Investigating commodity price interdependence with grancer causality networks

This paper investigates the interdependence among prices in the commodity and natural resource market segment. The analysis is performed using a large dataset made of about 50 commodity prices observed with monthly frequency over a period of almost half a century (1980-2024). These different commodities are clustered in five groups (energy, metals, agriculture, food, other raw materials) in order to discriminate the interdependence within and between groups. The adopted method consists in building a Commodity Price Network (CPN) defined via Granger causality tests. These tests are performed with two alternative empirical strategies: pairwise VAR models estimation (pairwise Granger Causality) and sparse VAR models estimation (sparse VAR Granger Causality). Both price levels and price first differences are considered in order to take the possible non-stationarity or price series into account. Network analysis is performed on the different networks obtained using these alternative series and modelling approaches. Results suggest relevant differences across series and methods but some solid results also emerges, particularly pointing to a generalized interdependence that still assigns a central role to some metals and agricultural products.