pith. sign in

arxiv: 2606.03828 · v1 · pith:K3T7KBJPnew · submitted 2026-06-02 · 📊 stat.ME

Network Time Series Models for Multivariate Volatility Forecasting

Chiara Boetti , Matthew A. Nunes This is my paper

Pith reviewed 2026-06-28 08:51 UTC · model grok-4.3

classification 📊 stat.ME
keywords GNHAR modelmultivariate volatilityrealized varianceGranger causalityconnectedness indicesHAR forecastingequity spilloversmarket stability
0
0 comments X

The pith

The GNHAR model uses a directed network of spillovers to improve multivariate realized volatility forecasts over standard HAR benchmarks.

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

The paper introduces the generalised network heterogeneous autoregressive (GNHAR) model for forecasting vectors of realized variances. It adds cross-sectional dependencies via a directed graph obtained from Granger-causality tests or connectedness indices. This keeps the specification parsimonious compared to full multivariate models. Applied to ten equities in tranquil and crisis regimes, it delivers better short- and long-term forecasts than common HAR benchmarks. The approach also yields parameters that reflect changing market interdependencies.

Core claim

By incorporating cross-sectional spillovers through a directed graph inferred from Granger-causality tests or connectedness indices, the GNHAR model yields a parsimonious multivariate time series specification that improves upon common HAR model benchmarks under both short- and long-term forecasting in an application to ten equities over tranquil and crisis regimes.

What carries the argument

The directed graph of cross-market spillovers, inferred from Granger-causality tests or connectedness indices, which structures the lag terms in the heterogeneous autoregressive equations.

Load-bearing premise

The inferred directed graph from causality tests or connectedness indices accurately represents meaningful economic spillovers without adding spurious connections that harm forecast accuracy.

What would settle it

If replacing the inferred graph with a random graph or no connections produces equally good or better out-of-sample forecasts in the same equity application.

Figures

Figures reproduced from arXiv: 2606.03828 by Chiara Boetti, Matthew A. Nunes.

Figure 1
Figure 1. Figure 1: Edge density of the Granger causal graphs at either [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Parameter estimates over the CI22 graph for the global- [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
read the original abstract

Realized volatility has become a standard tool for measuring latent variation in financial assets, and its forecasting is crucial for a wide range of financial applications. We propose a network-based model for forecasting a vector of realized variance processes through the heterogeneous autoregressive (HAR) approach. The generalised network HAR (GNHAR) model incorporates cross-sectional spillovers through a directed graph inferred from Granger-causality tests or connectedness indices, yielding a parsimonious multivariate time series model specification. In an application to ten equities over tranquil and crisis regimes, the proposed GNHAR model improves upon common HAR model benchmarks under both short- and long-term forecasting. We also compare the network-based specification when the jump-continuous decomposition or node-specific option-implied variances are considered. Finally, unlike overparameterised models, our approach yields a concise set of parameters that track the strengthening or weakening of cross-market dependencies, providing a time-varying quantitative assessment of market stability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper proposes the generalised network HAR (GNHAR) model, a multivariate extension of the heterogeneous autoregressive framework that incorporates directed cross-sectional spillovers through an adjacency matrix inferred from Granger-causality tests or connectedness indices. It claims that this specification yields improved short- and long-term forecasts relative to standard HAR benchmarks in an application to ten equities across tranquil and crisis regimes, while remaining parsimonious; additional comparisons are made with jump-continuous decompositions and node-specific option-implied variances, and the model parameters are presented as providing a time-varying quantitative measure of market stability.

Significance. If the empirical gains are shown to be robust, the GNHAR approach would supply a concise, interpretable multivariate volatility model that directly quantifies evolving cross-market dependencies, which could be useful for risk management and systemic-risk monitoring.

major comments (3)
  1. [Abstract] Abstract: the central claim that GNHAR improves upon HAR benchmarks supplies no numerical results, error metrics, baseline comparisons, or implementation details, rendering it impossible to evaluate whether the data support the forecasting improvement.
  2. [Model specification] Model construction (inferred from the description of graph inference): the directed spillover graph is built from Granger-causality tests or connectedness indices performed on the same data used for forecasting; the manuscript must demonstrate that this procedure is separated from forecast evaluation (e.g., via rolling windows with strict hold-out) to rule out spurious edges that mechanically improve in-sample fit but fail to generalize across regimes.
  3. [Empirical application] Empirical application: no details are given on how the adjacency matrix is re-estimated or validated separately in tranquil versus crisis sub-samples, which is load-bearing for the claim that the model improves forecasts under both short- and long-term horizons without introducing data-dependent artifacts.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by naming the ten equities and the exact sample period.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed report. We address each major comment below and outline revisions that will strengthen the clarity and rigor of the manuscript without altering its core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that GNHAR improves upon HAR benchmarks supplies no numerical results, error metrics, baseline comparisons, or implementation details, rendering it impossible to evaluate whether the data support the forecasting improvement.

    Authors: We agree that the abstract would be strengthened by including concrete numerical support. In the revised version we will add the key out-of-sample metrics (e.g., average MSE and QLIKE ratios versus the benchmark HAR models) and a brief statement of the rolling-window implementation used in the ten-equity application. revision: yes

  2. Referee: [Model specification] Model construction (inferred from the description of graph inference): the directed spillover graph is built from Granger-causality tests or connectedness indices performed on the same data used for forecasting; the manuscript must demonstrate that this procedure is separated from forecast evaluation (e.g., via rolling windows with strict hold-out) to rule out spurious edges that mechanically improve in-sample fit but fail to generalize across regimes.

    Authors: The graph is constructed inside each rolling estimation window using only the in-sample observations up to time t, with forecasts generated strictly out-of-sample; the adjacency matrix is therefore never estimated on data that enter the forecast evaluation. We will add an explicit paragraph in Section 2.2 that documents the exact window lengths, the hold-out rule, and the fact that Granger-causality or connectedness tests are re-run at every re-estimation step on the training subsample only. revision: yes

  3. Referee: [Empirical application] Empirical application: no details are given on how the adjacency matrix is re-estimated or validated separately in tranquil versus crisis sub-samples, which is load-bearing for the claim that the model improves forecasts under both short- and long-term horizons without introducing data-dependent artifacts.

    Authors: We will insert a new subsection (3.3) that reports the precise rolling-window scheme applied separately to the pre-crisis, crisis, and post-crisis windows, together with the resulting adjacency matrices and a brief check that the out-of-sample gains remain statistically significant in each regime. This will make transparent that the network structure is re-estimated without using future information. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper defines the GNHAR model by incorporating a directed graph (inferred via Granger-causality tests or connectedness indices) as an input to the HAR specification for multivariate realized variance. This is a modeling choice that uses a preprocessing step on the data rather than defining any quantity in terms of itself. The central empirical claim is an out-of-sample forecasting improvement over standard HAR benchmarks in an application to ten equities, which is presented as a comparison result rather than a quantity that reduces to the graph inference by construction. No self-citations, uniqueness theorems, ansatzes smuggled via citation, or renamings of known results appear in the provided abstract or description. The derivation chain consists of a standard network-augmented time-series specification whose forecasting performance is evaluated externally against benchmarks, satisfying the criterion for a self-contained modeling paper.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The central modeling step depends on a data-driven network whose construction is itself a statistical procedure performed on the forecasting sample; this adds fitted structure whose validity is not independently verified outside the paper.

free parameters (2)
  • HAR lag coefficients
    Standard autoregressive parameters estimated from data for each node and spillover term.
  • network adjacency weights
    Directed connections inferred via Granger-causality tests or connectedness indices on the same dataset.
axioms (2)
  • domain assumption Granger-causality or connectedness measures identify stable, economically relevant spillovers
    Invoked to justify using the inferred graph as the spillover structure in the GNHAR specification.
  • domain assumption The multivariate volatility vector follows a linear network-augmented autoregressive process
    Core modeling assumption underlying the entire GNHAR formulation.
invented entities (1)
  • directed spillover graph no independent evidence
    purpose: To encode cross-sectional dependencies among the volatility processes
    Constructed from statistical tests on the estimation sample; no external validation or falsifiable prediction outside the forecasting exercise is mentioned.

pith-pipeline@v0.9.1-grok · 5682 in / 1411 out tokens · 29135 ms · 2026-06-28T08:51:44.663148+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

59 extracted references · 1 canonical work pages

  1. [1]

    Leopoldo Catania , year =

  2. [2]

    Rob Hyndman and George Athanasopoulos and Christoph Bergmeir and Gabriel Caceres and Leanne Chhay and Mitchell O'Hara-Wild and Fotios Petropoulos and Slava Razbash and Earo Wang and Farah Yasmeen , year =

  3. [3]

    Journal of the American statistical association , volume=

    The distribution of realized exchange rate volatility , author=. Journal of the American statistical association , volume=. 2001 , publisher=

    2001

  4. [4]

    Econometrica , volume=

    Modeling and forecasting realized volatility , author=. Econometrica , volume=. 2003 , publisher=

    2003

  5. [5]

    Roughing it up:

    Andersen, Torben G and Bollerslev, Tim and Diebold, Francis X , journal=. Roughing it up:. 2007 , publisher=

    2007

  6. [6]

    International Journal of Forecasting , volume=

    The impact of sentiment and attention measures on stock market volatility , author=. International Journal of Forecasting , volume=. 2020 , publisher=

    2020

  7. [7]

    Journal of the Korean Statistical Society , volume=

    Sparse vector heterogeneous autoregressive modeling for realized volatility , author=. Journal of the Korean Statistical Society , volume=. 2021 , publisher=

    2021

  8. [8]

    Baek, Changryong and D. Local. The Annals of Statistics , volume=. 2023 , publisher=

    2023

  9. [9]

    Journal of Financial Economics , volume=

    Separating microstructure noise from volatility , author=. Journal of Financial Economics , volume=. 2006 , publisher=

    2006

  10. [10]

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

    Econometric analysis of realized volatility and its use in estimating stochastic volatility models , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2002 , publisher=

    2002

  11. [11]

    Journal of Applied Econometrics , volume=

    Nets: Network estimation for time series , author=. Journal of Applied Econometrics , volume=. 2019 , publisher=

    2019

  12. [12]

    Bekaert, Geert and Hoerova, Marie , journal=. The. 2014 , publisher=

    2014

  13. [13]

    Forecasting

    Blair, Bevan J and Poon, Ser-Huang and Taylor, Stephen J , journal=. Forecasting. 2001 , publisher=

    2001

  14. [14]

    arXiv preprint arXiv:2512.10446 , year=

    Long memory network time series , author=. arXiv preprint arXiv:2512.10446 , year=

    work page arXiv

  15. [15]

    Journal of Econometrics , volume=

    Generalized autoregressive conditional heteroskedasticity , author=. Journal of Econometrics , volume=. 1986 , publisher=

    1986

  16. [16]

    Journal of Finance , volume=

    Option prices, implied price processes, and stochastic volatility , author=. Journal of Finance , volume=. 2000 , publisher=

    2000

  17. [17]

    Volatility transmission in emerging

    Bub. Volatility transmission in emerging. Journal of Banking & Finance , volume=. 2011 , publisher=

    2011

  18. [18]

    Journal of Econometrics , volume=

    The role of implied volatility in forecasting future realized volatility and jumps in foreign exchange, stock, and bond markets , author=. Journal of Econometrics , volume=. 2011 , publisher=

    2011

  19. [19]

    Scientific Data , volume=

    Twitter sentiment geographical index dataset , author=. Scientific Data , volume=. 2023 , publisher=

    2023

  20. [20]

    Journal of Applied Econometrics , volume=

    Modelling and forecasting multivariate realized volatility , author=. Journal of Applied Econometrics , volume=. 2011 , publisher=

    2011

  21. [21]

    Journal of Financial Economics , volume=

    The relation between implied and realized volatility , author=. Journal of Financial Economics , volume=. 1998 , publisher=

    1998

  22. [22]

    Journal of Financial Econometrics , volume=

    A simple approximate long-memory model of realized volatility , author=. Journal of Financial Econometrics , volume=. 2009 , publisher=

    2009

  23. [23]

    International Journal of Forecasting , volume=

    A vector heterogeneous autoregressive index model for realized volatility measures , author=. International Journal of Forecasting , volume=. 2017 , publisher=

    2017

  24. [24]

    Journal of Business & Economic Statistics , volume=

    Comparing predictive accuracy , author=. Journal of Business & Economic Statistics , volume=. 1995 , publisher=

    1995

  25. [25]

    The Economic Journal , volume=

    Measuring financial asset return and volatility spillovers, with application to global equity markets , author=. The Economic Journal , volume=. 2009 , publisher=

    2009

  26. [26]

    Better to give than to receive:

    Diebold, Francis X and Yilmaz, Kamil , journal=. Better to give than to receive:. 2012 , publisher=

    2012

  27. [27]

    On the network topology of variance decompositions:

    Diebold, Francis X and Y. On the network topology of variance decompositions:. Journal of Econometrics , volume=. 2014 , publisher=

    2014

  28. [28]

    Journal of Empirical Finance , volume=

    A long memory property of stock market returns and a new model , author=. Journal of Empirical Finance , volume=. 1993 , publisher=

    1993

  29. [29]

    Autoregressive conditional heteroscedasticity with estimates of the variance of

    Engle, Robert F , journal=. Autoregressive conditional heteroscedasticity with estimates of the variance of. 1982 , publisher=

    1982

  30. [30]

    Journal of the American Statistical Association , volume=

    Making and evaluating point forecasts , author=. Journal of the American Statistical Association , volume=. 2011 , publisher=

    2011

  31. [31]

    Econometrica , pages=

    Investigating causal relations by econometric models and cross-spectral methods , author=. Econometrica , pages=. 1969 , publisher=

    1969

  32. [32]

    Journal of Business & Economic Statistics , volume=

    Realized variance and market microstructure noise , author=. Journal of Business & Economic Statistics , volume=. 2006 , publisher=

    2006

  33. [33]

    Econometrica , volume=

    The model confidence set , author=. Econometrica , volume=. 2011 , publisher=

    2011

  34. [34]

    International Journal of Forecasting , volume=

    Testing the equality of prediction mean squared errors , author=. International Journal of Forecasting , volume=. 1997 , publisher=

    1997

  35. [35]

    Hollo, Daniel and Kremer, Manfred and Lo Duca, Marco , year=

  36. [36]

    A multivariate

    Hwang, Eunju and Hong, Won-Tak , journal=. A multivariate. 2021 , publisher=

    2021

  37. [37]

    ArXiv e-prints , archivePrefix = "arXiv", volume =

    Modelling, Detrending and Decorrelation of Network Time Series. ArXiv e-prints , archivePrefix = "arXiv", volume =

  38. [38]

    Generalized network autoregressive processes and the

    Knight, Marina and Leeming, Kathryn and Nason, Guy and Nunes, Matthew , journal=. Generalized network autoregressive processes and the

  39. [39]

    Is implied volatility more informative for forecasting realized volatility:

    Liang, Chao and Wei, Yu and Zhang, Yaojie , journal=. Is implied volatility more informative for forecasting realized volatility:. 2020 , publisher=

    2020

  40. [40]

    Finance Research Letters , volume=

    Economic policy uncertainty and stock market volatility , author=. Finance Research Letters , volume=. 2015 , publisher=

    2015

  41. [41]

    Journal of Empirical Finance , volume=

    Forecasting multivariate volatilities with exogenous predictors: An application to industry diversification strategies , author=. Journal of Empirical Finance , volume=. 2025 , publisher=

    2025

  42. [42]

    A comparison of direct and iterated multistep

    Marcellino, Massimiliano and Stock, James H and Watson, Mark W , journal=. A comparison of direct and iterated multistep. 2006 , publisher=

    2006

  43. [43]

    Journal of the Royal Statistical Society Series B , volume=

    A test for second-order stationarity and approximate confidence intervals for localized autocovariances for locally stationary time series , author=. Journal of the Royal Statistical Society Series B , volume=. 2013 , publisher=

    2013

  44. [44]

    Quantifying the economic response to

    Nason, Guy P and Wei, James L , journal=. Quantifying the economic response to. 2022 , publisher=

    2022

  45. [45]

    Forecasting

    Nason, Guy P and Palasciano, Henry Antonio , journal=. Forecasting. 2025 , publisher=

    2025

  46. [46]

    Modelling clusters in network time series with an application to presidential elections in the

    Nason, Guy and Salnikov, Daniel and Cortina-Borja, Mario , booktitle=. Modelling clusters in network time series with an application to presidential elections in the. 2024 , organization=

    2024

  47. [47]

    Journal of Econometrics , volume=

    High-dimensional copula-based distributions with mixed frequency data , author=. Journal of Econometrics , volume=. 2016 , publisher=

    2016

  48. [48]

    Journal of Economic Literature , volume=

    Forecasting volatility in financial markets: A review , author=. Journal of Economic Literature , volume=. 2003 , publisher=

    2003

  49. [49]

    Journal of Forecasting , volume=

    A comparison of conditional predictive ability of implied volatility and realized measures in forecasting volatility , author=. Journal of Forecasting , volume=. 2020 , publisher=

    2020

  50. [50]

    Journal of Forecasting , volume=

    Forecasting global stock market volatility: the impact of volatility spillover index in spatial-temporal graph-based model , author=. Journal of Forecasting , volume=. 2023 , publisher=

    2023

  51. [51]

    Realized volatility transmission between crude oil and equity futures markets:

    Sou. Realized volatility transmission between crude oil and equity futures markets:. Energy Economics , volume=. 2013 , publisher=

    2013

  52. [52]

    Available at SSRN 5113698 , year=

    Higher Order Dynamic Network Linear Models for Covariance Forecasting , author=. Available at SSRN 5113698 , year=

  53. [53]

    Modeling stochastic volatility:

    Taylor, Stephen J , journal=. Modeling stochastic volatility:. 1994 , publisher=

    1994

  54. [54]

    Applied Economics Letters , volume=

    Realized volatility forecasting in an international context , author=. Applied Economics Letters , volume=. 2015 , publisher=

    2015

  55. [55]

    Applied Economics , volume=

    Forecasting realized volatility of crude oil futures with equity market uncertainty , author=. Applied Economics , volume=. 2019 , publisher=

    2019

  56. [56]

    International Journal of Forecasting , volume=

    Multivariate volatility forecasts for stock market indices , author=. International Journal of Forecasting , volume=. 2021 , publisher=

    2021

  57. [57]

    Journal of Financial Econometrics , volume=

    Graph-based methods for forecasting realized covariances , author=. Journal of Financial Econometrics , volume=. 2025 , publisher=

    2025

  58. [58]

    Forecasting realized volatility with spillover effects:

    Zhang, Chao and Pu, Xingyue and Cucuringu, Mihai and Dong, Xiaowen , journal=. Forecasting realized volatility with spillover effects:. 2025 , publisher=

    2025

  59. [59]

    The Annals of Statistics , number =

    Zhu, Xuening and Pan, Rui and Li, Guodong and Liu, Yuewen and Wang, Hansheng , title =. The Annals of Statistics , number =