A 35-year empirical study suggests that about one third of the monthly changes in a broad commodity price index can be attributed to a single global factor that is related to the business cycle. In fact, for a non-fuel commodity basket almost 70% of price changes can be explained by this factor. By contrast, oil and energy price indices have been driven mainly by a fuels-specific factor that is conventionally associated with supply shocks. Short-term price changes of individual commodities depend more on contract-specific events, but also display a significant influence of global and sectoral factors. The latent global factor seems to help forecasting commodity index prices at shorter horizons.
The post ties in with SRSV’s lecture on macro trends, particularly their importance at longer time horizons.
The below are excerpts from the paper. Emphasis and cursive text have been added.
Identifying latent commodity price factors
“Our empirical strategy consists in extracting latent factors from a large panel of commodity prices. In doing so, we decompose each commodity price series into a global (or common) component, block-specific components related to specific categories of commodity prices, and a purely idiosyncratic shock. The distinction between global, block-specific and idiosyncratic components reflects the underlying idea that different commodity price shocks have distinct consequences on the cross-correlation among commodity prices.”
“The data used in the estimation include the spot prices of 52 commodities from different categories, including food, beverages, agricultural raw materials, metals and fuel commodities. Commodity prices come from the IMF primary commodity price database and they cover the period from January 1980 to December 2015.”
“The model used here is an approximate dynamic factor model for large cross-sections… each series is the sum of two unobservable components, a common component – capturing the bulk of cross-sectional co-movements – and an idiosyncratic component reflecting specific shocks or measurement errors…The common factors and the idiosyncratic component are uncorrelated at all leads and lags… We further [decompose the idiosyncratic component] into factors that are specific to groups or blocks of commodities and a purely idiosyncratic component.”
“There is a single global factor driving [34%] of commodity price fluctuations [captured by a commodity index] and, by moving prices in the same direction, the global factor has limited effects on relative prices…Almost 70% of the variations in the index of non-fuel commodities is attributed to the global factor.…[By contrast] the bulk [around 80%] of the fluctuations in oil [and energy] prices is captured by fuel-specific shocks.”
N.B.: The contract prices for individual commodities depend less on the global factor than indices and much more on purely idiosyncratic factors. The global factor explains 1-40% of individual commodities’ price variance.
“[The figure below] shows the global factor estimated over the full sample. The figure plots the global factor along with the IMF global index of commodity prices. The main motivation for comparing the two indices is that one key feature of the estimated global factor…is that it filters out the idiosyncratic noise in the data to a greater extent than simple averages of commodity prices.”
“The global factor has homogeneous effects on all markets and hence limited effects on relative prices. Since the start of the new millennium, the relevance of the global factor has increased, especially for oil…Oil prices have become more correlated with other non-fuel commodities because of common global forces.”
What is driving the “global” commodity factor?
“The global factor estimated from a large panel of commodity prices is closely related to changes in global economic activity…The global factor is persistent and follows the major expansion and contraction phases in the international business cycle with the largest declines following recession periods…suggesting a close link with demand factors. “
“We…find that the global factor explains a large fraction of commodity price fluctuations during episodes typically associated with changes in global demand conditions, such as the world economic expansion that started around 2003 and the steep contraction during the Great Recession.”
“Recent studies on commodity prices based on principal components have related the estimated common factors to macroeconomic and financial data, such as the real interest rate, the US exchange rate and industrial production… Alternatively, studies in the financial literature…have related the co-movement among commodity prices to the growing participation of financial speculators in commodity markets.”
“By contrast, block components explain most of the fluctuations in commodity prices during episodes conventionally associated with supply or other commodity-specific shocks.”
Predicting commodity prices with the global factor
“We find that the factor model performs well in forecasting commodity prices and indices of commodity prices, in particular at short horizons.”
“We investigate whether the global factor has predictive power for commodity prices by conducting a real-time forecasting exercise. To this end, we take the perspective of a researcher who every month starting from January 2001, would have estimated the model using 20 years of past data, and used the estimates to compute out-of-sample forecasts of commodity prices each month from February 2001 to December 2015. The…forecasts for individual commodity prices are iterated from the state-space representation using the Kalman filter.”
“The main results of the out-of-sample forecasting exercise can be summarised as follows. First, we observe that the model performs well in predicting commodity prices at shorter horizons…At a horizon of 1 month the model outperforms the benchmark with gains in accuracy that range from 18% for the non-fuel index to 12% for oil. Looking at disaggregated commodity prices, the predictive component appears to be stronger for the group of commodities such as metals and food for which the common component is more important [such as] copper (19%), rice (19%), poultry (46%), cotton (17%) and aluminium (12%). Second, the gains fade out progressively over longer horizons and at a horizon of 12 months, we cannot reject the hypothesis of equal predictive performance.”