On the Predictive Ability of Time-Domain Modeling of Long Memory Data

Adeboye, Nurain Olawale and Ogunnusi, O.N (2020) On the Predictive Ability of Time-Domain Modeling of Long Memory Data. In: 4th International Conference of Professional Statisticians Society of Nigeria, 2020, Nigeria.

Text PSSN-CP-2020.pdf Download (1MB)

Abstract

The study of long memory data required the fitting of an appropriate time-domain model(s) which can be used to achieve a high level of precision in the forecast. To this end, Autoregressive Fractional Integrated Moving Average (ARFIMA) and Seasonal Autoregressive Integrated Moving Average (SARIMA) models were considered with special focus on the comparative predictive ability of the two techniques, using the Nigerian Stock Exchange All-share Index (ASI) as a study. The ASI series was subjected to a Unit Root test using Augmented Dickey-Fuller (ADF) approach and cross-examination of the ACF showed the presence of a long memory structure, which was confirmed using the Hurst exponent test. The Geweke and Porter-Hudak (GPH) method of estimation was used to obtain the long memory parameter of the ARFIMA model while the SARIMA model was also fitted for the ASI. However, based on minimum AIC and Maximum log-likelihoods, ARFIMA(4,0.204,1) and SARIMA (4,1,1)x(1,1,1)12 were found to be the best from several iterated models. Forecast evaluations of the best-fitted models were carried out using MAE, RMSE, and MAPE respectively. Results indicated that the SARIMA model was much better in prediction as against most established literature of ARFIMA’s superiority in the modelling of long memory data.

Item Type:	Conference or Workshop Item (Paper)
Subjects:	Q Science > QA Mathematics
Divisions:	Faculty of Engineering, Science and Mathematics > School of Mathematics
Depositing User:	Mr Taiwo Egbeyemi
Date Deposited:	08 Feb 2022 09:26
Last Modified:	08 Feb 2022 09:26
URI:	http://eprints.federalpolyilaro.edu.ng/id/eprint/1823

Actions (login required)

View Item