Crude oil returns and stock market returns

(1)

1

Crude oil returns and stock market returns

Bachelor Thesis Organization & Finance University of Amsterdam Floris Kooij 10203354 Supervisor: Philippe Versijp February 2, 2015

(2)

2

Statement of Originality

This document is written by Student Floris Kooij who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

1. Introduction

One of the most important resource on this earth is oil. The estimated consumption of oil in 2013 was 90.379.200 barrels a day. With an average price of 100 dollars per barrel in 2013, more than 9 billion dollars is consumed worldwide on oil, making it one of the largest markets on earth (U.S. EIA, 2015). Because of the large consumption of oil by consumers and

producers, it is often used as an explanatory variable to predict changes in the economy. The macroeconomic researcher have been using oil as a predicting variable since the Yom Kippur War in 1973 and there is an increasing interest for the impact of oil on the stock markets.

Hamilton (1983) discovered that prior to almost every recession in the U.S., there has been a shock in the oil price. He is one of the first economic researchers that finds a strong relation between the change in the price of oil and the change in the real GDP. Because oil prices predict changes in the real GDP, it could also predict changes in the stock markets (Sadorsky, 1999).

Between July 2014 and January 2015 the price of oil dropped with 55%. The demand for oil decreased because of the unexpected lower economic growth in Europe and China. In this same period, the supply of oil went up. A new stream of oil from the U.S. got on the market and the largest oil producing countries, the U.S., Russia, and Saudi Arabia did not reduce their production. These two effect together have caused the massive drop in the oil price over the last 6 months (Austin, Therramus, 2015). But what is the relationship between this changes in the oil price and the stock markets?

This thesis will focus on crude oil returns of Brent and Western Texas Intermediate (WTI) since January 2000 and test whether these returns can predict the returns on the stock market. Additionally, I will be testing if there is a different effect between industries. Previous literature has shown that there are some unexpected predictions between the returns on crude oil and non-oil industries. Because of these surprising effects, I will test 17 different industries within the European market.

The remainder of this thesis is organized as follows. Section 2 gives an overview of the previous literature about this subject. Section 3 gives a description about the data that is used and the explanation about the regression model. Section 4 contains the results of the OLS regressions and will explain the interpretation of these results. Finally, section 5 concludes this thesis paper.

(5)

5

2. Related Literature

This section discusses the theoretical background on the subject and contains the findings of previous empirical research. The first part explains why oil is used for explaining dynamics on macroeconomic level. The second part explains the relationship between oil returns and stock market returns, including empirical evidence from previous research.

2.1 Why oil matters.

Oil has a larger impact on the stock market than most people assumed in the past. Oil has been an important macroeconomic factor in previous research and is being used more as an explanatory variable in the finance. Merton (1973) is one of the first that discusses the use of macroeconomic variables when estimating asset prices. Hamilton (1983) discovered a direct link between the change of the oil price and the change of the real output. An increase in the oil price predicts a decline in the countries real output. He suggest, but does not confirm, that big changes in the oil price have been causing most of the recessions of the 20th century.

The effect of oil on the economy is not symmetrical. An increase in the price of oil has more effect on the GDP than a decrease in the price of oil has. This asymmetrical effect was discovered by Mork (1989) and a few years later by Mory (1993). Mork (1989) used the same research method as Hamilton (1983) and he finds that the correlation of price of oil on the real GDP in the U.S. has become less in the years after Hamilton’s research. Hooker (1996)

challenges this relationship of the shocks in oil price on the economy and also finds that the relationship has become weaker in data after 1973.

More recent research from Jiménez-Rodríguez and Sánchez (2005) studied the impact of oil price shocks on the real economic activity of industrialized countries. They discovered a non-linear relation between the oil price and the real GDP. Research from Jones et al. (2004) studied the effect of oil price shocks since 1996 and also found an asymmetric relationship between the price of oil and the real GDP in the U.S.

2.2 What is the relation between oil and stock markets?

Based on the link between oil and the costs of companies and consumers, it would be logical to hypothesize a negative correlation between oil price and stock market returns. Expensive fuel causes the transportation and production costs go up, which leads to less profits for firms and less money for the consumers to spend in the economy (Pescatori & Mowry 2008). Oil

(6)

6 price shocks can affect corporate cash flows and has an influence on the expected inflation and expected real interest rates (Miller & Ratti 2008).

Driesprong, Jacobsen, Maat (2008) find this expected negative relation between oil prices and stock markets. This predictability is even stronger in the developed markets. Their data starts from October 1973 and it ends at April 2003. Monthly data is used because it is less noisy than the daily data. Daily data is noisy because a lot of the oil trades use contracts, which set a fixed price for a period of time. Three types of oil are used as a benchmark (Brent, West Texas Intermediate, and Dubai). The basic regression model consists of:

𝑟𝑡 = 𝜇 + 𝛼1𝑟𝑡−1𝑂𝑖𝑙 + 𝜀𝑡 𝑤𝑖𝑡ℎ 𝜀𝑡 = 𝑟𝑡− 𝐸𝑡−1[𝑟𝑡]

They test whether 𝛼₁is significantly different from zero. The coefficient found is negatively and significant for most countries within their dataset.

Pollet (2003) finds that the returns of the market and the performance of an industry can be predicted by the expected oil prices. Just like the research of Driesprong, Jacobsen, Maat, Pollet tries to predict the stock market returns with the previous month returns on oil. Park and Ratti (2008) researched the effect of oil price shocks on the stock market in the U.S. and in European countries. The conclusion of the research is that oil price shocks have an impact on the stock returns in the same and/or within the next month over the period of January 1986 until December 2005. An increase in the volatility of oil prices predicts lower stock returns in the European countries.

In previous research by Hong and Stein (1999) there is evidence that the investors in the stock markets are underreacting to the changes in oil prices. Hong et al. (2007) find that many investors are not taking other asset markets into account, such as the oil market. Both Driesprong, Jacobsen, Maat and Pollet take this under reaction into account in their research. Driesprong, Jacobsen, Maat (2008) introduce a model with lagged monthly oil price

observation. The increase in the R-squared is largest when a lag of 6 trading days is used. Pollet (2003) uses the same under reaction, but finds that it effect is less plausible for stocks that are directly related to oil production. This is expected because investors will pay more attention to the change of oil prices and the information about the price of oil when they sell or buy an oil related stock. Finally, both Driesprong and Pollet have included an oil specific industry regression. Both with the same empirical conclusion. The (lagged) oil price changes cannot predict the returns of the stocks of oil-related companies, but it can predict returns in non-oil related industries.

(7)

7

3. Research Method

This section will explain which data is used for this research and why. It includes the research model used to explain the main research question. My research will be based on the research of Driesprong, Jacobsen, Maat (2008). The difference is the time period used, the stock market I am using to make the predictions on, and the use of other non-linear variables. Driesprong, Jacobsen, Maat (2008) use a time period from 1973 until 2003. I used a time period of 2000 until 2014 and a separate test for the period after the start of the financial crisis in 2008. The reason for the difference is to test whether I find a different outcome in the two time periods and to test if there is a difference with previous empirical research. I include a variable to test the model for non-linearity. There is an additional quadratic variable of the oil return and a dummy variable for when the price of the crude oil is higher than 80 dollars. I use a price higher than 100 dollars in the period after 2008, because the price of the crude oils are almost always higher than 80 dollars in that period.

3.1 Data

The data includes a time period of January 2000 until December 2014. The returns of the Stoxx 600 index and the oil prices of Brent and WTI are monthly. There are multiple reasons for this time period and frequency:

1. The data ends at December 2014. It is the most current data available at this moment.

2. The data includes the years of the financial crisis. Because of the changing

economic environment in the years after 2008, it is interesting to see whether there is a change in the relation between oil price returns and stock returns.

3. The data starts at January 2000. The data set is entirely complete after that point, giving a more reliable view on the prediction.

4. Monthly returns show the strongest relation in previous literature and are less noisy than daily data (Driesprong, Jacobsen, Maat 2008).

There are two oil prices that will be used to predict stock returns. Brent is globally used as the crude oil benchmark. The Brent crude oil is produced in the North Sea and it produces 0.86 million barrels per day. It is most commonly traded in Europe, the Mediterranean, and Africa. The Western Texas Intermediate (WTI) is produced in the United States. It is used as a benchmark for other crude oils that are produced in America (U.S. EIA, 2014). They are both

(8)

8 light, sweet crude oil products. This means that the process costs for this kind of oil products are low on average.

The Stoxx 600 index is used as the European stock market in this research. The Stoxx 600 includes 600 stocks from firms in Europe. The 600 different stocks give a balanced view on how the stock market in Europe would react on a change in the return on oil. The Stoxx 600 returns do not include dividend reinvestments. The returns are calculated by taking the difference of the price at (t) and the price (t-1) divided by the price of the index on (t-1).

I have divided the research into separate regressions. As stated above, there will be two kinds of crude oil prices to predict the returns on the European stock market. Secondly, there will be a Stoxx 600 index in euros and a Stoxx 600 index in dollars. The reason is that the crude oil prices are in dollars. Comparing stock returns in euros to crude oil returns in dollars will give a mismatch, because the exchange rate has its own return over time.

The first regressions will be in dollars. The dollar price of the Brent/WTI crude oil will be predicting the returns of the Stoxx 600 index in dollars. The second regressions will be in euros. The crude oil price per barrel is monthly corrected for the euro/dollar exchange rate. This will give the euro price of a Brent/WTI oil barrel predicting the Stoxx 600 index returns in euros.

3.2 Model

Two models will have to be tested to answer the research question. The first model will be testing whether the crude oil returns predict the returns of the European stock market. The second model will be testing whether this effect is different for specific industries.

The first basic regression model is: (1) 𝑅𝑡𝑀𝑟𝑘𝑡 = 𝛾0+ 𝛾1∗ 𝑅𝑡𝑂𝑖𝑙+ 𝜀𝑖

Where 𝑅_𝑡𝑀𝑟𝑘𝑡_{is the return of the Stoxx 600 market index at time t and 𝑅}

𝑡𝑂𝑖𝑙 is the monthly

return of crude oil at time (t). The coefficient 𝛾1 gives the prediction how much the oil returns

have influence on the Stoxx 600 market. The constant 𝛾₀ is not relevant for the prediction, because I only want to show the relation between the returns on the crude oil and the stock market.

(9)

9 The relation between crude oil returns and the returns on the Stoxx 600 market might not be linear. Jiménez-Rodríguez and Sánchez (2005) found evidence for a non-linear impact of oil prices on the real GDP. An increase in the oil price had more effect on the real GDP than a decrease, making their model asymmetrical. To test for this non-linearity I included two additional variables to the basic regression model:

(2) 𝑅_𝑡𝑀𝑟𝑘𝑡 _{= 𝜕} 0 + 𝜕1∗ 𝑅𝑡𝑂𝑖𝑙+ 𝜕2∗ (𝑅𝑡𝑂𝑖𝑙) 2 + 𝜀𝑖 (3) 𝑅_𝑡𝑀𝑟𝑘𝑡 _{= 𝜃} 0 + 𝜃1∗ 𝑅𝑡𝑂𝑖𝑙+ 𝜃2∗ (𝑃𝑟𝑖𝑐𝑒 > 80$) + 𝜀𝑖 (4) 𝑅_𝑡𝑀𝑟𝑘𝑡 _{= 𝜔} 0+ 𝜔1∗ 𝑅𝑡𝑂𝑖𝑙+ 𝜔2∗ (𝑃𝑟𝑖𝑐𝑒 > 80$) + 𝜔3∗ 𝑅𝑡𝑂𝑖𝑙∗ (𝑃𝑟𝑖𝑐𝑒 > 80$) + 𝜀𝑖

Equation (2) includes a variable(𝑅𝑡𝑂𝑖𝑙)2. If the relation between crude oil returns and stock

returns is quadratic, the coefficient 𝜕1 should be significant. Equation (3) and (4) include a

dummy variable. This variable is either zero when the price of the crude oil is equal or below 80$1 in that month or one if the price is above the 80$. When the coefficient 𝜔₃ is significant, it can be concluded that the relationship is non-linear.

For regression (1) to (4), there will be a separate regression that only includes the data after the financial crisis in 2008. The motivation for this split is to test whether the financial crisis caused any differences in the way oil returns predict the stock market returns.

The four regression will all be tested the same way. The basic hypothesis will be:

H0: The coefficient 𝑅_𝑡𝑂𝑖𝑙 = 0. This means that the change of the oil price of the previous month has no significant influence on the market return on time t.

H1: The coefficient 𝑅_𝑡𝑂𝑖𝑙 ≠ 0. The change in the oil price does have a significant effect on the market return. The coefficient of 𝑅_𝑡𝑂𝑖𝑙 is able to predict the market return on time t.

The second part of this research I need a new model to explain the effect of return of oil on the return on industry specific market. The reason for the industry specific test is to see whether oil has a different effect on the oil & gas industry than it has on other industries. The model that will be used contains the market return coefficient and the coefficient for the return on oil.

(10)

10 (5) (𝑅_𝑡Industrials_{− 𝑅}

𝑓) = 𝛿0+ 𝛿1∗ 𝑅𝑡𝑂𝑖𝑙+ 𝛿2∗ (𝑅𝑡𝑀𝑟𝑘𝑡− 𝑅𝑓) + 𝜀𝑖

Where (𝑅𝑡Industrials− 𝑅𝑓) is the industries monthly excess return, 𝑅𝑡𝑂𝑖𝑙 is the monthly return

of oil, and (𝑅_𝑡𝑀𝑟𝑘𝑡 _{− 𝑅}

𝑓) is the market monthly excess return. The constant 𝛿0 is important for

this regression, as it shows the returns uncorrelated to the risk. The 𝑅𝑓 is the monthly risk free

U.S. Treasury Bond return.

H0: The coefficient of 𝑅_𝑡𝑂𝑖𝑙 for a specific industry will be equal to zero.

H1: The coefficient of 𝑅_𝑡𝑂𝑖𝑙 for a specific industry will be unequal to zero. The specific industry can be predicted by the returns in crude oil.

(11)

11

4. Results and analyses

As previous discussed, the first results are based on an OLS regression on the Stoxx 600 index returns and crude oil returns of Brent or WTI as the explanatory variable. The second results are based on an OLS regression on the crude oil returns and a specific industry.

4.1 Relation between oil returns and stock market returns

The first regression is the relation between the Brent crude oil returns and the Stoxx 600 index in dollars. The table provides information about the coefficients found, the standard errors, and the p-value of the test. The p-values are in the parenthesis and are bold when the test gives a significant outcome.

The first number is the coefficient from the regression and the number below it provides the standard error. When the standard error is bold, the coefficient is significantly different from zero. The 𝑅2_{value gives information about how much of the data can be}

explained by the regression model. The higher the 𝑅2_{value, the more the model can explain}

the outcome.

The first 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 tells us if all the coefficients are zero. If this p-value below it is bold, it means that at least one of the coefficients in the model is different from zero. The second 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 tells us if the new regression is a significant improvement from the basic regression model (1). When the p-value is bold, it means that the new regression is better than regression (1) and that we can assume that model is stronger than model (1).

The last row contains the Breusch–Pagan test (Breusch & Pagan, 1979). The Breusch–Pagan test uses the original regression model, for example: 𝑅_𝑡𝑀𝑟𝑘𝑡 _{= 𝛾}

0+ 𝛾1∗

𝑅_𝑡𝑂𝑖𝑙_{+ 𝜀}𝑖_{. After the regression, it creates a new equation for the error term 𝑉𝑎𝑟(𝜀}𝑖_{) = 𝜎}2_∗

( 𝜇₀+ 𝜇₁∗ 𝑅_𝑡𝑂𝑖𝑙_{). Stata tests the new equation if all the coefficients are zero (𝜇}

0 = 𝜇1 = 0)

and calculates the 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒. When this test give a significant outcome, it means that the model is heteroskedastic and that the robust standard errors should be used.

(12)

12 Table 1. OLS Results Brent Dollar

(1) (2) (3) (4) (1) (2) (3) (4) Coefficient 01/2000-12/2014 (180) (65) 07/2008-12/2014 (78) (44*) 𝑅𝑡𝑂𝑖𝑙 .176028 .040144 .179878 .053853 .176018 .054426 .135091 .044250 .39828 .074228 .381209 .078544 .401202 .075085 .434378 .086015 𝑅𝑡𝑂𝑖𝑙2 -.00349 .003224 -.002908 .004243 𝑃𝑟𝑖𝑐𝑒 𝐷𝑢𝑚𝑚𝑦 -.193452 .860408 -.441055 .884158 -.532474 1.46247 -.51034 1.46630 𝐷𝑢𝑚𝑚𝑦 𝑥 𝑅𝑒𝑡𝑢𝑟𝑛 .216871 .101861 -.14162 .177717 𝑉𝑎𝑙𝑢𝑒 𝑅2 _0.0975 _0.1096 _0.0977 _0.1204 _0.2747 _0.2792 _0.2760 _0.2822 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 All Coefficients = 0 19.23 (0.0000) 5.91 (0.0033) 5.43 (0.0051) 8.03 (0.0000) 28.79 (0.0000) 14.53 (0.0000) 14.30 (0.0000) 9.70 (0.0000) 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 Regression (X) = Regression (1) 1.17 (0.2799) 0.05 (0.8224) 2.29 (0.1042) 0.47 (0.4953) 0.13 (0.7168) 0.38 (0.6828) 𝐻𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦 𝐶ℎ𝑖2 7.13 (0.0076) 12.47 (0.0004) 6.67 (0.0098) 1.92 (0.1660) 0.32 (0.5714) 0.31 (0.5758) 0.13 (0.7223) 0.24 (0.6254)

In the period 07/2008-12/2014 the dummy variable is based on a dollar price of 100$. When I use the 100$, the dummy includes almost the entire data set of that period, making it insignificant to test.

In all the regressions models, I find that the coefficient for 𝑅_𝑡𝑂𝑖𝑙_{is always significant different}

from zero. For example, a coefficient of 0.176028 in regression (1) should indicate that the Stoxx 600 index increases with 0.176028% when the returns on Brent crude oil increased with 1%. For the Brent crude oil there is a stronger relation in the period of 07/2008-12/2014, because of the much higher 𝑅2_{value. None of the regressions is significantly better than the}

basic regression (1), meaning that I can’t say for certain that the model is non-linear.

Compared to previous literature, I find it interesting that the coefficient 𝑅_𝑡𝑂𝑖𝑙_{is always}

positive. Until now, research has shown that a decrease in the oil returns lead to an increase in the stock returns.

(13)

13 Table 2. OLS Results WTI Dollar

(1) (2) (3) (4) (1) (2) (3) (4) Coefficient 01/2000-12/2014 (180) (66) 07/2008-12/2014 (78) (16*) 𝑅𝑡𝑂𝑖𝑙 .272319 .054064 .268731 .050702 .272151 .054007 .276268 .052232 .505299 .063506 .492305 .065340 .510864 .064112 .556693 .067288 𝑅𝑡𝑂𝑖𝑙2 -.004732 .003617 -.002981 .0034289 𝑃𝑟𝑖𝑐𝑒 𝐷𝑢𝑚𝑚𝑦 .070692 .802394 1.00672 .856779 -1.17292 1.55456 -.58649 1.55679 𝐷𝑢𝑚𝑚𝑦 𝑥 𝑅𝑒𝑡𝑢𝑟𝑛 -.014531 .006913 -.36918 .190981 𝑉𝑎𝑙𝑢𝑒 𝑅2 _0.1853 _0.1977 _0.1853 _0.2014 _0.4545 _0.4599 _0.4586 _0.4846 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 All Coefficients = 0 25.37 (0.0000) 21.81 (0.0000) 12.79 (0.0000) 11.04 (0.0000) 63.31 (0.0000) 31.93 (0.0000) 31.76 (0.0000) 23.19 (0.0000) 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 Regression (X) = Regression (1) 2.74 (0.0998) 0.01 (0.9299) 2.26 (0.1069) 0.76 (0.3873) 0.57 (0.4529) 2.16 (0.1221) 𝐻𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦 𝐶ℎ𝑖2 7.61 (0.0058) 4.11 (0.427) 7.76 (0.0054) 6.02 (0.0142) 0.18 (0.6674) 0.00 (0.9794) 0.33 (0.5686) 0.43 (0.5126)

In the period 07/2008-12/2014 the dummy variable is based on a dollar price of 100$. When I use the 100$, the dummy includes almost the entire data set of that period, making it insignificant to test.

When I use the WTI crude oil returns, I find similar results. The coefficient 𝑅_𝑡𝑂𝑖𝑙_is

significantly different from zero for every model. The 𝑅2_{value is higher in the period}

07/2008-12/2014. Overall, the WTI gives a better prediction than the Brent crude oil. The 𝑅2_{value is almost doubled for both time periods. The model has more explanatory power}

when I use the WTI crude oil as explanatory variable.

The second regressions are based on the Brent and WTI oil prices in euros compared to the Stoxx 600 index in euros. The tables are the same as used above. Bold numbers still mean it is significant and the p-values are in the parenthesis. The tables for the regression in euros can be found in the appendix I under Table 4.

(14)

14

4.2 Relation between oil returns and specific industry returns

In this section I will try to predict the effect of the returns of oil on a specific industries within the Stoxx 600 index. The results will show whether there is a significant correlation between the return of oil and industries like the oil & gas industry. Driesprong, Jacobsen, Maat (2008) discovered that the return on crude oil can sometimes predict the return of non-oil related industries. Because there might be a relation between crude oil returns and non-oil related industries, I will be testing a variety of industries within the Stoxx 600 market. It consists of 17 different industries including the oil & gas industry. This model will only be tested in dollars. I find a stronger relationship between the crude oil returns in dollars and the index returns in dollars than for the same regressions in euros.

The first column gives the information about which industry the test was on. The constant 𝛼 gives us information about the risk-neutral returns within that industry. The 𝛽 coefficient is for testing the industry on the market returns. This variable is included to make sure that there is no omitted variable bias. The 𝑅𝑂𝑖𝑙_{is the coefficient that shows us how much}

effect the return on the crude oil has on the specific industry. The last column gives the 𝑅2_{value, that gives us information about how much of the data outcome can be explained by}

the model.

Table 3. OLS Results Dollar Industry Specific

Industry 𝛼_{𝐵𝑟𝑒𝑛𝑡} 𝛽_{𝐵𝑟𝑒𝑛𝑡} 𝑅_{𝐵𝑟𝑒𝑛𝑡}𝑂𝑖𝑙 𝑅2 value 𝛼_𝑊𝑇𝐼 𝛽_𝑊𝑇𝐼 𝑅_𝑊𝑇𝐼𝑂𝑖𝑙 𝑅2 value

Oil & Gas (33) -.2367618 .2983082 .7802909 .053327 .1545293 .0339234 0.6791 -.2770269 .2903467 .7308448 .0554068 .1992529 .0410176 0.6901 Financials (143) .0975957 .2274715 1.311711 .0549465 -.0618589 .0259843 0.8814 .1099231 .2251255 1.329557 .0547495 -.0768627 .0285581 0.8820 Healthcare (35) .1284461 .2461411 .5435284 .0512479 -.0607905 .0260914 0.4916 .106776 .2504397 .5437077 .0562198 -.0496956 .030741 0.4820 Insurances (38) .3131707 .2914946 1.39312 .057617 -.1552775 .0355179 0.8342 .3118552 .2984849 1.421342 .0617284 -.1682666 .0437204 0.8294

This table only give the results of the industries that had a significant coefficient for 𝑅𝐵𝑟𝑒𝑛𝑡𝑂𝑖𝑙 and 𝑅𝑊𝑇𝐼𝑂𝑖𝑙 . The remaining results

(15)

15 The results are that the coefficient 𝑅𝑂𝑖𝑙_{in the oil & gas industry is significant for Brent}

and for WTI. This result is not in line with previous literature, where Driesprong, Jacobsen, Maat (2008) found a weaker effect in the sectors where an oil price change has more direct effect. They find that for non-oil related sectors, such as Consumer Goods and Technology, a stronger relation. I did find a significant prediction in three other sectors. The Financials, Healthcare, and Insurances all show that the crude oil returns can predict the returns from that industry. The complete table can be found in the appendix II under Table 5.

4.3 Further research

Given these results, I think there should be further research in the data and model. We should investigate on a year-by-year basis, if there is significant non-stationarity in the industry and oil returns. Time series are nonstationary when the joint distribution of the variable and its lags change over time. The problem in these time series may be a spurious regression and will give a misleading interpretation of the OLS estimations. This problem is caused by stochastic trends, where two independent series will appear to be related if they both have stochastic trends (Stock & Watson, 2012, pp. 591-592).

An alternative way to research this model is by using the Granger causality test. This can test whether current and lagged values of the crude oil returns help predict the next month values of the European stock returns. This method is used in previous research by Jiménez-Rodríguez and Sánchez (2005) and Hamilton (1983).

Omitted variable bias could be a problem in the model. Research from Hamilton (1983), Jiménez-Rodríguez and Sánchez (2005), and more macro-economic

research use additional variables such as the inflation, short, and long-term interest rates. The OLS estimations can give an over estimation or under estimation when leaving out variables that have an important effect on the model. The missing variables may cause the unpredicted results from the regression.

(16)

16

5. Conclusion and discussion

In this thesis, I focused on the relation between the returns on crude oil and the returns on the European stock market. The data consists of 179 observations between the time period of January 2000 and December 2014 for both crude oil returns and Stoxx 600 index returns. The regression are done in dollars and euros that are corrected for the exchange rate at time t.

The results from the basic model indicate that the returns on Brent and WTI both predict positive returns on the Stoxx 600 index. This result is opposite from previous literature, where the coefficient of crude oil returns was always negative. The different

coefficient might come from the data set. The data used in this research only uses the latest 15 years. It is possible that there has been a shift in the effect of crude oil returns on the stock returns over de last 15 years. Another explanation can that I do not take the under reaction into account. In previous literature it is shown that this under reaction effect is significant. When I leave this effect out, it can cause omitted variable bias.

Adding variable to test the model for non-linearity did not give significant evidence. In all three additional regression, the coefficient of 𝑅𝑡𝑂𝑖𝑙 is always significant, but the new model

is not significantly better than the basic regression model. We find stronger evidence in the period after the financial crisis of 2008 and better estimations when using the crude oil returns of the Western Texas Intermediate.

The results from the industry specific regressions show only four of the seventeen industries have a significant coefficient on the crude oil returns. I find that there is a positive correlation between the crude oil returns and the oil & gas industry. Previous empirical research shows that there is not direct link between the crude oil returns and oil-related stock returns.

(17)

17

References

Austin S., and Therramus T., (2015). Will Collapse in Oil Price Cause a Stock Market

Crash? Retrieved January 14, 2015, from

http://oil-price.net/en/articles/will-collapse-in-oil-price-cause-stock-market-crash.php

Breusch T., and Pagan R., (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrica, Vol. 47, No. 5, pp. 1287-1294

Driesprong G., Jacobsen B., and Maat B., (2008). Striking Oil: Another puzzle? Journal of

Financial Economics, Vol. 89, pp. 307–327

Jiménez-Rodríguez R., and Sánchez M., (2005). Oil price shocks and real GDP growth: empirical evidence for some OECD countries. Applied Economics, Vol. 37, pp. 201-228

Hamilton J., (1983). Oil and the Macroeconomy since World War II. Journal of

Political Economy, Vol. 91, pp. 228-48

Hong H., and Stein J., (1999). A Unified Theory of Underreaction, Momentum Trading and Overreaction in Asset Markets. The Journal of Finance, Vol. 54, pp. 2143–2148 Hooker M., (1996). What happened to the oil price-macroeconomy relationship? Journal of

Monetary Economics, Vol. 38, pp. 195-213

Lewitinn L., (2014). The surprising relationship between oil and stocks. Retrieved January 27, 2015, from http://finance.yahoo.com/blogs/talking-numbers/the-surprising-relationship-between-oil-and-stocks-220546670.html

Merton R., (1973). An Intertemporal Capital Asset Pricing Model. Econometrica, Vol. 41, No. 5, pp. 867-887

Miller J., Ratti R., (2009). Crude oil and stock markets: Stability, instability, and bubbles.

Energy Economics, Vol. 31, pp. 559-568

Mork K., (1989). Oil and the Macroeconomy When Prices Go Up and Down: An Extension of Hamilton's Results. Journal of Political Economy, Vol. 97, pp. 740-744

Mory J., (1993). Oil Prices and Economic Activity: Is the Relationship Symmetric?

(18)

18 Park J., and Ratti R., (2008). Oil price shocks and stock markets in the U.S. and 13 European

countries. Energy Economics, Vol. 30, pp. 2587–2608

Pescatori A., and Mowry B., (2008). Do Oil Prices Directly Affect the Stock Market? Economic Trends.

Pollet J., (2003). Predicting Asset Returns With Expected Oil Price Changes.

Sadorsky P., (1999). Oil price shocks and stock market activity. Energy Economics, Vol. 21, pp. 449-469

Stock J., and Watson M., (2012). Introduction to Econometrics (Third Edition), pp. 591-598.

Harlow: Pearson.

U.S. Energy Information Administration, (2015). International Energy Statistics. Retrieved January 27, 2015, from

http://www.eia.gov/cfapps/ipdbproject/IEDIndex3.cfm?tid=5&pid=5&aid=2 U.S. Energy Information Administration, (2014). Today in Energy. Retrieved January 15,

(19)

19

Appendix I

Table 4. OLS Results Brent & WTI Euro

(1) (2) (3) (4) (1) (2) (3) (4) Coefficient 01/2000-12/2014 (180) (65) 07/2008-12/2014 (78) (44*) 𝑅𝑡𝑂𝑖𝑙 .098431 .033587 .101035 .038195 .098437 .033682 .092281 .036680 .194364 .061574 .161028 .064004 .195409 .062299 .237352 .069647 𝑅𝑡𝑂𝑖𝑙2 -.002237 .001950 -.006246 .003719 𝑃𝑟𝑖𝑐𝑒 𝐷𝑢𝑚𝑚𝑦 -.020884 .7201818 -.063632 .728684 -.181089 1.10444 -.112063 1.10023 𝐷𝑢𝑚𝑚𝑦 𝑥 𝑅𝑒𝑡𝑢𝑟𝑛 .040271 .093817 -.201886 .152802 𝑉𝑎𝑙𝑢𝑒 𝑅2 _0.0460 _0.0544 _0.0460 _0.0470 _0.1159 _0.1480 _0.1162 _0.1366 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 All Coefficients = 0 8.59 (0.0038) 3.51 (0.0321) 4.27 (0.0154) 2.90 (0.0367) 9.96 (0.0023) 6.51 (0.0025) 4.93 (0.0097) 3.90 (0.0121) 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 Regression (X) = Regression (1) 1.32 (0.2530) 0.00 (0.9769) 0.09 (0.9116) 2.82 (0.0972) 0.03 (0.8702) 0.89 (0.4165) 𝐻𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦 𝐶ℎ𝑖2 3.54 (0.0597) 4.52 (0.0335) 3.48 (0.0620) 3.04 (0.0810) 0.00 (0.9831) 2.91 (0.0878) 0.02 (0.9021) 0.01 (0.9098) 𝑅𝑡𝑂𝑖𝑙 .158818 .043779 .154481 .040741 .15860 .043819 .150970 .051668 .308330 .059280 .283715 .063249 .312922 .059935 .357545 .063117 𝑅𝑡𝑂𝑖𝑙2 -.003430 .003402 -.00466 .004219 𝑃𝑟𝑖𝑐𝑒 𝐷𝑢𝑚𝑚𝑦 .089409 .662426 .049083 .687515 -.797453 1.23785 -.290296 1.24265 𝐷𝑢𝑚𝑚𝑦 𝑥 𝑅𝑒𝑡𝑢𝑟𝑛 .034188 .096419 -.340799 .174428 𝑉𝑎𝑙𝑢𝑒 𝑅2 _0.0880 _0.0945 _0.0880 _0.0887 _0.2625 _0.2743 _0.2666 _0.3026 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 All Coefficients = 0 13.16 (0.0004) 7.19 (0.0010) 6.55 (0.0018) 4.57 (0.0041) 27.05 (0.0000) 14.18 (0.0000) 13.63 (0.0000) 10.70 (0.0000) 𝐹 − 𝑇𝑒𝑠𝑡 𝑆𝑐𝑜𝑟𝑒 Regression (X) = Regression (1) 1.02 (0.3147) 0.02 (0.8928) 0.08 (0.9236) 1.22 (0.2729) 0.42 (0.5214) 2.12 (0.1268) 𝐻𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦 𝐶ℎ𝑖2 5.13 (0.0236) 4.04 (0.0445) 5.38 (0.0203) 4.81 (0.0283) 0.22 (0.6401) 0.00 (0.9771) 0.37 (0.5452) 0.93 (0.3351)

In the period 07/2008-12/2014 the dummy variable is based on a dollar price of 100$. The 80$ dummy includes almost the entire data set of that period, making it insignificant to test.

(20)

20

Appendix II

Table 5. OLS Results Dollar Industry Specific

Industry 𝛼_{𝐵𝑟𝑒𝑛𝑡} 𝛽_{𝐵𝑟𝑒𝑛𝑡} 𝑅_{𝐵𝑟𝑒𝑛𝑡}𝑂𝑖𝑙 𝑅2 value 𝛼_𝑊𝑇𝐼 𝛽_𝑊𝑇𝐼 𝑅_𝑊𝑇𝐼𝑂𝑖𝑙 𝑅2 value

Auto & Parts (14) .6706803 .3945 1.184679 .0833995 -.0175888 .0428599 0.6601 .6547989 .3993644 1.179793 .0827939 -.0070275 .0519992 0.6597 Banks (49) .02033 .2881852 1.325961 .0749479 -.0332093 .0314761 0.8265 .037174 .2833729 1.340796 .0746875 -.0490853 .0343527 0.8274 Chemicals (25) .6791591 .2453378 1.076502 .0518728 -.0345619 .0283541 0.8057 .6486218 .2453532 1.067244 .056916 -.0143213 .0338382 0.8036 Construction & Materials (20) .442911 .234836 1.140107 .0508026 -.0231433 .0278832 0.8347 .4467238 .2385946 1.146373 .0543612 -.0281453 .0326752 0.8348 Consumer Goods (70) .4074661 .1624004 .7715178 .0314334 -.0121549 .0164942 0.8168 .3923847 .1623568 .7660311 .0342872 -.0017157 .0200828 0.8162 Financials (143) .0975957 .2274715 1.311711 .0549465 -.0618589 .0259843 0.8814 .1099231 .2251255 1.329557 .0547495 -.0768627 .0285581 0.8820 Food & Beverages (27) .3991963 .2260804 .5945929 .045754 -.0243579 .0275341 0.5661 .3675283 .2294995 .5828552 .0495691 -.0023327 .0297073 0.5634 Healthcare (35) .1284461 .2461411 .5435284 .0512479 -.0607905 .0260914 0.4916 .106776 .2504397 .5437077 .0562198 -.0496956 .030741 0.4820 Industrials (128) .2318109 .1562426 1.118647 .0345659 -.0010551 .0181222 0.9188 .2376913 .1576473 1.121865 .0368705 -.0056477 .0212972 0.9188 Insurances (38) .3131707 .2914946 1.39312 .057617 -.1552775 .0355179 0.8342 .3118552 .2984849 1.421342 .0617284 -.1682666 .0437204 0.8294 Media (28) -.1041086 .3034263 1.053219 .0645661 -.00543 .0366009 0.7069 -.0838826 .2960535 1.064621 .0681129 -.0213888 .0451127 0.7074

Oil & Gas (33) -.2367618 .2983082 .7802909 .053327 .1545293 .0339234 0.6791 -.2770269 .2903467 .7308448 .0554068 .1992529 .0410176 0.6901 Retail (27) -.1727418 .2011137 .8374835 .0431066 -.0146258 .0211097 0.7678 -.1438739 .2016409 .8550375 .0454677 -.0380229 .0250825 0.7704 Technology (21) .0289172 .4760037 1.363279 .0987625 -.0723591 .0555222 0.6450 .0956341 .4789286 1.411024 .1088193 -.1299078 .0666876 0.6520 Telecom (23) -.3525087 .3423553 .8940926 .0729796 -.0338393 .0362281 0.5827 -.330271 .3410603 .9118157 .0780982 -.0538973 .0392803 0.5847 Travel & Leisure (22) .1852628 .2329248 .974045 .0437952 -.0419005 .0231427 0.7811 .2050005 .2300613 .9919838 .0453694 -.0607729 .0263634 0.7833 Utilities (26) .0640943 .240152 .86127 .0487421 -.0161181 .0247437 0.7325 .0323746 .2410449 .8479723 .051483 .0066893 .0281072 0.7319

Crude oil returns and stock market returns