The performance of long memory fractional series price model of essential trace element zinc

Abstract

Zinc is an essential trace element metal commodity that plays an important role in the economy because it is used as a raw material in various industries such as medicine, biological process, electronic devices, building construction, cement, and textiles. This study aims to build the pattern of zinc futures price movement in the form of linear and nonlinear model. An autoregressive fractionally integrated moving average (ARFIMA) and Double Exponential Smoothing (DES) as linear models were developed for time series data containing the exponential and long memory effects where the order can be formed fractionally, respectively. In contrast, Long Short-Term Memory as nonlinear model demonstrating the neurons system function also utilized to build the model of zinc price. In determining the prefered model of zinc price, three models of ARFIMA, DES, and LSTM are then compared by using MAE, RMSE, and MAPE, while AIC and BIC are used to measure the best selection model in ARFIMA model. Zinc futures price data shows the ARFIMA and LSTM models with a long memory effect with a high accuracy value, which are better than the classical model of DES. The result shows that zinc futures price has an important role in the industry because the price tends to be stable with a long memory effect as an industrial raw material.