Title:Mean-Reverting SABR Models: Closed-form Implied Volatilities and Application to Stock Indices

Abstract:In this paper, we consider three stochastic-volatility models, each characterized by distinct dynamics of the instantaneous volatility: (1) a CIR process for squared volatility (i.e., the classical Heston model); (2) a mean-reverting lognormal process for volatility; and (3) a CIR process for volatility. Previous research has provided semi-analytical approximations for these models in the form of simple (non-mean-reverting) SABR models, each suitably parameterized for a given expiry.
First, using symbolic integration packages, we derive closed-form expressions for these semi-analytical approximations, under the assumption that all parameters remain constant (but without the constraint of constant expected volatility). Although the resulting formulae are considerably lengthier than those in simpler SABR models, they remain tractable and are easily implementable even in Excel.
Second, employing these closed-form expressions, we calibrate the three models to empirical volatility surfaces observed in EuroStoxx index options. The calibration is well-behaved and achieves excellent fits for the observed equity-volatility surfaces with only five parameters per surface. Consequently, these approximate models offer a simpler, faster, and (numerically) more reliable alternative to the classical Heston model or to more advanced models, which lack closed-form solutions and can be numerically challenging, particularly in less sophisticated implementation settings.
Third, we examine the stability and correlations of our parameter estimates. In this analysis, we identify certain issues with the models - one of which appears to stem from the sub-lognormal behavior of the actual equity-volatility process. Notably, the CIR-volatility model (3), as opposed to the CIR-variance Heston model (1), seems to best capture this behavior, and also results in more stable parameters.