Gold Price Prediction Model Based on LSTM Neural Network and ARIMA

Abstract

This study proposes a composite model based on LSTM neural network and ARIMA for predicting future international gold prices and providing investors with the best timing strategy for gold from the present moment to the coming year. We successfully predicted monthly gold prices using a simple seasonal model. However, due to the non-stationarity of the daily gold price, we were unable to apply the time series model directly. For this reason, we introduced IQR and linear interpolation to smooth the data by seasonal differencing, providing reliable inputs to ARIMA and LSTM neural networks. Combining these two methods, we derive daily gold price forecasts. After that, we introduce a double SMA timing strategy, which utilizes a 10-day SMA and a 30-day SMA to smooth the curve through a simple moving average. The curves are used to provide appropriate investment solutions. This comprehensive approach not only improves the accuracy of the daily gold price, but also provides investors with more reliable decision support.