{"id":"565d0a4d-2573-4f7b-add4-95475d729451","arxiv_id":"2608.10336","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A skewed multinomial probit model with alternative-specific skewness parameters, an identified covariance reparameterization, and a Bayesian MCMC sampler improves prediction when choice responses are asymmetric and matches the standard model when they are symmetric.","lead":"This paper proposes a new version of the standard multinomial probit choice model that lets people react differently to price increases and decreases, using a skew-normal distribution for unobserved utilities. On two real purchase datasets it predicts choices at least as well as the standard model and changes the implied price elasticities and substitution patterns.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Identification of the SMNP parameter vector is asserted, not proven; the nonzero skew-normal mean plus alternative-specific intercepts creates an unexamined redundancy that Equation (7) does not resolve.","rationale":"Read in good faith: the paper proposes a useful extension of multinomial probit, the reparameterization in (7) is algebraically sound for scale normalization and positive definiteness, and the simulation evidence supports recovery of skewness when present. The stress-test concern is not a visible algebraic error but a missing formal proof of the paper's headline claim of identification. The nonzero mean of the skew-normal distribution and the presence of alternative-specific intercepts in X_i give a concrete mechanism by which a continuum of (δ, β_int) pairs could preserve the first two moments of the latent utilities; whether the discrete-choice likelihood breaks this continuum is precisely what is left unproven in Section 3.2. The proposed numerical rank test would resolve local identifiability in the paper's own simulation design, and the complementary starting-value check would address the global question. Secondary issues (the incomplete sentence in Section 5.1, absence of released code) are real but do not alter the assessment. The reader's CONDITIONAL verdict is appropriate; no verdict change is needed.","tokens_in":21490,"tokens_out":19226,"duration_ms":183002,"concrete_test":"At the Section 4.1 DGP, evaluate the Jacobian of the stacked choice-probability vector (P(Y_i=j) over a grid of X values and alternatives) with respect to θ at the true parameter values, using high-accuracy simulation or GHK under the augmented representation (5)-(6). Compute the singular values of this Jacobian. If the smallest singular value is not bounded away from zero, extract the null direction and verify whether it matches the δ_{2:J}-intercept compensation described above; if it is positive, local identification holds at that design. As a complementary check, run the sampler from two starting points that differ only by a candidate null direction and compare the posterior predictive distributions: identical predictions with different δ posteriors would indicate a global identification failure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that the reparameterized vector θ=(β,δ,γ,ψ) is identified and that the SMNP recovers asymmetric choice responses. Section 3.2 asserts identification from the scale normalization Σ_11=1 and the PD-preserving form of (7). But equation (7) only guarantees that every draw satisfies the normalization; it does not show that the map from θ to the multinomial choice probabilities is injective. The mechanism is concrete: for ε~SN(0,Σ,α), E[ε]=√(2/π)δ. The latent mean is X_iβ+√(2/π)δ, and X_i in (3) includes the J alternative-specific intercepts I_J. Therefore any shift δ→δ+t can be compensated by β_int→β_int−√(2/π)t without changing the conditional mean. Since Σ=Var(ε)+(2/π)δδ^T, the scale restriction Σ_11=1 fixes δ_1 only up to sign given Var(ε)_11, leaving δ_{2:J} free in the moment map; γ and ψ can be adjusted to keep Var(ε) and the (1,1) normalization intact over a continuum of δ values. The paper provides no rank condition or identifiability proof showing that the higher-order skewness structure is separately identified by polychotomous choices. This is not a claim that the model is unidentified—the simulations recover the DGP—but the identification assertion is load-bearing and currently unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a skewed multinomial probit (SMNP) model in which latent utilities follow a multivariate skew-normal distribution, allowing asymmetric choice responses. The authors introduce a reparameterization of the covariance matrix that enforces the MNP scale normalization and positive definiteness, a double data augmentation scheme yielding a Metropolis-Hastings within Gibbs sampler, and priors centered on a symmetric equicorrelated structure. Numerical experiments with symmetric and asymmetric DGPs and two empirical applications (laundry detergent and ketchup) are used to argue that SMNP recovers skewness, improves predictive accuracy, and changes implied price elasticities and substitution patterns.","tokens_in":21696,"tokens_out":17051,"duration_ms":146446,"significance":"The paper addresses a real gap in discrete choice modeling: the symmetry restriction of standard multinomial probit. If the identification and computational claims are correct, the SMNP model is a useful, parsimonious alternative to reference-price or random-coefficient approaches to asymmetry. The paper's numerical experiments are carefully designed with known DGPs and oracle comparisons, and the empirical evaluation uses repeated out-of-sample splits, which is a methodological strength. However, the identification proof is a necessary part of the contribution and is currently missing.","major_comments":[{"comment":"The claim that the reparameterization 'is key for identification' and that θ is 'the identified parameter vector' is not supported by a formal identifiability proof. The construction in Eq. (7) ensures only that every posterior draw satisfies the scale normalization Σ_11=1 and that Σ_u is positive definite; it does not establish that the map from θ to the multinomial choice probabilities is injective. Because the multivariate skew-normal error has nonzero mean √(2/π)δ, a change in δ can be compensated by an opposite change in the alternative-specific intercepts contained in X_i (Eq. 3) so that the conditional mean of Z_i is unchanged. The paper does not provide a rank condition or an argument showing that the higher-order cumulants of the skew-normal are separately identified from polychotomous choices. Without this, the MCMC sampler is operating on a parameter vector whose posterior may contain ridges, and the reported posterior densities for δ in Figures 2, 4, 5, and 7 cannot be interpreted as conclusive evidence of identification. This is load-bearing because the abstract and Section 6 list 'fully identified' as a main contribution.","section":"Section 3.2, Eq. (7)"}],"minor_comments":[{"comment":"The sentence 'In addition, we also apply the Giacomini–White test directly to the' is incomplete and should be finished or removed.","section":"Section 5.1"},{"comment":"The word 'difficult' appears in the introduction and should be corrected to 'difficult'.","section":"Section 1"},{"comment":"The expression for the expectation of ψ is written as V/((J+3)−(J−1)−1), which simplifies to V/3; the intermediate notation is confusing and could be simplified.","section":"Appendix A, Eq. (21)"},{"comment":"In the ketchup panel, the CLS3 entries for MNP and SMNP are identical (−0.4778) but the MNP entry is flagged with a star; the note should explain the convention for ties.","section":"Table 2"}],"recommendation":"major_revision","confidential_remarks":"The identification issue is the key block. If the authors can provide a rigorous identifiability proof for the SMNP parameter vector from multinomial choice probabilities, the paper would be a strong contribution. The computational and empirical work is otherwise solid, and the gap appears fixable within the manuscript's scope. I would encourage the editor to seek a major revision rather than reject."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth your time. It extends multinomial probit by allowing skew-normal latent utilities with a free scale matrix, and it has a genuinely nice technical core: the reparameterization in (7) builds the scale normalization into the covariance, plus a double data augmentation that leaves only one nonconjugate parameter. The sampler is carefully specified, and the simulations are encouraging—the model recovers the skewness parameters when they are in the DGP and collapses toward MNP when they are not. The empirical applications show small but mostly significant predictive gains and elasticity differences in the expected direction.\n\nThe soft spot is identification. The paper claims the parameter vector θ is identified, but the argument in Section 3.2 only shows that the parameterization satisfies the scale restriction and positive definiteness. It does not show that the map from θ to the multinomial choice probabilities is injective. The interaction between alternative-specific intercepts and the skewness-induced mean is the obvious place to look for trouble, and the stress-test note's continuum argument is not fully convincing because changing δ while compensating the intercepts changes the shape of the latent distribution, not just its mean. Still, the lack of a formal proof (or a rank condition) is a real gap, and it is load-bearing because the whole contribution claims a fully identified model. The fact that the MCMC recovers the DGP parameters in simulations is evidence, but not a proof.\n\nOther issues are minor by comparison: no code is provided, so the results aren't reproducible as-is; there is an incomplete sentence at the end of Section 5.1; the elasticity curves are presented without uncertainty bands; and the prior hyperparameters (τ_δ, τ_γ) are not sensitivity-analyzed, which matters given the hand-chosen value 0.1936. The predictive improvements in the applications are economically small, though statistically significant, and the paper is appropriately cautious about that.\n\nWho is this for? Anyone doing Bayesian discrete choice with asymmetric responses. It's a solid contribution that extends the MNP toolkit. I'd send it to peer review: a good referee can push for the identifiability proof and code release. With those, this would be a cite-worthy methods paper.","headline":"A genuinely useful reparameterization for skewed multinomial probit with a workable sampler, but identification is asserted rather than proven; deserves a referee round.","tokens_in":22311,"tokens_out":8387,"would_cite":true,"duration_ms":77971,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62F15","62H05","62P20"],"pacs":[],"model":"deepseek-v4-flash","headline":"A skewed multinomial probit model allows choice probabilities to respond asymmetrically to covariate changes, and the paper shows it is identified and estimable.","keywords":["skew-normal distribution","multinomial probit","asymmetric choice responses","data augmentation","Bayesian inference","price elasticities","substitution patterns","scale identification"],"falsifier":"For a small simulated three-alternative choice dataset generated with known skewness, compute the likelihood or posterior surface over the intercept and $\\delta_1$ with $\\Sigma_{11}$ fixed to 1; if a near-flat ridge of high likelihood runs along a curve intercept = $f(\\delta_1)$, the reparameterization does not identify skewness separately from the intercept, and posterior recovery of $\\delta$ would be prior-driven rather than data-driven.","tokens_in":21212,"feed_emoji":"🛒","tokens_out":8341,"duration_ms":64543,"temperature":0.7,"pith_summary":"This paper proposes a skewed multinomial probit (SMNP) model in which the latent utilities follow a multivariate skew-normal distribution, so a positive and a negative change of the same size in a covariate do not have to move choice probabilities equally. The contribution is showing that this extension can be made identified and computationally tractable: a covariance reparameterization fixes the scale and preserves positive definiteness by construction, and a double data-augmentation scheme turns the skew-normal utilities into conditionally Gaussian ones, yielding a Metropolis–Hastings within Gibbs sampler with a single one-dimensional Metropolis step. Numerical experiments and two consumer choice applications indicate that SMNP recovers asymmetric responses when they are present, improves predictive accuracy, and changes implied price elasticities and substitution patterns relative to standard multinomial probit. The paper also shows that when the data are symmetric, SMNP behaves like the standard MNP and does not sacrifice predictive performance.","feed_headline":"Skewed probit captures asymmetric consumer choices","feed_subtitle":"A multinomial probit variant lets choice probabilities respond differently to price rises and price cuts.","key_machinery":"The central object is the reparameterized covariance matrix of the double-augmented model, equation (7): $\\Sigma_u = \\begin{bmatrix} 1-\\delta_1^2 & \\gamma^\\top \\\\ \\gamma & \\psi + \\gamma\\gamma^\\top/(1-\\delta_1^2) \\end{bmatrix}$. This enforces the MNP scale normalization $\\Sigma_{11}=1$ by construction because $\\Sigma = \\Sigma_u + \\delta\\delta^\\top$, and it enforces positive definiteness through $|\\delta_1|<1$ and $\\psi \\succ 0$ via a Schur complement argument. The second key piece is the stochastic representation of the skew-normal distribution, $Z_i \\mid w_i \\sim N(X_i\\beta+\\delta w_i, \\Sigma_u)$ with $w_i$ a truncated standard normal, which turns the skew-normal utilities into a conditionally Gaussian regression and makes Gibbs updates available. The sampler then needs only a single Metropolis–Hastings step, for $\\delta_1$.","core_discovery":"On the paper's terms, the discovery is that the multivariate skew-normal distribution can replace the multivariate normal error in multinomial probit without giving up identification or tractable Bayesian computation. The identification problem is that the MNP scale normalization interacts with the skewness parameters, because the skew-normal error has a nonzero mean that can be absorbed into intercepts. The paper's reparameterization writes the doubly augmented covariance $\\Sigma_u$ with top-left entry $1-\\delta_1^2$, so that $\\Sigma=\\Sigma_u+\\delta\\delta^\\top$ automatically has $\\Sigma_{11}=1$; positivity follows from $|\\delta_1|<1$ and a positive-definite $\\psi$. With this parameterization, the double augmentation of latent utilities $Z_i$ and truncation variables $w_i$ gives closed-form full conditionals for all parameters except $\\delta_1$, which is updated by a single univariate Metropolis–Hastings step. The empirical message is that asymmetry is present in real purchase data and that ignoring it distorts own- and cross-price elasticities.","pith_inferences":["If the identification concern is real, $\\delta$ and alternative-specific intercepts are partially aliased, so posterior skewness should be checked for sensitivity to the intercept prior or to centering the skew-normal error at zero mean.","The distributional asymmetry is silent about mechanism: loss aversion, brand loyalty, and switching costs can all produce similar shapes; letting $\\delta$ depend on covariates would make mechanisms testable.","The double augmentation strategy should carry over to skew-$t$ or related skew-elliptical errors because the conditionally Gaussian representation generalizes, adding tail robustness.","A prior-predictive check for symmetry—comparing the observed difference in response to a price increase versus a price decrease with the range implied by the $\\tau_\\delta$ prior—would separate prior-induced from data-induced asymmetry."],"forward_implications":["Standard MNP is a special case at $\\delta=0$, so SMNP can be used as a drop-in generalization that reduces to the familiar model when asymmetry is absent.","In the paper's applications, allowing asymmetry changes own- and cross-price elasticities: the MNP understates price sensitivity for a focal detergent brand and misallocates substitution across ketchup brands.","The sampler is practical for the datasets studied: all full conditionals except $\\delta_1$ are closed-form Gibbs steps, with a single univariate Metropolis–Hastings update.","When the data generating process is symmetric, SMNP shrinks toward MNP and does not lose predictive accuracy, so the added flexibility has little cost in symmetric settings.","Because asymmetry is introduced through the latent utility distribution rather than observed gain/loss covariates, SMNP detects asymmetric responses without requiring reference prices or predictor transformations."],"supporting_citations":[{"why":"Supplies the latent-utility data augmentation that the SMNP sampler extends to skew-normal errors.","marker":"Albert & Chib 1993"},{"why":"Defines the multivariate skew-normal distribution used as the latent utility error.","marker":"Azzalini & Dalla Valle 1996"},{"why":"Establishes the affine-closure property that gives $Z_i$ a skew-normal distribution.","marker":"Azzalini & Capitanio 1999"},{"why":"Provides the conditionally Gaussian stochastic representation used in the second augmentation layer.","marker":"Frühwirth-Schnatter & Pyne 2010"},{"why":"Supplies the identified covariance parameterization whose skewness-zero case SMNP generalizes.","marker":"McCulloch et al. 2000"},{"why":"Motivates the equicorrelated prior mean that the SMNP prior is centered on.","marker":"Geweke et al. 1994"},{"why":"Source of the laundry detergent purchase data used in the first empirical application.","marker":"Chintagunta & Prasad 1998"},{"why":"Source of the ketchup purchase data used in the second empirical application.","marker":"Kim et al. 1995"}],"fun_headline_variants":["Skewed probit reveals asymmetric price responses","When a price hike isn't just a price cut in reverse","Choice asymmetry: why probit needs skew","Price rises vs cuts: skewed probit captures the difference"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that fixing the first diagonal element of the latent utility covariance to one separates intercepts from skewness; the paper asserts this identification but does not prove that no continuum of (intercept, skewness) pairs produces identical choice probabilities.","fun_headline_variants_meta":{"raw":{"variants":["Skewed probit reveals asymmetric price responses","When a price hike isn't just a price cut in reverse","Choice asymmetry: why probit needs skew","Price rises vs cuts: skewed probit captures the difference"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000484,"raw_usage":{"total_tokens":2396,"prompt_tokens":961,"completion_tokens":1435,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":577,"completion_tokens_details":{"reasoning_tokens":1372}},"tokens_in":577,"tokens_out":1435,"duration_ms":9548,"temperature":1.0,"reasoning_tokens":1372,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:23:25.259310+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a small simulated three-alternative choice dataset generated with known skewness, compute the likelihood or posterior surface over the intercept and $\\delta_1$ with $\\Sigma_{11}$ fixed to 1; if a near-flat ridge of high likelihood runs along a curve intercept = $f(\\delta_1)$, the reparameterization does not identify skewness separately from the intercept, and posterior recovery of $\\delta$ would be prior-driven rather than data-driven.","supporting_citations":[],"review_version":1}