arxiv: 2605.00196 · v1 · submitted 2026-04-30 · 📊 stat.ME · math.PR· math.ST· q-fin.ST· stat.TH

Recognition: unknown

Modeling Stock Returns and Volatility Using Bivariate Gamma Generalized Laplace Law

Tomasz J. Kozubowski , Andrey Sarantsev , James A. Spiker

Authors on Pith no claims yet

Pith reviewed 2026-05-09 19:51 UTC · model grok-4.3

classification 📊 stat.ME math.PRmath.STq-fin.STstat.TH

keywords bivariate generalized Laplacegamma mixturestock returns modelingvolatilitymaximum likelihoodlinear regressionconvergence rates

0 comments

The pith

Observing volatility with returns simplifies generalized Laplace estimation to linear regression.

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

This paper proposes using a bivariate version of the gamma generalized Laplace distribution to jointly model stock returns and their volatility. The key innovation is treating the gamma mixing variable, which represents volatility, as observed data. Under this assumption, the maximum likelihood estimators simplify dramatically to those from ordinary linear regression, complete with closed-form expressions. Certain parameters converge to their true values at rates faster than the typical square root of the sample size. The model is then applied to analyze returns and volatility from major stock indices.

Core claim

When the gamma mixing variable is observed in addition to the returns, maximum likelihood estimation for the parameters of the bivariate gamma generalized Laplace distribution reduces to classical linear regression, yielding explicit estimators that achieve nonstandard convergence rates exceeding the square-root rate for some configurations.

What carries the argument

Bivariate observation of the gamma mixing variable in the normal mean-variance mixture model, which allows the likelihood equations to be solved via linear regression.

If this is right

Explicit estimators are available and computable without numerical optimization.
Some estimators converge at rates faster than the standard n to the power of -1/2.
The framework applies to financial data consisting of paired return and volatility series.

Where Pith is reading between the lines

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

This observed-mixing approach could be tested on other mixture distributions common in finance.
Faster convergence rates may improve the reliability of volatility forecasts derived from the model.
Practitioners could benefit from using implied volatility or realized volatility measures to enable this simpler estimation.

Load-bearing premise

The gamma mixing variable that captures volatility must be observed together with the stock returns.

What would settle it

If maximum likelihood estimates calculated numerically on data with observed mixing variable do not coincide with the linear regression formulas, the claimed simplification would be falsified.

Figures

Figures reproduced from arXiv: 2605.00196 by Andrey Sarantsev, James A. Spiker, Tomasz J. Kozubowski.

**Figure 1.** Figure 1: Scatter plot and density plots of (X, Y ). We take α = 2, β = 1, δ = 0, µ ∈ {±3, 0}, σ = 1. 2.6. Curved exponential family and Fisher information. Straightforward derivations show that the BGGL family given by the PDF (2.4) is an exponential family. However, it is not a classic family, where the dimension of sufficient statistics is the same as that of the parameter space. Here, the parameter vector has fi… view at source ↗

**Figure 2.** Figure 2: The quantile-quantile plots of X and Z from financial data of S&P 500 returns and volatility versus their theoretical distributions. Stock Index Volatility Index α β δ µ σ S&P 500 VIX 1.1399 8.3909 0.0219 -0.0009 0.0048 DJIA VXD 1.2639 12.864 0.01112 -0.00029 0.00401 NASDAQ 100 VXN 1.82288 7.81255 0.02777 -0.00078 0.00502 [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗

read the original abstract

We consider a generalization of the variance-gamma (generalized asymmetric Laplace) distribution, defined as a normal mean - variance mixture with a gamma mixing distribution. While this model is typically studied in the univariate setting, we assume that the gamma mixing variable is observed alongside the primary variable, resulting in a bivariate framework. In this setting, maximum likelihood estimation becomes significantly simpler than in the standard univariate case, reducing to a form of classical linear regression. We derive explicit expressions for the resulting estimators. For certain parameter configurations, the estimators exhibit nonstandard convergence rates, exceeding the usual square-root rate. Finally, we illustrate the applicability of this model in financial contexts by analyzing stock index returns and associated volatility for several major indices.

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

By treating the gamma mixer as observed, the paper reduces MLE for this bivariate generalized Laplace model to ordinary linear regression with closed-form estimators and occasional faster-than-root-n rates.

read the letter

The core move is straightforward once the bivariate observation is granted: the joint likelihood factors so that maximum likelihood for the parameters reduces directly to least-squares regression on the observed mixer and the primary variable. They derive the explicit estimator formulas and note that for certain parameter values the convergence rate exceeds the usual n to the minus one-half. This is new relative to the standard univariate variance-gamma literature, where the mixer stays latent and estimation requires numerical work or EM-type algorithms. The financial section applies the setup to several major stock indices, pairing returns with an associated volatility series, which gives a concrete illustration of how the model might be used for joint return-volatility work at the index level. The derivations appear clean and the reduction follows directly from the model definition rather than from any fitted quantities or circular steps. The main limitation is the modeling choice itself. Treating the gamma variable as observed buys the regression simplification and the rate claims, but it requires a reliable volatility proxy in practice; for index data this may be feasible, yet the paper's gains would shrink if that proxy is noisy or if the assumption is relaxed back to a latent mixer. The nonstandard rates also apply only in specific configurations, so the practical payoff is narrower than the headline suggests. This is aimed at researchers working on mixture models for financial returns who want a lighter estimation route in the bivariate case. The technical claims are internally consistent and the application is relevant, so it deserves a serious referee to check the rate derivations and the proxy quality in the data example.

Referee Report

2 major / 2 minor

Summary. The paper proposes a bivariate extension of the gamma generalized Laplace (variance-gamma) distribution for jointly modeling stock returns and volatility. By assuming the gamma mixing variable is observed alongside returns, maximum likelihood estimation reduces to a linear regression problem with closed-form explicit estimators derived. For certain parameter configurations, these estimators are claimed to exhibit convergence rates faster than the standard sqrt(n). The model is applied to empirical analysis of stock index returns and associated volatility for major indices.

Significance. If the observed-mixing assumption holds in practice and the rate claims are verified, the explicit estimators could simplify joint modeling of returns and volatility in finance. The derivation of closed-form estimators is a potential strength for methodological work in stat.ME, though practical impact depends on how well proxies for the mixing variable align with the gamma assumption.

major comments (2)

[Abstract] Abstract: The reduction of MLE to classical linear regression follows directly by construction from the bivariate model definition with the gamma mixing variable treated as observed; this modeling choice (rather than a data-driven or derived property) should be stated more explicitly as the source of the simplification to avoid overstating the methodological novelty.
[Estimation section] Estimation (likely around the section deriving the estimators): The nonstandard convergence rates exceeding the usual sqrt(n) are asserted for certain parameter configurations, but the specific conditions on the parameters, the form of the estimators, and the asymptotic analysis establishing the rate (e.g., via information matrix or direct expansion) must be provided in detail for verification.

minor comments (2)

[Application] In the application section, specify the exact proxy or measurement used for the 'associated volatility' (i.e., the observed gamma mixing variable) and discuss its alignment with the model assumption.
[Model definition] Ensure all notation for the bivariate density and parameters is introduced consistently before the estimation derivations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments, which help clarify the presentation of our modeling assumptions and strengthen the asymptotic claims. We address each major comment below and outline the revisions we will make.

read point-by-point responses

Referee: [Abstract] Abstract: The reduction of MLE to classical linear regression follows directly by construction from the bivariate model definition with the gamma mixing variable treated as observed; this modeling choice (rather than a data-driven or derived property) should be stated more explicitly as the source of the simplification to avoid overstating the methodological novelty.

Authors: We agree that the reduction of MLE to linear regression arises directly by construction once the gamma mixing variable is treated as observed in the bivariate model. While the abstract and introduction already describe this as a deliberate modeling choice leading to the bivariate framework, we acknowledge that the source of the simplification could be emphasized more explicitly. In the revised manuscript we will update the abstract to state clearly that the simplification follows from the bivariate model definition with the observed mixing variable, thereby avoiding any implication of greater novelty than intended. revision: yes
Referee: [Estimation section] Estimation (likely around the section deriving the estimators): The nonstandard convergence rates exceeding the usual sqrt(n) are asserted for certain parameter configurations, but the specific conditions on the parameters, the form of the estimators, and the asymptotic analysis establishing the rate (e.g., via information matrix or direct expansion) must be provided in detail for verification.

Authors: The manuscript derives explicit estimators and states that nonstandard rates hold for certain parameter configurations. To address the request for verification, we will expand the estimation section in the revision to specify the exact parameter conditions, restate the closed-form estimators, and supply a detailed asymptotic analysis (via direct expansion of the estimators or the relevant information matrix) that establishes the claimed rates. This addition will make the claims fully verifiable while preserving the existing results. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper defines a bivariate model by explicitly assuming the gamma mixing variable (volatility) is observed alongside returns. From this modeling choice the joint likelihood is shown to reduce to a linear regression problem, yielding closed-form MLEs and, for some parameter values, nonstandard convergence rates. These steps are direct algebraic consequences of the bivariate construction rather than fitted quantities renamed as predictions or self-referential definitions. No load-bearing self-citations, ansatzes smuggled via prior work, or uniqueness theorems imported from the authors appear in the derivation chain. The result is therefore self-contained and independent of its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central simplification rests on treating the gamma mixer as observed and on standard properties of the generalized Laplace distribution; no new entities are postulated.

axioms (1)

domain assumption The mixing variable follows a gamma distribution and is observed jointly with the return variable.
This observation assumption is what converts the usual intractable univariate MLE into linear regression.

pith-pipeline@v0.9.0 · 5429 in / 1127 out tokens · 35261 ms · 2026-05-09T19:51:55.486624+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

27 extracted references · 1 canonical work pages

[1]

Amponsah, Tomasz J

Charles K. Amponsah, Tomasz J. Kozubowski, Anna K. Panorska (2021). A general stochastic model for bivariate episodes driven by a Gamma sequence. Journal of Statistical Distributions and Applications 8 (7)

2021
[2]

Olaf Barndorff-Nielsen, J. Kent, M. Sorensen (1982). Normal variance-mean mixtures and z -distributions, International Statistical Review 50 , 145--159

1982
[3]

Likelihood-Based Risk Estimation for Variance-Gamma Models

Marco Bee, Maria Michela Dickson, Flavio Santi (2018). Likelihood-Based Risk Estimation for Variance-Gamma Models. Statistical Methods and Applications 27, 69--89

2018
[4]

Stochastic volatility modeling

Lorenzo Bergomi (2016). Stochastic volatility modeling. Chapman & Hall/CRC

2016
[5]

Berk (1972)

Robert H. Berk (1972). Consistency and asymptotic normality of MLE's for exponential models. Annals of Mathematical Statistics 43 (1), 193--204

1972
[6]

Bickel, Kjell A

Peter J. Bickel, Kjell A. Doksum (2015). Mathematical statistics. Basic ideas and selected topics. Second edition. Chapman & Hall/CRC

2015
[7]

Probability Theory

Alexander Borovkov (2013). Probability Theory . Springer

2013
[8]

Hanson (2011)

Ronald Christensen, Wesely Johnson, Adam Branscum, Timothy E. Hanson (2011). Bayesian Ideas for Data Analysis. An Introduction to Scientists and Statisticians. Chapman & Hall/CRC

2011
[9]

Clark (1973)

Peter K. Clark (1973). A subordinated stochastic process model with finite varinace for speculative prices. Econometrica 41 (1), 135--155

1973
[10]

Stationary-Increment Variance-Gamma and t -Models: Simulation and Parameter Estimation

Richard Finlay, Eugene Seneta (2008). Stationary-Increment Variance-Gamma and t -Models: Simulation and Parameter Estimation. International Statistical Review 76 (2), 167--186

2008
[11]

Gaunt, Andrey Sarantsev (2024)

Adrian Fischer, Robert E. Gaunt, Andrey Sarantsev (2024). The variance-gamma distribution: A review. Statistical Science 40 (2), 235--258

2024
[12]

Gaunt, Andrey Sarantsev (2024)

Adrian Fischer, Robert E. Gaunt, Andrey Sarantsev (2024). Modified Method of Moments for Generalized Laplace Distributions.Communications in Statistics, Simulation and Computation

2024
[13]

Hoff (2009)

Peter D. Hoff (2009). A First Course in Bayesian Statistical Methods. Springer

2009
[14]

A novel weighted likelihood estimation with empirical Bayes flavor

Md Mobarak Hossain, Tomasz Kozubowski, Krzysztof Podg\'orski (2018). A novel weighted likelihood estimation with empirical Bayes flavor. Communications in Statistics: Simulation and Computation 47 (2), 392--412

2018
[15]

Johnson, Samuel Kotz, N

Norman L. Johnson, Samuel Kotz, N. Balakrishnan (1994). Continuous univariate distributions. Second edition. Wiley

1994
[16]

Kozubowski, Krzysztof Podg\'orski (2001)

Samuel Kotz, Tomasz J. Kozubowski, Krzysztof Podg\'orski (2001). The Laplace distribution and generalizations. Springer

2001
[17]

Kozubowski, Krzysztof Podg\'orski, Igor Rychlik (2013)

Tomasz J. Kozubowski, Krzysztof Podg\'orski, Igor Rychlik (2013). Multivariate generalized Laplace distribution and related random fields. Journal of Multivariate Analysis 113, 59--72

2013
[18]

Silvapulle (2012)

Svetlana Litvinova, Mervyn J. Silvapulle (2012). A test when the Fisher information may be infinite, exemplified by a test for marginal independence in extreme value distributions. Journal of Statistical Planning and Inference 142 (6), 1506--1515

2012
[19]

Madan, Eugene Seneta (1990)

Dilip B. Madan, Eugene Seneta (1990). The variance gamma model for share market returns. The Journal of Business 63 (4), 511--524

1990
[20]

Edited by Frank W. J. Olver, Daniel W. Lozier, Ronald F. Boisvert, Charles W. Clark (2010). NIST Handbook of Mathematical Functions. Cambridge University Press

2010
[21]

Log Heston model for monthly average VIX

Jihyun Park, Andrey Sarantsev (2024). Log Heston model for monthly average VIX. arXiv:2410.22471

work page arXiv 2024
[22]

Karl Pearson, G. B. Jeffrey, Ethel M. Elderton (1929). On the distribution of the first product moment-coefficient, in samples drawn from an indefinitely large normal population. Biometrika 21 (1-4), 164--193

1929
[23]

Wallin (2015)

Krzysztof Podg\'orski, J. Wallin (2015). Maximizing leave-one-out likelihood for the location parameter of unbounded densities. Annals of the Institute of Statistical Mathematics 67 (1), 19--38

2015
[24]

Cox, Matteo Bottai, James Robins (2000)

Andrea Rotnitzky, David R. Cox, Matteo Bottai, James Robins (2000). Likelihood-Based Inference with Singular Information Matrix. Bernoulli 6 (2), 243--284

2000
[25]

Fitting the variance-gamma model to financial data

Eugene Seneta (2004). Fitting the variance-gamma model to financial data. Journal of Applied Probability 41, 177--187

2004
[26]

Shannon (1948)

Claude E. Shannon (1948). A mathematical theory of communication. Bell System Technical Journal 27 (4), 623--656

1948
[27]

Spiker (2024)

James A. Spiker (2024). A bi-variate gamma generalized Laplace distribution. Thesis, University of Nevada, Reno

2024