Modeling and forecasting daily gold prices in Sri Lanka

Abstract

Gold, a popular and precious metal, is used in a variety of industries. Its demand has significantly impacted the global economy, making it a valuable financial instrument andalow-risk,safeinvestmentformanyinvestors. Crisessuchasthe2019EasterSun- day attacks and the 2020 COVID-19 pandemic significantly impacted the Sri Lankan economy, leading to a major economic crisis. In 2022, the Sri Lankan economy faced its most challenging period, and 2023 is seen as a year of recovery from the deep- est economic crisis. As a safe-haven investment, the uncertainty and instability in the financial markets have led to increased demand for gold during this economic cri- sis. Therefore, it is better to approach a new model for forecasting gold prices in Sri Lanka. Accordingly, daily gold price data from January 1, 2019, to May 31, 2023, which includes the period of the economic crisis, was used for this research. This research aims to develop a high-accuracy model to forecast daily gold prices in Sri Lanka using time series analysis. First, various Autoregressive Integrated Moving Av- erage (ARIMA) models were taken into consideration. Based on the minimum value oftheAkaike’sInformationCriterion(AICandAICc)withsignificantcoefficients,the ARIMA (2,1,4) model was selected as the best one. However, it has autocorrelation and ARCH effects, proving that it is not appropriate for fully describing data. The ex- ploringvolatilityclustering, usingresidualsofARIMA(2,1,4), selectedAutoreressive Conditional Heteroskedasticity (ARCH) model was fitted. However, the ARCH effect remains in ARCH model. Then various Generalized Autoregressive Conditional Het- eroskedasticity (GARCH) models were fitted, and the selected GARCH(2,1) model was based on the minimum value of Information Criterion (AIC and BIC) with sig- nificant coefficients. Finally, combining these two best models the ARIMA(2,1,4)- GARCH(2,1)hybridmodelwasfitted. Accordingly,itwasidentifiedasthebestmodel forforecasting,basedonsignificantcoefficientsandprovenassumptions. Theselected fitted model with 1.815% (< 10%) of Mean Absolute Percentage Error (MAPE) indi- cates high accuracy, and suitability for predicting future daily gold price fluctuations in Sri Lanka.