Long-Memory Processes and Exchange Rate Forecasting

Abstract

Modeling persistence in economic and financial time series using the autoregressive fractionally integrated moving average (ARFIMA) method has attracted the attention of many researchers in recent years [29, 16, 161, 17, 81, 103]. The frequency of the use of ARFIMA modeling in empirical research underscores the importance of efficient, both computational and statistical, estimation of the models. In estimating the parameters of the ARFIMA models, three approaches have been used: Maximum Likelihood(ML) [142], approximate ML [86, 46, 65, 66], and two-step procedures [51, 70]. Geweke and Porter-Hudak’s method [51], unlike the ML approach, is less computationally demanding but some analysts consider it to be inadequate for finite samples.