Predict Gold Yields with ARIMA and LSTM

Abstract

The volatility of the gold price has a profound effect on many financial activities around the world. Developing a reliable forecasting model can provide insight into the volatility, behavior and dynamics of the gold price and can ultimately provide the opportunity to make significant profits. Common forecasting models fall into two categories: classical linear models and deep learning models using neural networks. In this paper, the ARIMA model and the LSTM model are selected as representatives of the two types mentioned above to forecast the gold return on May 1, 2020.The experimental results show that the 90% confidence interval of ARIMA includes the actual gold return, but there is a substantial discrepancy between the point forecast and the gold return rate itself. And the prediction accuracy of the LSTM model depends heavily on the data size of the training set. Finally, by comparing the forecasting results with the actual results, this paper analyzes the shortcomings of each model and proposes improvement directions for future gold return rate forecasting research.