OUOU Model for FX: Two-Factor Volatility

• The OUOU model is a two-factor stochastic volatility framework where both volatility factors follow Ornstein–Uhlenbeck dynamics, enhancing FX pricing precision.
• It extends the Schöbel–Zhu model by offering a semi-analytical Fourier pricing method that efficiently captures European currency option surface dynamics.
• Empirical calibration using methods like ICM demonstrates lower RMSE compared to Bates under Feller restrictions, though model interpretability and broader validation remain challenges.

Searching arXiv for the OUOU model and closely related stochastic-volatility papers for supporting context. {"query":"OUOU model Schobel-Zhu currency options stochastic volatility arXiv", "max_results": 10} {"query":"Schobel-Zhu stochastic volatility currency option pricing arXiv", "max_results": 10} {"query":"Heston Bates two-factor variance model FX options arXiv", "max_results": 10} The OUOU model is a two-factor stochastic volatility model for currency options introduced as an extension of the Schöbel–Zhu model, with both volatility factors following Ornstein–Uhlenbeck dynamics. In the dissertation "Analytic estimation of parameters of stochastic volatility diffusion models with exponential-affine characteristic function for currency option pricing" (Łabędzki, 16 Jul 2025), it is positioned as a tractable alternative to Heston- and Bates-type specifications: it is intended to represent FX implied-volatility-surface dynamics more flexibly than one-factor models while preserving an analytic characteristic function and, therefore, a semi-analytical valuation formula for European options.

1. Conceptual definition and motivation

The OUOU model is a two-factor volatility model in which the spot price depends on the sum of two stochastic volatility processes, each evolving as an OU process (Łabędzki, 16 Jul 2025). It is named OUOU because both volatility factors are Ornstein–Uhlenbeck processes. The dissertation introduces it in the context of currency option pricing, especially for the EURUSD market, and describes it as “an extension of the Schöbel-Zhu model and has a semi-analytical formula for the valuation of european options” (Łabędzki, 16 Jul 2025).

The model is motivated by three linked considerations. First, the dissertation studies extensions of the Schöbel–Zhu model that can better represent the dynamics of the FX implied volatility surface. Second, it seeks a two-factor alternative to the Heston/Bates family with an analytic characteristic function. Third, the empirical analysis reported in the dissertation indicates that one-factor models, and even four-factor statistical approximations, leave important structure unexplained, especially at long maturities and at the wings of the smile (Łabędzki, 16 Jul 2025).

A plausible implication is that the OUOU specification is intended to occupy a middle ground between richer empirical flexibility and Fourier-based tractability. That interpretation is consistent with the dissertation’s emphasis on both surface fit and semi-analytical pricing.

2. Stochastic specification

The dissertation first places the OUOU model inside a broader family of stochastic-volatility diffusions. Its general one-factor framework is

$dS_t = \mu S_t\,dt + \nu^r S_t^\beta\, dW_t^S,$

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$

with correlation

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$

The mean-reversion transform is defined by

$h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$

Within that framework, the OUOU model itself is introduced as follows (Łabędzki, 16 Jul 2025):

$$dS_t = (r_d-r_f)S_t\,dt + \nu_{1,t}S_t\,dW_{1,t} + \nu_{2,t}S_t\,dW_{2,t},$$

$$d\nu_{1,t} = \kappa_1(\theta_1-\nu_{1,t})\,dt + \omega_1\,dW_{1,t},$$

$$d\nu_{2,t} = \kappa_2(\theta_2-\nu_{2,t})\,dt + \omega_2\,dW_{2,t}.$$

The dissertation treats the total volatility effect as the sum of the two volatility factors. Its parameter interpretation is explicit: $(r_d-r_f)$ is the FX risk-neutral drift; $\kappa_1,\kappa_2$ are the mean-reversion speeds; $\theta_1,\theta_2$ are the long-run means; $$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$0 are the volatilities of the volatility factors; and $$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$1 are the correlations between each volatility shock and the corresponding price shock. It further states that the model has ten parameters in total, like the Bates two-factor model (Łabędzki, 16 Jul 2025).

This formulation makes the OUOU model a two-factor volatility specification rather than a two-factor variance specification. That distinction is central to its placement within the stochastic-volatility literature.

3. Replication, affine structure, and pricing

The dissertation derives replication dynamics for a general one-factor model and then extends them to the multi-factor setting. For the one-factor case, the replicating portfolio is

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$2

and the derived PDE is

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$3

For the exponential-affine class, the dissertation imposes the restriction

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$4

and often also

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$5

so that the PDE admits an affine characteristic-function solution (Łabędzki, 16 Jul 2025).

European call valuation is written in the standard form

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$6

with $$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$7 and $$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$8 computed from the characteristic function. The dissertation then presents model-specific characteristic-function forms. For Heston,

$$d\nu_t = \kappa \nu_t^q \big(\theta - h(\nu_t)\big)\,dt + \omega \nu_t^p\, dW_t^\nu,$$9

For Schöbel–Zhu,

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$0

with boundary conditions

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$1

For the OUOU model, the dissertation gives

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$2

with detailed coefficients derived in Appendix A (Łabędzki, 16 Jul 2025).

The practical pricing machinery is Fourier-based. The probabilities are computed as

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$3

and, in the single-characteristic-function form,

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$4

For numerical efficiency, the dissertation also uses the Attari representation,

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$5

with

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$6

The significance of these formulas is straightforward: the dissertation treats the OUOU model as analytically tractable in the characteristic-function sense, so European options can be priced semi-analytically rather than by brute-force simulation alone (Łabędzki, 16 Jul 2025).

4. Relation to Schöbel–Zhu, Heston, and Bates

The OUOU model is defined relationally in the dissertation: it is the two-factor volatility extension of Schöbel–Zhu, just as the Bates two-factor model is the two-factor variance extension of Heston (Łabędzki, 16 Jul 2025). The relevant model taxonomy can be summarized as follows.

Model	Factor structure	Dynamics emphasized
Heston	One-factor variance	Square-root/CIR dynamics
Schöbel–Zhu	One-factor volatility	OU dynamics
Bates two-factor model	Two-factor variance	Extension of Heston
OUOU	Two-factor volatility	Extension of Schöbel–Zhu

In the dissertation’s notation, the Heston model is recovered under

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$7

with dynamics

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$8

$$\langle dW_t^S, dW_t^\nu \rangle = \rho\,dt,\qquad r\in\left\{\tfrac12,1\right\}.$$9

Schöbel–Zhu replaces square-root variance dynamics by OU volatility dynamics,

$h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$0

The Bates two-factor model then extends Heston by adding a second variance factor, whereas OUOU performs the parallel extension on the Schöbel–Zhu side, keeping the volatility dynamics linear OU-type rather than CIR-type (Łabędzki, 16 Jul 2025).

The dissertation also discusses Scott/Wiggins, Stein–Stein, GARCH diffusion, SABR, Duffie–Pai–Singleton, and Cheng–Scaillet, positioning OUOU as a new member of the affine/semi-affine family with two stochastic volatility factors and an analytic characteristic function (Łabędzki, 16 Jul 2025). This suggests that its novelty lies less in the general idea of multifactor stochastic volatility than in the specific OU-type two-factor construction combined with tractable Fourier pricing.

5. Calibration and parameter estimation

Calibration occupies a substantial part of the dissertation. For the OUOU model, the stated approach is to infer starting values analytically and then refine them with local numerical optimization (Łabędzki, 16 Jul 2025). The dissertation emphasizes estimation of the volatility-of-volatility and correlation parameters and introduces the Implied Central Moments (ICM) method as a new way to estimate $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$1 and $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$2.

The ICM method uses option-implied moments derived from power payoff portfolios:

$h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$3

and matches them to model-implied moments. Approximate formulas are derived for Heston and Schöbel–Zhu and then adapted to two-factor models, including Bates and OUOU (Łabędzki, 16 Jul 2025).

The dissertation also reviews the Durrleman method, which estimates $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$4 and $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$5 from local smile slope and curvature using the short-maturity smile expansion

$h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$6

and summarizes the Gauthier-Rivaille “smart parameters” approximation based on a Taylor expansion of Heston prices in $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$7.

For OUOU specifically, the calibration strategy proceeds through symmetrical model calibration as a first step and then combines ICM-based starting points with the Modified Equal Variance Parametrisations method. The dissertation states that formulas are derived to split a one-factor estimate into two-factor starting values; an example given under symmetry is

$h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$8

The stated purpose is to obtain a good starting point for the 10-parameter numerical calibration of the OUOU model (Łabędzki, 16 Jul 2025).

This calibration architecture indicates that the dissertation does not treat tractability as sufficient by itself. Analytical starting values are used to stabilize a high-dimensional nonlinear calibration problem, which is a standard concern in multi-parameter stochastic-volatility models.

6. Empirical behavior, advantages, and limitations

The empirical study uses EURUSD OTC FX options from Reuters over 2010–2015, with 30 implied volatility series across 5 deltas and 6 maturities (Łabędzki, 16 Jul 2025). Within that setting, the dissertation reports that the OUOU model fits the EURUSD option surface better than the Bates two-factor variance model with the Feller condition, and that the advantage is especially visible when using ICM-based starting values.

The quantitative summary reported after full calibration is:

• OUOU/ICM/MEVP mean RMSE $h(\nu)= \begin{cases} \lambda(\nu-1), & \lambda>0,\[4pt] \log(\nu), & \lambda=0. \end{cases}$9
• BatesFeller/ICM/MEVP mean RMSE $$dS_t = (r_d-r_f)S_t\,dt + \nu_{1,t}S_t\,dW_{1,t} + \nu_{2,t}S_t\,dW_{2,t},$$0

The dissertation therefore concludes that OUOU fits better than the Bates-Feller model (Łabędzki, 16 Jul 2025).

At the same time, the comparison is carefully bounded. The dissertation also states that OUOU does not outperform the unconditioned Bates model in fit. It further notes that the unconditioned Bates model violates the Feller condition on 100% of days in the sample, which makes that specification problematic for simulation and market-dynamics analysis, whereas OUOU does not require a square-root positivity condition (Łabędzki, 16 Jul 2025). A common misconception would therefore be to interpret the empirical results as showing unconditional dominance over all two-factor alternatives; the dissertation does not support that stronger claim.

The advantages attributed to OUOU are explicit:

1. Two stochastic volatility factors
2. Analytic characteristic function
3. Better fit than the Bates model with Feller condition on EURUSD options
4. Compatibility with the ICM calibration framework
5. Usefulness for modeling currency option surfaces, especially when longer-maturity dynamics matter

Its limitations are equally explicit:

1. It remains a parametric model
2. It requires calibration and can suffer from local minima
3. It does not outperform the unconditioned Bates model in fit
4. Interpretation of the two-factor parameters is less transparent than in one-factor models
5. It is designed and tested specifically for EURUSD FX options, so broader market validation is still needed (Łabędzki, 16 Jul 2025)

Taken together, these results position the OUOU model as a tractable FX stochastic-volatility specification whose main contribution is not merely adding another factor, but doing so in a Schöbel–Zhu-type OU framework that retains semi-analytical pricing and integrates naturally with the dissertation’s analytical initialization methods.