How developments during the Scottish Referendum of 2014 impacted the share prices of Scottish companies : an event study analysis

MSc Economics:

Monetary Policy, Banking and Regulation

Master Thesis

How Developments During the Scottish Referendum of 2014 Impacted

the Share Prices of Scottish Companies. An Event Study Analysis.

By

Joel Kirby

11086505

Submission Date: 31

st

_{January 2017}

Number of Credits: 15 Credits

Primary Supervisor/ Examiner:

Mr A.J. (Alex) Clymo MSc

Secondary Supervisor/ Examiner:

Dr. W.E. (Ward) Romp

Statement of Originality

This document is written by Joel Kirby (11086505) 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.

Abstract

The purpose of this paper is to extend event study analysis to the Scottish independence referendum of 2014. Literature on the Scottish referendum largely focuses on the speculative implications of Scottish independence, politically and economically. This paper wishes to extend this area of focus to the economic impact of developments during the referendum itself. By using standardized and

n-standardized tests, it was found that events that supported Scotland remaining in the United Kingdom (good news) had a negative impact in terms of the cumulative average abnormal returns

for Scottish companies. Whereas events supporting Scotland leaving the United Kingdom (bad news) had a negative but less conclusive impact in terms of statistical significance. With use of the

market model, the results retrieved appear to be at odds with current literature which examines unanticipated political events and its impact on the stock market performance.

Sole thanks throughout the course of the completion of this Master thesis is reserved for Alex Clymo. Alex consistently offered thorough guidance and assistance necessary for the completion of this thesis, for which I am very grateful for. His lessons and guidance is something I wish to build upon and extend after my time at the University of Amsterdam.

Table of Contents

Introduction...5

Section 1: Literature Review...8

- Theoretical Underpinnings of Stock Market Performance to New Information and Uncertainty...8

- Empirical Papers Examining Political Events and Impact on Stock Market Performance...9

Section 2: Objectives and Methodology...12

- Selection of Data and Scottish Companies...12

- Categorisation and Selection of Events...12

- Statistical Methodology...13

- Null Hypothesis and Statistical Testing...15

Section 3: Statistical Results...20

- Good News Results...21

- Interpretation of Results Representing Good News...21

- Robustness Check of Results Representing Good News...22

- Graphical Representation of Good News Results...23

- Bad News Results...24

- Interpretation of Results Representing Bad News...24

- Robustness Check of Results Representing Bad News...25

- Graphical Representation of Bad News Results...26

Section 4: Discussion of Results...27

- Analysis of Results...27

- Potential Limitations of Results...28

- Confounding Events...28

- Characteristics of Events...29

- Market Anticipation of Referendum Result...30

Section 5: Conclusion...31

Bibliography...32

Tables...35

- Table 1: List of Scottish Companies on London Stock Exchange...35

- Table 2: Chronological List of Events During Scottish Referendum...37

Introduction

From 1707 with the creation of Act of Union, Scotland became ‘united into One Kingdom by the Name of Great Britain’ with England. Since then Scotland has shared a prosperous yet at times fractious relationship with England and the United Kingdom. However, with the creation of the National Party of Scotland (the forerunners of the Scottish National Party) in 1928, this in due course had sown the seeds of an ever changing relationship with Scotland and the United Kingdom. The Scottish National Party (SNP) managed to break into Westminster politics in 1967 and since then devolution had been on the agenda ever since. From 1976, in order to obtain a majority in government, the Labour government lead by James Callaghan made an agreement with Scottish and Welsh nationals to implement legislation that would devolve powers from Westminster to Scotland and Wales in return for Labour support. This culminated in the Scottish referendum of 1979 which was to decide whether to create a deliberative assembly for Scotland. However, even though there was a majority in Scotland that voted in favour of an assembly, the act was not passed due to failing to meet the 40% voter turnout requirement. It took until 1997 for New Labour to implement a Scottish devolution referendum which would see the creation of a Scottish parliament with devolved powers over domestic affairs and tax varying powers. The Scots voted overwhelmingly for a Scottish Parliament (74.29% and 63.48% for questions over the creation of a Scottish

Parliament and the decision for tax-varying powers) and this in return passed The Scotland Act of 1998 which saw the creation of a Scottish Parliament in May 1999. It took until 2007 when the SNP was able to form its first minority government within the Scottish Parliament but was unable to implement a referendum on Scottish independence due to its minority position. However, this changed dramatically in 2011 when the SNP surprisingly won an overall majority. Given its majority position, the Scottish Independence Referendum Act of 2013 and Scottish Independence Referendum (Franchise) Act 2013 passed and subsequently enabled the Scottish independence referendum to take place on the 18th of September 2014.

Official campaigning for the Scottish referendum began 16 weeks prior to the referendum date on the 30th of May 2014. However, before the period of official campaigning, politicians and prominent figures were already making their case for Scotland’s relationship with the rest of the United Kingdom. This created a period of immense uncertainty as figures on both side of the debate gave conflicting arguments about the implications of Scottish independence. For example, on the same day in May 2014, the UK government provided analysis that Scots per head would be £1400 worse with independence whilst the Scottish government claimed Scots would be £100 per head better off (Bell et al, p7, 2014). Moreover, as the referendum date grew closer, questions such as what assets and liabilities Scotland would obtain remained unanswered. The amount of debt Scotland would obtain if it decided to split from the rest of the UK (rUK) and whether they would be able to keep the pound sterling remained ambiguous. Furthermore, Scotland’s position in the European Union was in question with EU and Scottish officials giving conflicting statements on whether Scotland would remain in the EU or would have to reapply as a new member state. Although these are only a select handful of questions that voters faced in their decision, this nonetheless highlights the great deal of uncertainty on Scotland’s future as an independent nation.

It was until 18th of September 2014 that voters took to the ballot box in order to determine the future of Scotland’s position with the rUK in which they had to answer Should Scotland be an independent country? With polls indicating that the Yes side (those in favour of Scottish

independence) was garnering support, the potential result was unpredictable. However, after polling took place the No side (those supporting Scotland remaining with the rUK) had won with 55.30% of the vote with Yes achieving 44.70% of the vote with a voter turnout of 84.59% (BBC, 2014). .

Although the No side won the referendum, there was still calls for Scotland to acquire more devolved powers. After the referendum the Prime Minister of the UK (David Cameron) pledged to

policies from both sides of the debate on what powers should be devolved further to Scotland. The Scotland Act 2016 implemented the recommendations of the Smith Commission and allowed Scotland the power to manage a variety of areas such as more tax-varying and devolved powers. Motivated by the historical relevance of the Scottish referendum and the lack of further research, the scope of this paper is to fill the gap in the literature in examining the impact itself of the Scottish referendum on the share prices of Scottish companies. In order to do so, an event study analysis will be implemented. Event studies are a commonly used tool in financial economics and other related fields in order to analyse the impact of specific events on the behaviour of firms stock prices. As Kothari and Warner (2006) emphasise, the usefulness of event studies is that they are commonly used to analyse the extent of abnormal performance of a stock market in the face of unanticipated events.

The empirical analysis of this paper focuses itself on Scottish companies registered on the London Stock Exchange. A total of 86 Scottish companies are analysed from the 14th _{of November}

2013 to the 18th _{of September 2014. In order to analyse the impact of the referendum, the events are}

classified into two groups: good news and bad news. Good news refers to events which support Scotland staying in the United Kingdom, whereas bad news refers to events focusing on Scotland leaving the United Kingdom. To perform the event study, an OLS market model is implemented with two sets of regressions: one for good news and the other for bad news. Necessary to analyse the extent to which events impacted Scottish share prices, standardized and non-standardized test statistics (time series t-test, standardized residual test, and standardized cross section test) are used to test the statistical significance of the cumulative average abnormal return (CAAR) retrieved from both regression sets. The cross-sectional t-test, Corrado rank test and generalized sign test are also provided as a means of a further robustness check. The null hypothesis essential for the purpose of this paper which is to be tested, is that good (bad) news has no effect on the stock prices of Scottish companies.

By implementing the market model, events which supported Scotland remaining in the UK (good news) appeared to have a largely negative impact in terms of the CAAR. Prior to the announcement of an event from (t = -5 to t = -1) the CAAR generated was positive at 0.27%. However, from the announcement of the event (t = 0 to t = +1), the result dropped to negative 0.33%. However, after the announcement of the event, this result begins to diminish over the course of (t = 2 to t = 3) to (t = 2 to t = 5) time-span from negative 0.10% to 0.04% in terms of the CAAR. In analysing the statistical significance of the results for good news, there was a lack of significance when one considers the time series t-test. However, when one includes the examination of the standardized test statistics, the results showed to be statistical significant around the announcement of the event.

Moreover, in terms of bad news, prior to the announcement of an event (t = -5 to t = -1), the CAAR was positive 0.08% and positive 0.05% from (t = 0 to t = 1). Moreover, during the course of the post-announcement period, the CAAR decreased to negative 0.28% for the (t = 2 to t = 3) window and increased to negative 0.17% over the (t = 2 to t = 5) event window. However, the results for bad news are hindered by a lack of statistical significance. Events representing bad news showed to be lacking in statistical significance for the CAAR prior to an event and for the announcement of a bad event. However, statistical significance did occur during the post-event period. Overall, the results prove to be contrary to the literature in examining the relationship of unanticipated political events on stock market performance.

As already mentioned, the contribution of this paper was to extend upon a lacking area of research with respect to the Scottish referendum. A key assumption in implementing the market model was that events were independent and not overlapping within respective windows of analysis. However, given the high volume of information over the Scottish referendum campaign, event clustering proved to be a troublesome issue in regards to the results. Although steps were

taken to minimise the impact of overlapping events, this is a serious issue as the abnormal returns generated for the sample of Scottish companies are no longer uncorrelated, which is what is

normally assumed. This becomes problematic in analysing the CAAR as confounding events result in bias standard errors and may lead to misleading results during hypothesis testing – therefore hindering the results retrieved. With literature examining the impact of political events and its impact on the stock market, the results retrieved appear to be contrary to the literature. In order to improve and extend upon the analysis of this paper, this can be done by taking into account the issue of overlapping events.

The remainder of this paper is as follows. Section (1) examines the literature relevant to the purpose of this study. The first part focuses on theoretical underpinnings. In light of this, the efficient market hypothesis is discussed in order to give some context to stock market performance and its

incorporation of information. Furthermore, stock market performance in times of uncertainty is addressed. Secondly, empirical papers which examine stock market performance to unanticipated political events are addressed in order to see if any parallels could be drawn to the Scottish

referendum. The conclusion derived from the literature is that unanticipated political events which create instability or uncertainty have a negative impact on stock market performance, whereas events which create stability and certainty have shown to have a positive effect.

Section (2) sets out the objectives and methodology relevant to carry out the research for this paper. Data retrieval and descriptions of how events are included and the sample of the Scottish companies are addressed. The statistical methodology explains in detail the econometric

underpinnings of the market model, with a description of the estimation period and of the event-windows included. Moreover, the relevant null hypothesis is presented and the necessary statistical tests used for its testing are presented and explained thoroughly.

Section (3) presents the results retrieved from following the methodology prescribed in section (2). The results are divided into good news and bad news and are discussed in turn. Provided in section (3) are the table of results, followed by an interpretation of the results retrieved.

Moreover, a robustness check of the results is provided with use of various statistical tests outlined in section (2). Finally, to visualise the results, a graphical representation of the results are provided.

Furthermore, section (4) gives an analysis of the the results retrieved in respect to the conclusions drawn from the literature. Moreover, potential limitations of the methodology of section (2) are addressed in order to rationalise and potentially explain the results retrieved from section (3).

Finally, section (5) offers concluding remarks in regards to the purpose of this paper, the results retrieved and highlights potential further research in this area of focus.

Section 1: Literature Review

Addressed in this section is a review of theoretical and empirical works necessary for the purpose of this paper. The first section examines the efficient-market hypothesis in order to obtain some theoretical context on how the stock market reacts to the injection of new information. Moreover, this analysis is extended to examine how the stock market performance reacts to uncertainty. This section addresses the theoretical context underpinning this paper. Furthermore, the following sub-section includes empirical papers which parallel the Scottish referendum experience. In particular, the empirical works examined focus on how the stock market of a respective country reacts to political developments and events. Given the similarities between these works and of this paper, preliminary conclusions will be drawn as to what one would expect during the Scottish referendum. Theoretical Underpinnings of Stock Market Performance to New Information and Uncertainty: The theory and underpinnings relevant to the purpose of this paper is derived initially from the work of Fama (1970) in regard to efficient capital markets. His thesis addressed the extent to which the stock market reflects information incorporated into the share price (Fama, 1970). To this extent, the market which fully reflected available information is deemed efficient. However, efficiency of the market (efficient-market hypothesis [EMH]) in terms information reflecting share prices was categorised into three variants: weak, semi-strong and strong. The weak form of the EMH claims that prices are set by information provided only by past prices and returns in which excess returns cannot be earned in the long run via this form of market efficiency. The semi-strong EMH implies that the share prices incorporate information that is publicly available in a manner in which no excess returns can be earned trading on this type of information. Finally, the strong form of the EMH reflects the hypothesis that one has monopolistic access to public and private information so that no one can earn excess returns (Fama, 1970).

Moreover, it is assumed that individuals are utility maximizing in which one forms their expectations rationally and that markets are able to aggregate this information in an efficient manner. Individuals make investment decisions based on what they expect future prices to be. When new information is introduced, then individuals update their expectations accordingly to reflect the injection of new information (Lewellan, Shanken, 1997) (Lo, 2007).

Although it is not the purpose of this paper to test the forms of the EMH, is important

however to reflect upon the theory behind share prices and information. What can be concluded, is that markets are able to update and convey the injection of new information to a varying degree in terms of share prices – in respect to the forms of the EMH. In regards to the UK markets, at a minimum it has been shown that the FTSE 100 of the London Stock Exchange, from January 2001 to November 2009 is consistent with the existence of following a random walk showing support of the weak form of the EMH (Konak, Seker, 2014). This result concur with the analysis provided by Borges (2008). Using the FTSE 100 again from the 1993 to the end of 2007, with a sub sample covering a five year period from 2003 to the end of 2007, it was found that one could not reject evidence of weak market efficiency (Borges, 2008).

Moreover, given the uncertainty of the Scottish referendum it is of importance to have a thorough understanding of how investors react under the conditions of uncertainty. The orthodox view taught in economics is to calculate the present value of expected profits of an investment and its associated costs. Then calculate the net present value of the investment. If this is found to greater than zero, then one should invest (Dixit, Pindyck, p2, 1994).

By taking a theoretical approach, Bloom, Bond et al (2001) examine short run firm-level investment with respect to demand shocks and to changes in uncertainty. Initially it was found that under no uncertainty, firms investment is positive in regards to a demand shock. However, if one is

to include uncertainty it was found that the derivative for investment is negative. This is largely put down to an increase in uncertainty increasing the real option value of investment and raising the investment threshold. Moreover, the theoretical underpinnings of their model are examined empirically and is found to be in line with their initial hypothesis. It was found that by examining 672 UK manufacturing firms during 1973 to 1991, that when uncertainty is high, their is a lower response of investment to sales growth. Furthermore, in the econometric results, the coefficient representing uncertainty is statistically significant indicating a negative effect on sales growth.

Extending this analysis further for demand and price uncertainty, Fuss & Vermeulen (2004) found similar results. By qualitatively examining individual investment by manufacturing firms, it was found that uncertainty does indeed have a negative impact on investment. More interestingly, after differentiating between demand and price uncertainty it was found that demand uncertainty has a negative and statistically significant impact on planned and realised investment. A one standard deviation in demand uncertainty (0.06) was negatively associated with the planned investment-capital ratio of 0.007. However, in regards to price uncertainty it was shown there was no evidence to suggest a negative impact on investment.

Taking the focus away from firms, Zhang (2006) examined information uncertainty and stock returns. In particular, the focus of the paper is the role of information uncertainty in price

continuation anomalies and cross-sectional variations in stock returns. In order to do so, Zhang differentiated between upwards or lower forecast revisions. If an upward forecast revision or a past winner occurred, then this was classed as good news and downward revisions or past lowers were referred to as bad news. From the econometric results, it was concluded that in the presence of information uncertainty, a high amount of uncertainty results in lower future returns following bad news and higher future returns in respect to good news.

Empirical Papers Examining Political Events and Impact on Stock Market Performance:

Examining the theory investment-uncertainty relationship is a necessary preliminary stage for the purpose of this paper. By assuming the negative relationship, would this transpire when one examines stock market performance with respect to unanticipated political events? Political events have the ability to influence the wider economy due to their implications. Relevant to the Scottish referendum is to ask how political events would have an impact on elements of the economy during times of uncertainty. As previously mentioned, the literature focusing on the Scottish referendum is primarily speculative but one can draw conclusions from literature which examines political events and their effect on stock market performance.

A vast array of literature examines unanticipated political events and its impact on the stock market. Comparing itself to the Scottish referendum, one can draw parallels with Quebec’s campaign for independence from Canada. Arguably the most relevant literature comes from Beaulieu, Cosset et al (2005) in their work on political uncertainty and stock market returns during the Quebec

Referendum of 1995. This is of interest as Scotland and Quebec both had a referendum on whether they should split from their former nation and given the similarities, one could draw parallels on its impact. By preforming event study analysis for firms headquartered in Quebec and listed on the Montreal Stock Exchange and/or the Toronto Stock Exchange, it was found that referendum result in Quebec (in which Quebec voted against independence) did indeed have a positive and

statistically significant effect for the sample of firms. Moreover it was found that the political uncertainty arisen from the referendum was of less importance for multinational firms compared to domestically based firms.

Quebec and Scotland are both democracies, have developed economies and have a similar experience in their strive for independence. Given the similarities of Scotland and Quebec with their experience on independence, Beaulieu et al's work offers powerful insight into what one would

expect with respect to the Scottish referendum. The positive stock market performance associated with the of the Quebec referendum was attributed to 'the resolution of the uncertainty regarding the uncertainty over Quebecs political future', in which the easing of this uncertainty had a positive effect. In reconciling the similarities of Quebec and Scotland with stock market performance and uncertainty, one would draw the conclusion in line with Beaulieu et al that Scotland leaving the UK would have a far more negative impact for Scottish stock market performance.In line with their work, this would be because of it introducing future political and economic uncertainty. This would be in contrast with Scotland remaining in the UK which speculatively would be more positive due to the continuation and stability of the status quo.

Furthermore, in examining how unanticipated political events have an impact on the stock market, one can look at the event study conducted by Dangol (2008) for Nepal. Dangol (2008) examines the Nepalese stock market and its reaction to unanticipated political events. The events themselves were categorised into good and bad news. Good news representing events such as an announced ceasefire between the government and Maoists, to a peace agreement between the two forces, and bad news representing events ranging from the Royal Massacre of 2001 to the imposition of capital gains tax. By employing event study methodology, it was found for the eleven banks under analysis good news generated positive abnormal returns for the post-event period, whilst bad news generated the opposite effect. Furthermore, Dangol (2008) recognised the speed of the adjustment of stock prices between 2 and 3 days from the announcement of an event.

The analysis provided by Dangol (2008) is insightful for this paper for primarily two reasons. The first is that it examines a variety of political events over a certain period of time using the same statistical method used in this paper. Secondly, like with this paper, it categorises the events into good and bad news and examines its impact on the stock market. However, where this paper departs from the work of Dangol (2008) is that instead of examining randomly occurring political events over a period of five years, this paper looks at events solely related to the Scottish referendum within a specific time frame. Given that Dangols work in conducting event studies includes the analysis for multiple events and categorising them into respective groups, this papers methodology is inspired by his work on the Nepalese stock market.

Moreover, further literature focusing on unanticipated political events show to generate similar findings. Mahmoud et al (2014), Zach (2003), Kim & Mei (1994) and He et al (2014) all examine the impact of political events across Pakistan, Israel, Hong Kong, and Taiwan. A recurrent theme throughout the literature is that unanticipated political events do indeed have an impact on the stock market. Moreover the kind of event, whether it is related to non-violent protests to the Israeli peace process with its Arab neighbours, show that the stock market indeed reacts given what happens. News that was shown to be disruptive generated negative abnormal returns, whereas news that generated an element of stability generated positive abnormal returns.

Taking a more extensive analysis, Diamonte et al (1996) examined political risk on stock returns over 45 developed and developing countries. Using an index for political risk provided by the International Country Risk Guide (ICRG) it was found that changes in political risk ‘represent an economically and statistically significant determinant on stock returns’. More interestingly, it was found that political risk had more of an effect on developing countries than developed. The difference between emerging markets experiencing an increase in political risk compared to those experiencing a decrease was approximately 11% per quarter. This was significantly lower for developed countries where the difference was only 2.5% (Diamonte et al, p75, 1996).

Overall, there is a consistent theme within the literature when it comes to analysing how political events impact stock market performance. The evidence suggests that political events do

indeed have an impact on the stock market. However, the extent to which it does is dependent on whether this creates an atmosphere of political and economic certainty. Events that create stability and certainty have shown to be positively associated with stock market performance, whereas unanticipated events which lead to instability and uncertainty create negative stock market performance. Where this paper wishes to extend this analysis, it is to take the conclusions and methods used in order to analyse the impact of political events, and see whether this holds for the Scottish referendum.

To summarise the literature, what is most apparent is that for stock market performance, certainty matters. The literature reviewed is intended to depict a picture in which the findings can be analysed in a way that could shed light on the Scottish referendum of 2014. The findings that can be drawn in parallel to the Scottish referendum is that stock market operates best in an environment in which the future is certain. When uncertainty is introduced, the literature has shown that this leads to a

negative impact on stock market performance as one is unclear about the future implications. In the case of the Scottish referendum, the referendum itself would have created an environment of uncertainty for Scotlands economic and political future. Scotland has enjoyed a union with the UK for some time, and by challenging the status quo would create huge uncertainties for Scotlands future. Where this uncertainty is addressed and stability is introduced, one would expect that the stock market performance of Scottish companies would react positively. However, with the status quo being challenged, it would be expected that this would have a negative impact in terms of stock market performance. The potential negative impact is speculated, because if the status quo is challenged with Scotlands economic future uncertain, this could result in the

expectations of Scotlands companies probability diminishing. This assumption is in line with the conclusions drawn from the theoretical and empirical works outlined above.

Section 2: Objectives and Methodology

The purpose of this paper is to analyse the extent to which the share prices of Scottish Companies, listed on the London Stock Exchange, reacted to developments during the Scottish referendum of 2014.

In order for this to be carried out, an event study will be implemented in line with previous literature. The benefit of using an event study methodology is that it can be a useful tool for analysing how events can impact a firm in regard to their share price. Share prices are highly insightful as one can use its predictive power for economic variables. Literature examining the predictive power on variables such as real GDP, private consumption and investment find that past stock prices are able to predict their trend (Andersson et al, 2011) (Andersen, Subbaraman, 1996). Selection of Data and Scottish Companies:

In gathering the relevant data, the list of Scottish companies was retrieved from the London Stock Exchange website with the Landmark selected as Scotland. The data retrieved comprises of daily share price data of all available Scottish firms and of the FTSE 100 which was obtained from theYahoo Finance website.

The criteria the selection of Scottish companies were:

- The address of the selected company had to be registered in Scotland,

- And subject to data availability for the event period. The period being 01/01/2012 to 01/01/2016. From this criteria, a total of 86 Scottish companies were selected. By categorising Scottish

companies into their respective FTSE sector, the data includes: Banks (1), Beverages (1), Construction & Materials (1), Electricity (1), Equity Investment Instruments (45), Financial Services (2), Food Producers (2), General Industrials (2), Health Care Equipment & Services (1), Household Goods & Home Construction (1), Industrial Engineering (1), Life Insurance (1), Media (3), Non-Equity Investment Instruments (2), Oil & Gas Producers (4), Oil Equipment & Services (1), Real Estate Investment & Services (2), Support Services (7), Software & Computer Services (3), and Travel & Leisure (5).

A full list of the name of the Scottish companies, their respective FTSE identification code and FTSE sector is provided in Table 1.

Categorisation and Selection of Events:

The events used for the purpose of this event study cover political and economic developments during the Scottish referendum period. The events themselves were retrieved from newspaper coverage of the Scottish referendum which was provided via LexisNexis - a provider of a variety of informational sources and materials. Given the plethora of information coverage during the Scottish referendum, the events included in this study were subject to a criteria. In order for the event to be included in the event study, the event themselves had to be from:

- Statements from UK based politicians and businessmen, and the Monarchy,

- Organised speeches from UK based politicians representing the Yes and No groups, - Statements from EU officials and EU member state officials,

- Statements from the Governor of the Bank of England (Mark Carney),

- Government reports on Scottish Independence from either the UK or Scottish government, - And polling data.

Additional information regarding government speeches and published reports on both sides of the debate were retrieved from the respective government websites. For Scottish government speeches and reports, they were retrieved from www.gov.scot and www.news.scotland.gov.uk. Conversely for the UK, this was retrieved from www.gov.uk.

Note that if the event was not obtained via the defined channels and not subject to above criteria, then the event itself was not included. This is largely in order to eliminate cognitive bias in picking and choosing events in order to skew the results in one way or another so one can obtain an unbiased and accurate depiction of the developments during the Scottish referendum.

All events occurred between the period from when the Scottish parliament passed the referendum bill on the 14th of November 2013 to the day of the referendum on the 18th of September 2014.

The events themselves are categorised into good news and bad news. Good news represent events that support Scotland remaining in the United Kingdom, and bad news represent events for Scotland leaving the United Kingdom. By categorising the events as such, one can examine how Scottish companies reacted to developments during the Scottish referendum. A full list of events is provided in Table 2.

In summary, there are in total 76 events starting from 14/11/2013 to 18/09/2014. There are 23 events representing good news with 8 events on UK government reports, 1 win on the televised debates, 11 speeches from UK and Scottish politicians, 2 business announcements backing Scotland remaining with the rUK, and Scotland voting against Scottish independence. In regards to bad news, there 53 events: the introduction of the Scottish Independence Referendum Act 2013, 6 statements from European officials, 1 win from the televised debates, 5 Scottish government reports, 12 speeches from either the Scottish First Minister or Deputy First Minister, 7 statements from either the Governor of the Bank of England or the BoE and 21 announcements from business leaders. Statistical Methodology:

In order to implement an event study, one can choose from a variety of statistical and economic models. However, given that the market model is regarded as the most widely used and highly respected method for an event study, this model will be implemented (Brown and Warner, 1985). The market model is a statistical method which measures the return of a given security relative to a market portfolio (MacKinlay, 1997, p18). In this case, the securities relates to the share price of Scottish companies, and the market portfolio relates to that of the FTSE 100. The FTSE 100 is used as the market portfolio as it is regarded as the benchmark for analysing the performance of the wider UK economy.

Furthermore, the methodological approach will follow that as provided by Dangol (2008). The market model takes the form of an ordinary least square (OLS) regression and is denoted as follows:

Ri ,t = αi + βiRm ,i + Ui ,t (1)

where Ri ,t = the return on the share price of company i at time t.

Rm t = the return on the share price of the market portfolio (FTSE 100) at time t.

αi andβi = the parameters of the market model for company i.

Ui ,t = the random error term on the share price of company i at time t.

The market model is estimated for each Scottish company in the sample using 200 daily returns. The estimation period begins 206 days up until 6 days before the announcement of an event. The estimated parameters of equation (1) and realized returns on the FTSE 100 are used to predict the normal returns before and after the announcement of an event.

The event windows used are repesented under figure 1: Figure 1: Parameter Estimation and Event Periods

In order to analyse the impact of a given event, one will examine 7 separate event windows with respect to a given event. The main event period is a 10 day window, which begins 5 days prior to the event and 5 days after (t = -5 to t = +5). Two sub-event windows are also included as (t = -3 to t = +3, and (t = -1 to t = +1). Before the announcement of the event, one will examine the returns during the pre-event period (t = -5 to t = -1). Furthermore the announcement period will cover the returns one day after the event from t = 0 to t = +1. The post-event period is broken down into two event windows: (t = 2 to t = 3) and (t = 2 to t = 5).

By introducing the overlapping (sub) event windows, this enables one to analyse the

cumulative abnormal return for the pre-event, announcement and post-event period at different time intervals and durations. In addition, given the high number of events that will be included in this research, the main and sub-event windows are kept small as to avoid the risk of event clustering. The event itself occurs during day t = 0 and all days are denoted as trading days. If the event does not occur on a trading day, then the estimate during the announcement period will commence on the next available trading day.

The coefficient estimates of equation (1) are used to generate normal returns for the seven event periods: (-5, -1), (-5, +5), (-3, +3), (-1, +1), (0, +1), (+2, +3) and (+2, +5). The standard error retrieved can be interpreted as the abnormal returns during the main event window for each Scottish company. The abnormal returns is simply the deviations of realization returns from normal returns around the main event window which is expressed as:

ARi , t = Ri , t − αi + βiRm ,i (2)

where the parameters of the model are the same as that of equation (1).

However, the abnormal returns in respect to equation (2), denote the abnormal returns for each individual Scottish company. Given one is dealing with a sample of Scottish firms (N), the average

T = -206 T = -6 T = -5 T = -3 T = -1 T = 0 T = +1 T = +2 T = +3 T = +5

Estimation Period

(-206, -6) Main Event Period (-5, +5)

abnormal return for all Scottish firms at time t, is calculated as: ARt = 1 N

∑

i=1 N ARi ,t (3)

Moreover, since this paper deals with multiple event periods across time, one has to obtain the cumulative abnormal return. The cumulative abnormal return is obtained by aggregating the sum of equation (3) over the main-event period (t = -5 to t = +5). Therefore, the cumulative average return is calculated as follows:

CAR_t =

∑

t=T1

T2

AR_t (4)

With the T subscripts denoting the 10 day main-event period (t = -5 to t = +5).

Likewise with the calculation of the average abnormal return, given that one is dealing with the analysis of a sample of firms over the given main-event period time interval, the cumulative average abnormal return (CAAR) has to be calculated. The CAAR is simply the average of the cumulative abnormal return for the sample of Scottish firms:

CAART₁, T₂ =

1

N

∑

i=1

N

CARt (5)

By following the process from equation (1) to equation (5) in order to obtain the CAAR, one is able to analyse how the cumulative average of the abnormal returns for the whole sample of Scottish firms across the main-event period reacted to good and bad news during the Scottish referendum. Null Hypothesis and Statistical Testing:

By obtaining the CAAR, the null hypothesis that is of central concern to this paper is:

H0 = the cumulative average abnormal return is equal to zero.

With the alternative hypothesis:

HA = the cumulative average abnormal return is not equal to zero.

In relation to this purpose of this study, the null hypothesis intuitively states that events with respect to Scottish independence had no effect on the stock prices of Scottish companies. The alternative hypothesis states that the events did have an effect on the stock prices. The hypothesis testing is implemented for both good and bad news.

In order to test the null hypothesis, the statistical analysis package Event Study Metrics, offers various statistical tests in order to carry out the test. The statistical tests offered by Event Study Metrics are: 1) time-series t-test, 2) cross-sectional t-test, 3) standardized residual test, 4) standardized cross-sectional test, 5) Corrado rank test and 6) generalized sign test. In order to enhance the robustness of the results, each of the statistical tests will be discussed:

-Time-Series t- Test:

The time-series t-test is the statistical significance test most familiar for those dealing with time series results. The time-series t-test is expressed as:

T_time = CAARt

(T2−T1+1)0.5 ̂σAARt

(6)

With the variance estimator based on the time-series of abnormal returns from the estimation window: ̂σAARt 2 = 1 M−d t=Est

∑

min Estmax [ AARt− 1 M Est

∑

min Estmax AARt] 2 (7) With M denoted the number of non-missing returns and d = 2 for the degrees of freedom for the market model. Moreover, given that the abnormal returns in the event window are an 'out of sample' prediction, the standard error is adjusted by the forecast error. The adjustment being:

√

1+ 1 M+ ( Rm , t−Rm , Es t) 2

∑

Es tmin Es tmax ( Rm ,t−Rm ,Es t) 2 (8) -Cross-Sectional t -Test:

The cross-sectional t-test tests the null hypothesis but with the variance estimator of the statistic being based on the cross-section of the abnormal returns. The test statistic is denoted as:

Tcross =

CAAR_{(T1 ,T2)} ̂σCAAR(T1 ,T2)

(9) with the variance estimator:

̂σCAAR(T1 ,T2) = 1 N( N−d )

∑

i=1 N [CARi(T −1 ,T2)−CAAR(T1 ,T2)] 2 (10)

-Standardized Residual Test:

The standardized residual test tests the null hypothesis under the assumption that the abnormal returns are uncorrelated and the variance is constant over time. The test was developed by Patell (1976) and the benefit of this test statistic is that it has shown to be robust in face of heteroscedastic abnormal returns in the event window.

The abnormal returns are standardized by its estimated standard deviation denoted as:

SARi ,T =

ARi ,T S( ARi)

with the standard deviation estimated from the time-series of abnormal returns of the estimation window: ̂σARi 2 = 1 Mi−dt= Est

∑

min Es tMa x ( ARi ,t) 2 (12) where M denotes the number of non-missing returns and d = 2 for the number of degrees of

freedom. Likewise, with the time-series t-test, one has to account for the 'out of sample' prediction of the event window abnormal returns. Therefore the standard error is adjusted by its forecast error:

S( ARi) = ̂σARi

√

1+ 1 M_i+ (Rm ,t−Rm , Es t) 2

∑

Es tmin Es tmax ( Rm , t−Rm , Es t) 2 (13)

Given that the abnormal returns have to be cumulated over time, the standardized version is expressed as: CSARi(T1 ,T2) =

∑

t=T1 T2 AR i , t S( ARi) (14) With respect to the null hypothesis, the distribution of SARi follows a 'students t-test' distribution.

Moreover, the expected value of CSARiis zero and the standard deviation is denoted as: S(CSARi)=

√

(T2−T1+1)

M_i−d

M_i−2d (15)

Following from this, the test statistic used in order to test the null hypothesis is expressed as:

TPatell = 1

√

N

∑

i=1 N _CSAR i(T1 ,T2) S(CSARi) (16) -Standardized Cross-Section Test:

The standardized cross-sectional test borrows from the standardized residual test in order to construct a test statistic that is robust to event-induced variance increases of stock returns. This is achieved by using an empirical variance estimate based on the cross section of abnormal returns during the event-window. This test statistic is developed by Boehmer, Musumeci and Poulson (1991). Likewise with the standardized residual test, it assumes that the abnormal returns are uncorrelated. The abnormal returns are standardized, and then the cross-sectional average of the

CSAR_i(T

1 ,T2) is obtained. This is expressed as:

CSAR_{(T1 ,}T2) =

1

N

∑

i=1

N

With the standard deviation estimated from the cross section abnormal returns during the course of the event-window: S CSAR =

√

1 N(N −1)

∑

i=1 N [CSARi(T1 ,T2)−CSAR(T1 ,T2)] 2 ₍₁₈₎

From this, one obtains the standardized cross-sectional test statistic used in order to test the null hypothesis:

T_{Boehmere ta l} = CSAR(T1 ,T2)

S(CSAR ) (19)

-Corrado Rank Test:

The Corrado Rank test was developed by Corrado (1989) and like the other statistical tests, it tests the relevant null hypothesis. However, the difference in regard to the Corrado Rank Test is that is a non-parametric rank test meaning that the data which will be implemented for the purpose of this study is not required to fit a normal distrubition. Instead the abnormal returns calculated are transformed into respective ranks. Tied ranks are treated by the method of midranks. This is done for each Scottish firm across the sample during the estimation window and main event-window:

K_{i ,T} = rank ( ARi , T) (20)

In order to adjust for any missing values, a uniform transformation of the ranks are implemented:

Ui , T =

Ki ,T

(1+Mi)

(21)

with Mi representing the number of non-missing returns for each asset.

The standard deviation, is denoted as:

S_{(U )=}

√

1 L_1+L2

∑

T[ 1

√

N_t

∑

i=1 Nt (Ui ,T−0,5 )] 2 (22)

Where Nt expresses the number of non-missing returns of the cross-section at T

Therefore, for the Corrado rank test, the test statistic is expressed as:

TCarrado =

1

√

N

∑

i=1

N

(Ui ,T−0,5 )/ S (U ) (23)

However, this test statistic is only valid for one day event windows. In order for it to be valid for longer event windows, this is done by taking the average of single day statistics, multiplied by the inverse of the square root of the periods length (Event Study Metrics, 2011).

-Generalized Sign Test:

The generalized sign test is a test constructed by Cowan (1992) and examines whether the number of stocks that experience positive CARs in the event window, is greater than the number expected in the absence of abnormal performance. It is based on the ratio of positive CARS over the event window with this ratio being a binominal random variable. The generalized sign test is expressed as: tG S = p0 +¿ − pest +¿

√

p+_est¿_{(1− p} est +¿_{)/ N} (24)

Event Study Metrics provides a variety of statistical tests in order to test the null hypothesis that the cumulative average abnormal return is equal to zero. The test statistics offered, propose alternative ways of testing the null hypothesis, with each different underlying assumptions. By testing the null hypothesis with respect to the test statistics provided, one hopes to analyse robustly the results achieved by following the outline expressed within the methodology.

Given that the initial test statistic used to test the null hypothesis is the time-series t-test, in order to improve on the robustness of the results, particular interest will be paid to the standardized residual test and the standardized cross-section test. The standardized tests offer particular use in terms of its robustness, as its been shown that standardized abnormal returns outperform non-standardized returns (Kolari, Pynnonen (2010). The non-standardized residual test offers an

improvement over the time series t-test as it is robust with respect to heteroscedastic event-window abnormal returns. Moreover, by standardizing the abnormal returns, the test grants a lower weight to the abnormal returns of share prices with large variances compared to the time series t-test. Furthermore, the standardized cross-sectional test builds upon the standardized residual test by allowing the variance of abnormal returns to differ between the estimation and event periods (Cowan, Sergeant, 1996). By running a series of simulations, Boehmer et al (1991) found that the standardized test statistics were able to avoid rejecting the null hypothesis too frequently compared to traditional event-study methods, whilst at the same time refraining from diminishing the test's power. Moreover, it was found that in the presence of event-date clustering, the results were unaffected. Given the improved performance of the standardized methods over traditional methods of statistical testing, particular interest will be granted to these two test-statistics in order to enhance the robustness of the results.

Section 3: Statistical Results

By employing the approach outlined within the methodology, two different sets of results were retrieved. The first set of results are those in regard to information with respect to good news i.e. Scotland remaining with the rUK. The second set of results were those in favour of Scotland leaving the rUK being represented as bad news.

As previously stated, the null hypothesis to be tested is that the cumulative average abnormal return is equal to zero. The time-series t-test is the initial test used to test the null hypothesis. However, given that Event Study Metrics offers other test statistics in order to test the null hypothesis, the test statistics outlined in the methodology will be incorporated into the interpretation of the results, in order to obtain a robust set of results. Given the outperformance of the Patell Z and Boehmer et al test, particular focus will be given to these test statistics in the analysis of the results.

The acceptance or rejection of the null hypothesis based on these test statistics will be done via the p-value (Prob.) on the 1%, 5% and 10% level of significance.

The results retrieved via Event Study Metrics for the Market Model are as follows keeping in mind that the estimation window was t = -206 to t = -6, the main event window being t = -5 to t = +5 (-5…5). The pre-event window is represented as t = -5 to t = -1 (-5...-1) which captures the CAAR prior to the announcement of the event. The window (0…1) represents the CAAR associated with the announcement of the event (0...1). The post-event window is represented as (2...3) and (2...5) which analyses the CAAR with respect to t = +2 to t = +3 and t = +2 to t =+5 days after the announcement of the event at day t = 0. In addition to the main event window, additional event windows of (-3…3) and (-1…1) are included. The purpose of including overlapping event windows are twofold: firstly so one can measure the CAAR for the pre-event, announcement and post-event period at different durations, secondly so one can obtain the CAAR before and after the event without the potential issue of event clustering.

Summary statistics for the sample of Scottish companies used to obtain the results are provided as Table 3 in the Table section. The Summary statistics include the number of observations, mean, median, standard deviation, and the minimum and maximum values of the share prices used in order to calculate the CAAR.

Good News

Results:

In respect to good news, table 4 provides the CAAR and the statistical significance tests for the given sub-sample.

Table 4

Cumulative Average Abnormal Returns for Good News

Table 4 presents cumulative average abnormal returns for good news. The sample consists of 1806 observations. The selected normal return model is 'Market Model' with the estimation window (-206;-6) and the maximum event window (-5;5).

(τ,τ) CAAR Pos : Neg T-test

(time-series)

Prob. T-test

(cross-sectional)

Prob. Patell Z Prob.

(-5...5) -0.0011 837 : 969 -0.2552 0.7986 -0.6872 0.4920 -0.8657 0.3867 (-3...3) -0.0034 790 : 1016 -1.0282 0.3038 -2.7495 0.0060 -2.9985 0.0027 (-1...1) -0.0039 786 : 1020 -1.7733 0.0762 -2.6355 0.0084 -1.5794 0.1143 (-5...-1) 0.0027 859 : 947 0.9574 0.3384 2.3933 0.0167 1.9793 0.0478 (0...1) -0.0033 775 : 1031 -1.8693 0.0616 -4.8180 0.0000 -3.0875 0.0020 (2...3) -0.0010 802 : 1004 -0.5487 0.5832 -1.2455 0.2129 -2.6502 0.0080 (2...5) -0.0004 820 : 986 -0.1718 0.8638 -0.4572 0.6475 -1.4654 0.1428

(τ,τ) CAAR Pos : Neg Boehmer et

al. Prob. CorradoRank Prob. Generalized Sign Prob.

(-5...5) -0.0011 837 : 969 -0.7971 0.4254 -1.2892 0.1973 0.8755 0.3813 (-3...3) -0.0034 790 : 1016 -2.9683 0.0030 -1.4325 0.1520 -1.3461 0.1783 (-1...1) -0.0039 786 : 1020 -1.3889 0.1648 -0.9246 0.3552 -1.5352 0.1247 (-5...-1) 0.0027 859 : 947 1.7569 0.0789 -0.2493 0.8031 1.9155 0.0554 (0...1) -0.0033 775 : 1031 -3.3645 0.0008 -1.1490 0.2506 -2.0552 0.0399 (2...3) -0.0010 802 : 1004 -2.5668 0.0103 -0.4688 0.6392 -0.7789 0.4360 (2...5) -0.0004 820 : 986 -1.4397 0.1500 -1.0466 0.2953 0.0720 0.9426 Interpretation of Results Representing

Good News:

Overall, in regards to the events representing the sub-sample for good news, the CAAR retrieved are surprisingly negative in respect to the return on Scottish companies share prices.

The CAAR for the pre-event period (-5...-1) was shown to be positive at 0.27% with a p value of 0.3384 indicating that one could not reject the null hypothesis that the CAAR is equal to zero. The CAAR for the announcement period (0...1) experienced a negative 0.33% change which was shown to be statistically significant at the 10% level. Moreover, the CAAR for the post-event period was shown to be statistically insignificant for the periods (2...3) and (2...5) with the respective CAAR showing a negative 0.1% change to an increase of negative 0.04%.

Furthermore, with respect to the overlapping event windows, the main-event window (-5...5) was shown to experience a negative 0.11% change in its CAAR with a statistically insignificant p

value. With respect to the sub-event windows (-3...3) and (-1...1), both experienced negative CAARs of 0.34% and 0.39%. The p value for the (-3...3) exceeded the necessary level of significance whilst the p value for the (-1...1) satisfied the 10% level of significance.

Overall, the level of insignificance in regard to the p value for good news was shown to be disappointing whilst examining the results with respect to the time-series t-test. The the time-series t-test for the main event windows (-5...-1) and (2...5) were shown to be statistically insignificant at any level of significance whilst the null hypothesis for the announcement period (0...1) could be rejected at the 10% level. By shortening the post-event period to (2...3), likewise with the main post-event period, the p value was shown to be statistically insignificant. Moreover, for the main-event windows and the sub-main-event windows [(-5...5), (-3...3), (-1...1)] the only main-event-window for the null hypothesis to be rejected was the (-1...1) window at the 10% level.

Robustness Check of Results Representing

Good News:

Aside from the time-series t-test, to check whether the results are robust, one can analyse the other test statistics provided.

In examining the cross-sectional t-test [t-test (cross-sectional)], one could reject the null hypothesis for the pre-event period (-5...-1), the announcement period (0...1) and the sub-event windows (-3...3) and (-1...1) at the 5% level of significance. However, for the main-event window and the post-event periods, (2...3) and (2...5), the null hypothesis could not rejected at any level of significance.

With respect to the standardized residual test (Patell Z), the null hypothesis could be rejected for the event-windows of (-3...3), (-5...-1), (0...1) and (2...3) at the 5% level. However, for the event-windows (-5...5), (-1...1) and (2...5), the null hypothesis could not be rejected at any level of significance.

For the standardized cross-section test (Boehmer et al.), one can reject the null hypothesis for the event-windows of (-3...3), (0...1) and (2...3) at the 5% level of significance, and the pre-event period (-5...-1) at the 10% level. However, for the other pre-event-windows (-5...5), (-1...1) and (2...5) the null hypothesis could not be rejected at any level.

The Carrado rank test (Corrado Rank) shows that the null hypothesis in regards to all of the event-windows could not be rejected with the p value exceeding the 1%, 5% and 10% level of significance.

Finally, the generalized sign test (Generalized Sign) shows that the null hypothesis could be rejected for the pre-event period (-5...-1) and the announcement period (0...1) at the 5% and 10% level of significance. For the remaining event-windows, the null hypothesis could not be rejected.

Graphical Representation of

Good News Results:

In order to simply and obtain a graphical representation of the results representing the sub-sample of good news, the CAAR is plotted onto the respective graph below:

By taking into account the non-overlapping event windows [ (-5...-1), (0...1), (2...3) and (2...5)], deviations in the CAAR from the pre-event period (t = -5 to t = -1) experienced a positive 0.27% change. However this result was hindered due to the statistical insignificance of the time-series t-test. Moreover, contrary to the literature, after the announcement of the event from (t = 0 to t = +1), it was shown that a statistically significant change in the CAAR of negative 0.33% occurred. Furthermore, by examining the post-event periods of t = +2 to t = +3 and t = +2 and t = +5, there was shown to be statistically insignificant changes in the CAAR of negative 0.1% and 0.04% respective. Again, likewise with the post-announcement period, the results retrieved appear to be contrary to the literature.

Bad News

Results:

Given the analysis of how Scottish companies reacted to events classified as good news, in turn one will now analyse the results with respect to how the share prices of Scottish companies reacted to bad news. For the sub-sample representing bad news, the statistical results are presented as follows:

Table 5

Cumulative Average Abnormal Returns for Bad News

Table 5 presents cumulative average abnormal returns for bad news. The sample consists of 4042 observations. The selected normal return model is 'Market Model' with the estimation window (-206;-6) and the maximum event window (-5;5).

(τ,τ) CAAR Pos : Neg T-test

(time-series)

Prob. T-test

(cross-sectional)

Prob. Patell Z Prob.

(-5...5) -0.0004 1884 : 2158 -0.1380 0.8903 -0.2804 0.7792 -1.9774 0.0480 (-3...3) -0.0024 1805 : 2237 -1.1030 0.2700 -2.2217 0.0263 -2.6568 0.0079 (-1...1) -0.0014 1760 : 2282 -0.9686 0.3327 -1.4494 0.1472 -2.3275 0.0199 (-5...-1) 0.0008 1840 : 2202 0.4295 0.6675 0.7530 0.4515 -0.3367 0.7364 (0...1) 0.0005 1768 : 2274 0.4663 0.6410 0.5089 0.6108 -0.2799 0.7795 (2...3) -0.0028 1763 : 2279 -2.3822 0.0172 -3.2184 0.0013 -3.7673 0.0002 (2...5) -0.0017 1813 : 2290 -1.0388 0.2989 -1.4644 0.1431 -2.7047 0.0068

(τ,τ) CAAR Pos : Neg Boehmer et

al. Prob. Corrado Rank Prob. Generalize d Sign Prob. (-5...5) -0.0004 1884 : 2158 -1.3718 0.1701 -1.8547 0.0636 1.2948 0.1954 (-3...3) -0.0024 1805 : 2237 -1.9452 0.0517 -1.5151 0.1297 -1.2001 0.2301 (-1...1) -0.0014 1760 : 2282 -2.4273 0.0152 -1.9143 0.0556 -2.6212 0.0088 (-5...-1) 0.0008 1840 : 2202 -0.3330 0.7392 -1.2915 0.1965 -0.0947 0.9245 (0...1) 0.0005 1768 : 2274 -0.2491 0.8033 -1.0917 0.2750 -2.3686 0.0179 (2...3) -0.0028 1763 : 2279 -1.8518 0.0641 -0.9362 0.3492 -2.5265 0.0115 (2...5) -0.0017 1813 : 2290 -1.3842 0.1663 -0.8597 0.3899 -0.9474 0.3434 Interpretation of Results Representing

Bad News:

Prior to the announcement of a bad event, the CAAR for the pre-event period (-5...-1) experience a positive change of 0.08% which is shown to be statistically insignificant at any level. Surprisingly, for the announcement period (0...1), there is a positive change of 0.05% for the announcement of a bad event. However, this result is statistically insignificant. For the post-event window of (2...3), there is shown to be a statistically significant decline of 0.28% in the CAAR. However, when the post-event window is extended to (2...5), the CAAR of negative 0.17% is shown to be statistically insignificant.

In analysing the overlapping event-periods, the main-event window (-5...5) experiences a negative 0.04% change from 5 days prior to a bad event to 5 days after. However, this result is

statistically insignificant. Moreover, for the sub-event windows of (-3...3) and (-1...1), the CAAR initially decreases from negative 0.04% to negative 0.24% and increases to negative 0.14%. However, for the sub-event windows, the results remain statistically insignificant.

Overall, by interpreting the results with the initial time-series t-test, the null hypothesis that the CAAR is equal to zero could only be rejected for the post-event period of (2...3) at the 5% level. In regards to the event windows (-5...5), (-3...3), (-1...1), (-5...-1), (0...1) and (2...5), the null

hypothesis could not be rejected at any level of significance. Robustness Check of Results Representing

Bad News:

Once again, one can check the robustness of the results by comparing the time-series t-test results to the other of statistical tests provided.

In analysing the t-test (cross-sectional), the null hypothesis could be rejected for the event-windows (-3...3) and (2...3). However in regard to the sub-event event-windows (-5...5), (-1...1), the pre-event period, the announcement period and the post-pre-event period (2...5), the null hypothesis could not be rejected.

According to the Patell Z test, the null hypothesis could be rejected at the 5% level for the main and sub-event windows (-5...5), (-3...3), (-1...1) and for both of the post-event periods (2...3) and (2...5). However, for the pre-event period and announcement period, (-5...-1) and (0...1), the null hypothesis could not be rejected.

For the Boehmer et al test, for the event window (-1...1) the null hypothesis could be rejected at the 5% level. The null hypothesis could be rejected at the 10% level for the

event-windows (-3...3) and (2...3). For the remaining event event-windows of (-5...5), (-5...-1), (0...1) and (2...5), the null hypothesis could not be rejected at any level of significance.

Moreover, for the Corrado rank test, the null hypothesis could be rejected for the event-windows of (-5...5) and (-1...1) at the 10% level. The event-event-windows of (-3...3), (-5...-1), (0...1), (2...3) and (2...5), the null hypothesis could not be rejected with the p values exceeding the necessary value.

Finally, with respected to the generalized sign test, the null hypothesis could be rejected for 1...1), (0...1) and (2...3) at the 5% level. However for the remaining event windows of 5...5), (-3...3), (-5...-1) and (2...5), the null hypothesis could not be rejected.

Graphical Representation of

Bad News Results:

By plotting the CAAR retrieved from the results, one can obtain a graphical representation of how the share prices of Scottish companies reacted to the announcement of bad news:

Likewise, with the analysis of Graph (1), the non-overlapping event-windows will be analysed with respect to the plotting of the CAAR graphically. With the pre-announcement period (t = -5 to t = -1), there was a statistically insignificant CAAR of 0.08%. Furthermore, during the announcement period (t = 0 to t = 1), there was a statistically insignificant increase in the CAAR of 0.05%. However, the CAAR from t = 2 to t = 3 experiences a statistically significant (at the 5% level) of negative 0.28%. This result, diminishes in statistical significance when the post-event window is extended to t = 2 to t = 5, where there is a further statistically insignificant increase in the CAAR up to negative 0.17%.

Section 4: Discussion of Results

Analysis of Results:

In reconciling the results retrieved with the conclusion drawn from the literature reviewed, the results retrieved proved somewhat troublesome. The conclusion one would wish to draw in regards to the reaction of Scottish companies to the Scottish referendum would be parallel with literature which examines how the stock markets react to unanticipated political events. From the literature discussed in section (1), stock market performance was shown to react positively to news and events which gave certainty and stability and react negatively to contrary information. One would assume that the uncertainty generated with Scotland leaving the United Kingdom would have a negative impact on the stock returns. Moreover, the expectation itself of Scotland leaving the United Kingdom may also have had a negative impact on Scottish share prices if one was to assume Scotland leaving would be less profitable than if it was to remain.

However, this conclusion of what one would expect could not be said for the Scottish referendum. In particular, after the announcement of a given event, one would expect the sample of Scottish companies to react positively to information classified as good news and negatively to that which represents bad news. However, by using the market model, it was found that the CAAR retrieved for the announcement of an event (at t = 0 to t =1) was statistically significant and negative (at the 10% level) for good news and positive for bad news however lacking in statistical significance. In particular for the announcement of an event, these result prove contrary to previous literature.

Moreover, in line with the literature one would have expected for the post-announcement period, there would be a positive and negative drift for good and bad news respective. Furthermore one would expect the post-announcement period of t = 2 to t = 3 to be statistically significant whilst the t = 2 to t = 5 window to be statistically insignificant. This is the case as one would expect that the sample of Scottish companies would take into account the introduction of an event, with the event itself losing its statistical impact over time. In the case of good news the statistical

significance of the announcement period diminished right after the t = 0 to t = 1 period with the post-announcement windows becoming statistically insignificant. Moreover, contrary with the conclusion drawn from the literature, no positive drift in the CAAR occurred with both post-announcement periods experiencing negative CAARs. However, with respect to bad news, the statistically insignificant result of the announcement period become statistically significant at the post-announcement period (t = 2 to t = 3) with a decline in the CAAR. This result, as one would expect became statistically insignificant as the post-announcement period was extended to the t = 2 to t = 5 event window.

In summary of the non-overlapping event windows, the results retrieved for the sub-sample representing good news was contrary to what one would expect in the literature. This result was contrary because the expected result of the announcement of an event was negative instead of positive, however the statistical significance was lost after the announcement of a good event over both post-announcement windows. Similarly with bad news, the results were not what one would expect as there was shown to be a positive increase in the CAAR (although statistically

insignificant) instead of negative. However, with the announcement of a bad event, there was shown to be a negative statistically significant impact after the announcement period at t = 2 to t = 3 which one would expect. Moreover, this result diminished over the t = 2 to t = 5 which one could hypothesis that the lack of statistical significance suggests that the share prices of the sample of Scottish companies incorporated bad news after the event occurred.

However, in expanding the scope in the analysis of the results towards the standardized statistical tests, the results in respect to the CAAR differ to that of the time-series t-test and prove somewhat