How does the stock market of 13 European countries react on a change in the oil price?

(1)

How does the stock market of 13 European countries react

on a change in the oil price?

Donglin Zhang

1401378

Anneke Tolsma

1487027

University of Groningen Faculty of Economics and Business

Supervisor: Dr. Ing. N. Brunia

Program: MSc BA Finance (Risk and Portfolio Management)

(2)

Preface

After many months we are proud to present our final version of our Master Thesis. It took a long road to finish this last part of our study Finance at the University of Groningen. Although it is not common to write a thesis together, this thesis is. Each of us had the final responsibility of a specific part. The final responsibility of the data, methodology, results of the impulse response function and the robusteness check sections lies with Donglin Zhang. Anneke Tolsma did have the final responsibility of the introduction, literature review, results of the variance decomposition and the conclusion section. Beside that she wrote the abstract and she combined the parts into a final report.

We are grateful to our supervisor, Dr. Ing. N. Brunia for his supervision and value comments. We would also thank the staff of University of Groningen for offering the advanced academic education. Finally we would like to thank our family and friends for the support.

Kind regards,

(3)

Table of contents

Abstract 4

1. Introduction 5

2. Theoretical background 7

2.1. The oil price 7

2.2. The relationship between cash flows and stock price and an oil price change 8

2.3. Overview of related literature 8

2.4. Macroeconomic variables 12 2.4.1. Industrial production 13 2.4.2. Interest rate 13 2.4.3. Exchange rate 13 2.4.4. Inflation 14 3. Data 15

3.1. Data sources and variable descriptions 15

3.2. Data descriptive statistics 17

3.3. Unit root tests 19

4. Methodology 21

4.1. The empirical VAR framework 21

4.2. Determining the lag length 21

4.3. The impact of shocks on the financial market 23

5. Results 24

5.1. Variance Decomposition 24

5.2. Impulse Response Function 30

5.3. Robustness check 31

(4)

Abstract

(5)

1. Introduction

The price of a barrel of oil depends on the supply and the demand of oil. However, the supply of the oil price is for 40 percent regulated by the organization of petroleum exporting countries (OPEC)1_.

This means that the price of a barrel of crude oil mainly depends on how much oil these countries are willing to deliver on the market. As a consequence, the price of one barrel of oil will be higher than when the market sets the price by demand and supply. Figure 1 shows the oil price development from January the first 1997 until the first of March 2008.

Figure 1 Price of one barrel of Brent crude oil

0 20 40 60 80 100 120 140 ja n-97 ja n-98 ja n-99 ja n-00 ja n-01 ja n-02 ja n-03 ja n-04 ja n-05 ja n-06 ja n-07 ja n-08 ja n-09 Date P r ic e o f o n e b a r r e l in $

Brent crude oil price

Source: IMF database

It is a general market perception that macroeconomic variables, like the return of the stock market, react on a change in the oil price. To be more specific, if the oil price increases, companies who are using oil will face higher costs for producing a particular good. In this case, the wealth of oil consumers partially shifts to producers of oil. How a change in the oil price affects macroeconomic variables is a widely examined topic. Rasche and Tatom (1981) are one of the earliest authors who wrote about the influence of an oil price increase on macroeconomic variables. Surprisingly, the influence of a change in the oil price on financial markets is not that well examined. The first research about the influence of an oil price change on the stock return is written by Kaul and Seyhun (1990).

For an investor it is of interest to know how a change in the oil price influences the stock market index. The interest of an investor is the trade off between risk and return. To lower the risk of holding a portfolio which is replicating a stock market index, an investor can use derivatives, like options and futures. A strategy to lower the risk of holding a portfolio which is replicating a stock market index is presented below. If an investor holds a portfolio of stocks which is replicating the stock market index,

(6)

the stock market index will decline in case of oil price increase since the companies which are listed in the index will face higher costs for its resources. In order to keep the loss on the portfolio as low as possible, the investor can protect his portfolio of stocks with buying long crude oil futures. If the oil price increases, the investor makes a gain on the futures, and a loss on the stocks market index. When the oil price decreases, a loss will be made on the future contracts and there will be a gain on the stock market index. To make this strategy work, it is of importance to know if there is a relationship between the price of a stock and a change in the oil price. Therefore, the following research question is formulated:

Does a change in the oil price influence the return of 13 European stock markets?

To examine the influence of an oil price change on the stock market index, most studies2_{use a}

standard market model. Within this market model the return of the stock market index depends on the return of the market portfolio and the extent of the stock market index responsiveness as measured by beta. In this paper, the influence of an oil price change on the stock market will be examined by using a Vector Autoregression (VAR) model. Within a VAR model all variables will be treated as endogenous. By treating all variables as endogenous it is not possible to make incorrect decisions about which variables are exogenous and which variables are endogenous. The result of a VAR model is driven by the data. Park and Ratti (2008) use this model to see if there is a relationship between a change in the oil price and the change in the return of the stock market index of 13 European countries and the United States. Beside the return of the stock market index, Park and Ratti (2008) also included the change in the oil price, change in the industrial production and the change in the short term interest rate. Hondroyiannis and Papapetrou (2001) examine the influence of macroeconomic variables on the return of stock market index of Greece. They conclude that a change in the exchange rate does have a substantial influence on the return of a stock market index. In our VAR model we will include changes of the following variables: return of the stock market index, oil price, industrial production, interest rate and exchange rate.

The remainder of this paper is organized as follows. In the second section a basic theoretical framework will be provided and articles about the influence of the oil price on stock markets will be discussed. In the third section the data and data restrictions will be presented. The VAR model will be discussed in detail in the forth section. In the fifth section the results will be presented and in the sixth section a robustness check will be preformed. The conclusion and recommendations can be found in the seventh section.

2_{Chen, Roll and Ross (1986), Hameo (1988), Ferson and Harvey (1994), Burmeister, Roll and Ross (2003),}

(7)

2. Theoretical background

2.1. The oil price

After the Second World War, several oil price shocks occurred. The first shock was during 1973 and 1974. During this period the OPEC came with its first oil embargo. As a consequence the crude oil price increased with almost 295% to $13.40 (Cuñado and Pérez de Gracia, 2003). The revolution in Iran during 1978-1979 leaded to a disruption in the oil supply and therefore the oil price increased to $30. This was an increase equal to 50%. When Iraq invaded into Kuwait the oil price went up with 62.5% to $26 in 1990.

In this paper we will examine whether or not the oil price has an influence on the return of the stock market index of 13 European countries. Table 1 shows the value of the net exports of mineral fuels, oils, distillation products and related goods between 1997 and 2006 for 13 European countries. Only Norway is a net exporter of mineral fuels, oils, distillation products and related goods in all years. In this paper Norway is an oil exporting country. Beside Norway, Denmark will also be treated as an oil exporting country since Denmark is a net exporter of mineral fuels, oils, distillation products and related goods after 1999. The remaining eleven countries are oil importing countries.

Table 1 The value of the net export of mineral fuels, oils, distillation products and related goods for 13 European countries

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Austria -2,630 -2,210 -2,604 -3,996 -3,854 -2,920 -4,995 -6,676 -9,227 -11,420 Belgium -4,926 -2,687 -3,048 -4,568 -5,328 -6,265 -6,801 -9,825 -16,750 -19,256 Denmark -659 -641 444 1,117 514 1,387 1,811 2,298 3,086 5,168 Finland -1,722 -1,199 -1,192 -1,730 -1,813 -2,303 -2,996 -3,579 -5,125 -6,571 France -15,549 -10,679 -13,735 -23,405 -20,568 -20,497 -25,425 -34,971 -45,861 -57,480 Germany -27,446 -20,733 -24,222 -41,153 -38,977 -32,419 -40,678 -48,847 -68,467 -82,999 Ireland -1,566 -1,300 -1,649 -2,715 -2,542 -1,363 -2,000 -2,796 -4,073 -5,284 Italy -12,555 -8,836 -16,682 -18,276 -17,116 -17,618 -20,387 -25,460 -32,828 -38,764 Netherlands 5,201 4,702 3,823 5,822 2,380 -7,074 -9,088 -5,254 -7,870 -8,706 Norway 26,510 17,565 23,071 39,171 36,102 34,883 41,149 50,353 67,898 79,932 Portugal -2,184 -1,658 -2,246 -3,478 -3,393 -3,348 -3,995 -5,103 -7,551 -7,806 Spain -7,168 -5,242 -7,304 -13,254 -13,720 -14,213 -16,369 -21,493 -32,281 -41,935 United Kingdom 9,025 4,660 7,072 10,172 7,270 8,320 7,179 1,108 -6,199 -11,079

Note: 1) Values represented in millions of US Dollar 2) The value represent the difference between the export and the import of mineral fuels, oils, distillation products and related goods (code 27) 3) Source: www.oecd.org

(8)

case of an oil price increase, income will transfer from the oil importing countries to oil exporting countries.

2.2. The relationship between cash flows and stock price and an oil price change

According to Chen, Roll and Ross (1986), the sum of the total expected discounted cash flows together equals the stock price. This relationship is shown in equation 1.

∑

∞ =

+

=

1 0

)

1 (

)

(

t t t

k

c

E

P

[1]

where P is the stock price, E(c) the expected stream of cash flows and k the discount rate. The discount rate is a combination of the expected inflation rate, the expected real interest rate and a component which reflects the market price for shouldering risk. The inflation rate is included in the discount rate to cover the possibility that future investment returns do not keep pace with future increases in prices and wages (Harris, 1983). A change in the oil price might influence equation (1) in three ways. In the first place an increase in the oil price leads to an increase in the costs to produce a particular good. If a company transfers the higher costs of raw material to the consumer, the expected cash flow will decline. Consequently, the stock price will decline.

An increase in the oil price also influences the discount rate since the inflation rate will be influenced. Huang, Masulis and Stoll (1996) provide the following argumentation. If the oil price increases, an oil importing country will face higher costs for producing the same amount of goods. This will lead to a decline in the oil importing country balance of payments. Due to the higher oil price, the amount of local currency on the exchange market increases. This leads to a downward pressure of the local currency and therefore the inflation increases. Since the inflation rate is a component of the discount rate, the stock price decreases.

According to Huang, Masulis and Stoll (1996), a change in the oil price also influences the interest rate. When the oil price is higher relative to the general price level, the oil price can cause the real interest rate to rise. If the real interest rate rises, the discount rate also increases, and this will lead to a lower stock price. Cologni and Manera (2008) examined the influence of an oil price change on macroeconomic variables of the G-7 countries. They find that an increase in the oil price does lead to an increase in the interest rate. Cologni and Manera (2008) argue that the governments of these G-7 countries increase the interest rate as a monetary policy instrument to force the inflation rate to its target level.

2.3. Overview of related literature

(9)

market return of the United States and the stock market return of Japan. The variables in their VAR model are based on the paper written by Chen, Roll and Ross (1986). Chen, Roll and Ross (1986) examine the influence of several variables on the stock market return of companies listed on the New York Stock Exchange during January 1958 to December 1983. Chen, Roll and Ross (1986) find a positive and significant effect of a change in the industrial production, risk premium, yield and yield curve on the stock return. The stock return will be statistically negatively influenced by a change in the unexpected and expected inflation (Chen, Roll and Ross (1986)). Chen, Roll and Ross (1986) find a statistically negative influence of an oil price change on the stock return of NYSE listed companies during the period 1958-1967. Kaneko and Lee (1995) included the same variables as Chen, Roll and Ross in their model. Beside these variables, Kaneko and Lee (1995) also included the international variables like the change in terms of trade (export price index/import price index) and the exchange rate in their model to examine the influence of an oil price change on the Japanese stock market return. They included these international variables in their model since Hamao (1988) find that these variables have an influence on the Japanese stock market return. Kaneko and Lee (1995) find that the term premiums, risk premiums and the growth rate of the industrial production have a significant effect on the U.S. and Japanese stock market returns. For the Japanese stock market they also conclude that a change in the oil price has a significant negative effect on the stock market return. Kaneko and Lee (1995) only examine the influence of a change in the exchange rate on the Japanese stock market return and they define the exchange rate as the change in the yen/U.S. dollar spot rate. This means that they do not make a distinction between appreciation and depreciation of the Yen. According to Kaneko and Lee (1995) a change in the exchange rate leads to a negatively significant effect on the Japanese stock market return. The influence of a change in the terms of trade does lead to a positively significant effect on the stock market return of Japan.

(10)

(11)

Sardosky (1999) examines the relationship between the stock return and macroeconomic variables and the oil price. In order to select the appropriate macroeconomic variables, Sardosky (1999) refers to the research conducted by Kearney and Daly (1998) and Koutoulas and Kryzanowksi (1996). Koutoulas and Kryzanowski (1996) examine which variables are of influence on the return of the Canedian stock market index by using an Arbitrage Pricing Model and find a negative significant effect of a change in the interest rate on the stock index return and a positive significant effect for the change in the industrial production on the stock index return. Kearney and Daly (1998) examine the influence of several variables on the Australian stock market index return by using Generalized Least Squares estimation procedure. They found that the interest rate, inflation and industrial production influence the return of the Australian stock market index. This influence is positively significant for the industrial production and negatively significant for the interest rate and inflation. Based on these two papers, Sardosky (1999) included the interest rate, industrial production and the oil price in his VAR model. He adjusted the oil price and the Standard and Poor 500 return with the inflation rate. Sardoksy finds that a change in the oil price does have a significant and detrimental effect on the return of the Standard and Poor 500. Beside that, in the post-1986 period, the oil price has a larger impact on the return of the Standard and Poor 500 compared with the pre-1986 period.

Hondroyiannis and Papapetrou (2001) examine the influence of a change of the oil price and the changes of macroeconomic variables on the Greek stock market. Beside the industrial production and the interest rate, they also included the exchange rate and a foreign stock market (Standard and Poor 500). The influence of a change in the foreign stock market does not have a significant impact on the Greek stock market. Of the variables under consideration, the influence of a change in the exchange rate on the stock market is the highest after all of the five periods and the influence of a change in the oil price on the Greek stock market is negative and significant.

(12)

Beside Gjerde and Sættem (1999), Park and Ratti (2008) also examine the influence of an oil price change on the Norwegian stock market index. Like Gjerde and Sættem (1999), they conclude that a change in the oil price affects the Norwegian stock market index return positively. While Gjerde and Sættem (1999) conclude that this effect is not significant, Park and Ratti (2008) find significant results. Beside Norway, Park and Ratti (2008) also investigate the influence of an oil price change on the return of the stock market index of 12 European countries and the United States. The framework which Park and Ratti (2008) use is based on Sardosky (1999). In the case of the 12 other European countries and the United States, a change in the oil price does have a significant negative effect on the return of the stock market index of these countries.

2.4. Macroeconomic variables

According Chen, Roll and Ross (1986) the macroeconomic variables they selected do have a significant effect on the return of a stock. Like Chen, Roll and Ross (1986), Gjerde and Sættem (1999) examine the influence of macroeconomic variables on the stock return. They use a different set of variables than Chen, Roll and Ross (1986). The VAR model does not specify which macroeconomic variables must be included. However, it is clear that beside the oil and the stock return other variables must be included in the VAR model to handle possible indirect effects of oil price changes on the stock market by macroeconomic variables. An example of this indirect effect is the increase in the interest rate by an increase in the oil price when the oil price is relatively higher than the general interest rate. This increase in the interest rate leads to a higher discount rate and therefore the stock price will go down. To incorporate the interest rate in the VAR model, this indirect effect will be cancelled out.

(13)

2.4.1. Industrial production

One of the systematic factors which influence the stock return is the cash flow. By discounting the expected future cash flow one will get the current stock price. This follows from the equation (1). It is possible to use profits or investments as a cash flow variable, however changes in these indicators for cash flows are small and often unreliable (Fama, 1990). To make the effects of data mining as low as possible, a production variable is a better proxy for the cash flow (Fama, 1990) since using cash flows will lead to more extreme values compared with a production variable. These extreme values lie within the rejection region of the null hypothesis. As a consequence there will be always a significant relationship. By using a production variable, the influence of extreme values will be lower and therefore a production variable is a more precise proxy for the cash flow variable. Several real activity variables (like industrial production, growth rate of real GNP and Gross Private Investment) can be used as a production variable. Of these real activity variables, the industrial production explains the most stock return variation and therefore, the industrial production is the best proxy for the cash flow (Fama, 1990).

2.4.2. Interest rate

According to equation (1), a change in the interest rate leads to a change in the discount rate and therefore it will lead to a change in the stock price. An increase in the interest rate influences the stock return in a negative way in several ways (Sardosky (1996). A change in the interest rate alters the relationship between competing financial assets, since the risk-free rate of return goes up and as consequence the risk premium declines. The risk premium is what an investor receives for bearing risk by holding a risky asset. When the risk premium declines, the willingness of an investor to buy a risky asset might decline. In order to attract debt holders, the corporation needs to increase the interest rate on debt. If the firm transfers these cost to the consumer, the cash flow will decline and, as follows from equation (1), the stock return will decrease. Beside that, an increase in the interest rate also affects the stock return in second way. When stocks are purchased on margin, a change in the costs of carrying margin debt influences the desire and/or the ability for investors to speculate. Consequently, an increase in the interest rate lowers real stock returns.

2.4.3. Exchange rate

(14)

(1995) included the exchange rate in his model since this follows from the results of Hamao (1988). Hamao (1988) concludes that the international trade is important for the Japanese economy. Therefore Kaneko (1995) concluded that a variable which affects the Japanese stock market should be included in the VAR model to cover the influence of the international trade on the cash flow. Kaneko (1995) included two international trade variables. These variables are the exchange rate and the changes in terms of trade (export price index/import price index). For the latter, Kaneko and Lee (1995) did not find any influence (see also Hamao (1988). Hondroyiannis and Papapetrou (2001) included the exchange rate in the VAR model because Greece depends upon the import of energy. As well Kaneko and Lee (1995) as Hondroyiannis and Papapetrou (2001) find that the influence of the exchange rate is the highest on the stock market when compared with the other variables.

When the domestic currency depreciates the domestic stock return will increase, since the competitive position of the domestic country increases and this will lead to a higher profitability. This is in line with the results of Hondroyiannis and Papapetrou (2001). By taking the exchange rate in consideration, this paper differs from the paper written by Park and Ratti (2008) about thirteen European countries and the United States. Like us, Park and Ratti (2008) examined the influence of a change in the oil price on a particular stock market by using a VAR model.

2.4.4. Inflation

(15)

Table 3 Nominal versus real variables

Nominal Real

Stock return Kaneko & Lee (1995)

Huang, Masulis & Stoll (1996) Sardoksy (1999) Gjerde & Sættem (1999)

Hondroyiannis & Papapetrou (2001) Park & Ratti (2008)

Oil price Kaneko & Lee (1995)

Huang, Masulis & Stoll (1996) Sardosky (1999)

Gjerde & Sættem (1999)

Interest rate Kaneko & Lee (1995)

Huang, Masulis & Stoll (1996) Sardosky (1999)

Industrial production and exchange rate

Kaneko & Lee (1995) Sardosky (1999)

Exchange rate Gjerde & Sættem (1999)

3. Data Description

3.1. Data sources and variable descriptions

This research examines the influence of oil price changes on stock markets in thirteen European countries by using a vector autoregressive (VAR) model. Monthly data are used on all variables. This follows several comparable researches (Brubridge and Harrison, 1984; Faff and Brailsford, 1999; Papapetrou, 2001). Due to data availability, the research period is from January 1997 to February 2008. The sources of data and variable descriptions used in this research are presented in table 4. Table 5 contains the codes associated with the raw data series of DataStream.

(16)

Table 4 Data description and its resources

Symbol Variables Data Source

S Total return index MSCI

Oil The price of one barrel of crude oil Brent crude oil price on NYMEX

IP Industrial production (base year 2000=100) IMF international financial statistics r Annualized monthly return on 3 month

government bond Austria: FBE & ACI Belgium: Banque Nationale de Belgique Denmark: Denmarks National Bank Finland: FBE & ACI

France: FBE & ACI Germany: FBE & ACI Ireland: FBE & ACI

Italy: Ministero Dell’Economia E Della Finance The Netherlands: FBE & ACI

Norway: Norges Bank Portugal: Banco de Portugal

Spain: Analistas Financerios International United Kingdom: British Bankers’ Association EX Exchange rate to US dollar GTIS-FTID and Eurostat

I Consumer price index (base year 2000=100) Main Economic Indicators, OECD

Notes: 1) The frequency is monthly and the period runs from January 1997 until February 2008 2) The abbreviations stands for: MSCI = Morgan Staley Capital International; IMF = International Monetary Fund; FBE = European Banking Federation; ACI = Financial Market Association; OMX AB =Aktiebolaget Optonsmäklarna (Helsinki Stock Exchange); GTIS = Global Trade Information Service; FTID = Financial Times Interactive Data; OECD = Organisation for Economic Co-operation and Development.

Table 5 Codes of the data series from Datastream

Notes: 1) Codes are derived from ThompsonDataStream 2) Source of each data series are presented in table 3. 3) Code for the crude Brent oil price is OILBRNP(P)

The empirical research of Chassard and Halliwell (1986) reveals that NYMEX (the New York Mercantile Exchange) crude oil futures are unbiased predictors of subsequent spot oil prices. However, due to data availability, spot oil prices are used in the research. The real oil price changes are the difference between changes of crude oil price and the changes of inflation rates. We calculate the changes in industrial production at time t as the first difference of the natural logarithms of the index, denoted by lnIPt. The interest rates used in this research are the monthly yields of 3-month

government Treasury bills. The interest rates are obtained by converting the annualized 3-month government Treasury bill yields into monthly yields. The transformation is shown in table 6.

Furthermore, in order to test the exposure to currency changes, the exchange rate of local currency to

Country Stock return Industrial

production

Interest rate Exchange rate Inflation

Austria MSATLRL(RI) OEI66…CE ASVIB3M USAUSCH OEOCP009F

Belgium MSBELGL(RI) BGI66…CE BGTBL3M USBELCM BGOCP009F

Denmark MSDENL(RI) DKI66…CE CIBOR3M USDANKR DKOCP009F

Finland MSDINDL(RI) FNI66…CE FNIBF3M USFINMK FNOCP009F

France MSFRNCL(RI) FRI66…CE PIBOR3M USFRNFR FROCP009F

Germany MSGERML(RI) BDI66...CE FIBOR3M USWGMRK BDOCP009F

Ireland MSIREL(RI) IRI66…CE EIRED3M USIRISH IROCP009F

Italy MSITALL(RI) ITI66…CE ITBT03G USITALR ITOCP009F

The Netherlands MSNETHL(RI) NLI66…CE AIBOR3M USNETHG NLOCP009F

Norway MSNWAY(RI) NWI66…CE NWIBK3M USNORGK NWOCP009F

Portugal MSPORDL(RI) PRI66…CE BBPTE3M USPORTE PTOCP009F

Spain MSSPANL(RI) ESI66…CE ESMIB3M USSPANP ESOCP009F

(17)

US Dollar is also included in the empirical analysis. Except Denmark, Norway and UK from January 2000, ten countries in our research adopted Euro. Thus, from January 2000, the exchange rates of local currency to US Dollar are obtained by first converting local currencies to Euro, then converting to US Dollar.

Table 6 Transformation of the variables

Symbol Variable Transformation

∆lnRS Real stock return

_ln(S

_it

_/S

_it-1

_)-

_{∆ln I}_it

∆lnROil Real oil price changes ln(Oilit / Oilit-1) - ∆lnI it

∆lnIP Real changes of the industrial production ln(IPtt / IPit-1) - ∆lnI it

∆lnr Real changes of the interest rate ln((1+(rit/12)) / (1+(rit-1/12) - ∆lnI it

∆lnEX Changes of the exchange rate ln (EXit / EXit-1)

∆lnI Changes of the inflation ln (Iit / Iit-1)

Notes: 1) Raw data sources are presented in table 3 and the associated codes from DataStream are in table 2) The abbreviations stands for: S = Monthly return on country equity index; I = Inflation; Oil = The price of one barrel of crude oil; IP = Industrial production; r = monthly return on government bond; EX = Exchange rate to US dollar;

3.2. Data descriptive statistics

Tables 7.1-7.5 show the statistical summary of all variables. Table 7.1 shows the statistical summary of real stock returns. The variable real stock returns proxies the performance of the stock market in each country. Finland has the highest mean stock return index (monthly yield 1.2%) while Ireland has the lowest means stock return index (monthly yield 0.1%) in the research period. The standard deviation is a proxy for volatility. Finland has the highest volatility, 9.8% on a monthly basis, and the stock return index of Austria has the lowest volatility, 2.8% on a monthly basis. The distribution of real stock returns is measured by skewness, kurtosis and Jarque-Bera statistics. Skewness measure the asymmetry of the distribution of the data series and kurtosis measures the extent to which the data series is peaked or flatted compared to a normal distribution. The skewness and kurtosis of a normal distribution is zero and three. Jarque-Bera statistics and its probabilities indicate whether the data series is normally distributed. Denmark is the only country with a positive skewness, which indicates a long right tail. All the thirteen have a kurtosis above three, which means the distributions of all data series are more peaked relative to the normal distribution. The probabilities of Jarque-Bera show that only the data of Portugal is normally distributed at a 5% significant level. Normal distribution of data is not necessary in a VAR model, so the non-normality of the real stock returns will not bias the results from the VAR model in the empirical research.

(18)

Table 7.1 Descriptive statistics of the real stock returns of 13 European countries

Notes: 1) The frequency is monthly and the period runs from 1997:1 – 2008:2. There are a total of 133 observations. 2) The mean is calcualted as the average of formula: (Natural log of (country i St / country i St-1)) - ∆lnI where S is the monthly return on country equity index, ln is the natural log, and I is the inflation rate (calculated as (Natural log of (country i It –It-1)) 3) The data are derived from Datastream

Table 7.2 Descriptive statistics of the real oil price of 13 European countries

Notes: 1) The frequency is monthly and the period runs from 1997:1 – 2008:2. There are a total of 133 observations. 2) The mean is calculated as the the average of formula (Natural log of (Oilt / Oilt-1) – ∆lnI where Oil is the price of one barrel of Brent crude oil and I is the inflation rate (calculated as (Natural log of (country i It –It-1)) 3) The data are derived from Datastream

Table 7.3 Descriptive statistics of the growth rate of industrial production of 13 European countries

Notes: 1) The frequency is monthly and the period runs from 1997:1 – 2008:2. There are a total of 133 observations. 2) The mean is calculated as the the average of formula (Natural log of (country i IPt / IPt-1)) where IP is the industrial production 3) The data are derived from Datastream

Mean Median Maximum Minimum Standard deviation

Skewness Kurtosis Jarque-Bera Probability

Austria 0.007 0.013 0.139 -0.202 0.028 -0.776 4.015 19.051 0.000 Belgium 0.005 0.016 0.132 -0.186 0.054 -0.991 4.443 33.330 0.000 Denmark 0.011 0.015 0.282 -0.167 0.067 0.104 4.432 11.602 0.003 Finland 0.012 0.017 0.274 -0.279 0.098 -0.333 3.919 7.138 0.028 France 0.006 0.013 0.121 -0.220 0.056 -1.000 4.754 39.214 0.000 Germany 0.006 0.013 0.162 -0.273 0.066 -0.824 4.944 35.986 0.000 Ireland 0.001 0.009 0.113 -0.205 0.059 -0.783 3.632 15.811 0.000 Italy 0.006 0.010 0.198 -0.237 0.060 -0.253 5.174 27.605 0.000 Netherlands 0.003 0.011 0.130 -0.226 0.058 -1.043 4.832 42.715 0.000 Norway 0.008 0.019 0.151 -0.292 0.061 -1.293 6.599 108.846 0.000 Portugal 0.006 0.012 0.154 -0.180 0.058 -0.302 3.598 4.000 0.135 Spain 0.009 0.014 0.145 -0.253 0.064 -0.771 4.915 33.510 0.000 United Kingdom 0.004 0.006 0.087 -0.148 0.042 -1.023 4.826 41.679 0.000

Austria 0.010 0.020 0.179 -0.226 0.096 -0.423 2.521 5.242 0.073 Belgium 0.010 0.020 0.179 -0.222 0.095 -0.416 2.520 5.106 0.078 Denmark 0.010 0.018 0.179 -0.233 0.095 -0.396 2.502 4.858 0.088 Finland 0.010 0.020 0.174 -0.222 0.095 -0.411 2.495 5.169 0.075 France 0.011 0.020 0.177 -0.225 0.095 -0.414 2.511 5.312 0.070 Germany 0.013 0.015 0.240 -0.244 0.108 -0.302 2.377 4.164 0.125 Ireland 0.008 0.020 0.175 -0.227 0.096 -0.418 2.513 5.178 0.075 Italy 0.010 0.025 0.201 -0.238 0.091 -0.376 2.832 3.285 0.193 Netherlands 0.009 0.022 0.199 -0.232 0.090 -0.369 2.814 3.214 0.200 Norway 0.009 0.023 0.201 -0.242 0.091 -0.376 2.829 3.304 0.192 Portugal 0.013 0.030 0.192 -0.241 0.085 -0.675 3.263 10.492 0.005 Spain 0.009 0.028 0.201 -0.241 0.092 -0.484 2.980 5.193 0.075 United Kingdom 0.009 0.022 0.196 -0.238 0.090 -0.372 2.825 3.231 0.199

(19)

Table 7.4 Descriptive statistics of the interest rate of 13 European countries

Notes: 1) The frequency is monthly and the period runs from 1997:1 – 2008:2. There are a total of 133 observations. 2) The mean is calcualted as the average of formula: (Natural log of (country i St / country i St-1)) - ∆lnI where S is the monthly return on country equity index, ln is the natural log, and I is the inflation rate (calculated as (Natural log of (country i It –It-1)) 3) The data are derived from Datastream

Table 7.5 Descriptive statistics of the exchange rate in 13 European countries

Notes: 1) The frequency is monthly and the period runs from 1997:1 – 2008:2. There are a total of 133 observations. 2) The mean is calculated as the the average of formula (Natural log of (country i EXt / EXt-1)) where EX is the exchange rate to US dollar 3) The data are derived from Datastream

3.3. Unit root tests

The standard theory of inference in regressions with stochastic regressors requires the data of all variables to be stationary. Stationarity means the data has a constant mean, constant variance and constant autocovariances for each specified lag length. Non-stationary data contains unit roots and regressions on variables with unit roots will increase conventional standard errors (Brooks, 2002). Therefore, unit root tests will be carried out to examine the stationary of the data of all variables. In these tests, we will use the critical values from augmented Dickey and Fuller (1979) (ADF test). Once the evidence of non-stationary data is observed, the stationarity of the first difference of the data will be tested. If the first difference of the data is observed to be stationary, it will be used in the VAR model.

A simple autoregressive (1) process of yt can be expressed as:

yt = ρyt-1 + δxt + εt , (2)

where xt is an optional exogenous regressor, which may consist of a constant or a constant and a trend,

Austria 0.000 0.000 0.049 -0.033 0.009 0.537 9.095 212.278 0.000 Belgium 0.001 0.000 0.045 -0.028 0.010 0.386 6.030 54.166 0.000 Denmark 0.000 0.000 0.049 -0.037 0.013 0.411 6.256 62.479 0.000 Finland 0.001 0.001 0.050 -0.033 0.040 0.496 8.735 187.722 0.000 France 0.000 0.000 0.049 -0.033 0.009 0.515 9.482 238.722 0.000 Germany 0.001 0.000 0.045 -0.028 0.010 0.294 6.610 74.150 0.000 Ireland -0.001 0.000 0.049 -0.082 0.014 -1.432 13.183 620.137 0.000 Italy 0.010 0.025 0.201 -0.238 0.091 -0.374 2.824 3.278 0.194 Netherlands 0.001 0.001 0.043 -0.024 0.009 0.446 7.463 11.779 0.000 Norway 0.000 0.001 0.124 -0.059 0.019 1.530 16.860 1116.424 0.000 Portugal -0.001 0.000 0.049 -0.033 0.010 0.403 7.219 102.225 0.000 Spain -0.001 0.000 0.043 -0.062 0.012 -0.711 9.861 272.115 0.000 United Kingdom 0.000 0.000 0.030 -0.033 0.010 -0.440 5.151 29.916 0.000

(20)

non-stationary series. If │ρ│< 1, yt is a stationary series. The ADF test involves an ordinary least

squares (OLS) estimation of the following equation:

∆yt =αyt-1 + δxt + β1∆yt-1 + β2∆yt-2 +…+ βp∆yt-p + νt (3)

Equation (3) subtracts yt-1 from both sides of equation (2) and adds p lagged difference terms of yt on

the right side of equation (2). α equals to ρ-1. The ADF test tests the null hypothesis H0: α=0, which

means non-stationarity, against the one-sided alternative hypothesis H1: α<0 (Dickey and Fuller (1979)

and Davidson and MacKinnon (1993)).

Table 8 reports the results of unit roots tests with the null hypotheses that each variable contains a unit root (non-stationary). The ADF test statistics of changes of industrial production in Germany indicate the null hypothesis of non-stationary can not be rejected. This means the data of changes of industrial production in Germany contains a unit root and is not stationary. The unit roots test is performed on the first difference of the changes of industrial production (∆∆ln IP) in Germany. The

result shows that the first difference of the changes of industrial production in Germany is stationary. Therefore, the first difference of the changes of industrial production in Germany will be used in the VAR model. The null hypothesis of the other data can be rejected at 1% significant level, so these data are not subject to unit roots and they can be used directly in the VAR model.

Table 8 Augmented Dickey and Fuller unit root results of 13 countries

∆lnRS ∆lnROil ∆lnIP ∆∆lnIP ∆lnr ∆lnEX

Austria -11.49* -11.50* -3.95* - -7.75* -11.05* Belgium -10.60* -11.46* -3.60* - -9.02* -11.01* Denmark -11.26* -11.49* -13.18* - -5.63* -10.60* Finland -9.40* -11.43* -15.39* - -7.74* -10.91* France -10.57* -11.52* -19.90* - -7.84* -11.04* Germany -10.57* -11.53* -2.15 -8.82* -8.28* -11.03* Ireland -10.71* -11.45* -11.82* - -5.52* -11.10* Italy -12.10* -11.29* -3.64* - -9.83* -11.02* Netherlands -10.74* -11.28* -3.99* - -6.92* -10.99* Norway -4.81* -11.38* -11.45* - -5.19* -11.76* Portugal -10.61* -11.55* -3.86* - -7.75* -11.01* Spain -11.48* -11.84* -18.67* - -5.32* -11.26* United Kingdom -11.76* -11.33* -3.91* - -4.67* -9.02*

Notes_: ∆lnRS is the real stock return, ∆lnOIL is the changes of oil prices, ∆lnIP is the changes of industrial productions, ∆lnR is the changes

of interest rates and ∆lnEX is the changes of exchange rates. * denotes that a test statistic is significant at the 1% level of significance. The

(21)

4. Research Methods

4.1. The empirical VAR framework

A vector autoregressive (VAR) framework is employed to examine the impact of changes of oil price on stock markets for the following reasons. First, VAR models are compact in presenting lag terms in equations and are able to capture the dynamic relationships among the economic variables. The impulse response function and the variance decomposition derived from the VAR model can trace out the response of all variables in the VAR model simultaneously to an one-time shock. This feature of VAR models is helpful for analyzing the historical movements in stock market and real oil price changes. Second, in VAR models it is not necessary to specify which variables are exogenous or endogenous because all variables are endogenous. As there are no restrictions on variables, VAR models are favourable in exploring the dynamic interaction between variables. Third, VAR models allow the value of one variable to depend on its own lags and lags of other variables (Brooks (2002)). This feature helps to examine the influence on stock markets from prior changes of oil price.

A VAR model is a system of equations, where each variable is expressed as a linear function of its own lagged values and lagged values of other variables in the system. A general VAR model with k variables can be expressed as:

yt= a0 +

∑

= p 1 i Ai yt-i + ut , (4)

where yt (y1t, y2t, ... ykt) is a column vector of the current values of all variables (∆lnRS, ∆lnOIL, ∆lnIP,

∆lnR and ∆lnEX) in the VAR model, a0 is a column vector of intercepts (constant terms), Ai is a k×k

coefficient matrix, yt-i is the vector of lag values of all variables, p is the lag length and ut (u1t, u2t, … ,ukt) is a vector of error terms. With stationary data, the error terms should be normally distributed. The

lag length p should be chosen the same for all variables in one country. The reason of an identical lag length of different variables and the determination of lag length p will be explained in the next sub-section.

In equation (4), the changes of yt are composed of two parts: the expected changes, come from the

prior changes of the k variables,

∑

=

p

1 i

Ai yt-i and the unexpected changes are due to changes in the

error term ut. The unexpected changes can be interpreted as shocks to yt. So the changes of one

(22)

4.2. Determining the lag length

Determining the lag length is a critical element in the specification of a VAR model. The importance of determining the lag length has been explored by several researches. Braun and Mittnik (1993) show that misestimates of the lag length of a VAR will bias the results of the impulse response functions and variance decompositions derived from the estimated VAR. A VAR with different lag lengths for each variable is viewed as a restricted VAR. Brooks (2002) and Sims (1972) prove that when the specification of all equations in one model is the same, the ordinary least squares estimation generates efficient parameter estimates. As well as for the simplicity of estimation, a VAR model should be estimated as unrestricted as possible. Therefore, we will use an unrestricted VAR in our research. The lag lengths of different variables in all equations of a VAR are the same and they are called symmetric lags. Since model (country)-specific lag length should be conducted, different countries may have different lag lengths.

The likelihood ratio (LR) test and the information criterion test are two easy and most widely used tools of lag length selection (Brooks, 2002). However, the LR test has a major limitation. As a Chi-squared distribution (χ2_{) test is included in the LR test, the LR test is subject to the assumption that the}

error terms from each equation are required to be normally distributed. The statistical summary of all variables in Section 3.2 shows that the data in this research do not follow a normal distribution, so the LR test is not applicable in this research. The information criterion is not constrained by the normal distribution assumption so we will apply the information criteria to select the lag length. The information criterion has two versions, the univariate criteria and the multivariate criteria. The multivariate information criteria is applicable to unrestricted VAR models so the multivariate information criteria will be used in this research. The definition of the modified Akaike’s information criteria (MAIC) used by Eviews can be presented as

MAIC = log │

∑

∧ +

T

2k'_{, (5)}

where

∑

∧ is the variance-covariance matrix of residuals, k’ is the number of regressors in the VAR model, which is equal to (k2_{p + k) for k is the number of equations in the VAR model and p is the lag}

(23)

Table 9 Lag length selection

Country Austria Belgium Denmark Finland France Germany Ireland

Lag length 12 12 11 12 12 12 12

Country Italy Netherlands Norway Portugal Spain United Kingdom

Lag length 12 12 11 12 12 12

Note: the lag lengths are determined by the modified Akaike’s information criteria (MAIC), MAIC = log │

∑

∧

+

T 2k'

4.3. The impact of changes on the financial market

Impulse responses can be used to examine the dynamic effect of oil price changes. Impulse responses trace out the responsiveness of variables on the left hand-side of the equations to shocks in each of the variables in the VAR model over time. An impulse response function traces the impact of a one-time shock in each variable on the current and (expected) future values of variables simultaneously in the VAR model (Pesaran and Shin (1998)). In this research, we use a period of 36 months to trace the impact of shocks. When examining the impact of shocks in one equation, the error terms of other equations are assumed to be constant. One unit shock is applied to the error term of each variable, and then the impacts on the real stock returns over 36 months are plotted.

One important issue in the impulse responses might be the correlation between shocks to different variables. If all the shocks at time t are uncorrelated, the impulse response functions are straightforward. The generalized impulse response can be applied. However, besides the shock originated variable- the oil price, the other four variables (∆lnOIL, ∆lnIP, ∆lnR and ∆lnEX) are usually correlated. Thus, the shocks to these variables are correlated. This correlation can be viewed as having a common component, which can not be associated with any individual variable. A proper transformation Q has to be applied to the shocks to make them uncorrelated.

χt= Q t ~ (0, D) (6)

In equation (6), χt (χ1t, χ2t, …, χkt) is a vector of uncorrelated shocks, Q is the proper transformation, it

can be interpreted as the effects of the components of the one-unit shock process t on the process χt at

the lag j, ηt (η1t, η2t, … , ηkt) is a vector of shocks to all variables, D is a diagonal covariance matrix.

(24)

The variance decomposition offers another way of examining the dynamic effects of oil price changes. It divides the variation in one variable into proportional component shocks to each of the variables in the VAR model. The variance decomposition shows the relative importance of each shock in affecting the variables in the VAR model. A forecast error in the VAR model is the difference between the actual value of one variable and the forecasted (expected) value of this variable. The source of this forecast error is the variation in the current and future values due to the shock to each variable. The variance decomposition shows the percentage of the forecast error variance caused by one unit shock to each variable in the VAR model by Monte Carlo simulation process. The Monte Carlo simulation process is used to construct the standard error. Like comparable researches we will conduct 1000 repetitions. Although many researches, like Sardosky (1999), Gjerde and Sættem (1999) and Park and Ratti (2008), only provide the results of the Monte Carlo simulation process, they do not use this result to determine whether or not the results of the variance decomposition are significant. Since it is appropriate to follow comparable researches, we will not use the Monte Carlo standard error to determine if the results are significant or not. This means that it is not possible to provide any information about the significance of the results.

A change to one variable will directly affect its own value and it will also be transmitted to all the other variables in the VAR model. In this section, we will document variance decompositions and orthogonalized impulse responses to examine the impact of shocks to the variables in the VAR model on real stock returns. Park and Ratti (2008) examine the influence of oil price changes on stock markets in thirteen European countries and the United States. They find the ordering of variables in the VAR model has a minor impact on the results of the variance decomposition and orthogonalized impulse response. We will derive the variance decomposition and orthogonalized impulse response from the VAR model. The ordering of variable in our VAR model is indicated as: real stock returns, the changes of the oil price, the industrial production, the interest rate and the exchange rate. The relationship between real stock returns and changes of oil price is the focus of this paper, so they are placed as the first two variables.

5. Results

5.1. Variance decomposition

(25)

rate on the real stock return after 1, 2, 3, 6, 12, 24 and 36 months. The sum of all of the changes, of a particular month, is equal to 100%.

The number which is presented in the tables represents the contribution of each source of change to the forecast error variance of the return of the real stock market. The number shown in parentheses is the Monte Carlo constructed standard error of the variable on the return of the real stock market after 1000 repetitions (the number of repetitions follows from comparable researches). Like comparable researches, we only present the Monte Carlo and therefore we will not interpret this result. This means we will not provide information about significance from the results of the variance decomposition.

The influence of a change in the return of the stock market on itself is shown in table 10.1. In the first month, 100 percent of the variability of the real stock return is attributed to changes in itself for all countries. For the Netherlands, the variability of the real stock return is the lowest after 2, 3, 6, 12, 24 and 36 months. This means that the real stock return of the Netherlands will be influenced the most by other variables compared with the other 12 countries. The return of the stock market of Austria explains itself the most compared with the other European countries after 6, 12, 24, 36 months.

Table 10.1 Variance decomposition of a change in the stock return on the stock return

Months 1 2 3 6 12 24 36 Austria 100 (0) 92.54 (6.04) 90.09 (6.64) 81.61 (7.98) 73.69 (8.51) 69.31 (9.51) 67.71 (11.20) Belgium 100 (0.00) 80.87 (7.86) 79.13 (7.95) 72.29 (8.20) 61.91 (7.92) 56.87 (9.31) 55.59 (10.95) Denmark 100 (0.00) 88.57 (7.02) 86.09 (7.46) 73.16 (8.01) 55.65 (8.00) 48.62 (8.74) 47.34 (10.71) Finland 100 (0.00) 91.98 (6.21) 87.00 (7.62) 77.21 (7.99) 64.19 (7.82) 57.77 (8.89) 57.42 (10.71) France 100 (0.00) 88.18 (7.05) 80.14 (7.99) 64.21 (7.97) 51.54 (7.53) 47.35 (8.02) 46.91 (9.41) Germany 100 (0.00) 90.52 (6.34) 88.78 (7.21) 77.75 (8.08) 66.02 (7.62) 60.43 (8.79) 59.02 (10.50) Ireland 100 (0.00) 95.27 (5.82) 90.27 (7.36) 70.33 (8.54) 61.39 (7.88) 57.79 (8.90) 56.97 (10.54) Italy 100 (0.00) 79.31 (7.89) 73.55 (8.40) 67.57 (7.90) 58.30 (7.83) 52.52 (8.80) 51.66 (10.53) Netherlands 100 (0.00) 77.64 (8.36) 73.79 (8.57) 55.53 (7.82) 48.37 (7.10) 42.52 (7.63) 41.17 (8.87) Norway 100 (0.00) 95.37 (5.03) 87.01 (6.84) 77.04 (7.94) 63.39 (8.28) 58.51 (8.46) 57.36 (10.36) Portugal 100 (0.00) 87.39 (7.13) 83.88 (7.62) 76.26 (7.98) 69.48 (8.28) 64.54 (9.27) 63.16 (10.68) Spain 100 (0.00) 84.64 (7.88) 81.42 (8.31) 69.20 (8.66) 60.03 (8.34) 55.98 (9.00) 53.73 (10.45) United Kingdom 100 (0.00) 87.20 (7.56) 85.76 (7.68) 72.63 (7.61) 61.08 (7.18) 54.48 (7.84) 53.51 (9.20)

(26)

This result indicates that influence of the other variables on the real stock return of Austria is the lowest and thus the real stock return of Austria has the highest self explaining power. For all of the countries, the highest contribution of forecast error variance does come from a change in the return of the real stock market itself after all of the months under consideration. This result is in line with the literature (Ferderer (1996), Sardosky (1999), Park and Ratti (2008)).

From table 10.2 we observe that a change in the oil price leads to volatility in the real stock returns. This table shows that an oil price change accounts for the lowest influence after 2 and 3 months for Denmark and Norway. These two countries are the only oil exporting countries under examination. This result indicates that the variability of the stock return of oil exporting countries will be influenced less by an oil price change compared with oil importing countries after a short period of time. The result for Denmark after 12, 24 and 36 is interesting. While the influence after 2, 3 and 6 months is low compared with the other countries, the result after 12, 24, and 36 is the highest.

Table 10.2 Variance decomposition of a change in the oil price on the stock return

Months 1 2 3 6 12 24 36 Austria 0.00 (0.00) 4.89 (4.89) 4.92 (4.86) 9.33 (6.45) 13.61 (7.31) 13.50 (8.16) 13.49 (9.62) Belgium 0.00 (0.00) 10.34 (6.36) 12.24 (6.71) 11.50 (6.51) 16.39 (6.42) 16.72 (7.00) 17.36 (8.27) Denmark 0.00 (0.00) 1.14 (3.02) 1.77 (3.68) 5.69 (5.17) 20.18 (7.88) 20.21 (8.90) 19.88 (10.95) Finland 0.00 (0.00) 4.93 (4.95) 6.92 (5.76) 9.41 (5.90) 12.29 (5.66) 13.86 (6.99) 13.73 (8.47) France 0.00 (0.00) 6.06 (5.36) 5.60 (5.17) 6.92 (5.04) 13.52 (5.87) 14.06 (6.24) 14.25 (7.54) Germany 0.00 (0.00) 3.74 (4.38) 4.10 (4.81) 4.65 (5.00) 8.52 (5.59) 9.22 (6.37) 9.29 (7.35) Ireland 0.00 (0.00) 2.87 (3.98) 4.13 (4.71) 5.75 (4.63) 6.99 (5.00) 7.55 (6.26) 7.61 (7.91) Italy 0.00 (0.00) 2.26 (3.52) 2.88 (4.04) 4.78 (5.03) 13.77 (6.25) 16.90 (7.48) 17.32 (9.24) Netherlands 0.00 (0.00) 3.02 (3.89) 3.01 (3.91) 6.67 (5.05) 12.31 (6.07) 13.54 (7.08) 13.92 (8.84) Norway 0.00 (0.00) 2.11 (3.83) 2.20 (3.62) 3.98 (4.74) 10.19 (5.69) 9.70 (6.01) 9.69 (6.96) Portugal 0.00 (0.00) 6.25 (5.10) 9.34 (5.65) 10.64 (5.66) 11.10 (5.95) 12.02 (7.09) 11.91 (8.69) Spain 0.00 (0.00) 6.25 (5.61) 6.72 (5.75) 9.65 (6.25) 13.47 (6.24) 14.20 (6.95) 14.61 (8.61) United Kingdom 0.00 (0.00) 7.54 (5.63) 7.23 (5.38) 8.12 (5.42) 13.78 (5.96) 14.10 (7.00) 14.18 (8.22)

Notes: 1) Monte Carlo standard errors after 1000 repetitions are presented in parentheses. 2) The ordering of the variables equals real stock return, real oil price, industrial production, interest rate and exchange rate.

(27)

consideration and is quite high after 2 and 3 months and quite low after 12, 24 and 36 months. The industrial production is used as a cash flow proxy in our model and therefore a change in the cash flow has a relative high impact on the stock return within/after a few months and low impact on longer periods in the future. The reverse holds for Norway. The result of the influence of a change in the industrial production on the real stock return for Norway after 24 months is in line with Gjerde and Sættem (1999).

Table 10.3 Variance decomposition of a change in the industrial production on the stock return

(28)

The last variable in our VAR model is the exchange rate and the results of the variance decomposition of the exchange rate is presented in table 10.5. According to Hondroyiannis an Papapetrou (2001), the exchange rate has the biggest impact on the forecast error variance of the real stock return of Greece after 1, 6, 12, 18 and 24 months. For the countries we examine, this holds for Germany, Italy, Netherlands and Norway. Germany, Italy and the Netherlands are open economies and therefore the exchange rate is of great importance (Kaneko and Lee (1995), Hondroyiannis and Papapetrou (2001)). A change in the exchange rate therefore does have a large impact on the real stock return since these economies highly depend on trade with other countries. A low forecast error variance of the exchange rate means that the real stock return does not depend that much on a change in the interest rate and is an indication of a less open economy.

Table 10.5 Variance decomposition of a change in the exchange rate on the stock return

Notes: 1) Monte Carlo standard errors after 1000 repetitions are presented in parentheses. 2) The ordering of the variables equals real stock return, real oil price, industrial production, interest rate and exchange rate.

(29)

the Netherlands (42.52%) and the highest for Austria (69.31%). For Austria the influence the variables oil price, industrial production, interest rate and exchange rate equals to a total of 30.69% and of this percentage, 13.50% is due to the oil price. This 13.50% is equal to 44% of the 30.69%. Table 10.7 shows the percentages of the other countries and variables.

Table 10.6 Variance decomposition of forecast error variance of European stock market returns after 24 months

Stock market return

Oil price Industrial

production

Interest rate Exchange rate

Austria 69.31 (9.51) 13.50 (8.16) 6.75 (4.41) 4.79 (7.53) 5.65 (6.37) Belgium 56.87 (9.31) 16.72 (7.00) 2.49 (3.75) 11.96 (7.47) 11.95 (6.50) Denmark 48.62 (8.74) 20.21 (8.90) 5.68 (3.95) 9.29 (7.08) 16.20 (7.70) Finland 57.77 (8.89) 13.86 (6.99) 10.21 (6.37) 10.88 (8.00) 7.29 (5.70) France 47.35 (8.02) 14.06 (6.24) 9.65 (5.30) 12.57 (6.99) 16.37 (7.47) Germany 60.43 (8.79) 9.22 (6.37) 2.59 (3.10) 9.01 (7.36) 18.74 (6.55) Ireland 57.79 (8.90) 7.55 (6.26) 10.87 (6.38) 15.95 (7.11) 7.84 (6.13) Italy 52.52 (8.80) 16.90 (7.48) 3.58 (3.01) 8.53 (6.56) 18.48 (7.52) Netherlands 42.52 (7.63) 13.54 (7.08) 4.94 (4.24) 19.54 (8.41) 19.45 (6.72) Norway 58.51 (8.46) 9.70 (6.01) 10.17 (7.68) 9.92 (7.86) 11.70 (6.12) Portugal 64.54 (9.27) 12.02 (7.09) 4.45 (2.79) 9.78 (7.26) 9.21 (7.34) Spain 55.98 (9.00) 14.20 (6.95) 3.66 (2.73) 16.27 (9.24) 9.88 (5.38) United Kingdom 54.48 (7.84) 14.10 (7.00) 9.75 (5.98) 9.97 (6.30) 11.70 (6.04)

(30)

Table 10.7 The influence of the oil price, industrial production, interest rate and exchange rate on the return of the real stock market expressed in percentages after 24 months

Percentage of influence other than real stock return

Oil price Industrial

production Interest rate Exchange rate Austria 30.69 0.44 0.22 0.16 0.18 Belgium 43.13 0.39 0.06 0.28 0.28 Denmark 51.38 0.39 0.11 0.18 0.32 Finland 42.23 0.33 0.24 0.26 0.17 France 52.65 0.27 0.18 0.24 0.31 Germany 39.57 0.23 0.07 0.23 0.47 Ireland 42.21 0.18 0.26 0.38 0.19 Italy 47.48 0.36 0.08 0.18 0.39 Netherlands 57.48 0.24 0.09 0.34 0.34 Norway 41.49 0.23 0.25 0.24 0.28 Portugal 35.46 0.34 0.13 0.28 0.26 Spain 44.02 0.32 0.08 0.37 0.22 United Kingdom 45.52 0.31 0.21 0.22 0.26 5.2. Impulse response

Orthogonalized impulse response is an alternative to obtain the information about the impact of shocks to the oil price, industrial production, interest rate and exchange rate on real stock returns. Table 11 presents the results of orthogonalized impulse responses of real stock returns to one unit shock to the other four variables in the VAR model in 36 months, with ordering: stock returns, changes of the oil price, the industrial production, the interest rate and the exchange rate. One unit shock is measured as an analytic one standard deviation. The orthogonalized impulse response functions of real stock returns to one unit shock to the oil price, industrial production, interest rate and exchange rate are presented in Figure A.1 in the appendix. In general, there is no robust evidence of significant impacts on stock returns from changes to the oil price, the industrial production, the interest rate or the exchange rate. Significant impacts are just found in partial countries in few months. In most of the periods of the 36 months, no significant impact on real stock returns is found.

(31)

Table 11 The impacts of shocks to the oil price, industrial production, interest rate and exchange rate on real stock returns in VAR(∆lnRS, ∆lnOIL, ∆lnIP, ∆lnR, ∆lnEX)

Austria Belgium Denmark Fin-

land France Germany Ireland Italy Nether- lands Norway Portu-gal Spain UK Shock to ∆lnROil N (2) P (9,10) N (2) P(7) N (2) N(2) N (2) Shock to ∆lnIP P (3) P (3) N (6,7) Shock to ∆lnr N (6) N (5,6) N(6) Shock to ∆lnEX P (2) P (2) P (2) N (6) P (2,3) P (2,3) P (3) N (4)

Notes: N/P (t) indicates negative/positive statistically 5% significant orthogonalized impulse response of real stock returns to shocks to oil

price, industrial production, interest rate and exchange rate shock with a lag of t months in the total 468 months. ∆lnROil is the changes of oil price. ∆lnIP is the changes of industrial production. ∆lnr indicates the changes of interest rate and ∆lnEX is the changes of exchange rate.

Shocks to the oil price have a negative impact on real stock returns in two months in only five countries (Belgium, France, Portugal, Spain and the United Kingdom). There is a positive impact on the real stock returns from the shock to oil price in Denmark only in nine and ten months. This can be explained by the fact that Denmark is an oil exporting country so it benefits from an increase in the oil price. This shock is consistent with the results from the variance decomposition. In Italy, the shock to oil price has a positive impact on real stock returns just in seven months. As an oil importing country, this positive impact is inconsistent with the theory in Section 2.

The impact of shocks to industrial production on real stock returns is not very robust in the thirteen countries either. Shocks to industrial production have a statistically significant impact on real stock returns in only three countries. In Austria and France, shocks to industrial production have a positive impact on real stock return after three months.

Shocks to interest rate have a negative impact on real stock returns only in Ireland, The Netherlands and Spain after five or six months. This means that when interest rate increases in these three countries, the real stock returns will decrease after five or six months. In the rest of countries, there are no significant impacts on stock returns. This result does not support the economic theory explained in Section 2.

Finally, the shocks to exchange rate do not have a strong impact on real stock returns as well. There is a positive impact of shocks to exchange rate on real stock returns merely in six countries in two or three months.

(32)

5.3. Robustness check

Koop et al. (1996) find that the order of variables in the VAR model has influences on the results of shock persistence and asymmetric effects of shocks derived from the variance decomposition and orthogonalized impulse responses from the VAR model. Alternative impulse response models with different ordering are examined as a robustness check of the responses of real stock returns to shocks to other variables. Many papers (Sadorsky 1999, Park and Ratti 2008) examining the relationship between oil prices and stock markets with VAR models put the variable real stock returns after all the other variables. Our first alternative VAR model will adopt this method. The variable real stock returns is put after the other four variables. The VAR model is indicated as: VAR (∆lnOIL, ∆lnIP, ∆lnR, ∆lnEX, ∆lnRS). The results of the orthogonalized impulse responses of real stock returns to one unit shock to the other four variables in the VAR model in 36 months are presented in Table 12. The second alternative VAR model follows the VAR model used by Park and Ratti (2008) in their robustness check. The interest rate is placed as the first variable and the rest of the variables are ordered the same as in the first alternative VAR model. The results of the orthogonalized impulse responses of real stock returns in VAR (∆lnR, ∆lnOIL, ∆lnIP, ∆lnEX, ∆lnRS) are presented in Table 13. The orthogonalized impulse response functions in these two alternative VAR models are presented in Figure A.2 and Figure A.3 in the appendix.

Table 12 The impacts of shocks to the oil price, industrial production, interest rate and exchange rate on real stock returns in VAR (∆lnOIL, ∆lnIP, ∆lnR, ∆lnEX, ∆lnRS)

Austria Belgium Denmark Fin- land

France Germany Ireland Italy Nether- lands

Norway Portugal Spain UK Shock to ∆lnROil N (1,2,3) P (9,10) N (2) N (1,2) N (1,2) N(1,2) P (1,2) N(1,2) N (2) Shock to ∆lnIP P (3) N (6) P(1,2) Shock to ∆lnr N (1) N(5,6) N (6) N (1,2) N(1) N(1) Shock to ∆lnEX P (2,3) P (2,3) P (2) N (6) P(2,3) P (2,3) P (3) N (4)

Notes: N/P (t) indicates negative/positive statistically 5% significant orthogonalized impulse response of real stock returns to shocks to oil price, industrial production, interest rate and exchange rate shock with a lag of t months in 36 months. ∆lnROil is the changes of oil price. ∆lnIP is the changes of industrial production. ∆lnr indicates the changes of interest rate and ∆lnEX is the changes of exchange rate.