Modeling Distributional Time Series by Transformations
Zhicheng Chen · viXra · 2018
Probability distributions play a very important role in many applications. This paper describes a modeling approach for distributional time series. Probability density functions (PDFs) are approximated by real-valued vectors via successively applying the log-quantile-density (LQD) transformation and functional principal component analysis (FPCA); state-space models (SSMs) for real-valued time series are then applied to model the evolution of PCA scores, corresponding results are mapped back to the PDF space by the inverse LQD transformation.