Pith. sign in

REVIEW 3 major objections 5 minor 39 references

Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance

T0 review · 3 major / 5 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Noncausal dynamics in the classic monetary VAR largely vanish once common macro factors are filtered out, and the price puzzle disappears with them.

desk verdict Solid applied paper: GCov + factor filtering turns Stock–Watson noncausality into an omitted-info diagnosis on the original sample, with coherent IRF changes; the post-2000 persistence and root-counting fragility are real limits, not fatal ones. read the letter →

arxiv 2607.28131 v1 pith:YIKP2UMU submitted 2026-07-30 econ.EM

Lison Christiaens , Julien Hambuckers , Alain Hecq This is my paper · ORCID
classification econ.EM
keywords VARmodelsnonfundamentalshocksGCovestimatorcausalandnoncausalnon-Gaussianityfactorfilteringmonetarypolicypricepuzzle
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Small monetary VARs often show noncausal dynamics, which can mean either true nonfundamental shocks or simply that agents see more information than the econometrician. This paper builds a mixed causal-noncausal VAR that partials out common factors from a large macro panel, then re-estimates with the GCov estimator. In Monte Carlo designs where noncausality is only missing information, the filter restores a purely causal classification. On the Stock–Watson inflation–unemployment–funds-rate system for 1960–2000, several noncausal roots appear in the baseline and Taylor-rule versions; after the first two or three FRED-QD factors are removed, those roots disappear in the policy-rule specifications. The cleaned systems yield impulse responses in which a contractionary policy shock no longer raises prices and unemployment responds more weakly or even negatively. The same exercise on data through 2025 leaves residual noncausality, so the paper treats noncausal VARs as a diagnostic for omitted aggregate information rather than a permanent feature of the economy.

What carries the argument

A factor-filtering mixed causal-noncausal VARX: contemporaneous common factors are projected out of the endogenous variables and their lags (Frisch–Waugh–Lovell), then the residual system is estimated by the portmanteau GCov estimator that uses nonlinear autocovariances to separate causal from noncausal roots.

What would settle it

Re-estimate the same three-variable system on 1960–2000 with an independent large information set or alternative factor extractor; if noncausal roots remain after partialling out the first two or three factors, or if they disappear even when the true DGP is known to be noncausal, the omitted-information reading fails.

Watch

Extended reading notes

Core claim

In the Stock–Watson monetary VAR on 1960:I–2000:IV, the noncausal dimensions recovered by GCov in the baseline and Taylor-rule systems largely disappear once the first two (or three) common FRED-QD factors are partialled out; the filtered Taylor-rule VARs become purely causal, and the associated impulse responses further eliminate the price puzzle and reverse the unemployment response to a contractionary policy shock.

Load-bearing premise

That the leading common factors from a large macro panel are a good enough proxy for the extra information agents hold, so that vanishing noncausal roots after filtering can be read as omitted information rather than over-filtering or estimator artifact.

Editorial extensions

If this is right

  • Small monetary VARs that detect noncausality should be re-checked after partialling out leading macro factors before nonfundamentalness is treated as structural.
  • Once factors restore a purely causal representation, standard recursive IRFs become valid again and can reverse classic anomalies such as the price puzzle.
  • Noncausal root counts can serve as a practical diagnostic of how much aggregate information a small VAR is missing.
  • On samples that include the financial crisis and COVID, residual noncausality after factor filtering signals either structural breaks or information not spanned by standard macro panels.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The post-2001 persistence of noncausality even with eight factors suggests the need for time-varying or regime-switching noncausal VARs rather than only richer static factors.
  • If factor filtering routinely restores causality in other canonical small SVARs (oil, fiscal, technology news), many published nonfundamentalness findings may be reclassified as omitted-information problems.
  • Combining the factor filter with nonlinear innovation filters could eventually deliver economically interpretable IRFs even when some noncausal roots remain.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper asks whether noncausal dynamics detected in small macro-finance VARs reflect genuine nonfundamentalness or omitted information available to agents. It develops a factor-filtering mixed causal–noncausal VARX procedure, estimates it with the GCov estimator of Gourieroux and Jasiak (2023), and studies finite-sample behavior in VAR(1)/VAR(2) Monte Carlos and a factor-omission design. Empirically it revisits the Stock–Watson (2001) three-variable monetary VAR (inflation, unemployment, interest rate, with backward- and forward-looking Taylor-rule residuals) on 1960:I–2000:IV. Baseline GCov finds several noncausal companion roots (n2 = 3–4); after partialling out the first two or three FRED-QD factors those roots vanish in the Taylor-rule systems (n2 = 0). Factor-filtered IRFs further mitigate the price puzzle and reverse the unemployment response. On the extended sample through 2025 noncausality persists even with eight factors.

Significance. If the original-sample result is robust, the paper supplies a practical diagnostic that links the noncausal-VAR literature to the factor/nonfundamentalness literature and shows that part of the noncausality in a canonical small monetary VAR can be attributed to omitted common information. The Monte Carlo design that starts from a purely causal factor-augmented DGP and recovers n2 > 0 when factors are omitted is a useful proof-of-concept. The application is transparent (same variables, transformations, and Taylor-rule constructions as Stock–Watson) and the IRF comparison is economically interpretable once the system is purely causal. Strengths include the multi-lag GCov evidence (Tables 1–2), the explicit factor-omission Monte Carlo (Table 3), and the clear original-sample tables (Tables 4–5). The extended-sample persistence of noncausality is reported honestly and correctly framed as an open issue.

major comments (3)
  1. [§4.1–4.2, Tables 4–5] Tables 4–5 and §4.1–4.2: the central claim that noncausal dimensions “largely disappear” after factor filtering rests on counting companion eigenvalues with modulus >1. Several retained roots sit on the unit-circle boundary (e.g. 0.9958±0.0564i, 0.9962±0.0590i, 0.9861±0.1145i in Panels C–D), and the interest-rate series are near-unit-root (ADF p-values 0.077/0.22). Online Appendix B already shows GCov misclassification rises when a true root approaches 0.99. The paper needs either (i) a formal modulus test / bootstrap confidence set for |root| > 1, or (ii) a Monte Carlo that matches the application’s near-unit-root + OLS-start + H=6 regime, before the drop from n2=3–4 to n2=0 can be read as recovery of a fundamental representation rather than stabilization of borderline moduli.
  2. [§3.2, Table 3] Table 3 and §3.2: the factor-omission Monte Carlo starts from a purely causal DGP with well-separated roots (0.85……0.3) and still leaves ~30% mass off n1=6 even with three factors. It does not stress the near-unit-root interest-rate environment or the OLS initialization used in the application. Without that stress test, the simulation supports consistency trends but does not underwrite the empirical claim that vanishing noncausal roots after partialling FRED-QD factors 1–2/3 identify omitted information rather than GCov root-counting fragility or over-orthogonalization.
  3. [§4.2, Figure 3] §4.2 and expanding-window Figure 3: on the extended sample through 2025, noncausality persists even after filtering with all eight FRED-QD factors, and noncausal dimensions reappear after the mid-2000s / COVID. The paper treats this as possible structural change or unmeasured forward-looking information, but does not reconcile it with the original-sample interpretation. Either the original-sample “omitted information” reading is sample-specific, or the factors are an incomplete proxy; the manuscript should state which interpretation is preferred and what additional evidence would distinguish them.
minor comments (5)
  1. [§3.1] Theorem 1 is labeled “Theorem 2.2” in the text of §3.1; numbering should be consistent.
  2. [§4.3, Figure 4] Figure 4 IRF panels would be clearer with confidence bands (even pointwise bootstrap) so that the unemployment sign reversal and price-puzzle removal can be assessed statistically.
  3. [§3.1] The choice of nonlinear transformations {et, e2t, e3t, etietj} and H=6 is described as “best simulations” but not justified against alternatives in the main text; a short robustness paragraph or appendix table would help.
  4. [§4.2.1] Clarify whether factors are extracted on the full FRED-QD sample and then restricted, or only on the information set available in 2000:Q4, in the main text (currently deferred to online Appendix D).
  5. Minor typos: “V ARs” spacing, “Göttingen” encoding, and occasional missing spaces before citations.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical noncausality-after-filtering claim is not forced by definition; mild self-use of GCov/representation theory is methodological scaffolding only.

  1. self citation load bearing [§2.2 Representation theorem; §2.4 GCov estimator (Eq. 14); empirical use in §4.1–4.2 Tables 4–5]
    "To prove the existence of a stationary solution of (5) and derive its two-sided moving average representation, it is necessary to introduce the Representation Theorem proposed by Gourieroux and Jasiak (2017). ... To estimate the proposed causal-noncausal model, we rely on the semi-parametric Generalized Covariance (GCov) estimator (Gourieroux and Jasiak (2023))."

    Root classification (n1 causal / n2 noncausal) and the estimator that delivers the companion eigenvalues are taken from prior work by overlapping authors. This is standard methodological scaffolding, not a circular reduction of the empirical claim: the paper does not define ‘omitted information’ as ‘whatever makes GCov report n2=0,’ and the drop in n2 after external factor residualization remains an independent data outcome. Mild only.

full rationale

The paper's central empirical claim—that noncausal dimensions detected by GCov in the Stock–Watson (2001) system largely vanish once FRED-QD factors are partialled out—is not circular. Noncausal dimension n2 is read from companion-matrix eigenvalues after GCov estimation (Tables 4–5, Representation theorem of Gourieroux–Jasiak 2017); factors come from an external large panel (McCracken–Ng 2021) via ordinary residualization (Frisch–Waugh–Lovell, Eq. 13). Monte Carlo Table 3 starts from a purely causal factor-augmented DGP and checks recovery of n1=6 under omitted vs full factors; that is a genuine (if imperfect) falsification design, not a tautology. Self-citations to GCov (Gourieroux–Jasiak 2023), VMAR/VAR(n1,n2,p) theory, and related work by overlapping authors supply the estimator and root-counting apparatus; they do not define the empirical outcome that n2 drops to 0 after filtering, nor do they force the IRF comparison. Interpretation that vanishing noncausality equals ‘omitted agent information’ is an economic reading of an external residualization step, not a definitional identity. No fitted parameter is renamed a prediction; no uniqueness theorem is imported to forbid alternatives; no ansatz is smuggled in as a first-principles result. Score 1 reflects only ordinary methodological self-citation that is not load-bearing for the headline claim.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

Identification and interpretation rest on non-Gaussian i.i.d. errors, root-modulus classification of causality, GCov tuning choices, and the substantive claim that FRED-QD principal components stand in for agents’ extra information. No new physical entities; free parameters are estimator and filter hyperparameters plus lag/policy-rule choices that affect the reported noncausal counts.

free parameters (5)
  • GCov nonlinear transformations and portmanteau lag H = {e,e2,e3,cross}, H=6 (H=10 in one panel)
    Baseline uses {e_t, e_t², e_t³, e_t i e_t j} and H=6 (sometimes H=10 for convergence); choice is simulation-tuned and can change optimization/root classification.
  • Number of retained FRED-QD factors J = J=2–3 primary; up to 8 in robustness
    Main claims use J=1,2,3 (sometimes 8); noncausal dimension falls sharply at J≥2 for Taylor-rule specs on the original sample.
  • VAR lag order p = p=2
    BIC-selected VAR(2) instead of Stock–Watson’s VAR(4); changes companion root count (n×p) and thus n2.
  • Student-t degrees of freedom in Monte Carlo errors = ν=4
    ν=4 used for non-Gaussian innovations; fourth moments fail to exist, justified by consistency-only identification.
  • Taylor-rule coefficients and expectation VAR for forward rule = Stock–Watson (2001) rule coefficients; expectations from VAR(4)
    Backward/forward policy filters follow Stock–Watson scalings and VAR(4) forecasts; they redefine the interest-rate series before GCov.
assumptions (5)
  • domain assumption Errors are i.i.d. non-Gaussian with enough moments for GCov consistency; second moments alone cannot separate causal and noncausal roots.
    Stated throughout §2 and required for GCov identification (Gourieroux–Jasiak).
  • domain assumption Roots of the companion matrix with modulus >1 count as noncausal dimensions; modulus <1 as causal.
    Classification rule used in all tables; near-unit roots are acknowledged as fragile (§4.1, Appendix B).
  • domain assumption Partialling Y and lags on contemporaneous factors (Frisch–Waugh–Lovell) yields autoregressive dynamics equivalent to a VARX for the purpose of root classification, without breaking GCov consistency when factors are stationary.
    §2.3–2.4; core of the factor-filtering design.
  • ad hoc to paper Leading principal components from transformed FRED-QD approximate the common information available to agents but missing from the small VAR.
    Interpretive bridge from factor residualization to “missing information vs genuine nonfundamentalness” (§1, §4.2).
  • standard math Standard matrix polynomial / companion-form representation theory for mixed causal–noncausal VARs (including Giancaterini conditions linking VAR(n1,n2,p) and VMAR).
    §2.2 Representation theorem and Theorem 1.
invented entities (1)
  • Factor-filtering mixed causal–noncausal VARX diagnostic
    purpose: Jointly allow noncausal roots and strip common macro factors to test whether noncausality is informational.
    Methodological construct combining existing noncausal VARX and factor partialling; not a new economic object with independent physical existence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance." pith.science (2026). https://pith.science/paper/YIKP2UMU

@misc{pith2026260728131,
  author       = {Pith},
  title        = {Pith review of: Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YIKP2UMU}},
  note         = {Machine review of arXiv:2607.28131}
}
read the original abstract

This paper studies the presence of noncausal dynamics in standard macro-finance VAR models and asks whether they reflect genuine nonfundamentalness or omitted information available to economic agents but unobserved by the econometrician. To that end, we introduce a factor-filtering mixed causal-noncausal VARX approach designed to account for common macroeconomic information. We assess its performance in simulated settings, while showing also that the generalized covariance (GCov) estimator correctly recovers causal and noncausal dynamics when using several lags. Empirically, we revisit the well-known Stock-Watson monetary policy (S)VAR and show that the noncausal components detected in the baseline specification largely disappear once common factors are filtered out. Finally, we compare impulse responses from the filtered and original data to assess the transmission of monetary policy shocks and show that filtering further removes the price puzzle.

Figures

Figures reproduced from arXiv: 2607.28131 by the authors.

Figure 1
Figure 1. U.S. Inflation, Unemployment rate and Interest rate from 1960:I to 2025:II. [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
Figure 2
Figure 2. Interest rate series filtered with backward and forward Taylor rules [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 3
Figure 3. Expanding-window estimates of the noncausal dimension based on the GCov estimator. The left [PITH_FULL_IMAGE:figures/full_fig_p028_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of impulse response functions to a contractionary monetary policy shock: unfiltered [PITH_FULL_IMAGE:figures/full_fig_p030_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 1 linked inside Pith

  1. [1]

    F` eve, A

    Beaudry, P., P. F` eve, A. Guay, and F. Portier (2019). When is nonfundamentalness in svars a real problem?Review of Economic Dynamics 34, 221–243

    2019

  2. [2]

    Breidt, F. J., R. A. Davis, K.-S. Lh, and M. Rosenblatt (1991). Maximum likelihood es- timation for noncausal autoregressive processes.Journal of Multivariate Analysis 36(2), 175–198

    1991

  3. [3]

    Brock, W. A., S. N. Durlauf, and G. Rondina (2013). Design limits and dynamic policy analysis.Journal of Economic Dynamics and Control 37(12), 2710–2728

    2013

  4. [4]

    Brockwell, P. J. and R. A. Davis (1987). Stationary arma processes. InTime Series: Theory and Methods, pp. 77–111. Springer

    1987

  5. [5]

    Cavaliere, G., H. B. Nielsen, and A. Rahbek (2020). Bootstrapping noncausal autoregres- sions: with applications to explosive bubble modeling.Journal of Business & Economic Statistics 38(1), 55–67

    2020

  6. [6]

    Giancaterini, A

    Cubadda, G., F. Giancaterini, A. Hecq, and J. Jasiak (2024). Optimization of the generalized covariance estimator in noncausal processes.Statistics and Computing

    2024

  7. [7]

    Cubadda, G. and A. Hecq (2011). Testing for common autocorrelation in data-rich environ- ments.Journal of Forecasting 30(3), 325–335

    2011

  8. [8]

    Davis, R. A. and L. Song (2020). Noncausal vector ar processes with application to economic time series.Journal of Econometrics 216(1), 246–267

    2020

Show all 39 references
  1. [9]

    Forni, M. and L. Gambetti (2014). Sufficient information in structural vars.Journal of Monetary Economics 66, 124–136

    2014

  2. [10]

    Gambetti, and L

    Forni, M., L. Gambetti, and L. Sala (2019). Structural vars and noninvertible macroeconomic models.Journal of Applied Econometrics 34(2), 221–246. 32

    2019

  3. [11]

    Giancaterini, F., A. Hecq, J. Jasiak, and A. M. Neyazi (2025b). Regularized generalized covariance (rgcov) estimator.arXiv preprint arXiv:2504.18678

    arXiv

  4. [12]

    Gourieroux, C. and J. Jasiak (2017). Noncausal vector autoregressive process: Represen- tation, identification and semi-parametric estimation.Journal of Econometrics 200(1), 118–134

    2017

  5. [13]

    Gourieroux, C. and J. Jasiak (2023). Generalized covariance estimator.Journal of Business & Economic Statistics 41(4), 1315–1327

    2023

  6. [14]

    Gourieroux, C. and J. Jasiak (2026). Nonlinear fore(back)casting and innovation filtering for causal-noncausal var models.Journal of Financial Econometrics

    2026

  7. [15]

    and J.-M

    Gourieroux, C. and J.-M. Zakoian (2017). Local explosion modelling by non-causal process. Journal of the Royal Statistical Society Series B: Statistical Methodology 79(3), 737–756

    2017

  8. [16]

    Greene, W. H. (2018).Econometric Analysis(8th ed.). New York: Pearson

    2018

  9. [17]

    Issler, and S

    Hecq, A., J. Issler, and S. Telg (2020). Mixed causal–noncausal autoregressions with exoge- nous regressors.Journal of Applied Econometrics 35(3), 328–343

    2020

  10. [18]

    Hecq, A. and D. Vel´ asquez-Gaviria (2025). Spectral estimation for mixed causal-noncausal autoregressive models.Econometric Reviews 44(7), 939–962

    2025

  11. [19]

    Hecq, A. and E. Voisin (2021). Forecasting bubbles with mixed causal-noncausal autoregres- sive models.Econometrics and Statistics 20, 29–45

    2021

  12. [20]

    Hecq, A. and E. Voisin (2023). Predicting crashes in oil prices during the covid-19 pandemic with mixed causal-noncausal models. InEssays in honor of Joon Y. Park: Economet- ric methodology in empirical applications, Volume 45, pp. 209–233. Emerald Publishing Limited. 33

    2023

  13. [21]

    Hencic, A. and C. Gouri´ eroux (2015). Noncausal autoregressive model in application to bitcoin/usd exchange rates.Econometrics of risk, 17–40

    2015

  14. [22]

    Jasiak, J. and A. Manafi Neyazi (2025). Gcov-based portmanteau test.Econometric Reviews. Forthcoming

    2025

  15. [23]

    Kilian, L. and H. L¨ utkepohl (2017).Structural Vector Autoregressive Analysis. Cambridge University Press

    2017

  16. [24]

    Lanne, M. and J. Luoto (2017). A new time-varying parameter autoregressive model for u.s. inflation expectations.Journal of Money, Credit and Banking 49(5), 969–995

    2017

  17. [25]

    Lanne, M. and P. Saikkonen (2011). Noncausal autoregressions for economic time series. Journal of Time Series Econometrics 3(3)

    2011

  18. [26]

    Lanne, M. and P. Saikkonen (2013). Noncausal vector autoregression.Econometric The- ory 29(3), 447–481

    2013

  19. [27]

    Lippi, M. and L. Reichlin (1993). The dynamic effects of aggregate demand and supply disturbances: Comment.The American Economic Review 83(3), 644–652

    1993

  20. [28]

    Lippi, M. and L. Reichlin (1994). Var analysis, nonfundamental representations, blaschke matrices.Journal of Econometics 63(1), 307–325

    1994

  21. [29]

    Lobato, I. N. and C. Velasco (2022). Single step estimation of arma roots for nonfundamental nonstationary fractional models.The Econometrics Journal 25(2), 455–476

    2022

  22. [30]

    Lof, M. and H. Nyberg (2017). Noncausality and the commodity currency hypothesis.Energy Economics 65, 424–433

    2017

  23. [31]

    (2021, November)

    Mavroeidis, S. (2021, November). Identification at the zero lower bound.Econometrica 89(6), 2855–2885. 34

    2021

  24. [32]

    McCracken, M. W. and S. Ng (2021). FRED-QD: A quarterly database for macroeconomic research.Federal Reserve Bank of St. Louis Review 103(1), 1–44

    2021

  25. [33]

    Ramey, V. A. (2016). Macroeconomic shocks and their propagation. In J. B. Taylor and H. Uhlig (Eds.),Handbook of Macroeconomics, Volume 2, pp. 71–162. Elsevier

    2016

  26. [34]

    Sims, C. A. (1980, January). Macroeconomics and reality.Econometrica 48(1), 1–48

    1980

  27. [35]

    Soccorsi, S. (2016). Measuring nonfundamentalness for structural vars.Journal of Economic Dynamics and Control 71, 86–101

    2016

  28. [36]

    Stock, J. H. and M. W. Watson (2001). Vector autoregressions.Journal of Economic Per- spectives 15(4), 101–115

    2001

  29. [37]

    Stock, J. H. and M. W. Watson (2002). Forecasting using principal components from a large number of predictors.Journal of the American Statistical Association 97(460), 1167–1179

    2002

  30. [38]

    Stock, J. H. and M. W. Watson (2012a). Disentangling the channels of the 2007–2009 recession.Brookings Papers on Economic Activity, 81–156

    2007

  31. [39]

    Taylor, J. B. (1993). Discretion versus policy rules in practice.Carnegie-Rochester Conference Series on Public Policy 39, 195–214. 35

    1993

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.