Directional returns for gold and silver: A cluster analysis approach

Abstract

This paper considers the directional predictability of daily returns for both gold and silver. These two metals have had a long history behaving sometimes as complements and other times as substitutes. We use daily data from June of 2008 through February of 2015. The last 2 years were removed as a set for validation of the model and the remainder, almost 5 years, was used as training. A cluster analysis yields six important clusters. An evaluation of these clusters leads to the formation of three strategies for directional predictions – up or down—for both gold and silver returns. The results of this analysis suggest that each strategy has its own advantages: the first strategy suggests that gold returns can be predicted better than those of silver; the second strategy shows that predicting up for gold also means predicting down for silver and the final strategy confirms that predicting up for silver also validates predicting down for gold.