The effect of elections on stock market performance
Master’s Thesis
Floris Johannes Gerardus Busscher
November 2014
Thesis Supervisor: Prof. Dr. E.C. Perotti
MSc Business Economic, Finance Track
Abstract
This paper re-examines the existence of the Political Business Cycle (PBC) for 15 OECD countries using stock market performance as an output variable of political manipulation and tests the direct weekly market responses to elections employing an event study. The results indicate the existence of the PBC as years prior to elections outperform years after elections. The model did not confirm the hypothesis that a cycle in fiscal balances plays an explanatory role in this mechanism. Weak statistical evidence was found investigating the behaviour of markets shortly around elections. Overall, markets seem to underperform during the election period as compared to the estimation period. Additionally, markets did not show significant responses to surprise elections and did not respond to favourability of election outcomes as anticipated. Further research is called on to resolve the controversial empirical evidence on the PBC and the statistical difficulties of political event studies.
Keywords: Political business cycle; Fiscal budget cycle; Elections; Uncertainty; Index returns;
Surprises; Election favourability
2
Statement of Originality
This document is written by Student Floris Busscher 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
Table of Content
I. Introduction ... 4
II. Literature ... 7
i. Political Business Cycle ... 7
ii. Partisan Theory ... 9
iii. Stock Markets ... 10
iv. Political Budget Cycle ... 11
v. Trading around election dates ... 13
vi. Surprise outcomes ... 14
vii. Favourability and Incumbent results ... 15
III. Data ... 16
i. Political Budget Cycle ... 16
ii. Event Study ... 21
IV. Methodology ... 25
i. Political Business Cycle ... 25
ii. Assumption tests ... 27
iii. Outliers ... 29
iv. Event study ... 30
v. Differences in subsamples ... 32
V. Results ... 34
i. Political Business Cycle ... 34
ii. Controlling for fiscal balances and economic performance ... 35
iii. Diminishing effect of the PBC through the sample period ... 38
iv. Robustness checks ... 39
v. Event Study Results ... 40
vi. Abnormal Return Election Day ... 42
vii. Election surprises and incumbent performance ... 42
viii. Election surprises and favourability ... 44
ix. Election surprises based on favourability ... 46
x. Election Day surprises and favourability ... 48
VI. Limitations and Future Research ... 49
VII. Conclusion ... 52
VIII. References ... 53
4
I.
Introduction
The impact of political institutions on economic prosperity is often highlighted in political debate and academic research. General consensus is that political uncertainty is undesirable for entrepreneurs and corporations. For example, the uncertainty related to frequent political changes in legislation, taxes and government expenditure would be detrimental to business (Przeworski & Limongi, 1993). A direct example of this is the research of Alesina and Perotti (1996 ), who found a broad measure of socio-political stability to directly influence domestic investments. The importance of stability is thus highlighted for economic affairs. However, even for the most stable democratic nations, the take-over of government position forced by elections remains an inevitable source of change and potential uncertainty.
For politicians, this transition is similarly inevitable. It can be assumed that most politicians, having run for office voluntarily, aim to be re-elected. While in office, incumbents have several policy instruments such as fiscal and monetary policy at their disposal which they can use to affect general economic conditions, to increase popularity amongst voters. This effect leads to the Political Business Cycle (PBC), which was proposed as a theoretical framework by Nordhaus (1975) and predicts a pre-electoral boost and a post-pre-electoral drop for several economic variables.
Additionally, the uncertain information hypothesis as developed by Brown et al. (1999), predicts markets to respond to the resolution of uncertainty. Since elections presumably carry valuable information for stockholders, markets potentially show various responses to elections. This can be both after elections, when the outcome can be responded to, or before the election, as the outcome often becomes increasingly predictable in the period preceding the electoral event.
Thus, multiple mechanisms exist that predict stock markets to be electorally affected. Therefore, the main research question of this paper becomes: Do elections impact long and short term stock
market performance?
In order to answer the research question, the following set of hypotheses is developed:
Hypothesis I: Markets perform better in the years prior to elections than after elections due to electoral alteration in fiscal policy
Hypothesis II: The effects of the PBC are more dominantly present in an earlier sample period
Hypothesis III: A positive abnormal short-term return exists in the weeks prior to elections
Hypothesis IV: Unexpected election outcomes will result in positive abnormal returns in the days after elections
5 Hypothesis V: Markets respond positively to favourability of election outcome and incumbent performance
Hypothesis VI: Markets respond more substantial to favourability and incumbent performance when the election outcome is a surprise.
After the initial introduction, the PBC has been extensively expanded and empirically tested using a wide variety of output variables, including GDP growth (Alesina, 1989), unemployment (Alesina et al., 1992) fiscal balances (Shi & Svensson, 2006), monetary fluctuations (Alesina et al., 1992) and stock market performance (Huang, 1985). Particularly, research on political cycles in the stock market remains limited and unresolved. Most of the research is focused on the United States (U.S.) or focuses on the difference between developed and developing nations. It is the aim of this paper to further verify the existence of a PBC in the stock market, by testing through a unique sample of 15 OECD countries with a forty year sample period of 1971-2011.
Current research shows that a cycle in fiscal balances exists for a wide panel of nations, but more prominently for lesser developed nations (Shi & Svensson, 2002, 2006). This seems to indicate that the developed nations tend to have more efficient policies and institutions that are more difficult to manipulate by incumbents. Following this line of reasoning, it is expected that in earlier years in the sample, the effects of the PBC are more dominantly present than in later year, since policies appears become more efficient when nations develop. Hypothesis II is used in order to test this rationale.
In addition, this paper aims to establish more insights in the causality behind existing cycles in stock markets. Evidence on the mechanics behind the PBC remain largely unexplained in literature. Due to the impact and manipulability of the measure, government fiscal balances will be used in order to further explain the cycle in stock markets. These conditions are hypothesised and tested through hypothesis I.
Elections are of relevance to shareholders in multiple ways. The preference of markets for different political parties has been analysed extensively through the rational partisan theory (Alesina, 1987), which gives mixed evidence on the different effects of Democratic and Republican administrations on stock markets. This paper tests for the party-independent general outlook for stocks in the years around elections. Elections are thus the uncertain beginning of years with favourable or unfavourable political leadership, and the theorised historical economic over- or underperformance.
Considering short term impact, the influence of the political cycle and partisan theory appears to result in higher stock market volatility around elections (Bialkowski et al., 2008). Investors appear to
6 be compensated for this by a price discount on the index, which dissolves in the weeks prior to elections, when uncertainty diminishes (Pantzalis et al., 2000). This paper aims to verify these findings by an event study around the elections in this sample. In addition, the weeks after elections will be analysed to see how markets behave and recover from different electoral effects. Hypothesis III presents the proposed existence of the abnormal return
Stock markets can generally be regarded as representing the expectations of different outcome. When these events take place, expectations become realities and market responses are expected. Therefore, this paper tests for the responses of markets for different levels of incumbent performance and election favourability. This is proposed through hypothesis IV.
However, it might seem difficult to find evidence for abrupt responses to electoral outcomes, as increasing availability of information is absorbed in stock prices prior to the election and markets reflect expected outcomes. Previous research has tackled this problem by analysing the impact of a surprise effect (Garfinkel et al., 1999). This paper conducts a surprise variable and examines general market response to surprises through hypothesis V. More importantly, the paper studies market behaviour to different variables around elections when the electoral outcome is a surprise. As markets did not account for correct information if elections are surprising, meaningful differences in response might be found. This reasoning is tested by the use of hypothesis VI.
The study contributes to the current field by supplementing the ambiguous evidence on the PBC through the use of a large and current sample. Through this sample allows for an actualisation of current evidence and a statistical comparison between time periods, investigating the diminishment of the PBC across time. After Huang (1985), who compared the evolving PBC for the late eighteenth and early nineteenth century, a similar comparison has not been conducted for electoral stock cycles.
More importantly, the paper provides a model which tests the role of fiscal balances in explaining the PBC. Although both the stock cycle and budget cycle have been examined, to the best of the author’s knowledge, no current research tests for electoral stock cycles while controlling for the effects of the budget cycle. Even though the literature provides theoretical and empirical evidence of fiscal effects on stock markets and both of these cycles have been shown in previous literature, the relation between these cycles is left unquestioned. Thus this paper enhances research on the PBC by analysing the impact of the budget cycle on political stock cycles.
The event study conducted in this paper supplements existing literature by the inclusion of the surprise variable. The results of previous papers were often ambiguous due to weak (Pantzalis et al., 2000) or insignificant (Bialkowski et al., 2008) statistical evidence. A surprise variable, as Garfinkel et
7 al. (1999) used for currency rate forecast errors, is included to categorise elections. The author expects that surprise elections will provide more evidence of the effects of certain predicting variables, such as favourability and incumbent performance, as also used in Pantzalis et al., as for surprise elections information was not efficiently incorporated in stock prices and thus more abrupt responses occur in the time period after elections. Additionally, the author expects that the hypothesis of uncertainty resolution prior to elections is more dominantly present for non-surprise elections. As non-surprises are often characterised by more accurate predictions of the outcome, more uncertainty resolution takes places ex-ante to elections than for ambiguous or surprise elections and the abnormal return should be more visible. Thus, this paper aims to verify previously unsupported hypotheses, explores new measures of return predicting variables and enhances research through the use of a surprise element. This has not been previously applied to political stock market event studies.
The findings of the paper will be of high relevance to investors, who observe both the direct impact of elections and the general performance in the electoral cycle. Policymakers and supervisory bodies witness the manipulability of instruments by electorally motivated politicians.
The remainder of this thesis is structured as follows. Section II provides an overview of current research in the field and provides theoretical motivation behind the hypotheses tested. Section III elaborately discusses the data sources, the data collection and the conduction of multiple variables. Section IV presents the methods that are used in order to test the hypotheses and discusses statistical considerations. Section V presents and discusses the results that were obtained from the analysis. Section VI discusses the limitations of the analysis and potential for future research, with section VII summarising and concluding the paper. Finally, a list of references used and the appendices are provided in Section VIII and IX respectively.
II.
Literature
i. Political Business Cycle
Plentiful research has analysed electoral cycles in equity markets. Niederhoffer et al. (1970) have examined stock markets in years around national elections. They found markets to perform significantly better in the last years of the political cycle. These results confirm the rationale behind the first hypothesis. This is confirmed by Allvine & O’Neill (1980), who similarly find election-timing based differences on a trimestral basis, with the quarters prior to elections outperforming. In this early research, a correlation was established between electoral tenure time and stock market
8 performance. The research however failed to theoretically explain the occurrence of this phenomenon. This gap was filled by the PBC.
The original model of Nordhaus (1975) has been heavily expanded throughout the years, but is still commonly referred to as the foundation of the PBC. Nordhaus proposed a theory of the Opportunistic Business Cycle, which argues that politician deliberately manipulate timing of certain policies to create a business cycle that matches the cycle boom with the timing of elections, to optimally capture the content of voters. The theory based itself on several assumptions and an overly simplified model to represent the electoral incentives that cause politicians’ biases. The model of the individual voter function is presented below:
𝑉𝑡𝑖 = ∅𝑖(𝑧𝑡, 𝑧̂𝑡) = 1 𝑖𝑓 𝑈 𝑖(𝑧 𝑡) 𝑈𝑖(𝑧̂ 𝑡) > 1 0 𝑖𝑓𝑈 𝑖(𝑧 𝑡) 𝑈𝑖(𝑧̂ 𝑡) = 1 −1 𝑖𝑓𝑈 𝑖(𝑧 𝑡) 𝑈𝑖(𝑧̂ 𝑡) < 1
With the aggregate voting function being:
𝑉𝑡= 𝑉(𝑧𝑡, 𝑧̂𝑡) = ∑ 𝑉𝑡𝑖 𝑛 𝑖=1 = ∑ ∅𝑖(𝑧 𝑡, 𝑧̂𝑡) 𝑛 𝑖=1
Where 𝑧𝑡 can be read as a measure of past incumbent performance, measured by matching a rational individual’s preference with a set of economic variables1. 𝑧̂𝑡 is the average performance of
the different electoral options that voters expect in the foreseeable future. Voters then predict future performance of governments based on their past behaviour and select their optimal utility option, which equals a re-election of incumbents (1) when incumbents past performance has been above their average expectation for the coming term.
Political parties are assumed to be interested in winning upcoming elections and thus aim to influence the voters utility satisfaction. Additionally, it is assumed they fully know the preferences of their voters. This leads to belief that they adjust their policies to Max 𝑉(𝑧𝑡, 𝑧̂𝑡) with respect to 𝑧𝑡.
The model heavily depends on the assumption of the short-sightedness of the utility function of voters. Shortly after Nordhaus, this was analysed by MacRae (1977), who found the myopic
1 These utility variables include inflation, unemployment and alternative economic variables such as GDP
9 hypothesis to hold for a U.S. sample period. This empirically confirmed the theoretical hypothesis of the short-term memory of the voters utility function.
The theoretical model of Nordhaus has been generously re-examined and revisited after its proposition. Although general consensus exists about the influence of economic variables on elections (Drazan, 2000) , a heated debate still rages on the reversed effect of political institutions on economic instruments and output variables. This debate is complicated, also due to many different variables which could reflect the existence of the PBC.
In the years after Nordhaus’ publication, the PBC was frequently examined and rejected. McCallum (1978), Golden and Poterba (1980), Beck (1982 & 1984), Hibbs (1977 & 1989) and Chapell and Keech (1986) all empirically reject the existence of the PBC, using outcome variables such as unemployment, inflation, GDP growth and political instruments such as monetary and fiscal balances.
A possible explanation why evidence is highly dissimilar and adverse is given by Schultz (1995). He portrays a model which controls for re-election uncertainty, i.e. the risk an incumbent party faces to be re-elected. The rationale is that incumbents who are certain of re-election are not likely to engage in policy adjustments, due to the negative long-term inefficiencies and ideological preferences. Schultz (1995) finds that controlling for re-election uncertainty shows more evidence of the PBC and addresses the problems that hold for many papers, in which re-election uncertainty biases and suppresses the true effect of elections on policy adjustments.
There is evidence indicating the existence of the PBC. Alesina et al. (1992) have tested for evidence of the opportunistic PBC in 16 OECD countries. They found no pre-electoral effects on GDP growth and unemployment. However, they did find evidence of expansionary monetary policy during election years. In addition, they observed a political budget cycle, i.e. loose fiscal policy prior to election. Inflation exhibited a post-electoral jump, which is likely related to the monetary and fiscal boosts prior to elections.
ii. Partisan Theory
Many authors instead aim to explain political cycles by the Partisan Theory, first developed by Hibbs (1977). This theory argues that socialist parties in Europe and the Democratic party in the U.S. are more likely to opt for low unemployment while allowing higher inflation, whereas Republicans and right-wing parties have a stronger focus on limiting inflation, resulting in a looser monetary policy during Democratic administration. MacRae (1977) found early evidence of the difference in inflation and unemployment through presidential cycles in the US, indicating the impact of
10 incumbents on these measures. The ideological cycle is also reflected through the appreciation of the dollar during Republican administration and depreciation during Democratic administration (Beck, 1984). Alesina (1987) extended on the Partisan Theory by proposing the rational partisan theory (RPT). In this it assumes that parties aim for ideological goals, only in times when they are confident about re-election and thus controls the general theory for re-election risk, as Schulz (1995) controlled the PBC. However, after proposing the theory, Alesina (1991) and Alesina et al. (1992) have empirically tested and rejected the strict RPT. Some evidence does remains of a general partisan theory, as shown by Hibbs (1992).
More partisan-evidence was found in Santa-Clara and Valkanov (2003). In an influential paper, they found stock markets to show a significant higher stock market return of at least nine percent for Democratic administrations, triggered by excess stock returns and lower real interest rates, which may partly confirm the proposed looser monetary policy. This difference could not be explained by accounting for different risk exposure. Thus, it appears Democratic administrations in the US are ultimately favourable for stock markets. This was recently confirmed by Belo et al. (2013), who further investigated the presidential puzzle that was proposed by Santa-Clara and Valkanov. They found that especially firms with high government exposure outperformed low-exposure firms with 6.1 percent during Democratic presidencies, while underperforming with -4.8 percent during Republican administrations. They generate investment strategies, based on the partisan cycle, that generate 6.9 percent annual abnormal returns.
The Partisan Theory is often used as a substitute to explain shortcomings of the PBC, although the theories are not mutually exclusive. However, both the rejection of the PBC and the support for the Partisan Theory is generally old research and is focused on the U.S. In addition, there is relatively recent research in support of the PBC. The Partisan Theory can be regarded as a highly relevant supplementary theory to the framework of this paper and useful in gaining insights as to the importance of elections to stock markets, which is a foundation for the event study conducted later in this paper.
iii. Stock Markets
Ample research was directed on different economic variables affecting the previously described voters function for incumbent re-election. Stock market performance is not one of the voter’s utility determinants as described in the original model of Nordhaus (1975). However, much research indicates the existence of an electoral cycle in domestic stock markets (Niederhoffer et al., 1970; Allvine & O’ Neill, 1980; Huang, 1985)
11 Although evidence is mixed, multiple recent indications exist of the presence of the PBC in stock markets. Problematically, the fast majority of the research is U.S. focused, where samples with more western nations show incongruities with U.S. data. This paper will enrich the current debate with a revision of the PBC in stock markets for a sample of 15 OECD countries over a sample period of 40 years, using a better proxy for stock market returns. This proxy, as will be discussed in the data section, better represents domestic business as it includes the fast majority of tradable mid-cap securities, rather than larger companies included in a typical index.
Furthermore, Huang (1985) analysed the PBC in stock returns for multiple periods in a U.S. sample and found significant differences between different periods. Indications exists in alternative research that the PBC, for multiple variables, is more dominantly present for developing nations than western countries (Svi & Svensson, 2002; Block et al., 2001) These findings lead to believe that the level of development or prosperity of a nation might increase efficiency of policy implementation. Accordingly, the difference between the first two decades and last two decades in the sample will be tested, to see if manipulability of policies is indeed declining when nations further develop.
iv. Political Budget Cycle
There are different sides as to how political cycles might impact stock markets. Döpke and Pierdzioch (2004) argue that stock market performance would qualify as a plausible candidate for testing the PBC due to its high transparency, since information on stock markets is more up-to-date and more widely available than any economic variable. Thus generating great influence on inattentive voters whose information is incomplete. Additionally, Gärtner and Wellershoff (1999) argue that stock markets are in general thought of as a leading indicator for real economic activity. Therefore, it would be in the interest of incumbents to stimulate stock market performance, influencing voter’s perception on economic performance. These arguments of economic reasoning would qualify stock market data as an output variable candidate for the PBC.
Another more technical argument can be made for spill over effects. Many of the outcome variables and instruments that are potentially sensitive to electoral cycles are of relevance to stock markets. Much debate in the field exists on the relationship between economic variables such as GDP growth and unemployment and stock market returns. Whereas there are papers indicating forms of (Granger) causality between these economic factors and stock markets (e.g. Wongbangpo & Sharma, 2002), the general consensus is that the relation between stock market and general economic conditions is relatively independent and unpredictable. However, the instruments that politicians can use to manipulate the voters function, i.e. monetary and fiscal policy, do give broad evidence of direct impact on stock markets. Considerable research shows monetary policy to directly
12 impact stock markets, both as a systemic effect (Bernanke & Kuttner, 2004) and as a corresponding direct response to announcements (e.g. Thorbecke, 1997). For fiscal policy, Belo et al. (2013) explain through the Partisan Theory how highly exposed firms show abnormal returns as large as 6.9% in times of high government expenditure, benefiting general market performance.
After reviewing the research on the PBC, it appears that unemployment and GDP growth, as outcome variables, are more difficult for politicians to manipulate than the instruments at their direct disposal (Howard & Chen, 2007). Since this research aims to increase understanding of the causality mechanisms of the PBC on stock markets, it is intuitive to investigate one of the instruments available to governments in the PBC, selecting from monetary and fiscal policy.
Even with the partial rejection of the political business cycle, there is convincing evidence of existence of a political budget cycle (Tufte, 1978; Golden & Poterba, 1980; Shi & Svensson, 2002, 2006), which is the increase in governments net expenditure prior to elections. Fiscal policy measures the extent to which governments adjust their fiscal balances, by adjusting spending or taxes, to impact the domestic economy or achieve administrative objectives. Although the alterations in fiscal policy are likely intended to affect economic utility conditions of voters, there is evidence for the policy’s (side)effects on stock markets, as explained by Belo et al. (2013). They found their previously described 6.9 percent annual abnormal return to be explained by the increased government spending, triggered by partisan cycles. Moreover, intuitively it seems conceivable to assume a relation between government expenditure and stock market performance. In the U.S., e.g., government spending accounts for 20% of GDP and appears a relevant factor.
Institutionally, monetary policy is controlled by central banks, who are assumed to operate independently. This is investigated by Cukierman et al. (1992) who measures independence of different central banks through legislative freedom and an index of international political experts. Although confirming independence for many countries using his indexation, he finds independence to be inversely correlated with inflation for industrial nations. Independence of central banks is thus a complex assumption, which has been neglected in most of the previous research in this field.
In this paper, the instrument of analysis is fiscal balances. The assumptions that would be required for the use of monetary policy are debatable and evidence of fiscal cycles and its proven effect on stock markets qualifies fiscal balances as a variable of interest when developing a model to gain further insights in the mechanisms behind the PBC. Results obtained by Belo et al. (2013), who found the partisan cycle to be explained by fiscal policy are similarly predicted in this research. To the best of the author’s knowledge, current literature does not provide a model in which electoral cycles are tested whilst controlling for the budget cycle. This paper hopes to further explore the existence
13 and interaction of the business and budget cycle, while analysing the relationship between election timing, fiscal budgets and stock markets.
v. Trading around election dates
There have been many academic efforts linking political situations to stock market performance. Asteriou (2000) analysed the effect of political instability using measures of political violence and found this specific form of instability to be detrimental to stock market performance. Closer to this research are Bialkowski et al. (2008), who examined the effect of upcoming elections on stock market volatility. They found markets to be more volatile around election dates, although their limited tests for compensating returns did not yield significant results. A further explanation of this is given by the empirical contributions of Boutchkova et al. (2012). They found domestic political and electoral risk to affect industry risk volatility, with industries being more dependent on trade, labour and contract enforcement being more affected by political risk.
Also closely connected to this paper is Mattozzi (2006), who showed political polls of elections to be directly linked to a portfolio of companies contributing to the campaign of the candidates. Using these portfolios as representative for stocks bearing highest political risk, they showed a significant positive relation between the portfolio returns and the polls resulting in daily linear shifts laterally with the opinion polls. Their results indicated that various industries carry direct political risk around elections. Combining this with overall preferences of the markets, economically justified by the partisan theory or business cycle argumentation, this indicates that the whole market bears electoral risk prior to elections.
Another example study of this is Gemmill (1992), who studied the relationship between election polls and the FTSE option market for the 1987 election. He explained how the option prices most efficiently reflected the development of the opinion polls, with the exception that in the weeks prior to the election an inefficiency occurred. He proposed the high media coverage to attract a great deal of ill-informed speculators, offsetting an option-market bubble prior to elections. This paper indicated a possible abnormal return due to increased speculative trading prior to election and called for future research.
Another proposition of a higher stock return prior to elections is the Uncertain Information Hypothesis (UIC) With the literature showing that volatility is higher around elections, markets are increasingly exposed to political risk prior to elections. Brown et al. (1988) predicted through the UIC that the electoral risk implies a risk-premium for stocks, translating into a lower asset prices whilst carrying this risk, generating relatively higher returns as a compensation for the carrier risk.
14 In most cases, election uncertainty will slowly resolve in the weeks leading to elections, when polls become more accurate and election outcome becomes increasingly predictable. Thus, the uncertainty resolves and asset prices return to their fundamental value. These efficient economic arguments predict a positive return in the ex-ante trading period.
This theorem has been tested in previous academic research. Both Bialkowski et al. (2006) and Pantzalis et al. (2000) have tested for a positive return prior to elections. Remarkably, although using similar methodology and comparable OECD samples, results differ slightly. Whereas Bialkowski et al. (2006) found no evidence of the resolution of uncertainty, Pantzalis et al. (2000) found some limited evidence of a risk-premium2.
This paper aims to contribute to the existing literature by giving a renewed verification of the resolution of uncertainty prior to elections. It updates the actuality of the results by using a renewed sample. Moreover, a more efficient determination of event period is used, contributing to the efficiency of the results. This will be discussed in the Data section.
vi. Surprise outcomes
Stock prices are commonly thought of as reflecting current value of stocks and a weighted expectation of future developments. Under an efficient market hypothesis, stocks incorporate information efficiently and immediately. Therefore, abrupt changes corresponding to immediate changes in present and future realities can heavily impact price levels.
Whilst information if often steadily incorporated when predicted, stock markets are always subject to the element of surprise. Examples of informative surprises on price levels exists throughout the literature. Relating stock markets to news regarding policy news, Bomfim (2003) found surprising monetary policy announcements to strongly boost stock market volatility and price levels, with a bigger effect existing for positive surprises. This is further confirmed by Pearce & Roley (1985), who also found stock markets to respond to surprises in various economic news variables.
Thus, intuitively, if stock markets respond to surprises in economic and policy news, assumedly markets would respond to election surprises, as much of these policies are directly influenced by electoral outcomes. A further indication of this is the research of Garfinkel et al. (1999), who have analysed the effect of electoral surprises on analysts’ forecasts in currency exchange rates for six of the G8 countries. They found surprises to be related to unusual large and significant forecast errors, underlying the relevance of electoral information for these measures.
2 Whereas Bialkowski et al. (2008) rejected the existence of an abnormal return prior to elections completely,
15 Thus, it is economically reasonable to explore the existence of the impact of an electoral surprise element on stock markets. To analyse if markets show a skewed response to surprises in general, i.e., a general tendency towards surprises that is unexplained by favourability of outcomes, general differences are tested between predicted and surprise elections throughout the sample. To the best of the author’s knowledge, a surprise variable as used in Garfinkel et al. (1999) is currently not being applied to stock indices similarly to the analysis of this paper.
vii. Favourability and Incumbent results
After testing for the general effects of the surprise element, it is of interest to classify election outcomes to investigate stock market preference and explain differences in market responses. Applying the Partisan theory as described previously would qualify as a way in which to investigate market’s preferences for different classifications of outcomes. However, large difference occur in the sample with regard to political systems. Most countries do not accommodate a two-party system and dissimilarities between right and left wing parties may be ambiguous.
Moreover, other variables are of interest to further research. The performance of incumbents during elections has been analysed by Chuang and Liu (2013) for the first trading day ex-ante elections. Their results show a strongly significant return for incumbents winning the elections, indicating overall preference for continuity in their sample.
This paper will not use partisan measures, as described previously, as a measure to predict abnormal return around elections. Santa-Clara and Valkanov (2003) already established an event study testing the difference in market responses to Democratic and Republican wins shortly around elections. They found no significant difference for their sample period, even with the strong differences in administration’s performance that markets are expected to respond to. Therefore, this paper will not use partisan theory in order to predict short term election responses.
Pantzalis et al. (2000) have classified their sample on multiple variables. One of these variables is the favourability of elections, in which they conduct a variable measuring the likeability of election outcome to investors3. Although their results presented some global indications, they failed to
establish statistically meaningful conclusions on this variable.
This paper aims to test the differences in market responses for these two variables, incumbent performance and favourability of elections, for a renewed sample and for different event periods. An
3 Favourability is defined by a combination of the country’s economic performance set against world average
performance and the incumbent performance. A favourable event occurs when strong (weak) economic growth is combined with an incumbent win (loss) during elections. Consequently, an unfavourable event occurs when a strong (weak) economic performance is combined with an incumbent loss (win).
16 event period prior to elections will be included to investigate if markets anticipated these outcomes differently.
This paper follows the methodology of Pantzalis et al. (2000) to a large extent. As an important addition, the conducted surprise variable will be used to create a surprise effect for these variables and see if differences in responses to favourability and incumbent performance are greater when the outcome was unanticipated by markets.
The results of the event study will give broad insights in the preferences of stockholders to different electoral outcomes. Furthermore, the market responses will provide investors with insights on the political risks they are exposed to during government transitions, whereas the excess returns will give merit to investment strategies around these events. This holds true at least for risk-neutral investors, since volatility is beyond the scope of this paper.
III.
Data
i. Political Budget Cycle
A sample of many developed democracies is needed to obtain valid results which can be generalised across the population and eliminate the dominance of country specific effects. For this, the author follows the paper of Anderson & Pontusson ( 2007) who address a sample of 15 OECD countries, testing political measures on job security. The authors believe their sample differs widely with regard to economic and cultural characteristics, next to public provisions. The sample includes the following countries: Canada, Denmark, France, Germany, Italy, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom and the United States. Although the Eurozone is strongly represented, the sample included both non-European countries and European countries using a different currency. The fifteen nations represent a wide range of political and economic diversity within the pool of developed nations. The sample period stretches from 1971 till 2011. Data availability for various variables and countries becomes sparse going back earlier than 1970. More importantly, a forty year period yields hundreds of different election years so that there are over a hundred of unique political cycles represented. The period leaves the sample with sufficient actuality and relevance, due to the comparability of electoral systems across the years and the inclusion of many recent years.
The election dates are obtained through the Comparative Political Data Set 1960-2011 (Armingeon et al., 2011), a collection of political and institutional data assembled by the Swiss
17 National Science Foundation for 28 democratic nations for a variety of research purposes. These election dates are used for both testing for the PBC and conducting the event study.
Missing observations occur mostly for Spain, Portugal and Switzerland in the earliest years of the sample. For the Iberian nations this is due to the transition to democracy which the countries experienced. Before and shortly after this transition, much information is not present in the dataset and the author does not pursue the missing observations, as these are not representative for the developed OECD democracies that this paper aims to test.
For the PBC analysis, a variable is conducted named election years. The election event is taken and years are categorised around this year. The year in which the election year takes places is given the value one, representing the first year after elections. Although especially the Anglo-Saxon countries in the sample have elections in the last quarter of the year, the lag of policy implementations and the elections in earlier quarters of many other countries in the sample validate a selection of the election year as a first year in the cycle. The two years previous to election are given the value -1 and -2, with the election year and the year after elections being given 1 and 2 respectively.
Often, elections are called early when incumbent governments resign or coalitions cease to rule. In this case, the cycle can be disrupted. A political cycle is eliminated from the sample if it is shorter than three years, since the effects of policy timing is assumed to need a representative cycle in order to be impactful4.
This variable is used as the most practical way of conducting election years. In this sample, an average election would take place in the second quarter of the year, thus the election year is approximately half prior to and half post of the elections. It is however not realistic to obtain reliable fiscal deficits for a random 52 week period in this sample. In addition, many policies implemented by governments are implemented on annual basis. Even with elections lost or won, these policies are not immediately reversed after the election date. For these considerations and the arguments mentioned previously, the author will treat the election year as year 1 and the preceding year as -1.
Fiscal balance changes are required to determine the budget cycle and measure its effect on stock returns. Fiscal balances reflect the combination of tax income and government spending, together representing fiscal policy. Fiscal balance data was obtained through the OECD economic
4 Three-year election cycles are treated differently than four-year cycles. For three year cycles the -2, -1 and 1
values are assigned to the cycle years. However, when an election is called early unexpectedly the 2, -2, 1 values are assigned. It is assume that in these cycles, governments behaved as theory predicted them to until the election year, as early elections were unexpected till that year.
18 outlook No. 92 (2012) by the use of the net lending variable, which measures the difference between government expenditure and income and is commonly used to represent fiscal balances. Then, the author manually constructed an absolute change in fiscal balances for each year. Also Gross Domestic Product (GDP) growth, discussed in the coming sections as a control variable, was obtained from this database5.
Thirdly, the OECD dataset provides primary balance data. The primary balances entails the government’s revenues minus its expenditure, excluding any interest payments and debt repayments. This is a measure of the operational balance of the government. Since typically governments cannot decide on immediately cutting interest expenses, it appears a more efficient measure of immediate policy change. Thus it is expected that these primary balances are more efficiently influenced by election timing better capturing the effects of manipulated policies as predicted by the PBC theory.
Most importantly, annual stock returns are needed, to be used as dependent variable for the first and second hypothesis. The MSCI indices, rather than common stock indices, have the advantage that they cover a significantly larger field of middle-sized publicly traded companies, and thus might be better suited more adequately reflecting economic impact of the effects of the PBC6. Therefore,
the MSCI index might more adequately reflect the true impact of the PBC, which is aimed at stimulating widespread economic prosperity, than the market index, which for many countries merely captures the largest tradable securities. Consequently, the MSCI index is expected to react more efficiently to government policy changes than the market index.
5 Variable used is the real GDP growth, which corrects for inflationary changes.
6 The standard index covers all investable large and mid-capitalization securities in many countries. It includes
normally around 85% of each country’s free-float adjusted market capitalization. Additionally, for multiple countries, availability of data is higher through the MSCI standard index than stock indices. The index returns are downloaded from the MSCI index performance databank (Morgan Stanley, 2014).
19 Figure I: Overview of variable mean development through the electoral cycle
Figure I graphically represents the means of the key variables throughout the political cycles in the OECD country sample. The left vertical axis represents the average stock returns throughout the election cycle. The right vertical axis represents the nominal changes in fiscal deficit and primary balance in percentages in these years. The horizontal axis follows the developments of these averages through the political cycle, represented as years around an election date.
The graph shows that the MSCI returns are higher for both years later in the political cycle then years immediately after elections. This is a first insight in the data that reinforces the theory of the PBC However, fiscal and primary balance changes show an unanticipated trend. Positive balance changes in year -1 and negative balance changes in the election year opposes the predictions of the political budget cycle. The mean development as visualised by Figure I is thus not consistent with presumptions behind the first hypothesis.
Table I represents descriptive statistics for each country for the variables used to test for a political fiscal cycle. The statistics were obtained ex-post the removal of incomplete election cycles.
20 Ta b le I D e sc ri p ti ve s ta ti sti cs f o r al l c o u n tr ie s in P an e l A d u ri n g sa mp le p e ri o d 1 97 0-20 10 . Fi sc al D e fi ci t Sta ti sti cs P ri ma ry B al an ce S ta ti sti cs M SCI R e tu rn S ta ti sti cs El e cti o n s Fu ll P o li ti ca l Cy cl e s M e an M in imu m M ax imu m M e an M in imu m M ax imu m M e an M in imu m M ax imu m P a n el A . A ll C o u n tr ie s 179 149 -0 .0 3 -6 .9 6 7. 03 -0 .0 4 -8 .8 1 9. 13 10 .4 5 -6 5. 22 17 3. 65 P a n el B . B y C o u n tr y Ca n ad a 13 9 0. 09 -2 .2 9 2. 47 0. 10 -3 .4 0 2. 46 10 .3 6 -2 1. 41 52 .7 1 D e n ma rk 16 7 0. 37 -2 .4 3 4. 79 0. 41 -5 .7 3 5. 63 10 .9 7 -4 8. 15 65 .8 5 Fr an ce 9 8 0. 04 -2 .7 2 1. 88 -0 .0 7 -4 .7 0 1. 99 10 .8 2 -3 4. 44 78 .4 5 G e rma n y 11 11 0. 15 -6 .9 6 6. 41 0. 17 -6 .8 2 6. 12 10 .9 7 -4 7. 24 13 1. 46 Ita ly 11 8 0. 00 -2 .8 0 3. 54 -0 .0 2 -3 .3 9 2. 78 9. 21 -3 4. 89 12 7. 54 Ja p an 13 11 -0 .3 0 -5 .9 9 4. 35 -0 .2 6 -7 .0 1 3. 88 11 .6 8 -3 6. 43 12 1. 16 N e th e rl an d s 13 10 -0 .1 1 -5 .5 3 7. 03 -0 .0 9 -6 .2 4 7. 28 9. 81 -5 0. 09 54 .2 4 N e w Ze al an d 14 14 -0 .4 0 -4 .3 0 2. 43 -0 .2 9 -4 .5 1 2. 69 10 .2 1 -5 6. 21 10 9. 81 N o rw ay 10 10 0. 03 -2 .0 7 3. 05 0. 22 -7 .6 5 9. 13 15 .0 2 -6 5. 22 17 3. 65 P o rtu ga l 14 8 -0 .0 1 -4 .9 2 3. 71 -0 .1 5 -6 .4 8 2. 92 5. 87 -5 3. 65 43 .8 8 Sp ai n 11 10 -0 .0 3 -5 .8 2 3. 22 -0 .1 7 -5 .8 2 3. 22 7. 08 -4 2. 97 11 2. 77 Sw e d e n 13 12 -0 .0 9 -6 .9 7 4. 72 -0 .1 0 -8 .8 1 4. 37 15 .0 5 -5 1. 44 77 .7 6 Sw itze rl an d 11 11 0. 09 -1 .2 4 1. 00 0. 01 -1 .8 1 1. 41 11 .7 8 -3 1. 57 10 2. 48 UK 10 10 -0 .1 2 -4 .0 3 2. 46 -0 .2 7 -6 .3 9 4. 79 6. 66 -5 2. 32 50 .1 9 U SA 10 10 -0 .1 5 -3 .8 2 1. 55 -0 .1 5 -5 .8 2 2. 24 11 .3 3 -3 7. 00 37 .5 8
21 Differences between various countries are substantial. Whereas maximum yearly index returns are over hundred percent for seven of the fifteen countries, big differences exists in arithmetic means and highest single year drops. The amount of political cycles included in the analysis per country ranges from a minimum of nine from France to sixteen from Denmark. Hence, all the countries are considerably represented in the analysis.
In Appendix I, the Pearson R correlation matrix is presented for the variables used in the PBC model. Again, the extremely high correlation coefficient of primary and fiscal balances (0.8717) was predicted, as these variables are basically inter-exchangeable. The high correlation between GDP growth, which will be discussed in methodology as a primary control variable, and changes in fiscal balances (0.4599) is similarly understandable. Through a Keynesian anti-cyclical argumentation, governments increase spending, thus decrease fiscal balances, in times of low GDP growth and depression whereas in times of high GDP growth financial positions are restored. Finally, the significant negative coefficient of MSCI returns and the election year variable (-0.0909) confirms figure 1 in that returns appear in the latter years of the sample.
Problematically is the insignificance of the fiscal balance variables and the election year variable. Not only does this preliminarily question the existence of a budget cycle in the sample, but it reduces the quality of fiscal balances as a control variable in the PBC regression. This will be further discussed in the methodology section. The positive correlation between MSCI returns and fiscal balance changes is intuitive when read as correlation. In times of high stock returns, economics tend to do well and the government can restore its financial position. For a causal effect a negative coefficient would be expected. Further insights here might be gained when controlling for GDP growth, as the next section will further specify.
ii. Event Study
For the second hypothesis, a sample size has to be determined. While elections are available from the data of the first hypothesis till 1971, accurate daily stock market returns are more difficult for multiple nations. Additionally, it is the aim of this paper to verify and extent the work of Pantzalis et al. (2000), who use a sample between 1974-1995. Thus, event study sample period will range between 1990-2011, yielding 74 elections7 both parliamentary and presidential elections8.
7 Elections with remarkable circumstances, leading to the elimination of trading have been eliminated. For New
Zealand, elections before 2003 are excluded due to the unavailability of data.
8 While countries throughout the sample have different political systems, France is most distinguished. With
the two-stage presidential elections requiring the country to vote for two separate days, the author preferred to use the second round of election to be used for the analysis, since this is the final and, widely regarded, more important round of the election process.
22 Both an expected and actual return is derived in order to obtain abnormal returns. Unlike the first hypothesis, market index returns are used for weekly stock market returns. Market index returns are assumed to be significantly more liquid and tradable thus representing more efficient and immediate responses to uncertainty resolution or exposure to political risk. An overview of the indices used is given in Appendix II. All returns required for the establishment of the event and estimation period were obtained through Thomson Financial DataStream. Both for conducting expected and actual return, net returns are used which include redistributed dividend and stock restructuring events, next to the closing price returns. Weekly returns were calculated on a percent basis using the product of the five daily returns.
The expected return for the event study is derived by the use of an estimation window. For the event period of (-4,4) weeks around the election date, the returns of the (N-51) preceding trading weeks were obtained and averaged across the estimation period, in order to conduct a weekly expected return. Actual weekly returns are conducted by obtaining daily returns and cumulating the returns of five trading days to form the aggregate week return. This methodology, unlike earlier papers, allow us to efficiently estimate five trading days as the week9. Using the week of in which the
election took place as week 1 gives inefficiency problems, since for Friday elections (e.g.) this would classify four trading days prior to the election as part of the post-election trading week. Thus, it is the author’s belief that this methodology provides a more accurate estimate of the weeks around elections and potentially captures the effects under study more efficiently.
The return on election day itself is used separately, to calculate a daily abnormal return for this trading day. Expected return is the geometric daily average of the total estimation period. The 20 trading days prior to and after this day are used to obtain weekly returns for the eight week period of abnormal returns. The data is then used in the model specified in the Methodology section, to calculate a cumulative abnormal return(CAR).
Multiple countries facilitate elections on non-trading days, mainly Sundays. These include: France, Germany, Italy, Japan, New Zealand, Portugal, Spain, Sweden, Switzerland and Norway during the 1993 election. For countries organising the elections on a trading day, the trading day is being analysed separately from the week window, as discussed later in the paper10.
9 For elections taking place on Tuesday (e.g.), Monday and Tuesday till Friday of the previous week jointly form
trading week -1.
10 Thus for countries with weekend elections, the Monday after the election and the Friday before are used as
the first (last) trading day after (before) elections. For trading day elections, the previous day is used as the last day for week -1 and the day after election is used as the start of week 1 when calculating weekly returns.
23 Table II reports primary descriptive statistics of the cumulative abnormal returns for the total event period and the dispersion amongst countries in the sample.
The number of elections per country ranges from three Norwegian and New Zealand elections to seven Japanese elections included. Cumulative abnormal stock returns for the full event period range from -22.58% to 14.71% for the eight week event period throughout the sample. Ten out of fifteen countries show a general negative mean value for the CARs in the event period, which, in case of significance, is a remarkable finding regarding to the overall election period.
For the fourth hypothesis, a surprise variable is needed to classify elections . This will be done with secondary qualitative data; expert reports in newspapers commenting on the database. The New York Times is used as the primary source of these reports and these reports are cross-checked with other quality papers to determine the level of surprise. In addition, the earlier results of Garfinkel et al. (1999), who determined the level of surprise for 28 elections in 7 G8 countries, can partly be used to cross-check the output of this methodology and see whether it confirms to previous literature.
Table II
Number of elections
Mean Median Minimum Maximum
Panel A. All Countries
74 -1.52 -1.38 -22.58 14.71 Panel B. By Country Canada 6 -4.69 -4.25 -16.24 3.44 Denmark 5 2.53 -3.88 -5.66 14.71 France 4 -2.95 -2.47 -15.89 9.03 Germany 5 4.51 -2.58 -12.30 9.41 Italy 3 5.48 -2.93 -0.99 14.51 Japan 7 -0.10 -1.65 -7.56 8.53 Netherlands 6 -4.04 -3.17 -11.2 0.57 New Zealand 3 -1.85 -3.71 -3.74 1.91 Norway 5 -5.74 -5.18 -22.58 4.99 Portugal 5 -3.31 -0.38 -12.62 2.25 Spain 4 2.88 -0.38 -1.76 14.04 Sweden 6 -1.92 0.42 -12.79 8.47 Switzerland 5 -0.24 -0.32 -0.41 4.30 UK 5 0.33 1.75 -10.03 6.44 US 5 -1.04 0.56 -8.92 3.61
Descriptive statistics for CAR (-4, 4)
1990-24 Elections which are defined by phrases such as ‘unexpected outcome, ‘surprise’ or ‘polls went wrong’ will be defined as a surprise (X=S). Elections characterised by reviews including ‘overwhelming victories’ or if the losing candidate had indeed been given ‘small chance of winning’ will be defined as a no surprise(X=N). Additionally, elections which are unclear in terms of surprise and hard to determine based on the information will be determined as ambiguous (X=A).
Similar qualitative research is used to determine incumbent performance, simply defined as ‘Win’ or ‘Loss’. For presidential elections, this is a straightforward process. For complex coalition formations in parliamentary elections, result is dependent on keeping majority in parliament for governing parties, or in exceptional cases for the possibility of staying in office based on minority-coalitions.
Favourability of elections, to be used for the final hypothesis, is created by the use of Incumbent performance and Economic performance. Two proxies of economic performance are created. Firstly, a world-economic performance measure is conducted through comparing the GDP growth of the country in the election year to the MSCI world index11. The returns from the world index are
downloaded from the MSCI databank (MSCI, 2014). Secondly, an average-economic performance measure is conducted through comparing the GDP growth of election year with the average domestic GDP growth of the three years preceding election year. Whereas the previous literature made use of the world economic comparison (Pantzalis et al., 2000), the author beliefs that a historical country comparison of a few years might better represent favourability, since this is more relevant to domestic investors and better memorised by voters. GDP growth is derived from the data used for the PBC, whereas world economic performance is obtained through the MSCI world index.
For both measures, economic performance is defined as “above (below) average” when GDP growth is at least 0.5 percent larger (smaller) than the benchmark. For observations within a 0,5 percent range, economic performance is determined as average and is not used in the favourability study.
If Economic performance has been above average, an incumbent win (loss) is determined as a favourable (unfavourable) event. If economic performance is below average, an incumbent loss (win) is defined as an unfavourable (favourable) event.
11 For elections taking place in the first four months of the year, a one year lag is used, and the economic
performance of the year prior to election year is used, as a better measure of incumbents recent economic performane.
25
IV.
Methodology
i. Political Business Cycle
With the prior literature giving mixed evidence and indications of the political stock cycle, the aim of this analysis is to verify its existence for a panel of 15 developed nations over the last four decades. Methodological issues have to be addressed in order to obtain valid and reliable results. Firstly, to test for the existence of a political business cycle, a primary regression is specified through Equation I:
𝑌𝑖𝑡 = 𝛽0+ 𝛽1𝐸𝐿𝐸𝑖𝑡+ 𝜀𝑖𝑡 (I)
Whereas 𝑌𝑖𝑡 represents the yearly index returns, 𝐸𝐿𝐸𝑖𝑡 equals election timing, 𝛽0 the sample intercept and 𝜀𝑖𝑡 a (N-0, σ2 ) distributed error term. 𝐸𝐿𝐸𝑖𝑡 is a variable which equals one of four
values: -2, -1, 1, 2. Negative integers represent the years leading up to elections, with -1 equalling the year prior to election year. Year 1 and 2 equal the election year and the year following elections.
The coefficient of this regression is meaningful but not clearly interpretable, due to the categories it uses as an ordinal variable. Thus, four dummy variables are created to reflect each of the above described years. This yields the following model:
𝑌𝑖𝑡 = 𝛽1𝐷−2𝑖+ 𝛽2𝐷−1𝑖+ 𝛽3𝐷1𝑖+ 𝛽4𝐷2𝑖+ 𝜀𝑖𝑡 (II) The constant term is suppressed in order to prevent multicollinearity. The four dummy variables equal 1 for observations in their respective election year. Therefore, the interpretation of the coefficients is equal to the mean values of the individual years in the sample. The statistical significance of the dummies has no relevant meaning, since it is the significance of having an index return different than zero for that election year. An F-test will be used in order to test for statistical differences between the dummies.
Next, the political cycle is divided in two halves, with the two years before the elections serving as one half and the years after elections as the other. A dummy variable is created and the following linear regression is presented:
𝑌𝑖 = 𝛽0+ 𝛽1𝑃𝑟𝑒 − 𝐸𝑙𝑒𝑐𝑡𝐷𝑢𝑚𝑚𝑦𝑖+ 𝜀𝑖𝑡 (III) The equation III output provides insights in the general differences between the two halves of the cycle. The intercept 𝛽0 , with 𝑃𝑟𝑒 − 𝐸𝑙𝑒𝑐𝑡𝐷𝑢𝑚𝑚𝑦𝑖 = 1 for the pre-election years, can be read as the mean of the two years after election, whereas the 𝛽1 coefficient is read as the difference between
26 the two halves of the cycle. The significance of the dummy is a test for the significance of a difference in means.
In addition, it is of interest to see the development of the cycle through the past decades. Therefore, the sample is classified between two sample periods, 1971-1991 and 1991-2011. Equation IV is run for both of the subsamples in order to test for the development of the political cycle12.
After results have been established to test for the existence of the political business cycle through equation I till III, an endeavour is made to increase insight in the causality behind the PBC. In order to test for the factors involved, two control variables are included in the model. As discussed in the literature, yearly changes in fiscal balances of central governments is included. Due to the manipulability of this instrument for incumbents seeking re-elections, and its direct effect on economic conditions, this is a variable that is expected to explain the political cycle. Growth of GDP is included as an additional control variable. The motivation of this inclusion consists of the postulated aim of politicians to directly affect voters and stimulate general economic prosperity, rather than equity markets, as has been elaborated on in the literature. Additionally, it tackles the potential omission bias of economic cycles, as it is plausible that during harsh (prosperous) economic times, both fiscal balances and stock markets decrease (increase) simultaneously
The inclusion of the control variables results into the following model:
𝑌𝑖𝑡 = 𝛽0+ 𝛽1𝐸𝐿𝐸𝑖𝑡+ 𝛽2∆𝐹𝐼𝑆𝐶𝑖𝑡+ 𝛽3𝐺𝐷𝑃𝑖𝑡+ 𝜀𝑖𝑡 (IV) With ∆𝐹𝐼𝑆𝐶𝑖 and 𝐺𝐷𝑃𝑖 representing changes in government’s fiscal balances and national GDP growth respectively13. As discussed in the data, equation IV is run using two measures of fiscal
balances, adjusted fiscal balances and primary balances, in order to capture the potentially amplified predicting value of the primary balance, due to its efficient and more rapid changes.
If after the inclusion of the control variables in equation IV the β1 coefficient loses significance, the results indicate that the effect of election timing on stock indices is, at least partially, explained by either of the control variables, since controlling the other variables isolates the effect of election timing from the other measures. If the results of equation I and equation IV are similar, control variables do not appear to mediate the effects of the political cycle.
Since the observations are, inherently, uniquely identified by country and year, there are two ways of treating the data. While using OLS coefficient estimators is most often used in linear
12 This methodology follows Huang (1985), who analyzed the PBC for different periods in the U.S.
13 Fiscal deficit changes are calculated as a nominal change in fiscal deficits, expressed in percentages. GDP
27 regression models, the option remains to follow a fixed effect model, which would allow the model to include for entity- and time-specific effects and thereby control for country- and year-specific effects whilst testing the relation between election timing and index performance. The fixed effects model is shown in Equation V:
𝑌𝑖𝑡 = 𝛼𝑖+ 𝛽1𝐸𝐿𝐸𝑖𝑡+ 𝛽2∆𝐹𝐼𝑆𝐶𝑖𝑡+ 𝛽3𝐺𝐷𝑃𝑖𝑡+ 𝜀𝑖𝑡 (V) Where 𝛼𝑖 is the time-invariant country specific effect.
To test for the presence of fixed country effects, as a potential bias to the results, a Hausman-test is used in order to analyse the necessity of employing fixed effects regression. The Hausman-test compares the coefficients between fixed effect regression and random effect regression, which is more efficient but inconsistent if fixed effects occur. The results of the test are presented in Appendix III. As the coefficients are highly similar, no significant difference occurs and the Ho of consistency is not rejected. Thus, no fixed effects regression is demanded by the Hausman test.
In order to double check robustness of the model to fixed effects, equation V is run and the output is compared to the general model specified in equation IV. The output is given in Appendix IV. It can be observed that the results are comparable to a great extent, with entity specific effects being insignificant in explaining the variations of the model. Thus, in the rest of this paper, the results are obtained from a pooled OLS regression model, as fixed effects appear not to threaten the consistency of the estimators. Next, certain assumptions underlying least square regression are discussed.
ii. Assumption tests
In order to draw statistically correct conclusions, the data may have to be corrected for the assumptions behind a multi-linear regression.
Firstly, a linear relationship is required between dependent and independent variables. This is tested for by a closer examination of the data and a comparison with alternative models. After examining scattered data plots and the descriptive data, no immediate non-linearity problem is witnessed for both fiscal balances and GDP growth. Additionally, the Pearson r correlation coefficients as reported in Appendix I give supplementary evidence of the linear relationships between the variables.
To test the linearity assumption more formally, a model transformation will be applied in order to check for quadratic relationships. Quadratic relationships, or often referred to as polynomial relationships, are common in financial and macroeconomic data (Stock and Watson, 2011) and an
