Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Let the Tree Decide: FABART A Non-Parametric Factor Model

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes FABART, a factor-augmented VAR whose factor loadings are estimated by Bayesian additive regression trees, and claims that this nonparametric mapping improves industrial-production forecasts and reveals sign asymmetries…

desk verdict The sign-asymmetry claim cannot arise from the model as written: the BART nonlinearity is linearized away before the GIRFs are computed, even though the BART-in-FAVAR idea itself is worth a careful look. read the letter →

arxiv 2506.11551 v1 pith:OUVISUCP submitted 2025-06-13 econ.EM

Sofia Velasco This is my paper · ORCID
classification econ.EM
keywords Bayesianadditiveregressiontreesfactor-augmentedVARnonparametricfactormodeloilsupplynewsshockssignasymmetrygeneralizedimpulseresponsefunctionsforecastevaluationlargemacroeconomicpanels
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

This paper builds a factor-augmented vector autoregression in which the link between observable macro-financial series and latent factors is estimated by Bayesian additive regression trees rather than by a linear mapping. It argues that this nonparametric FAVAR, called FABART, recovers latent factors under nonlinear data-generating processes without overfitting when the truth is linear, and that it forecasts U.S. industrial production more accurately than linear benchmarks during turbulent periods. The paper also claims that oil supply news shocks transmit asymmetrically: a positive shock that raises real oil prices contracts real activity and raises inflation more strongly and persistently than a negative shock expands activity, with the same sign asymmetry visible in state-level employment. If true, the work matters because linear factor models would miss both the forecast gains and the asymmetric propagation.

What carries the argument

The load-bearing object is the sum-of-trees function $f_i(Y_t)=\sum_{s=1}^{S} g_{is}(Y_t\mid \tau_{is},\mu_{iS})$ with $S=250$ regression trees, each a recursive binary partition of the factor space with constant terminal-node values, estimated under the BART regularizing prior by Metropolis-Hastings tree moves and Bayesian backfitting. What carries the argument is a two-step bridge: the fitted tree function is projected onto the observables through $\tilde A_i = F^\dagger X_i$, the Moore-Penrose projection of $X_i$ on the matrix of fitted nonlinear values $F$, producing a linear measurement equation $X_{i,t}=\tilde A_i' Y_t+\epsilon_{i,t}$ that permits Gaussian state-space draws of the latent factors. Impulse responses are then computed as generalized impulse response functions conditional on the long-run mean, which is how the paper lets shock sign and history enter the responses.

What would settle it

Fit the FABART model to a dataset generated by a known nonlinear factor model, then compute GIRFs two ways: through the linear projection $\tilde A_i = F^\dagger X_i$ and through direct simulation of the fitted BART function. If the linearized responses are statistically indistinguishable from a linear FAVAR's symmetric responses, or if the direct and projected responses diverge, the claimed sign asymmetry is not being carried by the model as described.

Watch

Extended reading notes

Core claim

The central claim is that replacing the linear factor-loading equation with a sum-of-trees approximation, where each observable is a function of the latent factors estimated by 250 regularized regression trees, lets latent factors inherit nonlinear structure from the data and preserves that structure when computing generalized impulse responses. In the simulations, FABART delivers lower forecast errors for the latent factor than a linear FAVAR under a linear DGP and two nonlinear DGPs, and Monte Carlo replications show posterior factor estimates with average correlations above 0.96. In the empirical application, FABART improves forecasts of industrial production and keeps predictive densities stable during the COVID-19 period, while the linear BVAR and FAVAR benchmarks deteriorate. The paper's headline empirical finding is sign asymmetry in oil supply news shock transmission: positive shocks generate stronger and more persistent contractions in real activity and inflation than negative shocks generate expansions, and state-level employment contracts more after positive shocks than it improves after negative shocks.

Load-bearing premise

The sign-asymmetry result rests on the assumption that the linear projection $\tilde A_i = F^\dagger X_i$ used to draw the latent factors keeps the nonlinear information in the BART function; if the projection acts as an ordinary linear loading matrix, the measurement equation is linear, the impulse responses are symmetric, and the reported asymmetry cannot arise.

Editorial extensions

If this is right

  • Under nonlinear data-generating processes, FABART recovers the latent factor with lower RMSE than a linear FAVAR and with posterior correlations above 0.96 across 100 Monte Carlo replications.
  • In the empirical application, FABART forecasts U.S. industrial production with lower RMSE and better log scores than a linear BVAR and a linear FAVAR at one-, three-, and twelve-month horizons, and its density forecasts stay stable when the COVID-19 shock reaches the data.
  • The estimated GIRFs imply that a positive oil supply news shock of 10 percent raises inflation and contracts U.S. and global industrial production more strongly and persistently than an equal-sized negative shock expands them.
  • State-level employment responses show the same sign asymmetry: most states fall below the 45-degree symmetry line, and the average difference is significantly negative; manufacturing-heavy states contract more while mining-intensive states are more insulated.
  • For real oil prices the forecast gains are small, and for financial variables such as the excess bond premium and the S&P 500 the linear BVAR remains competitive, so the nonparametric factor stage mainly helps real-activity forecasts and density stability.

Reading between the lines

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

  • The reported sign asymmetry may partly come from the linear projection used to sample factors; a direct comparison of GIRFs from the full tree function against GIRFs from the projected linear measurement equation would show how much of the asymmetry is genuinely produced by the BART model.
  • The same nonparametric factor-loading device could be applied to other observed factors, such as the federal funds rate or an uncertainty index, where the paper's identification strategy would transfer without change.
  • The state-level cross-sectional regressions suggest a sectoral mechanism, but a panel GIRF design with time-varying state exposure could test whether the manufacturing-share channel is causal rather than a slow-moving proxy.
  • Because the external instrument covers only part of the sample, the asymmetry estimates use a shorter identification window; extending the instrument or using narrative sign restrictions would reveal whether the result is sample-dependent.
Share X Bluesky LinkedIn Reddit HN

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 proposes FABART, a factor-augmented VAR in which the measurement equation linking observables to latent factors is modeled with Bayesian additive regression trees. Estimation proceeds by constructing a linear approximation to the BART function and then drawing factors from a linear Gaussian state space via Carter-Kohn. The framework is applied to recursive forecasting of U.S. macro-financial variables and to generalized impulse responses to oil supply news shocks identified with the Känzig (2021) external instrument. The paper claims improved forecasts for industrial production, especially during the COVID-19 period, and pronounced sign asymmetries in the transmission of oil supply news shocks at both aggregate and state levels.

Significance. If the claims were valid, FABART would be a useful addition to the nonlinear factor-model literature and the empirical oil-shock literature, and the forecasting evidence for industrial production would be of practical interest. The paper engages with a relevant literature, spells out priors and the estimation algorithm in detail, and connects its results to existing theoretical explanations of oil-price asymmetries. However, the central empirical contribution—sign asymmetry in GIRFs—is not obtainable from the model as described, because the measurement equation used for inference and impulse responses is a linear projection of the BART fit, and the transition equation is linear Gaussian. The simulation evidence also labels a linear DGP as nonlinear. These issues affect the paper’s core claims rather than only its presentation.

major comments (3)
  1. [Sections 2.2.2 and 2.3.1, Eqs. (12)–(14)] The estimation algorithm replaces the BART function F(Y_t) with the linear approximation X_{i,t} = \tilde A_i' Y_t + e_{i,t}, where \tilde A_i = F^\dagger X_i, and the transition equation (2) is linear with Gaussian shocks. The Carter-Kohn step then conditions on this linear Gaussian state space. Consequently, the conditional expectation in the GIRF definition (14) is linear in the shock; for a positive and a negative shock of equal magnitude, the responses must be exact mirror images. The sign asymmetries reported in Section 4.3.1 and Figure 7 therefore cannot arise from the model as specified. The BART nonlinearity is not used in the GIRF simulation, so the reported asymmetry is not a model output but an artifact of the implemented procedure.
  2. [Section 3, Eq. (26)] The “quadratic nonlinearity” DGP is X_t = B^2 F_t + V_t. Since B is a fixed 20×1 vector, B^2 is a fixed set of loadings, and this equation is linear in the factor F_t; it does not introduce any nonlinearity or sign dependence. Only the tanh specification in Eq. (27) is genuinely nonlinear. The simulation and Monte Carlo sections therefore do not establish recovery under two distinct nonlinear DGPs; they compare one linear and one nonlinear DGP. The near-identical RMSEs for “Nonlinear I” and “Nonlinear II” in Table 2 (0.648 vs. 0.650) are unsurprising given this structure.
  3. [Section 4.2, Table 4 and Figures 2–3] The forecast evaluation reports RMSEs and log scores without any measure of uncertainty, such as standard errors, Diebold-Mariano tests, or Bayesian posterior intervals. For example, the industrial production 12-month RMSE improvement (1.725 vs. 2.773) is presented as a substantial gain, but without uncertainty quantification it is impossible to assess whether the difference is statistically meaningful. This is a load-bearing limitation for the forecasting claim, which is a prominent part of the abstract and conclusions.
minor comments (5)
  1. [Sections 2.3.2 and 2.3.3] The two subsections both provide the identification of oil price shocks and contain nearly identical equations; one appears to be a leftover draft version and should be removed.
  2. [Figure 7 note] The note describes the figure as illustrating “size asymmetry” while the surrounding text discusses sign asymmetry; the terminology should be made consistent.
  3. [Section 4.3.1 vs. Table 3] The text states that the structural analysis uses the pre-pandemic sample 1974:01–2016:12, while Table 3 and the dataset description report 1974M1–2024M08; the sample used for the GIRFs should be stated unambiguously.
  4. [Section 2.2.2, algorithm Step 5] Step 5 says “Conditional on the draws obtained in Steps 1.-5.”, which references itself; it should presumably refer to Steps 1–4.
  5. [General] The manuscript contains several typos, including “hightened”, “postive”, and “F ABART” in the title, and no replication code or data are provided, which is particularly unfortunate given that the GIRF results cannot be verified from the text alone.

Circularity Check

2 steps flagged · score 7.0 of 10

The 'nonparametric' factor loadings are by construction OLS projections and the state space used for inference is linear, so the claimed sign asymmetries are not derived from the model as specified.

  1. self definitional [Section 2.2.2, Eq. (12)]
    "Without additional regularization, the projection naturally implies that F˜Ai ≈ Xi. ˜Ai allows to produce a linear approximation to the non-parametric multivariate model and obtain the elements of Γ Factor loadings matrix for each N-variable from the measurement equation (1), such that Xi,t = ˜Ai′Yt + ϵi,t (12)."

    The model is introduced as nonlinear in Eq. (3), Xt = F(Yt)+ηt, but the loadings used in the measurement equation are defined as the Moore-Penrose/OLS projection A~i = F†Xi. This makes the 'recovery of nonlinear relationships' tautological: by construction F A~i ≈ Xi, so the loadings are the best linear fit to the same data. The BART nonlinear function F(Yt) does not enter the linear measurement equation (12) used for estimation; the claimed non-parametric factor model reduces to a linear FAVAR with OLS loadings.

  2. other [Section 2.3.1, Eq. (14); Section 2.2.2, Step 5]
    "Based on this approximation to a linear model with Gaussian shocks the latent factors are drawn based on the Carter and Kohn (1994) algorithm. ... Since the model is nonlinear, the impulse responses are computed via Monte Carlo simulation, accounting for history-dependent effects and potential asymmetries in shock transmission."

    The only state space described for posterior inference is linear: transition Eq. (2) is a Gaussian linear VAR and measurement Eq. (12) is linear. The GIRF in Eq. (14) is a difference of conditional expectations; for a linear Gaussian system the response to a positive shock is the exact mirror image of the response to a negative shock. The paper's central empirical result—pronounced sign asymmetries in Figures 7-8—is therefore not a property of the estimated model as specified; the 'nonlinearity' invoked to justify asymmetric GIRFs was removed by the Crawford projection in Eq. (12).

full rationale

The paper's claimed contribution is a nonparametric BART factor model, but the estimation section replaces the BART measurement equation with a linear projection (Eq. 12) and then draws latent factors with Carter-Kohn from the resulting linear Gaussian state space. Consequently, the simulation 'recovery' of loadings is an identity of the projection, and the state space used for inference is linear. The GIRF asymmetry, the headline empirical finding, cannot be derived from a linear transition and linear measurement as described; the paper does not specify any alternative nonlinear simulation step that would reintroduce the BART function. The forecast evaluation is genuinely out-of-sample and the external instrument of Känzig (2021) anchors identification, so the paper is not wholly circular; nevertheless, the central nonlinearity and asymmetry claims reduce, by the paper's own equations, to a linear model. A further technical issue (not circularity): the 'quadratic nonlinear' DGP in Eq. (26), Xt = B^2 Ft + Vt, is linear in Ft, so only the tanh DGP is actually nonlinear.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The model introduces no new physical or conceptual entities; latent factors are standard. The main auxiliary assumptions are the linear projection approximation and the external instrument validity. All priors and hyperparameters follow standard practice or are chosen by hand.

free parameters (7)
  • Number of factors J = 7
    Chosen to explain approximately 47% of variance in the dataset, following Stock and Watson (2005); not estimated within the model.
  • Number of trees S = 250
    Fixed following Chipman et al. (2010) and Huber et al. (2020).
  • Prior hyperparameters alpha, beta, kappa = 0.95, 2, 2
    Recommended by Chipman et al. (2010); set by hand.
  • Prior tightness iota = 0.1
    Standard for US data per Banbura et al. (2010).
  • Sum-of-coefficients prior tightness lambda = 10*iota
    Follows Banbura et al. (2010).
  • Training sample length for prior means = 40 observations
    Used for OLS AR(1) prior means; standard Minnesota prior construction.
  • Shock size calibration = 10% oil price change
    Chosen to match Kanzig (2021); normalizes the impulse response comparison.
assumptions (4)
  • ad hoc to paper The BART function F(Y) can be replaced by a linear projection \tilde A = F^dagger X without losing the nonlinear information relevant for impulse responses.
    Invoked in Section 2.2.2 to enable Carter-Kohn sampling; the loss of nonlinearity is not analyzed.
  • domain assumption The external instrument m_t identifies the oil supply news shock and is orthogonal to other structural shocks.
    Equation (22) follows Kanzig (2021); standard external instrument assumption.
  • domain assumption Seven latent factors are sufficient to capture the common dynamics of the 187-variable panel.
    Chosen based on variance explained; no formal test.
  • standard math The prior on tree structures (Chipman et al. 2010) regularizes the BART fit adequately.
    Standard BART prior; used as recommended.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Let the Tree Decide: FABART A Non-Parametric Factor Model." pith.science (2026). https://pith.science/paper/OUVISUCP

@misc{pith2026250611551,
  author       = {Pith},
  title        = {Pith review of: Let the Tree Decide: FABART A Non-Parametric Factor Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OUVISUCP}},
  note         = {Machine review of arXiv:2506.11551}
}
read the original abstract

This article proposes a novel framework that integrates Bayesian Additive Regression Trees (BART) into a Factor-Augmented Vector Autoregressive (FAVAR) model to forecast macro-financial variables and examine asymmetries in the transmission of oil price shocks. By employing nonparametric techniques for dimension reduction, the model captures complex, nonlinear relationships between observables and latent factors that are often missed by linear approaches. A simulation experiment comparing FABART to linear alternatives and a Monte Carlo experiment demonstrate that the framework accurately recovers the relationship between latent factors and observables in the presence of nonlinearities, while remaining consistent under linear data-generating processes. The empirical application shows that FABART substantially improves forecast accuracy for industrial production relative to linear benchmarks, particularly during periods of heightened volatility and economic stress. In addition, the model reveals pronounced sign asymmetries in the transmission of oil supply news shocks to the U.S. economy, with positive shocks generating stronger and more persistent contractions in real activity and inflation than the expansions triggered by negative shocks. A similar pattern emerges at the U.S. federal state level, where negative shocks lead to modest declines in employment compared to the substantially larger contractions observed after positive shocks.

Figures

Figures reproduced from arXiv: 2506.11551 by the authors.

Figure 2
Figure 2. Forecast evaluation across time: 1-month ahead performance for macroeconomic [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Forecast evaluation across time: 1-month ahead performance for financial variables [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 6
Figure 6. Response to a positive oil price shock. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗
Figures from the paper (6 more)
Figure 7
Figure 7. Figure 7: Sign asymmetry in the response to oil price shocks. [PITH_FULL_IMAGE:figures/full_fig_p030_7.png]
Figure 8
Figure 8. Figure 8: Cross-state symmetry in employment responses to oil price shocks [PITH_FULL_IMAGE:figures/full_fig_p031_8.png]
Figure 9
Figure 9. Figure 9: Linear and non-linear DGPs based on one unobservable Factor. Straight black [PITH_FULL_IMAGE:figures/full_fig_p040_9.png]
Figure 10
Figure 10. Figure 10: Comparison of Posterior Factor Estimates with the True Factor [PITH_FULL_IMAGE:figures/full_fig_p043_10.png]
Figure 11
Figure 11. Figure 11: Responses to a negative oil price shock. [PITH_FULL_IMAGE:figures/full_fig_p050_11.png]
Figure 12
Figure 12. Figure 12: Posterior Density of State-Level Employment Response Differences to Oil Price [PITH_FULL_IMAGE:figures/full_fig_p051_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Clustered Local Projections for Short and Ultra-Short Time Series -- A Hierarchical Bayesian Framework

    econ.EM 2026-08 conditional novelty 6.0 of 10

    A hierarchical Bayesian local-projection estimator with sparse mixture clustering lets short and ultra-short time series borrow impulse-response information from longer, similar series, substantially reducing estimati...

Reference graph

Works this paper leans on

53 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    and Mumtaz, H

    Alessandri, P. and Mumtaz, H. (2017). Financial conditions and density forecasts for US output and inflation . Review of Economic Dynamics , 24:66--78

    work page 2017

  2. [2]

    Arce-Alfaro, G. (2025). The economic implications of oil supply uncertainty. Energy Economics , page 108425

    work page 2025

  3. [3]

    and Ng, S

    Bai, J. and Ng, S. (2008). Forecasting economic time series using targeted predictors. Journal of Econometrics , 146:304--317

    work page 2008

  4. [4]

    S., Brown, S

    Balke, N. S., Brown, S. P., and Yucel, M. K. (2002). Oil price shocks and the us economy: Where does the asymmetry originate? The Energy Journal , 23(3)

    work page 2002

  5. [5]

    Ba \'n bura, M., Giannone, D., and Reichlin, L. (2010). Large bayesian vector auto regressions. Journal of Applied Econometrics , 25(1):71--92

    work page 2010

  6. [6]

    and Hamilton, J

    Baumeister, C. and Hamilton, J. D. (2019). Structural interpretation of vector autoregressions with incomplete identification: Revisiting the role of oil supply and demand shocks. American Economic Review , 109(5):1873--1910

    work page 2019

  7. [7]

    Baumeister, C., Huber, F., and Marcellino, M. (2024). Risky oil: It's all in the tails. Technical report, National Bureau of Economic Research

    work page 2024

  8. [8]

    and Kilian, L

    Baumeister, C. and Kilian, L. (2015). Forecasting the real price of oil in a changing world: a forecast combination approach. Journal of Business & Economic Statistics , 33(3):338--351

    work page 2015

Show all 53 references
  1. [9]

    Baumeister, C., Korobilis, D., and Lee, T. K. (2022). Energy markets and global economic conditions. Review of Economics and Statistics , 104(4):828--844

    2022

  2. [10]

    Baumeister, C., Liu, P., and Mumtaz, H. (2013). Changes in the effects of monetary policy on disaggregate price dynamics. Journal of Economic Dynamics and Control , 37(3):543--560

    2013

  3. [11]

    Berisha, E., Chisadza, C., Clance, M., and Gupta, R. (2021). Income inequality and oil resources: Panel evidence from the united states. Energy Policy , 159:112603

    2021

  4. [12]

    S., Boivin, J., and Eliasz, P

    Bernanke, B. S., Boivin, J., and Eliasz, P. (2005). Measuring the effects of monetary policy: a factor-augmented vector autoregressive (favar) approach. The Quarterly Journal of Economics , 120(1):387--422

    2005

  5. [13]

    and Martinez-Bruera, P

    Caravello, T. and Martinez-Bruera, P. (2024). Disentangling sign and size non-linearities. SSRN 4704050

    2024

  6. [14]

    Carter, C. K. and Kohn, R. (1994). On gibbs sampling for state space models. Biometrika , 81(3):541--553

    1994

  7. [15]

    A., George, E

    Chipman, H. A., George, E. I., and McCulloch, R. E. (1998). Bayesian cart model search. Journal of the American Statistical Association , 93(443):935--948

    1998

  8. [16]

    A., George, E

    Chipman, H. A., George, E. I., and McCulloch, R. E. (2010). BART: Bayesian additive regression trees . The Annals of Applied Statistics , 4(1):266--298

    2010

  9. [17]

    R., Runcie, D

    Crawford, L., Flaxman, S. R., Runcie, D. E., and West, M. (2019). Variable prioritization in nonlinear black box methods: A genetic association case study . The annals of applied statistics , 13(2):958

    2019

  10. [18]

    C., Zhou, X., and Mukherjee, S

    Crawford, L., Wood, K. C., Zhou, X., and Mukherjee, S. (2018). Bayesian approximate kernel regression with variable selection. Journal of the American Statistical Association , 113(524):1710--1721

    2018

  11. [19]

    and Otrok, C

    Del Negro, M. and Otrok, C. (2008). Dynamic factor models with time-varying parameters: measuring changes in international business cycles. FRB of New York Staff Report , (326)

    2008

  12. [20]

    Doan, T., Litterman, R., and Sims, C. (1984). Forecasting and conditional projection using realistic prior distributions . Econometric Reviews , 3(1):1--100

    1984

  13. [21]

    J., Chen, L., and Gonzalo, J

    Dolado, J. J., Chen, L., and Gonzalo, J. (2020). Quantile factor models

    2020

  14. [22]

    Forni, M., Franconi, A., Gambetti, L., and Sala, L. (2023). Asymmetric transmission of oil supply news. Technical report, CEPR Discussion Papers

    2023

  15. [23]

    Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine. Annals of statistics , pages 1189--1232

    2001

  16. [24]

    and Amisano, G

    Geweke, J. and Amisano, G. (2010). Comparing and evaluating Bayesian predictive distributions of asset returns . International Journal of Forecasting , 26(2):216--230

    2010

  17. [25]

    Hamilton, J. D. (2003). What is an oil shock? Journal of econometrics , 113(2):363--398

    2003

  18. [26]

    Hauzenberger, N., Huber, F., and Klieber, K. (2023). Real-time inflation forecasting using non-linear dimension reduction techniques. International Journal of Forecasting , 39(2):901--921

    2023

  19. [27]

    Huber, F., Koop, G., Onorante, L., Pfarrhofer, M., and Schreiner, J. (2020). Nowcasting in a pandemic using non-parametric mixed frequency VARs . Journal of Econometrics

    2020

  20. [28]

    and Rossini, L

    Huber, F. and Rossini, L. (2022). Inference in Bayesian additive vector autoregressive tree models . The Annals of Applied Statistics , 16(1):104--123

    2022

  21. [29]

    Ish-Horowicz, J., Udwin, D., Flaxman, S., Filippi, S., and Crawford, L. (2019). Interpreting deep neural networks through variable importance. arXiv preprint arXiv:1901.09839

    2019 arXiv

  22. [30]

    Jacquinot, P., Kuismanen, M., Mestre, R., and Spitzer, M. (2009). An assessment of the inflationary impact of oil shocks in the euro area. The Energy Journal , 30(1):49--84

    2009

  23. [31]

    K \"a nzig, D. R. (2021). The macroeconomic effects of oil supply news: Evidence from opec announcements. American Economic Review , 111(4):1092--1125

    2021

  24. [32]

    Kase, N., Melosi, L., and Rottner, M. (2025). Estimating nonlinear heterogeneous agent models with neural networks. BIS Working Papers 1241, Bank for International Settlements

    2025

  25. [33]

    H., and Potter, S

    Koop, G., Pesaran, M. H., and Potter, S. M. (1996). Impulse response analysis in nonlinear multivariate models. Journal of econometrics , 74(1):119--147

    1996

  26. [34]

    Korobilis, D. (2013). Assessing the transmission of monetary policy using time-varying parameter dynamic factor models. Oxford Bulletin of Economics and Statistics , 75(2):157--179

    2013

  27. [35]

    and Schr \"o der, M

    Korobilis, D. and Schr \"o der, M. (2024). Monitoring multi-country macroeconomic risk: A quantile factor-augmented vector autoregressive (qfavar) approach. Journal of Econometrics , page 105730

    2024

  28. [36]

    Litterman, R. B. (1986a). Forecasting with bayesian vector autoregressions—five years of experience. Journal of Business & Economic Statistics , 4(1):25--38

    1986

  29. [37]

    Litterman, R. B. (1986b). A statistical approach to economic forecasting. Journal of Business & Economic Statistics , 4(1):1--4

    1986

  30. [38]

    McCracken, M. W. and Ng, S. (2016). Fred-md: A monthly database for macroeconomic research. Journal of Business & Economic Statistics , 34(4):574--589

    2016

  31. [39]

    Miescu, M., Mumtaz, H., and Theodoridis, K. (2024). Nonlinear dynamics of large oil supply news shocks. Cardiff Economics Working Papers

    2024

  32. [40]

    and Ricco, G

    Miranda-Agrippino, S. and Ricco, G. (2023). Identification with external instruments in structural vars. Journal of Monetary Economics , 135:1--19

    2023

  33. [41]

    Mohammed, R. I. and Mohammad, A. A. (2022). Stability conditions for limit cycle of smooth transition hyperbolic tangent autoregressive model. Journal of Algebraic Statistics , 13(2):2346--2357

    2022

  34. [42]

    Mumtaz, H. (2010). Evolving UK macroeconomic dynamics: a time-varying factor augmented VAR . Bank of England Working Paper

    2010

  35. [43]

    and Piffer, M

    Mumtaz, H. and Piffer, M. (2022). Impulse response estimation via flexible local projections. arXiv preprint arXiv:2204.13150

    2022 arXiv

  36. [44]

    Mumtaz, H., Sunder-Plassmann, L., and Theophilopoulou, A. (2018). The state-level impact of uncertainty shocks. Journal of Money, Credit and Banking , 50(8):1873--1913

    2018

  37. [45]

    and Surico, P

    Mumtaz, H. and Surico, P. (2012). Evolving international inflation dynamics: world and country-specific factors. Journal of the European Economic Association , 10(4):716--734

    2012

  38. [46]

    Mumtaz, H., Theophilopoulou, A., and Drossidis, T. (2024). The distributional effects of oil supply news shocks. Economics Letters , 240:111769

    2024

  39. [47]

    and Xiong, Y

    Pelger, M. and Xiong, Y. (2022). State-dependent factor models for us bond yields and stock returns. Journal of Financial Economics , 144:730--748

    2022

  40. [48]

    Robertson, J. C. and Tallman, E. W. (1999). Vector autoregressions: forecasting and reality . Economic Review-Federal Reserve Bank of Atlanta , 84(1):4

    1999

  41. [49]

    Sami, H. M. and Rahman, M. M. (2021). Determining the best activation functions for predicting stock prices in different stock exchanges through multivariable time series forecasting of lstm. American Journal of Engineering and Information Technology , 3(1):1--12

    2021

  42. [50]

    Sims, C. A. (1992). Interpreting the macroeconomic time series facts: The effects of monetary policy. European economic review , 36(5):975--1000

    1992

  43. [51]

    Sims, C. A. and Zha, T. (1999). Error bands for impulse responses. Econometrica , 67(5):1113--1155

    1999

  44. [52]

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

    2002

  45. [53]

    Stock, J. H. and Watson, M. W. (2005). Implications of dynamic factor models for var analysis. Technical report, National Bureau of Economic Research

    2005

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.