Application of Time Series Modelling and Change Point Detection Algorithms for Gold Price Forecasting

Abstract

"Gold is a precious metal which has become a crucial investment tool across the globe. Due to the high significance of gold for both as a commodity and financial asset, investors are more focused on obtaining gold price forecasts for taking more informative decisions regarding their investments. However, due to the dynamic nature of the gold prices it has been challenging to provide accurate forecasts. Therefore, this study is mainly focused on obtaining more reliable model to make gold price forecasts with the application of Change Point Detection algorithms. Data for the study is taken from the Yahoo Finance website containing data from 2013-01-01 to 2023-09-29. Five change point detection algorithms that are available under “Ruptures” Python library is used to detect the change points and separate datasets are created according to the identified latest change point date by each of the algorithms. ARIMA, CNN, RNN and LSTM models are employed to obtain the predictions, and the models are compared using the RMSE and MSE. According to the results, LSTM model applied on the dataset created by the Window Based change point detection algorithm (LSTM_WB_Model) has the lowest RMSE and the MSE values indicating a positive impact on application of change point detection algorithms on making forecasts for gold prices. The identified best model can make forecasts for at least next 3 to 5 months. Moreover, an application is developed to obtain the forecasts for the stakeholders."