arxiv: 2604.15811 · v2 · submitted 2026-04-17 · 💰 econ.EM

Recognition: unknown

The realized copula of volatility

Kim Christensen , Wenjing Liu , Zhi Liu , Yoann Potiron

Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:35 UTC · model grok-4.3

classification 💰 econ.EM

keywords realized copulastochastic volatilityhigh-frequency datacopula estimationgoodness-of-fit testGumbel copulamultivariate asset pricesrealized variance

0 comments

The pith

The realized copula of volatility consistently estimates the dependence structure of latent stochastic volatilities from high-frequency returns.

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

The paper introduces the realized copula of volatility, a nonparametric statistic that measures codependency in the second moments of continuous-time multivariate asset price processes. It constructs this measure from local volatility estimates derived from high-frequency asset returns, providing an estimator for the empirical copula of the latent stochastic volatility. Consistency of the estimator is established under in-fill asymptotics for both fixed and expanding time spans, along with a functional central limit theorem for the empirical process of measurement error in the time-invariant marginal copula. A goodness-of-fit test is developed to evaluate hypotheses on the shape of that marginal copula. In an application to high-frequency futures data on U.S. equity and treasury bonds, the dependence between realized variance processes is found to be nearly perfectly described by a Gumbel copula.

Core claim

The realized copula of volatility is defined as the empirical copula computed from local volatility estimates obtained via high-frequency returns. This statistic converges in probability to the true copula of the latent volatility processes under in-fill asymptotics. When the observation period lengthens, the associated empirical process for the measurement error satisfies a functional central limit theorem. The framework further supplies a goodness-of-fit test for the time-invariant marginal copula of volatility.

What carries the argument

The realized copula of volatility, formed by applying the empirical copula function to locally estimated volatility measures constructed from high-frequency asset returns.

If this is right

The estimator serves as a reliable proxy for both the empirical and marginal copulas of volatility even when only moderate amounts of high-frequency data are available over short samples.
The associated goodness-of-fit test exhibits correct size and high power in finite-sample simulations.
In U.S. equity and treasury bond futures markets the dependence structure between realized variances is captured almost exactly by a Gumbel copula.
The asymptotic results cover both fixed-horizon and expanding-horizon settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The nonparametric nature of the estimator allows examination of volatility codependence across asset classes without committing to a specific parametric copula family in advance.
The Gumbel shape implies stronger joint movements during high-volatility episodes than during low-volatility episodes, which could guide the design of tail-risk hedging strategies.
Relaxing the time-invariant marginal copula assumption would permit study of how volatility dependence evolves over longer horizons.
The framework could be applied to other continuous-time processes, such as interest rates or exchange rates, to map out cross-market volatility linkages.

Load-bearing premise

The latent volatility processes are continuous and high-frequency returns permit consistent estimation of local volatility.

What would settle it

A Monte Carlo experiment in which the true volatility process is discontinuous would show whether the realized copula estimator fails to converge to the known copula of the latent processes.

Figures

Figures reproduced from arXiv: 2604.15811 by Kim Christensen, Wenjing Liu, Yoann Potiron, Zhi Liu.

**Figure 4.** Figure 4: Histogram of log-realized variance. Panel A: ES. Panel B: TY. Log-variance -8 -7 -6 -5 -4 -3 -2 -1 0 1 Relative frequency 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 ES Log-variance -8 -7 -6 -5 -4 -3 Relative frequency 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 TY Note. In Panel A, we show a histogram (plotted in terms of relative frequency) of the log-realized variance of ES, whereas Panel … view at source ↗

**Figure 5.** Figure 5: reports the average daily log-spot realized variance for ES in Panel A and TY in Panel B. The smoothing attenuates the measurement error in the spot volatility estimator a bit to better capture any comovement. There is forceful evidence of a positive relationship, most notably during periods of elevated market distress. To support this claim, in Panel A of [PITH_FULL_IMAGE:figures/full_fig_p026_5.png] view at source ↗

**Figure 6.** Figure 6: Scatter plot of log-realized variance and realized copula of [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗

**Figure 7.** Figure 7: Graphical representation of the realized copula of volatilit [PITH_FULL_IMAGE:figures/full_fig_p028_7.png] view at source ↗

read the original abstract

We study a new measure of codependency in the second moment of a continuous-time multivariate asset price process, which we name the realized copula of volatility. The statistic is based on local volatility estimates constructed from high-frequency asset returns and affords a nonparametric estimator of the empirical copula of the latent stochastic volatility. We show consistency of our estimator with in-fill asymptotic theory, either with a fixed or increasing time span. In the latter setting, we derive a functional central limit theorem for the empirical process associated with the measurement error of the time-invariant marginal copula of volatility. We also develop a goodness-of-fit test to evaluate hypotheses about the shape of the latter. In a simulation study, we demonstrate that our estimator is a good proxy of both the empirical and marginal copula of volatility, even with a moderate amount of high-frequency data recorded over a relatively short sample. The goodness-of-fit test is found to exhibit size control and excellent power. We implement our framework on high-frequency transaction data from futures contracts that track the U.S. equity and treasury bond market. A Gumbel copula is found to offer a near-perfect bind between the realized variance processes in these data.

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.

Desk Editor's Note private letter to a colleague

The paper defines a realized copula of volatility from high-frequency local-vol estimates, proves consistency plus a functional CLT, and reports a near-perfect Gumbel fit on equity and bond futures, but the whole thing rests on continuous price paths.

read the letter

The main thing here is a new nonparametric statistic that turns local volatility estimates into an empirical copula for the latent volatility processes. They show consistency under in-fill asymptotics for both fixed and growing time spans, plus a functional central limit theorem for the marginal copula estimator when the sample lengthens. A goodness-of-fit test comes with it. Simulations indicate the estimator tracks the true copula reasonably well even with moderate high-frequency data, and the test keeps size and has decent power. The empirical piece applies this to U.S. equity and treasury futures and finds the Gumbel copula binds the realized variance series almost perfectly. That is the concrete takeaway a reader would remember. The soft spot is the maintained continuity assumption on the latent price process. The local-volatility step and all the asymptotic results rely on consistent spot-vol recovery via in-fill theory, which breaks down at jumps. Futures data routinely contain jumps, so the measurement error goes straight into the copula and the reported Gumbel result. The paper does not appear to supply jump-robust alternatives or checks that would show how sensitive the finding is. This is aimed at financial econometricians who work with multivariate high-frequency volatility models and want a nonparametric dependence measure. A reader who already uses realized volatility tools and cares about copula restrictions on volatility will get the most out of the asymptotics and the test. It deserves a serious referee because the estimator is new, the theory is carried through, and there is both simulation evidence and a real-data illustration. I would send it out for review.

Referee Report

3 major / 3 minor

Summary. The paper introduces the realized copula of volatility, a nonparametric estimator of the empirical copula of latent stochastic volatilities constructed from high-frequency local volatility estimates. It establishes consistency of the estimator under in-fill asymptotics (fixed or expanding time span), derives a functional central limit theorem for the associated empirical process, develops a goodness-of-fit test for the shape of the time-invariant marginal copula, validates performance via simulations, and applies the method to high-frequency futures data on U.S. equity and Treasury bonds, concluding that a Gumbel copula provides a near-perfect fit to the dependence between the realized variance processes.

Significance. If the continuity assumption holds, the framework supplies a new nonparametric tool for measuring volatility codependence with explicit asymptotic theory and a GoF test, supported by simulations that demonstrate reliable proxy performance even with moderate high-frequency samples. The empirical finding of strong upper-tail dependence via the Gumbel copula between equity and bond volatility processes would be of interest for multivariate volatility modeling and risk applications.

major comments (3)

[§2, Theorem 1] §2 (model setup) and Theorem 1 (consistency): the estimator and its consistency rest on the maintained assumption that the latent price process is a continuous semimartingale, allowing consistent local-volatility recovery by in-fill asymptotics. The empirical application in §5 uses equity and bond futures transaction data, which standardly contain price jumps; the paper does not report jump-robustness checks or simulations with jumps, so the measurement-error propagation into the copula estimator and the Gumbel fit remain unexamined.
[§5] §5 (empirical results): the headline claim that 'a Gumbel copula is found to offer a near-perfect bind' is obtained by feeding the local-volatility estimates directly into the empirical copula. Because the consistency and FCLT results are derived only under continuity, any jump-induced inconsistency in the local-vol estimates directly affects the reported copula shape and GoF statistic; no alternative jump-robust local-vol estimator or sensitivity analysis is provided.
[Theorem 3] Theorem 3 (functional CLT): the limiting process is derived for the measurement error of the time-invariant marginal copula under the continuous-semimartingale assumption. It is unclear whether the result extends when the local-volatility estimator is inconsistent at jump times, which would alter the rate and the form of the empirical process used for the GoF test.

minor comments (3)

[Abstract, §5] The phrase 'near-perfect bind' in the abstract and §5 is informal; replace with a quantitative statement of the GoF statistic or distance to the Gumbel copula.
[§3] Notation for the realized copula estimator and the local-volatility kernel should be introduced once and used consistently; some passages mix 'realized variance' and 'local volatility' without explicit linkage.
[§4] The simulation design in §4 could usefully report coverage of the functional CLT bands or finite-sample bias of the copula estimator under increasing jump intensity.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the careful and constructive report. We address each major comment below and indicate the revisions we will make to strengthen the manuscript while remaining faithful to its theoretical scope.

read point-by-point responses

Referee: [§2, Theorem 1] §2 (model setup) and Theorem 1 (consistency): the estimator and its consistency rest on the maintained assumption that the latent price process is a continuous semimartingale, allowing consistent local-volatility recovery by in-fill asymptotics. The empirical application in §5 uses equity and bond futures transaction data, which standardly contain price jumps; the paper does not report jump-robustness checks or simulations with jumps, so the measurement-error propagation into the copula estimator and the Gumbel fit remain unexamined.

Authors: We agree that Theorem 1 and the consistency of the realized copula estimator are derived under the continuous semimartingale assumption. Futures transaction data can contain jumps, and we did not previously examine their effect on the copula estimates. We will add Monte Carlo experiments that inject jumps into the price process and report the resulting bias in the empirical copula and the Gumbel goodness-of-fit statistic. A brief discussion of this robustness exercise will be inserted in §5. revision: yes
Referee: [§5] §5 (empirical results): the headline claim that 'a Gumbel copula is found to offer a near-perfect bind' is obtained by feeding the local-volatility estimates directly into the empirical copula. Because the consistency and FCLT results are derived only under continuity, any jump-induced inconsistency in the local-vol estimates directly affects the reported copula shape and GoF statistic; no alternative jump-robust local-vol estimator or sensitivity analysis is provided.

Authors: The referee is correct that the reported Gumbel fit and GoF p-value rest on the local-volatility estimates. We will revise §5 to include a sensitivity analysis that replaces the baseline local-volatility estimator with two jump-robust alternatives (truncated realized volatility and bipower-variation-based local volatility). We will report the resulting changes, if any, in the estimated copula and the GoF test outcome. revision: yes
Referee: [Theorem 3] Theorem 3 (functional CLT): the limiting process is derived for the measurement error of the time-invariant marginal copula under the continuous-semimartingale assumption. It is unclear whether the result extends when the local-volatility estimator is inconsistent at jump times, which would alter the rate and the form of the empirical process used for the GoF test.

Authors: Theorem 3 is stated and proved under the continuous-semimartingale assumption maintained throughout the paper. Deriving the corresponding functional CLT when local-volatility estimates are inconsistent at jump times would require a new theoretical argument that lies outside the present contribution. We will add an explicit caveat in the text of §4 and §5 stating that the asymptotic justification of the GoF test applies under continuity and that the empirical results should be interpreted with this qualification. revision: partial

standing simulated objections not resolved

Extension of the functional central limit theorem in Theorem 3 to price processes containing jumps

Circularity Check

0 steps flagged

No circularity: estimator and asymptotics are defined from high-frequency data and standard in-fill theory without self-referential reduction.

full rationale

The realized copula is constructed directly as the empirical copula of local volatility estimates extracted from high-frequency returns. Consistency and the functional CLT for the measurement-error process are derived from in-fill asymptotics under the maintained continuous semimartingale assumption; these steps invoke standard stochastic calculus results rather than any fitted parameter or prior self-citation that encodes the target copula shape. The Gumbel-copula finding is an empirical observation on futures data, not a prediction forced by the estimator definition. No self-definitional, fitted-input, or uniqueness-imported steps appear in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard continuous-time semimartingale assumptions for asset prices and the existence of a well-defined copula for the latent volatility vector; no free parameters or invented entities are introduced in the abstract.

axioms (2)

domain assumption Asset price processes are continuous semimartingales with stochastic volatility
Invoked to justify local volatility estimation from high-frequency returns
domain assumption The marginal copula of volatility is time-invariant
Required for the functional CLT on the empirical process of measurement error

pith-pipeline@v0.9.0 · 5504 in / 1253 out tokens · 21006 ms · 2026-05-10T07:35:18.103045+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references

[1]

Answering the skeptics: Yes, standard volatility models do provide accurate forecasts,

Andersen, T. G., and T. Bollerslev, 1998, “Answering the skeptics: Yes, standard volatility models do provide accurate forecasts,” International Economic Review , 39, 885–905. Andersen, T. G., T. Bollerslev, F. X. Diebold, and H. Ebens, 2001, “T he distribution of realized stock return volatility,” Journal of Financial Economics , 61, 43–76. Andersen, T. ...

1998
[2]

Variance Risk Premiums,

Carr, P., and L. Wu, 2009, “Variance Risk Premiums,” Review of Financial Studies , 22, 1311–1341. Chong, C. H., and V. Todorov, 2023, “The ﬁne structure of volatilit y dynamics,” Working paper. Christensen, K., R. C. A. Oomen, and R. Ren` o, 2022, “The drift bu rst hypothesis,” Journal of Econometrics, 227, 461–497. Christensen, K., M. Thyrsgaard, and B. ...

2009