Browsing by Author "Liu, Peng"

Filter results by typing the first few letters
Now showing 1 - 2 of 2
  • No Thumbnail Available
    Compactly Supported Radial Basis Function Kernels
    (North Carolina State University. Dept. of Statistics, 2004) Zhang, Hao Helen; Genton, Mark G.; Liu, Peng
  • No Thumbnail Available
    A Stochastic Volatility Model and Inference for the Term Structure of Interest
    (2007-04-25) Liu, Peng; A. Ronald Gallant, Committee Member; Denis Pelletier, Committee Member; William H. Swallow, Committee Member; Peter Bloomfield, Committee Chair; David Dickey, Committee Member
    This thesis builds a stochastic volatility model for the term structure of interest rates, which is also known as the dynamics of the yield curve. The main purpose of the model is to propose a parsimonious and plausible approach to capture some characteristics that conform to some empirical evidences and conventions. Eventually, the development reaches a class of multivariate stochastic volatility models, which is flexible, extensible, providing the existence of an inexpensive inference approach. The thesis points out some inconsistency among conventions and practice. First, yield curves and its related curves are conventionally smooth. But in the literature that these curves are modeled as random functions, the co-movement of points on the curve are usually assumed to be governed by some covariance structures that do not generate smooth random curves. Second, it is commonly agreed that the constant volatility is not a sound assumption, but stochastic volatilities have not been commonly considered in related studies. Regarding the above problems, we propose a multiplicative factor stochastic volatility model, which has a relatively simple structure. Though it is apparently simple, the inference is not, because of the presence of stochastic volatilities. We first study the sequential-Monte-Carlo-based maximum likelihood approach, which extends the perspectives of Gaussian linear state-space modeling. We propose a systematic procedure that guides the inference based on this approach. In addition, we also propose a saddlepoint approximation approach, which integrates out states. Then the state propagates by an exact Gaussian approximation. The approximation works reasonably well for univariate models. Moreover, it works even better for the multivariate model that we propose. Because we can enjoy the asymptotic property of the saddlepoint approximation.