F=0 (Assumption 4.3(ii)).
    This converts the price-side conditional moment into identification of psi0; without it, psi0 is not separately identified from h0t.
  • domain assumption Excluded instruments are valid: E[h0t | Wt, Xt] = E[h0t | Xt] and E[h0t | Xt] = 0 (Assumption 4.1(i) and 4.3(i)).
    Centers the price-side moment at zero and is inherited from the Berry-Haile instrument assumption.
  • domain assumption Markets are i.i.d. and both the number of markets and the minimum within-market sample size grow: T to infinity and min_t n_t to infinity (Assumption 3.3).
    This double-asymptotic sampling scheme is what makes the incidental market intercepts asymptotically harmless.
  • domain assumption The parameter space is a compact normalized Holder ball and the sieves approximate with criterion preservation (Assumption 6.1 / E.1-E.2).
    Compactness and sieve approximation are needed for the extremum-estimation consistency argument and for the moving-criterion population problem.
  • domain assumption Moving population separation and uniform convergence of the profiled moving criterion (Assumptions 6.4-6.5, with primitive routes in Appendix E).
    These high-level conditions identify the truth from the population criterion and control the feasible criterion uniformly; Appendix E states them as assumptions underlying the main theorem rather than fully deriving all parts from primitives.

pith-pipeline@v1.3.0-alltime-deepseek · 47560 in / 11069 out tokens · 117875 ms · 2026-08-01T07:47:12.586259+00:00 · methodology

0 comments
read the original abstract

This paper develops a profiled sieve minimum-distance estimator for a semi-nonparametric differentiated-products demand model with micro-level choice data. Building on Berry and Haile (2024), the estimator uses within-market variation in consumer covariates to recover a flexible consumer-heterogeneity function and market-specific composite intercepts. Excluded price instruments then separate these intercepts into a flexible price-side function and structural demand shocks. The main statistical challenge is that the number of profiled market intercepts grows with the number of markets. I show that, when both the number of markets and the minimum within-market sample size grow, this profiling step is asymptotically negligible and the common structural functions are uniformly consistently estimated. Monte Carlo evidence supports the consistency result and illustrates the value of flexible price-side estimation for counterfactual demand analysis.

Figures

Figures reproduced from arXiv: 2607.21323 by Richard Grigorian.

Figure 1
Figure 1. Figure 1: Normalized Monte Carlo errors for the full profiled estimator. [PITH_FULL_IMAGE:figures/full_fig_p018_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Counterfactual performance against Logit-IV, BLP, and Micro-BLP. [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Fixed versus growing price-side sieve under approximation error. [PITH_FULL_IMAGE:figures/full_fig_p056_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Oracle price-side robustness: GMM/SMD versus OLS. [PITH_FULL_IMAGE:figures/full_fig_p057_4.png] view at source ↗

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

30 extracted references · 1 canonical work pages

  1. [1]

    Ai, C. and X. Chen (2003). Efficient Estimation of Models with Conditional Moment Restrictions Containing Unknown Functions.Econometrica 71(6), 1795–1843

    2003

  2. [2]

    Andrews, D. W. K. (1994, January). Chapter 37 Empirical process methods in econometrics. In Handbook of Econometrics, Volume 4, pp. 2247–2294. Elsevier

    1994

  3. [3]

    Berry, S. T. (1994). Estimating Discrete-Choice Models of Product Differentiation.The RAND Journal of Economics 25(2), 242–262

    1994

  4. [4]

    Berry, S. T., A. Gandhi, and P. A. Haile (2013). Connected Substitutes and Invertibility of Demand. Econometrica 81(5), 2087–2111

    2013

  5. [5]

    Berry, S. T. and P. A. Haile (2014). Identification in Differentiated Products Markets Using Market Level Data.Econometrica 82(5), 1749–1797

    2014

  6. [6]

    Berry, S. T. and P. A. Haile (2024). Nonparametric Identification of Differentiated Products Demand Using Micro Data.Econometrica 92(4), 1135–1162

    2024

  7. [7]

    Berry, S. T., J. Levinsohn, and A. Pakes (1995). Automobile Prices in Market Equilibrium. Econometrica 63(4), 841–890

    1995

  8. [8]

    Berry, S. T., J. Levinsohn, and A. Pakes (2004). Differentiated Products Demand Systems from a Combination of Micro and Macro Data: The New Car Market.Journal of Political Econ- omy 112(1), 68–105

    2004

  9. [9]

    Chen, X. (2007). Large Sample Sieve Estimation of Semi-Nonparametric Models.Handbook of Econometrics 6B

    2007

  10. [10]

    Chen, X. and D. Pouzo (2015). Sieve Wald and QLR Inferences on Semi/Nonparametric Conditional Moment Models.Econometrica 83(3), 1013–1079. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.3982/ECTA10771

    work page doi:10.3982/ecta10771 2015

  11. [11]

    Compiani, G. (2022). Market Counterfactuals and the Specification of Multiproduct Demand: A Nonparametric Approach.Quantitative Economics 13(2), 545–591

    2022

  12. [12]

    Donald, S. G., G. W. Imbens, and W. K. Newey (2003, November). Empirical likelihood estimation and consistent tests with conditional moment restrictions.Journal of Econometrics 117(1), 55–93. Dub´ e, J.-P., J. T. Fox, and C.-L. Su (2012). Improving the Numerical Performance of Static and Dy- namic Aggregate Discrete Choice Random Coefficients Demand Estim...

    2003

  13. [13]

    (2015, March)

    Freyberger, J. (2015, March). Asymptotic theory for differentiated products demand models with many markets.Journal of Econometrics 185(1), 162–181

    2015

  14. [14]

    Freyberger, J. and M. A. Masten (2019, October). A practical guide to compact in- finite dimensional parameter spaces.Econometric Reviews 38(9), 979–1006. eprint: https://doi.org/10.1080/07474938.2018.1514025

    arXiv 2019

  15. [15]

    and J.-F

    Gandhi, A. and J.-F. Houde (2025). Measuring Substitution Patterns in Differentiated-Products Industries. 62

    2025

  16. [16]

    Hahn, J. and J. Meinecke (2005). Time-Invariant Regressor in Nonlinear Panel Model with Fixed Effects.Econometric Theory 21(2), 455–469

    2005

  17. [17]

    Hahn, J. and W. K. Newey (2004). Jackknife and Analytical Bias Reduction for Nonlinear Panel Models.Econometrica 72(4), 1295–1319

    2004

  18. [18]

    Hall, P. and J. L. Horowitz (2005, December). Nonparametric methods for inference in the presence of instrumental variables.The Annals of Statistics 33(6), 2904–2929

    2005

  19. [19]

    Horowitz, J. L. (2011). Applied Nonparametric Instrumental Variables Estimation.Economet- rica 79(2), 347–394

    2011

  20. [20]

    Shi, and J

    Lu, Z., X. Shi, and J. Tao (2023, August). Semi-nonparametric estimation of random coefficients logit model for aggregate demand.Journal of Econometrics 235(2), 2245–2265

    2023

  21. [21]

    (1974, November)

    McFadden, D. (1974, November). The measurement of urban travel demand.Journal of Public Economics 3(4), 303–328

    1974

  22. [22]

    Monardo, J. (2025). Measuring Substitution Patterns with a Flexible Demand Model

    2025

  23. [23]

    Nevo, A. (2001). Measuring Market Power in the Ready-to-Eat Cereal Industry.Econometrica 69(2), 307–342

    2001

  24. [24]

    Newey, W. K. (1997, July). Convergence rates and asymptotic normality for series estimators. Journal of Econometrics 79(1), 147–168

    1997

  25. [25]

    Newey, W. K. and D. McFadden (1994, January). Chapter 36 Large sample estimation and hypothesis testing. InHandbook of Econometrics, Volume 4, pp. 2111–2245. Elsevier

    1994

  26. [26]

    Newey, W. K. and J. L. Powell (2003). Instrumental Variable Estimation of Nonparametric Models. Econometrica 71(5), 1565–1578

    2003

  27. [27]

    Petrin, A. (2002). Quantifying the Benefits of New Products: The Case of the Minivan.Journal of Political Economy 110(4), 705–729

    2002

  28. [28]

    Reynaert, M. and F. Verboven (2014). Improving the Performance of Random Coefficients Demand Models: The Role of Optimal Instruments.Journal of Econometrics 179(1), 83–98

    2014

  29. [29]

    Torgovitsky, and H

    Tebaldi, P., A. Torgovitsky, and H. Yang (2023). Nonparametric Estimates of Demand in the California Health Insurance Exchange.Econometrica 91(1), 107–146. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.3982/ECTA17215. Van Der Vaart, A. W. and J. A. Wellner (1996).Weak Convergence and Empirical Processes. Springer Series in Statistics. New York, NY: Springer

    doi:10.3982/ecta17215 2023

  30. [30]

    (2023, August)

    Wang, A. (2023, August). Sieve BLP: A semi-nonparametric model of demand for differentiated products.Journal of Econometrics 235(2), 325–351. 63

    2023