Modeling and Forecasting Sri Lankan Gold Prices

Abstract

The movements of the prices of gold are both interesting and important. It can be forecasted and used for making future decisions. The main objective of this research study is to develop a forecasting model to predict gold prices in Sri Lanka with high accuracy. This study took monthly data of the gold prices per troy ounce in Sri Lanka from January 2005 to May 2014 which consists of, 113 observations. 90% observations were used for modeling and 10% observations were used for testing. This research study developed two models; Auto Regressive Integrated Moving Average (ARIMA) model and Vector Auto Regressive (VAR) model for forecasting monthly gold prices per troy ounce in Sri Lanka and a comparison was done to find the best model among them. Based on the literature and preliminary analysis, monthly data of Exchange rate (Sri Lankan rupees per dollar), inflation rate and narrow money supply (Rupees in million) were selected as the explanatory variables to build the VAR model. Mean Absolute Percentage Error (MAPE) value was used to asses the suitability of fitted models. ARIMA (2,1,2) model was selected as the best model to forecast monthly gold prices in Sri Lanka as this model comply with all conditions and assumptions of ARIMA. MAPE value of fitted data from ARIMA (2,1,2) model is 9.8855. According to the fitted VAR model it can be concluded that the change in the gold price of current month is affected by 94.03% of the change in the gold price of previous month. Percentage change of exchange rate, inflation rate and narrow money supply of the previous month are not individually affected to the percentage change of gold price of the current month meanwhile those variables are jointly affected to the percentage change of gold price of the current month.