The partisan effect of elections on stock markets

Transcription

1 The partisan effect of elections on stock markets Bas Gerrits S Tilburg School of Economics and Management Department of Finance Dr. Paul Sengmuller

2 Master Thesis: The partisan effect of elections on stock markets K.W. Gerrits S Tilburg School of Economics and Management Department of Finance Dr. Paul Sengmueller 30 th August 2011

3 Table of Contents 1. Introduction Theory and previous research Political business cycle theory Rational partisan theory Practical implications of the rational partisan theory Structure of the thesis Hypotheses construction and expectations Main research goal Study boundaries Research questions and hypotheses Regression variables Methodology Dependent variables Independent variables Event study Regressions Data Elections Independent variables Results Event study Results by election outcome Industry results by election outcome Results by decade... 31

4 6.2 Regressions Regression results Robustness Industry regressions Conclusion Discussion and recommendations References

5 1. Introduction Political decisions affect the economy of every country. For example in China where the government has decided to focus more on economic growth and to open up to the world. This decision to open up has transformed the world, for instance since China is so large that prices have increased due to increasing demand from China. In every country governments have to make decisions like the Chinese government that affect their economy and possibly the global economy. These governmental decisions and policies affect a lot of people worldwide. Think about entrepreneurs, managers, investors and also the population of a country. Therefore it could be very useful for some of these stakeholders to know what the effects of partisanship are on the economy of a country. In a world were political decisions influences the lives of almost everyone, elections have become very important events. Elections determine what party will be in charge for the next term. The ideology of the party in office determines a large part of the decision making. The ideology of a party determines the decisions it makes, these ideologies can be grouped in different orientations. In general left wing parties are thought to defend the rights of the labour force, whereas right wing parties are thought to create a stimulating climate for entrepreneurs. The goals of the left wing parties express itself in certain policies, like imposing regulations, increasing taxation and increasing government size. While the right wing goals express themselves in other policies, like less government intervention, lower taxes and a smaller government. Previous research has focused solely on the opposing forces of left wing parties and right wing parties. The recent history of political unification, like the European Union and global financial integration has caused these opposing forces to converge. Governments are forced to synchronize their policies in order to achieve goals set by their political union, in general this allows for less extreme left or right wing policies. Therefore the power of left and right wing policies decreases and more and more policies converge to become more center oriented. This study tries to identify the partisan effects of elections on stock markets. In this study the possible orientations are not only left and right but also center. These possible orientations enable identification of trends toward center orientation, which in past research has never been studied. The focus of this study is not only to identify the partisan effects of elections on stock markets, but it also focuses on trends in time due to changing circumstances like the end of the cold war, increasing political globalisation and global financial integration. 1

6 2. Theory and previous research In order to study the effects elections have on stock markets some assumptions have to be made. The first assumption is that stock market prices are based on expectations of future values of the underlying assets of the stock. When expectations change due to changing circumstances, for instance elections, the expectation of the future value could change. This change in expectation of the future value is represented by a change of the stock market price today. There are three models about which information influences stock market prices; weak form efficient market hypothesis, the semi-strong form efficient market hypothesis and the strong form efficient market hypothesis. The weak form of the efficient market hypothesis (EMH) states that only past public information is incorporated into stock prices. The semi-strong form of the EMH states that all publicly available information is incorporated immediately into the stock price, while the strong form EMH suggests that all information, including insider information, is incorporated into the stock price. Since strong form EMH is commonly believed to be incorrect, in this study the semi-strong form EMH is assumed. The assumptions made above allow for theories concerning the effects of elections on stock markets. When elections are held future expectations about companies are changed and therefore the future value of those companies can change. According to the semi-strong form EMH information about elections and election results could impact stock prices, but only if they deviate from previous expectations. 2.1 Political business cycle theory The first to pose a theory concerning partisanship and the effects of partisanship was Downs (1957). His theory suggests that the sole interest of political parties is getting or staying in office. As a result of this one goal, the parties have no interest in good governance or the effects of their policies on the economy. Their only interest is keeping the voter happy enough such that they will vote for their party (again) the next time. Good policy is only executed if it has the expected effect that they would stay in office. The interest of the voter is not pursued in this model. The effects of partisanship in this model are absent. This is since the parties have the same objectives and therefore pursue the same policies when they get in office. The only possible difference arises when there are differences in information about the electorate and their preferences. 2

7 This theory has led to some macroeconomic models, one of the most important is called the political business cycle, used by Nordhaus (1975) and McRae (1977). This model suggests recessions at the start of the tenure of governments, while it suggests inflationary expansions at the end of their tenure. The political business cycle theory is tested in later empirical research. Although Nordhaus found evidence for the political business cycle theory, most empirical research found no evidence. For example McCallum (1978) and Golden & Poterba (1980) rejected the theory based on empirical research. This approach isn t pursued much further in later research, due to the lack of evidence. 2.2 Rational partisan theory Later theories concerning partisanship and economic effects of partisanship have a different approach. This approach suggests that different parties have inherently different preferences concerning the intrinsic properties of their economic policies. They care about the outcomes of the policy they apply. The different parties have different constituencies and therefore they have different points of interests, since the different constituencies have different needs and goals. So even if the parties have the same information about the voters preferences they still pursue different policies. The only way that parties get to implement their vision, their policies, is by participating in elections with uncertain results. Theories that make these assumptions in most cases also assume two groups of orientations, left wing parties and right wing parties. The properties of left wing party-policy are considered to be a focus on equality and growth, while right wing party-policy is mostly focussed on consumption and growth. The effects of these different objectives are that a left wing government is expected to have amongst others a higher tax rate, higher minimum wages (Leigh, 2007) and higher government spending (Barro, 1991; 1996) compared to a right wing government. A large part of research concerning partisanship is conducted in the United States. The two parties there are the Republicans and the Democrats. Their preferences and their ways of achieving their goals are consistent with the left-right division made in other research. The Republicans are considered to have the same preferences as parties marked as right wing. The Democrats are considered to have the same preferences as left wing parties. The Republicans pursue consumption driven growth, while the Democrats opt for investment driven growth (Quinn & Shapiro, 1991). This is consistent with the assumption that left wing parties have a higher government spending as part of GDP (Barro, 1991; 1996). Implementing the policies 3

8 of each party result in different macroeconomic outcomes. Democrats generally have higher growth and lower unemployment, while Republicans in general have lower inflation (Hibbs, 1987; Alesina & Rosenthal, 1995). Taking these assumptions about right and left wing parties into consideration, Alesina (1987) poses the rational partisan theory (RPT). This theory is built upon a model which assumes there is a polarised party system (left vs. right or Democrats vs. Republicans) in a repeated game. The left (or Democratic) parties in this system have the preference to minimize unemployment, while the right (or Republican) parties have the preference to minimize the inflation. Each party thus has incentives to implement different policies. However in the long run policy changes due to changes in power, diminish the effect of policies put in place just for their own constituencies. Therefore better results are expected when both parties slightly converge in their policies. When parties converge in their policies fewer and smaller policy changes will be made and therefore better results for both parties in the long run will be realised. Alesina constructs a model for this interaction between polarised parties, he suggests that wage setters are the base for the economic effects of different policies by different parties, since both parties have different preferences about minimum wage (Leigh, 2007). The model suggests precise empirical implications. When a right government gets elected into office the first part of their tenure there will be low inflation, while a left government will experience growth and higher than average inflation. In the last part of their tenure both will experience the same growth number, however left governments will experience higher inflation. These empirical implications are confirmed for the US by Alesina & Sachs (1988). 2.3 Practical implications of the rational partisan theory One of the practical implications of the RPT on firm performance is that firms are expected to perform better under a right wing government. Bechtel & Fuss (2008) studied the implications of elections on stock market returns and volatilities; they found that firms indeed perform better under right wing governments. In anticipation of right wing election success stock markets anticipate and have positive abnormal returns, while left wing success generates negative abnormal returns. This effect is stronger for small firms, which could be since they are less diversified and therefore more sensitive to macro-economic changes and since these firms are less likely to be multinationals (Cooley, 2009). Bechtel & Fuss further found that uncertainty before the election reduces volatility and that expected success for right wing 4

9 parties increases volatility and with volatility the returns increase accordingly. According to Leblang & Mukherjee (2007) left wing electoral success generates lower stock market volatility after the election which corresponds to lower returns. The RPT suggests that differences only appear in the first years after an election, Cooley tested this assumption and confirmed that it holds. He also confirms that growth figures in later years after the election are not distinctly different, which is also in accordance to the RPT. Cooley suggests that the main driver behind the differences in growth are caused by differences in unemployment figures between left and right wing governments. Cooley used stock returns as a proxy for growth in his study. Cooley also found that surprising election results generate greater after election abnormal results. The effect of surprising election results on abnormal returns can be explained by the semi strong form of the efficient market hypothesis (EMH). The semi strong form of the EMH poses that publicly available information is incorporated into the stock price. Thus when election results are anticipated this is reflected in stock prices. However if election results do not match pre-election expectations or if there were unclear expectations, the election information is not correctly incorporated into the stock market prices. Therefore after a surprising election result excessive abnormal returns are expected in order to incorporate changed expectations and public information into the stock prices. In the last decade abnormal returns as a consequence of elections have decreased, since party differences have become smaller. This convergence of party differences is caused by economic integration and political unification. This effect was suggested by Castle & Mair (1984), later Bechtel & Fuss found evidence for this decreasing difference in their study. Past research and theories concern country wide effects, however elections also influence the stock market results of specific industries. Left wing parties have an ideology that states that everybody should be equal, meaning equal rights and equal possibilities. In order to achieve this left wing parties are considered to impose regulations on certain industries, increase taxation and increase government size. For certain industries the imposed regulations by left wing governments are limiting their abilities and decreasing their profitability. Therefore certain industries are more influenced by election results than other industries that are not affected by regulations imposed by left wing governments. 5

10 2.4 Structure of the thesis This thesis is structured as follows. In the next chapter the different hypotheses that are the focus of this paper are presented. The fourth chapter will consist of the methodology of the study. The data that is used in this study will be presented in the fifth chapter, while the sixth chapter consists of the results that are produces by this study. The seventh chapter will draw conclusions based on the results found and presented in the sixth chapter, while in the eighth chapter the results are discussed and recommendations about further research are made. 6

11 3. Hypotheses construction and expectations 3.1 Main research goal The main goal of this research is to study the effects of elections on stock market returns and in particular the effects of partisanship in elections on stock market returns. To get a full understanding of the effect of elections on stock markets and what the determinants of this effect are this research poses and tries to answer different research questions. The results and answers of these research questions should provide insight into the effects of elections on stock market returns. 3.2 Study boundaries Before these research questions can be answered it is important to determine what exactly is being studied. The precise subjects of this research are elections for houses of representatives in OECD 1 countries in the period of 1975 till Houses of representatives are used, since it is the only house that is always publicly elected for every form of government present in the OECD countries. OECD countries are observed since these countries are in general politically stable and the needed data is available for most of the elections in these countries during the specified period. For other sets of countries these characteristics do not hold, especially the availability of information is often a problem. The OECD countries are therefore chosen for practical reasons. The time period is also chosen for practical purposes, since the study requires data on the orientation of political parties. This data is obtained from the DPI 2 and is only available for the time period. 3.3 Research questions and hypotheses In order to investigate the effects of elections on stock markets we first have to investigate whether there is in fact an effect of elections on stock markets. Therefore the first research question is: Is there in fact an effect of elections on stock markets in OECD countries in the time period? This effect can be identified by performing an event study on elections. For a number of elections data is gathered about amongst others cumulative abnormal returns (CAR s) of these 1 Organization for Economic Cooperation and Development 2 Database of Political Institutions, World Bank 7

12 elections. These CAR s can be checked whether or not they are different from zero. If these CAR s are different from zero an election effect on stock markets is present. Thus the hypothesis corresponding to the first research question is: H1: The average CAR of elections is different from zero. This hypothesis is expected to hold, since elections have a major influence on how countries and their economies develop. Different policies and regulations affect different companies and industries, therefore the stock markets are expected to show a reaction to elections. This is in line with the rational partisan theory (RPT) posed by Alesina (1987), which suggests that a left orientation of a party in office produces different effects on the economy than when the party in office has a right orientation. If investors expect these effects to take place they are likely to adjust their expectations about the future according to the orientation of the party in charge. At the moment of the election result the largest party is the most likely candidate to form a government and thus to become the party in charge. Therefore an election effect on stock markets is likely, however if this is positive or negative is hard to predict. Positive CAR s are realized if the election result is more beneficial to investors than they expected and negative CAR s occur when the election result is worse than investors expected it to be. In past research election effects have been found, however no expectations on the magnitude can be formed since most research involved case studies with election specific effects that are not relevant to other elections. The second part of gaining insight in the effects of elections on stock markets is determining whether or not there is a partisan effect. Thus the second research question is: Is there a partisan effect on stock markets after elections? To answer this research question partisan information on the parties participating in elections is collected as well as the actual election results. The partisan information states the orientation of a political party, the orientations used in this research are left, right and center. The corresponding hypotheses to this research question are: H2a: The average CAR of elections with left parties obtaining the most votes is different from the average CAR of elections with right parties obtaining most votes. H2b: The average CAR of elections with center parties obtaining most votes is different from the average CAR of election with right parties obtaining most votes. 8

13 H2c: The average CAR of election with left parties obtaining most votes is different from the average CAR of elections with center parties obtaining most votes. It is difficult to form an expectation about whether or not these hypotheses will hold. Popular conceptions are that left parties defend the interests of the labour force, and therefore are creating additional costs for companies. While right orientated party defend the interest of entrepreneurs, therefore creating a more favourable climate for companies. These popular conceptions would suggest that elections with left parties obtaining most votes realize lower CAR s than elections with right parties obtaining most votes. In contrast to this the RPT suggests that left oriented parties in office generate higher growth numbers for the economy in the first part of their tenure than do right oriented parties. These higher growth numbers are beneficial to (listed) companies. Therefore one could, based on this theory expect higher average CAR s for elections where left oriented parties obtained most votes than for elections where right oriented parties obtained most votes. Past empirical research does not provide a clear result for the question which party orientation outperforms the others. Alesina & Sachs (1988) found evidence that supports the RPT and thus that elections with left parties obtaining most votes realize higher CAR s than elections that have right oriented parties obtaining most votes. Bechtel & Fuss (2008) however find that elections with right oriented parties obtaining most votes have positive abnormal returns while elections with left parties obtaining most votes realize negative abnormal returns. These theories and expectations are all based on just two possible orientations, left and right. However in this research there is a third orientation added; center. The orientation of parties can be left and right, however when they are center or close to center they are in previous research classified as being left or right. The center category is added in order to distinguish the orientations of the parties better. The difference within the different categories has decreased and therefore these categories are more uniform. This is expected to lead to more robust and clear results, however it is not clear what these results will be. After identifying this partisan effect of elections on stock markets, these partisan effects can also be studied for the different industries. This leads to the third research question: Are different industries differently affected by elections and partisanship? To identify differences between industries, index returns of the corresponding industries are needed to calculate the CAR after an election for the different industries. The election result 9

14 data and party orientation data collected earlier are also needed to test the following hypothesis: H3: The average CAR of the different industry indices react the same to partisan effects as the country indices. The industries that are dependent on government policies and regulations are expected to have a larger magnitude of CAR, since the implications of left or right orientated parties coming into office are more severe for these industries. Industries that are affected by these government policies and regulations are expected to have a more severe reaction to election results and therefore greater CAR s. The oil and gas industry is affected by the government subsidies on energy efficient and green cars and subsidies on green energy. The health care industry is subject to government regulation concerning amongst others medicines. The telecom industry is also subject to government regulation in order to protect the consumer, while the utilities industry is highly government regulated (and often government owned) to ensure sufficient access for every consumer. The financial industry has always been regulated, however regulations have increased when the industry developed and as a consequence of the financial crisis that started in 2008, the regulations have increased even more. The effects of the financial crisis cannot be identified in the results of this research, since there is a low number of observations after the start of the crisis and the data is divided by decade not by year. Popular conceptions are that left parties are in favour of large governments, they are also thought to impose more regulations on industries. Right oriented parties are thought to be more liberal, have fewer regulations and are in favour of market forces. Regulations restrict companies in their abilities and as a consequence possibly lower their profits. Therefore industries that are subject to government regulations are expected to realise higher average CAR s after an election with a right party obtaining most votes. However it is unclear what to expect from the CAR for elections with center parties obtaining most votes. The last aspects of the effect of elections on stock markets that are examined in this study are the possible changes through the decades as a consequence of changing circumstances. Therefore the fourth and last research question is: Do the CAR results of elections change through time? To be able to identify changes over time, data about the dates of the elections has to be collected. These dates in combination with the earlier used data allows for time trends to be 10

15 identified. The corresponding hypothesis for this research question is as follows: H4: The average CAR of elections is equal over time Politics and economy have become more global and integrated over the last decades. The expected CAR is therefore expected to be lower in later decades, since integrated global economics suggest that the factors that affect the economy of a country are also more global, therefore the effect of elections should become smaller in later decades. Global politics, for example the forming of the European Union, force countries to synchronise their policies more. The consequence of this synchronizing of policies is the fact that the extremes of left and right produce sub optimal results, since they are not in confirmation with the larger unions. Therefore more central orientations are more optimal and should realize higher average CAR s in later decades. Therefore hypothesis H4 is not expected to hold. 3.4 Regression variables The effects of elections on stock markets are also being examined by means of fixed effects regressions to determine which factors are of influence when elections are held. The effect of elections is measured by calculating cumulative abnormal returns (CAR s) for every election event. These CAR s are regressed on several factors that are thought to be of influence. The most important factor that could be of influence on a CAR of an election is the orientation of the party that obtained the largest percentage of votes in the election. The three possible categories within this research are left, right and center. The party that obtained most votes is at the moment of the election results the most likely to form a government. For investors the orientation of that party is very interesting. Parties with a left orientation are expected to defend the rights of the labour force, increase the size and spending of the government and impose regulations on certain industries to protect consumers, while right oriented parties are expected to be in favour of entrepreneurs, decrease the government and its spending and let market forces regulate markets. Parties with center orientation have more moderate views on these subjects and are less extreme in their ideology. The first factor is a measure of surprise. This surprise variable measures the volatility relative to an estimation window and the worldwide volatility. The effect of the surprise variable should be large, since stock prices reflect expectations and when expectations shift stock prices change. Thus when a surprise happens, stock prices should change and CAR s are realised. 11

16 When a majority is obtained by one party the CAR could be higher, since decisions can be made more quickly and the government is possibly more efficient. On the other hand, when one party obtains more than half of the votes it is very probable that the election result is no surprise and therefore the CAR could be lower, since the expectations are incorporated into the stock prices somewhere in the past before the elections. Another factor that might influence the CAR of an election is whether or not the previous government completed its term or that dissolution of government has taken place. If dissolution has taken place new elections are likely to produce shifts in the amounts of votes obtained by the different parties. These shifts increase the probability that results differ from the expectations that were formed and therefore that CAR s can be greater when dissolution has taken place. On the other hand when a government has completed its term there is less probability of change, since there it is less likely that there were major problems within the government and therefore major shifts in the percentage of votes obtained by parties are less likely to occur. If after an election another party has become the largest that has a different orientation than the previous one, investors can experience this as a good thing or as a bad thing. Since the orientation of a party in general determines the expectations of investors about the economic policies of the parties, a change in orientation of the largest party with respect to the orientation of the previous largest party indicates a change in expectations for the investors. Changes in expectations usually generate CAR s. The variable that measures this change is orientation change. Almost all governments consist of two houses, a house of people s representatives that is chosen by the population of a country and a senate. The house of representatives forms laws, while the senate is a controlling entity that has to approve of these laws before they are passed. If after an election one party controls both houses, their power is greater than when they control only one of the two. When both houses are controlled the largest party has the ability to make and pass laws, without having to negotiate in one of the houses. The party in office is able to change more and to exert its own policies as they promised in the campaign before the elections. When one party controls both houses the CAR is likely to be higher, since the party in office is possibly more efficient in implanting its policies. Some elections have just two relevant parties participating in them. A good example of such an election are the United States, over 90 per cent of the votes in the United States are usually 12

17 obtained by just two parties. The effect of just two relevant parties within an election should be that the CAR is lower, since there is less surprise possible in the election result. The percentage of votes the largest party has obtained could also be of influence on the CAR. The greater the percentage is that the largest party obtained the more strength the party has to implement their plans and policies. A stronger party could have more influence on the CAR since the result of the election is clearer. The same arguments hold for the effects of the margin between the largest and the second largest party. When this margin increases the election result is clearer and the largest party is supposed to be stronger, than with a smaller margin. The above mentioned factors are all thought to have an influence on the CAR corresponding to an election. However the effects of these factors could change across time, due to the change in circumstances in which the elections are held. Think of changes in the political climate due to the end of the cold war, increasing wealth and technological progress. These factors contribute to a continuously changing society and therefore continuously changing circumstances through time. Therefore decades could also have an influence on the CAR and these are also considered in this study. Some studies have shown that stock markets are subject to some cycles or effects that occur at specific times. Think of higher index levels with no particular reason at the end of the year, or periods when companies produce their financial figures of the previous year. The effects of these specific periods could also possibly influence the CAR of an election, therefore months are also considered to be a factor of influence in this study. 13

18 4. Methodology The first step in performing this research is generating the data sample. The data is organised as panel data. The panel consists of the 34 member countries of the OECD. For every country the daily stock market data since 1974 is collected, if available in Thomson Reuters Datastream 3. For a worldwide index the same information is collected. Data for elections in these 34 countries within the timespan 1975 to 2011 is collected from the Parline database 4. This data is used to generate cumulative abnormal returns (CAR s) which in turn are needed in the event study and the subsequent regressions. The data needed for this research is obtained from several resources, like Thomson Reuters Datastream and the Parline database. Below the different variables that are obtained are described. 4.1 Dependent variables First of all the dependent variable of this research is collected. This is the variable that is tested in the event study and the regressions that are performed. Cumulative abnormal return: This variable measures the cumulative abnormal returns (CAR) of an election with respect to the expected returns (normal returns) over a period of time. The daily data for the calculation of the CAR is obtained from Thomson Reuters Datastream. For the actual country returns the Datastream Country Index (DCI) is used. This DCI is constructed out of minimal 80 per cent of the market capitalization of that country. 5 The DCI is value weighted. This index is used since it is available for every country and is computed in the same way for every country. The worldwide index used is the Datastream Worldwide Index (DWI). This DWI index is calculated in the same way as the DCI s and therefore ideal for calculating normal returns. These CAR s are also calculated for ten different industries. These industries are defined by the Industry Classification Benchmark (ICB) jointly created by FTSE and Dow Jones. 6 The sectors identified by this classification are: Oil & Gas, Basic Materials, Industrials, Consumer Goods, Health Care, Consumer Services, Telecom, Utilities, Financials and Technology. 3 Thomson Reuters Datastream is a database of financial figures maintained by Thomson Reuters. 4 The Parline database is a database of elections maintained by the Inter-Parliamentary Union (IPU) 5 The specifications of the Datastream indices can be found in the Datastrean Global Equity Manual page 8. 6 The specification about how the ICB is used can be found in the Datastream Global Equity Manual page 3. The complete ICB breakdown can be found on pages 45 till 54. In this research the level 2 ICB categories are used. 14

19 These different industries have their own Datastream country indices, computed in the same way as the DCI and the DWI, if the respective industry is represented on the stock market of a country. 4.2 Independent variables These variables are factors that could influence the CAR. These variables are the independent variables that are used in the event study and regressions. Election result: This variable represents the political orientation of the largest party after the election. This party is not by definition the party that comes into government after the election. The Parline database is checked to see which party obtained the largest percentage of votes in the election. Then the DPI is checked to obtain the political orientation of that particular party. This variable can take on three different results: left, center and right. For the regressions this variable is turned into two dummy variables; Dleft and Dright. Dleft is one if the Election result is left and zero otherwise and Dright is one if Election result is right and zero otherwise. In these regressions the base state of the regression is a center election result. Surprise: This variable measures to what extent the election result was a surprise. In order to construct this variable the daily volatility during the event window of the election, two days before till five days after, is measured. This is then compared to an expected daily volatility. This expected daily volatility is constructed over the period of one month, this month ends one week before the election, so that the election does not influence this expected volatility too much. The daily volatility over the estimation period is measured for the country index returns, which is represented by as well as for the worldwide index return, which is represented by. The daily country volatility is then divided by the daily worldwide volatility to measure the coefficient of the country volatility to the worldwide volatility. Thus: where stands for the coefficient of the country volatility for country i at time t. Then the expected volatility is constructed by multiplying the sensitivity of the country to the daily worldwide volatility over the event window; where represents the actual daily worldwide volatility during the event window of the election. Finally to get a measure of surprise, the actual daily country volatility, which is represented by is divided by the expected daily volatility. Thus: 15

20 Majority (dummy): This variable takes the value of one if the result of an election states that one party has 50 per cent or more of the total votes. The total vote percentages are obtained from the Parline database, constructed by the Inter-Parliamentarian Union (IPU). The IPU is an international union of parliaments. The IPU maintains the Parline database, this database contains election data about elections held in the countries that are members of the IPU. Complete term (dummy): This variable takes the value of one if the preceding government has completed the entire term as planned. When there has been a premature dissolution the dummy has the value of zero. The information about completing the term is obtained from the Parline database. Orientation change (dummy): This variable measures if there is a change in the political orientation of the party that has obtained most votes during the election, for example if a right orientated party was the largest party and is succeeded by a left or center orientated party as the largest party. When there is a change in party orientation this dummy variable is one, otherwise it is zero. The information about the orientation of the different parties is obtained from the Database of Political Institutions (DPI) 7 constructed by the Worldbank. The classification is based solely upon the economic ideology they describe in their party programs. For instance when a party describes itself as a socialist party they are classified as left oriented, when a party describes itself as conservative they are classified as right oriented. Divided government (dummy): This variable is parallel to the all houses variable used in the DPI. Depending on the political system a country has, this variable takes on the value of one if the party controlling the parliament also controls the senate. These are the houses that have law making powers. When the senate is appointed it is considered to be controlled by the party that controls the House of Representatives, however when the senate is made up along the lines of ethnic or tribal representation it is not considered to be in control of the executive. The only change made to this variable to adapt it to this study is the fact that the party in charge is not considered, but the largest party after the election is considered, since at the moment of the election result this party is the most likely to become the next party in charge. Two parties (dummy): This variable measures if there are two or more relevant parties within the election. An election with two relevant parties is defined by the top two parties. If these top two parties have obtained 80 per cent of the total votes or more, then the value of this 7 The Database of Political Institutions is constructed by the Worldbank. It was first constructed in 1995 and is updated every 5 years. The database can be found on the Worldbank internet site; 16

21 variable is one, otherwise it is zero. The information about vote percentages is obtained from the Parline database. Largest party: This variable measures the percentage of votes obtained by the largest party after the election. Information about this vote percentage is obtained from the Parline database. Margin: This variable measures the difference between the largest party after the election and the second largest party after the election. This is obtained by extracting the vote percentage of the second largest party of that of the largest party. This information can be found in the Parline database. Orientation changedleft (dummy): This variable is constructed by multiplying the variable Orientation change with the variable Dleft. This dummy has the value one if and only if there is a change in the orientation of the largest party and this change was from right or center to left. Orientation changedright (dummy): This variable is constructed in the same way as Orientation changedleft. This represents a change in the orientation of the largest party to the right. Decade: This is a variable that states in which decade the elections were held. The elections that are tested are between 1975 and 2011, therefore five decades are specified; 70 s, 80 s, 90 s, 00 s and the 10 s. In order to use this variable in regressions it is recoded into dummy variables; D80, D90, D00 and D10. The dummy variables correspond to the decades in which the elections are held and the 70 s are the base state in these regressions. Information about election dates is obtained from the Parline database. Month: This is a variable that corresponds to the month the election is held. This variable can range from one to twelve corresponding to the months in a year. These values are generated by Stata, using the date the election is held. This data is obtained from the Parline database. This variable is recoded into dummy variables for regression purposes. Djanuary, Dfebruary etc. are used, corresponding to the values of Month. 17

22 4.3 Event study An event study generates and studies the effect of specific events. The events in this case are elections within OECD countries between 1975 and The effects of these election events can be measured by abnormal returns (AR) and cumulative abnormal returns (CAR). In order to generate CAR s first AR s are needed. AR s are specified as returns above or below the expected return. Expected return is in general called normal return (NR). The general formula to obtain AR s is specified as follows:, where i is the country identifier and t stands for time in days. Therefore NR i,t is the normal returns for country i on date t. Calculating CAR s out of these AR s is simply adding the AR s for the specified time window around the event;, where eventdate(j) gives the date for event j. In this case the event window is from two days before the event, represented by till five days after the event. This window can be varied. The AR s needed to calculate the CAR s can be calculated using several methods. The first step in every method is generating the returns out of the daily index levels. This is done by a simple return formula: This is done for both the country indices as well as for the worldwide index. One of the methods to calculate AR s is simply subtracting the worldwide return from the country return;, where DWI t is the Datastream worldwide index on date t. In this case the worldwide index acts as a benchmark for the country indices and is specified as the normal return for these country indices. This method is however not very sophisticated, since country specific characteristics that depress or stimulate daily returns are not included. In order to try to include these effects into the NR s the Capital Asset Pricing Model (CAPM) formula for calculating asset returns is adapted to be used to specify NR s. The CAPM formula uses a to correct for the sensitivity to the market risk premium;, where R i,t represent the expected returns for asset i at time t, Rf t stands for the risk free rate at time t, represents the sensitivity of asset i to the market risk premium at time t which is represented by Rm t Rf t. When this is converted for use on index levels the Rm t is replaced by DWI t and R i,t no longer stands for the expected asset return for asset i on time t, but represents the expected return for country i on time t. The formula now looks like this: 18

23 The country specific is estimated using regressions. Regressing ( ) on for every single election event estimates a for every election. The estimation period for this is based on 250 trading days, which is on average the amount of trading days in a year. The estimation window starts at 260 trading days before the election event and stops ten trading days before the election event. This way estimates the sensitivity of the country index to the worldwide index over the past year without letting the election event influence. This should generate more accurate expected returns when there is no election and therefore provide more accurate AR s caused by the election event. The model used to generate expected country returns is tested to check whether it holds in its new form. The method to check whether the model holds is the same as can be used to check whether the CAPM holds. When regressing ( ) on a is obtained, a constant, and an error term. The model holds when the constant is not significantly different from zero and the expected value of the error term is also zero. Testing the model generates an average of 0,936 (with a standard deviation of 0,233) which is a value that is reasonable, since for the complete market should be one and the OECD represents a large portion of the total market. The constant has an average value of 0,018 with a standard deviation of 0,114. This generates a T-value of The chance of the constant being different from zero is therefore 3,5 per cent. The average value of the error term is 0,027 with a standard deviation of 0,101. This generates a T-value of 0,267. The chance of the error term being different from zero is about 3,8 per cent. Taking these tests into consideration the model is considered to hold for this application. The factors of interest in this event study are whether or not the election outcome with respect to orientation (left, center or right) influences the results, when the effects take place, and whether or not these effects change over months or decades. The event window is specified as two days before the event till five days after. This period of time is chosen since it should incorporate most of the effects of the election event. The two days before the event are expected to incorporate information leakage, which are polls and expectations about the election result, while the five days after should incorporate the actual election result. Five days after the election are used since in some countries it can take some time to count the votes and announce the results. 19

24 The factors of interest are represented by hypotheses, these hypotheses are tested by means of this event study. The first hypothesis tested is: H1: The average CAR of elections is different from zero. To test this hypothesis we calculate the CAR s of all the election observations and we take an average. This average over all the CAR observations is then tested using a T-test. This T-test determines if the average CAR is different from zero and with what amount of certainty it can be concluded that the average CAR is different from zero. The second set of hypotheses (H2a, H2b and H2c) concerns the effects of partisanship and tests whether or not the CAR s of elections with parties of different orientations obtaining the most votes are different from each other. In order to test if the elections with largest parties of different orientations are different a T-test is used again. This T-test then determines whether or not the different CAR s are different from each other and with what amount of certainty it can be concluded. The third hypothesis identifies the effects of elections and partisanship on the ten different industries identified by the ICB. To test this hypothesis the CAR s for the ten industries are broken down by election result. For every industry is tested whether or not the different orientations realise significantly different CAR s by T-tests. If the order of the CAR s by orientation is different from the country industry and the magnitudes differ as well, the hypothesis does not hold. If the hypothesis fails it suggests that other factors influence the industry indices with respect to the country indices. The expectation is that the hypothesis does not hold, since regulations and policies are expected to influence the expectations investors have about the future of specific industries. The fourth hypothesis to be tested concerns the changes that might have taken place over the decades. Hypothesis H4 tests if the average CAR s that are realised by elections in the seventies are equal to those of latter decades. Those latter decades are tested against each other also to determine when changes have occurred, if there are any. To test these differences T-tests are used again. 4.4 Regressions The event study provides a first glance into the effects of election events on stock market returns. However the event study does neither identify variables nor quantify their effect on CAR s. In order to get an idea about which factors are of influence for the CAR s and what 20