The method with Markov-Switching and GARCH errors: 1. allows to characterize different types of fluctuation dynamics, along with a sequence of regimes. 2. could be suitable for 1-step ahead forecasts and outperforms Benchmark models with respect to various quality forecast criteria. Further work: 1. Since the process is bounded, we need to investigate on other types of distributions: censored or truncated normal, ... 2. Comparison of the sequence of regimes with meteorological time series over the same period.
展开▼