The Variance-Gamma Process for Option Pricing

Basic Information

• Paper ID: 2510.14093
• Title: The Variance-Gamma Process for Option Pricing
• Authors: Rohan Shenoy (Imperial College London), Peter Kempthorne (MIT)
• Classification: q-fin.MF (Quantitative Finance - Mathematical Finance)
• Publication Date: August 2024
• Paper Link: https://arxiv.org/abs/2510.14093

Abstract

This paper explores the concept of stochastic time subordination in modeling stock price dynamics. It first investigates the Laplace distribution as a result of Gaussian variance mixture, particularly providing more efficient volatility estimation procedures through absolute moments. The Laplace model is generalized to characterize the robust variance-gamma model proposed by Madan et al. as Gamma-subordinated Brownian motion, with European call option pricing via the Esscher transform method. The findings demonstrate that the variance-gamma model can empirically explain excess kurtosis observed in log-return data, rejecting Black-Scholes assumptions in hypothesis tests.

Research Background and Motivation

1. Core Problem

Traditional financial market models, particularly the Black-Scholes option pricing model, assume that asset returns follow a lognormal distribution. However, empirical financial data reveals that daily return distributions exhibit the following characteristics:

• Higher kurtosis than normal distribution (leptokurtic)
• Thicker tails
• Sharper peaks near zero

2. Problem Significance

These deviations from normality lead to systematic biases in the Black-Scholes model, particularly:

• Implied volatility smile phenomenon
• Moneyness bias in option pricing
• Underestimation of extreme market events

3. Limitations of Existing Methods

• Black-Scholes Model: Assumes lognormal distribution, unable to capture fat tails in actual return distributions
• Traditional Parameter Estimation: Under non-normal distribution assumptions, sample mean and variance are no longer optimal estimators
• Risk-Neutral Measure: The risk-neutral measure under Black-Scholes is oversimplified, losing the drift term

4. Research Motivation

By introducing the concept of stochastic time subordination, construct models that better describe the characteristics of actual financial data, improving the accuracy of option pricing.

Core Contributions

1. Proposed efficient parameter estimation methods for the Laplace distribution: Demonstrated that under Laplace distribution assumptions, the sample median and mean absolute deviation are more efficient than traditional sample mean and standard deviation
2. Established a complete theoretical framework for the variance-gamma process: Characterized it as Gamma-subordinated Brownian motion
3. Developed option pricing methods based on the Esscher transform: Constructed a risk-neutral measure for the variance-gamma model
4. Provided empirical validation: Using S&P 500 data, verified model superiority, rejecting Black-Scholes assumptions in 92.3% of weeks

Methodology Details

Task Definition

The core task of this research is:

• Input: Stock price time series data
• Output: Theoretical prices for European options
• Constraints: Construct arbitrage-free option pricing models capable of explaining statistical characteristics of actual return distributions

Model Architecture

1. Laplace Distribution Foundation

Definition 1 (Classical Laplace Distribution): A random variable X follows a classical Laplace distribution CL(θ,s) with parameters θ (mean) and s (scale parameter), with probability density function:

$f_X(x; θ, s) = \frac{1}{2s}e^{-\frac{|x-θ|}{s}}, \quad x ∈ \mathbb{R}$

Characteristic function: $\psi_X(t) = \frac{e^{itθ}}{1 + s^2t^2}$

2. Gaussian Variance Mixture Representation

Proposition 2: Let Z follow a standard normal distribution and V be an independent standard exponential distribution. Then the random variable $X = \sqrt{2V}Z$ follows a standard Laplace distribution.

This establishes an important characterization of the Laplace distribution as a Gaussian variance mixture: $X = θ + σ\sqrt{V}Z$ where V is random variance.

3. Variance-Gamma Distribution

Definition 4 (Variance-Gamma Distribution): Let $X = c + θV + σ\sqrt{V}Z$, where:

• c, θ ∈ ℝ, σ > 0 are fixed parameters
• Z follows a standard normal distribution
• V is an independent Gamma(α,β) distribution

Then X follows a variance-gamma distribution VG(c, θ, σ, α, β).

Characteristic function: $\psi_X(t) = \left(1 - \frac{iθt}{β} + \frac{t^2σ^2}{2β}\right)^{-α}$

4. Variance-Gamma Process

Definition 7 (Variance-Gamma Process): Let $(b_t)_{t≥0} = b(t; θ, σ)$ be a Brownian motion and $(γ_t)_{t≥0} = γ(t; 1, ν)$ be a Gamma process. Then the time-subordinated process:

$X_t = X(t; θ, σ, ν) = b(γ(t; 1, ν); θ, σ) = θγ_t + σW_{γ_t}$

is called a variance-gamma process VG(t; θ, σ, ν).

Technical Innovations

1. Improved Parameter Estimation

Under Laplace distribution assumptions:

• Location Parameter: The asymptotic variance of the sample median is half that of the sample mean
• Scale Parameter: The asymptotic variance of mean absolute deviation is 20% smaller than that of scaled sample standard deviation

2. Time Subordination Mechanism

Through random transformation of time via the Gamma process:

• Economically relevant time itself becomes a stochastic process
• Capable of explaining fat tails and sharp peaks in return distributions
• Preserves the Markov property

3. Esscher Transform Pricing

Definition 8 (Esscher Transform): For the density $f(x,t)$ of a Lévy process $(X_t)_{t≥0}$, its Esscher transform is:

$\hat{f}(x, t, h) = \frac{e^{hx}f(x,t)}{M(h,t)}$

where $M(·,t)$ is the moment generating function at time t.

The risk-neutral parameter $h^*$ is determined by the relationship: $e^{rt} = \frac{M(h^* + 1, t)}{M(h^*, t)}$

Experimental Setup

Dataset

1. Daily Return Data: S&P 500 index daily log returns from January 2022 to January 2024 (262 observations)
2. Option Data: CBOE-traded S&P 500 futures options, August 2022 to August 2023, European call options expiring in 7 days (46,135 option prices)

Evaluation Metrics

1. Statistical Tests: Likelihood ratio test using Wilks' theorem to construct χ² test
2. Parameter Estimation Accuracy: Asymptotic variance comparison
3. Pricing Performance: Log-likelihood comparison with Black-Scholes model

Comparison Methods

1. Baseline Model: Black-Scholes geometric Brownian motion model
2. Nested Testing: Variance-gamma model degenerates to Brownian motion as ν→0

Implementation Details

• Maximum likelihood estimation for all parameters
• Risk-neutral parameters re-estimated weekly
• Numerical methods used to compute Esscher transform parameter h*

Experimental Results

Main Results

1. Daily Return Fitting

Likelihood ratio test statistic: 2(logL_VG - logL_Gaussian) = 15.5445 p-value < 0.0001 under χ²₁ distribution, strongly rejecting the Gaussian hypothesis.

2. Option Pricing Performance

• 5% Significance Level: 48 out of 52 weeks (92.3%) reject Black-Scholes hypothesis
• 1% Significance Level: 45 out of 52 weeks (86.5%) reject Black-Scholes hypothesis

3. Parameter Estimation Efficiency

Monte Carlo simulations show:

• Under Laplace distribution, the variance of sample median is consistently smaller than sample mean
• Asymptotic variance of mean absolute deviation is 4/5 of scaled standard deviation

Ablation Studies

By setting ν = 0, the theoretical result that the variance-gamma process degenerates to Brownian motion is verified, confirming the nested nature of the model.

Case Analysis

Figure 7 shows differences between variance-gamma and Black-Scholes option prices:

• In-the-money/Out-of-the-money Options: VG prices significantly higher than BS prices
• At-the-money Options: BS prices higher than VG prices
• This pattern is consistent with the implied volatility smile phenomenon

Experimental Findings

1. Excess Kurtosis Explanation: VG model captures excess kurtosis observed in actual data (excess kurtosis = 3)
2. Implied Volatility Smile: BS implied volatility backed out from VG simulated prices exhibits smile shape
3. Risk-Neutral Drift: VG risk-neutral measure retains drift term, while BS model completely eliminates it

Related Work

Main Research Directions

1. Fat-Tail Distribution Modeling: Student-t distribution, Cauchy distribution as alternatives to normal distribution
2. Jump-Diffusion Models: Merton jump-diffusion model
3. Stochastic Volatility Models: Heston model, SABR model
4. Lévy Processes: Generalized hyperbolic distribution, normal inverse Gaussian distribution

Contributions of This Paper

Compared to existing work, this paper's advantages include:

• Provides a complete theoretical framework from statistical distribution to option pricing
• Offers intuitive economic interpretation through time subordination
• Establishes efficient parameter estimation methods

Conclusions and Discussion

Main Conclusions

1. Theoretical Contribution: Established complete theory of variance-gamma process as time-subordinated Brownian motion
2. Methodological Improvement: Proposed efficient parameter estimation under Laplace distribution
3. Empirical Validation: Significantly outperforms Black-Scholes model on S&P 500 data
4. Economic Interpretation: Time subordination provides new understanding of market microstructure

Limitations

1. Computational Complexity: Esscher transform parameters require numerical solution
2. Model Selection: Risk-neutral measure is not unique, requiring additional criteria
3. Parameter Stability: Requires frequent re-estimation of parameters
4. Scope of Application: Primarily applicable to European options; American option pricing is more complex

Future Directions

1. Extension to Other Derivatives: Asian options, barrier options, etc.
2. Multi-Asset Generalization: Joint modeling of correlated assets
3. Machine Learning Integration: Neural networks to accelerate parameter estimation
4. High-Frequency Data Application: Microstructure noise handling

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Establishes complete mathematical framework from probability theory foundations
2. Practical Value: Provides implementable option pricing formulas
3. Empirical Support: Large-sample data validates theoretical predictions
4. Clear Writing: Logical structure and detailed mathematical derivations

Weaknesses

1. Limited Novelty: Variance-gamma model already proposed by Madan et al.; this paper is primarily a systematic summary
2. Shallow Empirical Analysis: Lacks comparison with other advanced models
3. Insufficient Economic Intuition: Economic meaning of time subordination not deeply explained
4. Computational Efficiency: Insufficient discussion of computational challenges in practical implementation

Impact

1. Academic Value: Provides teaching material for application of Lévy processes in finance
2. Practical Guidance: Provides quantitative analysts with implementable modeling framework
3. Methodological Contribution: Parameter estimation methods have generalization value

Applicable Scenarios

1. European Option Pricing: Particularly suitable for short-term options
2. Risk Management: VaR calculation and stress testing
3. Quantitative Trading: Strategies based on statistical arbitrage
4. Academic Research: Serves as benchmark for more complex models

References

1 Madan, D. and Seneta, E. The variance gamma (v.g.) model for share market returns. The Journal of Business, 1990. 2 Black, F. and Scholes, M. The pricing of options and corporate liabilities. Journal of political economy, 1973. 6 Madan, D., Carr, P., and Chang, E. The variance gamma process and option pricing. European Finance Review, 1998.

Overall Assessment: This is a well-structured and theoretically sound financial mathematics paper. Although the core model is not original, the authors provide valuable reference material for variance-gamma model applications through systematic exposition and empirical validation. The paper's main contribution lies in establishing a complete chain from fundamental probability distributions to option pricing, with practical implementation guidance.