Bachelor Thesis Economics and Finance Universiteit van Amsterdam
February 2016
The impact of elections on stock return and volatility:
A study of the UK markets around election time.
Abstract: Previous studies conclude that stock returns are higher under right wing governments.
Furthermore, electoral uncertainty is expected to increase stock volatility. Using OLS regression, event
study methodology and a GARCH (1,1) model, this study examines the effect of political uncertainty on
stock returns and volatility in election periods from 2001 until 2015. Election polls are used to estimate
the expected political regime, and healthcare stock returns are examined for an opposite effect caused by
an expected left wing government. The results suggest that there is no specific effect of elections on stock
return, and that electoral uncertainty decreases stock volatility.
Keywords: Stock market performance, healthcare stock, elections, GARCH model, political risk. JEL Classification: G14, G18
Tamir Mohamed 10274839 Supervisor:
Table of contents
The impact of elections on stock return and volatility: 1
Table of contents 2
1. Introduction 3
2. Theoretical framework 4
3. The Electoral System of the United Kingdom General Elections 8
4. Methodology and data 9
4.1 Data 9
4.2 Methodology 10
5. Results 15
5.1 Results for OLS regression 15
5.2 Results for GARCH (1,1) model 20
6. Conclusion 23
References 25
Appendix A: STATA output for GARCH model 27
Appendix B: Output for GARCH (1,1) models 28
Appendix C: STATA output for OLS Regression: 29
Appendix D: UK Healthcare expenditure 30
1. Introduction
In the past two decades the effect of elections and change in political regime on stock market returns has been researched frequently. In 2008 Füss and Bechtel concluded that a rightwing political regime positively influences stock market returns for small stock, but their results were inconclusive for large companies’ stock returns. In 2003 Santa-Clara and Valkanov did a similar research about elections influences on stock market returns concluding that stock market excess returns are higher under Republican administrations, than under Democratic party administrations.
To an investor, knowing what the effect of a change in political regime has on the stock market is viable knowledge. However, for investors it is adamant to know what influences political risk and change in political regime. Frequently used indicators of what the expected political regime will be are voting polls. According to the efficient market hypothesis (EMH) all available information on a stock should be reflected in its price (Fama, 1970). Fama’s theory states that the information contained in the voting polls are incorporated in a stock prices. For this reason, this thesis will use polls to explain the effects of elections on stock market returns and volatility.
In contrast with the previously stated conclusions, healthcare industries are arguably not negatively affected by a left wing political regime. An explanation for this can be the increased government expenditure on healthcare generally associated with a left wing political regime (Vatter & Rüefli, 2003). Firms as well as private investors that invest in healthcare industries are likely to be interested in the effect of change in political regime on that particular industry. In light of this particular interest it is relevant to investigate the effect of change in political regime on returns on both regular and healthcare stock
This thesis will focus on how elections in the United Kingdom (UK) affect UK stock market returns and volatility, and in particular returns on UK healthcare industry stocks. The goal of this research is to answer the following research question:
Do elections in the UK have an effect on stock returns and volatility, and in particular healthcare stock returns?
In this thesis daily UK stock returns of general, small cap and healthcare stock will be regressed to election polls using OLS regression. For this study combined samples of four 60, 20 and 10 day intervals surrounding the election days of 2001, 2005, 2010 and 2015 will be used. In the combined sample the proportion of Labour votes to conservative votes will be used as an explanatory variable. To control for omitted variables stock returns of European markets excluding UK companies will be used as control variables. In addition to an OLS regression, the event study methodology introduced by MacKinlay is used to infer whether elections significantly affect stock market returns (1997). The event study methodology tests whether abnormal returns are significantly different in election periods, with the goal of inferring whether a Conservative win affects abnormal stock returns. Regarding
the effect on volatility, an analysis of estimated volatility of the stock returns will be conducted using a generalized autoregressive conditional heteroskedasticity (GARCH) model. Using a GARCH (1,1) model, the relation between volatility of stock returns and elections will be examined. The main findings of this thesis are that increased electoral uncertainty decreases stock return volatility, and that elections do not significantly affect both nominal and cumulative abnormal returns.
The further structure of the thesis is as follows. Section 2 discusses previous studies related to this research and forms the expected results. Then in section 3 the UK electoral system and the use of election polls in this research will be addressed. In section 4 the methodology and data of this research are described. In section 5 the results of the research are stated and discussed. Finally, in section 6 a conclusion is given based on the results aiming to answer the research question. Additionally, the limitations of this thesis are given and some suggestions for further research will be made.
2. Theoretical framework
A prevalent theory on how stock prices are affected is the efficient market hypothesis by Eugene Fama. This theory states that all available information is incorporated in the prices of stocks, and therefore the market processes information efficiently (Fama, 1970). With regard to a change in political regime, this implies that information related to the change of political regime should also be incorporated in the prices of stocks. In this section results of previous studies investigating the effect of elections on stock returns will be discussed.
Regarding the precise effect of elections on stock prices in general there has been related research. Füss and Bechtel researched the effect of the German elections of 2002 on stock market returns. Their main findings were that only small companies’ returns were significantly negatively (positively) correlated with left (right)-wing governments (2008). They argued that larger companies diversify their political risk, and therefore their returns are not affected by a change in political regime. Furthermore, they hypothesized that right leaning governments induce low tax and low welfare expenditure policies. Such policies generally result in lower minimum wage and lower corporate taxes, which are beneficial to firm profitability (Füss & Bechtel , 2008). Additionally, they argued that linking political regime and stock market returns to inflation policy only seems adequate if government has significant control over monetary policy (Füss & Bechtel , 2008). Similarly, the House of Commons, the governments of the UK, doesn’t exert significant control over monetary policy. The level of independence of The Bank of England is such that election cycles don’t affect monetary policy (Mishkin, Matthews, & Giuliodori, 2013).
A study by Gordon Gemmill regarding the FTSE100 index during the 1987 elections studied the relation between elections and the FTSE100 index, using opinion polls. This study concluded that the probability of a Conservative
win was positively correlated with the index’s price. Further he added that this information was reflected in the price efficiently, supporting the efficient market hypothesis by Fama (Gemmill, 1992).
In 1998 Hudson, Keasey and Dempsey analyzed stock returns under Conservative and Labour governments in the UK using data from 1945 until 1994. Their main findings were that stock returns tend to react positively to Conservative governments being elected. However, there was no significant evidence that stock markets performance was overall better in periods with Conservative governments (Hudson, Keasey, & Dempsey, 1998). In ordnance with the conclusions of previous research this study expects to find a positive relation between right wing government and stock market returns. Based on these expectations the following hypothesis is formed:
Hypothesis 1:
The null hypothesis for the returns on the stock market is that the coefficient of left wing political regime equals zero, and is smaller than zero alternatively. The alternative hypothesis is synonymous with the explanation that a left wing political has a negative effect on returns. This hypothesis can be expressed mathematically as:
H0: βs= 0 , H1: βs< 0
Where βs represents the coefficient of the binary variable that equals 1 when a left wing political regime is
expected.
As opposed to the previously stated consensus that a Labour (Conservative) government is negatively (positively) correlated with stock market returns, the effect on healthcare stock in particular can be challenged to be different. Rüefli et al. state that increased healthcare expenditure is correlated with partisanship, which means a left-winged government (2003). Healthcare expenditure provides relief for those whom can’t entirely pay their own medical bill. Donald Freeman studied the price elasticity of healthcare goods in two studies. Both studies found that expenses on health and pharmaceuticals have an elasticity below unity, arguing that patients healthcare expenses are not elastic to price changes (2003) (2012). Digesting this, it is plausible that an increase in healthcare expenditure allows patients to buy more or better medicine that they couldn’t afford previously. This meaning that an additional source of income for healthcare companies is supplied, which in turn positively effects profitability. Increased income and profitability are determinants of future dividends paid out by the company, which according to the dividend discount model should be incorporated into the price (Berk & DeMarzo, 2014).
In 1974 the year in which government expenditure became increasingly centralized in government terms, due to the inception of the National Health Service (Harker, Rachael; Social and General Statistics, 2012). On average Labour governments do tend to have higher healthcare expenditure as percentage of GDP (see table in Appendix D). The average growth rate of government expenditure on healthcare as a percentage of GDP is also larger than that of the Conservative party. These findings corroborate with those of Füss & Bechtel, which state that government welfare expenditure is increased under left wing governments (2008).
The above-mentioned statements are the basis of the following hypothesis:
Hypothesis 2:
The null hypothesis for the returns on healthcare stock is that the coefficient of left wing political regime equals zero, and is larger than zero alternatively. The alternative hypothesis is synonymous with the explanation that a left wing political regime is has a positive effect on healthcare stock returns. This hypothesis can be expressed mathematically as:
H0: βh= 0 , H1: βh> 0
Where βh represents the coefficient of the binary variable that equals 1 when a left wing political regime is
expected.
In addition to the implications of the expected political regime, the overall effect of elections on stock returns is examined. A previous study by Jones and Banning using samples of a period of over 104 years examines the relation between elections and stock returns (2009). Their study concluded that there is no significant effect of elections on stock market returns. When examining a specific event in time, such as an election, event study methodology is a tool that can describe the effects of that event on stock returns. Event study methodology focuses on cumulative abnormal returns, and the average thereof (MacKinlay , 1997). In order to test a possible relation between elections and cumulative abnormal returns the following hypothesis is formulated:
Hypothesis 3:
The null hypothesis for the effect of elections on cumulative abnormal returns is that they do not significantly differ from zero. This hypothesis can be expressed mathematically as:
H0: 𝐶𝐴𝑅! = 0, H1: 𝐶𝐴𝑅!≠ 0
Where 𝐶𝐴𝑅! represents the cumulative abnormal returns in election i. Overall the alternative hypothesis is that
abnormal returns differ in election periods.
Based on the studies of the FTSE100 return by Hudson et al. and Gemmill, this study expects to find positive cumulative abnormal returns in election i, if the Conservative party wins election i (1998) (1992).
Regarding the effect of elections and political regime on volatility of stock returns there have been several studies. In 2005 Beaulieu et al. investigated the relation between political risk and the volatility of stock returns using a GARCH (1,1) model by estimating the conditional variance of the return. Following the dividend discount model they hypothesized that the expected cash flows of a firm are impacted by political risk. This impact influences the future expected dividends which in term futures the stock price and makes it more volatile in states of higher
political risk (Beaulieu, Cosset, & Essaddam, 2005). Their main conclusions were that political risk and uncertainty increases stock return volatility but that political risk can be diversified so that it doesn’t affect an investor’s portfolio.
Leblang and Mukherjee studied the effect of United States (US) and UK elections on monthly stock return volatility using a GARCH (1,1) model (2005). They analyzed the relationship between voting intentions (polls) and the volatility of return. Their main findings were that a higher probability of Democratic and Labor win is positively correlated with lower volatility. Also higher electoral uncertainty increases volatility for both the US and UK (Leblang & Mukherjee, 2005). Leblang and Mukherjee used trading volume as an explanatory variable for volatility. Trading volume had a significant effect on volatility in the both samples.
The aforementioned study by Füss and Bechtel found that an increased probability of a right wing government increases volatility, and argues that future political prospects are reflected by price volatility, which complies with the efficient market hypothesis (2008). Contrary to other studies, Füss and Bechtel found that electoral uncertainty decreases volatility using a GARCH (1,1) model. Their analysis of the effect of uncertainty on volatility they did not use robust standard errors, which in their opinion rendered their findings inconclusive.
Using the implications of previous studies on effects of elections on stock return volatility, this study will use the subsequent hypotheses regarding stock return volatility:
Hypothesis 4.1:
The null hypothesis for the volatility of stock returns is that the coefficient of left wing political regime equals zero, and is smaller than zero alternatively. The alternative hypothesis is synonymous with the explanation that a left wing political regime decreases stock return volatility. This hypothesis can be expressed mathematically as:
H0: βv= 0 , H1: βv< 0
Where βv represents the coefficient of the probability of a left wing government winning the election.
Hypothesis 4.2:
The null hypothesis for the volatility of stock returns is that the coefficient of electoral uncertainty regime equals zero, and is larger than zero alternatively. The alternative hypothesis is synonymous with the explanation that increased electoral uncertainty increases stock return volatility. This hypothesis can be expressed mathematically as:
H0: βv= 0 , H1: βv> 0
3. The Electoral System of the United Kingdom General Elections
To determine how an election period is related to stock market returns and stock price volatility, the variable determining the expected political regime is useful. In order to determine the expected political regime, the electoral system of the general elections needs to be clearly defined. In this part of this essay the electoral system of the United Kingdom and the use of election polls will be discussed.
In the United Kingdom there are several types of elections, in this study we will regard the General Elections. General elections are held once every five years as of 2011 following the Fixed-term Parliaments Act 2011 (FTPA). Before the FTPA, elections were held after a parliament was dissolved. In ordnance with the Parliament Act 1911, parliaments had to expire within five years. Before the expiration date parliaments were dissolved by royal decree. After the dissolution general elections were held. In this study a total of four general elections will be examined: the General elections in 2001, 2005, 2010 and 2015. For the General Elections all citizens that are of voting age are allowed to vote for a candidate in their constituency. The candidate representing their constituency will go on to claim a seat in the House of Commons. After the general elections a new government is formed. This formation consists of a coalition of parties or a single party; either of which must hold the majority of seats. If no single party holds the majority of seats, then a coalition must be formed that contains a majority in the House of Commons. The amount of constituencies and therefore the amount of seats in the House of Commons are determined by the Boundaries Commissions and fluctuate over time.
Table 1: Governments formed and election dates in UK from 2001-2015
Election dates Winning party Prime Minister Coalition or Majority
June 7th 2001 Labour Tony Blair Majority government
May 5th 2005 Labour Tony Blair Majority government
May 6th 2010 Conservative James Cameron Coalition government with the Liberal Democratic party May 7th 2015 Conservative James Cameron Majority government
*Source: http://www.parliament.uk/about/how/elections-and-voting/
In the periods regarded by this study the amount of seats is 659, 648, 650 and 650 for election years 2001, 2005, 2010 and 2015 respectively. Using polls can approximate the expected government. More concisely, in order to approximate the expected government, the amount of seats per party needs to be estimated. The most accurate way to do this would be to take data from polls in every constituency, and use this data to estimate the amount of seats won per party. Such data is not available for most election years regarded in this study, at least not abundantly (Hanretty, Lauderdale, & Vivyan, 2015). National polls will be used to estimate the expected government. It is
clear that polls portraying percentages of total votes per party are not a precise measure indicating the amount of seats a party will obtain. These polls are an indicator of the probability of either a right or left winged government, and are more frequently available during election periods. According to the theory of Fama all information available will be incorporated in the price. Given that the closest estimate of the election results is given by national polls, and not by polls per constituency due to lack of data, a national poll will be information that can affect the price.
4. Methodology and data
In this thesis two effects of elections on stock market returns will be examined: the effect on stock market returns, and the effect on stock market return volatility. In this section the methodologies used to analyze these effects and data used in this study are described.
4.1 Data
In this study stock returns are measured by using market indices. Market indices are weighted composites of a selection of stocks. The benefit of using market indices to measure performance, is that the performance is not firm specific and thus systematic (Berk & DeMarzo, 2014). Besides performance, indices contain less idiosyncratic risk as opposed to singular firm stocks. To accurately measure the returns of a stock index over a period of time a total return index will be used. A total return index is an index that represents the total value of the index including directly reinvested dividends since the inception of the index. Meaning that all dividends are directly reinvested in the underlying portfolio, which is not measured by a regular index. The benefit of using a total return index is that it incorporates the benefits of dividends paid and therefore more accurately reflects the portfolio returns over time. Using data from Bloomberg the historical prices of the indices are used. By using the historical prices quoted by Bloomberg, the daily return is calculated by:
𝑅! = 𝑃! 𝑃!!!− 1
Where 𝑅! represents the daily return on day t, 𝑃! represents that price of the index on day t, and 𝑃!!!represents the price on day t-1. As proxy for the return on UK stock market return the daily returns on the FTSE100 total return index (ticker: TUKXG) and the FTSE Allshare total return index (ticker: ASXTR) are used. The proxies used for UK healthcare stock return are the FTSE ASX Healthcare total return index (ticket: TFTNCYCG) and the FTSE EPRA/NAREIT Healthcare total return index (ticker: TELUHE). The proxy with ticker TELUHE was founded in 2006 therefore this sample has data only for the elections of 2010 and 2015. In addition to general stock market and healthcare stock indices, indices that capture a different market capitalization are included. The FTSE Smallcap total return index (ticker: SCXG) is included to investigate whether the effect of elections differs for different levels of market capitalization.
Besides the stated indices, several indices are used as control variables to explain returns of stock and healthcare stock. The control variables are solely ex-UK indices whose constituents are companies in Europe that reflect either the general stock market or the healthcare industry. The control variable for the general stock market is the MSCI Europe Ex UK total return index (Ticker: MSDE15XN). The STOXX 600 Europe Ex UK Healthcare total return index (Ticker: SXDG) is used as a control variable for the healthcare industry. Also the MSCI Europe ex-UK Small cap total return index (ticker: NLCDUK) is used to control for the general stock market performance of lower market capitalization firms. The descriptive statistics and correlation table of the dataset are given in Appendix C. To estimate the expected political regime polls published on YouGov researcher Anthony Well’s website:
http://ukpollingreport.co.uk. Polls for the elections of 2001, 2005, 2010 and 2015 are used and organized in the following way: if a YouGov poll is available, this will be used, otherwise the aggregate of the other available polls is used for the missing dates. Similar to the study by Jensen and Schmidt, if no poll is available for a particular day, the most recent poll is used for that day (2005). In order to avoid major inaccuracies, the polling data has no gaps between published polls greater than 4 business days.
4.2 Methodology
In order to determine the effect of left- or rightwing political regime on the UK stock market returns the following OLS regression model will be used:
𝑅(𝑀𝐾𝑇)! = 𝛼 + 𝛽𝐿𝐴𝐵!+ 𝜆𝑅(𝑀𝐾𝑇𝑒𝑥𝑈𝐾)! + 𝜀!
The dependent variable 𝑅(𝑀𝐾𝑇)! represents the returns on the UK stock market (FTSE100 and All share indices) at time t. The first independent variable 𝐿𝐴𝐵! will assume values of 1 if the Labour party leads the polls, 0.5 if there is a tie between the Labour and Conservative party and 0 otherwise. The independent variable 𝑅(𝑀𝐾𝑇𝑒𝑥𝑈𝐾)! represents the returns on the stock market in Europe excluding the UK (ticker: MSDE15XN) at time t. Lastly the constant term is represented by 𝛼 and the error term at time t is represented by 𝜀!. The first independent variable, the constant term and error term are identical for each of the regression models, and are not individually addressed further on.
The model used to determine the effect of political regime on returns on UK healthcare industry stocks is given as: 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! = 𝛼 + 𝛽𝐿𝐴𝐵!+ 𝜃𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥𝑈𝐾)! + 𝜀!
The dependent variable 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! represents the returns on the UK stock market at time t, which will be represented by indices with ticker: TELUHE and TFTNCYCG. The second independent variable
𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥𝑈𝐾)! represents the returns on healthcare stock in Europe excluding the UK market (ticker: SXDG) at time t. Additionally a model used analyzing the effect of political regime on returns on UK Smallcap stock is given as:
The dependent variable 𝑅(𝑆𝑚𝑎𝑙𝑙)! represents the returns on FTSE Smallcap index (ticker: SCXG) stock at time t. The second independent variable 𝑅(𝑆𝑚𝑎𝑙𝑙𝑥𝑈𝐾)! represents the returns on the MSCI Europe ex-UK Small cap index (Ticker: NLCDUK) at time t.
These three models will be used performing OLS regression over three time periods surrounding elections: t=(-30,+30), t=(-10,+10) and t=(-5,+5). Where t is the amount of days before, and when positive, after election day.
Subsequently the results from this regression are used to infer whether the previously stated hypotheses are true or false. All models are tested for heteroskedasticity using the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity. In the case of heteroskedasticity, robust standard errors are used to estimate the coefficients of the model. The overall significance of the model is tested using the F-statistic.
Apart from implications of political regime change, the overall effect of UK elections on stock returns is investigated by using the event study methodology introduced by MacKinlay (1997). MacKinlay suggests that in studying the effect of an event on stock returns, abnormal returns need to be examined.
In order to estimate whether abnormal returns are affected by an event, MacKinlay specifies an estimation period (T0,T1) and an event period (T1,T2).
Figure 1: The event study timeline
The estimation period starts at T0 and end on T1, this period is comprised of 280 days. In the original event study methodology presented by MacKinlay a sample of 250 business days is used, however in this study a sample of 196 days is used in the estimation period. According to Kothari and Warner, a period of 180 days is sufficient for an estimation period (2004). The event period (t=-20, t=20) consists of 41 days, starting 20 days before, and ending 20 days after the election day.
A simple regression will be performed on returns from the estimation period, estimating the parameters 𝛼, 𝛽. Returns are given as:
𝑅!" = 𝛼 + 𝛽𝑅(𝐸𝑈𝑀𝐾𝑇)!" + 𝜀!" with 𝐸 𝜀!" = 0, 𝑣𝑎𝑟 𝜀!" = 𝜎!!
Where 𝑅! represents the return, 𝑅(𝐸𝑈𝑀𝐾𝑇)!" the external market return, and 𝜀! the error term on day t in election
i. The estimated parameters 𝛼 and 𝛽 are estimators of population parameters 𝛼 and 𝛽 respectively. In order to determine the return, in this study the aforementioned total return indices for the UK and external market will be used. Next the abnormal returns 𝐴𝑅!" for the event period is calculated using the estimated parameters:
𝐴𝑅!" = 𝑅!"− 𝛼 − 𝛽𝑅(𝐸𝑈𝑀𝐾𝑇)!" The cumulative abnormal return of election i is given by:
𝐶𝐴𝑅! = 𝐴𝑅!" !!
!!!!
To test whether stock returns differ in election period i, the specified hypothesis is tested: H0: 𝐶𝐴𝑅! = 0, H1: 𝐶𝐴𝑅! ≠ 0
MacKinlay states that if sample size is large, the variance of the cumulative abnormal return is given as: 𝑣𝑎𝑟 𝐶𝐴𝑅! = 𝜎!! 𝑇!, 𝑇! = 𝑇!− 𝑇!+ 1 𝜎!!!
Using the above, the variance of the cumulative abnormal returns 𝜎!! 𝑇!, 𝑇! will be estimated using the variance of the error term 𝜎!!!. To test if abnormal stock returns differ in an event period, De Jong uses the following test statistic:
!"!!
!(!"!!)~𝑁(0, 𝜎 !
! 𝑇!, 𝑇! ),
where 𝜎(!"!!)= 𝑑 ∗ (𝑆𝑡. 𝑑𝑒𝑣 𝐴𝑅!" ), with 𝑑 the amount of days in period i.
In order to infer whether, on average, elections have an effect on abnormal returns the average abnormal returns are examined. The average abnormal return of N elections at time t is given as:
𝐴𝑅!= 1 𝑁 𝐴𝑅!" ! !!!
The average of the cumulative abnormal returns is given as:
𝐶𝐴𝑅 = 𝐶𝐴𝑅!
! !!!
Under the null hypothesis the average abnormal returns are equal to zero. The alternative hypothesis states that on average the cumulative abnormal returns differ in election time. Shin et al. and de Jong have studied average
cumulative abnormal returns using MacKinlay’s definitions (2013) (2007). The test statistic they used to test average of cumulative abnormal returns follows a student-t distribution and is specified as:
𝑡 = 𝑁 𝐶𝐴𝑅
𝑠 ~ 𝑡!!!
Where 𝑠 = !!!! !!!!(𝐶𝐴𝑅!− 𝐶𝐴𝑅)!, and N is the number of elections examined.
Regarding the effects of elections and electoral uncertainty on stock return volatility, the GARCH (1,1) model is used to estimate the conditional variance. The reason a GARCH model is used to analyze the effect of exogenous variables, such as electoral uncertainty, on return volatility is because of volatility clustering. Volatility clustering is the occurrence that periods of low (high) volatility are followed by periods of low (high) volatility. This quantifies as a relation between previous time period variance and returns and today’s variance.
According to Leblang and Mukherjee, in order to use the GARCH model some conditions need to exist in the independent variable (2005). One condition is the phenomenon of volatility clustering, which can be examined graphically. Another condition is serial correlation, and in specific the existence of ARCH effects. The existence of ARCH effects in the residuals can be tested with the Lagrange-Multiplier test for ARCH effects. In Appendix B the test statistic for the FTSE100 returns is given, and implies there are ARCH effects in the residuals. Figure 2 displays the existence of volatility clustering of the residuals of the FTSE100 return, which means according to Leblang and Mukherjee the preconditions are satisfied.
Figure 2: Volatility clustering of the residuals of the FTSE100 return.
-. 1 -. 0 5 0 .05 .1 F T SE1 0 0 re tu rn
2001 election 2005 election 2010 election 2015 election timeline
The GARCH (1,1) model used in this thesis comprises of two equations: the mean equation and the variance equation. The mean equation is represented by:
(𝑀𝐾𝑇)! = 𝛼 + 𝛽!(𝐸𝑥𝑜𝑔𝑒𝑛)! + 𝜀!
The dependent variable (𝑀𝐾𝑇)! represents the returns on the UK stock market at time t. The independent variable (𝐸𝑥𝑜𝑔𝑒𝑛)! represents a possibly added exogenous variable. Lastly the constant term is represented by 𝛼 and the error term at time t is represented by 𝜀!. The error term 𝜀! follows a Gaussian distribution with mean 0 and variance σ2. The variance equation of the model is given as:
𝜎!
! = 𝑒!!!!(!"#$!")!!!!+ 𝛽𝑟!!!!+ 𝛾𝜎!!!!
The variable 𝑅(𝑀𝐾𝑇)! represents the conditional variance of stock return at time t. The variable (𝐸𝑥𝑜𝑔𝑒𝑛)!! represents a possibly added exogenous variable i and α! its corresponding coefficient. The ARCH (1) term is represented by 𝑟!!!! which represents the effect of the previous days return on the variance of today’s return. The GARCH (1) term is represented by σ!!!! which represents the effect of the previous days volatility on today’s volatility. The magnitude of the effects of the ARCH (1) and GARCH (1) terms are given by coefficients α! and α! respectively.
Two key assumptions are important for the GARCH (1,1) model. The first assumption is stationarity, which imposes the constraint that 𝛽+ 𝛾<1. The stationarity constraint implies that that the effect of a past shock on current volatility decreases over time (Verbeek, 2004). The second assumption is non-negativity of the coefficients. This is a specification necessary to ensure that the estimated conditional variance is positive, and restrains 𝛽 and 𝛾 from taking negative values. Normally non-negativity also applies to the constant term (ω), however in this study the constant term is included as an exponential term which cannot assume a negative value.
The GARCH model uses maximum likelihood estimation to fit an estimated conditional variance to the stock market returns. In order to reduce the bias of the estimation of the conditional variance, the dependent variable will be tested for normality using the Jacque-Bera test. If this test shows that the population of the dependent variable significantly differs from the null of normality, Bollerslev and Wooldridge semi-robust standard errors will be used for the estimation. Also a test on the residuals of the mean equation are tested by using the ARCH LM test which is chi squared distributed. The ARCH LM test is used to determine whether significant “ARCH effects” exist in the residual, meaning that there exists serial correlation in the residual of the dependent variable. Furthermore, Leblang and Mukherjee, and Füss and Bechtel use the Ljung-Box test to test whether there exists remaining serial correlation in the residuals of the GARCH (1,1) model, both studies conclude that their models have no remaining autocorrelation. This study will refrain from using the Ljung-Box test statistic, due to the findings of these studies that have shown the GARCH (1,1) model has no remaining autocorrelation.
For the GARCH model the poll data is used to determine the probability of a Labour victory. This probability is estimated similarly as been done by Leblang and Mukherjee. Leblang and Mukherjee noted that since 1940 every party that obtained 35% of the national vote won the election, this is still true today (2005). Because no other parties than the Conservatives and Labour have obtained such a proportion of the popular vote, only these two will be incorporated in the calculation of the probability of a Labour win. Leblang and Mukherjee define the probability of a Labour win at time t as follows:
𝑃!"#!= 𝜙(!!"#!! !!!" ! ! )
Where 𝑉!"# represents the voting intention noted in the polls at time t, 𝜇 represents the mean difference in voting intention in one day, and 𝜎 represents the variance of the voting intention. The probability of a Conservative win is given by 𝑃!"#= 𝑃!"#− 1.
Entropy, a measure indicating the level of political uncertainty used in research of Leblang and Mukherjee, as well as the research of Füss and Bechtel will be used. Entropy is given by:
𝐸𝑛𝑡𝑟𝑜𝑝𝑦! = 1 − 4 ∗ (𝑃!"#!− 0.5)!
Entropy is a measure of electoral uncertainty, and reaches its maximum when the probability of a Labour win equals 0.5.
The data used to fit the GARCH (1,1) model comprises of three samples. The first sample is the entire poll data set until election day, from all election years. The second sample includes 50 observations before election dates from all years. The third and final sample contains all poll data until election day from years 2005, 2010 and 2015 (excluding the 2001 election polls). In the last sample the 2001 election polls are excluded because the poll data preceding the 2001 elections is far less frequent, meaning that for extended periods only one poll per week was supplied. The poll data of the other elections is considerably more frequent; if a deviation occurs between the results of the third sample and those of other samples, this can be caused by the frequency of poll data.
5. Results
5.1 Results for OLS regression
All OLS models were tested for heteroskedasticity with the Breusch-Pagan / Cook-Weisberg test, and did not significantly differ from the null of homoskedasticity. All OLS regressions used normal standard errors. The results for the OLS regressions are shown in tables 2-4, first the coefficients of the model are shown and directly below the amount of observations, the p-value of the heteroskedasticity test (BP/CW test), the R2 and the F-statistic are displayed. The following models will be examined:
𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! = 𝛼 + 𝛽𝐿𝐴𝐵!+ 𝜃𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥)! + 𝜀! (1.2) 𝑅(𝑆𝑀𝑋)! = 𝛼 + 𝛽𝐿𝐴𝐵!+ 𝜑𝑅(𝑆𝑚𝑎𝑙𝑙𝑒𝑥)! + 𝜀! (1.3)
The corresponding tickers of the dependent variables are displayed in tables 2-4, this is considering that the models are regressing the data of that particular ticker.
Table 2: Results for the regression examining the 10 days surrounding the 2001-2015 elections, t=(-5,+5).
(1) (2) (3) (4)
Dependent var:
𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)!
𝑅(𝑀𝐾𝑇)!
𝑅(𝑀𝐾𝑇)!
𝑅(𝑀𝐾𝑇)!
Ticker: TFTNCYCG TUKXG ASXTR ASXTR
𝛽
.00496** 0.00429*** 0.00343** 0.00330*
𝜃
(0.00242) (0.00147) (0.00140) (0.00163) 0.812***
𝜆
(0.120)
0.699*** 0.697***
𝜑
(0.0403) (0.0385)
0.678***
𝛼
-0.00180 -0.00273** -0.00223** -0.00210*
(0.0445)
(0.00182) (0.00111) (0.00106) (0.00123) Diagnostics:
Observations 40 40 40 40 BP/CW test 0.9131 0.9913 0.8785 0.8725 R-squared 0.576 0.891 0.899 0.863 F 25.11 152.0 164.9 116.6 p-value 0.00 0.00 0.00 0.00
Notes: Standard errors of coefficients in parentheses, * p<0.1, ** p<0.05, *** p<0.01,
where 𝛽, 𝜃, 𝜆 and 𝜑 represent the coefficients of 𝐿𝐴𝐵!, 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥)!, 𝑅(𝑀𝐾𝑇𝑒𝑥)! and 𝑅(𝑆𝑚𝑎𝑙𝑙𝑒𝑥)! respectively. BP/CW test represents the p-value of the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity.
Regarding Table 2, the models with dependent variable 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! have results fairly similar to those of the 𝑅(𝑀𝐾𝑇)! regressions. The coefficient 𝛽, representing the effect of an expected left wing government is significantly positive for all models in Table 2 (p<0.1). This coincides with a positive effect of an expected left wing government on stock market returns. In Table 3 the coefficients 𝛽 are less significant and only statistically significant (p<0.05) for the market returns in models (7) and (8). It is reasonable, given sample size of regressions in Table 2 and Table 3, that obtaining significant results for a small sample makes inference about the entire population less reliable. However, an explanation for the increased significance of coefficients of an expected left wing government may lie in risk implications of electoral polls. Approximating the election day, polls will tend to deviate less from actual election results. An explanation for this decreased deviation is that voters are less likely to
change their mind when polls are closer to the election day (Ipsos Mori Final Poll, 2015). Following this reasoning the expected government based on polls is a more reliable variable when approaching the election day. Also comparing the coefficient 𝛽 of regressions on dependent variable 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! to those of 𝑅(𝑀𝐾𝑇)! it is noticeable that the 𝛽 coefficients are larger in all models regressing 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)!. This implies that the effect of expected left wing government is larger for healthcare stock return than general stock return.
Table 3: Results for the regression examining the 20 days surrounding the 2001-2015 elections, t=(-10,+10).
(5) (6) (7) (8) (9) (10)
Dependent
var: 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! 𝑅(𝑀𝐾𝑇)! 𝑅(𝑀𝐾𝑇)! 𝑅(𝑀𝐾𝑇)! 𝑅(𝑆𝑀𝑋)!
Ticker: TFTNCYCG TELUHE TUKXG ASXTR ASXTR SCXG 𝛽 0.00304* 0.00723* 0.00253** 0.00205** 0.00157 -0.00652 (0.00159) (0.00387) (0.000999) (0.000935) (0.00110) (0.00464) 𝜃 0.719*** 0.757*** (0.0781) (0.110) 𝜆 0.694*** 0.689*** (0.0321) (0.0300) 𝜑 0.680*** 0.837*** (0.0360) (0.0792) 𝛼 -0.000917 -0.00120 -0.00164** -0.00136* -0.000955 0.00270 (0.00121) (0.00146) (0.000757) (0.000709) (0.000839) (0.00177) Diagnostics: Observations 80 40 80 80 80 40 BP/CW-test 0.6268 0.3881 0.3694 0.4746 0.5998 0.6733 R-squared 0.537 0.567 0.861 0.873 0.824 0.755 F 44.73 24.19 237.6 265.5 179.9 57.02 p-value 0.00 0.00 0.00 0.00 0.00 0.00
Notes: Standard errors of coefficients in parentheses, * p<0.1, ** p<0.05, *** p<0.01, where 𝛽, 𝜃, 𝜆 and 𝜑 represent the
coefficients of 𝐿𝐴𝐵!, 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥)!, 𝑅(𝑀𝐾𝑇𝑒𝑥)! and 𝑅(𝑆𝑚𝑎𝑙𝑙𝑒𝑥)! respectively. BP/CW test represents the p-value of the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity.
Apart from estimated coefficients of expected left wing government for healthcare and general stock, results for lower market capitalization stock are not significant. This is shown by the results of the regressions on 𝑅(𝑆𝑀𝑋)!, where the coefficients of 𝐿𝐴𝐵!are not significant, meaning that an expected left wing government doesn’t have a significant effect on small cap stock returns. These results are inconclusive and fail to confirm the expectations that small cap firm’s return is negatively affected by elections, due to their inability to diversify political risk.
Table 4: Results for the regression examining the 60 days surrounding the 2001-2015 elections, t=(-30,+30). (11) (12) (13) (14) (15) (16) Dependent var: 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ)! 𝑅(𝑀𝐾𝑇)! 𝑅(𝑀𝐾𝑇)! 𝑅(𝑀𝐾𝑇)!
𝑅(𝑆𝑀𝑋)!
Ticker: TFTNCYCG TELUHE TUKXG ASXTR ASXTR SCXG
𝛽 0.00109 0.00298 0.000700 0.000556 0.000160 -0.00122 (0.000945) (0.00205) (0.000691) (0.000616) (0.000671) (0.00210) 𝜃 0.621** 0.705** (0.0470) (0.0639) 𝜆
0.728** 0.713**
(0.0285) (0.0254) 𝜑
0.692** 0.806**
(0.0277) (0.0463) 𝛼 -0.000702 -0.000760 -0.000786 -0.000616 -0.000503 0.00161* (0.000709) (0.000766) (0.000519) (0.000463) (0.000504) (0.000784) Diagnostics: Observations 240 120 240 240 240 120 BP/CW-test 0.1656 0.4819 0.5700 0.4247 0.5795 0.2079 R-squared 0.428 0.512 0.733 0.768 0.725 0.722 F 88.78 61.48 325.7 393.0 312.8 151.8 p-value 0.00 0.00 0.00 0.00 0.00 0.00
Notes: Standard errors of coefficients in parentheses, * p<0.1, ** p<0.05, *** p<0.01, where 𝛽, 𝜃, 𝜆 and 𝜑 represent the
coefficients of 𝐿𝐴𝐵!, 𝑅(𝐻𝑒𝑎𝑙𝑡ℎ𝑒𝑥)!, 𝑅(𝑀𝐾𝑇𝑒𝑥)! and 𝑅(𝑆𝑚𝑎𝑙𝑙𝑒𝑥)! respectively. BP/CW test represents the p-value of the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity.
In the results shown in Table 4, none of the coefficients of 𝐿𝐴𝐵! are significant, which implies that an expected left wing government does not affect stock returns. Of the models examining the effect of elections on general stock market return, only models (2) and (3) in Table 2, and models (6) and (7) shown in Table 3 have statistically significant coefficients 𝛽. In total four out of eight models with dependent variable 𝑅(𝑀𝐾𝑇)! have a significant 𝛽, and these four have relative smaller sample size. Additionally, in contrast with expectations the coefficients are positive, meaning that an expected left wing government positively affects stock market returns. Interpreting our results displayed in tables 2-4, at odds with expectations, no findings confirm that an expected right wing political regime has a positive effect on stock returns. Moreover only results from the smallest sample suggest that healthcare stock returns are positively affected by an expected Labour government, which is not sufficient evidence to infer whether or not elections effect healthcare stock returns.
Besides the analysis of daily stock returns, cumulative abnormal returns have been calculated and presented in Table 6. Remember that the abnormal return at time t is given as 𝐴𝑅!" = 𝑅!"− 𝛼 − 𝛽𝑅(𝐸𝑈𝑀𝐾𝑇)!" , and the cumulative abnormal returns for period i : 𝐶𝐴𝑅! = !!!!! !𝐴𝑅!". The estimated coefficients 𝛼 and 𝛽 and their respective standard errors are shown in Table 6. The two-sided p-value shown in Table 6 is that of the t statistic for testing cumulative abnormal returns. In the results of the event study are that cumulative abnormal returns do not
significantly deviate from zero, this shows from the two-sided p-values in Table 6, which all are larger than 0.05. Based on the research of Hudson et al. and Gemmill, the expectations were that a Conservative win would positively affect cumulative abnormal returns. This would be noticeable if the 2015 and 2010 elections would have significant higher cumulative abnormal returns. The findings suggest that there is no significant effect of elections on cumulative abnormal returns, which confirms the findings of Jones and Banning (2009). On average, cumulative abnormal returns are not significantly affected by elections. The results for the average cumulative abnormal returns are shown in Table 7. Even though on average, abnormal returns are negative (as shown by component 𝐶𝐴𝑅 in Table 7), this result is not significant with p<0.05. So overall the results suggest that elections do not affect cumulative abnormal returns. An explanation for these findings is that firms diversify their political risk, so that election outcomes do not affect their returns. Both Beaulieu et al. and Füss and Bechtel gave this explanation for their results that suggested that elections do not significantly affect stock returns (2005) (2008). Another explanation is that election results are merely a shift in political regime, but the tax changes imposed by a new government are not effective immediately. Tax changes are said to influence overall firm profits and dividends, which in turn influences firm’s (abnormal) returns. The effect of said tax changes might influence the price when the legislation is implemented as opposed to proposed by a new government. It might be worthwhile to examine the effect of the day legislation is implemented with an event study.
Table 6: Results of the event study for cumulative abnormal returns:
Year (i) Index 𝛽 𝜎! 𝛼 𝜎! 𝐶𝐴𝑅! Two-sided p-value Observations
2015 FTSE100 0.700*** (0.0371) -0.000372 (0.000363) 0,0093 0,534 41 2010 FTSE100 0.868*** (0.0199) 0.000463** (0.000223) -0,0180 0,728 41 2005 FTSE100 0.788*** (0.0325) 0.000209 (0.000200) -0,0437 0,144 41 2001 FTSE100 0.815*** (0.0522) 0.000261 (0.000711) 0,0240 0,468 41 2015 ASX 0.673*** (0.0326) -0.000342 (0.000320) -0,0191 0,662 41 2010 ASX 0.860*** (0.0185) 0.000488** (0.000207) -0,0107 0,431 41 2005 ASX 0.743*** (0.0286) 0.000264 (0.000176) -0,0462* 0,093 41 2001 ASX 0.742*** (0.0452) 0.000262 (0.000617) 0,0340 0,197 41
Notes: * p<0.1, ** p<0.05, *** p<0.01, standard errors 𝜎!, 𝜎! in parentheses. Value Index indicates which index is used to calculate the abnormal returns.
Table 7: Results for average cumulative abnormal returns.
FTSE100 ASX
𝐶𝐴𝑅 -0,010505275 -0,0070847 t-statistic -0,69542338 0,422233438 p-value 0,268408737 0,350644193
5.2 Results for GARCH (1,1) model
The GARCH (1,1) results are shown in tables 8-10, first showing the mean equation and its components, secondly showing the variance equation and lastly the amount of observations. In Appendix A the ARCH LM test for ARCH effects of the residuals of both examined mean models are shown. Both tests of the residuals show that there is a significant ARCH effect. The included results for the GARCH (1,1) have a significant fit, meaning that previous day’s return and volatility are significantly correlated with the estimated conditional variance. Also all included models satisfy the stationarity (𝛽+ 𝛾<1) and non-negativity conditions. Models that do not satisfy these conditions or that do not have a significant fit (p<0.1 for 𝛽 and 𝛾) are shown in Appendix B. The two equations of the GARCH (1,1) model and their coefficients:
𝑀𝑒𝑎𝑛: (𝑀𝐾𝑇)! = 𝛼 + 𝛿(𝐸𝑥𝑜𝑔𝑒𝑛)! + 𝜀! (2.1) 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒: 𝜎!
! = 𝑒!!!!(!"#$%&)!!!!+ 𝛽𝑟!!!!+ 𝛾𝜎!!!! (2.2)
Regarding the results shown in tables 8-10, the effect of the estimated coefficients of 𝑃!"#! and 𝐸𝑛𝑡𝑟𝑜𝑝𝑦! will be examined, and afterwards explained. The results for the sample including all poll data are shown in Table 8, where models (1), (2) and (3) are the estimates for the return of the FTSE100 total return index. Models (4), (5) and (6) represent the estimates for the FTSE Allshare total return index. Regarding the effect of political uncertainty on volatility, all statistically significant coefficients of 𝐸𝑛𝑡𝑟𝑜𝑝𝑦! are negative, implying that volatility decreases when uncertainty increases. This can be explained as a negative effect of increased electoral uncertainty on the estimated conditional variance. This contradicts the alternative hypothesis that electoral uncertainty increases volatility, and aligns with the findings of Füss and Bechtel (2008).
Table 8: Results for sample using all polls and returns from 2001-2015
(1) (2) (3) (4) (5) (6) (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! Variance 𝐸𝑛𝑡𝑟𝑜𝑝𝑦! -2.169*** -2.165*** -2.124*** -2.043*** (0.710) (0.665) (0.686) 𝑃!"#! 9.758*** -0.469 -10.30*** -0.712 (2.905) (0.758) (3.856) (0.778) 𝜔 -9.344*** -19.64*** -8.996*** -9.695*** -9.687*** -9.233*** (0.592) (2.424) (0.683) (0.589) (0.721) (0.617) 𝛽 0.204*** 0.189*** 0.209*** 0.199*** 0.180** 0.203*** (0.0719) (0.0696) (0.0724) (0.0748) (0.0704) (0.0754) 𝛾 0.624*** 0.785*** 0.596*** 0.662*** 0.791*** 0.628*** (0.120) (0.0543) (0.131) (0.102) (0.0624) (0.112) Observations: 464 464 464 465 465 465
Notes: Coefficient of constant in mean equation not reported to conserve space. Values in parentheses are
Bollerslev and Wooldridge semi-robust standard errors. * p<0.1, ** p<0.05, *** p<0.01
The effect of the probability of a Labour victory represented by 𝑃!"#!, is estimated to have a positive effect on return volatility in model (2), and a negative effect in model (5). This might imply that for firms with higher market capitalization a Labour victory is a determinant of increased stock return volatility. This might be due to the lower
market capitalization of the FTSE Allshare index used in model (4), which has lower market capitalization than the FTSE100. However, model (3) estimates the coefficient of 𝑃!"#! as a negative value, implying that an increased probability of a Labour victory decreases volatility. Because of these contradictory findings, the evidence found in this sample can be interpreted as inconclusive.
Table 9: Results for sample using all polls and returns from 2005-2015
(7) (8) (9) (10) (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! Mean 𝑅(𝑀𝐾𝑇𝑒𝑥𝑈𝐾)! 0.780*** 0.779*** 0.773*** 0.770*** (0.0235) (0.0234) (0.0222) (0.0218) 𝛼 0.0000199 0.0000125 0.0000819 0.0000590 (0.000197) (0.000197) (0.000178) (0.000177) Variance 𝐸𝑛𝑡𝑟𝑜𝑝𝑦! 0.999 2.228* (1.144) (1.221) 𝑃!"#! -4.483*** -5.509*** -3.396*** -5.447*** (1.383) (1.488) (0.960) (1.506) 𝜔 -10.81*** -11.18*** -11.47*** -12.28*** (0.628) (0.889) (0.668) (0.854) 𝛽 0.109* 0.107* 0.127** 0.122** (0.0561) (0.0575) (0.0542) (0.0571) 𝛾 0.664*** 0.640*** 0.673*** 0.590*** (0.0839) (0.107) (0.0887) (0.106) Observations: 407 407 408 408
Note: Values in parentheses are Bollerslev and Wooldridge semi-robust standard
errors. * p<0.1, ** p<0.05, *** p<0.01
Table 9 displays the results for the third sample that uses data from 2005 until 2015. Model (9) clearly estimates a negative effect of 𝑃!"#! on stock return volatility with all coefficients being significant at a 1% confidence level. The findings of model (13) suggest that increased electoral uncertainty increases the stock return volatility. This is contradicted by other results of the same sample displayed in Table 10. All the models in Table 10 estimate a negative coefficient for 𝐸𝑛𝑡𝑟𝑜𝑝𝑦!, which suggests electoral uncertainty decreases volatility.An explanation for this is that increased electoral uncertainty decreases the probability of a majority government. A “hung” government is deemed less likely to pass many new laws, as it is internally divided among political issues. In this perspective, lesser risk would exist that the current tax regime will change significantly.
A possible explanation of the findings for the coefficient of 𝑃!"#! being inconsistent between the two samples (2001-2015 and 2005-2015) is the discontinuity of the poll sample in 2001. The 2001 election polls report far less often than the 2005, 2010 and 2015 polls, meaning that for several days voting intentions are unchanged (constant) more frequently in the 2001 sample. Moreover the first sample (results in Table 8) has no unambiguous estimation of the effect of an increase in a Labour victory on the estimated conditional variance of stock returns.
The results of the second sample are not significant and are shown in Appendix B. The GARCH (1,1) models estimating the variance for the second sample have no significant estimation of the previous days effect on the conditional variance and thus volatility of stock returns. Results for this sample thus yield no inferential value, as it is constrained to a smaller sample size as well as the discontinuity off polls in 2001. This suggests that larger sample size with continuous polling data may benefit the estimation of the effect of elections on stock return volatility.
The overall findings suggest that the probability of a Labour win has no significant effect on the stock return volatility. An explanation for these findings is that this study does not specifically look at the probability of a Labour win when a Conservative government is in office and when the Labour party holds office. It might not be unlikely that the effect of an increased probability of a Labour win on stock return volatility differs as to which party currently holds office. This might be a significant factor that is not accurately incorporated in this study’s analysis.
As to how to quantify the results given by tables 8-10, it is slightly less straight forward quantifying coefficients of the GARCH (1,1) model than those of the OLS regressions. Due to the use of multiplicative heteroskedasticity, which means the independent variable and the constant are represented in an exponential term 𝑒!!!! !"#$%&!!!!, the magnitude of the effect of an exogenous variable on volatility is dependent on the constant term 𝜔. Using additive heteroskedasticity would be useful for a more straight forward interpretation of the effect of an exogenous variable on variance. A GARCH (1,1) model with additive heteroskedasticity is given as:
𝜎
!!
= 𝜔 + 𝛼
!𝐸𝑥𝑜𝑔𝑒𝑛
!!!!+ 𝛽𝑟
!!!!+ 𝛾𝜎
!!!!Table 10: Results for sample using all polls and returns from 2005-2015 (11) (12) (13) (14) (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! (𝑀𝐾𝑇)! Mean 𝑅(𝑀𝐾𝑇𝑒𝑥𝑈𝐾)! 0.783*** 0.775*** (0.0235) (0.0221) 𝛼 0.0000368 0.000489 0.000557 0.000104 (0.000203) (0.000422) (0.000408) (0.000182) Variance 𝐸𝑛𝑡𝑟𝑜𝑝𝑦! -3.551* -3.550*** -3.364*** -2.866 (2.159) (1.291) (1.006) (2.692) 𝜔 -11.00*** -9.217*** -9.458*** -12.67*** (1.080) (0.793) (0.694) (1.516) 𝛽 0.115** 0.147** 0.152** 0.129** (0.0502) (0.0603) (0.0627) (0.0507) 𝛾 0.825*** 0.786*** 0.782*** 0.854*** (0.0896) (0.0850) (0.0764) (0.0602) Observations: 407 407 408 408
Note: Values in parentheses are Bollerslev and Wooldridge semi-robust
6. Conclusion
The main focus of this study was to determine whether elections have an effect on stock market returns and volatility, and in particular healthcare stock returns. In order to explain the relation between elections and both regular and healthcare stock returns, an OLS regression was performed. This regression used proxies for stock returns by using stock returns of several total return market indices and estimated the expected political regime by using election polls.
Overall the effect of an expected left wing government is found to positively affect stock returns of regular and healthcare stock, in contrast with previous studies. Findings suggest that an expected left wing government more positively affects healthcare stock. No statistically significant effect was found on returns of low market capitalization stocks. In addition, statistically significant effects were found near election date, for larger sample sizes no significant effects were detected. Given the small sample size of the samples used for regression, it would be bold to wholeheartedly conclude that the findings of this study hold much inferential value. Besides the sample size, the main explanatory variable for the OLS regressions was a discrete variable, which hardly captures fluctuations of the overall voting sentiment. The results of the OLS regressions can’t conclude whether a Labour government positively impacts stock market returns, even though the results suggest so. In future studies it would be worthwhile to look at the effects of polls using a larger sample, with a continuous explanatory variable. Also looking at healthcare stock returns specifically, a relation between healthcare expenditure and healthcare stock return might be interesting to investigate. The implications by previous studies that show left wing governments have increased healthcare expenditure formed the basis to investigate the effect of an expected left wing government on healthcare stock. However, one could look to the slightly more obvious relation between healthcare expenditure and healthcare stock return.
Besides a simple regression, event study methodology has been used to test whether the cumulative abnormal returns of stock returns differ in election periods. It was expected that in elections where the Conservative party won, cumulative abnormal returns would be higher. The results of the event study showed no significant evidence to confirm this expectation. As mentioned earlier, the effect of an election is namely an impact on the current tax structure that applies to firms. However, if a new government is elected its future policy is not entirely determined from day one. In addition to this perspective, it is interesting to look at the dates when actual legislation is implemented. An example of this is the voting of the United States congress on Obamacare.
To determine the effects of elections and electoral uncertainty on stock market volatility, the GARCH (1,1) model was used to estimate the conditional variance of stock returns. Using the estimated GARCH (1,1) model, there is significant evidence to conclude that an increased probability of a Labour victory reduces stock market volatility. The findings suggest that increased electoral uncertainty is negatively correlated with variance of stock returns, disagreeing with earlier studies. However this effect of electoral uncertainty on volatility is that electoral uncertainty increases the probability of a “hung” government, which is less likely to implement significant policy
changes due to its internal dividedness. When such a government holds office, it will not significantly change the current legislature, such that lesser risk on firm’s future dividends is imposed. Furthermore examining what party holds office at the time of the election might hold additional explanatory value when studying the effects of elections on the stock market. Supposedly the implied risk of a Labour win is higher when the Conservative party is currently in office, and vice versa.
The GARCH (1,1) model in this study was not tested for serial correlation in the residuals because it was assumed to be nonexistent. This assumption can be false, therefore it is recommended to test the residuals for serial correlation in future studies. There are alternatives to the GARCH (1,1) model that in some occasions more accurately estimate the conditional variance of stock returns. One of which is the EGARCH model, as used by Leblang and Mukherjee (2005). In this thesis multiplicative heteroskedasticity was used to estimate the effect of elections and electoral uncertainty on stock return volatility. Using multiplicative heteroskedasticity, it is difficult to quantify the implications of the model. In future research, to better the interpretations of the findings, it is advised to use additive heteroskedasticity.
References
Beaulieu, M.-C., Cosset, J.-C., & Essaddam, N. (2005). The Impact of Political Risk on the Volatility of Stock Returns: The Case of Canada. Journal of International Business Studies , 6 (36), 701-718.
Berk, J., & DeMarzo, P. (2014). In Corporate Finance (3rd Global edition ed., pp. 272-276). Essex: Pearson. Bialkowski, J., Katrin, G., & Wisniewski, T. P. (2008). Stock market volatility around national elections. Journal of Banking & Finance , 32 (9), 1941-1953.
David, L., & Mukherjee, B. (2005). Government Partisanship, Elections, and the Stock Market: Examining American and British Stock Returns, 1930–2000 . American Journal of Political Science , 4 (49), 780–802. de Jong, F. (2007). Event studies methodology. Lecture Notes for Empirical Methods in Finance , 1-44. Tilburg, Netherlands: Tilburg University.
Füss, R., & Bechtel , M. M. (2008). Partisan politics and stock market performance: The effect of expected government partisanship on stock returns in the 2002 German federal election . Public Choice , 135, 131–150 . Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empircal Work. The Journal of Finance , 25 (2), 383-417.
Freeman, D. G. (2003). Is health care a necessity or a luxury? Pooled estimates of income elasticity from US state-level data. Applied Economics , 35, 495-502.
Freeman, D. (2012). Is health care a necessity or a luxury? New evidence from a panel of U.S. state-level data. Sam Houston State University, Department of Economics and International Business , 1203.
Gemmill, G. (1992). Political risk and market efficiency: Tests based in British stock and options markets in the 1987 election. Journal of Banking and Finance , 16, 211-231.
Hanretty, C., Lauderdale, B., & Vivyan, N. (2015, May 7). Combining national and constituency polling for forecasting. Retrieved from Election Forecast: http://electionforecast.co.uk/resources/electoral-studies-final.pdf Harker, Rachael; Social and General Statistics. (2012). NHS funding and expenditure. Retrieved from NHS History: http://www.nhshistory.net/parlymoney.pdf
Hicks, A., & Misra, J. (1993). Political Resources and the Groth of Welfare in Affluent Capitalist Democracies, 1960-1982. American Journal of Sociology , 99 (3), 668-710.
Hudson, R., Keasey, K., & Dempsey, M. (1998). Share prices under Tory and Labour governments in the UK since 1945. Applied Financial Economics , 8, 389-400.
Ipsos Mori Final Poll. (n.d.). Retrieved from Ipsos Mori:
https://www.ipsos-mori.com/researchpublications/researcharchive/3572/The-Ipsos-MORI-Final-Election-poll.aspx
Jensen, N. M., & Schmith, S. (2005). Market repsonses to politics. The rise of Lula and the decline of the Brazilian stock market. Comparative Political Studies , 10 (38), 1245-1270.
Jones, S. T., & Banning , K. (2009). US elections and monthly stock market returns. Journal of Economics and Finance , 33, 273-287.
Kothari, S. P., & Warner, J. B. (2004). In Handbook of Corporate Finance: Empirical Corporate Finance. North Holland: Elsevier.
Leblang , D., & Mukherjee, B. (2005). Government Partisanship, Elections, and the Stock Market: Examining American and British Stock Returns, 1930–2000. American Journal of Political Science , 4 (49), 780-802.
Leblang, D., & Mukherjee, B. (2004). Presidential Elections and the Stock Market: Comparing Markov-Switching and Fractionally Integrated GARCH Models of Volatility . Political Analysis , 3 (12).
MacKinlay , C. A. (1997). Event studies in Economics and Finance. Journal of Economic Literature, , 1 (35), 13-39.
Mishkin, F. S., Matthews, K., & Giuliodori, M. (2013). In The Economics of Money, Banking & Financial Markets (European edition ed., pp. 281-282). Pearson.
Santa-Clara, P., & Valkanov, R. (2003). The Presidential Puzzle: Political Cycles and the Stock Market. The Journal of Finance , 58 (5), 1841-1872.
Shin, H., Richardson, R., & Soluade, O. (2013). Review of J. D. Power Quality Measurement of U. S. and Japanese Automotive Industry. Communications of the IIMA , 1 (13), 26-42.
Stock, J. H., & Watson, M. W. (2015). In Introduction to Econometrics (Updated 3rd edition, Global Edition ed., pp. 710-715). Pearson.
Vatter, A., & Rüefli, C. (2003). Do Political Factors Matter for Health Care Expenditure? A Comparative Study of Swiss Cantons. Journal of Public Policy , 23 (3), 301 - 323.
Verbeek, M. (2004). In A guide to Modern Econometrics (2 ed., pp. 297-301). West Sussex: John Wiley & Sons, Ltd.
Data:
• Bloomberg
• http://ukpollingreport.co.uk
Statement of originality: I hereby declare that:
• All sources used for this study are properly referred to, and mentioned as references.
• This thesis is an original study that in its entirety is in no way based on material that has not been mentioned as a source.
