Forecasting of Gold Coin Future Contract Return in Iran Mercantile Exchange: with Approach of Comparative between ARCH Model and Neural Network

Given the importance and role of gold as a tool for investing, especially in developing countries, various approaches have been used to predict gold future returns. Hence, the main purpose of the present study is prediction the daily return of gold coin future contract by using multilayer feed-forward neural network and auto-regressive conditional heteroskedasticity (ARCH) models in Iran mercantile exchange. For this purpose, the daily data on 20 the gold coin futures contracts for periods July 2014 to September 2016 which has been continued using the method “back-adjusted”, is used. Also, after investigating results of previous studies, dollar price return, gold coin price return and global gold price return have been used as effecting variables on gold coin future contracts return. In addition, Predictive accuracy the neural network and the ARCH models were evaluated using root mean squared error (RMSE), mean squared error (MSE), mean absolute error (MAE) and coefficient of determination (R2). The results showed that in the period under review, the neural network model performs better than the ARCH model in the prediction out of sample. But based on the results of the paired t-test, the prediction accuracy of the two models hasn’t been the statistically significant difference.