The US dollar exchange rate and the demand for oil

Transcription

1 The US dollar exchange rate and the demand for oil Selien De Schryder Ghent University Gert Peersman Ghent University BoE, CAMA and MMF Workshop on Understanding Oil and Commodity Prices" 25 May 2012

2 Motivation Increased attention for demand side of oil market Oil price fluctuations mainly driven by changes in oil demand in recent times E.g. Peersman (2005), Hamilton (2009), Kilian (2009), Peersman and Van Robays (2009), Kilian and Murphy (2010) Low price elasticity of oil demand implies that supply shocks have considerable impact on oil prices resulting in very high oil price volatility Baumeister and Peersman (2009): Price elasticity has declined a lot over time BVAR with time-varying parameters Van Robays (2012): Price elasticity lower in periods of macroeconomic uncertainty Threshold VAR Gately and Huntington (2002): Price elasticity is higher for oil price increases versus decreases Multi country panel with fixed effects

3 This paper Explores the role of US dollar exchange rate for oil demand Typically ignored in existing studies (focus on price and income elasticity) Surprising: influence of 1% increase in oil prices (expressed in US dollar) should be similar for oil demand (outside US) as 1% appreciation of US dollar For a lot of oil consumption, it is the local price that matters If exchange rate matters, changes in US dollar shift oil demand curve Evolution exchange rate, oil production and oil prices over time suggests that this could indeed be the case: oil production and oil prices decline when the US dollar exchange rate appreciates, and vice versa (see figures)

4 This paper Explores the role of US dollar exchange rate for oil demand Real effective US dollar exchange rate Detrended world oil production Real effective US dollar exchange rate Real oil price in US dollars

5 This paper Explores the role of US dollar exchange rate for oil demand Panel dataset of 23 OECD and 42 non-oecd countries over sample period (annual data) Oil exporting countries are excluded Using recent advanced in panel data estimation techniques Existing panel studies on oil demand suffer a number of problems, e.g. Gately and Huntington (2002), Griffin and Schulman (2005) Are there similar nonlinearities as for price elasticity of oil demand?

6 Benchmark specification Δdem i, t = α + τ trend + λ dem i + γ Δgdp i i i, t + β Δrac i i t i, t 1 i + γ gdp l i + θ Δrer t + ε i, t 1 i, t + β rac l i t 1 + θ rer l i t 1 dem : total oil demand (consumption) per capita gdp : real income per capita rac : international real crude oil price in US dollars rer : real effective US dollar exchange rate Panel error correction model All variables are non-stationary (ADF and PANIC tests) Null hypothesis of no error correction between variables is rejected (GUW test)

7 Existing panel studies Estimate oil demand function with standard fixed effects estimator E.g. Gately and Huntington (2002), Griffin and Schulman (2005) FE estimators are biased when there is heterogeneity in slope coefficients Likely to be the case given differences in economic structures across countries Cross-sectional dependence in error terms is ignored Results are biased when observed explanatory variables are correlated with unobserved common factors E.g. common global business cycle, monetary policy or speculation All countries in panel are equally weighted Result is average elasticity, which is not the same as global elasticity

8 Fixed effects estimator OECD countries Real income 0,741*** (0,094) Oil price -0,045*** (0,009) Exchange rate -0,156*** (0,037) Non-OECD countries 0,537*** (0,110) -0,013 (0,014) -0,101* (0,059) US dollar exchange rate has a significant impact on demand for oil Magnitude is considerably higher than price elasticity But: Homogeneity in slope coefficients clearly rejected by Hausman tests Respectively 87,04*** and 19,65*** for OECD and Non-OECD sample Also likelihood ratio tests reject homogeneity

9 Mean Group estimator OECD countries Non-OECD countries FE MG FE MG Real income 0,741*** 0,672*** 0,537*** 0,615*** (0,094) (0,073) (0,110) (0,113) Oil price -0,045*** -0,049*** -0,013-0,018 (0,0089) (0,007) (0,014) (0,014) Exchange rate -0,156*** -0,144*** -0,101* -0,072 (0,037) (0,033) (0,059) (0,055) MG estimator does not impose homogeneity Does not matter much for magnitudes of coefficients, except for impact exchange rate in non-oecd countries But: CD test of Pesaran (2004) indicates there is cross-section dependence in the error terms

10 Reducing cross-section dependence in error terms Pesaran (2004) common correlated effects estimator (CCE estimator) Eliminates asymptotically cross-section dependence by using the cross-section averages of the dependent and independent variables as observable proxies for the linear combination of unobserved common factors Not applicable in our study because oil price and US dollar exchange rate are already common variables by construction Reduction of degree of cross-sectional dependence only marginal when we add mean real income per capita as regressors to the model Bai, Kao and Ng (2009) Include estimated common component of residuals of MG regression to eliminate cross-sectional dependence in error terms of cointegrating model Results in much larger decline of cross-sectional correlation, in particular for OECD sample

11 MG estimator corrected for cross-section dependence OECD countries Non-OECD countries MG MG_Ft MG MG_Ft Real income 0,672*** 0,568*** 0,615*** 0,495*** (0,073) (0,066) (0,113) (0,085) Oil price -0,049*** -0,051*** -0,018-0,013 (0,007) (0,0067) (0,014) (0,010) Exchange rate -0,144*** -0,204*** -0,072-0,039 (0,033) (0,040) (0,055) (0,048) Taking into account cross-section dependence matters for the results Elasticity of exchange rate much larger in OECD countries and smaller (but still insignificant) in non-oecd countries Also decline of income elasticity, while irrelevant for oil price elasticity

12 Average versus global elasticity Estimated elasticity is (unweighted) average of individual countries Not very useful for the analysis of global oil market E.g. Kilian and Murphy (2010) implement estimated price elasticity oil demand as a boundary restriction in a global VAR to identify different types of oil shocks Since we are using MG estimator, we can accommodate this by deriving the Weighted Mean Group estimates Shares of individual countries in total oil demand are used as weights Also more reliable for relative small samples (Pesaran 2006)

13 Weighted Mean Group estimator OECD countries Non-OECD countries Overall MG_Ft W_MG_Ft MG_Ft W_MG_Ft W_MG_Ft Real income 0,568*** 0,529*** 0,495*** 0,405*** 0,484*** (0,073) (0,108) (0,085) (0,062) (0,060) Oil price -0,051*** -0,050*** -0,013-0,035*** -0,045*** (0,007) (0,010) (0,010) (0,005) (0,006) Exchange rate -0,204*** -0,200*** -0,039-0,101*** -0,168*** (0,040) (0,041) (0,048) (0,015) (0,021) Does not matter for OECD countries, but results change a lot for Non-OECD countries (indicates lower elasticities for countries with small share) Exchange rate elasticity is still more than 3 times larger than price elasticity

14 Some robustness checks Results robust for using instrumental variables (lags and FFrate) Results robust for using bilateral real exchange rates and IV variables (only available for OECD countries) Elasticity (not surprisingly) declines when we include Federal Funds rate Note we already take into account output, oil price and common unobserved factors in all estimations (and dependent variable is oil consumption) OECD countries Non-OECD countries Overall Real income 0,539*** (0,110) Oil price -0,034*** (0,007) Exchange rate -0,048*** (0,010) Interest rate -0,001*** (0,000) 0,324*** (0,049) -0,023*** (0,003) -0,023*** (0,004) -0,002*** (0,000) 0,400*** (0,049) -0,033*** (0,004) -0,050*** (0,006) -0,001*** (0,000)

15 US dollar exchange rate and nonlinearities Gately and Huntington (2002): Price elasticity is higher for oil price increases versus decreases Same is true for appreciation versus depreciation US dollar exchange rate in OECD countries (not for non-oecd countries) OECD countries Real income 0,566*** (0,116) Oil price -0,058*** (0,012) USD appreciation -0,425*** (0,087) USD depreciation -0,006*** (0,002) Non-OECD countries 0,431*** (0,117) -0,036*** (0,014) -0,104*** (0,016) -0,108*** (0,016)

16 US dollar exchange rate and nonlinearities Baumeister and Peersman (2009): price elasticity has declined over time Same is true for US dollar exchange rate in OECD countries Interaction of the exchange rate with a simple linear trend This is also the case in Non-OECD countries OECD countries Real income 0,473*** (0,097) Oil price -0,042*** (0,009) Exchange rate -0,453*** (0,093) Exchange rate * trend 0,014*** (0,003) Non-OECD countries 0,426*** (0,065) -0,033*** (0,005) -0,227*** (0,035) 0,008*** (0,001)

17 US dollar exchange rate and nonlinearities What happens if we combine both nonlinearities? We even find a significant difference of appreciation versus depreciation in non- OECD countries (but magnitude of difference is very small) OECD countries Real income 0,514*** (0,105) Oil price -0,051*** (0,010) USD appreciation -0,674*** (0,138) USD depreciation -0,238*** (0,049) Exchange rate * trend 0,013*** (0,003) Non-OECD countries 0,466*** (0,068) -0,034*** (0,005) -0,227*** (0,035) -0,209*** (0,032) 0,007*** (0,001)

18 Economic relevance A back of the envelope calculation Our (simplified) benchmark oil demand: q = -0,045 p - 0,168 USD Kilian and Murphy (2010) oil supply: q < 0,025 p In the data, the monthly average of USD = 1,00% To clear the market, this implies that the exchange rate on average leads to: q > 0,06% (whereas monthly average q is 0,96% in the data) p > 2,40% (monthly average p 4,46% in the data) Exchange rate is hence very important contributor to oil price volatility

19 Economic relevance A (preliminary) structural VAR in the spirit of Kilian (2009), Peersman and Van Robays (2009), Kilian and Murphy (2010) Four variables: oil production, real oil prices, world economic activity and the real effective US dollar exchange rate Identification of oil demand shock ( q>0 and p>0) which is characterized by depreciation of the US dollar exchange rate ( er<0) and orthogonal to world economic activity ( y=0), while oil supply price elasticity < 0,025 This shock explain approximately 4-5% of oil production volatility and 60-65% of oil price fluctuations Implied exchange rate elasticity of the shock is between -0,16 and -0,21 Results very similar when we also orthogonalize to interest rate innovations

20 Conclusions The US dollar exchange rate is a crucial determinant for oil demand Appreciation of US dollar consistently leads to decline in oil demand The elasticity turns out to be much higher than the price elasticity of oil demand (expressed in US dollar) Economically important to explain fluctuations in the oil market: should be taken into account in the analysis of the global oil market Open issue: why is the elasticity much higher than price elasticity? Stronger responsiveness to a less volatile variable (e.g. signal to noise)? Short term responsiveness is constrained, which implies a larger estimated elasticity of smaller shocks (another nonlinearity in oil demand)? Future research: does exchange rate also influence the supply side of the oil market?