Heston Model Example

Updated 24 April 2026

Heston Model is a stochastic volatility framework defined by semi-closed-form solutions and robust calibration techniques.
It employs Fourier methods, maximum likelihood estimation, and adaptive simulation for efficient option pricing.
Recent advancements include joint market data calibration and extensions such as Stationary Heston and Stochastic Vol-of-Vol for improved smile reproduction.

The Heston model is a stochastic volatility framework widely employed in quantitative finance for pricing options and calibrating volatility surfaces. Core to its appeal are semi-closed-form solutions for European claims, robust calibration methodologies, tractable simulation algorithms, and extensions enabling joint calibration and improved volatility smile reproduction. This article provides a detailed exposition of the Heston model through worked examples, highlighting practical implementation workflows, calibration, numerical schemes, and recent advancements.

1. Core Heston Model and Semi-Closed-Form Pricing

Under the risk-neutral measure $\mathbb{Q}$ , the dynamics of the spot price $S_t$ and its instantaneous variance $V_t$ are governed by: $\begin{cases} dS_t = r S_t dt + \sqrt{V_t} S_t dW_{1,t}, \ dV_t = \kappa (\theta - V_t) dt + \eta \sqrt{V_t} dW_{2,t}, \ d\langle W_1, W_2 \rangle_t = \rho\, dt, \end{cases}$ where $r$ is the risk-free rate, $\kappa$ the variance mean-reversion speed, $\theta$ the long-run variance, $\eta$ the volatility of volatility, and $\rho$ the spot-variance correlation (Boukai, 2021).

For a European call, the model admits a semi-closed-form price using characteristic functions: $C(S, V, t; K, T) = S\, P_1 - K e^{-r(T-t)}\, P_2,$ where the probabilities $S_t$ 0 are given by one-dimensional Fourier inversions of exponential-affine transforms of the log-moneyness (Boukai, 2021). This tractability makes the Heston model particularly suitable for calibration to market data and fast computation of vanilla prices.

2. Maximum Likelihood Estimation and Empirical Calibration

Parameter estimation for the Heston SDEs can be performed using explicit closed-form maximum likelihood estimators (MLEs) based on discretized likelihoods. The calibrated parameters capture mean reversion ( $S_t$ 1), long-run variance ( $S_t$ 2), volatility-of-volatility ( $S_t$ 3), and correlation ( $S_t$ 4). For observations $S_t$ 5 at discretized times: $S_t$ 6 where $S_t$ 7 are computed from observed variance time series (Azencott et al., 2014).

Applications to S&P 500 daily data or minute-level equity data yield parameter estimates that are robust under large-sample regimes, provided the canonical condition $S_t$ 8 (for Gaussian asymptotics) is satisfied.

3. Risk-Neutral Densities and Explicit Approximations

The risk-neutral density (RND) for $S_t$ 9 implied by the Heston model falls within a scale-family with scale parameter equal to the forward price. Five explicit one-parameter RND families can be used for efficient approximation:

Family	RND Parameterization	Calibration Note
Log-Normal	variance $V_t$ 0	Good fit, matches positive skew
Gamma	$V_t$ 1	Slightly inferior to Log-Normal/IG
Inverse-Gaussian	mean 1, var $V_t$ 2	Comparable to Log-Normal with positive skew
Weibull	( $V_t$ 3)	Generally less accurate
Inverse-Weibull	( $V_t$ 4)	Good for positive skew

Numerical calibration on real equity options (e.g., AMD) demonstrates that Log-Normal and Inverse-Gaussian RNDs achieve mean squared errors near those of full Heston, and deliver significant computational savings (Boukai, 2021).

4. Advanced Calibration: Joint Equity and Volatility Index Fitting

For robust parameter inference, especially the vol-of-vol, practitioners increasingly incorporate volatility index (VIX/VVIX) data. Four approaches for calibrating the vol-of-vol ( $V_t$ 5) are demonstrated:

Closed-form: Based on variance moments; fast and accurate within 5% of full PDE solution.
Log-contract and replication integrals: Use transition densities for exact or approximate estimation.
PDE-based double-replication: Numerically solve the full Heston PDE for forward-starting option portfolios, extracting VVIX by discrete replication.

Calibration to both SPX and VVIX data stabilizes $V_t$ 6 estimates, with values for $V_t$ 7 remaining within a narrow band across methodologies (0.33–0.37), in contrast to the broad variability under SPX-only calibration (Healy, 22 Dec 2025, Fouque et al., 2017).

5. Efficient Simulation and Numerical Schemes

Simulation of Heston paths for pricing or calibration relies on accurate schemes due to the nonlinearity of the variance process:

Adaptive Simulation: The “ADAPT” framework recasts the CIR variance in terms of a bridge of squared-Bessel processes, adaptively refining quadrature intervals to meet a prescribed variance tolerance while maintaining positivity (Iscoe et al., 2011).
Hybrid Tree-Finite Difference: Combines a recombining Markov chain for $V_t$ 8 with 1D finite-difference solves for the log-price, allowing fast, stable weak convergence for European and American options (Briani et al., 2013).
Artificial Boundary Methods: Enhanced boundary condition schemes (ApABC, MApABC1/2) facilitate high-accuracy PDE solutions on truncated domains, significantly reducing boundary-induced bias (<0.5% relative error versus standard Heston BCs) (Li et al., 2019).

Table: Numerical accuracy with various boundary conditions (Li et al., 2019)

Boundary Condition	Relative Error (%)
Heston Standard	2.5
MApABC1	0.6
MApABC2	0.3

6. Extensions: Stationary Heston and Multiscale Vol-of-Vol

Two notable Heston extensions improve calibration to short-term smile/skew or enable joint VIX/SPX fitting:

Stationary Heston: Samples the initial variance from the model's invariant Gamma law, correcting for insufficient steepness of short-maturity implied volatility skews, and is calibrated on the same market data as the standard model, yielding significantly improved near-expiry fits (Lemaire et al., 2020).
Heston Stochastic Vol-of-Vol (SVV): Introduces fast and slow stochastic factors into $V_t$ 9. The first-order perturbative pricing expansion reduces to Heston’s formula plus Fourier-corrective terms constructed from ODE systems in the Fourier variable. Joint calibration to SPX and VIX option data yields MSE reductions of ~50% compared to classic Heston, at minimal incremental computational cost (Fouque et al., 2017).

7. Practical Implementation and Robustness

Robust calibration approaches include:

Space Mapping: Use a PDE-calibrated surrogate to accelerate the calibration of SDE-based fine models (e.g., Asian options), with convergence to within 90% of residual error in <4 outer iterations (Clevenhaus et al., 24 Jan 2025).
Iterative Splitting for PDEs: Decompose the option pricing PDE into tractable Black-Scholes and correction problems, solved with rapid 1D tridiagonal solvers, yielding reduced computational complexity and improved accuracy of Greeks (Li et al., 2020).

Implementation in MATLAB/Python is facilitated by direct translation of the PDE discretization, ODE solvers for Fourier corrections, and adjoint-based gradient descent for PDE-constrained calibration.

The Heston model ecosystem thus supports semi-closed-form vanilla pricing, stable calibration to rich data sources, robust simulation schemes, and further admits powerful extensions for joint smile/skew and volatility-index calibration, making it a cornerstone of contemporary quantitative finance practice (Boukai, 2021, Fouque et al., 2017, Healy, 22 Dec 2025, Briani et al., 2013).