Compare Gold Bullion Trading Forecasts with ARIMA and ARIMAX Models

Keywords: the exchange rate of USD (baht), Nominal Effective Exchange Rate (NEER), SET index Leading Economic Index (LEI) and Dubai Crude Oil Price inverse index

Abstract

The purpose of this article was to find the best model by comparing the average gold price forecast of gold jewelry with the ARIMA and ARIMAX models since January 2012 to April 2023, a total of 136 months. The preliminary variable of ARIMA forecasting was the average gold price trading data of gold jewelry; the dependent variable was the trading forecasting result of the ARIMA model and forecasting with the ARIMAX model, the initial variables were trading forecast results with the ARIMA model, the exchange rate of USD (baht), Nominal Effective Exchange Rate (NEER), SET index Leading Economic Index (LEI) and Dubai Crude Oil Price inverse index. The dependent variables are trading forecast results with the ARIMAX model. MAPE, MAE, MSE, RMSE, Ljung-Box Q (18) and BIC criteria were selected from the lowest values ​​and p-values ​​greater than or equal to 0.05. The study found that forecasting the purchase price of gold jewelry using the ARIMA (1,1,0) model was more accurate than the model. ARIMAX (0,1,0) with MAPE equal to 2.040 %, MAE equal to 447.450, RMSE equal to 447.450, Ljung-box Q(18) equal to 6.384, BIC equal to 12.873 and p-value equal to 0.990. It means that the built-in correlation coefficient at Lag is not significantly different from zero. Forecasting the selling price of gold jewelry using the ARIMA (1,1,0) model is more accurate than the ARIMA (1,1,0) model. ARIMAX (0,1,0), with MAPE equal to 1.993%, MAE equal to 455.887, RMSE equal to 592.078, Ljung-box Q (18) equal to 6.139, BIC equal to 12.885, and p-value equal to 0.992, meaning that the built correlation coefficient at Lag is not significantly different from zero