Gold Price Forecasting Based on the Improved GM(1,1) Model with Markov Chain by Average of Middle Points

Abstract

In this paper, we combine Markov chain and GM(1,1) model, which will be called MCGM(1,1), to forecast the gold price in London. The transition probabilities of Markov chain are constructed by the error of GM(1,1) and original data from January 1990 to December 2011. We compare the accuracy of the prediction using the testing data from January 2012 to June 2014. We find that MCGM(1,1) is better than GM(1,1) in terms of three criterions, namely, mean square error, absolute mean error and absolute relative error.