State Space Approach to Adaptive Fuzzy Modeling: Application to Financial Investment

Abstract

This paper proposes a new state space approach to adaptive fuzzy modeling under the dynamically changing environment, where Bayesian filtering sequentially learns parameters including model structures as state variables. Moreover with a particle filtering algorithm, our approach is widely applicable to the machine learning for real-time observation data flows. To show the effectiveness of our framework, a Takagi-Sugeno-Kang fuzzy model is concretely designed for financial portfolio construction based on a benchmark return, that is stock market index (e.g. S\&P 500 index) return with non-negative lower bound, and successfully attains fine risk-return profiles. An out-of-sample simulation with our proposed portfolio construction demonstrates the validity of our framework.