LAUSR

This paper analyses the causal relationship between commodities funds and returns using monthly data for the period May 1997 to August 2015. Linear Granger causality tests fail to detect any evidence of causal relationships. However, our data reveals strong evidence of nonlinearity and structural breaks, making the results from the linear model unreliable. Given this, we use wavelets to analyse the causality between the two variables at both time and frequency domains. Wavelet coherency reveals that these two variables are primarily positively related in the short-run and over the period of 2008 to 2015. When we investigate the phase differences over this period, we observe that flows have predicted returns over the period of 2008 to 2012, with causality running in the other direction thereafter. Our results highlight the importance of using a time-varying approach across the frequency domain to draw correct inferences between commodity returns and flows, especially in the presence of nonlinearities and structural breaks, which results in a misspecified linear model of Granger causality.