Discussion Papers 2020

CIRJE-F-1142	"A New Interval Type-2 Fuzzy Logic System under Dynamic Environment: Application to Financial Investment"
Author Name	Takahashi, Akihiko and Soichiro Takahashi
Date	February 2020
Full Paper
Remarks	Published in Engineering Applications of Artificial Intelligence, Volume 100, April 2021.

Abstract
This paper proposes a new interval type-2 fuzzy logic system (IT2 FLS) for financial investment with time-varying parameters adaptive to real-time data streams by using an on-line learning method based on a state-space framework. Particularly, our state-space approach regards parameters of IT2 FLSs as state variables to sequentially learn by Bayesian filtering algorithms under dynamic environment, where time-series data are continuously observed with occasional structural changes. Moreover, our proposal is helpful for financial applications, which often involve various practical complex constraints, because general state-space model makes it possible to flexibly deal with non-linearities. In our empirical experiment with time-series data of financial assets, our approach is applied to on-line parameter learning of type-1 and type-2 FLSs for portfolio decision making. As a result, it is shown that the IT2 FLS holds its advantage against the type-1 FLS, even though the both type-1 and type-2 models have the adaptive time-varying parameters, which is an unexplored topic for empirical studies of this area.

Abstract

This paper proposes a new interval type-2 fuzzy logic system (IT2 FLS) for financial investment with time-varying parameters adaptive to real-time data streams by using an on-line learning method based on a state-space framework. Particularly, our state-space approach regards parameters of IT2 FLSs as state variables to sequentially learn by Bayesian filtering algorithms under dynamic environment, where time-series data are continuously observed with occasional structural changes. Moreover, our proposal is helpful for financial applications, which often involve various practical complex constraints, because general state-space model makes it possible to flexibly deal with non-linearities. In our empirical experiment with time-series data of financial assets, our approach is applied to on-line parameter learning of type-1 and type-2 FLSs for portfolio decision making. As a result, it is shown that the IT2 FLS holds its advantage against the type-1 FLS, even though the both type-1 and type-2 models have the adaptive time-varying parameters, which is an unexplored topic for empirical studies of this area.