Efficient simulation of prices for European call options under Heston stochastic-local volatility model: a comparison of methods

Basic Information

• Paper ID: 2509.24449
• Title: Efficient simulation of prices for European call options under Heston stochastic-local volatility model: a comparison of methods
• Authors: Meng Cai, Tianze Li
• Affiliations: School of Statistics and Mathematics, School of Economics, Central University of Finance and Economics
• Classification: q-fin.CP (Computational Finance)
• Publication Date: October 16, 2025
• Paper Link: https://arxiv.org/abs/2509.24449

Abstract

This paper investigates the Heston stochastic-local volatility (HSLV) model, which comprises an asset price process and a Cox-Ingersoll-Ross (CIR) type variance process, widely applied in the financial industry. In contrast to the weakly convergent non-central chi-squared approximation employed in reference 25, this paper considers two methods that achieve strong convergence and positivity preservation for the CIR process under the Lamperti transformation: the truncated Euler method and the backward Euler method. The explicit truncated Euler method is computationally efficient and remains robust under high volatility, while the implicit backward Euler method provides high computational accuracy and stable performance. Numerical experiments on European call options demonstrate the advantages of different methods.

Research Background and Motivation

Problem Background

1. Model Requirements: Financial engineering requires balancing perfect calibration with correct dynamics; the stochastic-local volatility (SLV) model provides a solution to this trade-off
2. Limitations of Existing Models:
   • Pure stochastic volatility models have limited parameters and struggle to perfectly fit implied volatilities across all maturities and strike prices
   • Local volatility models, while capable of calibrating all current market options, have issues predicting future volatility behavior, resulting in forward volatility smiles that tend to flatten

Core Challenges

Numerical simulation of the CIR process is a critical bottleneck in model evaluation because:

• The diffusion term is unbounded and coefficients are not globally Lipschitz continuous
• Traditional Euler-Maruyama methods may produce negative variances, which are meaningless both mathematically and financially
• Existing non-central chi-squared approximations only achieve weak convergence

Research Motivation

Develop strongly convergent methods with positivity preservation to ensure that trajectories of exact and approximate solutions remain close to each other, thereby improving overall option pricing accuracy.

Core Contributions

1. Proposes two novel numerical methods: The truncated Euler method and backward Euler method, both possessing strong convergence and positivity preservation properties
2. Theoretical Analysis: Proves strong convergence of the truncated Euler method (convergence order 1/2)
3. Comprehensive Numerical Comparison: Systematically compares performance of different methods through European call option pricing experiments
4. Practical Guidance: Provides method selection recommendations for different application scenarios

Methodology Details

Problem Formulation

Solve the European call option pricing problem under the Heston stochastic-local volatility model, where the core task is to efficiently and accurately simulate the CIR variance process.

Model Architecture

HSLV Model Definition

$\begin{cases} dS_t = rS_t dt + \sqrt{V_t} S_t \sigma(t, S_t)\left(\rho dW_t + \sqrt{1-\rho^2} d\tilde{W}_t\right) \\ dV_t = \kappa(\theta - V_t)dt + \gamma\sqrt{V_t} dW_t \\ dW_t d\tilde{W}_t = \rho dt \end{cases}$

Where:

• $S_t$: Asset price process
• $V_t$: Variance process (CIR dynamics)
• $\sigma(t,s)$: Local volatility function
• $\kappa, \theta, \gamma$: CIR process parameters

Lamperti Transformation

To overcome numerical difficulties of the CIR process, introduce the transformation $L_t = \sqrt{V_t}$: $dL_t = \frac{1}{2}\kappa\left(\frac{\theta}{L_t} - L_t\right)dt + \frac{1}{2}\gamma dW_t$

Core Numerical Methods

1. Truncated Euler Method

Define truncation mapping: $\pi_\tau(x) = (b\tau^{1/4}) \vee x$

Numerical scheme: $L^N_{t_{n+1}} = L^N_{t_n} + \frac{\kappa}{2}\tau\left(\frac{\theta}{\pi_\tau(L^N_{t_n})} - \pi_\tau(L^N_{t_n})\right) + \frac{1}{2}\gamma\Delta W_n$

Characteristics:

• Explicit scheme with high computational efficiency
• Guarantees positivity through truncation
• Maintains robustness under high volatility

2. Backward Euler Method

$L^N_{t_{n+1}} = L^N_{t_n} + \frac{\kappa}{2}\tau\left(\frac{\theta}{L^N_{t_{n+1}}} - L^N_{t_{n+1}}\right) + \frac{1}{2}\gamma\Delta W_n$

Characteristics:

• Implicit scheme requiring iterative solution
• Provides high computational accuracy
• Stable performance but with higher computational cost

Technical Innovations

1. Strong Convergence: Unlike weakly convergent non-central chi-squared methods, ensures trajectory-level convergence
2. Positivity Preservation Design: Ensures variance remains positive through mathematical transformation and truncation techniques
3. Method Complementarity: Two methods offer complementary advantages in efficiency and accuracy, suitable for different scenarios

Experimental Setup

Model Parameters

• Benchmark Heston Parameters: $\gamma = 0.95$, $\kappa = 1.05$, $\rho = -0.315$, $\theta = 0.0855$, $S_0 = 1$, $r = 0$, $v_0 = 0.0945$, $T = 5$
• HSLV Adjustment Parameters: Introduce adjustment parameter $p = 0.25$ to modify various parameters

Evaluation Metrics

• Relative error percentage
• Computational time
• Robustness under different market conditions

Comparison Methods

• Standard Euler method
• AES method (Andersen Exact Simulation)
• Truncated Euler method
• Backward Euler method

Experimental Design

1. Conditional Expectation Calculation: Use 2D-COS method as benchmark with 20 bins, 10,000 paths, and step size 0.001
2. Parameter Sensitivity Analysis: Test performance under different $\theta$ and $p$ values
3. Strike Price Comparison: Three cases with $K = 70\%, 100\%, 150\%$

Experimental Results

Main Results

Error Comparison Under Different Step Numbers

Computational Time Comparison (40 steps)

• AES method: 0.1114 seconds
• Euler method: 0.0939 seconds
• Truncated Euler method: 0.0733 seconds
• Backward Euler method: 137.9476 seconds

Key Findings

1. With Fewer Steps: Truncated Euler method typically exhibits the lowest error
2. With More Steps: Backward Euler method achieves the highest accuracy
3. Parameter Robustness: Truncated Euler method shows excellent robustness with slow error growth as $\theta$ increases
4. Computational Efficiency: Truncated Euler method is fastest; backward Euler method is significantly more time-consuming due to iterative solution requirements

Ablation Studies

Impact of Volatility Parameter $\theta$

Experiments show that as $\theta$ increases (simulating extreme market conditions):

• AES and standard Euler methods show significantly increased errors
• Truncated Euler method maintains robustness with lower error growth rates
• This indicates truncation methods are more suitable for stress testing and tail risk modeling

Impact of Adjustment Parameter $p$

As $p$ increases, the backward Euler method consistently performs best, but shows higher sensitivity to step size.

Related Work

Main Research Directions

1. SLV Model Calibration: Pioneering work by Guyon and Henry-Labordère (2011)
2. CIR Process Numerical Methods: Including finite differences, Monte Carlo particle methods, etc.
3. Deep Learning Applications: Cuchiero et al. (2020) using generative adversarial networks for calibration

Advantages of This Work

Compared to existing work, this paper provides strongly convergent positivity-preserving methods, filling the gap left by weakly convergent methods and offering more reliable numerical tools for practical applications.

Conclusions and Discussion

Main Conclusions

1. Method Complementarity: Truncated Euler method is suitable for large-scale simulations and real-time risk management; backward Euler method is suitable for stress testing and high-precision scenarios
2. Robustness Advantages: Truncated Euler method demonstrates excellent performance under extreme market conditions
3. Accuracy-Efficiency Trade-off: Appropriate step size and method selection should be based on specific application requirements

Limitations

1. High computational cost of backward Euler method
2. Step size selection requires balancing accuracy and efficiency
3. Method stability for certain extreme parameter combinations requires further verification

Future Directions

1. Develop adaptive step-size strategies
2. Explore parallel implementations
3. Extend to more complex SLV model variants

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Provides complete convergence proofs
2. High Practical Value: Both methods offer advantages, providing flexible choices for practical applications
3. Comprehensive Experiments: Systematically compares different methods under various conditions
4. Clear Presentation: Accurate mathematical notation and clear logical structure

Weaknesses

1. Theoretical Depth: Relatively limited theoretical analysis of the backward Euler method
2. Real Market Validation: Lacks validation with real market data
3. Computational Complexity Analysis: Lacks detailed theoretical complexity analysis

Impact

1. Academic Contribution: Provides new strongly convergent methods for CIR process numerical simulation
2. Practical Value: Directly applicable to option pricing and risk management in financial institutions
3. Reproducibility: Detailed method descriptions facilitate implementation

Application Scenarios

1. Quantitative Trading: High-frequency option pricing and hedging strategies
2. Risk Management: Stress testing and tail risk assessment
3. Model Calibration: SLV model parameter estimation and validation

References

The paper cites 25 relevant references covering key works in SLV model theory, CIR process numerical methods, and option pricing, providing a solid theoretical foundation for the research.