Long-memory Modelling and Forecasting of the Returns and Volatility of Exchange-traded Notes (ETNs)

Abstract

This research provides evidence in determining the predictability of exchange-traded notes (ETNs). It utilises commodity, currency and equity ETNs as data samples, and examines the performance of the three combinations of long-memory models, that is, autoregressive fractionally integrated moving average and generalised autoregressive conditional heteroskedasticity (ARFIMA-GARCH), autoregressive fractionally integrated moving average and fractionally integrated generalised autoregressive conditional heteroskedasticity (ARFIMA-FIGARCH) and autoregressive fractionally integrated moving average and hyperbolic generalised autoregressive conditional heteroskedasticity (ARFIMA-HYGARCH), and three forecasting horizons, that is, 1-, 5- and 20-step-ahead horizons, to model ETNs returns and volatilities. The article finds long-memory processes in ETNs; however, dual long-memory process in returns and volatilities is not verified. The research also poses a challenge to the weak-form efficiency hypothesis of Fama (1970) because lagged changes determine future values, especially in volatility. The findings also show that differences in the characteristics of commodity, currency and equity ETNs are not concluded because of similarities in ETN traits and several insignificant results. However, the presence of intermediate memory was identified, and should serve as a warning sign for investors not to keep these investments in the long run. Lastly, the ARFIMA-FIGARCH model has a slight edge over the ARFIMA-GARCH and ARFIMA-HYGARCH specifications using 1-, 5- and 20-forecast horizons. JEL Classification: G11, G17