World Output and Commodity Price Cycles*

Abstract

This study investigates the cyclical patterns of energy, agriculture, and metals and minerals (MetMin) commodity prices. We identify three super cycles since 1960, and a potential fourth arising from the Ukraine crisis and global COVID-19 pandemic. Employing a Structural Vector Autoregression (SVAR) approach, we establish an empirical relationship between output, CPI, and commodity prices. Our analysis reveals that an output shock leads to a general increase in all commodity prices, where the highest impact is on energy inflation. Moreover, we examine the heterogeneous effects of commodity inflation on overall inflation, uncovering ‘second round' effects across all commodities. Notably, agriculture inflation has the most significant impact on aggregate inflation, potentially explaining the destabilizing nature of food inflation in many countries. Our findings enhance understanding of these dynamics, offering important insights for policymakers and informing the public.

Highlights

• We analyze the cyclical patterns of energy, agriculture and MetMin commodity prices.

• Real output, CPI and commodities exhibit the same cyclical patterns.

• A shock to output increases all commodities, where the highest response is energy inflation.

• We find ‘second-round' effects, where agriculture prices have the highest impact on inflation.

KEYWORDS:

• Super cycles
• Global Commodity Prices
• Global Macroeconometric Modeling

JEL Classifications:

• Q43
• O13
• L61
• E23

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We convert nominal global GDP (FRED mnemonic NYGDPMKTPCDWLD), which is denominated in U.S. Dollars (FRED mnemonic CPALTT01USA661S), to real GDP by dividing the former with the U.S. CPI.

2 The data is publicly available https://www.worldbank.org/en/research/commodity-markets. The agricultural index is a weighted average of prices of food (e.g., cereals, oils), beverages (e.g., coffee, cocoa and tea), agricultural raw materials (e.g., timber, cotton), and metals and minerals (e.g., aluminum, copper, iron ore, lead, nickel, steel, tin, zinc). The oil price is based on the average of the Brent, Dubai and WTI crude oil price.

3 The cycle trend is consistent with Christiano and Fitzgerald (2003) who use the asymmetric CF band pass filter of up to 8 years for output.

4 The 25 economies include: Australia (‘AUS’), Austria (‘AUT’), Belgium (‘Belgium’), Canada (‘CAN’), Switzerland (‘CHE’), Germany (‘DEU’), Spain (‘ESP’), Finland (‘FIN’), France (‘FRA’), United Kingdom (‘GBR’), Greece (‘GRC’), India (‘IND’), Iceland (‘ISL’), Italy (‘ITA’), Japan (‘JPN’), South Korea (‘KOR’), Luxembourg (‘LUX’), Netherlands (‘NLD’), Norway (‘NOR’), New Zealand (‘NZL’), Portugal (‘PRT’), Sweden (‘SWE’), Turkey (‘TUR’), United States (‘USA’) and South Africa (‘ZAF’).

5 Our results are similar to Ratti and Vespignani (2016), who show that the first principal component captures 89.6% of the variation for prices relating to the G5 countries.

6 While the focus is on World, we take the USA as a secondary benchmark.

7 Filtering methods have been used in business cycle research to isolate cyclical component of a time series. While the Hodrick-Prescott (HP) filter (Hodrick and Prescott, 1997) is arguably the most commonly used filtering technique, the HP filter is semi-parametric and has a drawback that the researcher needs to choose an appropriate smoothness parameter (Baxter & King, 1999). The Baxter-King filter facilitates an alternative approach to extracting cyclical components based on a symmetric approximation (i.e., no phase shifts, which can result in trimming a series). The Christiano-Fitzgerald filter extends the Baxter-King filter to include an asymmetric band pass filter can utilize the full sample period.

8 As Cuddington and Jerrett (2008) mention, the trend is not constant and can evolve slowly over time.

9 Note that a minimum of 2 is selected considering that is the minimum value that can be used in to estimate a BP filter.

10 The first principal component of the commodity price non-trend and super cycle components account for 95.3% and 89.4%, respectively (Scree plots are provided in Figure A2 and Figure A3 in the Appendix). The contemporaneous correlation between the non-trend and super cycle components for first principal component of energy, agriculture and MetMin commodity prices and output is 95.2% and 97.4%, respectively.

11 In the case of expected degrees of freedom equal to 1 would indicate that the estimated penalized smooth term would be consistent with a simple linear relationship.

12 The complete IRFs are provided in the Appendix via Figures A4–A8, relating to models that include the energy, non-energy, agriculture, metals and minerals, and oil price index, respectively. We provide the variance decomposition for all five models in the Appendix via Figure A13. We find that output growth and aggregate inflation have a considerable influence on energy and the oil inflation, jointly accounting for circa 30% of the variance.

13 The impact on oil inflation is consistent.

14 Defined as real commodity prices to a change in real global economic growth.

15 Buyuksahin et al. (2016) estimate the price elasticity of global output growth for oil, metals and agriculture to be 14.0, 9.2 and 7.2 percentage points, respectively.

16 We extend the SVAR to a time-varying parameter VAR (TVP-VAR). Consistent with the SVAR model, one lag was chosen for the TVP-VAR estimation for all three models. The IRFs between the SVAR and TVP-VAR are consistent, see Figure A11 in the Appendix.

17 The standard deviation over the entire sample is 0.254, 0.096 and 0.152 for energy, agriculture and MetMin, respectively. The standard deviation since 1990 is 0.245, 0.071 and 0.165 for energy, agriculture and MetMin, respectively.

18 According to Buyuksahin et al. (2016), agricultural production reacts much more quickly, usually within the next growing season, considering investment costs are generally higher for oil and metal projects which can have long gestation periods (e.g., Radetzki, 2006; Cuddington & Jerrett, 2008).

19 According to Amaglobeli et al. (2022), ‘[a] demand response can be sizable for energy, but much less so for food because people need to eat about the same amount.’

20 Ginn & Pourroy, 2019 and Ginn & Pourroy, 2022 show an endogenous fiscal policy response in regards to sizable producer and consumer food price subsidies designed to cushion the effects of rising prices, which may be a policy-induced price smoothing mechanism that is different to, yet in parallel with, the classic Calvo price stickiness approach.

21 Wiggins et al. (2010) notes that once food prices started to increase in 2007, there were amplifying reactions that accelerated the price increases such as export restrictions, country-imposed increase in import taxes on food goods.

22 The results remain the same if one compares the IRFs for food inflation and oil inflation, see Figure A6.

23 Based on IEA data, the energy share is 3.2% for the USA. According to the USDA, the food share is 10.3% in the USA for 2021.

24 According to Pourroy et al. (2016), changes in food prices can induce significant variations in inflation, where the expenditure share on food is typically higher in lower income economies. Pourroy et al. (2016) show that the expenditure share on food in low-income, middle-income and high-income economies is 48%, 31% 20%, respectively. This relationship is known as Engel’s law.

25 See https://www.federalreserve.gov/newsevents/speech/brainard20191108a.htm.

Additional information

Notes on contributors

William Ginn

William Ginn is a Sr Economist/Data Scientist at Labcorp. He is originally from the USA and has spent numerous years abroad in Europe, where he is currently residing in Germany. Ginn completed a B.S. in Economics at East Carolina University; a M.A. in Economics at Duke University; an MBA at Jesus College at University of Oxford; and a doctorate in Economics at University of Erlangen-Nuremberg.