The Effect of Oil Price, Demand and Supply Shocks:
A Comparative Study on Norway and Sweden
This paper assesses the impact of oil price, demand, and supply shocks on real GDP and unemployment in Norway and Sweden. A vector autoregression model is used to examine the impact of the shocks. A Cholesky decomposition is used to estimate the impulse response functions, in addition variance decompositions also are estimated. The response of real GDP to an oil price shock is positive in Norway, and negative in Sweden. However, both responses are not significant. The outcomes from the variance decompositions also indicate that oil price shocks only have a limited impact. Overall, unemployment responds stronger to the three shocks than real GDP.
Keywords: Oil price; VAR; impulse response function; variance decomposition JEL classification: Q30; Q32; Q43
Renate Meekenkamp Groningen, December 2007 University of Groningen
1. Introduction
In the economic literature oil price shocks receive much attention for their presumed macroeconomic effects. However, there is not much consensus on how oil price shocks affect the economy and the empirical findings are not conclusive. Several studies find that oil price shocks do not have an impact on the economy and that recessions cannot be explained by oil price shocks (Darby, 1982, and Lee, et al., 1995). However, other studies find that oil price shocks do affect the economy and see oil price shocks as a direct cause for past economic recessions (Hamilton, 1983, and Gisser and Goodwin, 1986).
From a theoretical point of view economists have given different explanations for the inverse relationship between oil price changes and economic variables (Brown, et al., 2003). The classic supply shock explanation, in which rising oil prices are a sign of scarcity of energy, is the most intuitive explanation. Oil is an important factor of production and when this factor becomes less available, output and labor productivity will be reduced (or their growth rate will be reduced). If consumers expect the rise in the oil price to be temporary, they could decide to save less or borrow more. This would cause a fall in real balances and a further increase in the price level. Another explanation is the influence of oil price shocks on other economic variables, as it transfers income from oil-importing countries to oil-exporting countries. The reduction in the domestic demand can be offset by a rise in export demand from the oil-exporting countries, but in net terms there will be a negative impact on the consumer demand for goods produced in the oil-importing countries. Furthermore, the real balance effect states that after an increase in oil prices there will be an increase in money demand, as people wish to rebalance their portfolios toward liquidity. Interest rates will be boosted when monetary authorities fail to meet the growing money demand, as real balances decrease when the price level rises without a corresponding increase in the money supply.
However, more explanations for this inverse relationship exist (Brown, et al., 2003).
This paper examines the differences in the effects of oil price shocks, and demand and supply shocks in explaining fluctuations in real GDP and the unemployment rate between an oil importing and an oil exporting country. In this paper Sweden is the oil importing country, and Norway the oil exporting country. The main motivation for using these two countries is that other economic aspects and underlying institutions of these two countries are similar, which facilitates a comparison between these two countries. Moreover, for these two countries a sufficient amount of data is available. I use the analysis of Bjornland (2000) as a starting
point. Bjornland examined the effects of demand, supply and oil price shocks in five countries. Norway also is included in her analysis. Bjornland uses a Structural Vector Autoregression (VAR) model to examine the effects of the shocks, and she bases the restrictions on economic theory.
In this paper a VAR model is used to identify the different shocks. Energy shocks can influence the economy in a complexity of ways and this motivates the use of a VAR model (Bjornland, 2000). In this paper the VAR model is identified through a Cholesky decomposition. The VAR model is applied to estimate impulse response functions and variance decompositions for real GDP and unemployment in order to determine how these two variables respond to the three types of shocks and what proportion of the variance of real GDP and unemployment can be explained by each variable.
This paper is a contribution to the existing literature, since it makes an explicit comparison between the responses in an oil-importing and an oil-exporting country. Furthermore, most analyses focus on the U.S. and in this analysis two European countries are considered. The focus in this paper also is on the effect of oil price shocks on unemployment, whereas in many other analyses unemployment is not included as a variable of interest.
The responses of real GDP and unemployment to an oil price shock differ between Norway and Sweden. As expected, the response of real GDP in Norway to an oil price shock is positive and real GDP in Sweden responds negatively to an oil price shock. However, these two responses are not significant. Unemployment in Norway responds more to an oil price shock than unemployment in Sweden. Overall, unemployment responds stronger to the three shocks in both countries than real GDP.
This paper is structured as follows. Chapter 2 contains a brief description of the economies of Norway and Sweden. The theoretical background is included in chapter 3, and chapter 4 contains the data description and derives the VAR model. In chapter 5 the results of the impulse response functions and variance decompositions are summarized. Finally, chapter 6 concludes.
2. Country description
This section briefly describes the economies of the two countries in this research. As mentioned in the introduction, I selected Norway and Sweden for this research, as both countries have similar economic aspects. The main difference between the two countries is the wealth of available natural resources of Norway.
2.1. Norway
Norway is one of the wealthiest countries in the world, where life expectancy and health conditions are high. Two main reasons for this high welfare level are the redundancy of natural resources, and the fact that Norway is situated close to the markets of the industrialized West-European countries. Norway has a highly industrialized, open- and export-oriented economy. Norway is not a member of the European Union, however it has signed a free-trade deal with the Union. Furthermore, the economy of Norway can be described as a mixed, capitalist market economy with an evident element of state interference.
The expansion of most part of the industry is led by private ownership, but some industrial activities (or the exploitation of) are owned by the government.
In 1996 Norway was the third largest oil exporting country of the world. Now, Norway is the second largest producer of oil and the fourth producer of gas in Europe. All production of oil is offshore, in the North Sea and the Norwegian Sea. The domestic demand for oil in Norway is in the hands of Statoil Norge AS (Norwegian state oil company), Hydro Texaco AS (possessed by Texaco Inc and Norsk Hydro) and six private corporations. Figure 1 shows the level of real GDP (seasonally adjusted) and the unemployment rate of Norway.1
2.2. Sweden
The Swedish economy is an open economy, and the presence of several large multinationals (Ikea, Ericsson, H&M) make that the economy is simultaneously dependent on and a beneficiary of the international business cycle. The main export products of Sweden are:
transport equipment and machines, Swedish vehicles, and chemistry products. Sweden is a member of the European Union, but decided against joining the common currency (Euro).
Sweden has rich natural supply of coniferous forests, water power, uranium, iron ore and other minerals. Nevertheless it lacks significant oil and coal deposits. In the far north are the
1 www.eubusiness.com
only iron ore mines which are still in production. Their production is mainly exported. Figure 2 shows the level of GDP (seasonally adjusted) and the unemployment rate of Sweden.2
Figure 1: Real GDP and unemployment in Norway
50000 100000 150000 200000 250000 300000 350000 400000
1980 1985 1990 1995 2000 2005
GDP (in NOK)
1 2 3 4 5 6 7
1980 1985 1990 1995 2000 2005
Unemployment (in %)
Figure 2: Real GDP and unemployment in Sweden
350000 400000 450000 500000 550000 600000 650000
1980 1985 1990 1995 2000 2005
GDP (in SEK)
1 2 3 4 5 6 7 8 9
1980 1985 1990 1995 2000 2005
Unemployment (in %)
3. Literature Review
Literature on the macroeconomics of oil price shocks had a prosperous history since the first oil price shock in the 1970s and economists found empirical evidence which suggests that oil
2 www.sweden.se
price shocks are closely related to macroeconomic performance. In this section an overview of relevant theoretical and empirical articles on the subject of oil price shocks is given. First, I provide an overview of the first empirical studies on oil price shocks, and then a review of previous results on the effect of oil price shocks on unemployment and GDP is given.
3.1. Overview of the first empirical literature on oil price shocks
After the first oil price shock in 1973 the empirical literature on the impact of the oil price developed. The authors who first examined and estimated the impact of oil price increases on real income in the U.S. and other developed economies were Darby (1982) and Hamilton (1983). Darby (1982) concludes that the increase in the oil price was not the main reason for the economic slowdown in the mid 1970s and early 1980s in the U.S. This result contradicts other authors which concluded that these slowdowns do originate in higher oil prices (Hamilton, 1983, and Gisser and Goodwin, 1986). Other studies argue that it were not the oil price shocks, but the monetary policy responses which caused the fluctuations in aggregate economic activity (Bohi, 1989, and Bernanke, et al., 1997).
Darby (1982) used a Lucas-Barro real income equation to test the significance of oil price variables for eight countries.3 He notes that the 1973 oil price shock was not the only occurrence; the world was also emerging from the international monetary arrangements established by the Bretton Woods agreement. Darby’s results are supportive of the hypothesis that oil price effects had no effect on real GNP. Limitations of his research are that oil price effects only can occur through the aggregate production function, while other effects, which operate through aggregate demand and employment, could also be present.
In contrast to Darby’s study, Hamilton (1983) indicates that oil price shocks were a contributing factor in almost all the post World War II recessions in the U.S. He found that the correlation between the oil price and economic performance in the U.S. was statistically significant for the period 1948-72. Hamilton assumed three possible explanations for this correlation, the first explanation is that this correlation is a historical coincidence. The second explanation for this correlation is that it is the result off an endogenous explanatory variable, the oil price increases and recessions are caused by some third set of influences. The final explanation is that at least some of the recessions were caused by an exogenous increase in the price of crude oil. For testing Hamilton used traditional tests to test for the absence of statistical correlation, Granger-causality tests, and institutional and historical evidence. He
3 This model was derived, following Barro (1973), by combining a standard Lucas (1973) aggregate supply function with nominal money, real government spending, and real exports as arguments.
concludes that there are few grounds for assuming that this correlation is a statistical coincidence, and he finds modest evidence for a third set of influences that could be responsible for both the oil price increases and the following recessions. Gisser and Goodwin (1986) support Hamilton’s findings for the U.S., and Burbidge and Harrison (1984) for the United Kingdom, Japan, and the U.S.
After the failure of the 1986 oil price decline to generate an economic boom, the idea emerged that there exists an asymmetric relation between oil price changes and economic performance.
As a reaction, Mork (1989) wrote an extension on Hamilton’s results. Mork’s analysis focused on the fact that during the period examined in Hamilton’s paper there were only large oil price increases, and therefore the question remained what the correlation would be in a period of oil price declines. Mork also adjusted the dataset of the oil prices. Hamilton used the Producer Price Index (PPI) for crude oil, whereas Mork used a composite Refiner Acquisition Cost (RAC). His results show that the effects of oil price declines differ significantly from oil price increases and could be zero. However, his results confirm that a negative correlation with oil price increases exists. Mork, Olsen and Mysen (1994) extended the analysis of Mork to other countries. They found that all the other countries, except Norway, experienced a negative relationship between oil price increases and GDP growth. The relation between the oil price and GDP growth is measured through correlation coefficients. In contrast to these analyses, other analyses obtained results which are not supportive of Hamiltons results (Lee, et al., 1995, and Hooker, 1996). In these papers the data sample is longer, and after the mid- 1980s oil prices typically fail to Granger cause macro variables.
3.2. Review of previous results
In this section a review of previous results of the response of unemployment and GDP to oil price shocks is given. It also is discussed which type of models and variables are used in these analyses.
3.2.1. Results on unemployment
First I will examine the results found in other analysis on the response of unemployment to oil price shocks. Not many papers focused explicitly on the response of unemployment to oil price shock. Davis and Haltiwanger (2001) examine the response of the labor market (job creation and destruction) to separately defined positive and negative oil price shocks, in this analysis VAR models are used. Two transmission channels are distinguished, namely an aggregate (potential output, income transfer, wage stickiness) and an allocative (closeness of
the match between actual and desired factor input levels across regions, sectors, and firms) channel. An oil price increase would increase job destruction and reduce creation through aggregate channels, and an oil price decrease would decrease job destruction and increase creation. In contrast, allocative channels would increase both creation and destruction, asymmetrically in response to price increases and decreases. The authors find that in nearly every industry oil price and monetary shocks cause larger responses in destruction than in creation. Keane and Prasad (1996) examined the effect of oil price shocks on unemployment and real wages with micro panel data. The use of panel data sets makes it possible to correct for compositional effects by constructing fixed-weight wage indices that hold fixed the efficiency units of labor per man-hour, and they use a fixed effect ordinary least square (OLS) model. Their result is that oil price increases cause real wages to decline (between 3% and 4%
in the long run) at the aggregate level and in almost every sector. However, aggregate unemployment in the long-run is not reduced, although oil price increases reduce wages.
Absolute wage cuts occur for workers at all skill levels, an exception is the relative wage of skilled workers, their relative wage increases after an oil price increase. Keane and Prasad find evidence that oil price changes affect employees with different experience levels differently. This empirical study also indicates that skilled labor may be a good substitute for energy in the production functions, as employment probabilities rise for skilled labor after an oil price increase.
3.3.2. Results on GDP
The literature on the response of GDP to oil price shocks is more extended than the literature on the response of unemployment. Several papers analyzed the effect of oil price shocks on GDP, and in these analyses different types of VAR models have been used. In this analysis the paper by Bjornland (2000) is used as a starting point. Bjornland analyzed the dynamic effects of aggregate demand, supply and oil price shocks on GDP and unemployment in four countries. In her analysis three variables are included, namely real GDP, the real oil price and the unemployment rate. A structural VAR model identifies the different shocks. The Structural VAR model is identified through nine short- and long-run restrictions. These restrictions are implied by an economic model. The advantage of an ordering based on an economic model is that the restrictions do not follow a recursive structure. VAR models in which the ordering is based on a recursive structure are very sensitive to how the identification was achieved (for example Ahmed, et al., 1988). Bjornland finds that for all countries, except Norway, oil price shocks have significant negative effects on output.
Cologni and Manera (2005) also use a Structural cointegrated VAR model to study the direct effects of oil price shocks on output and prices, and the reaction of monetary variables to external shocks for the G-7 countries. Seven variables are included in their analysis, namely the short-term interest rate (treasury bill or lending rate), monetary aggregate (M1), the consumer price index, real gross domestic product, the oil price and a variable for the exchange rate. Structural cointegrated VAR models have been used extensively to analyze the changes in GDP and to explain exchange rate anomalies (Dolado and Jimeno (1997) and Kim and Roubini (2000)). In the analysis of Cologni and Manera this type of model is preferable because it allows both long-run (cointegrated) and short-run (covariance) restrictions. The results found by Cologni and Manera indicate that unexpected oil price shocks have an impact on interest rates, suggesting a contractionary monetary policy response directed to fight inflation. The increases in interest rates are transmitted to the real economy by reducing output growth and the inflation rate. An unconstrained VAR model has been used by Burbidge and Harrison (1982) to examine the effect of oil price changes on the U.S. economy.
In their model seven variables are included, namely the oil price, a measure for industrial production in OECD countries, the yield on treasury bills, currency and demand deposits, average hourly wage, index of consumer prices and industrial production. The ordering of the time-series in the model is based on Sims (1980), and the oil price is ordered first. No economic theory is used for the ordering of the variables and the authors have not examined outcomes which result from a different ordering. The authors find that the price of oil is especially important in its impact on nominal wages.
Eltony (2001) estimates a VAR model, a VECM and a Structural VAR model using seven key macroeconomic variables for the state of Kuwait. The results from the VAR model can be less accurate because of the non-theoretical approach developed by Sims (1980) to decompose VAR residuals into orthogonal shocks implying the difficulty in grating these shocks structure interpretations (Eltony (2001)). Therefore Eltony also estimates a VECM and a Structural VAR. The variables included in the analysis are the oil price of Kuwaiti Blend Crude, oil revenue, government development expenditure and current expenditure, the consumer price index, money demand (M2) and the value of imports of goods and services. All three models indicate a high degree of interrelation between the macroeconomic variables. Furthermore, the models indicate the causality running from the oil prices and oil revenues, and government development and current expenditure, towards other variables. Government fiscal stimuli are the main determinant of domestic prices, while monetary stimuli have the least results. Eltony
concludes from these findings that fiscal policy can be used more effectively to stabilize the domestic economy after an oil shock. The three models lead to qualitatively similar results.
Multivariate VAR analysis using both linear and non-linear models is used in a paper by Rodríguez and Sanchez (2004). They assess the effects of oil price shocks on the real economic activity of the main industrialized countries. Seven variables are included in the analysis, namely real GDP, the real effective exchange rate, the real oil price, real wage, inflation, and short and long-term interest rates. The authors use Cholesky decomposition for the impulse response functions. Real GDP is ordered first, then the real oil price, followed by inflation, the short and long-term interest rate, real wages and finally the effective exchange rate. This ordering assumes that real GDP is not contemporaneously affected by shocks to the remaining variables. However, shocks to the first variable do affect the other variables in the system. Both oil-importing and exporting countries are included in the analysis, and Japan is the only oil-importing country which economic activity is not negatively affected by an oil price increase. The two oil-exporting countries are affected differently by an oil price increase. In the UK the effect is negative, and in Norway positive. In addition, evidence for a non-linear impact of oil prices on real GDP is found.
In these papers different types of VAR models have been used. A Structural VAR model is often preferred, since this type of model is identified by restrictions which are based on economic theory.
Table 1: Response GDP to oil price shock
Author Country Largest response in % Quarters
Bjornland Germany -0,3 6
Bjornland UK -0,4 6
Bjornland US -0,5 6
Bjornland Norway 0,4 8
Rodríguez and Sanchez US -3,9 8
Rodríguez and Sanchez Japan 1,7 8
Rodríguez and Sanchez Canada -0,8 6
Rodríguez and Sanchez France -1,5 8
Rodríguez and Sanchez Italy -2,2 6
Rodríguez and Sanchez Germany -1,8 8
Rodríguez and Sanchez UK -1,9 10
Rodríguez and Sanchez Norway 1,8 6
Cologni and Manera Canada 0,2 4
Cologni and Manera France 0,1 4
Cologni and Manera Germany 0,5 3
Cologni and Manera Italy -0,1 2
Cologni and Manera Japan 0,1 4
Cologni and Manera UK 0,1 3
Cologni and Manera US 0,2 3
An overview of the responses of GDP to oil price shocks found in the abovementioned papers is given in table 1. Bjornland finds that in all countries, except Norway, real GDP responds negatively in the second quarter. The responses lie around the -0.4%. In the analysis of Rodríguez and Sanchez the effect of oil price shocks on the GDP growth rate is examined, and they find that the GDP growth rate in Norway and Japan does not respond negatively to an oil price shock. The largest positive response of GDP to an oil price shock in Norway is 1.8%. Cologni and Manera find for all countries, except Italy, that real GDP responds positively to an oil price shock. This is in contrast to the results found in the other analyses.
4. Methodology
In this section I will describe the data and identify the VAR model. The use of a VAR model is motivated by the complexity of the channels through which energy can affect the economy.
VAR models are able to capture the dynamic structure of many time series variables.
4.1. Data description and model specifications
The data sample consists of real GDP (non-oil GDP for Norway), real oil prices and the unemployment rate.4 All data is obtained from DataStream, and for the oil price the Brent crude oil price is taken. For Norway non-oil GDP is taken, because then real GDP includes the same for both countries. Real GDP and the unemployment rate are given as quarterly data, however the oil price is stated in monthly prices. I use the average oil price of the three months per quarter to obtain a quarterly time series. The oil price series is stated in U.S.
dollars, and this price is converted to the national currency of the two countries.5 Data from the first quarter of 1980 until the second quarter of 2005 is used. Real GDP and the real oil price are entered as logarithms. A common problem with quarterly observed time series are the cyclical movements they often display. The time series of GDP are seasonally adjusted for both countries. For adjusting the moving average method is used, this method assumes that the seasonal factors are constant from year to year.
It is important to test the presence of unit roots in order to avoid the problem of spurious regression. If a variable contains a unit root (is non-stationary) and does not form a stationary
4 Data statistics are included in Appendix A
5 The exchange rate on November 24, 2005 was used for this conversion
cointegrated relationship with another non-stationary time series, then regressions using these series can incorrectly imply the existence of a meaningful economic relationship.
To test for a unit root in the underlying processes I use the Augmented Dickey-Fuller (ADF) test of a unit root against a stationary alternative. Three possibilities exist for the ADF test; a model with a constant, a model with a constant and trend and a model without a constant. The results from the three ADF tests are included in table 2, and the null hypothesis is that there is a unit root. The log of real GDP, the log of real oil price and unemployment are stationary integrated variables I(0) for both countries.
Table 2: Results ADF tests
Log GDP Log Oil Unemployment
Norway Sweden Norway Sweden Norway Sweden
Model with constant -3.70* -4.37* -6.56* -9.15* -5.53* -5.38*
Model with constant and
trend -4.82* -4.37* -6.53* -9.10* -5.60* -5.39*
Model without constant -3.37* -1.95*** -6.05* -8.78* -5.52* -5.39* Note: *denotes rejection of the null hypothesis at the 1% critical level, ** at the 5% critical level and *** at the 10% critical level
I also tested for co-integration between the variables with the Johanson co-integration test, since the VAR is estimated in logarithms (Sims, et al., 1990). The Johanson test examines whether there are long-run relationships among the variables. According to the ADF test the time series of the log of real GDP, the log of the oil price, and the unemployment rate are stationary. The results of the Johanson test are given in table 3. For Norway the null hypothesis r < 2 can be rejected at the 5% significance level, the trace statistic (7.029) exceeds the critical value (3.84). Also the max value of r = 2 exceeds the critical value, thus the null hypothesis can be rejected. I therefore conclude that there are three long-run relationships between the variables. For Sweden I find two long-run relationships. The null hypothesis r < 2 cannot be rejected, since the trace value (1.583) does not exceed the 5% critical value (3.84).
The max value of r = 2 does not exceed the critical value, so the null hypothesis cannot be rejected. The max value of r = 1 does exceed the critical value, thus the null hypothesis can be rejected.
The lag order of the VAR models is determined by the Akaike and Schwarz information criteria. Based on these two criteria I decide to use 4 lags for Norway and 2 lags for Sweden.
Both estimated VAR models are stable and stationary, according to the inverse roots of the characteristic AR polynomial.
Table 3: Results Johanson’s co-integration test
Norway Sweden
H0 H1 trace value
5% critical
value Prob. H0 H1 trace value
5% critical
value Prob.
r = 0 r > 0 59.91427* 29.80 0.0000 r = 0 r > 0 39.13391* 29.80 0.0001 r < 1 r > 1 21.36957* 15.49 0.0058 r < 1 r > 1 18.85601* 15.49 0.0150 r < 2 r > 2 7.029129* 3.84 0.0080 r < 2 r > 2 1.583215 3.84 0.2083
H0 H1 max value
5% critical
value Prob. H0 H1 max value
5% critical
value Prob.
r = 0 r = 1 38.54470* 21.13 0.0001 r = 0 r = 1 30.27792* 21.13 0.0001 r = 1 r = 2 14.34044* 14.26 0.0486 r = 1 r = 2 17.27279* 14.26 0.0162 r = 2 r = 3 7.029129* 3.84 0.0080 r = 2 r = 3 1.583215 3.84 0.2083
* indicates significance at the 5% level
4.2. VAR model
Sims (1980) developed the VAR model to address the interrelationships among time series.
All the endogenous and exogenous variables form a simultaneous system, and each endogenous variable in the system is a function of the lagged values of all the endogenous variables in the system, and
yt = A1 yt-1 + … + Ap yt-p + Bxt + εt (10)
is the mathematical representation of a VAR model. Here yt is a vector of endogenous variables, xt is a vector of exogenous variables, and A1, … , Ap and B are matrices of coefficients to be estimated, and εt is a vector of innovations. Since these variables may have an influence on each other, the VAR model is appropriate since it captures dynamic relationships better. Other advantages of the VAR model are that there is no simultaneity bias and OLS estimates are consistent, efficient, and equivalent to GLS estimates (Bjornland, 2000).
In this analysis impulse response functions for real GDP and unemployment are presented to show the responses of these two variables to the three types of shocks. Furthermore, variance decompositions are estimated to indicate what proportion of variance of real GDP and unemployment is explained by each variable. Variance decomposition separates the variation
in an endogenous variable into the component shocks to the VAR, it gives insights in the relative importance of the effects of each random innovation on the variables in the VAR.6
In this analysis three variables are the focus, namely, log real GDP, log real oil price, and unemployment. Which leads to the following VAR model for Norway and Sweden
+
+ +
=
−
−
−
−
−
−
t t t
p t
p t
p t
t t
t
t t
t
U OIL GDP
p p p
p p p
p p p
U OIL GDP
a a a
a a a
a a a
U OIL GDP
3 2 1
33 32 31
23 22 21
13 12 11
1 1
1
33 32 31
23 22 12
13 12 11
...
ε ε ε
(11)
here GDP is log real GDP, OIL is log real oil prices, U is unemployment, and a, p and c are coefficients to be estimated. εt is a coefficient for innovation which has to be estimated. p denotes the number of lags included in the VAR model, and p is 4 for Norway and 2 for Sweden.
In this paper I use a recursive Cholesky orthogonalization to obtain a non-recursive orthogonalization of the error terms. Three structural disturbances are investigated, namely oil price, aggregate demand and aggregate supply shocks. Since VAR models have a dynamic lag structure a shock to an individual variable does not only directly affect this variable, but this shock also is transmitted to the other endogenous variables in the VAR. In this paper the measurement of the effects of a shock to one of the variables on current and future values of the endogenous variables in the model is based on Cholesky one standard deviation shock to the residuals (innovations). Cholesky uses the inverse of the Cholesky factor of the residual covariance matrix to orthogonalize the impulses.7 This option imposes an ordering on the variables in the VAR model and attributes all of the effect of any common component to the variable that comes first in the VAR model. So, this ordering assumes that the variable that is ordered first is not contemporaneously affected by shocks to the remaining variables, but shocks to the first variable do affect the other variables in the system.
Real GDP is ordered first, which means that this variable is assumed to be driven by supply, demand and oil price shocks. This ordering assumes that real GDP does not react contemporaneously on impact to the rest of the variables (Rodríguez and Sanchez, 2004). The real oil price is ordered second and unemployment third. Unemployment is ordered last,
6 Eviews
7 Eviews
because I assume that unemployment responds slower than real GDP and the real oil price to the shocks. This leads to
=
U OIL GDP
OIL AS AD
u u u
e e e
a a a
a a a
1 0 0
0 1 0
0 0 1
0 0 0
33 23 22
13 12 11
(12)
here et are the observed residuals, and ut are the unobserved structural innovations. For the Cholesky decomposition the dof adjustment option is taken, this option makes a small sample degrees of freedom correction when estimating the residual covariance matrix used to derive the Cholesky factor.
A problem with the Cholesky ordering is that the world may not be recursive (Bessler and Akleman (1998)). As a robustness check I also impose the restrictions according to generalized impulses. Pesaran and Shin (1998) developed a generalized impulse response analysis for VAR models; this method avoids orthogonalization of shocks and generates order-invariant results. According to Doan (2002) the results from generalized impulse responses must be interpreted with caution because of the difficulty in interpreting impulses from highly correlated shocks within a non-orthogonalized setting. Doan also states that the generalized impulse response method is equivalent to computing shocks with each variable in turn being set atop a Cholesky ordering.
From the review in chapter 3 it became clear that a structural VAR model is preferred. In a structural VAR model the ordering is imposed by short- and long-run restrictions. These restrictions are based on economic theory, and an advantage of this approach is that the outcomes of the model can be interpreted on the basis of economic theory. However, a problem with this approach is that the true contemporaneous orderings that the author claims to know by assumption may be in fact unknown (Bessler and Akleman (1998)). In the current version of Eviews that I am using it is not possible to impose both short- and long-run restrictions simultaneously. Therefore I am not able to develop a SVAR model, and I only use the Cholesky decomposition.
5. Results
This section discusses the findings of VAR models. The impulse response functions are given in several figures, and the variance decompositions of real GDP and unemployment are summarized. The figures in this chapter show the impulse response functions with Cholesky decomposition, and a standard deviation band around the point estimate is included.8 The impulse response functions with generalized impulses are included in Appendix B. I begin by analyzing the responses of real GDP, then I analyze the responses of unemployment, and finally I examine the variance decompositions of real GDP and unemployment for both countries.
5.1. Responses of real GDP
In this section the responses of real GDP to the three shocks and the accumulated response of real GDP to an oil price shock are given. Figure 3 shows the dynamic effect of an oil price shock on real GDP with a standard deviation band around the point estimate. It is expected that an increase in the oil price has a positive effect on real GDP in Norway, since Norway exports oil. A negative effect on real GDP is expected in Sweden.
In Norway an adverse oil price shock (the real oil price increases) has a small positive effect, and this effect remains positive until it dies out in the fourteenth quarter. The adverse oil price shock initially has a positive effect on real GDP in Sweden.
Then in the third quarter the effect becomes negative, and the effect dies out in the twelfth quarter. According to the transfer income explanation it is possible that an oil price shock has a positive effect on real GDP besides a negative effect in an oil-importing country, since export demand from an oil-exporting country can increase.9 In Norway the effect of an adverse oil price shock is longer present than in Sweden. In both Norway and Sweden these effects are small and the standard deviation bands indicate that the responses are not significant (bands become wide).
In Appendix B the results from the impulse response functions with generalized impulses are included. Differences between generalized impulses and the Cholesky ordering are that in
8 Standard errors are calculated as analytic response standard errors. The impulse responses are significant at the 95-percent confidence level when both standard error bands are simultaneously above or below zero on the y- axis (Eviews).
9 See introduction
Norway the effect becomes zero in the second quarter and thereafter positive again, and for Sweden the effect in the third quarter is more negative.10
Figure 3: Dynamic effects of oil price shock on GDP, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
Figure 4: Dynamic effects of a demand shock on real GDP, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
The response of real GDP to a demand shock starts in both countries above zero, and then becomes negative in the second quarter (figure 4). This negative response is larger in Norway.
In both countries the effect of a demand shock is negative in the second quarter. The effect of the demand shock fluctuates in both countries, and in Norway the effect of the demand shock dies out quicker than in Sweden.
10 Response of real GDP to an oil price shock in Norway also is not significant with generalized impulses.
a
-.006 -.004 -.002 .000 .002 .004 .006 .008
2 4 6 8 10 12 14 16 18 20 22 24
b
-.006 -.004 -.002 .000 .002 .004 .006 .008
2 4 6 8 10 12 14 16 18 20 22 24
a
-.02 -.01 .00 .01 .02 .03
2 4 6 8 10 12 14 16 18 20 22 24
b
-.02 -.01 .00 .01 .02 .03
2 4 6 8 10 12 14 16 18 20 22 24
In contrast to the demand shock, a supply shock has a negative effect on real GDP in both countries (figure 5). However, in Sweden the effect of the supply shock is positive in the second and fifth quarter. The response of real GDP in Norway is not significant as indicated by the widening standard deviation bands. Bjornland (2000) finds positive responses of real GDP to supply shocks for all countries included in her analysis. The responses of real GDP with generalized impulses are for Norway similar to those obtained with Cholesky. For Sweden the responses to a supply shock differ to some extent. The effect of a supply shock starts negative in the first quarter compared to positive in the case of Cholesky, also the effect of a supply shock is less negative in the third quarter (Appendix B).
Figure 5: Dynamic effects of a supply shock on real GDP, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
a
-.010 -.008 -.006 -.004 -.002 .000 .002 .004 .006
2 4 6 8 10 12 14 16 18 20 22 24
b
-.008 -.004 .000 .004
2 4 6 8 10 12 14 16 18 20 22 24
Figure 6: Accumulated dynamic effects of an oil price shock on real GDP, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
a
-.005 .000 .005 .010 .015
2 4 6 8 10 12 14 16 18 20 22 24
b
-.004 .000 .004 .008 .012
2 4 6 8 10 12 14 16 18 20 22 24
In figure 6 the accumulated dynamic effects of an oil price shock are given for both countries.
The accumulated impulse response function is the cumulative sum of the impulse response function (Eviews). For Norway it is found that in the long-run the effect of an oil price shock is approximately a 0.5% increase in real GDP, and in Sweden the long-run effect of an oil price shock on real GDP is zero.
5.1.1. Summary
In Norway the response of real GDP to a demand shock is stronger than in Sweden, and the response of real GDP to a supply shock is only significant in Sweden (and this effect also is stronger). The response of real GDP to an oil price shock is insignificant in both countries, and the responses also are small. I expected to find stronger en significant responses of real GDP to oil price shocks in Norway. In table 3 an overview of the largest responses of real GDP to an oil price shock found in this analysis is given. The responses found in my analysis are not significant, however I want to compare the largest response to those found in other analyses. For Norway I find that the largest response is 0.1% and this is similar to the responses found in the other analyses (these responses are included in table 1). None of the other analyses has included Sweden, therefore it is not possible to compare this response. The other analyses found in general that this largest response is around the eight quarter, whereas I find that this is around the fourth quarter. The accumulated responses of real GDP indicate that only in Norway an oil price shock has an effect in the long-run, again this effect is small.
In Sweden the effect is zero in the long-run.
Table 3: Overview largest response of real GDP to oil price shock
Country Largest response (in %) Quarters
Norway 0.1 4
Sweden -0.2 3
5.2. Unemployment responses to shocks
In this section the responses of the unemployment rate to the three shocks and the accumulated response of the unemployment rate to an oil price shock are given. All the responses of the unemployment rate to the three shocks start above zero, only the response to a demand shock in Sweden starts below zero. An explanation for this is that the unemployment rate is ordered last in the Cholesky ordering.
The response of the unemployment rate to an oil price shock is given in figure 7. The largest response of the unemployment rate is negative in both countries. In Norway the response is largest in the fifth quarter, -1%, and in Sweden the response is largest in the third quarter.
Figure 7: Dynamic effects of oil price shock on unemployment, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
Thereafter the negative effect becomes smaller, and even becomes positive in the ninth and twelfth quarter. In Sweden the response dies out quicker than in Norway (respectively, the eight and the fourteenth quarter). The response of unemployment differs with generalized impulses for Sweden, and for Norway the response is identical (Appendix B). The response of unemployment to an oil price shock in Sweden is more negative with generalized impulses.11
Figure 8: Dynamic effects of a demand shock on unemployment, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
11 Response of unemployment in Sweden to an oil price shock with generalized impulses is not significant.
a
-.04 -.03 -.02 -.01 .00 .01 .02 .03 .04
2 4 6 8 10 12 14 16 18 20 22 24
b
-.04 -.03 -.02 -.01 .00 .01 .02 .03 .04
2 4 6 8 10 12 14 16 18 20 22 24
a
-.05 -.04 -.03 -.02 -.01 .00 .01 .02 .03
2 4 6 8 10 12 14 16 18 20 22 24
b
-.05 -.04 -.03 -.02 -.01 .00 .01 .02 .03
2 4 6 8 10 12 14 16 18 20 22 24
An aggregate demand shock has an initial negative effect on unemployment in both countries (figure 8). In Sweden the response begins negative and dies out in the fourteenth quarter. The effect of a demand shock on unemployment becomes positive after the third quarter in Norway, and dies out in the tenth quarter.
In both countries the dynamic effects of an aggregate supply shock are positive (figure 9). In Sweden this effect remains longer positive, the effect dies out around the sixteenth quarter compared to the tenth quarter for Norway. In Norway the response fluctuates more, and in the second quarter the effect is zero.
Figure 9: Dynamic effects of a supply shock on unemployment, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
a
-.04 -.02 .00 .02 .04 .06 .08 .10 .12
2 4 6 8 10 12 14 16 18 20 22 24
b
-.04 .00 .04 .08 .12
2 4 6 8 10 12 14 16 18 20 22 24
Figure 10: Accumulated dynamic effects of an oil price shock on unemployment, One Standard Error Band (dotted lines), for (a) Norway, and (b) Sweden
a
-.08 -.04 .00 .04 .08
2 4 6 8 10 12 14 16 18 20 22 24
b
-.08 -.04 .00 .04 .08
2 4 6 8 10 12 14 16 18 20 22 24
Figure 10 gives the accumulated dynamic effects of an oil price shock for Norway and Sweden. In Norway the response of the unemployment rate to an oil price shock is positive in the long-run, however this effect is small. In the long-run the response of the unemployment
rate is negative in Sweden, this effect also is small. Keane and Prasad (1996) also find that unemployment is not reduced as a result of an inverse oil price shock in the long-run.
5.2.1. Summary
As mentioned before, all the responses of the unemployment rate to the three shocks start at non-zero, and an explanation for this is that the unemployment rate is ordered last. When comparing the responses of unemployment of the two countries, the response to an oil price shock and a supply shock are larger in Norway. In the long-run a positive effect is found in Norway, and a negative effect in Sweden to an oil price shock. However, both long-run effects are small. Comparing the responses of the unemployment rate with the responses of real GDP, unemployment has stronger response to the three shocks than real GDP in both countries.
5.3. Variance Decomposition
This section analyses the results of the variance decompositions. Table 4 and 5 present the results for the variance decompositions for real GDP and unemployment in Norway and Sweden, for the first, fourth, eighth and twenty-fourth quarter.
Table 4: Variance Decomposition of real GDP for Norway and Sweden in percentages, (AD shock is an aggregate demand shock, and AS shock is an aggregate supply shock)
Norway Sweden
Quarter AD Shock AS Shock
Oil
Shock AD Shock AS Shock
Oil Shock
1 95.6 4.2 0.2 99.4 0.5 0.1
4 94.7 4.4 0.9 99.0 0.6 0.4
8 94.1 4.8 1.1 98.9 0.6 0.5
24 94.0 4.8 1.2 98.9 0.6 0.5
An oil price shock only has a small effect on output, both in Norway and Sweden (table 4). In Norway this effect increases in some extent, while in Sweden it remains rather constant at 0.5%. So oil price shocks only explain a small part of the fluctuations of real GDP in both countries. In both countries aggregate demand shocks are the most important source of output
fluctuations.12 Aggregate supply shocks have a stronger effect on output fluctuations in Norway than they have in Sweden. In the short-run around 4% of the output fluctuations can be explained by the effect of aggregate supply shocks, increasing to 5% in the long-run.
Table 5: Variance Decomposition of Unemployment for Norway and Sweden in percentages, (AD shock is an aggregate demand shock, and AS shock is an aggregate supply shock)
Norway Sweden
Quarter AD Shock AS Shock
Oil
Shock AD Shock AS Shock
Oil Shock
1 14.9 84.8 0.3 19.2 80.4 0.4
4 26.9 71.9 1.2 27.5 72.2 0.3
8 31.8 64.3 3.9 29.1 70.7 0.2
24 32.8 61.0 6.2 29.9 69.9 0.2
Oil price shocks have more effect on the unemployment rate fluctuations in Norway, around 6% after six years (table 5). In Sweden the effect of oil price shocks on unemployment rate fluctuations is minimal. In Norway aggregate demand shocks effect unemployment fluctuations with approximately 20% in the short-run, and 30% in the long-run. Aggregate demand shocks explain in a similar manner the unemployment fluctuations in Sweden, in the short-run 20% and increasing to 30% in the long-run. The largest effect on unemployment is by aggregate supply shocks. In Sweden this effect is relatively larger in the long-run, 70%
compared to 60% in Norway.13
6. Conclusion
This is the concluding section and here I will discuss the results of the analysis. In this paper I examined the effects of different shocks on the fluctuations in real GDP and unemployment in Norway and Sweden. Oil price, demand and supply shocks were included in the analysis. The focus is on the differences in the effects between Norway (an oil-exporting), and Sweden (oil- importing).
12 These results are robust to extending the period to forty or sixty quarters.
13 These results are robust to extending the period to forty or sixty quarters.
The effect of an oil price disturbance differs between an oil-exporting and an oil-importing country. As expected, real GDP in Norway responds positively and in Sweden negatively.
However, in both countries the responses are not significant. In Sweden the largest response is negative, but in two periods the response is positive, however this response also is not significant. This can be explained by the fact that export demand from an oil exporting country can increase, and this can increase real GDP. In the long-run a positive effect of an oil price shock on real GDP is found in Norway, and in Sweden the long-run effect is zero. The effect of an oil price shock on unemployment is stronger in Norway, with an effect of -1% in the fifth quarter. In both countries the effect of an oil price shock on unemployment in the long-run is approximately zero.
In Norway and Sweden the responses to the other two disturbances are similar, however these responses are not always significant. An explanation for this result could be that the economies of these two countries are comparable. Comparing the responses of real GDP with the responses of the unemployment rate, unemployment responds stronger to the three shocks than real GDP in both countries.
The variance decompositions confirm that an oil price shock only explains a relative small part of the fluctuations in real GDP and unemployment. Demand shocks explain a large part of the fluctuations in real GDP, while supply shocks explain a larger part of the fluctuations in unemployment. In Norway oil price shocks explain a larger part after 24 quarters, and in both countries the explanatory power of demand shocks increases.
Possible extensions and further research could lie in the inclusion of more non-U.S. countries in the analysis, and the examination on how these economies respond to the increases in the oil price. In 2007 the oil price reached approximately $96 per barrel, and the discussion on the oil reserves remains. In the coming years it is possible to examine the consequences and effects of this oil price increase. It is also interesting to examine if there is an optimal monetary policy for managing oil price increases. Several studies already examined the possibility that monetary policy responses affect the fluctuations of the aggregate economy (Bernanke, et al., 1997, and Hunt, et al., 2002), however no optimal policy is given.
Reference List
Ahmed, E., Rosser, J., and Sheehan, R. (1988), ‘A global model of OECD Aggregate Supply and Demand using Vector Autoregressive Techniques’, European Economic Review, Vol. 32, p. 1711-1729.
Bernanke, B.S., Gertler, M., and Watson, M. (1997), ‘Systematic Monetary Policy and the Effects of Oil Price Shocks’, Brookings Papers on Economic Activity, Vol. 1, p. 91-157.
Bessler, D., and Akleman, D. (1998), ‘Farm Prices, Retail Prices, and Directed Graphs:
Results for Pork and Beef’, American Journal of Agricultural Economics, Vol. 80, p. 1144- 1149.
Bjornland, H.C. (2000), ‘The Dynamic Effects of Aggregate Demand, Supply and Oil Price Shocks – A Comparative Study’, The Manchester School, Vol. 68, p. 578-607.
Bohi, D.R. (1989), ‘Energy Price Shocks and Macroeconomic Performance’, Washington, D.C.: Resources for the Future.
Brown, S.P.A., Yücel, M.K., and Thompson, J. (2003), ‘Business Cycles: The Role of Energy Prices’. Working Paper, Federal Reserve Bank of Dallas.
Burbidge, J., and Harrison, A. (1984), ‘Testing for the Effects of Oil-Price Rises Using Vector Autoregression’, International Economic Review, Vol. 25, p. 459-484.
Cologni, A., Manera, M. (2005), ‘Oil prices, Inflation and Interest Rates in a Structural Cointegrated VAR model for the G-7 Countries’, Working Paper, International Energy Markets.
Darby, M.R. (1982), ‘The price of Oil and World Inflation and Recession’, The American Economic Review, Vol. 72, p. 738-751.
Davis, S.J., and Haltiwanger, J., (2001), ‘Sectoral Job Creation and Destruction Response to Oil Price Changes’, Journal of Monetary Economics, Vol. 48, p. 465-512.
Doan, T. (2002), ‘Generalized Impulse Responses’, The RATSletter, Vol. 15, p. 4.
Dolado, J., and Jimeno, J. (1997), ‘The causes of Spanish unemployment: A Structural VAR approach’, European Economic Review, Vol. 41, p. 1281-1371.
Finn, M.G. (2000), ‘Perfect Competition and the Effects of Energy Price Increases on Economic Activity’, Journal of Money, Credit and Banking, Vol. 32, p. 400-416.
Gisser, M., and Goodwin, T.H. (1986), ‘Crude oil and the Macroeconomy: Tests of Some Popular Notions’, Journal of Money, Credit, and Banking, Vol. 18, p. 95-103.
Hamilton, J.D. (1983), ‘Oil and the Macroeconomy since World War II’, The Journal of Political Economy, Vol. 91, p. 228-248.
Hooker, M.A. (1996), ‘What Happened to the Oil Price-Macroeconomy Relationship?’, Journal of Monetary Economics, Vol. 38, p. 195-213.
Hunt, B., Isard, P., and Laxton, D. (2002), ‘The Macroeconomic Effects of Higher Oil Prices’, National Institute Economic Review, Vol. 179, p. 87-103.
Jones, D.W., Leiby, P.N., and Paik, I.K. (2004), ‘Oil Price Shocks and the Macroeconomy:
What Has Been Learned Since 1996’, The Energy Journal, Vol. 25, p. 1-32.
Keane, M.P., Prasad, E.S. (1996), ‘The Employment and Wage Effects of Oil Price Changes:
A Sectoral Analysis’, Review of Economics and Statistics, Vol. 78, p. 389-399.
Kim, S., and Roubini, N. (2000), ‘Exchange rate anomalies in the industrial countries: A solution with a Structural VAR approach’, Journal of Monetary Economics, Vol. 45, p. 561- 586.
Lee, K., Ni, S., and Ratti, R.A. (1995), ‘Oil Shocks and the Macroeconomy: The Role of Price Variability’, Energy Journal, Vol. 16, p. 39-56.
Mork, K.A. (1989), ‘Oil and the Marcroeconomy When Prices Go Up and Down: An extension of Hamilton’s Results’, The Journal of Political Economy, Vol. 97, p. 740-744.
Mork, K.A., Olsen, O., and Mysen, H.T. (1994), ‘Macroeconomic Responses to Oil Price Increases and Decreases in Seven OECD Countries’, Energy Journal, Vol. 15, p. 19-35.
Eltony, M. (2001), ‘Oil price fluctuations and their impact on the Macroeconomic Variables of Kuwait: A Case Study Using VAR Model for Kuwait’, International Journal of Energy Research, Vol. 25, p.939-959.
Pesaran, H., and Shin, Y. (1997), ‘Generalized impulse response analysis in linear multivariate models’, Economic Letters, Vol. 58, p. 17-29.
Rodríguez, R., and Sanchez, M. (2004), ‘Oil price shocks and real GDP growth, empirical evidence for some OECD countries’, Working Paper, European Central Bank.
Sims, C. (1980), ‘Macroeconomics and Reality’, Econometrica, Vol. 48, p. 1-48.
Sims, C., Stock, H., Watson, M. (1990), ‘Inference in Linear Time Series Models with some Unit Roots’, Econometrica, Vol. 58, No. 1, p. 113-144.
Sims, C. (2002), ‘Structural VAR’s’, Time Series Econometrics, (Fall 2002), Econ. 513.
