A0260
Title: Modeling nonlinear dynamics of functional time series for large-scale data
Authors: Hannah Lan Huong Lai - National University of Singapore (Singapore) [presenting]
Maria Grith - Erasmus University Rotterdam (Netherlands)
Ying Chen - National University of Singapore (Singapore)
Abstract: In numerous empirical applications, financial and economic data can be naturally represented as curves or surfaces exhibiting nonlinear dynamics. To address this challenge, we propose a Nonlinear Functional Autoregressive model (NFAR) that leverages neural networks to capture the nonlinear serial dependence of functional time series data. The estimation of NFAR models involves a two-stage procedure. In the first stage, we extract low-dimensional components from the covariance operator of the response and explanatory variables. In the second stage, we use neural networks to approximate the complex and nonlinear patterns in the data. We further explore the asymptotic properties of the NFAR models. To demonstrate the effectiveness of our proposed approach, we apply the NFAR model to the daily implied volatility surfaces of the S&P 500 index options from 2009 to 2021. Our results showcase superior prediction accuracy and substantial economic gains, which we illustrate via several trading strategies. These findings suggest the potential of our proposed method in capturing complex dynamics of financial and economic data, thereby providing valuable insights for investors.
