Memory and Predictability of Bitcoin Prices : An Arfima-Aegas Approach

Accurate prediction of Bitcoin prices is a purported boon for risky investors, more so, if the forecasts are largely unconditional. This paper introduces a class of autoregressive fractionally integrated moving average model with asymmetric exponential generalized autoregressive score errors to accommodate a complex interplay of ‘memory’ with information asymmetry towards producing a reliable forecast of prices. To build a true forecast function, our conditional variance includes leverage effects, jumps and fat tail-skewness. We show superior predictive performance of our model over competing approaches. Further, we find that Bitcoin markets are not weakly efficient, a conclusion that is commensurate with literature