Oil price shocks and the US economy: What makes the latest oil price episode different

Oil price shocks and the US economy: What makes the latest oil price episode different Mar´ıa Dolores Gadea∗ Ana G´omez-Loscos† Universidad de Zarag...
2 downloads 0 Views 268KB Size
Oil price shocks and the US economy: What makes the latest oil price episode different Mar´ıa Dolores Gadea∗

Ana G´omez-Loscos†

Universidad de Zaragoza

Bank of Spain

May 2013

Abstract This paper distinguishes different periods in the relationship between oil price shocks, economic growth and inflation for the US economy. Focusing on the latest period, covering mainly the noughties, a change is found in the exogeneity pattern associated with recent oil price episodes. A clear effect of GDP growth on oil price movements is identified, while there is no evidence of any influence of the latter on either GDP growth or inflation. JEL classification: C32, E31, E32, Q43 Keywords: oil shocks, inflation, production, structural breaks, exogeneity

∗ University of Zaragoza, Gran V´ıa, 4, 50005 Zaragoza, Spain. Tel: +34 976761842; fax: +34 976761840 and email: [email protected] † Corresponding author: Bank of Spain, Alcal´ a, 48, 28014 Madrid, Spain. Tel: +34 913385817; fax: +34 913385193 and email: [email protected]

1

Oil price shocks and the US economy

1

2

Introduction

The macroeconomic impact of oil price shocks has been widely studied1 and there is a broad consensus among macroeconomists that an increase in oil prices had a significant negative impact on output growth and contributed to high inflation in oil-importing countries during the 1970s, whereas this effect seemed to progressively lessen from the mid-1980s during the tranquil times of the Great Moderation. However, the oil price movements during the noughties have reopened this issue. The past decade has been characterised by a long escalation of oil prices that began in 2002-03 and reached its peak with the oil price spike of 2007-08. Indeed, some economists consider the latter as one of the underlying factors that contributed to the Great Recession (GR).2 Oil prices collapsed at the end of 2008, climbed anew soon afterwards until early 2010 and then began to edge down again. Kilian (2009) considers that the oil price surge that started in 2002, to which the US economic proved resilient, was driven by a series of positive demand shocks associated with shifts in global economic activity3 (his analysis goes up to 2007). More specifically focused on the events of 2007-08, Hamilton (2009) explores their causes and consequences, concluding that this episode contributed to the US economic downturn.4 Furthermore, Kilian’s discussion of this paper links the fall from mid-2008 only in part to an unexpected reduction in global real economic activity (other factors, such as the worsening of the financial crisis, may have also played a role).5 Much of the existing literature, both theoretical and empirical, sustains that major oil price movements were caused by physical disruptions of supply brought about by geopolitical events in the Middle East6 and so they are considered exogenous when evaluating the response of macroeconomic variables to oil price changes. This view is held in the extensive oil-related literature developed by Hamilton (see, for example, Hamilton (1983) or Hamilton (2003)). However, Hamilton (2009) considers that the 2007-08 oil price surge was caused by strong demand colliding with stagnating 1

Two comprehensive surveys of this literature are Hamilton (2008) and Kilian (2008). See Clark (2009) or Hamilton (2009). 3 Kilian and Hicks (2012) analyse the causes of the surge in oil prices from 2003 to 2008, which they link to the growing demand for industrial commodities arising, mainly, from emerging Asia. 4 In this regard, Hamilton (2011) states that the GR is one of the 10 (out of 11) post-war US recessions preceded by an increase in the oil price. 5 For the G7 countries, considering the whole of the 2000s, Gomez-Loscos et al. (2012) find that the impact of oil price shocks on macroeconomic variables, particularly on inflation, has reappeared, but to a lesser extent than during the 1970s. Jimenez-Rodriguez and Sanchez (2010) also find that oil price increases prompted higher inflation and output losses in the new millennium. 6 The Yom Kippur War in 1973, the Iranian revolution in 1978, Iraq’s invasion of Iran in 1980 and Iran’s invasion of Kuwait in 1990. 2

3

Oil price shocks and the US economy

world oil production. Barsky and Kilian (2002) and Kilian (2009) argue against the notion that earlier oil price episodes were driven primarily by unexpected supply disruptions and point out that demand factors could have contributed to increasing oil prices in several of them, challenging the exogeneity of this variable. Summing up, on the one hand, there is evidence of a changing relationship over time both in the responses of output growth and inflation to oil price shocks and a debate on the causality direction and, on the other hand, there is a renewed interest in oil price movements during the noughties. The contribution of this paper is threefold. Focusing on the US economy (1970.1-2012.4), we first analyze oil price exogeneity with respect to GDP growth and inflation through a rolling exercise. Second, we look for endogenous structural breaks in the relationship between the three variables. Given that oil prices could be either endogenous or exogenous with respect to the macroeconomic variables considered, we do not assume either of the two possibilities a priori but we let the data speak, which allows us to choose the correct specification for each period identified. Third, we use a wider span than in previous studies, as our sample runs to 2012 and, consequently, also includes the most recent oil price rise (2009-10), the economic developments during the GR and the subsequent nascent recovery.

2

Methodology and results

2.1

Exogeneity analysis

Taking into consideration the debate about the exogeneity of oil prices, we test the causality of oil prices in relation to output growth and inflation for our whole sample. Our US data span is quarterly and runs from 1970.Q1 to 2012.Q4. Oil prices (OILP) quoted in this paper refer to the Producer Price Index for crude petroleum from the US Bureau of Labor Statistics. GDP and consumer price index (CPI) are from OECD’s Main Economic Indicators.7 Data are displayed in Figure 1. To analyze causality, we perform a rolling exercise. The following stationary VAR system is defined, yt = υ +

p X

Ai yt−i + ut

(1)

i=1 7

Unit root tests do not reject that the growth rates of GDP, consumer price index and oil price are stationary.

Oil price shocks and the US economy

4

where yt = (y1t , ..., ykt )0 is a kx1 random vector, Ai are fixed kxk coefficient matrices, υ = (υ1 , ..., υk )0 is a fixed kx1 vector of intercept terms, ut = (u1t , ..., ukt )0 is a k−dimensional innovation process P with E(ut ) = 0, E(ut ut 0) = u , E(ut u0s ) = 0 for s 6= t and, finally, p is the order of the VAR.8 In our case, yt = (∆GDP, ∆CP I, ∆OILP )0 . In this context, a Wald test for Granger causality is proposed, H0 : α13,1 = α13,2 = 0 for causality from ∆GDP to ∆OILP H0 : α23,1 = α23,2 = 0 for causality from ∆CP I to ∆OILP

(2)

H0 : α13,1 = α13,2 = α23,1 = α23,2 = 0 for causality from ∆GDP and ∆CP I to ∆OILP where αij,l with i,j=1,...,k and l=1,...,p are the coefficients of matrix Al . This test is applied across the entire sample using a rolling method with a window size of 40 quarters. Results of p-values are shown in Figure 2 including a dashed line at the value corresponding to 5%. The years shown on the abscissa correspond to the last value of the rolling window. For example, the figure of 2012.4 reflects the value of the test corresponding to the period 2003.1-2012.3. The results show that oil price exogeneity does not remain stable across the sample. The endogeneity that appears in two different periods, one at the beginning of the sample, due to the influence of inflation reversals in the 1970s, and the other at the end of the sample, as a result of the effect exerted on the oil price by the output growth, is particularly noteworthy. Notice that this latter effect occurs when the rolling window moves across the period 1998.4-2008.3 to 2000.3-2010.1, that is, for about ten years the oil price is endogenous. However, during the last part of the sample, as we move into the period following the GR, the identified causality of GDP growth on oil prices vanishes.

2.2

Identification of structural breaks

From the findings of the previous section, it is immediate to question whether the evolution and causal links among the variables could help us to identify different periods in the relationship between oil prices and macroeconomic variables. To address this question, we consider the methodology developed by Qu and Perron (2007) (QP), which allows us to estimate and test for multiple structural changes that occur at unknown dates in a system of equations. The method of estimation we use 8

According to the information criteria and the diagnosis of the residuals we have selected p=2. However, the results are also robust to higher orders.

5

Oil price shocks and the US economy

is quasi-maximum likelihood based on Normal errors. The model considered is as follows: yt = (I ⊗ zt0 )Sβj + ut

(3)

There are n equations and T observations, excluding the initial conditions if lagged dependent variables are used as regressors. The total number of structural changes in the system is m and the break dates are denoted by the m vector, T = (T1 , ..., Tm ) taking into account that T0 = 1 and Tm+1 = T . A subscript j indexes a regime (j = 1, ...m + 1), a subscript t indexes the temporal observation (t = 1, ..., T ), and a subscript i indexes the equation ( i = 1, ..., n) to which a scalar dependent variable yit is associated. The parameter q is the number of regressors and zt is the set which includes the regressors from all equations zt = (z1t , ..., zqt ). Furthermore, ut has mean 0 and P covariance matrix j for Tj+1 + 1 ≤ t ≤ T j . The matrix S is of dimension nxq with full column rank. We use a selection matrix that involves elements that are 0 and 1 and, thus, indicates which regressors appear in each equation. For our VAR model, we further have zt = (yt−1 , ..., yt−q ), which contains simply the lagged dependent variables q = p and the deterministic terms and S, which is an identity matrix. We consider two specifications. Firstly, for a full VAR system where all the variables are endogenous, p = 1, zt =(1, ∆GDP, ∆CP I, ∆OILP ) and S = I12 . Secondly, we apply the QP method to a bivariate VAR with two endogenous variables (∆GDP and ∆CP I inflation) and an exogenous variable (∆OILP ). Hence, our model is,

yt = υ +

p X i=1

Ai yt−i +

p X

Bi xt−i + ut

(4)

i=0

where yt = (∆GDP, ∆CP I)0 , xt = (∆OILP ) and S = I10 .9 The number of breaks has been selected following approximated critical values derived from response surface regressions.10 The results are quite robust with either of the two specifications: 3 breaks are found and their location from the global optimization is in 1979.4, 1991.2 and 2000.4 when the oil price variable is exogenous and in 1982.3, 1991.1 and 1999.3 when it is considered 9 10

We have chosen to impose 2 lags in accordance with several information criteria as in the full VAR system. The details of the results, which are not presented to save space, are available from the authors upon request.

Oil price shocks and the US economy

6

endogenous. Then, we test the causality in the different periods defined by the breaks. Results in Table I confirm the findings of the rolling procedure; there is endogeneity at 10% at the beginning of the sample due to the link between inflation and oil prices and a strong causality from economic growth to oil price in the last period.

2.3

Focusing on recent oil price episode

Bearing in mind the discussion in the literature about recent oil price shocks, we present more detailed figures for the last identified period that mainly covers the noughties and pinpoints a different pattern for these events. Results in Table II, considering the causality effect of each variable separately, support the previous finding of the effect of GDP growth on oil prices and provide a basis for ordering the variables properly, making the Cholesky decomposition and calculating the impulse response functions, which are displayed in Figure 3.11 We observe a strong positive and significant effect of GDP growth on oil prices while there is no significant influence of oil prices on the macroeconomic variables.12

3

Concluding remarks

This paper shows that recent episodes of oil price movements that occurred during the noughties have a different nature to previous shocks. During this period, there is a clear effect of economic activity on oil prices (and also on inflation), which is not identified in previous periods, while there is no evidence of any influence of the oil prices on either output growth or inflation. New causal relationships between macro-aggregates and oil prices call for a reexamination of the role of oil price shocks on macroeconomic outcomes. The large shocks of oil prices do not seem to be a potential cause of the recent rise in the volatility of the US economy in the light of these results.

4

Acknowledgements

Financial support from Ministerio de Ciencia e Innovaci´on under grant ECO2011-30260-C03-02 is gratefully acknowledged. The views expressed in this paper are the responsibility of the authors and do not necessarily represent those of the Banco de Espa˜ na or the Eurosystem. 11 12

The confidence intervals have been estimated by bootstrapping. Regarding the two macro variables of the system, we find bidirectional causality between them.

Oil price shocks and the US economy

7

References Barsky, R. B., and Kilian, L. (2002). “Do we really know that oil caused the great stagflation? a monetary alternative.” In B. S. Bernanke, and K. Rogoff (Eds.), NBER Macroeconomics Annual 2001, vol. 16, 137–83, Cambridge, MA: MIT Press. Clark, T. E. (2009). “Is the great moderation over? an empirical analysis.” Economic Review Federal Reserve Bank of Kansas City, Q IV, 5–42. Gomez-Loscos, A., Gadea, M. D., and Monta˜ nes, A. (2012). “Economic growth, inflation and oil shocks: are the 1970s coming back?” Applied Economics, 44 (35), 4575–4589. Hamilton, J. D. (1983). “Oil and the macroeconomy since world war ii.” Journal of Political Economy, 91 (2), 228–48. Hamilton, J. D. (2003). “What is an oil shock?” Journal of Econometrics, 113 (2), 363–398. Hamilton, J. D. (2008). “Oil and the macroeconomy.” In S. N. Durlauf, and L. E. Blume (Eds.), The New Palgrave Dictionary of Economics, Palgrave Macmillan. Hamilton, J. D. (2009). “Causes and consequences of the oil shock of 2007-08.” Brookings Papers on Economic Activity, 40 (1), 215–283. Hamilton, J. D. (2011). “Nonlinearlities and the macroeconomic effects of oil prices.” Macroeconomic Dynamics, 15, 364–378. Jimenez-Rodriguez, R., and Sanchez, M. (2010). “Oil-induced stagflation: a comparison across major g7 economies and shock episodes.” Applied Economic Letters, 17 (15), 1537–1541. Kilian, L. (2008). “The economic effects of energy price shocks.” Journal of Economic Literature, 46 (4), 871–909. Kilian, L. (2009). “Not all oil price shocks are alike: Disentangling demand and supply shocks in the crude oil market.” American Economic Review, 99 (3), 1053–69. Kilian, L., and Hicks, B. (2012). “Did unexpectedly strong economic growth cause the oil price shock of 2003–2008?” Journal of Forecasting, n/a–n/a. Qu, Z., and Perron, P. (2007). “Estimating and testing structural changes in multivariate regressions.” Econometrica, 75 (2), 459–502.

8

Oil price shocks and the US economy

Tables TABLE I Causality Granger test by periods Exogenous VAR Endogenous VAR Period Wald test Period Wald test 1970.1 − 1979.4 5.41 1970.1 − 1982.3 9.37 (0.2474)

1980.1 − 1991.2

7.69

(0.0524)

1982.4 − 1991.1

(0.1034)

1991.3 − 2000.4

7.60

1991.2 − 1999.3

(0.1072)

2001.1 − 2012.4

8.81

1.29 (0.8623)

2.27 (0.6855)

1999.4 − 2012.4

(0.0659)

13.16 (0.0105)

Note: p-values in brackets.

TABLE II Causality Granger test 1999.4-2012.4 GDP CPI OILP ALL GDP 6.54 1.12 9.48 (0.0380)

CPI OILP

(0.5726)

(0.0502)

9.48

1.19

11.72

(0.0087)

(0.5503)

(0.0195)

7.25

3.41

13.16

(0.0267)

(0.1817)

(0.0105)

Notes: Causality effects by rows; p-values in brackets.

Figures

−100 1970.1

0

100

200

1970.1

0

5

10

15

20

−5 1970.1

0

5

10

1976.1

1976.1

1976.1

1994.4

1994.4

1988.3

1994.4

Growth rate of oil price

1988.3

Inflation growth rate

1988.3

Figure 1. Year-on-year growth rates

1982.2

1982.2

1982.2

GDP growth rate

2001.1

2001.1

2001.1

2007.2

2007.2

2007.2

2012.4

2012.4

2012.4

Oil price shocks and the US economy 9

0 1979.4

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1985.4

1998.2

Figure 2. Rolling causality test

1992.1

2004.3

2010.4

2012.4

all gdp cpi

Oil price shocks and the US economy 10

10

15

10

15

10

15

5

10

15

Response of CPI to POIL impulse

5

Response of CPI to CPI impulse

5

Response of CPI to GDP impulse

20

20

20

−0.5

0

0.5

1

−40

−20

0

20

−20

−10

0

10

20

10

15

10

15

5

10

15

Response of POIL to POIL impulse

5

Response of POIL to CPI impulse

5

Response of POIL to GDP impulse

Figure 3. Impulse-response functions (2001.1-2012.3)

−0.02

−0.02 20

−0.01

−0.01 15

0

10

0.01

0

−1

0.01

5

20

−0.5

0

0.02

Response of GDP to POIL impulse

5

1 0.5

0.02

−2

−1

0

1

Response of GDP to CPI impulse

−0.5

20

−1 15

0

0

10

0.5

1

5

1

Response of GDP to GDP impulse 2

20

20

20

Oil price shocks and the US economy 11