A1440
Title: A semiparametric semi-strong FARIMA with heteroskedastic errors applied to energy market
Authors: Yuanhua Feng - Paderborn University (Germany)
Aliyu Abubakar Musa - University of Paderborn, Germany (Germany) [presenting]
Abstract: Usually, the FARIMA and the SEMIFAR (semiparametric fractional autoregression) models are assumed to be linear with i.i.d. errors. The FARIMA-GARCH and SEMIFAR-GARCH models provide semi-strong, non-linear extensions of them with uncorrelated GARCH errors. A SEMIFAR-SemiGARCH model is proposed by introducing a latent smooth scale function into the uncorrelated GARCH errors. This provides an approach with trend and long-memory in the mean part, as well as a (latent) smooth scale function and stationary volatility in the errors, where the last component can be analyzed by different GARCH models. The properties of this novel model and its estimation theory are investigated. A multi-step algorithm that combines the SEMIFAR and SemiGARCH estimation procedures has been developed for practical implementation. The performance of this algorithm is justified. This proposal is applied to a selected time series in the energy market. Results show that the proposed model is useful in practice and can be applied to forecast different stationary and non-stationary components in a time series. It can also be extended to include long memory in volatility.
