A0450
Title: A cylindrical hidden Markov model based on skewed circular distributions
Authors: Yoichi Miyata - Takasaki City University of Economics (Japan) [presenting]
Takayuki Shiohama - Nanzan University (Japan)
Toshihiro Abe - Hosei University (Japan)
Abstract: Hidden Markov models are known to be a useful method for estimating the timing of structural change and the structure of each population for time series data. Cylindrical hidden Markov models are applied when data consist of pairs of non-negative values and values on the circle. If angular data show somewhat asymmetric patterns, it is required to model a circular marginal distribution to have a slightly more strongly skewed shape. A hidden Markov model whose components are cylindrical distributions in which a circular random variable (representing the angle) has an extended sine-skewed circular distribution. In particular, some conditions are clarified for the consistency of the maximum likelihood estimator and present numerical examples of the model applied to real data.
