Title: Forecasting with information extracted from the residuals of ARIMA in financial time series using continuous wavelet transform

Authors: Heng Yew Lee; Woan Lin Beh; Kong Hoong Lem

Addresses: Faculty of Information and Communication Technology, Universiti Tunku Andul Rahman, Jalan Universiti, Bandar Barat, 31900 Kampar, Perak, Malaysia ' Faculty of Science, Universiti Tunku Andul Rahman Jalan Universiti, Bandar Barat, 31900 Kampar, Perak, Malaysia ' Faculty of Science, Universiti Tunku Andul Rahman Jalan Universiti, Bandar Barat, 31900 Kampar, Perak, Malaysia

Abstract: Time series of financial or economic data are often considered to have certain trends and patterns. It is believed that the study of historical patterns helps in the forecasting into the future. ARIMA model is one of the popular models for the task. However, long-term forecasting with ARIMA often appears as a straight line. This is due to ARIMA's dependency on previous values and its tendency to omit the outliers that lie outside of the captured general trend. This paper sought to capture useful outlier information from the residual of ARIMA modelling by using continuous wavelet transform (CWT). The CWT captured information was then added to the ARIMA forecasted values to form non-homogenous long-term forecasting. The results were encouraging. It was also found that choices of certain CWT related parameters have positive or negative effect to the forecasting outcomes.

Keywords: wavelet; forecasting; autoregressive integrated moving average; ARIMA; time series; continuous wavelet transform; CWT.

DOI: 10.1504/IJBIDM.2023.127313

International Journal of Business Intelligence and Data Mining, 2023 Vol.22 No.1/2, pp.70 - 99

Received: 15 Oct 2021
Accepted: 13 Dec 2021
Published online: 30 Nov 2022 *