Commodity prices: how important are real and nominal shocks?

Abstract

We consider the response of both nominal and real commodity prices on world markets to real and nominal shocks by hypothesizing that nominal shocks can permanently affect nominal commodity prices, but can have only temporary effects on real commodity prices. Real shocks, in contrast, can have permanent as well as temporary effects on both nominal and real commodity prices. When nominal and real shocks are decomposed in this manner, real shocks are found to be of much greater importance to the observed movements in commodity prices. We use the Blanchard and Quah (BQ, 1989) decomposition to obtain the permanent and temporary components of the real commodity price series and relate this to the rate of growth of world industrial production as an indicator of business cycle movements. The results suggest that the impact of the business cycle is self-stabilizing in that there is an initial positive effect on growth in commodity prices followed by a fully offsetting negative effect.

Keywords:

• commodity prices
• business cycle
• nominal and real shocks

JEL Classification::

• E32
• C22

Notes

¹ As Enders and Lee (1997, p. 239) acknowledge, the purpose of their theoretical model is to demonstrate the identifying restriction with respect to the transient characteristics of nominal shocks on real variables, rather than offering ‘the correct model of the US economy’.

² Lag length choice for the VAR in first differences is based on the validity of a model reduction sequence, information criteria and tests for residual autocorrelation. Since the data are monthly, we test down from a lag length of 13 in the VAR, information criteria (SIC, Hannan–Quinn (HQ) and AIC) all select a lag length of 1, F-tests for the model reduction from lag 13 to 1 do not reject (the F-test for the reduction is F(48,554) = 1.2493, with a p-value of 0.1277) and the VAR(1) shows no evidence of serial correlation (with tests for autocorrelation in the individual equations yielding p-values of 0.23 and 0.28).

³ The deterministic component was also decomposed. While this had a negative trend, values were of very small magnitude having a mean of −0.123.

⁴ Results for the temporary component were of less interest and are available from the authors on request.

⁵ Full results of the cointegration tests are available from the authors on request.

⁶ The lag order is initially set at 13 and with model reduction tests (both information criteria and F-tests) of the reduction sequence. The tests are conflicting. Both the SIC and HQcriterion suggest lower order lags of 3 in the VAR, however the VAR(3) shows significant evidence of autocorrelation, whilst the AIC selected a larger lag order (12). The lag length of 12 is further supported by tests on reducing lag length.

⁷ Hansen's (1992) tests are applicable to regression models with stationary regressors as is the case here.