Gold plays an important role as a precious metal with portfolio diversification; also it is an underlying asset in which volatility is an important factor for pricing option. The aim of this paper is to examine which autoregressive conditional heteroscedasticity model has the best forecast accuracy applied to Chinese gold prices. It seems that the Student's t distribution characterizes better the heavy-tailed returns than the Gaussian distribution. Assets with higher kurtosis are better predicted by a GARCH model with Student's distribution while assets with lower kurtosis are better forecasted by using an EGARCH model. Moreover, stochastic models such as Stable processes appear as good candidates to take heavy-tailed data into account. The authors attempt to model and forecast the volatility of the gold prices at the Shanghai Gold Exchange (SGE) during 2002–2016, using various models from the ARCH family. The analysis covers from as in-sample and out-of-sample sets respectively. The results have been estimated with MAE, MAPE and RMSE as the measures of performance.