Mean-Reverting SABR Models: Closed-form Surfaces and Calibration for Equities

Basic Information

• Paper ID: 2502.07766
• Title: Mean-Reverting SABR Models: Closed-form Surfaces and Calibration for Equities
• Author: Vlad Perederiy (Perederiy Consulting)
• Classification: q-fin.PR (Quantitative Finance - Pricing of Securities)
• Publication Date: December 2024, revised in October 2025
• Paper Link: https://arxiv.org/abs/2502.07766

Abstract

This paper investigates three stochastic volatility models, each with distinct instantaneous volatility dynamics: (1) CIR process for variance (classical Heston model); (2) mean-reverting lognormal process for volatility; (3) CIR process for volatility. The author first derives semi-analytical closed-form expressions using computer algebra systems, then calibrates these three models to empirical implied volatility surfaces of EuroStoxx index options, and finally analyzes the stability and correlation of parameter estimates. The study finds that the CIR-volatility model performs best in capturing the sub-lognormal behavior of equity volatility processes.

Research Background and Motivation

Problem Definition

Stock and equity index returns exhibit significant volatility clustering and mean-reversion characteristics, particularly a strong negative correlation between return changes and volatility changes. These features manifest in equity derivatives as:

1. Term Structure Effects: During high volatility periods, implied volatility of short-term ATM options typically significantly exceeds that of long-term options
2. Volatility Skew: Due to negative correlation, volatility smiles typically exhibit skewness around forward prices
3. Mean-Reversion Effects: Long-term option volatility smiles tend to flatten rapidly

Limitations of Existing Methods

1. Heston Model: While capturing key characteristics of equity markets, it lacks closed-form solutions and requires complex numerical integration, prone to numerical issues
2. Standard SABR Model: Possesses closed-form solutions but insufficiently reflects term structure and mean-reversion characteristics of equity markets
3. Simple Interpolation Methods: Less effective for equity markets compared to foreign exchange markets

Research Motivation

This paper aims to develop models that capture the main characteristics of equity markets while possessing closed-form solutions, providing simpler, faster, and numerically stable alternatives for practical applications.

Core Contributions

1. Derivation of Closed-form Expressions: First-time derivation of complete closed-form expressions for three mean-reverting SABR models (hSABR, mrSABR, CIR-ZABR) without constant expected volatility constraints
2. Empirical Calibration Verification: Validates model performance using EuroStoxx index option data, achieving excellent fitting with only 5 parameters
3. Parameter Stability Analysis: In-depth analysis of parameter estimate correlations and stability, revealing that CIR-ZABR model exhibits optimal parameter stability
4. Enhanced Practicality: Provides formulas implementable in Excel, significantly improving model accessibility and usability

Methodology Details

Task Definition

The research objective is to develop stochastic volatility models with closed-form solutions for equity option pricing that can:

• Input: 5 model parameters (α, θ, λ, ρ, ν) and option characteristics (strike price, time to maturity)
• Output: Black implied volatility
• Constraints: Maintain numerical stability and computational efficiency

Three Model Architectures

1. Heston-SABR (hSABR)

SABR approximation based on the classical Heston model:

dF = F√V dW₁
dV = λ(θ² - V)dt + ν√V dW₂

where V is variance following a CIR process.

2. Mean-Reverting SABR (mrSABR)

Mean-reversion incorporated directly into the SABR framework:

dF = F^β A dW₁
dA = λ(θ - A)dt + νA dW₂

where A is volatility following a mean-reverting lognormal process.

3. CIR-ZABR

Generalized model with volatility following a CIR process:

dF = F^β A dW₁
dA = λ(θ - A)dt + νA^γ dW₂

When γ = 0.5, volatility follows a CIR process.

Technical Innovations

1. Closed-form Solution Derivation

Using symbolic computation systems (Python sympy) to derive complex polynomial expressions:

• Each effective coefficient τₑₓ, Gᵢₙₜ, b̄, c̄ can be expressed as polynomial ratios of 7 variables (α, θ, λ, ρ, ν, Tₑₓ, E=e^λTₑₓ)
• Polynomials reach up to 8th degree, expanding to over 200 terms

2. Binomial Expansion Approximation

For the CIR-ZABR model, employing reparameterized binomial expansion:

α(x) = μ + μδ(2e^(-λx) - 1)

where μ = (α+θ)/2, δ = (α-θ)/(α+θ), ensuring convergence condition |y| < 1 is always satisfied.

3. Numerical Stability Improvements

• Eliminates complex logarithm branch issues in the Heston model
• Provides Excel-implementable formula formats
• Removes dependence on numerical integration

Experimental Setup

Dataset

• Data Source: Bloomberg EuroStoxx index options
• Time Range: Monthly data from 2021-2024
• Data Scale: 15 implied volatility quotes per volatility surface (5 strikes × 3 maturities)
• Calibration Method: Excel Solver minimizing RMSE

Evaluation Metrics

• Primary Metric: Root Mean Square Error (RMSE), measured in implied volatility units
• Goodness of Fit: Explained variance ratio (R²)
• Parameter Stability: Spearman correlation coefficients, standard deviation/mean ratios

Comparison Methods

• Cross-model comparison among three models
• Comparison with constrained versions (α = θ)
• Performance comparison between fixed and re-estimated parameters

Experimental Results

Main Results

Fitting Performance

All models achieve excellent fitting with only 5 parameters, significantly outperforming constrained versions with constant expected volatility (RMSE increase of approximately 0.5%).

Parameter Stability Analysis

CIR-ZABR Model Performs Best:

• Correlation between α and ν approaches 0 (-0.05), compared to 0.45 for hSABR and -0.83 for mrSABR
• Relatively low parameter coefficients of variation
• Highest satisfaction rate of non-degeneracy conditions

Parameter Correlation Findings

Common Patterns

1. Positive Correlation between θₜ and αₜ: Long-term volatility positively correlates with initial volatility
2. Negative Correlation between λₜ and θₜ, αₜ: Mean-reversion speed is slower during high volatility periods
3. Positive Correlation between λₜ and νₜ: Reflects moderate redundancy among model parameters

Model Differences

• mrSABR: Strong negative correlation between α and ν (-0.83), indicating over-compensation in lognormal diffusion
• hSABR: Positive correlation between α and ν (0.45), reflecting insufficiency in normal diffusion
• CIR-ZABR: Weakest correlation between α and ν, suggesting CIR diffusion better approximates the true process

Non-degeneracy Condition Verification

• hSABR: Feller condition (2λθ² > ν²) violated across all surfaces
• mrSABR: Non-degeneracy condition (λ > 0.5ν²) satisfied in all estimates
• CIR-ZABR: Corresponding condition (2λθ > ν²) satisfied in most cases

Related Work

SABR Model Development

• Hagan et al. (2002): Original SABR model and closed-form approximations
• Hagan et al. (2016): Improved universal smile formula
• Hagan et al. (2018, 2020): Semi-analytical solutions for hSABR and mrSABR

Stochastic Volatility Models

• Heston (1993): Classical stochastic volatility model
• Felpel et al. (2020): Effective stochastic volatility approach for ZABR models

Contribution of This Paper

This paper provides the first complete closed-form solutions without constraints and reveals through empirical analysis how different diffusion assumptions affect parameter stability.

Conclusions and Discussion

Main Conclusions

1. Technical Feasibility: All three mean-reverting SABR models provide excellent volatility surface fitting
2. Practical Advantages: Closed-form solutions enable easy implementation in simple environments like Excel
3. Diffusion Process Insights: CIR volatility diffusion is more suitable for equity markets than lognormal or normal diffusion

Limitations

1. Formula Complexity: Although closed-form, expressions are quite lengthy
2. Parameter Correlation: Significant parameter correlation issues exist in models other than CIR-ZABR
3. Approximation Accuracy: CIR-ZABR requires binomial expansion approximation, potentially affecting accuracy

Future Directions

1. Nonlinear Mean-Reversion: Parameter estimates suggest markets may exhibit nonlinear mean-reversion patterns
2. Stochastic Long-term Volatility: Consider time-varying long-term volatility parameters
3. Multi-asset Extensions: Extend models to multi-asset portfolio volatility surface modeling

In-depth Evaluation

Strengths

1. Outstanding Theoretical Contribution: First provision of unconstrained closed-form solutions, filling an important theoretical gap
2. High Practical Value: Excel implementation significantly lowers usage barriers
3. In-depth Empirical Analysis: Parameter correlation analysis reveals essential characteristics of volatility processes
4. Strong Numerical Stability: Avoids numerical integration issues in the Heston model

Weaknesses

1. Poor Formula Readability: Closed-form expressions are overly complex, affecting theoretical understanding
2. Sample Limitations: Uses only EuroStoxx data, lacking cross-market validation
3. Insufficient Approximation Error Analysis: Binomial expansion error in CIR-ZABR not fully quantified
4. Computational Complexity: While avoiding numerical integration, polynomial calculations remain complex

Impact

1. Academic Value: Provides new technical pathways for stochastic volatility modeling
2. Practical Significance: Offers financial institutions more practical volatility surface modeling tools
3. Methodological Contribution: Successful application case of symbolic computation in financial modeling

Applicable Scenarios

1. Risk Management: Suitable for risk management systems requiring fast, stable volatility surface estimation
2. Option Pricing: Particularly suitable for computationally constrained environments
3. Monte Carlo Simulation: Provides efficient volatility models for large-scale simulations
4. Teaching and Research: Excel implementation makes it an ideal teaching tool

References

• Heston, S.L. (1993). A closed-form solution for options with stochastic volatility with applications to bond and currency options. Review of Financial Studies, 6(2): 327-343
• Hagan, P.S. et al. (2002). Managing Smile Risk. Wilmott Magazine, September 2002: 84-108
• Hagan, P.S. et al. (2018). Implied Volatility Formulas for Heston Models. Wilmott Magazine, November 2018: 44-57
• Felpel, M. et al. (2020). Effective Stochastic Volatility: Applications to ZABR-type Models. Quantitative Finance, 21(5): 837-852