关键词: SCD模型, 持续期, 伪似然估计, MCMC估计, 核估计

Abstract: The SCD model can effectively describe the changes of the durations in the ultra-high time series, but it is based on the assumption that the financial durations are generated by a fixed mechanism, and the estimation of the parameters is not easy. Combing with the kernel estimation, this paper proposes the nonparametric SCD model and its iteration algorithm neither relies on the form of the conditional mean of the duration nor the distribution of the error term. Then, based on the simulated data generated by the TEACD(1,1) model, the non-parametric SCD model is compared with the parametric SCD model which is estimated by the quasi-maximum likelihood with the using of the Kalman filter and by the Markov chain Monte Carlo with the using of the Gibbs sampling, proving that the non-parametric SCD model performs as well as the parametric SCD model by the MCMC estimation and yields better fitted effects than parametric SCD model by the QML estimation with large samples and proposes a valuable preliminary suggestion on the choice of the parametric specification.

Key words: SCD Model, Duration, QML Estimation, MCMC Estimation, Kernel Estimation