{"id":"71d8b268-45c4-4f72-b010-96af3bbc78ea","arxiv_id":"2608.05676","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A dynamic factor model with endogenous stochastic volatility shows that common macroeconomic factors with shifting volatility, combined with heterogeneous sector exposures, explain over half of the cross-sectional variation in asymmetric tail risk across 116 U.S. variables.","lead":"This paper builds a statistical model of the U.S. economy that tracks both the level and the volatility of more than one hundred indicators, and shows that a single set of common forces explains which sectors face extreme upside or downside risk. The result offers a practical way to see where financial vulnerabilities concentrate, which is useful for policymakers and investors.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table 1's 55% R² is computed from model-generated tail asymmetry regressed on model-estimated loadings pooled across categories and horizons; a data-based cross-sectional regression is needed before taking the headline claim at face value.","rationale":"The paper is methodologically serious: the Gibbs/particle-Gibbs algorithm is described in detail, Monte Carlo validation indicates the sampler recovers the true parameters in a small-scale version, predictive densities are competitive with quantile regressions for most variables, and the BDS microdata correlation provides genuine external support for the sectoral pattern in industrial production. None of these, however, validates the specific number in the headline. The qwCRPS comparison (Section 6.3) evaluates full predictive densities, not the cross-sectional decomposition; the BDS exercise covers a subset of IP sectors, not the pooled 116-variable R². The weakest link is that the Table 1 regression uses model-generated tail asymmetry and model-estimated loadings, so under the maintained Gaussian-idiosyncratic assumption (eq. 2) a high R² is partly a statement about the model's own design. The proposed check is feasible because the quantile regressions are already estimated for all 116 variables; computing their implied tail asymmetry and regressing it on the loadings directly severs the circularity. If the data-based R² is close to 0.55, the central claim is materially strengthened; if it drops, the paper should reframe the result as a property of the estimated model rather than an empirical fact. This does not overturn the paper—the external validation and density comparisons remain valuable—but it keeps the reader's CONDITIONAL verdict unchanged pending the check.","tokens_in":37955,"tokens_out":9515,"duration_ms":102872,"concrete_test":"Use the quantile-regression model already estimated in Section 6.3 (equation 8) to compute fully data-based conditional 5th and 95th percentiles for all 116 variables and form the same tail-asymmetry statistic (log ratio of the time-series standard deviations of the two quantiles). Regress this data-based asymmetry on the estimated loadings from Table 1: (i) pooled, (ii) with growth/inflation/financial category dummies, and (iii) within the 60 growth variables only. If the pooled R² falls substantially below 0.55 or is insignificant within categories, the headline claim is an artifact of using model-generated objects; if the data-based R² remains near 0.55, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—that factor exposures explain 55 percent of the cross-sectional variation in tail asymmetry (Table 1)—is supported by a regression in which both the dependent variable (Section 3.2) and the regressors are outputs of the same estimated model. Equation (2) imposes symmetric Gaussian idiosyncratic shocks, so the model cannot attribute any asymmetry to idiosyncratic noise; all asymmetry must come through factor mean-volatility correlations. The Table 1 R² therefore partly certifies the model's internal mapping rather than an empirical regularity in the data. The permutation placebo in Section 6.2 randomizes loadings within the same model and does not break this circularity. The concern is heightened by pooling 116 variables with different horizons (12-month for growth and inflation, 3-month for financial) and very different average asymmetry across categories; between-category differences can inflate the pooled R² even if loadings have no within-category predictive power. Section 3.4 offers a within-IP association, but Table 1 itself does not show that the 55 percent survives category controls or a data-based definition of asymmetry.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper estimates a seven-factor dynamic factor model with endogenous stochastic volatility on 116 U.S. macroeconomic and financial series and uses the implied predictive distributions to document pervasive but heterogeneous tail asymmetry. Its central claim is that a single mechanism—common factors whose levels and volatilities move together, transmitted through heterogeneous loadings—unifies growth-at-risk, inflation-at-risk, and sectoral tail-risk heterogeneity, with factor exposures explaining 55 percent of the cross-sectional variation in tail asymmetry (Table 1). The paper also constructs aggregate risk indices, provides counterfactual shock analyses, compares tail forecasts to quantile regressions, and validates sectoral patterns against Census microdata.","tokens_in":38207,"tokens_out":5728,"duration_ms":60262,"significance":"If the headline claim holds, the paper offers a parsimonious explanation for where tail risk concentrates and a practical forecasting framework for a large panel. The manuscript has notable strengths: the estimation algorithm is described in unusual detail with a Monte Carlo validation; the tail-forecast comparison to quantile regressions and the calibration checks are informative; the permutation placebo provides a useful benchmark; and the external validation using BDS microdata is a valuable step beyond pure model-internal evidence. These elements make the paper a potentially important contribution to the macro-finance risk literature. However, the central cross-sectional claim currently rests on a regression in which both the dependent variable and the regressors are outputs of the same estimated model, so the headline number needs a data-based cross-sectional check before it can be taken at face value.","major_comments":[{"comment":"The headline R² = 0.55 in Table 1 is computed by regressing a model-implied tail asymmetry measure on model-estimated factor loadings, with both quantities generated from the same estimated model. Because Equation (2) constrains idiosyncratic shocks to be Gaussian and symmetric, the model cannot assign any tail asymmetry to idiosyncratic noise; all asymmetry is mechanically channeled through the common factors. The regression therefore partly certifies the model's internal mapping rather than an empirical regularity in the data. The footnote that generated regressors make inference 'mildly conservative' addresses standard errors, not the mechanical dependence of the dependent variable. Please add a data-based cross-sectional regression—for example, using conditional quantiles from the quantile regressions already estimated in Section 6.3, or realized sample quantiles—and report whether the loading structure survives. Without such a check, the abstract's 'factor exposures explain over half' claim is overstated.","section":"§4.3, Table 1; §3.2; Eq. (2)"},{"comment":"The pooled regression in Table 1 mixes three categories with different horizons (12-month for growth and inflation variables, 3-month for financial variables) and with very different average asymmetry levels across categories. A pooled R² can be inflated by between-category mean differences even if loadings have no within-category predictive power. Please report within-category regressions or include category fixed effects and horizon interactions. The within-IP association in Section 3.4 is encouraging but covers only one category and does not establish the 55 percent claim for the full panel.","section":"Table 1, note; §3.2"},{"comment":"The Gaussian, symmetric assumption on idiosyncratic shocks is load-bearing for the interpretation that heterogeneity in tail risk reflects factor exposures rather than idiosyncratic noise. The robustness exercise that shuts off idiosyncratic shocks shows only that common factors are sufficient within the model's own structure; it does not test whether idiosyncratic innovations have asymmetric tails in the actual data. Please provide diagnostics on the estimated idiosyncratic innovations, or relax the distributional assumption, to support the claim in the abstract and Section 7 that the heterogeneity is not idiosyncratic.","section":"§2.1, Eq. (2); §6.2"}],"minor_comments":[{"comment":"The dimension of Σ is written as N+n, but Σ is the covariance of the N factor shocks and the N volatility shocks, so the dimension should be 2N (or, if a different convention is intended, please define n explicitly). The current notation is confusing because n is used elsewhere for the number of observed variables.","section":"Appendix B.2, Step 6"},{"comment":"The text quotes a 55 percent R² while Table 1 reports an adjusted R² of 0.522 and does not state the unadjusted R². Please clarify which quantity is being reported so readers can reconcile the number with the table.","section":"Table 1 and §4.3 text"},{"comment":"The two panels use very different y-axis scales and orders, which makes cross-category comparison difficult; please label the panels more explicitly and align the displayed nominal-coverage reference lines so that the reader can compare coverage across growth, inflation, and financial variables.","section":"Figure 13(b)"},{"comment":"The title 'Volatility Decomposition' could be misread; since the table reports variance shares (R²) as well as standard deviations, consider renaming it 'Variance Decomposition' to match the standard terminology.","section":"Table A.4 title"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is well within the journal's scope and the authors are transparent about many limitations. The main worry is that the headline 55 percent R² is presented as an empirical finding although it is partly a property of the model's own structure; I would urge the editor to require a data-based cross-sectional regression before final acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is the first large-panel dynamic factor model with endogenous stochastic volatility — factors' levels and volatilities move together, with in-mean effects and correlated shocks — and it produces full predictive distributions for 116 variables. That is a real technical contribution. Second, the headline claim — factor exposures explain 55% of cross-sectional tail asymmetry — is computed from model outputs on both sides: the asymmetry measure comes from the model's predictive distributions and the loadings come from the same estimation. So that R² partly certifies the model's internal mapping, not an external regularity. The paper is aware of the issue and runs a placebo that reshuffles loadings, but that is still inside the model.\n\nWhat does well: the estimation algorithm is detailed, with a Monte Carlo check that recovers the DGP in a two-factor version. The external validation is the most convincing part: sectoral tail asymmetry lines up with Census microdata on job creation and establishment entry skewness, and those data are not used in estimation. The tail forecast comparison to quantile regressions is honest: the factor model wins for inflation, ties for growth, and is competitive for financial variables. Calibration is mostly in line with nominal rates.\n\nThe soft spots, in proportion: the R² is the central empirical claim, and it needs a data-based counterpart — e.g., regress realized tail asymmetry from quantile regressions on the model loadings, or at least include category controls. The mixed horizons (12-month for growth/inflation, 3-month for financial) pooled together can inflate the R² if categories differ in both loading patterns and average asymmetry. The Gaussian symmetric idiosyncratic assumption is load-bearing; if idiosyncratic shocks had asymmetric tails, the decomposition would shift. The paper does not release code or data, which is a problem for a paper this computational. The pandemic exclusion is a minor issue — the through-2019 estimate gives an even higher R², so it is not doing the work.\n\nMy take: the central mechanism — common factors with mean-volatility comovement transmitting asymmetric risk through heterogeneous loadings — is credible and well-supported by the external BDS evidence. The paper deserves a serious referee. I would send it out, and ask for code/data plus a data-based cross-sectional regression, or a clear statement that the R² is a model-implied quantity with a different interpretation.","headline":"A serious, well-executed paper whose headline 55% R² is partly internal consistency; still worth refereeing because the model, external validation, and honest evaluation are substantial.","tokens_in":38736,"tokens_out":2745,"would_cite":true,"duration_ms":29111,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that asymmetric tail risk across the U.S. macroeconomy has a single source: common factors whose levels and volatilities move together, with heterogeneous factor loadings transmitting the resulting asymmetry unevenly…","keywords":["dynamic factor model","tail risk","growth-at-risk","inflation-at-risk","stochastic volatility","leverage effect","sectoral heterogeneity","mean-volatility correlation"],"falsifier":"Estimate the model with idiosyncratic shocks drawn from a skewed-t distribution and check whether the posterior skewness parameters are large and whether the cross-sectional R-squared of tail asymmetry on loadings falls materially; alternatively, feed the model data generated with asymmetric idiosyncratic shocks and see whether it misattributes their asymmetry to factor loadings.","tokens_in":37726,"feed_emoji":"📉","tokens_out":7028,"duration_ms":64559,"temperature":0.7,"pith_summary":"This paper tries to establish that asymmetric tail risk across more than one hundred U.S. macroeconomic and financial variables is not a collection of idiosyncratic phenomena but the output of a single mechanism. The mechanism is a small set of common factors—financial conditions, inflation, credit, consumption—whose levels and volatilities move together, so that an adverse shock raises uncertainty at the same time it lowers the level. Each variable inherits a particular risk profile through its factor loadings, and these loadings explain about 55 percent of the cross-sectional variation in tail asymmetry across the panel. If true, tracking a few common factors tells you where in the economy downside and upside risks concentrate, and why some sectors are stable while others swing sharply. This unifies growth-at-risk, inflation-at-risk, and sectoral heterogeneity under one model rather than separate explanations.","feed_headline":"Factor exposures explain 55% of tail asymmetry across the economy","feed_subtitle":"One dynamic factor model yields growth-at-risk, inflation-at-risk, and sectoral differences at once.","key_machinery":"The load-bearing object is a dynamic factor model with endogenous stochastic volatility: levels of seven latent factors evolve subject to past volatility (in-mean effects), volatilities evolve with lagged factor movements, and contemporaneous shocks to factor levels and volatilities are correlated through a block of the covariance matrix. The paper defines the mean-volatility correlation as this connection between a factor's level and its conditional variance, and shows that heterogeneous factor loadings, meaning each variable's exposure to each factor, convert the factor-level correlation into variable-specific tail asymmetry. A named-factor normalization anchors each factor to an observable variable—excess bond premium, 1-year Treasury rate, S&P 500 index, real PCE, real PCE housing and utilities, nonrevolving credit, and PCE inflation—giving the mechanism an economic interpretation.","core_discovery":"On the paper's own terms, the central discovery is that tail risk is organized by the same common macroeconomic dynamics that drive business cycles. In a linear factor model with symmetric shocks, no factor structure can produce asymmetric risk; asymmetry appears only when a factor and its volatility move together—adverse shocks raising uncertainty, and elevated uncertainty feeding back to activity. The paper builds a seven-factor model with endogenous stochastic volatility in which this mean-volatility correlation is estimated, and shows that factor loadings transmit it unevenly: loadings on financial conditions and inflation generate most of the cross-sectional tail asymmetry, explaining 55 percent of the variation in tail asymmetry across 116 variables, with financial loadings contributing 38 percent on their own. Loadings on the aggregate consumption factor, despite being large, contribute nothing to asymmetry. In this view, growth-at-risk, inflation-at-risk, and sectoral risk heterogeneity are not separate phenomena but different projections of the same factor structure.","pith_inferences":["A testable extension: if the mechanism is general, estimating the same model on other countries' panels should yield a similar fraction of tail asymmetry explained by factor loadings, with financial and inflation factors dominant.","The paper's assumption of symmetric Gaussian idiosyncratic shocks means any true idiosyncratic tail asymmetry would be absorbed into the loadings; a version allowing skewed idiosyncratic shocks would reveal how much of the 55 percent is genuinely common-factor-driven.","A practical consequence the authors do not spell out: the estimated loadings give a ready-made ranking of sectors by sensitivity to a factor-specific stress, which could support scenario-based risk monitoring for portfolios or sectoral policy."],"forward_implications":["Growth-at-risk and inflation-at-risk become two sides of one mechanism: the same estimated factor structure produces downside risk in real activity and upside risk in prices, with the sign determined by the factor's mean-volatility correlation.","Because factor exposures explain over half of the cross-sectional variation in tail asymmetry, researchers and policymakers can track a small number of factors to learn where tail risk is concentrating rather than monitoring each series separately.","Counterfactual exercises attribute most of the GFC's downside risk to financial-condition shocks and most of the Great Inflation's upside price risk to inflation shocks, implying that the source of the dominant shock determines which tail stretches.","The model produces full predictive distributions for all 116 variables from a single estimated system, matching or beating quantile regressions for inflation and financial variables and performing comparably for real activity."],"supporting_citations":[{"why":"Defines growth-at-risk as the volatile lower tail of output growth and provides the quantile-regression baseline the paper compares against.","marker":"Adrian et al. (2019)"},{"why":"Documents the leverage effect, negative shocks raising volatility, which the paper generalizes to the macroeconomy.","marker":"Black (1976)"},{"why":"Establishes the two-way feedback between uncertainty and activity that motivates the endogenous volatility specification.","marker":"Ludvigson et al. (2021)"},{"why":"Defines the standard macro uncertainty measure whose independence from factor levels the paper contrasts with its own correlated specification.","marker":"Jurado et al. (2015)"},{"why":"Estimates common volatility with mean effects as a separate factor; the paper's contribution is letting the factors' own volatility respond to their levels.","marker":"Carriero et al. (2018)"},{"why":"Provides the dynamic factor model framework and the named-factor normalization used to identify the latent factors.","marker":"Stock and Watson (2016)"},{"why":"Documents inflation-at-risk, the upside-price-tail counterpart the paper reproduces with the same mechanism.","marker":"López-Salido and Loria (2024)"}],"fun_headline_variants":["Financial and inflation loadings drive most tail risk asymmetry","One factor mechanism unifies growth, inflation, and sector tail risk","55% of tail asymmetry explained by factor exposures","Common dynamics, heterogeneous loadings: source of tail risk"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The decomposition that attributes tail asymmetry to factor loadings assumes idiosyncratic shocks are Gaussian and symmetric; if sector-specific shocks also had skewed tails, the factor-loading mechanism would not be the unique source of heterogeneity.","fun_headline_variants_meta":{"raw":{"variants":["Financial and inflation loadings drive most tail risk asymmetry","One factor mechanism unifies growth, inflation, and sector tail risk","55% of tail asymmetry explained by factor exposures","Common dynamics, heterogeneous loadings: source of tail risk"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00061,"raw_usage":{"total_tokens":2778,"prompt_tokens":821,"completion_tokens":1957,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":437,"completion_tokens_details":{"reasoning_tokens":1891}},"tokens_in":437,"tokens_out":1957,"duration_ms":16751,"temperature":1.0,"reasoning_tokens":1891,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T04:14:47.539782+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate the model with idiosyncratic shocks drawn from a skewed-t distribution and check whether the posterior skewness parameters are large and whether the cross-sectional R-squared of tail asymmetry on loadings falls materially; alternatively, feed the model data generated with asymmetric idiosyncratic shocks and see whether it misattributes their asymmetry to factor loadings.","supporting_citations":[{"cited_title":"and Watson, Mark W","cited_arxiv_id":null,"evidence_quote":"Provides the dynamic factor model framework and the named-factor normalization used to identify the latent factors."}],"review_version":1}