The effect of oil price on stock returns
Abstract
This study uses both ordinary least squares and a random effects panel data methodology to empirically determine the relationship between oil prices and the Dutch stock market. It covers the AEX index and 13 firms listed on the AEX index over the period from 1997 to 2017. Many studies report a negative significant effect between oil prices and stock returns. The results from this study reveal a positive significant relationship between oil prices and stock market returns. The relationship is less significant when controlling for firm level differences on the effect of oil prices.
31th of January 2017
Robin Zijsling 10778799
University of Amsterdam
Thesis supervisors: Mr. S. Changoer MSc and dr. P.J.P.M. Versijp Organization and Finance
Statement of Originality
This document is written by Robin Zijsling 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
1. Introduction
Before 1973 oil prices were not volatile at all. A few large oil companies known as the Seven Sisters set production controls stabilizing the oil price. OPEC took over a large part of the market after 1973 and prices became more volatile and started fluctuating like other commodities (Driesprong, Jacobsen, Maat, 2008).
According to Mussa (2000), oil prices play an important role in financial markets and the economy. Supporting this idea, Driesprong et al. (2008) find that when oil prices increase, economic activity tends to decrease and when oil prices decrease economic activity tends to increase. Driesprong et al. (2008) provide several explanations for these findings. The first is that changes in oil price have an impact on real output of firms as it affects a firm’s production costs and expected earnings. The second is that investors have difficulty in evaluating the impact of changes in oil prices.
If these explanations are valid, then oil prices should be inversely related to stock returns. Several studies examine whether this is the case. However, most of those studies focus on countries with large economies and ignore smaller countries such as the Netherlands.
This study is related to the work of Driespong et al (2008), Sadorsky (1999) and Hamilton (1983) and is a follow up to the existing literature but focusses on a rather small economy, the Dutch stock market. This study strives to answer the following research question: “What is the effect of oil prices on AEX index returns?” To answer this question this study starts by discussing the following sub-research questions. How do stock markets react to changes in oil prices? Is information on oil prices efficiently reflected in oil prices? Is the stock market efficient? Which factors affect the stock market?
My study is interesting for several reasons. First, because the effect of oil prices on AEX-index returns has not been researched yet. Second, because most research on the effects of oil prices on stock markets is prior to 2003. Researching the effect on a smaller economy contributes to the literature. Investigating the period after 2003 could bring possible new evidence on changing oil price dynamics and how this affects the stock market.
For my analysis, I use monthly data from Datastream and CBS to estimate two models. First, I regress AEX index monthly returns on Brent oil monthly returns together with other control variables. Then, I estimate a random effects model using panel data and run a regression on 13 individual AEX index stocks. Both regressions are performed on the total timeframe between 1997 and 2017.
The remainder of this study is organized as follows. Section 2 discusses the relevant background information and presents a review of literature with respect to the research questions. Section 3 outlines the methodology and data to be used. In section 4, I present the empirical results. In section 5, I discuss the findings. In section 6, I discuss the conclusion and limitations of this study.
2. Literature review
2.1 Background information The oil market:
Oil is the worlds most traded commodity and is an important resource in our world economy (ECB, 2016). Three main types of oil can be distinguished and serve as a benchmark for other types of oils: Brent oil, Dubai crude, West Texas intermediate (WTI). The three types of oil differ in price, which is caused by a difference in density and Sulphur content (Driesprong et al., 2008). When the content of Sulphur in oil is relatively low the oil is called ‘sweet‘, a low density is referred to as light (Driesprong et al. , 2008). Brent oil can be classified as a sweeter and lighter type of oil, making it more expensive to process compared to WTI.
Brent oil accounts for 60 % of the total daily consumption, making it the worlds most consumed type of oil. WTI and Dubai account for 22 % and 18 % respectively (Driesprong et al, 2008). This study focuses on Brent oil as it represents the largest consumed type of oil.
The price of Brent oil was relatively stable before the establishment of OPEC in 1973. Since then, oil prices began rising significantly, reaching a record high at $143 a barrel in the early part of 2008 (see figure 1). In the later part of 2008 oil prices plunged back to $43 a barrel as a result of the sub-prime mortgage crisis also known as the financial crisis of 2008. In the following years, prices acted volatile and ranged between $34 and $120 a barrel (ECB, 2016). New lows were hit in 2016 at $28 a barrel (ECB, 2016). Nevertheless, prices recovered rapidly and stabilized around $50 a barrel (OPEC, 2016).
Fig. 1. Brent oil price development over the period January 1987-December 2016.
2.2 The effect of oil price on stock returns
Driesprong et al. (2008) find that changes in oil prices are negatively related to stock market returns. Consistent with this view, Hamilton (1983) and Sadorsky (1999) conclude that changes in oil prices significantly help in predicting stock market returns.
Driesprong et al. provide several explanations for these findings. The first is that an increase in oil prices has a negative impact on economic performance and real output of firms. Nandha and Faff (2008) support this idea. The reason is that firms that use relatively more oil in their production process are negatively affected by oil price shocks as it increases their production costs and lowers their expected earnings. The second is that investors have difficulty in evaluating the impact of changes in oil prices.
Miller and Ratti (2009) argue that an oil price shock is positively related to inflation and higher inflation figures are generally followed with higher interest rates. This negatively affects expected discounted cash flows from firms and lowers expected stock returns. If these explanations are valid, then I expect a negative relation between oil price and stock market returns.
Sadorsky (1999) shows that the effect of oil price shocks on stock markets is asymmetric. In his research he shows that a positive oil price shock explains more of the forecast error variance than negative shocks (Sadorsky, p.465). In the total sample period of his study 51% of the shocks were negative and 49% were positive. These statistics show that the occurrence of negative shocks is larger than positive shocks. Yet, positive oil price shocks seem to have a greater impact on economic performance. He concludes that price shocks appear to be asymmetric in explaining economic performance.
Contrarily, another prediction assumes that investors react rationally to information on oil prices. Several studies have examined whether this is the case. For instance, Jones&Kaul (1996) examine whether stock markets react rationally or overreact on oil price shocks. Their results suggest that the US and Canada have rational markets. On the other hand, they did not find proof for rationality in the United Kingdom and Japanese markets.
From these results it is fair to assume that not all markets react the same on information on oil prices. This result makes it interesting to examine how different countries react to information on oil prices. Because other studies ignored smaller markets like the Dutch stock market, this study strives to examine the effect of oil prices on the Dutch Stock market.
My hypothesis is that the oil price has a positive significant effect on AEX index returns. This hypothesis is in contrast with the results from previous studies (Driesprong et al., 2008; Hamilton, 1983; Sadorsky, 1999). This is because the composition of the AEX index consists of a relatively large share of oil & gas industry companies see figure 2, (appendix A). Royal Dutch Shell accounts for approximately 15% of the composition of the index. I think I might find positive correlations between changes in oil prices and future stock returns for the AEX index since the weight of Shell is 15%. I also think that the effect of oil prices on stock returns might have changed in the last 20
years. This is caused by improved technology in the financial sector. This has enabled information to be reflected in securities pricing more efficiently (Ito, Noda and Wada, 2014).
2.3 Is the stock market efficient?
According to Fama (1970) a market is called efficient when prices fully and immediately reflect all available information. Some studies such as DeBondt and Thaler (1985) show that stock markets are inefficient. Other studies, such as Ito, Noda and Wada (2014) conclude that stock market efficiency is time varying.
DeBondt and Thaler (1985) demonstrate the inefficiency of the stock market by showing how abnormal returns can be preserved from an investors overreaction to information. This means that an investor can over perform that market at the same risk level. Conrad and Kaul (1998) support this idea and conclude that securities prices underreact to certain events and information.
Ito, Noda and Wada (2014) conclude that stock market efficiency is time varying. They suggest that market efficiency has increased over time, because of improved technology in the financial sector. This has enabled information to be reflected in securities pricing more rapidly, they find no unambiguous results.
2.4 Which factors affect stock markets?
I include three sets of control variables in regression 1. First, I include the Dutch GDP growth rate, the Euro/Dollar exchange rate and the German yield on 10-year government bonds to control for macro-economic factors such as economic growth, monetary policy and exchange rate fluctuations.
I include these variables, because studies such as those of Levine and Zervos (1996) find a significant relationship between stock markets and macro-economic variables such as GDP,
monetary policy and exchange rates. This is also confirmed by Flannery and Protopapadakis (2002), who find that interest rates negatively affect stock market returns. This means that a rise in interest rates has a negative effect on stock market returns. Moreover, a rise in interest rates causes bond prices to fall and lower bond prices make investing in bonds more attractive relative to stocks (Volper, 2013). Comparable research in this field suggests that the same negative relationship coexists between inflation rates and stock market returns (Pearce and Koley 1983; Basher and Sadorsky;2006; Volpert, 2013).
Filis (2010) also addresses the relationship between GDP growth and stock-market movements. Earlier research in this field claims that GDP growth is a lagging indicator of stock market growth. Moreover, GDP growth is followed by delayed stock market growth (Glen, 2002). The effect is reflected in subsequent periods. Another factor affecting stock market returns is researched by Bahmani and Sohrabian (1992). In their empirical study they found a bidirectional relationship in the short run between the American S&P500 index and the exchange rate of the dollar denominated with respect to multiple other currencies. In the long run however no significant effects were found. The rationale behind exchange rates affecting stock market returns is directly linked to the balance of trade. The devaluation of a currency implies consequences on import and export prices. Both international competitiveness and the balance of trade are affected by it.
3. Methodology and data
3.1 Model Specification
To test 𝐻𝐴(1), I estimate the following model, which is based on those of Driesprong et al (2008) and Sadorsky (1999): “Striking oil: Another puzzle?” and “Oil price shocks and stock market activity”, respectively (see appendix B). Regression 1 will test whether the coefficient β1 is significantly different from zero.
𝑅𝑡= β0 + β1𝐵𝑅𝐸𝑁𝑇𝑡 + β2 𝐸𝑋𝐶𝐻𝑡 + β3 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡+ β4 𝐺𝐷𝑃𝑡+ 𝜀𝑡, (1),
where:
𝑅𝑡 = Reinvestment return AEX-index in month t (at the end of the month); 𝐵𝑅𝐸𝑁𝑇𝑡= Brent oil return in month t (at the end of the month);
𝐸𝑋𝐶𝐻𝑡 = Euro/Dollar exchange rate return in month t (at the end of the month);
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 = 10-year German government bond yield in month t(at the end of the month); 𝐺𝐷𝑃𝑡 = GDP growth rate Netherlands at quarter t (at the end of the quarter);
𝜀𝑡 = Disturbance term.
To test 𝐻𝐴 (2), I estimate the following model, equation 2 presents a random effects model that allows to control for differences across entities.
𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡= α0 + α1 𝐵𝑅𝐸𝑁𝑇𝑡 + α2 𝐸𝑋𝐶𝐻𝑡 + α3 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡+ α4 𝐺𝐷𝑃𝑡 + α5 𝑂𝐼𝐿𝑖 (2),
+ α6 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 + 𝑢𝑖𝑡 + 𝜀𝑖𝑡,
𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡= Stock return individual stock i at month t, at the end of the month;
𝐵𝑅𝐸𝑁𝑇𝑡= Brent oil return coefficient at month t, at the end of the month;
𝐸𝑋𝐶𝐻𝑡 = Euro/Dollar exchange rate return coefficient at month t, at the end of the month; 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡= 10-year German government bond yield at month t, at the end of the month; 𝐺𝐷𝑃𝑡 = GDP growth rate Netherlands at quarter t, at the end of the quarter;
𝑂𝐼𝐿𝑖 = Dummy variable oil effect for stock i (1 = Oil price effect for stock i); 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 = Interaction variable per stock i at month t
𝑢𝑖𝑡 = Between-stocks errors stock i at month t 𝜀𝑖𝑡 = Within-stocks errors stock i at month t
3.2 Interpretation
𝑅𝑡 is the return of the AEX reinvestment index in month t. The AEX index is a representative
of the 25 largest Dutch companies listed on Euronext Amsterdam. I use the AEX reinvestment returns as the dependent variable in my model because it acts as a good representative for the Dutch market. By using the reinvestment index a dividend control variable can be ruled out.
𝐵𝑅𝐸𝑁𝑇𝑡is the return of Brent oil in month t. I focus on the return of Brent oil because it is the most consumed type of oil and has the largest production share. This makes Brent oil is a good representative for the oil market and a useful predicting variable in my model. I include multiple control variables to the model, because Studies such as those of Levine and Zervos (1996) find a significant relationship between stock markets and macro-economic variables such as GDP, monetary policy and exchange rates.
𝐸𝑋𝐶𝐻𝑡 is the return on the Euro/Dollar in month t. As discussed in section 2, exchange rates and stock markets are negatively correlated. If this assumption is true, the variable 𝐸𝑋𝐶𝐻𝑡
negatively correlates with 𝑅𝑡. This study focusses on the Dutch market which is situated in the
Euro zone. Because of this the Euro/Dollar exchange rate is a good representative for the control
variable 𝐸𝑋𝐶𝐻𝑡 in the model.
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 is the 10-year German government bond yield in month t. I am adding this variable
to the model because interest rates tend to correlate negatively with stock returns. An increase of interest rates is followed by lower stock market returns. When interest rates increase investors tend to flee to saver investments like government bonds. The German 10-years bund acts as a good bench mark for the Dutch market since Germany is the largest economy of Europe. 𝐺𝐷𝑃𝑡 is the quarterly growth rate of the Netherlands in month t. The GDP growth rate of a country is a good indication of the business cycle at a certain point in time. When the economy is in a recession GDP growth rate will likely be lower and when the economy is in a higher phase of the business cycle the GDP growth rate will likely be higher. As discussed in section 2 the current phase of the economy affects stock market returns.
Additional information equation 2:
Equation 2 implies a random effects panel data model to empirically determine the
relationship between oil prices and the Dutch stock market. 𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡 represents the stock return for the 13 stocks presented in table 1 (see appendix C). The variables 𝐵𝑅𝐸𝑁𝑇𝑡, 𝐸𝑋𝐶𝐻𝑡 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 and 𝐺𝐷𝑃𝑡 are the same as in equation, as they do not vary across stocks but only vary over time. 𝑂𝐼𝐿𝑖 is a dummy variable to control for an oil price effect per stock i, (1= significant oil price effectfor stock i). Table 1 presents the results from a set of regressions conducted on equation 3 (see appendix C) when considering the predictability of the main
explanatory variable and control variables on individual stock returns. The variable 𝐵𝑅𝐸𝑁𝑇𝑡 is
significant at the 5% level for the stocks Akzo Nobel (0.048<α= 0.05), DSM (0.000<α= 0.05), ING (0.032<α= 0.05) and Shell (0.000<α= 0.05). 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 represents an interaction
variable for the firms for which 𝐵𝑅𝐸𝑁𝑇𝑡 is significant. This is variable helps to control for firm
level differences on the effect of oil prices.
3.3 Evaluating hypothesis
Regression 1 tests whether the coefficient β1 is significantly different from zero. If the variable β1is significantly different from zero the null hypothesis is rejected, indicating an oil price effect on AEX index returns.
Hypotheses set 1: The null hypothesis is:
𝐻0: β1 = 0 Oil price effect not significant
Which is tested against the alternative hypotheses: 𝐻𝐴: β1 ≠ 0 Oil price effect significant
Regression 2 tests whether the coefficient α1 is significantly different from zero. If the variable
α1 is significantly different from zero the null hypothesis is rejected, indicating an oil price effect
Hypotheses set 2: The null hypothesis is:
𝐻0: α1 = 0 Oil price effect not significant
Which is tested against the alternative hypotheses: 𝐻𝐴: α1 ≠ 0 Oil price effect significant
4.Empirical study
The AEX index is a representative of the 25 largest Dutch companies listed on Euronext Amsterdam. The AEX is a market capitalization weighted index, meaning that the individual stock performance is reflected in the index with respect to the weight of the stock (Euronext, 2017).
Company Symbol Cty Sector Weight (%)
AALBERTS INDUSTRY AALB NL Industry good& services 0.80
ABN AMRO GROUP ABN NL Banks 1.77
AEGON AGN NL Insurance 1.79
AHOLD DELHAIZE AD NL Retail 4.48
AKZO NOBEL AKZA NL Chemicals 3.51
ALTICE ATC NL Telecommunications 1.00
ARCELOR MITTAL MT NL Basic resources 3.43
ASML HOLDING ASML NL Technology 11.93
BOSKALIS WESTMIN BOKA NL Construction and Materials 0.51
DSM KONINKLIJK DSM NL Chemicals 2.75
GALAPAGOS GLPG NL Health Care 0.56
GEMALTO GTO NL Technology 0.73
HEINEKEN HEIA NL Food and beverages 3.82
ING GROEP N.V. INGA NL Banks 11.33
KPN KONINKLIJK KPN NL Telecommunications 1.89
NN GROUP NN NL Insurance 2.10
PHILIPS KONINKLIJK PHIA NL Healthcare 5.59
RANDSTAD RAND NL Industrial goods & services 1.16
RELX REN NL Media 3.53
ROYAL DUTCH SHELL RDSA NL Oil & Gas 15.92
SBM OFFSHORE SBMO NL Oil & Gas 0.48
UNIBAIL RODAMCO UL NL Real Estate 3.98
UNILEVER UNA NL Personal & Household goods 14.09
VOPAK VPK NL Industrial goods & services 0.49
WOLTERS KLUWER WKL NL Media 2.38
Table 2 shows the components of the AEX by company at the start of 2017. The composition of the index is dynamic and changes over time. Only the 25 stocks listed at Euronext with the highest market capitalization are included in the index.
Table 2 indicates that the largest sector in the AEX index is represented by oil and gas industry companies. The banking, personal & household goods and technology sector also represent a large share of the index. Figure 2 presents a plotted sector allocation of the AEX index (see appendix A).
4.1 Data description
The empirical analysis is limited to the period between 1997 and 2017. I focus on this period, because most research on the effects of oil prices on stock markets is prior to 2003. Driesprong et al. (2008) examined the period 1973 until 2003. Sadorsky studied the period in post-World War II (1950-1996). Hamilton focused on the timeframe between 1948 and 1972. The period after 2003 seems not to be investigated in the literature.
Table 3 provides descriptive statistics for the variables used in regression 1. The dependent variable 𝑅𝑡, the main explanatory variable 𝐵𝑅𝐸𝑁𝑇𝑡 and the relevant control variables
𝐸𝑋𝐶𝐻𝑡,𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡, 𝐺𝐷𝑃𝑡. The minimum and maximum values can be observed together with the standard deviation and the mean. Table 3 also provides insight on the data’s distribution by evaluating the kurtosis and skewness.
Table 3: Descriptive statistics
Variable N Mean Std. Minimum Maximum Kurtosis Skewness
𝑅𝑡 252 0.0050 .0600 -.2207 .1503 4.2779 -.7078
𝐵𝑅𝐸𝑁𝑇𝑡 252 0.0087 .0885 -.3142 .3376 3.8552 -.1737
𝐸𝑋𝐶𝐻𝑡 252 -0.0005 .0290 -.0888 .1036 3.6989 .0235
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 252 0.0345 .0170 -.0015 .0660 2.2914 -.4525
𝐺𝐷𝑃𝑡 252 0.0074 .0097 -.0163 .0666 17.8087 2.2458
This table present descriptive statistics for the variables of equation 1. The variable 𝑅𝑡 implies the AEX reinvestment index returns in month t. The
variable 𝐵𝑅𝐸𝑁𝑇𝑡 represents Brent oil returns. The control variables 𝐸𝑋𝐶𝐻𝑡,𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡, 𝐺𝐷𝑃𝑡 indicate Euro/Dollar returns, 10-year German yield and
GDP growth rate of the Netherlands (all in month t), respectively. (Period 1997-2017)
The average monthly AEX return in the period 1997 until 2017 is positive 0.450%. For Brent oil and Euro/Dollar exchange rate the average monthly returns are 0.870% and -0.0503%. The average yield on the 10-year German government bond is 3.450% and the GDP monthly growth rate is 0.736%. The minimum return on the AEX index is -22.07% and the maximum observation is 15.03%. The minimum return was recorded in September 2018 at the start of the subprime mortgage crisis. The highest return was recorded in the early part of 2003 when stock markets started recovering from the stock market collapses caused by the dot com bubble. The minimum and maximum values reported for Brent oil are -31.424% and 33.760%, respectively. The decrease
of 31.424% was recorded for November 2008 and was a direct result of the financial crisis that year. The highest monthly increase in price happened in March 1999.
Table 3 also presents results with respect to kurtosis and skewness. With exception from the variable 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 all coefficients reject the null hypothesis on kurtosis (𝐻0: 𝐾1,2,3,4 ≤ 2; 𝐻𝐴: 𝐾1,2,3,4> 2). This means the data in regression 1 includes some ‘extreme ‘values or outliers. Yet,
these rather extreme outcomes do not need to be excluded from the regression as the sample size is large enough (n=252). In the last column outcomes on skewness are presented. All values except for 𝐺𝐷𝑃𝑡 are non-skewed (𝐻0: -1 ≤ 𝑆1,2,3,4 ≤ 1; 𝐻𝐴; 𝑆1,2,3,4 > 1, or: 𝑆1,2,3,4 < - 1). The value on skewness for 𝐺𝐷𝑃𝑡 equals 2.246 indicates that there is a higher probability for a growth rate in the Netherlands, this outcome is obvious as we assume economies tend to grow on average.
Table 4 provides the correlations between the variables used in regression 1. The highest correlation presented is between 𝐺𝐷𝑃𝑡 𝑎𝑛𝑑 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 (0.348). The magnitude of the coefficient
is of similar size as in the study of Ang, Piazzesi and Wei (2006). In their study they argue that a positive relationship between GDP rate and yield curves coexists. The rationale behind this finding is that yields are used as a tool on GDP growth stimulation by central banks. Table 4 also demonstrates a positive relationship between AEX index returns and Brent oil returns in line with the main hypothesis of this study. Another correlation between EXCHt and BRENTt (0.279) is
presented in table 4. Wang, Shan and Ling (2009) argue that a depreciation of the dollar (a rise in the euro/dollar returns) results in a price increase of commodities like gold and oil.
Table 4: Correlation Matrix
Variable 𝑅𝑡 𝐵𝑅𝐸𝑁𝑇𝑡 𝐸𝑋𝐶𝐻𝑡 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 𝐺𝐷𝑃𝑡 𝑅𝑡 1.000 𝐵𝑅𝐸𝑁𝑇𝑡 0.1726 1.000 𝐸𝑋𝐶𝐻𝑡 -0.0334 0.2790 1.000 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 -0.0076 0.0663 0.0379 1.000 𝐺𝐷𝑃𝑡 -0.0146 0.0307 0.0200 0.3482 1.000
Table 5: Number of observations regression 2
Variable N Number of observations per
group 𝐵𝑅𝐸𝑁𝑇𝑡 3263 251 𝐸𝑋𝐶𝐻𝑡 3263 251 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 3263 251 𝐺𝐷𝑃𝑡 3263 251 𝑂𝐼𝐿𝑖 3263 251 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 3263 251
This table present descriptive statistics for the variables of equation 2. 𝑅𝑡 implies the AEX reinvestment index returns in
month t. The variable 𝐵𝑅𝐸𝑁𝑇𝑡 represents Brent oil returns. The control variables 𝐸𝑋𝐶𝐻𝑡,𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡, 𝐺𝐷𝑃𝑡 indicate
Euro/Dollar returns, 10-year German yield and GDP growth rate of the Netherlands (all in month t), respectively. The variable 𝑂𝐼𝐿𝑡 represents a dummy variable (1= significant oil price effect for stock i).
𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡. represents an interaction variable to control for firm level differences on the effect of oil price. Regression
2 consists of 13 groups, with 251 obs. per group (Period 1997-2017).
4.2 Multi collinearity
Table 6 provides the results from a multicollinearity check. The right column represents the outcomes of the variance inflation factors. All coefficients except the null-hypothesis of no multicollinearity (𝐻0:VIF(𝛽̂1,2,3,4) ≤ 4; 𝐻𝐴: 𝑉IF(𝛽̂1,2,3,4) > 4). Moreover, table 5 presents the correlation matrix between the variables used in regression 1. None of the variables are highly correlated and all coefficients accept the null hypotheses ( 𝐻0: | ρ |1,2,3,4 ≤ 0.7, 𝐻𝐴: | ρ |1,2,3,4 > 0.7)
Table 6: Multicollinearity check regression 1
This table presents the standard errors and variance inflation factors from regression 1
4.3 Regression analysis and results
The main finding in this study show that oil prices have a positively significant relationship with stock returns in the period between 1997 and 2017. In the literature multiple studies report a negative significant response from oil prices on stock returns (Hamilton, 1983; Driesprong et al, 2008; Sadorsky, 2009; Nandha and Faff 2008; Miller and Ratti, 2009). Table 7 provides the results from regression 1 when considering the predictability of the main explanatory variable and control
Variable Std. Error VIF
𝐵𝑅𝐸𝑁𝑇𝑡 .0441 1.0900
𝐸𝑋𝐶𝐻𝑡 .1344 1.0800
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 .2381 1.1400
variables on the dependent variable. The joint F test value is 2.41, this indicates the coefficients are jointly unequal to zero and significant at the at the 5% level (0.050>α= 0.05.
The first column is table 7 shows the coefficient estimates from regression 1. The variable 𝐵𝑅𝐸𝑁𝑇𝑡 is significant at 5% significance level, (0.003/2<α= 0.05), a 1% change in the price of Brent Oil results in a 0.1346% change of the AEX index. The coefficient EXCHt is not significant at the 5% level (0.177/2>α= 0.05). At the 10% level the variable is significant. The coefficient estimate indicates that a 1% change in EXCHt results in a decrease of 0.182% in 𝑅𝑡. This result is supported by a study from Bahmani and Sohrabian (1992). In their empirical results they found a significant negative relationship between stock market returns and the exchange rates. 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡
(0.854/2>α= 0.05) and 𝐺𝐷𝑃𝑡 (0.8264/2>α= 0.05) are both insignificant. Table 7: Regression 1
Variable Coefficient estimate t-statistic p-value
a .0059 0.69 0.49
𝐵𝑅𝐸𝑁𝑇𝑡 .1346 3.05 0.003
𝐸𝑋𝐶𝐻𝑡 -.1819 -1.35 0.177
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 -.0439 -0.18 0.854
𝐺𝐷𝑃𝑡 -.0905 -0.22 0.826
This table presents the estimation results from regression 1.
Table 8 presents the regression results from regression 2. The dummy variable 𝑂𝐼𝐿𝑖 is not
significant (0.308>α= 0.05). This is because the dummy variable only functions as a tool to demonstrate which stocks are more sensitive for changes in oil prices.
𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 is very significant at both the 5% and the 1% level (0.000<α= 0.01). This result
is obvious because the interaction variable only reacts to stocks for which the oil effect is significant (1 = Significant oil effect for stock i).
The coefficient for the main regressor 𝐵𝑅𝐸𝑁𝑇𝑡 indicates an average effect of .0484 on
𝐵𝑅𝐸𝑁𝑇𝑡. Interpretation of the coefficients is tricky because the coefficient includes both the within
entity and between entity effects. The coefficient is estimate is significant at the 10% level (0.05<α= 0.1).
Table 8: Regression 2
Variable Coefficient estimate z-statistic p-value
a .0137 2.40 0.016 𝐵𝑅𝐸𝑁𝑇𝑡 .0484 1.91 0.050 𝐸𝑋𝐶𝐻𝑡 -.0911 -1.40 0.161 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 .1692 1.47 0.143 𝐺𝐷𝑃𝑡 -.6747 -3.40 0.001 𝑂𝐼𝐿𝑖 -.0079 -1.02 0.308 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 .1742 3.93 0.000
This table presents estimation results for the variables of equation 2. The variable 𝑅𝑡 implies the AEX reinvestment index
returns in month t. The variable 𝐵𝑅𝐸𝑁𝑇𝑡 represents Brent oil returns. The control variables 𝐸𝑋𝐶𝐻𝑡,𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡, 𝐺𝐷𝑃𝑡
indicate Euro/Dollar returns, 10-year German yield and GDP growth rate of the Netherlands (all in month t), respectively. The variable 𝑂𝐼𝐿𝑡 represents a dummy variable (1= significant oil price effect for stock i).
𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡. represents an interaction variable to control for firm level differences on the effect of oil price. (Period
1997-2017). For this regression to be valid we assume that differences among regressors are uncorrelated the between entity errors, corr(𝑢𝑖, 𝑂𝑖𝑙𝑖 ) = 0.
4.4 Heteroscedasticity check
To control for heteroscedasticity, I run regression 1 and 2 using robust standard errors. Table 9 and 10 present the results. 𝐵𝑅𝐸𝑁𝑇𝑡 is still significant (0.007/2<α= 0.05), but increased slightly from p=0.003 to p=0.007 (regression 1). The control variables 𝐸𝑋𝐶𝐻𝑡, 𝐺𝐷𝑃𝑡 𝑎𝑛𝑑 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 are
still insignificant. This indicates that the error term is equal among all variables and that the coefficient estimates are not overestimating.
In regression 2, 𝐵𝑅𝐸𝑁𝑇𝑡 is still significant (0.016<α= 0.05) and decreased from p=0.056 to
p=0.016. The control variables 𝐸𝑋𝐶𝐻𝑡, 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 are still insignificant. 𝑂𝐼𝐿𝑖 is still insignificant but increased slightly from p=0.308 to p=0.118. The variable 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 is still very
significant with p=0.000. This indicates that the error term is equal among all variables and that the coefficient estimates are not overestimating.
Table 9: Regression 1 (robust std. errors)
Variable Coefficient estimate Robust Std. Error t-statistic p-value
a .0059 .0062 0.94 0.347
𝐵𝑅𝐸𝑁𝑇𝑡 .1346 .0495 2.72 0.007
𝐸𝑋𝐶𝐻𝑡 -.1820 .1432 -1.27 0.205
𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 -.0440 .2169 -0.20 0.840
𝐺𝐷𝑃𝑡 -.0904 .5833 -0.16 0.877
This table represents the results from regression 1 using robust std. errors.
Table 10: Regression 2 (robust std. errors)
Variable Coefficient estimate Robust Std. Error z-statistic p-value
a .0136 .0059 2.33 0.020 𝐵𝑅𝐸𝑁𝑇𝑡 .0484 .0201 2.40 0.016 𝐸𝑋𝐶𝐻𝑡 -.0911 .1103 -0.83 0.409 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 .1692 .0700 2.42 0.016 𝐺𝐷𝑃𝑡 -.6747 .1185 -5.69 0.000 𝑂𝐼𝐿𝑖 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 -.0079 .1742 .0050 .0271 -1.56 6.42 0.118 0.000 This table represents the results from regression 2 using robust std. errors.
4.5 Robustness
Lu and White (2014) argue that a robustness check is conducted by adding or removing regressors from the model to examine how the core regression coefficient estimates behave. To examine the validity of the core coefficients in the models (1) and (2), I drop a part of the control variables.
I regress AEX index monthly returns on Brent oil monthly returns together with 𝐸𝑋𝐶𝐻𝑡. I
drop the control variables 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 and 𝐺𝐷𝑃𝑡 because they are non-significant. The estimation
results are presented in table 14.1. The coefficient estimates for 𝐵𝑅𝐸𝑁𝑇𝑡 decreased from .1346
Table 11.1: Robustness check
Variable Coefficient estimate t-statistic p-value
a .0037 1.00 0.318 (.0037) 𝐵𝑅𝐸𝑁𝑇𝑡 .1338 3.05 0.003 (.0439) 𝐸𝑋𝐶𝐻𝑡 -.1829 -1.37 0.173 (.1339)
This table presents the estimation results for the variables of equation (1) using the following model: 𝑅𝑡= β0 + β1𝐵𝑅𝐸𝑁𝑇𝑡
+ β2𝐸𝑋𝐶𝐻𝑡 + 𝜀𝑡.
I regress 𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡 on Brent oil monthly returns together with 𝑂𝐼𝐿𝑖 and 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡. I drop the control variables 𝐸𝑋𝐶𝐻𝑡, 𝑌𝐼𝐸𝐿𝐷𝑡𝑒𝑛𝑡 and 𝐺𝐷𝑃𝑡 because they are non-significant. The
estimation results are presented in table 14.2 The coefficient estimate for 𝐵𝑅𝐸𝑁𝑇𝑡 decreased from
.0484 to .0402 and remains significant. The coefficient estimate for 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 remains
unchanged. This is obvious because the dummy variable does not react to changes over time but only reacts to changes in entity.
Table 11.2: Robustness check
Variable Coefficient estimate z-statistic p-value
a .0147 3.40 0.001 (.0043) 𝐵𝑅𝐸𝑁𝑇𝑡 .0402 1.63 0.103 (.0246) 𝑂𝐼𝐿𝑖 -.0079 -1.02 0.308 (.0078) 𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 .1742 3.92 0.000 (.0444)
This table presents the estimation results for the variables of regression 1 using the following equation: 𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡= α0 + α1 𝐵𝑅𝐸𝑁𝑇𝑡 + α5𝑂𝐼𝐿𝑖+ α6𝑂𝐼𝐿𝑖∗ 𝐵𝑅𝐸𝑁𝑇𝑡 + 𝑢𝑖𝑡 + 𝜀𝑖𝑡.
The estimation results presented in table 11.1 and 11.2 demonstrate that the core coefficients are plausible and robust because they remain significant and do not react much to changes in the models.
5. Discussion and conclusion
Many studies report a negative significant effect between oil prices and stock returns (Driesprong et al., 2008; Hamilton, 1983; Sadorsky, 1999). The results from this study reveal a positive significant relationship between oil prices and stock market returns. This study is related to the work of Driespong et al (2008), Sadorsky (1999) and Hamilton (1983), but focused on a smaller economy. To empirically determine the relationship between oil prices and AEX index returns, this study used both ordinary least squares and random effects panel data methodology. This methodology allows to control for firm level differences. This study covered the AEX index and 13 firms listed on the AEX index over the period from 1997 to 2017.
The results from regression 1 indicate that the coefficient 𝐵𝑅𝐸𝑁𝑇𝑡 is significant at the 5%
significance level. A 1% change in the price of Brent Oil results in a 0.1346% change of the AEX index. Regression 2 controls for firm level differences on the effect of oil prices. The coefficient for the main regressor 𝐵𝑅𝐸𝑁𝑇𝑡 indicates an average effect of 0.0484 % on 𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡, for changes in Brent across time and between individual stocks for a 1% change in 𝐵𝑅𝐸𝑁𝑇𝑡. Careful
interpretation of the coefficients is needed because the coefficient includes both the within entity and between entity effects. These results show a positive significant effect of oil prices on the Dutch stock market. Results from regression 2 show that the relationship between oil prices and stock returns is less significant when controlling for firm level differences on the effect of oil prices. This indicates that when controlling for firms which react more to the effect of oil prices the predictability of the main explanatory variable 𝐵𝑅𝐸𝑁𝑇𝑡 decreases but still indicates a significant effect at the 10% level.
Driesprong et al. (2008) argued that changes in oil price have an impact on real output of firms as it affects the firms production costs and expected earnings. This results in a negative relationship
between oil prices and stock returns. Figure 2 appendix (A) shows that the construction and materials, industrial goods & services and basic resources, only account for approximately 10% of the composition of the total index. Since especially these industries are negatively affected by increases in oil prices this does not have a great effect the AEX index returns.
Another possible explanation for this result could be the relatively high weight of Royal Dutch Shell, which accounts for approximately 15% of the index. Regression results from table 3 (see appendix C) show that the effect of oil price is highly significant on the stock returns for Royal Dutch Shell. This table shows that a 1% change in oil price results in a 0.2705% in its stock return.
Sadorsky (2001) concludes that oil prices are positively correlated to companies in operating in the oil & gas industry. This because there cashflows heavily depend on the price of oil.
In this study multiple of limitations arise. A bidirectional causality between stock market returns and the regressors might exist. e.g. a relationship that exists in two ways (A→B; B→A). The Granger test is a technique to measure the bidirectional concept of causality. In this study this test was not conducted.
There are multiple other factors affecting stock markets like corporate earnings, inflation figures and market sentiment. For simplicity, not all these variables were incorporated in the models used in this study.
Appendix A:
Figure 2 represents the sector allocation of the AEX index in %
Appendix B:
My model (1) is based on those of Driesprong et al (2008), and Sadorsky (2009), which are shown below:
𝑅𝑡 = µ + 𝐴1𝑟𝑡−1+ 𝑒𝑡 with 𝑒𝑡 = 𝑟𝑡− 𝐸𝑡−1 [𝑟𝑡 ] (B1), where:
𝑅𝑡 = MSCI- world reinvestment index;
𝑟𝑡−1 = Lagged oil price, t indicates the lag in days; µ = Constant;
𝑒𝑡 = Disturbance term.
Driesprong et al. used the equation above to test for an existing effect of oil price on stock market returns. The only explanatory variable in their model is the lagged oil price 𝑟𝑡−1.
0 2 4 6 8 10 12 14 16 18
Oil & Gas Personal & Household goods Banks Technology Chemicals Health Care Media Retail Real Estate Insurance Food and beverages
Basic resources Telecommunications Industrial goods & services Construction and Materials
𝑋𝑡 = ∑𝑝𝑖=1𝐴𝑖𝑋𝑡−𝑖 + 𝑒𝑡 with 𝑋𝑡 = (Δl𝑟𝑡, Δl0𝑡, Δ𝑙𝑖𝑝𝑡, 𝑟𝑠𝑟𝑡) (B2),
where:
𝑒𝑡 = The disturbance term acts as a factor of all disturbances (𝑒𝑟, 𝑒𝑜, 𝑒𝑝 , 𝑒𝑘 ) with
E(𝑒𝑡, 𝑒𝑡′) = ∑
𝑟 = Interest rates of the 3-month T-Bill; 𝑜 = West Texas Intermediate oil price; 𝑝 = Industrial production;
𝑘 = Reel stock returns.
The disturbance term acts as a factor of all disturbances (𝑒𝑟, 𝑒𝑜, 𝑒𝑝 , 𝑒𝑘 ) with E (𝑒𝑡, 𝑒𝑡′) = ∑.
The terms of disturbances indicate shocks in interest rates of the 3-month T-Bill, oil prices, industrial production and reel stock returns. They find a negative correlation between changing oil prices and stock returns. They also find a negative correlation between interest rates and stock market returns.
Appendix C:
Table 1 presents the results from a set of regressions conducted on equation 3 when considering the predictability of the main explanatory variable and control variables on individual stock returns. The variable 𝐵𝑅𝐸𝑁𝑇𝑡 is significant at the 5% level for the stocks Akzo Nobel (0.048<α= 0.05), DSM (0.000<α= 0.05), ING (0.032<α= 0.05) and Shell (0.000<α= 0.05).
Table 1: Regression results from equation 3
𝑆𝑡𝑜𝑐𝑘𝑖 Coefficient estimate t-statistic p-value
Aegon .0657 0.67 0.501 Ahold .1241 1.60 0.112 Akzo Nobel .1320 1.99 0.048 ASML .1460 1.40 0.162 DSM .2340 3.96 0.000 Heineken .0233 0.48 0.633 ING .1968 2.16 0.032 KLM -.0246 -0.23 0.819 KPN .0093 0.10 0.921 Philips .1334 1.62 0.106 Shell .2705 5.48 0.000 Unilever .0080 0.17 0.867 Kluwer .0079 0.69 0.493
This table represents the regression results from equation 3 of the variable 𝐵𝑅𝐸𝑁𝑇𝑡.Equation 3 is 𝑆𝑡𝑜𝑐𝑘𝑟𝑒𝑡𝑢𝑟𝑛𝑖= β0 + β1𝐵𝑅𝐸𝑁𝑇𝑡 +
Reference list:
Bahmani, M.O., “Sohrabian, A Stock prices and the effective exchange rate of the dollar”
Applied Economics, 1992, 24 (4), p.459-464.
Basher, S.A., Haug, A.A., Sadorsky, P. “Oil prices, exchange rates and emerging stock markets” Energy Economics, 2012, 34(1), p.227.240.
Basher, S.A., Sadorsky, P., “Hedging emerging market stock prices with oil, gold, VIX, and bonds: A comparison between DCC, ADCC and GO-GARCH” Energy Economics, 2016, 54, p.235-247.
DeBondt, W. F. M. and Thaler, R. H. (1985) Does the stock market overreact? Journal of Finance, 40 (1), 793-805.
Dow, J., Gorton, G.” Stock market efficiency and economic efficiency: is there a connection?”
Cambridge, Mass.national bureau of economic research, 1995
Driesprong, G., Jacobsen, B. & Maat, B. “Striking oil: Another puzzle?” The Journal of Financial Economics, 2008, 89 (2), pp.307-327.
Euronext (2017): www.euronext.com/nl/products/indices/NL0000000107-XAMS/market-information
ECB, “Global implications of low oil prices” Economic Bulletin 2016 (4)
https://www.ecb.europa.eu/pub/pdf/other/eb201604_focus01.en.pdf?0deb0ff14d70fa91bdeb086c 2c1c49d1
Filis, G., “Macro economy: stock market and oil prices: Do meaningful relationships exist among their cyclical fluctuations” Energy Economics, 2010, pp.877-886.
Hamilton, J.D., “Oil and macro-economy since world war II", Journal of Political Economy, 1983, 90 (2), pp. 228-247.
Ito, M., Noda, A., Wada, T.,” International stock market efficiency: a non-Bayesian time-varying model approach” Applied Economics, 2014, p.1-11.
Jones, C.M., Kaul, G., “Oil and stock Markets”, Journal of Finance, 1996, 51 (2), pp. 463-491.
Killian, L., Park, C. “The impact of oil price shocks on the U.S. stock market” International Economic Review, 2009, 50 (4), pp. 1267-1287.
Lu, X. and White, H. (2014) Robustness checks and robustness tests in applied economics, 178 (1), p.194-206
Park, J., Ratti, R., “Oil price shocks and stock markets in the U.S. and 13 European countries” Energy economics, 2008, 30 (5), pp. 2587-2608.
Pearce, D.K., Roley, V.V. “The Reaction of Stock Prices to Unanticipated Changes in Money” Journal of Finance” 1983, 38(4), pp.1323-1333.
(Pilbeam, 2006, p.113).
Sadorsky, P., “Oil price shocks and stock market activity” Energy Economics, 1999, 21 (5), pp.449-469.
Sadorsky, P., “Risk factors in stock returns of stock returns among Canadian oil and gas companies” Energy Economics, 2001, 23 (1), pp.17-28
Volpert, K., “How well do long-term bond interest rates predict stock market returns?” The journal of investing, 2013, 22 (2), pp. 23-28.
