Asymmetric pass-through of oil prices : an econometric analysis of the effects of oil price changes on inflation in the European Union

(1)

Asymmetric Pass-Through of Oil Prices

An Econometric Analysis of the Effects of Oil Price Changes on Inflation in the European Union

By Hans Mulder

August 2014

Presented to the Faculty of Economics and Business University of Amsterdam

In Partial Fulfillment of the Requirements for the Degree of Master of Science in Economics

Supervisor: Dr. D.J.M. Veestraeten Co-reader: Dr. M. Micevska Scharf

Faculty of Economics and Business University of Amsterdam

(2)

– 2 –

ABSTRACT

This paper examines the effect of changes in the oil price on inflation in the EU and investigates whether there exists asymmetry in pass-through between member countries. A dataset for the 1993-2012 period is constructed which includes macroeconomic measures of economic growth, unemployment, the exchange rate, and the energy intensity of a EU member state economy. Results show that an increase in the nominal oil price leads to an increase in inflation across the EU. This significant positive relationship is also found when controlling for energy intensity. Further, Central and Eastern European Countries (CEECs) experience a significantly stronger increase in inflation following an oil price change than non-CEECs.

(3)

– 3 –

ACKNOWLEDGEMENTS

When I decided to study Economics I had to submerge myself in the world of mathematics. It was September 2011 and I had just started a Master in International Relations at the University of Amsterdam. Pursing a double degree was an ambitious undertaking, which went hand-in-hand with passing one course but sometimes failing another because I could not give it my undivided attention. Four years of studying at the University of Amsterdam now come to a close. It was a great time in my life during which I met a lot of people and fully experienced student life. I also had the opportunity to develop myself personally through a full-time board year at the Financial Study Association Amsterdam. This paper marks the end of my Economics master curriculum. The thesis topic relates strongly to the field of International Economics, and combines my interests in macro-economics and econometrics. Specifically, this research focuses on the effects of changes in the world oil price on inflation in the European Union (EU). Completing this work would not have been possible without the advice and help of different people.

First, I would like to express my gratitude to Dr. D.J.M. Veestraeten for supervising my thesis. I have experienced his guidance throughout the course of the thesis as very constructive, providing me with helpful insights on estimation issues and scope of research. Second, I thank fellow student Maarten Tellegen, who followed the same path as I did, and with whom I shared many discussions on the study material over the years. Also, I would like to thank my parents who have always supported me and the choices I have made. Finally, I want to thank my lovely girlfriend Dominique, for everything.

This paper is written in the context of partial fulfilment of the requirements for the degree of Master in Economics at the University of Amsterdam.

(4)

– 4 –

1. Introduction

By the end of the 1960s, the European Economic Community (EEC) acknowledged the need to protect its member states from oil supply disruptions. Over the past decades, legislation on this issue has developed into a comprehensive set of rules and has strengthened the cooperation between member states and the European Commission with respect to EU energy security policy.1_{For example, member states are required to}

continuously hold an amount of petroleum products equal to at least 90 days of the average daily internal consumption during the previous year.2_{These strategic}

stock-holding requirements enable member states to readily access these petroleum products in case of supply shortages, where responsibility lies with the respective member state government to allocate resources to those sectors that are most affected. This coordinated EU approach to energy security reflects the key role of oil in the EU economy. Oil is important through its direct role in private consumption (i.e. transportation and heating fuel) and from its indirect role as a production factor for firms. Figure 1 shows energy consumption in the EU between 1965 and 2012. Clearly, crude oil has been the most important source of fuel over the past decades. According to the U.S. Energy Information Administration (EIA), oil will remain the largest energy source through 2040, albeit its share in the total energy mix will go down. This development is visible in figure 2.

Source: BP Statistical Review of World Energy 2013 Source: Energy outlook 2013 – EIA

1_{Council Directive 2006/67/EC} 2_{Council Directive 68/414/EEC}

Crude oil Coal Natural gas Nuclear energy Mi lli on s of to nn es Qu ad ril lio n B tu Crude oil

Coal Natural gas

Renewables

(9)

– 9 – 0 2 4 6 8 10 12 14 16 18 1965 1975 1985 1995 2005 0 5 10 15 20 25 30 1965 1975 1985 1995 2005

Due to a lack of large natural endowments, the energy consumption of the EU member countries is largely dependent on suppliers outside the EU for various industrial fuel categories (figure 2).3_{Because of its position as a net importer, the EU has paid much}

attention to its energy security. Between 2006 and 2010, oil imports accounted for 36% of EU energy consumption whereas oil production in the EU declined by 42% between 1999 and 2009.4_{Today, the EU imports crude oil mainly from African countries, the}

Middle East and Former Soviet Union (FSU) states, including the Russian Federation. Intra-EU production is largely attributable to Norway, United Kingdom (UK) and Denmark.5_{Figure 3 shows the difference in consumption and production of crude oil in}

the EU since 1965 (panel A) and the position of the EU as a producer of crude oil relative to other regions in the world (panel B).

FIG 3: EU and global production of oil

Panel A: EU consumption and production Panel B: Production by region

Source: BP Statistical Review of World Energy 2013

The dependence of the EU on oil imports exposes member states to external shocks, such as oil price increases or supply disruptions due to political uncertainty in producer regions. An oil price change can therefore have an effect on the macroeconomy. One macroeconomic variable that may be affected following an oil price change is inflation. Intuitively, the argument that oil price changes affect inflation seems appealing if we assume flexible prices. If the price of oil goes up, so must the consumer price level since

3_{Between 2006 and 2010, 54% of energy consumed in the EU came from imports, which was}

significantly higher than a decade earlier (in 1999 it was 45%).

4_{See European Economy. Member States’ Energy Dependence: An Indicator-Based Assessment. Occasional Papers}

145, April 2013.

5_{Based on 2013 figures by the European Commission Oil Bulletin, FSU countries accounted for 39.68%}

of imports, African nations 24.05%, Europe 18.76%, Middle East 12.89%.

Consumption Production Th ou sa nd s o f b ar re ls da

ily Middle East

N-America FSU/Russia EU Africa Asia Th ou sa nd s o f b ar re ls da ily

(10)

– 10 –

it becomes costlier for firms to produce. More directly, gasoline prices and heating oil may rise. The effects of an oil price change may, ceteris paribus, not be equal across countries. Especially in the context of the EU, which consists of a heterogeneous group of member states, one can expect that the effects oil price changes on inflation are different between countries. For example, there are likely differences in oil production and consumption that could make one country more prone to an oil price change than others (see Égert et al. 2010). One way to approximate the exposure of a country to oil price changes is to look at the energy intensity of the economy, measured as the ratio of total primary energy consumption to GDP (Sun, 2003). For example, among EU countries in 2010, the member states with higher degrees of energy intensity were Eastern European countries, including the Czech Republic, Poland, Romania and Bulgaria. By comparison, Western and Southern European countries, including the Netherlands, the UK, France, Spain and Italy, among others, have lower degrees of energy intensity.6_{These differences may lead to different responses in inflation following}

an oil price change.

The majority of studies on the oil price-inflation relationship consider U.S. data. A number of papers focus specifically on a subset of EU member states or the Eurozone, including the work of Blanchard and Galí (2007), de Gregorio et al. (2007), Chen (2009), LeBlanc and Chinn (2004), Meyler (2009), Égert et al. (2010), Venditti (2013), Galeotti et al. (2013), Jacquinot et al. (2009) and Cologni and Manera (2008). However, there is only a small amount of research that looks at this relationship in the context of the EU as a whole. This paper analyses quantitatively what the effects of oil price changes on inflation are in the current 28 EU member states. Further, it attempts to measure whether the effects of oil price changes lead to asymmetric responses in inflation for CEECs compared to non-CEECs. Throughout this thesis the term ‘asymmetric pass-through’ is defined as ‘uneven pass-pass-through’. This means that inflation rates in EU member states simply respond differently to an oil price change. For example, inflation rates may rise faster when the economy is more energy intensive. Understanding asymmetry in pass-through of oil price changes may help EU authorities with optimal energy policy-making.

(11)

– 11 –

Therefore, the central question in this paper is: do oil price changes have an effect on inflation

in the EU? And if so, to what extent does the pass-through differ between member states?

Panel data regression is applied for a sample consisting of the EU member countries, for the time period between 1993 and 2012. First, the effects of oil price changes on inflation are measured for the entire EU. Second, we examine the differences in oil price pass-through into inflation between economies that are more energy intensive and those that are less energy intensive. As the energy intensity and inflation level is substantially higher in Central and Eastern European Countries (CEECs) compared to non-CEECs, we examine whether the effects of oil price changes on in inflation are larger for CEECs.7

The empirical results show that an oil price increase positively impacts inflation in the EU, even when we control for the energy intensity level of the economy. Also, CEECs experience a greater rise in inflation after an oil price change than non-CEECs. This suggests that oil price changes asymmetrically pass-through across member states.

The paper is structured as follows. Chapter 2 provides a descriptive background on how oil prices have moved over the past decades and discusses inflation in the EU. This is followed by an overview of the academic literature on the oil price-inflation relationship in chapter 3. Subsequently, the empirical model is discussed. It starts with a brief discussion on the data and sample, as well as on the variables used in the models. This is followed by an overview of the models including a number of comments on the applied regression technique. Then, the regression results are discussed, followed by a discussion of the validity of the results. Finally, chapter 5 presents the conclusions of this paper and briefly discusses a number of avenues for future research.

7_{CEEC countries include Bulgaria, Czech Republic, Estonia, Croatia, Latvia, Lithuania, Hungary, Slovakia, Poland,}

(12)

– 12 –

2. Oil Prices in the European Union

The Oil Shocks of the 1970s and the global recessions that followed have spurred interest in the relationship between oil prices and economic cycles. Economists in the 1980s analyzed the link between oil prices and the macroeconomy in various econometric studies, demonstrating a negative correlation for the US and other industrial countries (Segal, 2011). Before focusing on inflationary effects of oil price changes we need to develop an understanding of how oil prices move. This chapter gives a brief history of oil price movements. After that we discuss both the role of oil and inflation in an EU context.

2.1 Oil price developments

To analyze the movement of international oil prices it seems reasonable to focus on the post-1945 period. Before the Second World War the global oil industry was in its infancy because of its relatively low level of technology. The share of oil in energy consumption was small (compared to coal), suggesting only marginal impacts on the balance of payment of oil importing countries (Yan, 2012). After the Second World War, however, construction activity went up in major economies. By 1967, the proportion of oil in the energy consumption mix surpassed coal and reached over forty percent (Yan, 2012). Against this background a number of oil-exporting countries, primarily in the Middle East, formed the Organization of Petroleum Exporting Countries (OPEC), an intergovernmental organization to protect the interests of oil producing countries in the Middle East and Africa, in particular. At that time, Western transnational oil companies (also grouped together as “the Seven Sisters”) were still “vertically” dominating the oil industry by controlling supply, transportation, refining and oil pricing. In this way they were able to keep the oil price stable (Spero and Hart, 2010). This began to change as oil-producing countries were increasingly exerting influence on the international crude oil market and various petroleum companies in OPEC countries were nationalized by their respective governments. These developments led to a gradual loss of control over oil resources by Western oil companies (Yan, 2012).

(13)

– 13 – 0 20 40 60 80 100 120 1920 1935 1950 1965 1980 1995 2010 -100 -50 0 50 100 150 200 250 1920 1935 1950 1965 1980 1995 2010

During the 1970s, a several decades-long period of stable oil prices came to an end. The 1973 Oil Shock, following the Yom-Kippur War between Arab states and Israel, and the 1979-1980 Iranian Revolution are widely believed to have caused a steep increase in the price of oil (Wirl, 2008). However, between 1986 and 1997 oil price volatility went down, although the First Gulf War and the collapse of the Soviet Union in 1990-1991 coincided with a sharp increase in the oil price. Several years later, during the 2000s, the so-called “Oil Bubble” took place, which reflected a period of rapidly increasing oil prices. The Oil Bubble showed that movements of oil prices can be hard to predict. According to Maugeri (2009) the oil market is complex, involving a wide set of interacting market players, complicated interactions between crude oil and its derivative products and unique geopolitical pressures that affect the industry. Further, reliable data on the oil market is scarce and often estimates are used rather than facts (see also Sornette et al., 2009). To get a sense of how the oil price moves, figure 4 provides an overview of the oil price level since 1920.8_{It clearly indicates the rise in the oil price in}

the 1970s and the upswing during the 2000s. Figure 5 exhibits the volatility of the oil price and shows that during the 1970-1975 period the oil price became more volatile. In fact, oil prices are said to be one of the most volatile commodity prices in the post-1973 period (Regnier, 2007). This unpredictable behavior of the oil price led to a vast increase in empirical studies on the effects of oil prices on the economy and has led researchers to focus on the channels through which these effects play out.

8

The BP Statistical Review of Energy 2013 has published these data using a US average of oil prices between 1920-1944, the Arabian light crude oil price between 1945-1983, and the U.S. dollar spot price of Brent Dated between 1984-2012. Brent dated is a benchmark assessment of the price of physical, light North Sea crude oil.

Source: BP Statistical Review of World Energy 2013

Source: BP Statistical Review of World Energy 2013 FIG 4: Crude oil price

development ($ per barrel) FIG 5: Crude oil price volatility

% Ch an ge P rice ($ )

(14)

– 14 –

2.2 Transmission channels of oil price movements

The transmission of oil price changes on the economy is complex. With respect to inflation, we can identify two channels through which oil prices have an effect on the price level: first-round (direct and indirect) effects and second-round effects. We follow the distinction made by the European Central Bank (ECB) in its August 2010 Monthly Oil Bulletin.

2.2.1 First-round effects

The first-round effects are split up into direct and indirect effects. Direct effects reflect the immediate pass-through of oil price changes into prices of refined oil products, such as fuel or heating oil, which are consumed by households. The direct impact of oil prices is dependent on the share of household spending on refined products as a share of total expenditures (Álvarez et al., 2011). The ECB states that the largest and most immediate impact of oil price changes on inflation comes from direct first-round effects. Along the same line, Meyler (2009) reasons that direct first-round effects are most immediately passed-through into consumer prices and uses it as the channel of focus for measuring the effects on euro area consumer liquid fuel prices. Doroodian and Boyd (2003) argue that a change in the crude oil price of the magnitude of the 1973 oil shock leads to a significant effect on gasoline and refinery prices in the U.S., but that these effects largely dissipate over time at the aggregate level.

Indirect effects occur when costs of producing goods and services which use petroleum products as an input factor go up (down) and this increase (decrease) is passed through to consumer prices. Álvarez et al. (2011) argue that through this channel oil prices have slower-pass through and the effect depends on factors such as cyclical movements of the economy, market competition and whether the shock is transitory or persistent. Including euro zone countries in their sample, and in particular focusing on Spain, they claim that the direct effects of oil price changes account for more than fifty percent of the variance of Spanish inflation and forty-five percent in case of the euro area as a whole. Indirect effects, however, seem to have limited effects. This is because indirect effects depend on the oil intensity of the economy. They argue that these effects are today less

(15)

– 15 –

significant than before. Oil price shocks in the early 1970s triggered firms to have more energy-efficient production functions, thereby reducing the use of oil per unit of output of the economy. The role of energy intensity of the economy will be further discussed in section 2.3.2.

2.2.2 Second-round effects

Second-round effects arise when workers successfully raise nominal wages to maintain the same level of purchasing power when energy prices rise. Consequently, costs for firms go up as wages go up. When firms pass on the rise of wages to the end consumer by increasing the price of their goods and services, there is upward pressure on the overall price level. Baumeister et al. (2009) reason that while direct effects only result in a permanent shift of the price level, second-round effects could cause a self-sustaining downward spiral of higher wages and prices leading to a more persistent effect on inflation. The models in this paper do not account for first and second round effects, since only aggregate data are applied.

2.3 Oil Use in the European Union

To understand the degree of exposure of a EU member state to oil price changes, we need to identify differences in the consumption and production of oil and the usage of crude oil across the EU. For instance, we can expect that economies of member states with high energy intensity levels to be more vulnerable to oil price changes than member states with lower energy intensity levels. This hypothesis is supported by Bachmeier and Cha (2011), who investigated the role of energy intensity in pass-through effects of oil price changes on inflation in the U.S. in the 1973–1985 and 1986–2006 period. Not only the role of crude oil, but also that of alternative sources of energy may be relevant for asymmetric effects following oil price shocks. When oil prices rise, prices of other energy sources tend to rise as well due to increase demand for these alternative sources, especially when the oil price shock is driven by an increase in overall economic activity. For an exogenous oil supply shock or an oil-specific demand shock the price rise of alternative energy sources depends on the substitutability of oil and other sources of energy (Peersman and Van Robays, 2012).

(16)

– 16 – 0 100,000 200,000 300,000 400,000 500,000 600,000 700,000 800,000 1990 1995 2000 2005 2010

FIG 6: Gross inland energy consumption

Thousand tonnes of oil equivalent (TOE)

EU 28 Non CEECs CEECs EU Primary Production 0 2000 4000 6000 8000 10000 12000 14000 1990 1995 2000 2005 2010

FIG 7: EU import and export of crude oil Thousand barrels per day

Imports Exports 2.3.1 EU Consumption and Production of Oil

The consumption of energy in the EU as a whole lies substantially above production, as shown in Figure 6.9_{For comparison purposes the figure includes a breakdown of the EU}

consumption level into member states excluding CEECs (non CEECs) and the remaining eleven CEECs. This split shows that consumption in the CEECs stayed fairly constant over the past twenty years. Figure 7 shows imports relative to exports of crude oil for the entire EU. As a whole, the EU is a net importer of oil.

Table 1 in appendix I shows that Denmark, Germany and Italy were the largest oil-producing countries in 2012 (in absolute terms). The largest consumers were Germany and Italy, but also France and Spain are relatively large consumers of oil. Further, all members states except Denmark are net importers of crude oil. All CEECs have been net importers over the last twenty years. In 2012, the average net amount of imports was about 119 thousand barrels.10

2.3.2 Energy intensity

Apart from consumption and production of oil we can differentiate between member states in terms of energy intensity, which is defined as the amount of energy (including oil) required to generate a unit of economic output, measured as the ratio of gross inland consumption of energy to GDP. It captures the energy consumption of the economy and its overall oil efficiency. The majority of studies on the role of energy intensity in the oil

9

The Gross Inland Consumption of Energy is calculated as the sum of consumption levels of five types of energy: coal, electricity, oil, natural gas and renewable energy sources. In addition, each of these figures is calculated as an aggregation of different data on production, storage, trade (imports/exports) and consumption/use of energy.

10_{See U.S. EIA country data.}

(17)

– 17 – 0 100 200 300 400 500 600 700 800 1995 2000 2005 2010 Energy intensity ratio (kg of oil equivalent per 1000 EUR)

FIG 8: Aggregate Energy intensity

Euro area CEECs 0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 1995 2000 2005 2010

FIG 9: CEEC Energy intensity

Czech Republic Estonia Croatia Latvia Lithuania Hungary Poland Romania Slovenia Slovakia MS excl CEEC

price-macroeconomy relationship argue that energy intensity has declined through time (Bachmeier and Cha, 2011). Voight et al. (2014) perform a study on energy intensity in 40 major economies and argue that changes in the level of energy intensity are due to either technological or structural adjustments. They find that technological improvements are the key component of sustained energy efficiency improvement, leading to a decline in energy intensity. Cornillie and Fankhauser (2004) reason that improvements in energy intensity are directly linked to economic progress. For the purpose of measuring asymmetric effects of oil price changes on inflation we include energy intensity in the model. Figure 8 shows differences in energy intensity between euro area countries and CEECs. The figure indicates that CEECs have a substantially higher energy intensity level. This observation is confirmed in figure 9, where an average of all member states excluding CEECs is compared to the individual eleven CEECs. Another observation is that energy intensity levels have gone down over the past years in the CEECs but not so much in the rest of the European Union.

Source: EIA

(18)

– 18 –

2.4 Inflation

This paragraph discusses some of the differences in inflation rates across the EU. Inflation dynamics within the EU are a widely discussed topic. Article 121 of the 1992 Maastricht Treaty requires candidate countries that want to join the Economic and Monetary Union (EMU) – and adopt the euro – to curb their inflation rate to not higher than 1.5 percentage points above the average of the three lowest inflation rates in the EU (Kocenda and Papell, 1997). Generally, the price level is lower in less developed (lower-income) countries compared to developed countries. This holds for the EU as well, where new member countries, who typically have lower GDP levels, have lower prices compared to the “older” EU member states. In 2013, price levels for consumer goods and services differed widely. Denmark had the highest price level (140% of EU average), followed by Sweden (130%), Luxembourg and Finland (both 123%). Central and Eastern European economies had the lowest price levels at around 30% to 35% of the EU average.11_{Apart from price levels, we see that inflation rates between EU members differ}

strongly (appendix II). Over the past decades, CEECs have experienced higher – and more erratic – rates of inflation compared to non-CEECs.

Égert (2011) performs an empirical investigation of the drivers of inflation rates across the EU. He shows that the Balassa-Samuelson effect is not an important driver, although this concept is traditionally used to explain price level differentials between countries that have unequal productivity growth rates in the tradable goods sector. The Balassa-Samuelson theory states that, for an emerging economy, productivity growth in the sector of tradable goods and services exceeds that of non-tradables. Assuming that wages equalize across sectors due to perfect labor mobility, a higher productivity growth rate in the tradables sector pushes up wages in all sectors, leading to a higher price of non-tradables. With a fixed exchange rate, the relative price increase in fast-growing countries that catch up with more slow-growing developed countries may result in higher inflation (Pilbeam, 2006).

(19)

– 19 –

3. The Oil Price-Inflation Relationship

For a long time economists have been interested in the effects of oil prices on the macroeconomy. Attention to this issue grew significantly in the 1970s, because this period showed growing dependence on imported oil and unprecedented disruptions in the world oil market, coinciding with a downturn of the U.S. economy. Hence, it was not surprising that academics started to expect a causal relationship between oil prices and macroeconomic indicators (Darby, 1982; Barsky and Kilian, 2004). In an influential paper by Hamilton (1983), he argues that all but one of the U.S. recessions that took place after the Second World War were preceded by a significant rise in the oil price, albeit typically with a lag of three quarters. Hamilton’s findings triggered a rise in empirical research on oil price effects on the macroeconomy. The majority of these studies find that oil prices indeed affect macroeconomic aggregates such as output and inflation (Burbidge and Harrison, 1984; Rotemberg and Woodford, 1996; Leduc and Sill, 2004). Although there is mixed evidence on whether oil price shocks cause economic downturns, there are multiple studies that claim that oil price changes at least partially impact inflation levels (Mork and Hall, 1980; Darby, 1982; Sheehan and Kelly, 1983; Chen, 2009; Caraballo and Usabiaga, 2009). This chapter continues to discuss the findings of the most relevant studies on the oil price-inflation relationship. It attempts to identify the key issues in the debate which will form the basis for the empirical model in chapter 4.

3.1 Declining pass-through effect

During the 1990s, studies on the effects of oil prices on the macroeconomy have pointed out that the relationship has weakened over the past decades (Darrat et al., 1996; Hooker, 1996). This holds for the oil price-inflation relationship as well. The main finding is that more recent oil price shocks have had a reduced impact on inflation when compared to earlier shocks in the 1970s and 1980s. Blanchard and Galí (2007) use a VAR approach for the U.S., France, Germany, the UK, Italy and Japan for the period 1970-2005 and measure the effect of oil price shocks on the CPI. They use three proxies for

(20)

– 20 –

inflation, namely CPI, GDP deflator inflation and wage inflation. It appears that CPI inflation was relatively sensitive to oil price changes in the late 1970s, when inflation is estimated to have risen over 1 percentage point during a period of two to three quarters after a 10 percent rise in the oil price. However, the inflation response became muted and less persistent over time, especially over the last years of the sample period. This finding also holds when using GDP deflator inflation. Over the last years, however, the effect is negligible. The response of wage inflation is muted along the entire period, except for late 1970s and early 1980s.

Hooker (2002) provides evidence supporting this argument by using Phillips curve models that allow for a structural break (pre- and post-1981 period) and nonlinearities for U.S. data. He argues that pass-through effects on inflation have weakened after 1981, but significantly contributed to inflation before that time. In contrast, Gisser and Goodwin (1986) find no evidence of a break in the way oil prices affect U.S. macroeconomic variables, including inflation. In addition, they argue that the relationship has been stable in the 1970s. Also Gómez-Loscos et al. (2012) find evidence that conflicts with that of Blanchard and Galí (2007) and Hooker (2002). They estimate the impact of oil price shocks on the economic activity and prices of the G-7 economies. Using VAR methods for a sample running from 1970 until 2008, they identify four differentiated periods and prove the existence of breaks. They find that the response of output and inflation to oil price swings is less strong around the 1990s compared to the strongest response in the 1970s, but that the impact of oil prices recovers in the 2000s (albeit smaller than in the 1970s).

In another paper, De Gregorio et al. (2007) use a sample of thirty-three countries, primarily industrial and emerging market economies. By applying a comparable approach as Hooker (2002), they find evidence of a decreasing effect of oil price changes on inflation for industrial countries and to a lesser degree for emerging economies. They come up with a number of explanations why pass-through effects have gone down. First, the origin of the oil price shock plays a role. In earlier shocks the driving force was supply shortages. They claim that oil price changes nowadays are to a large extent ascribed to demand shocks, a finding that holds for both oil importing and oil exporting counties.

(21)

– 21 –

This result is supported by the work of Archanskaïa et al. (2012), who argue that the main driving force behind oil price shocks has changed from supply-driven shocks in 1970-1992 to demand driven shocks in 1992-2006. The second explanation put forward by de Gregorio et al. (2007) is that the persistence of an oil shock is important. A smaller pass-through is expected for transitory shocks compared to persistent shocks. Over the last decade persistence expectations have gone up, and this equally holds for the actual persistence of the shock. Third, countries may have implemented institutional price stabilization mechanisms to protect themselves against oil price fluctuations, such as countercyclical oil taxes, stabilization funds and strategic oil reserves. In this way, the gasoline price at the pump does not reflect the actual volatility in the oil price. Fourth, monetary policy responses to oil shocks may have changed, due to a stronger inflation-targeting policy than before.

Chen (2009) uses data on 19 industrialized countries for which he estimates time-varying oil price pass-through coefficients. He finds evidence of declining effects across almost all countries in the sample and suggests explanations for the smaller impact of oil price changes. He argues that lower pass-through is not due to pre-shock inflation levels, but is instead linked to domestic currency appreciation, a more active monetary policy in response to inflation and a higher degree of trade openness. This finding is disputed by Taylor (2000), who argues along the same line as De Gregorio et al. (2007) by providing evidence that inflation is positively correlated with persistence to inflation, suggesting that low pass-through is caused by low ex ante inflation levels.

In the paper by De Gregorio et al. (2007) the model controls for the economy’s oil intensity and they find that the declining use of oil in economic activity helps explain the decreased pass-through effects. The economic importance of oil has declined as industrial economies have become more service oriented and previous oil price shocks have driven these economies to adopt more energy efficient technologies and diversify energy consumption. The idea of declining oil intensity is supported by other scholars. Bachmeier and Cha (2011) investigate whether the change in responsiveness of core CPI inflation in the U.S. to oil price shocks can be explained by changing energy intensity in the economy for the 1973-1985 period and the 1986-2006 period. They argue that firms

(22)

– 22 –

have “substituted away” from oil as a factor of production. Although production of a given amount of output currently uses less energy input than it did in the years following the 1973 oil shock, it is at least partially offset by higher energy prices. They find that about two-thirds of the reduced response in inflation was due to lower oil intensity in the production process. One-third of the reduced effect came from changes in monetary policy by the Federal Reserve. In the case of the U.S., Katayama (2013) argues that energy intensity is an explanatory factor for the decline in pass-through due to technological advancement. Also, deregulation of the transportation industry have led to a more competitive environment and less reliance on petroleum. Although energy intensity seems a logical control variable in this paper’s regression model, Hooker (2002) argues to the contrary of Bachmeier and Cha (2011) and Katayama (2012), by providing empirical evidence of instable coefficients on oil share of the U.S. economy and finding significantly lower pass-through.

3.2 Asymmetric pass-through of oil prices

Since this paper examines asymmetry in effects of oil price changes between CEECs and non-CEEC, we now briefly discuss a number of studies that have focused on asymmetry in pass-through effects in particular. LeBlanc and Chinn (2004) take into account asymmetric and nonlinear effects by adopting a Phillips curve framework. Using data for the U.S., United Kingdom, France, Germany and Japan, oil price increases of as much as 10 percentage points will lead to an increase in inflation of about 0.1-0.8 percentage points in the United States and the EU. For euro area countries, Meyler (2009) does not find systematic and widespread asymmetric effects of increases or decreases of either upstream (crude and refined products) and downstream (consumer). He points out, however, that this holds at the aggregate level and this may not be the case at a sub-regional level. Égert et al. (2010) investigate the effects of a positive commodity price shock on CPI levels of ten Eastern European transition economies, using a VAR model for a time period that runs from the early 1990s until 2009. Their results indicate that a positive commodity supply shock – leading to higher energy prices – gives rise to higher inflation rates, but effects are asymmetric across countries. The results are more pronounced for Eastern European countries at a lower stage of economic development,

(23)

– 23 –

because poorer countries consume a relatively higher share of commodity-related goods in the CPI basket. This finding would suggest that economic structure plays a role in explaining asymmetric pass-through of oil price changes into inflation. Second, they find that the response to the commodity price shock is weaker for inflation targeting countries. This would imply that a credible monetary policy and well-anchored inflation expectations may contain the effects of a shock.

(24)

– 24 –

4. Empirical Analysis

This chapter focuses on the empirical model. First, the data and sample are discussed. This is followed by a discussion of the dependent and independent variables. In the third paragraph a discussion of the models is provided, followed by the regression results. Finally, a number of threats to the validity of the models is provided and tests are run to address some of the threats to validity.

4.1 Data and Sample

This paper makes use of regression analysis to examine the effects of oil price changes on inflation in the EU. The sample time period covers the years running from 1993 to 2012 and for each variable we have yearly data. In the years preceding 1993, in fact since the late 1980s, countries in Eastern and Central Europe have been in transition from a planned economy to a market economy (Wertel et al., 1997). The ensuing economic turmoil could lead to either over- or underestimating the true effect of the regressors in the model. This reasoning is supported by Blanchard (1996), Égert et al. (2010) and Frenkel and Nickel (2005).

The sample consists of 25 member states of the European Union. To avoid outliers, the sample excludes Luxembourg, Malta and Cyprus. Luxembourg is excluded since the country has the largest GDP per capita in the EU due to its financial sector and could therefore be considered an outlier. This argument is supported by the findings of Mattila (2006). Further, Malta and Cyprus are omitted because these economies are too small in terms of GDP but are strongly dependent on (energy) imports due to their limited natural resources.12_{The total number of observations equals 592. Working with many}

observations increases the chance of missing values, which is often a feature of financial or economic econometric analysis (Baum, 2006). For some explanatory variables a small number of consecutive years are missing, which applies to all countries in the sample.

12_{Malta makes up 0.5% of the EU total GDP, while Cyprus accounts for 0.1%. See International Monetary Fund,}

(25)

– 25 –

4.2 Variables

This paragraph discusses the variables that are included in the models, including the way these variables are operationalized. It starts with the dependent variable, inflation, and continues with all explanatory variables.

4.2.1 Dependent variable

The variable to be explained is inflation. Inflation is measured by the annual growth of the GDP implicit deflator, instead of CPI inflation. Although these two measures of inflation tell a similar story, there are two important differences. The GDP deflator reflects the prices of all goods and services produced domestically, whereas the CPI reflects the goods and services bought by consumers. The GDP implicit deflator is the ratio of GDP in current local currency to GDP in constant local currency. The data comes from World Development Indicators (WDI), based on World Bank national accounts and OECD national accounts data. A one period lag of the inflation rate is included in the baseline model to capture persistence in inflation, since inflation in the current period probably depends on its value in the previous period (Meller and Nautz, 2012).

4.2.2 Baseline model variables

The baseline model includes those variables that are expected to explain inflation best. First, GDP per capita and GDP growth are included as separate variables. Low-income countries tend to have higher inflation levels following the Balassa-Samuelson argument which justifies adding GDP per capita. Also the GDP growth rate is included. Higher growth is likely to increase inflationary pressures (Coffinet and Frappa, 2010). Conversely, one can expect that inflation impacts economic activity and growth as well, which suggests that causality runs two ways. Koulakiotis et al. (2012) investigates the endogeneity between inflation and economic activity for 14 European countries. They find that inflation causes GDP at the 5% significant level and GDP causes inflation at the 10% level. For this paper we extract GDP per capita data from WDI. It is measured as gross domestic product divided by midyear population. GDP is the aggregate of value added by all domestic producers plus any product taxes and minus subsidies not included

(26)

– 26 –

in the value of the products. Data is measured in current U.S. dollars. GDP per capita growth data is also extracted from WDI, measured as the annual percentage growth rate of GDP per capita based on constant local currency. Aggregates are based on constant 2005 U.S. dollars. Just as with GDP per capita, GDP growth is calculated without making deductions for depreciation of fabricated assets or for depletion and degradation of natural resources.

Subsequently, the exchange rate and unemployment are added to the baseline. Unemployment is included as it affects inflation through a Phillips curve framework, an approach followed by other studies (Hooker, 2002; LeBlanc and Chinn, 2004). The Phillips curve shows that unemployment and inflation share an inverse relationship (Romer, 2012). National output increases when more people work, causing wages to rise which causes consumers to spend more. Demand for goods and services goes up, causing prices to rise. In this paper, data on unemployment is extracted from WDI and is an estimate by the International Labour Organization (ILO), defined as the percentage of the total labor force that is without work but available for and seeking employment. The exchange rate is included since a depreciation of the exchange rate means the currency buys less foreign exchange, leading to more expensive imports and cheaper exports, thus increasing inflation (Dornbusch, 1988). Incorporating the exchange rate in the models follows the line of reasoning of de Gregorio et al. (2007). They argue that increased exchange rate flexibility among oil importers should increase the volatility of oil price inflation in terms of domestic currency. When we assume that governments aim at inflation stability, flexible exchange rates may help to absorb external oil price shocks (de Gregorio et al., 2007). Exchange rate data is extracted from Penn World Tables (PWT), version 7.1, and is measured as national currency units per US dollar.

The key independent variable for this study is the crude oil price, which will be included when we expand the baseline model. Selecting the right crude oil price variable is not straightforward. National oil prices are influenced by price-controls, taxes on petroleum products, exchange rate fluctuations and national price index variations (Cuñado and Perez, 2005). In an EU context, this can lead to differential characteristics which impacts the effective oil price each economy faces. Therefore, most of the empirical work which

(27)

– 27 –

analyses the effects of oil price changes on macroeconomic variables use benchmark crude oil prices, such as North Sea Brent and West Texas Intermediate (WTI), or Dubai crude oil price. Each of these three refers to a crude oil of high quality with a specific production or trading location. The WTI and Dubai are mainly traded in the US and Asia, respectively, whereas North Sea Brent is used for world reference (Chevillon and Rifflart, 2009). According to the EIA, Brent is a blended crude stream produced in the North Sea region which serves as a reference or “marker” for pricing a number of other crude streams. As Brent is used for the price of two-thirds of worldwide exchanged crude oil, and because it is also used for the EU’s reporting in its monthly oil bulletin, this paper makes us of this crude oil price.

Although the different prices are not equal, they are strongly correlated. Our choice for the Brent crude oil price should therefore not substantially affect estimates. This is supported by the conclusions of Fattouh (2007), who argues that the Brent-WTI price differential is stationary. Data on the Brent crude oil price is extracted from EIA and includes only spot price data, where EIA has calculated annual data based on the unweighted average of the daily closing spot prices. The data is constant for all countries in the sample as each economy faces the same crude oil price, but variable over time. This implies that we have a variable that is similar to a time-fixed effect. To measure the effects of a change in the oil price on inflation the first difference transformation of the Brent oil price is constructed, following Cuñado and Perez (2005).

In other papers, a number of alternative oil price specifications are employed. Cuñado and Perez (2005) use different oil-CPI specifications in order to measure the effects of oil price shocks on the price level in six Asian countries over the period 1975-2002.13_In

particular, they use a separate variable that only reflects real oil price increases. In this paper, regressions are performed with similar variables, which will be briefly discussed in 4.5. Also, Scholtens and Yurtsever (2012) apply different types of oil specifications. They use the Net Oil Price Increase (NOPI) as a proxy for an oil price shock, following Hamilton (2009), which seems especially relevant for spending decisions of consumers

(28)

– 28 –

and firms, and monthly changes of real oil prices with a conventional first log difference transformation of real oil price variables.

4.2.3 Other explanatory variables

The models employ ‘Energy intensity of the economy’ as a proxy for the exposure of member states to oil price changes. It is expected that EU economies which rely more on the use of oil for production are more likely to have a larger pass-through effect of oil price changes on inflation. Energy intensity is measured by the quantity of energy required per unit of output, which implies that using less energy to produce one unit of output reduces energy intensity and increases energy efficiency. Data is extracted from EIA, which defines energy intensity as Total Primary Energy Consumption per U.S. dollar divided by GDP. Energy consumption is measured in quadrillion “British thermal units” (Btu) per year in 2000 U.S. dollars (at market exchange rates) and includes the consumption of petroleum, dry natural gas, coal, and nuclear energy, hydroelectric, and non-hydroelectric renewable electricity but also net electricity imports (electricity imports minus electricity exports). Finally, a dummy variable is created for countries belonging to Central and Eastern Europe. Since energy intensity and inflation levels differ vastly between CEECs and non-CEECs in the EU (see graph 8 and 9), we expect oil price changes to affect inflation to a different extent.

4.3 Model

This section discusses the econometric specifications of the regression analysis, where attention is paid to the estimation technique that is applied throughout all models. After that, each of the models is briefly discussed.

4.3.1 Econometric specification

Since the 25 EU member states are observed for more than two time periods, in this case twenty time periods (i.e. years) running from 1993 through 2012, a panel (or longitudinal) data approach is used for repeated cross-sections. Panel data includes multiple entities where each entity is observed in two or more time periods (Stock and Watson, 2012). The models control for omitted variables when these variables vary

(29)

– 29 –

between entities (EU members) but do not change over time. With entity-fixed effects each country has a different intercept (Baker and Riddick, 2013). If we let equal a

vector of all independent variables and equal a vector of all controls, the model

would boil down to:

Where _{are the country specific intercepts;} _; _and _{is the}

error term. The model includes 24 country dummies to avoid the ‘dummy variable trap’. Fixed effects are assumed to be country-specific, since EU member states are structurally different and we can assume that this does not change over time. Country-specific effects allows to estimate the pure effect of the regressor more accurately as each dummy in fact “absorbs” the effects that are particular to a member state. One can also adopt random effects, for which variation across entities is assumed to be random and uncorrelated with the predictor or independent variables in the model. This variation can be summarized in a random error term. This means that the regression error consists of two parts: an individual specific component, which does not vary over time, and a remainder component, which is assumed to be uncorrelated over time. Correlation of the error terms over time is attributed to the individual effect component of the error term (Verbeek, 2008).

In deciding between entity-fixed effects and random effects a Hausman test is applied, where the null hypothesis is that the preferred model uses random effects instead of fixed effects (Green, 2008). It tests whether the errors are correlated with the regressors and where the null hypothesis states that they are not. The Hausman test statistic for our model gives a p-value of 0.0001 < 0.05. Hence, we reject the null hypothesis and adopt entity-fixed effects estimation.

Just as entity-fixed effects control for variables that are constant over time but differ across entities, so can time-fixed effects control for variables that are constant across entities but evolve over time (Stock and Watson, 2012). It can be appropriate to include both time- and country-specific effects when potential omitted variables can be assumed to be fairly constant over time but vary across countries, while other potential omitted

(30)

– 30 –

variables can be assumed to be constant across states but vary over time. EU members are structurally different from each other and this does not alter over time. Also, there are no year effects that affect all countries in the sample to the same extent which justifies not using time-fixed effects. The use of the oil price change in this paper is a fixed-time effect. Because the variables are measured through time we need to control for possible correlation between variables and their lagged values, also known as autocorrelation. This is done by applying Prais-Winsten estimation.

4.3.2 Prais-Winsten Estimation

In times series data the value of any variable is potentially correlated with its own lagged value. Omitted variables – and possibly included variables as well – may change slowly over time causing the error term to be serially correlated (from one period to the next). This is known as autocorrelation of errors. The consequences of autocorrelation are the same as with heteroskedasticitiy: OLS remains unbiased, yet estimates become inefficient and standard errors are not correctly estimated (Stock and Watson, 2012). To address autocorrelation the models apply Prais-Winsten estimation. This method uses Generalized Least Squares (GLS) to estimate the parameters of the linear regression with serially correlated errors. It follows a first-order autoregression model, abbreviated AR(1), which accounts for the first lag of a variable. The error term is assumed to depend on the value of the error term in the previous period. In simple terms, the AR(1) is expressed as:

If the errors are auto-correlated the usual standard error formula for cross-section is not valid. Errors that are valid in case of autocorrelation are Heteroskedasticity and Autocorrelation Consistent (HAC) standard errors. These errors are also termed ‘clustered’ , because it allows regression errors to have an arbitrary correlation within a cluster – or in this case a member state – but restricts these errors to be correlated across entities. Clustered standard errors are valid in case of heteroskedasticity, autocorrelation, or both (Stock and Watson, 2012).

(31)

– 31 –

4.3.3 Baseline model and extensions

The first model is the baseline for the empirical analysis. This model includes the variables that are expected to best explain the change in inflation, expressed as:

Where equals inflation for country at time ; equals GDP per capita;

_{equals the real annual percentage growth rate of GDP per capita based on}

constant local currency; equals the percentage of the total labor force that is

unemployed in a particular member state; is a one-period lag of the dependent

variable; measures the exchange rate as national currency units per US dollar

and is the error term. This model uses country-specific fixed effects, which also holds

for all of the following models. For brevity, the country-specific intercepts will be omitted in the notation but are included in the estimations. The expected signs for ,

and are positive and is expected to be negative. The sign on is likely to be positive. When GDP growth increases, demand grows causing prices for goods and services to rise. The variables in the baseline model are all macroeconomic aggregates and are likely to be correlated. To account for this we test for multicollinearity in section 4.5.

Model 2 will extend the baseline model with the change in the oil price as additional independent variable. The oil price level variable is not included in any of the models since the oil price level likely has a trend (unit root), which leads to insignificant coefficients. Instead, the oil price change variable is used, which reflects the relative oil price change with respect to the previous year. With this model the first proposition will be tested which states that oil price increases lead to higher inflation across the EU. The model is expressed as:

(32)

– 32 –

Where _{reflects the exogenous variable that indicates the relative oil price change}

with respect to the previous period (year). Oil prices are the same for each country but differ across time. The sign on this variable is expected to be positive. Larger positive changes in the oil price are expected to push up inflation.

Following the discussion on energy intensity in chapter 2 we expect that member states that have more energy intensive economies are more exposed to oil price changes than energy efficient economies. Hence, a high energy intensity level implies that the economy is likely to be affected differently (to a larger extent) by oil price changes than those economies that are more efficient in their use of energy. We can distinguish between CEECs and non-CEECs, since we have seen that energy intensity levels differ strongly between these two groups (graph 4 and 5). Model 3 will test the second proposition which states that oil price changes have different effects on inflation depending on the level of energy intensity of the member state’s economy. The model includes energy intensity as an explanatory variable. The model boils down to:

Where _{equals the energy intensity level of the economy. The expected sign on} _{is positive.}

In the fourth model we test whether inflation rates in CEECs respond differently to oil price changes than inflation rates in non-CEECs. Not only do we see differences in energy intensity levels between these two groups, also the inflation rate movements over the past decade differ substantially (see appendix II). A dummy is created for model 4 to test whether the effects of oil price changes on inflation is different for CEECs compared to non-CEECs. Also the interaction term with the oil price change is included, since the parameter coefficient on the dummy variable is the same for different countries.

(33)

– 33 –

[ ]

Where is a dummy variable which equals ‘1’ if a country belongs to the CEEC region and ‘0’ otherwise; and _{is an interaction term between this dummy}

and continuous variable of the oil price change. Including the interaction term implies that if _{equals zero the regression function estimates only the effect of}

(along with the intercept and the remaining independent variables). However, if a country is a CEEC then the intercept becomes and we get . Hence, it changes both the slope and the intercept. This model allows us to test the third proposition which states that the effect oil price changes on inflation is dissimilar between CEECs and non-CEECs.

4.4 Empirical Results

This section contains the empirical analysis. It starts with a number of comments on descriptive statistics of each of the variables. Next, the regression output is discussed for each of the four models.

4.4.1 Descriptive Statistics

Table 2 and 3 in appendix III contain descriptive statistics for each variable as well as correlation coefficients. Inflation is negatively correlated with GDP per capita and its growth rate, positively related to unemployment and negatively related to the exchange rate. Overall, the signs are as expected. However, contrary to expectations the GDP growth rate is positively related to inflation. Further, oil price changes and inflation are inversely related, but the correlation seems to be quite small. Also, energy intensity and inflation are relatively strongly and positively correlated.

(34)

– 34 –

Dep. var: Inflation (1) (2) (3) (4)

GDP per capita 0.00 0.00 0.00 0.00 (2.41)** (2.42)** (-0.93) (-1.02) GDP p.c. growth -2.70 -3.24 -2.77 -2.85 (-2.06)* (2.13)** (2.37)** (2.38)** Unemployment -0.53 -0.63 -1.95 -1.95 (-1.47) (-1.66) (-1.7) (-1.7) Inflation lag 0.17 0.17 0.03 0.03 (7.03)*** (7.64)*** (-0.87) (-1.08) Exchange rate 0.07 0.09 0.05 0.05 (-1.59) (-1.76)* (-1.42) (-1.35)

Oil price change 0.33 0.38 0.24

(-2.03)* (2.33)** (2.12)**

Energy intensity 0.01 0.01

(2.35)** (2.38)**

CEEC -2.25

(-0.63)

CEEC*Oil price change 0.31

(-1.91)*

Constant 34.85 42.62 -77.51 -79.47

(2.63)** (2.60)** (-1.97)* (-2.01)*

R2 0.18 0.18 0.28 0.28

N 437 437 415 415

Fixed effects Yes Yes Yes Yes

-0.11 -0.11 -0.08 -0.08

MSE 49.87 49.80 47.74 47.76

Table 4 - Regression results

* p < .10; ** p < .05; *** p < .01 4.4.2 Regression Analysis

A total of four models is developed that aims to identify whether oil price changes influence inflation in the EU and to what extent we see asymmetric effects between member states. Table 4 displays the results of these four models. The table reports t-statistics between parentheses.

The baseline model provides evidence of the effect on inflation of a number of controls. The first observation is that the signs on the coefficients are precisely as we expected. It shows that GDP per capita is statistically significant at the 5% level (p < .05). This

(35)

– 35 –

implies that a 1% rise in GDP per capita leads to a fall in inflation of .001%. GDP growth is also significant – albeit at a lower level of significance. At the 5% level we conclude that a 1% increase in GDP growth leads to a decline in inflation of 2.7%. The lag of inflation is highly significant (p < .01). A 1% increase in the previous period inflation rate leads to an increase in the current period inflation rate of .167%. The _{is .18, which}

implies that about 18% of the variation in inflation is explained by the model.

Model 2 adds the oil price change, our variable of interest. We conclude that the oil price change is significant at the 10% level (p < .10). The coefficient should be interpreted with care. A 1% increase in the oil price change – which means an increase in the growth rate of the nominal oil price – leads to an increase in inflation of .33%. Model 2 shows that oil prices in fact do explain inflation, albeit for the 1993-2012 period and only for the EU. Again, GDP per capita is significant at the 5% level (p < .05) and carries a similar coefficient. GDP per capita is now significant at the 5% as well, implying that a 1% rise in GDP growth lowers inflation by 3.2% approximately. The inflation lag is again strongly significant at the 1% level (p < .01). Contrary to the baseline, the exchange rate is significant at the 10% level (p < .10). Since the exchange rate is measured as national currency units per US dollar, this means that a rise in the exchange rate implies a depreciation of the domestic currency, which fuels inflation. Here, a 1% rise in the exchange rate (or a 1% depreciation), leads to an increase in inflation of .09%. The is identical to model 1. The Mean Squared Error (MSE) is slightly lower, so the model seems to be only marginally more accurate.

To test whether ‘energy intensity of the economy’ matters for the impact of oil prices on inflation we include the energy intensity variable in model 3. We estimate whether the effect of an oil price change on inflation is different for economies that are more exposed to oil. The oil price change is now significant at the 5% level, and so is ‘energy intensity’. We conclude that a 1% rise in the energy intensity level leads to a .01% rise in inflation. The variable carries a positive coefficient, as we expected, which implies that energy intensity positively contributes to the inflation rate. The lag of inflation and the exchange rate are no longer significant and the increases substantially to .28. The results from this model are a key finding for this paper. Where Bachmeier and Cha

(36)

– 36 –

(2011) argue that energy intensity have led oil prices to lose their explanatory power, we find that oil prices are indeed significant when we control for energy intensity.

The last model includes a dummy for CEECs and an interaction term with the oil price. The oil price change and energy intensity are both strongly significant at the 5% level (p

< .05). A 1% rise in the oil price change leads to a .24% rise in inflation. The effect of

energy intensity is similar to model 3. Once more, GDP growth is significant at the 5% level (p < .05). The interaction term relating the CEEC dummy with the oil price change is significant at the 10% level and carries a positive sign. The sign on the CEEC dummy is positive, following our expectations. The model confirms this paper’s prediction that there are indeed asymmetric effects of oil price changes between CEECs and non-CEECs, when controlling for energy intensity. Just as with model 3, the is .28, and implies a substantial improvement to model 2.

4.5 Internal Validity

This paragraph addresses a number of validity issues that are relevant to the models in section 4.4. Two important issues are potential autocorrelation and multicollinearity, which are briefly discussed below. After that, potential omitted variable bias, misspecification of the regression equation and simultaneous causality is addressed. Using variables over time may cause the error terms to be correlated from one period to the next. To test for autocorrelation we use the Durbin-Watson (DW) statistic.For each of the models the DW-statistic indicates that there is no autocorrelation present. Plotting the residuals shows us that the error term is approximately normally distributed (appendix IV). As discussed before, multicollinearity could pose a problem to the validity of the model if regressors are strongly correlated. Regardless of a high this could lead to substantially higher standard errors and lower t-statistics. To test for multicollinearity we employ Variance Inflation Factors (VIF). Table 4 in appendix III shows the VIF values for all variables and confirms that there are no problems with multicollinearity. Using VIF values is preferred over the use of simply the Pearson correlations, since the latter

Asymmetric pass-through of oil prices : an econometric analysis of the effects of oil price changes on inflation in the European Union