Density-Based Inference of Continuous-Time Models of Equity Volatility and Short-Term Interest Rates
Gurdip Bakshi,Nengjiu Ju,Hui Ou-Yang
2004-01-01
Abstract:The treatment of this article renders closed-form density approximation feasible for univariate continuous-time models. Offering methodological convenience, the approximation method relies on numerically evaluating one-dimensional integrals and therefore circumvents existing dependence on intractable multidimensional integrals. Implementation steps depend directly on the parametric-form of drift and diffusion of the primitive process and not its transformation to a unit-variance process. Densitybased inferences can now be drawn for broader set of models of equity volatility and short-term interest rates. Our empirical results provide insights on crucial outstanding issues related to the ranking-ordering of continuous-time models, the absence/presence of non-linearities in the drift function, and the desirability of pursuing more flexible diffusion specifications. ∗Bakshi and Ju are both at Department of Finance, Smith School of Business, University of Maryland, College Park, MD 20742. Ou-Yang is at Faqua School of Business, Duke University, Durham, NC 27708. Bakshi can be reached at Tel: (301) 405-2261, email: gbakshi@rhsmith.umd.edu, Website: www.rhsmith.umd.edu/finance/gbakshi/; Ju at Tel: (301) 405-2934, email: nju@rhsmith.umd.edu, and Website: www.rhsmith.umd.edu/finance/nju/; Ou-Yang can be reached at Tel: (919) 660-3790, email: huiou@duke.edu, and Website: www.fuqua.duke.edu/faculty/alpha/huiou.htm. The authors gratefully acknowledge several useful discussions with Charles Cao, Steve Heston, Dilip Madan, and Matt Pritsker. Computer codes used to implement the continuous-time models are available from any of the authors. Looking back over the history of financial economics there is little doubt that continuoustime stochastic processes have played a prominent role in producing theoretically plausible and empirically testable models of portfolio selection, fixed income, equity derivatives, and expected stock returns. Subject of extensive research efforts, modeling the evolution of equity prices and spot interest rates as a one-dimensional process have brought enormous successes in understanding the market behavior of equity options, volatility, and yield curves. Whether the end-goal is martingale pricing or maximum-likelihood estimation, theory invariably requires the knowledge of the transition density of the economic forcing variable which is generally unamenable to closed-form characterization. In this sense the lack of analyticity of the density function has hampered empirical testing and the validation of alternative hypotheses about continuous-time models. To remedy this deficiency, Äıt-Sahalia (1999, 2002) has proposed a method to approximate the transition density in a one-dimensional diffusion setting. Given its potential utility to the researcher in applied and theoretical work, the thrust of this article is to expand on the analytical density approach of Äıt-Sahalia (1999, 2002) and our treatment renders the original method feasible for a substantially larger class of one-dimensional models. Armed with this approximation we empirically implement the method to study the plausibility of general models of equity volatility and spot interest rates. Density-based inferences allow us to disentangle issues connected with (i) the ranking-ordering of continuous-time models, (ii) the presence of nonlinearities in the drift function, and (iii) the desirability of adopting more flexible diffusion specifications. The motivation for our analysis to expand on Äıt-Sahalia (1999, 2002) derives from two considerations. First, in the context of one-dimensional diffusions dXt = μ[Xt] dt + σ[Xt] dWt, for economic-variable Xt, extant density approximations crucially hinge on transforming Xt to a unit-variance process via ∫X du σ[u] and then on inverting ∫X du σ[u] . This requirement has proved analytically challenging for some interesting economic models (see Bakshi and Ju (2002) for examples). Second, in the enhanced-method of Äıt-Sahalia (1999, 2002), the recursively defined coefficients that fulfill the forward/backward equation have a multidimensional integral dependence and are seldom tractable outside of constant elasticity of variance diffusion class. The framework of this paper overcomes both the hurdles associated with implementing Äıt-Sahalia (1999, 2002). More precisely, the contribution of our approach lies in determining the recursively defined expansion coefficients that exhibit