A partial correlation-based connectedness approach: Extreme dependence among commodities and portfolio implications

Abstract

We propose a partial correlation-based connectedness approach to study the directional connectedness under normal and extreme market conditions among the returns of 22 commodities and compare it with the well-known Diebold and Yilmaz (i.e. generalized forecast error variance decomposition (GFEVD)) connectedness approach estimated at the mean and tails. Considering four groups of commodities, namely energy, agricultural, precious metals, and industrial metals, and daily data from September 1, 2005 to June 5, 2024, covering various crisis periods, we draw filtered networks and measures of directional connectedness. The main results are summarized as follows. Firstly, the total connectedness index captures the significant commodities related shocks, and intensifies during crises episodes, notably at the extreme lower quantile. Secondly, using partial correlations in the approach of connectedness leads to a surge of the total connectedness level at the extreme lower quantile and identifies the beginnings of major crises earlier than the GFEVD measure of connectedness. Thirdly, the connectedness structure of commodities based on partial correlation is unstable during turbulent market conditions, a feature that is ignored when the GFEVD approach of connectedness is used. Fourthly, in terms of practical implications, the partial correlation-based connectedness portfolio outperforms the GFEVD based minimum connectedness portfolio on a risk adjusted basis.