Predictability of structural co-movement in commodity prices: the role of technical indicators

Abstract

This paper investigates the financialization and structural co-movement of several commodity futures using factor variance decomposition and predictability of technical indicators and macro variables. We find that financialization is still a dominant player in the commodity market and that recent commodity price fluctuations can be significantly and robustly forecasted by technical analyses of commodity index investments. Moreover, the co-movement of commodities is demonstrated by variance decomposition and explained as commodity index investment, which provides evidence of financialization. The overall empirical analysis reveals that technical indicators and macro variables can statistically and economically forecast the indexed investment and off-index trading, respectively, which indicates that they are suitable predictors of the commodity markets.

Keywords:

• Price predictability
• Technical indicators
• Commodity returns
• Co-movement
• Financialization of commodity

JEL:

• C32
• E30

Acknowledgement

This work was supported by the National Natural Science Foundation of China [project number 71671193], [project number 71401193], [project number 71473279]; the key projects of the National Social Science Foundation of China [project number 15ZDC024]; the program for new century excellent talents of the Ministry of Education of China [project number NCET-13-1055]; Program for Innovation Research in Central Univers.

Notes

¹ The choices of the values considered for the parameters of the three technical indicators are referenced in existing studies; see Miffre and Rallis (2007), Marshall et al. (2008), Szakmary et al. (2010), Fuertes et al. (2010), Shynkevich (2012), Neely et al. (2014) and Narayan et al. (2015).

² To ensure that ${\mathit{\beta }}_i^w$, ${\mathit{\beta }}_i^s$ and ${\mathit{\beta }}_i^c$ sum to one, we follow Kose et al. (2003) and orthogonalize the factors when computing the variance decompositions at each replication. Because the sample correlations are small, this has little influence on the results. The idiosyncratic errors ɛ_i,t are assumed to be normally distributed but may be serially correlated.