Pith. sign in

REVIEW 3 major objections 7 minor 103 references

Time-Varying Multi-Seasonal ARMA Models

T0 review · 3 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A new exact-likelihood Gibbs sampler fits time-varying multi-seasonal ARMA models and finds that Covid-19 nearly eliminated seasonal variation in New York–Miami air travel.

desk verdict Real extension of TV-AR to seasonal TV-ARMA with an approximate sampler; needs an exact-sampler benchmark before the central claim is fully supported. read the letter →

arxiv 2608.11481 v1 pith:UZX7SCWG submitted 2026-08-11 stat.ME

Ganna Fagerberg , Mattias Villani , Robert Kohn This is my paper · ORCID
classification stat.ME MSC 62M1062F1562M20
keywords time-varyingARMAmulti-seasonalmodelsBayesianinferencedynamicshrinkageprocessstochasticvolatilityexactlikelihoodGibbssamplingseasonality
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 establishes that a time-varying multi-seasonal ARMA model can be estimated from the exact likelihood by a Gibbs sampler, extending earlier work that handled only pure autoregressions. The model forces the AR and MA polynomials to stay stable and invertible at every time point, so the process never enters explosive or non-identifiable regions, provided no roots cancel. Parameter evolution and volatility are both governed by dynamic shrinkage priors, which allow long flat stretches, gradual drift, and sudden shifts. Applied to monthly New York–Miami air passengers from 1990 to 2024, the model finds that the strong seasonal pattern was essentially wiped out during the Covid-19 pandemic and has only partly returned. This matters because standard global-stationarity or differencing-based seasonal models cannot attribute such changes to specific time periods.

What carries the argument

The load-bearing machinery is the multiplicative seasonal TV-SARMA specification together with a recursive reparameterization: unrestricted parameters $\theta_t$ are mapped to partial autocorrelations, then to AR coefficients $\phi_t$ in the stability region, with a sign change to enforce MA invertibility, applied separately to each seasonal polynomial at every time $t$. This keeps the state-space structure conditionally linear-Gaussian once the error path and pre-sample values are known, which is what lets the authors sample the ARMA parameters jointly with the FFBSx algorithm based on the extended Kalman filter. The dynamic shrinkage process prior on the increments of $\theta_t$ is what allows long periods of constancy punctuated by jumps, and the same prior structure is used on the log-volatility innovation variance. The Gibbs sampler alternates between the parameter paths, the latent errors, the initial observations, and the volatility states, with data-driven priors on the initial state.

What would settle it

Re-run the air-passenger analysis and, from the fitted posterior, compute the roots of the relevant AR and MA polynomials at each time point; if during the pandemic months any pairwise root distance falls near zero, the 'seasonality wiped out' conclusion is confounded by near root cancellation. Alternatively, simulate artificial series with the same coefficient path but no root cancellation and check that the posterior seasonal coefficients remain separated from zero.

Watch

Extended reading notes

Core claim

The central claim is that the time-varying multi-seasonal ARMA model, written as $\phi_t(L)\prod_{j=1}^M \Phi_{jt}(L^{s_j}) y_t = \psi_t(L)\prod_{j=1}^M \Psi_{jt}(L^{s_j})\varepsilon_t$, can be fitted with the exact likelihood by adding two Gibbs steps to the existing pure-AR sampler: sampling the latent error path $\varepsilon_{-q_{\max}:T}$ and the pre-sample observations $x_0$, while keeping the ARMA parameters updated by the extended Kalman filter FFBSx. The parameterization via partial autocorrelations constrains all AR and MA polynomials to the stable and invertible region at every time point, and dynamic shrinkage priors allow the parameters to be constant, gradual, or abruptly changing. In simulations the sampler recovers the true time evolution and outperforms a nonparametric spectral benchmark on seasonal moving-average data; in the air-passenger case study both seasonal MA coefficients move near zero during Covid-19 and the first one partially recovers afterward.

Load-bearing premise

The whole inference stands on the assumption that the AR and MA polynomials do not share nearly common roots at any time point; when they nearly cancel, the parameters are unidentified and the sampler's credible intervals widen, so a period of near cancellation could be mistaken for a genuine change in seasonality.

Editorial extensions

If this is right

  • Users get posterior credible intervals for every AR and MA coefficient at every time point, so they can date when seasonality or persistence changes rather than assuming a fixed structure.
  • The exact-likelihood sampler makes the model usable with multiple seasonal periods and stochastic volatility, so high-frequency data with several seasonalities can be analyzed without pre-differencing.
  • Because the model constrains parameters to the stable and invertible region at each time point, it avoids explosive forecasts and parameter drift out of identifiable regions.
  • On seasonal moving-average data the model's spectral density estimates were better than the AdaptSpec benchmark, suggesting it can serve as a practical alternative for nonstationary seasonal series.
  • The air-passenger analysis attributes the pandemic-era drop in seasonality to the seasonal MA coefficients rather than to a change in noise variance.

Reading between the lines

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

  • Editorial inference: the reported loss of seasonality during Covid-19 could be contaminated by near root cancellation, since the paper does not compute root-distance diagnostics for the air-passenger series; a near-cancelling AR/MA pair can mimic a parameter shift.
  • Editorial inference: the same machinery could be used to build a formal test for a 'seasonal break' by comparing posterior probabilities of the seasonal coefficients crossing a threshold, something the paper does not implement.
  • Editorial inference: the DSP-prior stochastic volatility has smoother estimates than ordinary SV, which may improve short-term volatility forecasts in other macroeconomic and epidemiological series.
  • Editorial inference: a vector extension would let analysts measure whether seasonal shifts across multiple airports or cities are synchronized; the paper only mentions multivariate stability as future work.
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 / 7 minor

Summary. The paper proposes a time-varying multi-seasonal ARMA (TV-SARMA) model with regular and seasonal AR/MA polynomials, parameterized through partial autocorrelations to enforce local stability and invertibility at every time point, with parameter evolution governed by dynamic shrinkage process (DSP) priors and a DSP stochastic volatility component. Inference is by a Gibbs sampler that uses the exact likelihood by augmenting with latent error paths and pre-sample observations, while sampling the time-varying AR/MA parameter block with FFBSx, an extended Kalman filter-based approximate sampler inherited from the authors' pure-AR work. The method is evaluated on two simulation experiments (a non-seasonal TV-ARMA(1,1) and a seasonal TV-SMA(1,1)_12) against the AdaptSpec benchmark, and applied to monthly JFK-Miami air passenger data, where the seasonal MA coefficients are reported to collapse during the Covid-19 pandemic.

Significance. If the posterior sampler is correct, the paper fills a genuine gap: it provides a tractable, locally stable/invertible time-varying seasonal ARMA framework with an exact likelihood, and the empirical finding of a pandemic-induced change in seasonal travel patterns is substantively interesting. The paper is careful in its DGP-based simulation design, reports posterior intervals and multi-start convergence checks, and uses a spectral-density benchmark appropriate for nonstationary comparisons. The main uncertainty is not the model formulation but the validity of FFBSx as a posterior sampler for ARMA-type observation equations, which is currently supported by analogy to the pure-AR case and by self-convergence checks rather than by comparison with a simulation-consistent reference sampler. The manuscript would be strengthened by such a check, and by correcting the internal inconsistencies in the pre-sample observation update and the noise-variance notation.

major comments (3)
  1. [Section 3 (inference), Eq. (3.4); Experiments 1-2] The central claim that the Gibbs sampler draws from the joint posterior of the TV-SARMA model relies on FFBSx being accurate for the ARMA observation equation, but this is not established. Conditional on the error path and pre-sample observations, the observation equation is not the pure-AR form used in Fagerberg et al. (2026): the expanded coefficient vectors in (3.3) contain products of the underlying AR and MA parameters (e.g., -phi_{1t}*Phi_{1t} and psi_{1t}*Psi_{1t}), so the EKF linearization is more complex. The multi-start convergence checks in Figures 1 and 6 and the MSE comparisons with AdaptSpec in Figures 2 and 7 cannot detect a systematic bias shared by all runs. Please add a comparison with a simulation-consistent sampler (e.g., PGAS with ancestor sampling or PMMH) for at least the TV-ARMA(1,1) and TV-SMA(1,1)_12 DGPs, reporting posterior medians and HDIs for the parameters and for the spectral density. Without this, the 'exact' in the exact-likelihood claim is not matched by verification of exact posterior sampling.
  2. [Section 3.2, Eq. (3.7)] The state-space representation for the pre-sample observations is internally inconsistent as written. With alpha_t = (y_t, ..., y_{t-(pmax-1)}) and f = (1,0,...,0)^T, one has f^T alpha_t = y_t, so Eq. (3.7) becomes y_t = z_t^T tilde{psi}_t + y_t, which would force z_t^T tilde{psi}_t = 0. The state should presumably consist of lagged values (y_{t-1}, ..., y_{t-pmax}) with an appropriate f, or the transition innovation should include z_t^T tilde{psi}_t + epsilon_t and the observation equation should be y_t = f^T alpha_t. Please correct the equation and the state definition; this step is part of the exact-likelihood Gibbs update in Algorithm 1.
  3. [Section 5 (air passenger application)] The Covid-19 seasonality finding is presented as a robust feature of the data, but the paper does not report any identification diagnostics for the fitted TV-SMA(1,2)_12 model. Section 2.2 notes that (near) root cancellation can cause instability and poor MCMC convergence, and Experiment 1 (Section 4.3, Figures 4-5) shows exactly such behavior under near cancellation. Although the case-study model is pure MA rather than ARMA, near common roots between the regular and seasonal MA factors can still create near-non-identifiability. Please report posterior checks (e.g., distances between reciprocal roots of the MA factors, or trace and HDI width diagnostics over the pandemic window) to show that the decline in seasonal coefficients is not an artifact of a poorly identified parameter region.
minor comments (7)
  1. [Section 2.3, Eq. (2.5)] The index in theta_t = (theta_{1t}, ..., theta_{t,p+q})^T should be theta_{p+q,t}; as printed the subscript is nonsensical.
  2. [Sections 3.1 and 3.3] The noise variance notation is inconsistent: model (3.1)/(3.4) writes epsilon_t ~ N(0, sigma^2), but Section 3.3 and Algorithm 1 sample a time-varying sigma^2_{1:T}; use sigma_t^2 or exp(h_t^*) throughout for consistency.
  3. [Algorithm 1] Algorithm 1 contains several index typos: 'varepsilon_{-(qmaxl-1):T}', 'varepsilon_{-(qmax-1):T}', and 'z_0^top(0), defaulting to varepsilon_{-(qmaxl-1):0}' use inconsistent qmax versus qmaxl and -qmax versus -(qmax-1); make the lower index of the error path consistent (e.g., -qmax:T).
  4. [Section 3.1] The vector alpha_t = (varepsilon_t, ..., varepsilon_{t-(qmax+1)}) is said to lie in R^{qmax+1}, but the index range from t to t-(qmax+1) has qmax+2 elements; this affects the dimensions of G and the FFBS state vector.
  5. [Section 4.3 and Figure 4] The MA parameter is denoted psi_t in the model (4.3), but Figure 4's label and surrounding text refer to the 'sum of AR, phi, and MA, theta'; use psi_t to avoid confusion with the unrestricted parameters theta_t.
  6. [Section 3.2, matrix G_t] The second row of the companion matrix G_t is printed as '0 1 1 ... 0' and should presumably be '0 1 0 ... 0'.
  7. [General] The title and abstract emphasize multiple seasonal periods, but Experiments 1-2 and the case study all use a single seasonal period (s=12); a small simulation with two seasonal periods would better demonstrate the multi-seasonal extension.

Circularity Check

1 steps flagged · score 4.0 of 10

FFBSx accuracy for ARMA is transferred from the authors' prior pure-AR paper by self-citation, but the main modeling and empirical content is otherwise self-contained.

  1. self citation load bearing [Section 3, paragraph after Eq. (3.4)]
    "Fagerberg et al. (2026) demonstrate that the FFBSx algorithm is fast, robust, and accurate for models like (3.4); the accuracy of the sampler is verified by comparing the posterior from FFBSx to that from the simulation-consistent PGAS sampler (Lindsten et al., 2014). ... The same properties hold for the posterior of the AR and MA parameters in the more general ARMA case here, since conditional on the error path ε_{−qmax:T} and the pre-sample observations x⊤0, the observation equation (3.4) is of the same non-linear regression type as in Fagerberg et al. (2026)."

    The central claim that FFBSx is a trustworthy sampler for the ARMA parameter block is justified solely by transferring validation from Fagerberg et al. (2026), a paper by the same authors that treated the pure-AR special case. The transfer is asserted, not demonstrated: Eq. (3.3) contains products of time-varying AR and MA coefficients, e.g. −φ1tΦ1t and ψ1tΨ1t, so the observation equation is not literally the pure-AR regression with an added linear MA regressor. No PGAS or PMMH comparison is reported for any ARMA model, so the 'same non-linear regression type' claim is not independently checked. Because every downstream posterior summary, including the Covid-19 seasonality finding, is obtained from these FFBSx draws, this self-citation is load-bearing.

full rationale

No fitted parameter is renamed as a prediction: the simulations evaluate known data-generating processes, and the air-passenger finding is a posterior summary of the fitted model, not a predictive claim that reduces to its own input. The exact-likelihood Gibbs steps for latent errors and pre-sample observations, the dynamic shrinkage priors, and the stochastic volatility component are independently constructed. The main circularity concern is the FFBSx block: the paper's only support for applying the approximate EKF-based sampler to ARMA is a self-citation to the authors' prior pure-AR paper, with no simulation-consistent benchmark for the ARMA case. This is a genuine load-bearing self-citation, but it is not a definitional equivalence or a fitted-input-as-prediction reduction, so a moderate score of 4 is appropriate rather than 6 or higher. The root-cancellation assumption is acknowledged and tested by the authors, and is a standard identifiability condition rather than a circular step.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The model is a Bayesian state-space construction; it introduces no new physical or conceptual entities. The free parameters are tuning and prior choices, and the axioms are standard time-series and numerical assumptions. The load-bearing assumptions are the absence of root cancellation and the approximate accuracy of the EKF-based sampler.

free parameters (6)
  • Initial error state variance prior (alpha0 prior scale) = sigma_hat^2 estimated from static ARMA on first 30 observations
    The prior for the initial error state in the FFBS update (Section 3.1) is data-driven. The inference depends on this prior, especially for short series.
  • Initial observation state prior variance (gamma_hat_0) = estimated variance of y_t from first 30 observations
    The prior for pre-sample observations (Section 3.2) is chosen from the first 30 observations; it affects the exact likelihood.
  • Prior variance scale factor for the initial coefficient state in FFBSx = 0.5
    Section 4.1 scales the Gaussian prior for the initial AR/MA coefficients by 0.5 to keep the algorithm away from parameter boundaries.
  • Termination time t* for the error-path filter = 8 (Experiment 1), 20 (Experiment 2)
    Section 4.1: the FFBS filter for the error path is stopped at t* when the posterior covariance is essentially zero; chosen via inspection.
  • Offset value in h updates = machine precision in R
    Section 4.1: offset added in the log-volatility update to avoid zeros.
  • Number of spline knots for AdaptSpec benchmark = 15 (default is 10)
    Section 4.1: increased so AdaptSpec can better capture the seasonal spectral density; this affects the comparison outcome.
assumptions (7)
  • domain assumption The AR and MA polynomials share no common or near-common roots at any time t.
    Stated in Section 2.2; near root cancellation makes the parameters non-identified. Experiment 1 (Section 4.3) shows the sampler struggles when this is violated.
  • domain assumption The EKF linearization inside FFBSx yields draws from the correct full conditional posterior of the time-varying coefficients.
    The paper relies on Fagerberg et al. (2026) for the AR case and argues the same structure holds conditionally for ARMA (Section 3). No derivation or formal verification is given for the ARMA case.
  • standard math The Barndorff-Nielsen-Schou / Monahan mapping, with the sign modification, maps the unrestricted parameters onto the stable AR and invertible MA regions at every time t.
    Definition 1 in Section 2.2; standard result from Barndorff-Nielsen and Schou (1973) and Monahan (1984).
  • domain assumption The time-varying spectral density formula in Eq (4.1) is a valid local approximation for the TV-SARMA process.
    Used in Section 4.2 to define the MSE metric for all methods; assumes local stationarity at each time point.
  • domain assumption Detrending the air passenger series with locfit (bandwidth chosen by bw.ucv) does not distort the seasonal variation of interest.
    Section 5 pre-processes the data by subtracting a local mean before fitting the TV-SMA model; the analysis conditions on this detrending.
  • domain assumption Initial data-driven priors based on the first 30 observations are adequate for the exact likelihood inference.
    Sections 3.1 and 3.2; follows Marriott et al. (1993). If the first 30 observations are not representative, the priors on initial states could bias the posterior.
  • domain assumption The Gaussian innovation assumption for epsilon_t is appropriate for the data.
    Model (2.1) and (3.4) assume Gaussian errors; heavy-tailed alternatives are not considered.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Time-Varying Multi-Seasonal ARMA Models." pith.science (2026). https://pith.science/paper/UZX7SCWG

@misc{pith2026260811481,
  author       = {Pith},
  title        = {Pith review of: Time-Varying Multi-Seasonal ARMA Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UZX7SCWG}},
  note         = {Machine review of arXiv:2608.11481}
}
read the original abstract

We propose an ARMA model that allows for multiple seasonal periods and time varying parameters in both regular and seasonal components, building upon previous work for pure AR processes and the conditional likelihood. The model is parameterized to ensure stability and invertibility at each time point. The parameter evolution is governed by dynamic shrinkage processes, enabling extended periods of essentially constant parameters, gradual changes, and abrupt shifts. The model includes a stochastic volatility component to account for potentially heterogeneous noise, also modeled by a dynamic shrinkage process. A Gibbs sampler is developed using the exact likelihood, with separate updating steps for the latent errors and the unobserved pre-sample history of the process. The time-varying AR and MA parameters are sampled jointly using a fast posterior sampler based on the extended Kalman filter. The model and the efficiency of the Gibbs sampler are evaluated using simulated and real data. A case study on monthly air passenger data in the US during 1990-2024 reveals significant changes in seasonality during the Covid-19 pandemic.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

103 extracted references · 75 canonical work pages

  1. [1]

    Journal of Statistical Computation and Simulation , volume=

    A note on reparameterizing a vector autoregressive moving average model to enforce stationarity , author=. Journal of Statistical Computation and Simulation , volume=. 1986 , publisher=

    1986

  2. [2]

    Journal of Multivariate Analysis , year =

    Barndorff-Nielsen, O and Schou, G , title =. Journal of Multivariate Analysis , year =

  3. [3]

    International Statistical Review , pages=

    Normal variance-mean mixtures and z distributions , author=. International Statistical Review , pages=. 1982 , publisher=

    1982

  4. [4]

    , title =

    Bertolacci, M. , title =. GitHub repository , howpublished =. 2021 , publisher =

    2021

  5. [5]

    Brillinger, David R , year=. Time

  6. [6]

    Brockwell, Peter J and Davis, Richard A , year=. Time

  7. [7]

    Econometrics , volume=

    Triple the gamma—A unifying shrinkage prior for variance and variable selection in sparse state space and TVP models , author=. Econometrics , volume=. 2020 , publisher=

    2020

  8. [8]

    Biometrika , volume=

    The horseshoe estimator for sparse signals , author=. Biometrika , volume=. 2010 , publisher=

    2010

Show all 103 references
  1. [9]

    Journal of

    Estimation and comparison of multiple change-point models , author=. Journal of. 1998 , publisher=

    1998

  2. [10]

    The Annals of Statistics , volume=

    A likelihood approximation for locally stationary processes , author=. The Annals of Statistics , volume=. 2000 , publisher=

    2000

  3. [11]

    Handbook of statistics , volume=

    Locally stationary processes , author=. Handbook of statistics , volume=. 2012 , publisher=

    2012

  4. [12]

    Econometric reviews , volume=

    Forecasting and conditional projection using realistic prior distributions , author=. Econometric reviews , volume=. 1984 , publisher=

    1984

  5. [13]

    Everitt, Richard Geoffrey and Andrieu, Christophe and Davy, Manuel , journal=. Online. 2013 , publisher=

    2013

  6. [14]

    Journal of time series analysis , volume=

    An introduction to long-memory time series models and fractional differencing , author=. Journal of time series analysis , volume=. 1980 , publisher=

    1980

  7. [15]

    2020 , publisher=

    Time series analysis , author=. 2020 , publisher=

    2020

  8. [16]

    Econometrica , pages=

    A new approach to the economic analysis of nonstationary time series and the business cycle , author=. Econometrica , pages=. 1989 , publisher=

    1989

  9. [17]

    Journal of Computational and Graphical Statistics , volume=

    Enforcing stationarity through the prior in vector autoregressions , author=. Journal of Computational and Graphical Statistics , volume=. 2023 , publisher=

    2023

  10. [18]

    Journal of Econometrics , volume=

    Time-varying sparsity in dynamic regression models , author=. Journal of Econometrics , volume=. 2014 , publisher=

    2014

  11. [19]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume=

    Dynamic shrinkage processes , author=. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume=. 2019 , publisher=

    2019

  12. [20]

    Biometrika , volume=

    A note on enforcing stationarity in autoregressive-moving average models , author=. Biometrika , volume=. 1984 , publisher=

    1984

  13. [21]

    Journal of Econometrics , volume=

    Fully Bayesian analysis of ARMA time series models , author=. Journal of Econometrics , volume=. 1983 , publisher=

    1983

  14. [22]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    The analysis of multiple stationary time series , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1953 , publisher=

    1953

  15. [23]

    M. C. Jones , journal =. Randomly Choosing Parameters from the Stationarity and Invertibility Region of Autoregressive-Moving Average Models , urldate =

  16. [24]

    The Review of Economic Studies , volume=

    Time varying structural vector autoregressions and monetary policy , author=. The Review of Economic Studies , volume=. 2005 , publisher=

    2005

  17. [25]

    International Journal of Forecasting , volume=

    Forecasting costs incurred from unit differencing fractionally integrated processes , author=. International Journal of Forecasting , volume=. 1994 , publisher=

    1994

  18. [26]

    Bayesian Analysis , volume=

    Bayesian Lattice Filters for Time-Varying Autoregression and Time-Frequency Analysis , author=. Bayesian Analysis , volume=. 2016 , publisher=

    2016

  19. [27]

    M Lee and Gabriel A Rodriguez-Yam , title =

    Richard A Davis and Thomas C. M Lee and Gabriel A Rodriguez-Yam , title =. Journal of the American Statistical Association , volume =. 2006 , publisher =. doi:10.1198/016214505000000745 , URL =

    2006 doi

  20. [28]

    Journal of Computational and Graphical Statistics , volume =

    Sally Wood and Ori Rosen and Robert Kohn , title =. Journal of Computational and Graphical Statistics , volume =. 2011 , publisher =. doi:10.1198/jcgs.2010.09174 , URL =

    2011 arXiv

  21. [29]

    Journal of Time Series Analysis , volume =

    PRADO, RAQUEL and HUERTA, GABRIEL , title =. Journal of Time Series Analysis , volume =. doi:https://doi.org/10.1111/1467-9892.00280 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1111/1467-9892.00280 , abstract =

  22. [30]

    , year =

    Marriott, John and Ravishanker, Nalini and Gelfand, A. , year =

  23. [31]

    Shrinkage in the time-varying parameter model framework using the

    Knaus, Peter and Bitto-Nemling, Angela and Cadonna, Annalisa and Fr. Shrinkage in the time-varying parameter model framework using the. Journal of Statistical Software , volume=. 2021 , publisher=

    2021

  24. [32]

    Econometrics , VOLUME =

    Cadonna, Annalisa and Frühwirth-Schnatter, Sylvia and Knaus, Peter , TITLE =. Econometrics , VOLUME =. 2020 , NUMBER =

    2020

  25. [33]

    Doucet, Arnaud and Arnaudismacjp, Email and Johansen, Adam M , biburl =

  26. [34]

    Journal of the American Statistical Association , volume =

    Simon J Godsill and Arnaud Doucet and Mike West , title =. Journal of the American Statistical Association , volume =. 2004 , publisher =. doi:10.1198/016214504000000151 , URL =

    2004 doi

  27. [35]

    Backward simulation methods for

    Lindsten, Fredrik and Sch. Backward simulation methods for. Foundations and Trends in Machine Learning , volume=. 2013 , publisher=

    2013

  28. [36]

    Journal of Machine Learning Research , year =

    David Barber , title =. Journal of Machine Learning Research , year =

  29. [37]

    Godsill , keywords =

    Simo Särkkä and Pete Bunch and Simon J. Godsill , keywords =. A Backward-Simulation Based Rao-Blackwellized Particle Smoother for Conditionally Linear Gaussian Models , journal =. 2012 , note =. doi:https://doi.org/10.3182/20120711-3-BE-2027.00115 , url =

    2012

  30. [38]

    Dynamic Linear Model with Markov Switching , volume =

    Kim, Chang-Jin , year =. Dynamic Linear Model with Markov Switching , volume =. Journal of Econometrics , doi =

  31. [39]

    IEEE Transactions on signal processing , volume=

    Marginalized particle filters for mixed linear/nonlinear state-space models , author=. IEEE Transactions on signal processing , volume=. 2005 , publisher=

    2005

  32. [40]

    Particle

    Lindsten, Fredrik and Jordan, Michael I and Schon, Thomas B , journal=. Particle. 2014 , publisher=

    2014

  33. [41]

    and Jenkins, Gwilym M

    Box, George.E.P. and Jenkins, Gwilym M. , publisher =. Time Series Analysis: Forecasting and Control , edition =

  34. [42]

    2019 , publisher=

    Time Series Analysis Univariate and Multivariate Methods , author=. 2019 , publisher=

    2019

  35. [43]

    2019 , note =

    tvReg: Time-Varying Coefficients Linear Regression for Single and Multi-Equations in R , author =. 2019 , note =

    2019

  36. [44]

    2023 , note =

    tvReg: Time-Varying Coefficients Linear Regression for Single and Multi-Equations , author =. 2023 , note =

    2023

  37. [45]

    Journal of the American Statistical Association , volume=

    AdaptSPEC: Adaptive spectral estimation for nonstationary time series , author=. Journal of the American Statistical Association , volume=. 2012 , publisher=

    2012

  38. [46]

    Fast sampling of

    Rue, H. Fast sampling of. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume=. 2001 , publisher=

    2001

  39. [47]

    Particle

    Andrieu, Christophe and Doucet, Arnaud and Holenstein, Roman , year =. Particle. Journal of the Royal Statistical Society Series B , volume =

  40. [48]

    2010 , publisher=

    Time series: modeling, computation, and inference , author=. 2010 , publisher=

    2010

  41. [49]

    arXiv preprint arXiv:1405.1668 , year=

    Bayesian inference of time varying parameters in autoregressive processes , author=. arXiv preprint arXiv:1405.1668 , year=

    arXiv

  42. [50]

    Economic Quarterly , year=

    Time-varying parameter vector autoregressions: Specification, estimation, and an application , author=. Economic Quarterly , year=

  43. [51]

    Resenhas do Instituto de Matem

    Bayesian time-varying autoregressions: Theory, methods and applications , author=. Resenhas do Instituto de Matem

  44. [52]

    Journal of Computational and Graphical Statistics , volume=

    Bayesian mixtures of autoregressive models , author=. Journal of Computational and Graphical Statistics , volume=. 2011 , publisher=

    2011

  45. [53]

    2020 IEEE International Conference on Big Data (Big Data) , pages=

    Forecasting with multiple seasonality , author=. 2020 IEEE International Conference on Big Data (Big Data) , pages=. 2020 , organization=

    2020

  46. [54]

    Journal of the Operational Research Society , volume=

    Short-term electricity demand forecasting using double seasonal exponential smoothing , author=. Journal of the Operational Research Society , volume=. 2003 , publisher=

    2003

  47. [55]

    European Journal of Operational Research , volume=

    Forecasting time series with multiple seasonal patterns , author=. European Journal of Operational Research , volume=. 2008 , publisher=

    2008

  48. [56]

    Journal of the American Statistical Association , volume=

    Forecasting time series with complex seasonal patterns using exponential smoothing , author=. Journal of the American Statistical Association , volume=. 2011 , publisher=

    2011

  49. [57]

    Journal of time series analysis , volume=

    Data augmentation and dynamic linear models , author=. Journal of time series analysis , volume=. 1994 , publisher=

    1994

  50. [58]

    Carter, Chris K and Kohn, Robert , journal=. On. 1994 , publisher=

    1994

  51. [59]

    On particle

    Chopin, Nicolas and Singh, Sumeetpal S , year=. On particle

  52. [60]

    Particle

    Berntorp, Karl and Di Cairano, Stefano , journal=. Particle. 2017 , publisher=

    2017

  53. [61]

    Bayesian inference for logistic models using

    Polson, Nicholas G and Scott, James G and Windle, Jesse , journal=. Bayesian inference for logistic models using. 2013 , publisher=

    2013

  54. [62]

    Handbook of nonlinear filtering , volume=

    A tutorial on particle filtering and smoothing: Fifteen years later , author=. Handbook of nonlinear filtering , volume=

  55. [63]

    An overview of existing methods and recent advances in sequential

    Capp. An overview of existing methods and recent advances in sequential. Proceedings of the IEEE , volume=. 2007 , publisher=

    2007

  56. [64]

    arXiv preprint arXiv:1505.06356 , year=

    Particle ancestor sampling for near-degenerate or intractable state transition models , author=. arXiv preprint arXiv:1505.06356 , year=

    arXiv

  57. [65]

    2005 , publisher=

    Probabilistic robotics , author=. 2005 , publisher=

    2005

  58. [66]

    2009 , publisher=

    Dynamic linear models with R , author=. 2009 , publisher=

    2009

  59. [67]

    Package ‘stochvol’ , author=

  60. [68]

    Dealing with stochastic volatility in time series using the

    Kastner, Gregor , journal=. Dealing with stochastic volatility in time series using the

  61. [69]

    Ancillarity-sufficiency interweaving strategy (

    Kastner, Gregor and Fr. Ancillarity-sufficiency interweaving strategy (. Computational Statistics & Data Analysis , volume=. 2014 , publisher=

    2014

  62. [70]

    2001 , publisher=

    Probability and Statistical Models with Applications , author=. 2001 , publisher=

    2001

  63. [71]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    A skew extension of the t-distribution, with applications , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2003 , publisher=

    2003

  64. [72]

    Journal of Econometrics , volume=

    Stochastic volatility with leverage: Fast and efficient likelihood inference , author=. Journal of Econometrics , volume=. 2007 , publisher=

    2007

  65. [73]

    Stochastic volatility: likelihood inference and comparison with

    Kim, Sangjoon and Shephard, Neil and Chib, Siddhartha , journal=. Stochastic volatility: likelihood inference and comparison with

  66. [74]

    Electronic Journal of Statistics , volume=

    Sparsity information and regularization in the horseshoe and other shrinkage priors , author=. Electronic Journal of Statistics , volume=

  67. [75]

    Journal of Machine Learning Research , volume=

    Automatic differentiation in machine learning: a survey , author=. Journal of Machine Learning Research , volume=

  68. [76]

    Efficient

    Gerlach, Richard and Carter, Chris and Kohn, Robert , journal=. Efficient. 2000 , publisher=

    2000

  69. [77]

    2020 , publisher=

    An Introduction to Sequential Monte Carlo , author=. 2020 , publisher=

    2020

  70. [78]

    1997 , publisher=

    Makridakis, Spyros and Hibon, Michele , journal=. 1997 , publisher=

    1997

  71. [79]

    2015 , edition =

    Time series analysis: forecasting and control , author=. 2015 , edition =

    2015

  72. [80]

    arXiv preprint arXiv:2312.10487 , year=

    The Dynamic Triple Gamma Prior as a Shrinkage Process Prior for Time-Varying Parameter Models , author=. arXiv preprint arXiv:2312.10487 , year=

    arXiv

  73. [81]

    Time-varying multi-seasonal

    Fagerberg, Ganna and Villani, Mattias and Kohn, Robert , journal=. Time-varying multi-seasonal. 2026 , publisher=

    2026

  74. [82]

    Chan, Joshua CC and Eisenstat, Eric and Koop, Gary , journal=. Large. 2016 , publisher=

    2016

  75. [83]

    Bayesian Analysis of

    Kleibergen, Frank R and Hoek, Henk , year=. Bayesian Analysis of

  76. [84]

    Bayesian analysis of

    Marriott, John and Ravishanker, Nalini and Gelfand, Alan E and Pai, J , year=. Bayesian analysis of

  77. [85]

    IEEE Transactions on Automatic Control , volume=

    Input-output modeling of nonlinear systems with time-varying linear models , author=. IEEE Transactions on Automatic Control , volume=. 2000 , publisher=

    2000

  78. [86]

    Time-dependent

    Grenier, Yves , journal=. Time-dependent. 1983 , publisher=

    1983

  79. [87]

    Barnett, Glen and Kohn, Robert and Sheather, Simon , journal=. Robust. 1997 , publisher=

    1997

  80. [88]

    Bayesian estimation of switching

    Billio, Monica and Monfort, Alain and Robert, Christian P , journal=. Bayesian estimation of switching. 1999 , publisher=

    1999

  81. [89]

    Bayes inference in regression models with

    Chib, Siddhartha and Greenberg, Edward , journal=. Bayes inference in regression models with. 1994 , publisher=

    1994

  82. [90]

    Biometrika , volume=

    Maximum likelihood estimation of regression models with autoregressive-moving average disturbances , author=. Biometrika , volume=. 1979 , publisher=

    1979

  83. [91]

    Communications in Statistics-Theory and Methods , volume=

    Bayesian inferences and forecasts with moving averages processes , author=. Communications in Statistics-Theory and Methods , volume=. 1984 , publisher=

    1984

  84. [92]

    Triantafyllopoulos, Kostas and Nason, Guy P , journal=. A. 2007 , publisher=

    2007

  85. [93]

    Efficient estimation of

    Chan, Joshua CC and Eisenstat, Eric , journal=. Efficient estimation of. 2017 , publisher=

    2017

  86. [94]

    2010 , publisher=

    Introduction to time series modeling , author=. 2010 , publisher=

    2010

  87. [95]

    Journal of Time Series Analysis , volume=

    Time-varying autoregressions with model order uncertainty , author=. Journal of Time Series Analysis , volume=. 2002 , publisher=

    2002

  88. [96]

    Journal of the American Statistical Association , volume=

    Monte Carlo smoothing for nonlinear time series , author=. Journal of the American Statistical Association , volume=. 2004 , publisher=

    2004

  89. [97]

    1990 , publisher=

    Forecasting, structural time series models and the Kalman filter , author=. 1990 , publisher=

    1990

  90. [98]

    2006 , publisher=

    Bayesian forecasting and dynamic models , author=. 2006 , publisher=

    2006

  91. [99]

    Time-varying

    Huang, Renke and Zheng, Hao and Kuruoglu, Ercan E , journal=. Time-varying. 2013 , publisher=

    2013

  92. [100]

    Cox, Benjamin and Elvira, V. Sparse. IEEE Transactions on Signal Processing , volume=. 2023 , publisher=

    2023

  93. [101]

    Bayesian time-varying autoregressive models of

    Giudici, Paolo and Tarantino, Barbara and Roy, Arkaprava , journal=. Bayesian time-varying autoregressive models of. 2023 , publisher=

    2023

  94. [102]

    Macroeconometrics and time series analysis , pages=

    Bayesian time series analysis , author=. Macroeconometrics and time series analysis , pages=. 2010 , publisher=

    2010

  95. [103]

    Trend and Variance Adaptive

    Wu, Haoxuan and Schafer, Toryn LJ and Matteson, David S , journal=. Trend and Variance Adaptive. 2024 , publisher=

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.