A new discrete fractional AMAR model for finance time series forecasting by machine learning

Abstract

This study analyzes and addresses the modeling problem of short-term dependent time series. Firstly, discrete fractional calculus is proposed to enhance the performance of the classical model. A fractional Autoregressive Moving Average model is proposed. Then, the neural network is adopted to construct an optimization problem. The automatic model selection algorithm is used to find an optimal solution, along with optimal neural network architectures. Furthermore, the neural network is trained, and the parameter estimation of the proposed model for stock price forecasting is obtained. Through the robust testing, model verification, and comparison with traditional models, the experimental results demonstrate the new model’s efficiency and reliability.