Forecasting Gold Price Using a Novel Hybrid Model with MEEMD-ConvLSTM

Gold occupies a significant position in the world economy and financial market as a precious metal. In order to forecast gold spot and futures price more accurately, this paper introduces a new prediction model called MEEMD-convLSTM. The IMF sequence of the original gold price was decomposed by using a modified ensemble empirical mode decomposition (MEEMD), and convLSTM was utilized to predict each IMF, then the predicted IMF was constructed to obtain the final forecasting results. Model confidence set test (MCS) statistically validated MEEMD-convLSTM and compared it with the traditional neural network (BP), support vector regression (SVR) and hybrid prediction models. We can draw the conclusion that MEEMD-convLSTM can improve the forecasting accuracy and be the best of all the models chosen. The results of robustness tests showed that the model had better forecasting accuracy before and after the Covid-19 outbreak. Overall, the MEEMD-convLSTM model distinguishes itself as a very promising method for gold spot and gold futures price forecasting.