The economic importance of rare earth elements volatility forecasts

Rare earth elements (REEs) have gained increasing attention recently for several key reasons: 1) they are of paramount and strategic importance for a variety of green- and high-technology products, such as hybrid and electric cars, wind energy turbines, photovoltaic cells, mobile phones, hard and CD drives, and permanent magnets, 2) they are relatively scarce as concentration is generelly low, 3) they frequently exhibit high price fluctuations, 4) China holds a quasi-monopoly on their mining as well above 80% of global REE mine production and more than 50% of worldwide REE reserves are located in China, and 5) China’s REE policy of setting export quotas, which was overly restrictive and led to a formal complaint from the U.S., Japan, and the EU at the World Trade Organization (WTO) in 2012. In this paper we use non-parametric as well as maximum likelihood procedures to estimate autoregressive fractional integration moving average (ARFIMA) models to analyze the long memory behaviour of the four major rare earths elements (REEs), i.e., cerium (ce), lanthanum (la) neodymium (Nd) and yttrium (y). We find both, returns and volatilities to exhibit a significant degree of long memory. Accordingly, the REEs in our sample are neither stationary (I(0)) nor unit root processes (I(1)) but fractionally integrated. Furthermore, we use the class of ARFIMA models to generate one step ahead forecasts for both, rare earth returns and volatilities. We compare the accuracy of the long memory ARFIMA forecasts with those generated from short memory ARMA models as well as volatility forecasts from a GARCH(1,1) model. Our findings show that with respect to the return series, the ARFIMA model explicitly allowing for long memory generally is outperforming the traditional ARMA forecasts. This result holds true for both, rare earth oxides as well as metals. More importantly, the out-of-sample forecasting performance with respect to the volatility of rare earth elements is outperforming ARMA forecasts as well as forecasts generated from a GARCH(1,1) model which is the workhorse of volatility forecasting.