Commodity prices have fallen dramatically over the past two years. Figure 1 shows the IMF’s commodity price indices for fuels, metals and agricultual raw materials.
Figure 1: IMF Commodity Price Indices for Fuels, Metals, and Agricultural Raw Materials.
A popular explanation for the commodities rout is the slowdown in China, and more generally, the slowdown in the global economy. A second explanation, favored by Liberty Street Economics, is the strength of the dollar. Since commodities are priced in dollars, a stronger dollar requires commodity prices to fall in order for markets to clear. The dollar has strengthened is large part because the Fed has embarked on a tightening cycle even as the European and Japanese central banks are easing. Another implication of US monetary tightening is a slowdown in the creation of international dollar credit, in turn implying weaker demand for primary commodities. Figure 2 shows the trade-weighted US dollar Index (Dollar Strength) and US dollar credit to non-bank non-residents (Global Dollar Liquidity), compiled by the Bank of International Settlements.
Figure 2: Dollar Strength and Global Dollar Liquidity.
Commodities are also financial assets, so that their prices are affected by the market price of risk. Etula (2013) showed that the risk-bearing capacity of US broker-dealers—Wall Street banks—is an important determinant of commodity returns. This is because commodity derivatives are largely traded over-the-counter (OTC); that is, in markets where dealers serve as market-makers. Greater dealer risk appetite implies lower expected commodity returns while increased dealer risk aversion implies higher expected commodity returns. Figure 3 shows Etula’s measure of broker-dealer risk appetite (Effective Risk Aversion).
Figure 3: Effective Risk Aversion (detrended).
In order to understand the contributions of these different factors, I estimated linear models for quarterly changes in Fuel and Non-Fuel commodity price indices compiled by the IMF over 2000-2015. I found that a parsimonious model with only three variables (Effective Risk Aversion, Dollar Strength and Global Dollar Liquidity) explains 28 percent of the variation in Fuels and 51 percent of the variation in Non-Fuels. I also tried other reasonable predictors for which quarterly data is available outside paywalls. None had significant explanatory power. In particular, OECD, Chinese and EM growth rates were insignificant for both indices even at the 10 percent level.
Figure 4: Contributions to YoY% changes in the Fuel Price Index.
Figure 4 shows the decomposition of year-on-year percentage changes in the Fuel Price Index. We see that Dollar Strength and Global Dollar Liquidity have been major factors pulling down energy prices. Still, there is a big residual that presumably contains the large-scale effects of geopolitical and oversupply factors. This is certainly the case with energy prices. In 2012-2015, US shale gas production increased by 10 billion cubic feet per day. In the same period, US oil production rose by 3 million barrels a day. In addition, the Saudis essentially declared a price war on their Russian and Iranian rivals as well as on American oil firms. I have no supply-side predictors in the model, meaning that if commodity prices fell due to negative supply shocks then that variation would not be captured by the model. Indeed, it would very sketchy if it did!
Figure 5: Contributions to YoY% changes in the Non-Fuel Price Index.
Figure 5 shows the decomposition of year-on-year percentage changes in the Non-Fuel Commodities Price Index. The model performs much better for non-fuel commodities, where it is able to explain half the price variation in 2000-2015. The residuals here are much smaller, showing that a supply glut has been less of a factor for non-fuel commodities than for energy. Dollar Strength and Global Dollar Liquidity each explain roughly a quarter of the price decline in non-fuel commodities over the past two years. On the other hand, the impact of Effective Risk Aversion has been largely positive over the past two years.
The raw correlation between the fuel and non-fuel price indices in the period under consideration is 71 percent, while that of the fitted values is 98 percent. But the correlation between the two residuals is still 57 percent, meaning that while some of the comovement of two series is accounted for by the monetary factors in our model, much of it still begs explanation. The obvious explanation that comes to mind is market expectations of future demand growth. Since commodity prices are forward-looking, market expectations of future demand growth for commodities is likely the dominant factor driving their residual covariation. And that takes us back to China.
 Etula, Erkko. “Broker-dealer risk appetite and commodity returns.” Journal of Financial Econometrics 11.3 (2013): 486-521.
 I also modeled Metals and Agricultural Raw Material Price Indices with similar results. These results are omitted here for the sake of brevity.
 More precisely, the independent variables in the model were one quarter lagged, detrended Effective Risk Aversion, changes in the trade-weighted dollar index, and changes in the natural log of US dollar credit to non-bank non-US residents. All independent variables were normalized to have zero mean and unit variance.
 I considered the possibility of a supply-side cycle. If there were a detectable supply-side cycle, then the residuals would oscillate about zero. But subjecting the residuals to the runs test failed to reject the null hypothesis of randomly generated errors. This was also the case with Metals and Agricultural Raw Material Price Indices.