Efficiency Comparison of Gold Price Forecasting Models with Data Mining Techniques

Abstract

Gold is a valuable asset. When crises happen, the price of gold will rise, but in a normal situation, the prices of gold may decrease or remain stable. For this reason, gold prices can go up and down depending on the situation. This research aimed to study and compare the efficiency of gold price forecasting models using data mining techniques. It used gold price data from 2 January 1990 to 31 December 2020 for a 30-year learning period of 7,885 items and data from 2 January 2021 to 30 June 2021 of 131 items as data for various model tests. The review of related research found that the most popular forecasting models were the Box-Jenkins model, Holt-Winters model, and Artificial Neural Network (ANN). The researcher selected Linear Regression Analysis and Support Vector Machine (SVM) to differentiate from previous research. To measure the efficiency of the model, it used Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The results of the study found that using data for learning, Box-Jenkins model, where the optimal parameters are window size = 120, p = 0, d=1, q = 1 (ARIMA(0,1,1)) generates minimal RMSE. It can be used to forecast the gold price appropriately. The results showed that Box-Jenkins can be used to forecast gold prices most appropriately. The gold trade association determines and announces the gold price of Thailand. Box-Jenkins model can be used to forecast gold prices to plan decision-making up-down gold price futures and Gold traders can use the forecasted gold price to plan gold purchases to manage their warehouses to achieve reasonable costs.