A0824
Title: Financial time series analysis with weighted quantile approach
Authors: Tomas Tichy - VSB-TU Ostrava (Czech Republic) [presenting]
Michal Holcapek - University of Ostrava (Czech Republic)
David Nedela - VSB - Technical University of Ostrava (Czech Republic)
Abstract: A rather complex sub-task of the research is to design inference mechanisms in a fuzzy-probability environment that will be well justified and interpreted because probability and fuzziness are different measures, and care must be taken when combining them. Inference mechanisms proposed in previous works are analyzed and integrated into the new probabilistic fuzzy models. They are also analyzed and extended to other inference mechanisms designed for fuzzy systems in fuzzy-probabilistic settings. In particular, in this contribution, weighted quantiles and the introduction of a very simple and fast method for their determination are studied. This method is implemented in an algorithm for probabilistic-fuzzy inference systems that allows the deriving of weighted quantiles for any element or interval in a given input domain for a fixed probability, for example, a median or median function, if the probability is equal to 0.5. In addition, a quantile function can be determined for any fixed element in a given domain. Subsequently, several examples of time series are investigated, and statistical properties of proposed probabilistic-fuzzy models are discussed and compared to recently popular models in financial time series analysis and forecasting.
