The effect of the Brexit Referendum on the value of British firms with different levels of foreign exposure to the EU

MSc. Finance – Asset Management

Master thesis

The effect of the Brexit Referendum on the value of British firms with

different levels of foreign exposure to the EU

Name:

Fabian van Ginkel

Student number:

10354921

University of Amsterdam: Economics and Business

Supervisor:

S. Arping

2

Content

3.1 Hypothesis I –British firms experienced negative abnormal stock returns after the Brexit Referendum ...14

3.2 Hypothesis II – British firms with higher levels of foreign exposure reacted significantly different to the Brexit Referendum ...16

3.3 Hypothesis III – Firms from different British countries reacted significantly different to the Brexit Referendum ...19 4. Data...20 5. Results ...25 5.1 Hypothesis 1...25 5.2 Hypothesis 2...27 5.3 Hypothesis 3...34 6. Robustness Checks ...35

6.1 British crown dependencies and overseas territories ...35

6.2 Firms located in London ...37

6.3 Industry sectors ...38

6.4 Trading volume...39

6.5 Firms from other European countries...40

6.6 Event-induced variance ...41 7. Conclusion ...43 Appendix A ...46 Appendix B ...49 Appendix C ...50 Bibliography ...55

3

Statement of Originality

Hereby I, Fabian van Ginkel, declare that I take the full responsibility for the content of this thesis. I declare that the content presented in this thesis is original and that only the references mentioned in the text are used to create this thesis. The Faculty of Economics and Business is only responsible for providing supervision on completing the work, not for the content of this document.

4

Abstract

This research estimates the stock market reaction of British firms to the Brexit Referendum and aims to examine if the reaction was different among British firms. The main hypothesis states that British firms with high levels of foreign exposure were affected significantly worse by the Referendum outcome compared to domestic focused firms. The study uses all publicly listed firms located in the UK and uses an event study and the crude dependence adjustment method to estimate abnormal return performance. There is significant proof of negative abnormal stock returns after the Referendum. This result is robust to event clustering and to event-induced variance increases. Moreover, firms with foreign exposure did experience significantly more severe abnormal returns. However, this observation clearly depends on the industry sector in which the firm is operating and on market size. Furthermore, there is little evidence that Scottish firms reacted less severe than British firms. Finally, the short-run effect of the Brexit was focused on British firms and firms in the financial, manufacturing and mining sector had significantly less severe abnormal returns compared to other industry sectors.

5

1. Introduction

On the 23th _{of June 2016 the United Kingdom (UK) chose in a Referendum to withdraw from the} European Union (EU). Although this decision was heavily criticized, on the 29th _{of March 2017 Prime} Minister Theresa May sent the corresponding letter to the EU to officially start Article 50 and such the Brexit negotiations. Examples of arguments in favour of the Brexit are that the belief that the UK can build a competitive economy outside the EU, better regulate migration and stimulate local industries. Examples of negative impacts are a possible capital flight from the UK and the uncertainty about the new trade agreements with the EU. The next two years of negotiations will have some major consequences on the bargaining power of the UK in worldwide trade, the relation between the UK and the EU, the position of London as financial centre of Europe and the levels of foreign direct investment in the UK. Key in the negotiations will be if the UK can maintain its position as a single market within the EU (Moore, 2016). This unique settings and uncertainty about the future puts attention to the stock market to see the translation of this important decision on the financial markets.

The Brexit not only entails uncertainty about the future, it also interrupted stock markets immediately after the Referendum. In July investors pulled out 5.7 billion pounds from UK-based stock market funds showing their lack of confidence in the British markets. Also the FTSE 100 Index dropped significantly (Rodionova, 2016). Despite some major drops the FTSE 100 Index increased with 16% after the Brexit Referendum (Moore, 2016). This demonstrates that events of political nature only temporarily increase systematic stock return risk and therefore this research only estimates the short-run effect of the Brexit on the stock market.

The aim of this research is to show that British companies are affected significantly different by the Brexit Referendum. The research question therefore is: to what extent is foreign exposure to other European countries an explanatory variable for UK firms in analysing their cross -sectional abnormal returns after the Brexit? The first hypothesis is that there are abnormal returns observed on the stock market after the Brexit. The second hypothesis of this thesis will be the expectation that firms more focused on having businesses with the rest of the EU are affected significantly worse by the Brexit outcome than companies more domestic focused. The third hypothesis is that the Brexit affected the stock returns in the four countries differently. The Brexit was not supported with a large majority, unique in the history of Europe and entails uncertainty about the next years which provides unique settings which are not studied before using the current dataset and method. The outcome of the abnormal returns study will contain valuable information about the next two years of uncertainty on the stock market in the UK. Also a recent increase of populism in the rest of Europe brings a major threat to the EU (BBC News, 2016a). This increases the demand for studies estimating the effects of an EU-exit on the stock market.

6 Taking a closer look at the Referendum outcome shows that it is only achieved with a majority of 51.9% of the votes (BBC News, 2016b). Also the Brexit was not supported among young adults. In the age group 18-24 a total of 73% wants to remain in the EU and this is still 62% in the age group 25-44 (BBC News, 2016c). Moreover, voting polls did not predict a Brexit and such the Brexit was a surprise to the stock market. This is the reason for expecting negative abnormal return performance after the Brexit. In addition, several international banks admitted that the Brexit lets them to rethink their investment position in the UK (especially London) (Inman, 2016). This puts pressure on the position of London as financial office centre of Europe. Furthermore, due to the Brexit the sterling jumped to a 31-year low rate on the 4th _{of October 2016 and this devaluation will support domestic industries in at} least the short-run (The Guardian, 2016). This two observations are the basis of the second hypothesis of this research. Despite the main division across the whole UK, the Brexit also shows a critical situation for the existence of the UK as a sovereign country. In Scotland only 38% of the people voted to leave the EU. A referendum in 2014 already showed in Scotland only a majority of 55.3% wanted to stay within the UK (BBC News, 2014). This shows that there are frictions between England and Scotland and this supports the third hypothesis.

This is one of the first studies relating the political uncertainty of the Brexit directly to the stock market of the UK. Earlier research by Schiereck et al. (2016) showed that the effect of the Brexit on the bond market was substantial but smaller than the impact of the Lehman Brothers default in 2008. Another study looked at the effects of the Brexit on stock returns in the logistics market and concluded that air transportation companies were affected the most (Schiereck & Tielmann, 2017). Also Canada was exposed to the threat of Quebec leaving the national federation during multiple years which provides a case study closely related to the current research. The study showed that the political risk mostly affected domestic focused firms (Beaulieu et al., 2005).

Despite the contribution of the research to the Brexit literature, it relates closely to the existing literature on political uncertainty. Goetzamnn and Jorian (1999) showed that stock markets are interrupted by events of a political nature. Many studies indeed proof that political uncertainty increases stock return volatility: (Boutchkova et. Al, 2012) (Pástor an Veronesi, 2013) (Brogaard and Detzel, 2015). Brogaard and Detzel (2015) and Pástor and Veronesi (2012) mentioned that stock returns prices decrease in times of political uncertainty and that this effect is stronger if the event came as a surprise. Among others Pantzalis et al. (2000) showed abnormal reactions of stock prices to elections. Li and Born (2006) even mentioned that closely contested elections have a bigger impact on the stock markets. These papers showed the importance of analysing the effects of political uncertainty on the stock market.

This current research continues on the existing literature that estimated the effect of political uncertainty on the stock market. Most studies looked at elections as an event of political uncertainty

7 and also did not take into account cross-sectional differences. This study however looks at a Referendum which is an unique case of political uncertainty since the uncertainty is unlikely to be resolved after the outcome. Furthermore, this study estimates cross-sectional abnormal return differences by defining different levels of foreign exposure as explanatory variable. Finally, the Brexit was characterized as a surprise which does not hold for most of the earlier mentioned studies and therefore makes the current settings unique and allows the research to add valuable information to the existing literature.

As stated before this research looks at levels of exposure towards other members of the EU for British firms. This exposure means that the British firms are performing business in other EU countries. This exposure is measured with three different variables: UK firms having a foreign subsidiary in the EU, firms being cross-listed on main exchanges in the EU and UK firms which have segmentation sales or profits in the EU. Boutchkova et al. (2012, p. 1117) showed that firms which are more dependent on international trade experience higher stock return volatilities in response to both higher domestic and foreign political uncertainty. This supports the hypothesis that firms with foreign exposure are expected to react significantly stronger to the Referendum outcome.

The research is conducted using a normal event study method. First, a market model is used to predict normal returns. After generating the abnormal returns for firms, the crude dependence adjustment method is applied. This controls for event clustering and is performed since all firms are exposed to the same event. The crude dependence adjustment method is performed for the two different subsamples: firms with foreign exposure to the EU and firms without that same exposure. Subsequently Wilcoxon rank-sum tests are performed for two subsamples, which is a nonparametric test used to estimate the hypothesis that two subsamples are from the same distribution. To finally estimate cross-sectional abnormal return differences several regressions are performed with foreign exposure as explanatory variable. To take into account differences in market capitalization also matching by market size is performed for several regressions.

Data is collected from the Orbis dataset of Bureau Van Dijk and contains all publicly listed UK firms with stock and market capitalization data available in 2016. These 981 firms in the dataset consist of mainly British firms (913). Despite consisting of many high capitalized firms the dataset also presents listed firms which are traded infrequently even in unique settings. Foreign exposure turns out to be highly correlated with market size. Also most internationally focused firms operate in the manufacturing sector, whereas most financial and insurance companies are domestically focused.

The study finds proof for negative abnormal stock performance on the event date June 24th and after the weekend on June 27th _{and this result is robust to event clustering. Furthermore, the} conclusion is the same for all portfolios of firms. This result is in line with the current literature that stock markets are interrupted by political events and that the magnitude is bigger if the event came as

8 a surprise. The negative abnormal reaction of the stock market shows what the event indeed was a surprise and that financial markets were unable to anticipated the short-run effect of the Referendum on the stock market.

There is little evidence for the hypothesis that firms with foreign exposure to the EU experienced more severe abnormal returns. There is significant proof of lower cumulative abnormal returns for the foreign subsidiaries and – segmentation sample. Several regressions show that for those two samples foreign exposure significantly decreases returns between 2.3% and 1.9%. Overall, firms having foreign exposure to the EU did react more negative to the Referendum tho ugh this observation clearly depends on the industry sector the firms are operating and the market size of the firms. Finally, there is little evidence that Scottish firms reacted less severe to the Referendum outcome than British firms.

For additional statistical power multiple robustness checks are performed. First, all listed firms located on the British Channel Islands or in Gibraltar show that those firms located outside the UK were affected less worse by the Brexit Referendum. Secondly, city locations of firms are used as explanatory variable and surprisingly show that firms located in London observe significantly less severe abnormal returns. A check based on industry dummies shows that real estate firms experience more severe abnormal returns. But firms in the financial, manufacturing and mining sector had significantly less severe abnormal returns. The analysis is continued with a robustness check for trading volume showing similar results compared to the second hypothesis. Moreover, the effect of the Brexit to firms from other European countries again shows that the effect of the Brexit on firms located outside the UK is less severe. Finally, the event-induced variance method is added to test for volatility differences between the event- and estimation window. The method shows that the crude dependence adjustment method is robust to event-induced variance increases.

In Section 2 the related literature is summarized. Subsequently the method of the research is described in Section 3, followed by the description of the Data in Section 4. Section 5 analyses the results of this study, Section 6 provides robustness checks and Section 7 concludes.

9

2. Literature Review

In the field of the Brexit there are only two studies available looking at the stock market reaction after the Referendum. Schiereck et al. (2016) estimated the effect of the Brexit on the bond- and stock market. The authors found that CDS (Credit Default Swaps) spreads increased after the Brexit announcement but the effect was smaller than after the default of Lehman Brothers in 2008. The study also estimated abnormal return differences for financial and nonfinancial institutions and for British and other European companies only. The effect on the stock market was more severe than after the Lehman Brothers default and EU financial banks were hit harder than non-EU financial banks. The impact of the Brexit Referendum was concentrated on financial institutions. They concluded that the Brexit had only a small effect on the riskiness of global European banks in general. Another study relates the uncertainty of the Brexit to the stock market by looking at cross-sectional abnormal return differences for different sectors in the logistics market (Schiereck & Tielmann, 2017). Again the authors proof that there were statistically significant negative abnormal returns for UK firms. In addition, they demonstrate that in the logistics sector air transportation companies were affected the most. The current study will be more extensive since it first looks at all the publicly listed UK firms and subsequently estimates cross-sectional abnormal return differences for firms based on foreign exposure to the EU. In addition, it will estimate differences in abnormal returns among multiple industry sectors.

Despite its contribution to the Brexit literature this study will also contribute to studies related to the effects of economic and political uncertainty on the stock market. It is evident that the Brexit Referendum caused uncertainty on the British stock market. A separation of the UK from the EU will change both the expected future cash flow and the discount rate of UK firms and such affect firm values (Beaulieu, Cosset, & Essaddam, 2005, pp. 703-704). Uncertainty about renegotiations of international trade agreements, a tax increase to finance the transaction costs and a reduce of investment in the UK will decrease future cash flows to UK firms. Also a capital flight and subsequently an exchange rate decrease and interest rate increase will cause a rise in the cost of capital and reduce firm values.

Other drawbacks from exiting the EU can be found in literature estimating the consequences of entering the EU on the financial markets of these particular countries. An important note here is that the UK has a more developed economy (compared to for example Croatia, Hungary, Slovenia and Czech Republic) and is additionally a worldwide player in international business which makes it harder let the conclusions assert for the UK. Data on GDP per capita from Eurostat shows that this variable is more than double for UK than for the recent EU member states (Eurostat, 2016). Moore and Wang (2007) concluded that the volatility of the stock market of new member states moves from high to low

10 when entering the EU. Dvorak (2007, pp. 15-17) shows that after entering the EU, the yields on bonds are in line with those of older EU member states and also interest rates have converged towards EU-levels. Furthermore, the banking sector is now dominated by foreign banks, the cost of capital is reduced and the stock market rally shows that the financial integration into the EU has started. These advantages contribute to the most regular improvements of greater product and labour market integration and the benefits of political integration. Another gain is that investors view an EU-membership as a characteristic of stability and this increases FDI. Finally, Dvorak and Podpiera (2006, pp. 21-22) demonstrated that stock prices increased in new EU member states and this was caused by upward revisions of future expected earnings and changes in systematic risk.

On the other side, the majority of the UK inhabitants voted pro-Brexit. The major reason for the Brexit is the idea that the UK can build a competitive economy outside the EU supported by the UK being a stable economy and for having its own currency. Although the sterling experienced a huge drop after Brexit, the UK has never defaulted on its currency. Also the UK is not closely related to the economic and political influence of the EU on the sterling anymore. Moreover, the UK can return to the world markets on their own and possibly generate bargaining power by not being part of the EU (Riley-Smith, 2016). Another argument supporting the outcome of the Brexit is the stimulation of local industries since a sovereign United Kingdom can better control its regional and agricultural policies (O'Grady, 2016). A sterling devaluation also stimulates the demand for domestic industries. Theory suggests that a country’s currency decreases in times of political uncertainty, however political instability can also cause a currency crisis (Beaulieu, Cosset, & Essaddam, 2005, p. 702). Overall, since the outcome of the negotiations is still uncertain at this moment, the UK stock market is left in a uncertain environment. It is not yet clear if the Brexit will be a success or a failure and this depends on the negotiations with the EU. The next paragraphs will focus on the influence of this economic and political uncertainty on the financial markets.

Goetzmann and Jorion (1999, pp. 978-979) reported that stock markets are interrupted by events of a political nature. This points towards the involvement of political uncertainty on the stock market. Brown, Harlow, and Tinic (1988, pp. 383-384) showed that increases in systematic risk are temporarily and resolved in the long-run. This conclusion supports the argument of only looking at the short-run impact of the Brexit Referendum on the stock market. The interruptions of political uncertainty on the stock market are translated into an increased stock return volatility. Boutchkova et al. (2012, pp. 1150-1151) showed that domestic political uncertainty is associated with systematic volatility to stocks, while global political uncertainty brings larger idiosyncratic volatility. This argues that UK firms will experience systematic volatility increases. The authors also mentioned that these return volatility increases will have different effects for different industries. They mention that industries with a higher export-orientation suffer higher return volatility when political uncertainty is

11 higher. In addition, Davis et al. (2016, pp. 2-3) created an index for this economic policy uncertainty (EPU) and showed that a rise in EPU is associated with an increase in share price volatility and a decrease in investment. Furthermore, Pástor and Veronesi (2013, p. 41) showed that political uncertainty leads to higher volatility and correlation between stocks, but the effect is weaker for developed economies. Their model implies risk premia caused by political uncertainty. Other research, performed by Chan and John Wei (1996), shows an increase in stock return volatility in association with both good or bad political news. More proof on stock return volatility comes from Schwert (1989), who demonstrated that uncertainty about future macroeconomic circumstances causes a significant stock return volatility change.

The first hypothesis that stock returns are expected to decrease after the announcement of the Brexit is supported by the literature. Brogaard and Detzel (2015, pp. 32-33) mentioned that decisions of economic agents are based on the economic policy environment of the future. The authors estimated that an increase of one percent in the economic policy uncertainty leads to a contemporaneous 2.8% decrease in the one-month market index return, but it slightly increases in the following month. The market volatility increased 18% due to the economic policy uncertainty (Brogaard & Detzel, 2015, pp. 16-17). Pástor and Veronesi (2012, pp. 37-39) demonstrated that stock prices fall due to policy changes and that this fall is expected to be larger for higher government policy uncertainty. They state that stock prices are expected to fall after a policy change announcement, conditional on the fact that the policy change came as a surprise. This is the case in the Brexit outcome, where the last poll before voting gave a 10 percent lead for ‘Remain’ (55% against 45%) (Cooper, 2016). Also bookmakers predicted on the 23th _{of June that there was a 90% chance of a Brexit rejection, such} that the Brexit can be labelled as a surprise (Doyle, 2016).

Despite these articles related to political uncertainty, many researches provide evidence of stock return volatility around national elections. Since the Brexit was caused by a voting Referendum these articles relate to the Brexit case and continue on the earlier mentioned evidence. Bialkowski et al. (2008, pp. 24-26) support this view and estimated that stock return volatility is elevated temporarily around elections, in particular caused by the return on the election day. The authors find that due to country-specific components the volatility can double in the week of the elections. Also the effect increases with uncertainty about the outcome: when it was a closely contested election, when the outcome will change the governmental orientation and when the outcome does not guarantee parliamentary majority. Especially the first two relate to the Brexit settings. Contributing on this topic, Li and Born (2006) researched elections with no clear outcome of the winner and found that this increased stock return volatility.

The contribution of this current research to the existing political uncertainty literature will be that it looks at a Referendum which is not a regular form of an election and this particular Referendum

12 can potentially have a significant large effect on the economy of the whole country. In addition, the studies mentioned before mostly looked at the overall economy. This study will take into account cross-sectional differences for a large set of UK firms and therefore has additional statistical power. Finally, as mentioned before the outcome of the Referendum was a surprise which is not always the case for regular political events and elections where accurate polls are available. Furthermore, in normal elections uncertainty is resolved after the outcome. This is not the case for the Brexit Referendum since a vote to exit will not resolve the political uncertainty existing ex-ante and even increases this uncertainty.

However, there does exist another set of studies focusing on the stock market surrounding times of political uncertainty after a Referendum (Beaulieu, Cosset, & Essaddam, 2005, pp. 713-715). In Canada there was a threat that Quebec wanted to withdraw from the Canadian federation which makes that particular study directly comparable to the Brexit. Around the political uncertainty in Canada investors experienced increased stock market volatility. However, this stock market volatility varied for different types of firms. Political news mostly affected the stock return volatility of purely domestic companies and it did not affect the volatility for companies with i nternational operations. The authors argue that political risk is diversifiable since investors did not require a risk premium. A difference between the Canadian study and the current research is the outcome. For the UK there is no way back after the Referendum outcome. The inhabitants of Quebec voted the stay within the Canadian federation. The authors concluded that the Quebec firms did experience positive abnormal returns since the outcome of the Referendum resolved the political uncertainty. This reasoning clearly does not hold in the Brexit case.

This current study estimates if firms with higher levels of foreign exposure were affected worse by the Brexit. To measure the orientation of UK firms towards the rest of the EU, foreign segmentation sales and profits can be used as indicator. Boutchkova et al. (2012, p. 1116) argued that firms more dependent on exporting are more vulnerable to demand disruptions than firms domestically focused. The demand is decreased due to political uncertainty with foreign partners of the country. This argument is based on the observation of Engel and Wang (2011) in which trade is two or even three times more volatile than GDP. In addition, Boutchkova et al. (2012, p. 1117) expected that firms which are more dependent on international trade experience higher stock return volatilities in response to both higher domestic and foreign political uncertainty. The expectation that export-orientated firms react significantly stronger to higher political risk was supported empirically. These findings show that foreign segmentation sales and profits are a valid way to measure cross-sectional return differences in response to political uncertainty.

Another measure indicating the international orientation of a particular British firm are foreign subsidiaries. In this research having a foreign subsidiary means that a UK firm has a majority of shares

13 in another firm located outside the UK and thus measures Foreign Direct Investment (FDI). This study expects that the international focus of UK firms towards the EU is an indicator of a more severe abnormal stock market reaction. Related literature already proofs that the levels of FDI in the UK are expected to decrease after the political instability. Brada et al. (2003, pp. 9-10) mentioned that political stability plays an important role in sustaining high levels of FDI and political uncertainty plays a major role in the investment decision. This demonstrates that the levels of FDI in the UK are expected to decrease due to political instability.

Finally, also cross-listing can be used as an explanatory variable of foreign exposure. Cross-listing means that besides listed on an exchange in its domestic country, a company is also listed on an exchange abroad. Lang et al. (2003, p. 342) used US data to show that cross-listed firms have better information environments and this increases their market value. Bailey et al. (2006, pp. 41-42) showed that after cross-listed in the US, firms experience higher trading volumes and higher volatility reactions to earnings announcements due to additional disclosure requirements. This demonstrates that cross-listed firms are expected to react stronger on the Brexit Referendum compared to companies with a domestic UK-focus. However, as indicator of foreign exposure the cross-listing results should be interpreted with caution. Being cross-listed does not directly mean that a firm actually performs business in a foreign country. Also some exchanges have only weak requirements to cross-list on their exchange.

14

3. Method

3.1 Hypothesis I – British firms experienced negative abnormal stock returns after the

Brexit Referendum

This first hypothesis is the basis of all other estimations in this research. It is a typical event study with an intention to show abnormal return performance on the British stock market on the days surrounding the Brexit Referendum. Subsequently those abnormal returns can be used in regressions for the second and third hypothesis. Event studies play an important role in stock market literature and estimate the effect of information becoming public knowledge on the stock market. In the basis an event study estimates if stock prices behave different around certain events when comparing their returns to normal periods, or the expected returns if there was no event. The difference between those returns is called the abnormal return. Since the returns of one firm are stochastic, event studies look at the effect of an event to multiple firms on aggregate. Finally, the hypothesis that returns around event dates are not statistically different from their expected returns in the same period is tested (Fama, Roll, Fischer, & Jensen, 1969).

In this study the event is the Brexit Referendum on the 23th _{of June 2016. The outcome became} public knowledge on the 24th _{of June 2016. For the event window at least the announcement day and} the day after the announcement need to be included (MacKinlay, 1997, pp. 14-15). Average trading volume per week from 20 May 2016 until 9 September 2016 shows a peek due to the two weeks 20-24 June and 27 June-1 July, therefore these weeks are the event window. In econometric terms this means that the event window is t1=-4 until t2=7 with the event on the 24th _{of June (t=0). The estimation} window is used to calculate the normal returns with a benchmark model. Following MacKinlay (1997) the estimation window is typically equal to the 120 trading days before the event took place. The event window starts at the 20th _{of June 2016 an therefore the last day of the event window is the 17}th _of June. To have an estimation window of 120 trading days the start is therefore the 4th _{of January} (weekends excluded). On the days 25 March, 28 March, 2 May and 30 May the Exchange was closed for different reasons. This leads to an event window of 116 trading days starting January 4, 2016 and stopping at June 17, 2016. In econometrics terms this gives an estimation window of T1=-172 until T2=-7.

To estimate the normal returns in the estimation window period a benchmark model is needed. A market model is used as benchmark which assumes the beta of each stock to be different from one (de Goeij & de Jong, 2011, pp. 5-6). The market model looks as follows:

15 In this model the return of each firm (𝑅𝑖𝑡) is regressed by the market return (𝑅𝑚𝑡) over time. This provides a β-measure which measures the sensitivity of the stock returns to the variation in the market return and is therefore used to estimate the normal returns (Fama & French, 2004). 𝛼𝑖 represents the residuals used to calculate the predicted returns and 𝜀𝑖𝑡 measures the error term. The normal returns are equal to the fitted value of the β-measure times the market return plus the fitted value of the residuals per firm over time.

The predicted return that can be expected from the market return is called the normal return. Subsequently the abnormal return is equal to the difference between this normal stock return and the actual return observed for the specific company in the event window. These abnormal returns form the basis of statistical tests to estimate abnormal return performance (de Goeij & de Jong, 2011, pp. 7-9). For testing abnormal return performance two measures are key in estimating test-statistics:

𝐴𝐴𝑅𝑡= 1

𝑁 ∑ 𝐴𝑅𝑖𝑡 𝑁

𝑖=1 (2), 𝐶𝐴𝑅𝑖= ∑𝑡2𝑡=𝑡1𝐴𝑅𝑖𝑡 (3) 𝐴𝑅𝑖𝑡 measures the abnormal return per company and per event date, 𝐴𝐴𝑅𝑡 is the average abnormal return per event date and 𝐶𝐴𝑅𝑖 is the cumulative abnormal return over the event window per firm. The abnormal return performance is tested against the hypothesis:

𝐻0: 𝐸(𝐴𝑅𝑖𝑡) = 0; 𝐻1: 𝐸(𝐴𝑅𝑖𝑡) ≠ 0

The major problem of a Referendum as event is that all stocks experience the same event at the same time which is called event clustering. The consequence is that abnormal returns are not cross-sectional uncorrelated anymore. This correlation leads to higher variances of single abnormal returns and therefore the usual variance estimator underestimates the actual abnormal return variance. This leads to an upward bias in the test-statistics and subsequently to an over rejection of the null hypothesis. The solution for this bias is to use the crude dependence adjustment method introduced by Brown and Warner (Brown & Warner, 1980, pp. 233-234). In the crude dependence adjustment method the variance of the abnormal return on the event day is estimated using the time series of average abnormal returns of the estimation window. In a normal abnormal return performance test, only the abnormal returns of the event window are used in the estimation (de Goeij & de Jong, 2011, pp. 9-10). One drawback of the crude dependence adjustment method is that the estimates are not robust to an increase in stock return volatility and cross-sectional correlation due to the event. For this reason the method of Boehmer et al. (1991) is used as an additional robustness check to take into account stock return volatilities.

To calculate the test-statistic of the abnormal return performance using the crude dependence adjustment method equation (2) is needed. Equation (2) shows average abnormal returns over time for all firms in the dataset. Now the average of those average abnormal returns over time is calculated to estimate the standard deviation-measure (de Goeij & de Jong, 2011, pp. 13-14):

16 𝐴𝑅∗₌ 1

𝑇 ∑ 𝐴𝐴𝑅𝑡 𝑇2

𝑡=T1 (4)

An important note for equation (4) is that it uses the estimation window to calculate the average return 𝐴𝑅∗_{. The standard deviation used for the test-statistic is computed as follows:}

𝑠̅ = √ 1

𝑇−1∑ (𝐴𝐴𝑅𝑡− 𝐴𝑅 ∗₎2 𝑇2

𝑡=𝑇1 (5) Again the standard deviation is also computed using only the abnormal returns observed in the estimation window. To calculate the t-test statistic used to measure abnormal return performance the average abnormal return observed on the event date is divided by this standard deviation measure:

𝑇𝑆1 = 𝐴𝐴𝑅𝑡

𝑠̅ ≈ 𝑁(0,1) (6) The distribution of the t-test statistic is approximately standard normal in large samples (N=981). For using the cumulative abnormal returns of equation (3) the t-test statistic is different and equal to:

𝑇𝑆2 = 1 √𝑇∗ 1 𝑁 ∑ 𝐶𝐴𝑅𝑖 𝑁 𝑖=1 𝑠̅ ≈ 𝑁(0,1) (7) Looking at cumulative abnormal returns is interesting when looking at the abnormal returns of the total pre-event or post-event window (de Goeij & de Jong, 2011, p. 10). For the basis student t-test which does not take into account event clustering the standard de viation is different per date:

𝑇𝑆3 = √𝑁 𝐴𝐴𝑅𝑡 𝑠𝑡 ≈ 𝑁(0,1) (8), 𝑠𝑡= √ 1 𝑁−1∑ (𝐴𝑅𝑖𝑡− 𝐴𝐴𝑅𝑡 ) 2 𝑁 𝑖−=1 (9) The hypothesis states that abnormal returns are expected to be negative after the Brexit Referendum. This is not in line with the hypothesis of the Quebec case study where returns were expected to be and turned out positive (Beaulieu, Cosset, & Essaddam, 2006). A significant difference with the current study is that in Quebec the majority of voters wanted to stay within the Canadian federation. The reason for expecting negative abnormal returns after the Brexit Referendum is that the event came as a surprise to the UK stock market. Therefore the Referendum failed to resolve the political uncertainty existing ex-ante and it even increased this uncertainty. Due to this higher uncertainty on the stock market, British firms are expected to experience significant drops in firm value.

3.2 Hypothesis II – British firms with higher levels of foreign exposure reacted significantly

different to the Brexit Referendum

The belief that firms are affected significantly different by the Referendum comes from the new set of rules expected to apply after the Brexit negotiations. Financial markets belief that in exchange for the current trade agreements the UK should accept free movement of goods and people between the EU and UK (Moore, 2016). These two conditions are both really important for the internationally focused

17 companies in the UK. Also several multinationals and banks admitted that the Brexit lets them to rethink their position in especially London (Inman, 2016). On the other hand, a decrease in the import levels of the UK stimulates the position of domestically focused firms. The outcome of the negotiations is not yet clear though, which leaves the UK firms with a focus on business with other EU countries in uncertainty. Summarizing, the hypothesis states that UK firms with an exposure to other European countries are expected to react significantly worse to the outcome of the Referendum. The most valid argument in favour of this hypothesis is that after the Referendum the political uncertainty was not resolved and even increased. This uncertainty is expected to mostly affect international ly focused companies since the new trading rules and future movement of workers with the EU is under pressure. The hypothesis is first tested in the same way as hypothesis one. After generating the average abnormal returns per event date, these returns are again judged using the crud e dependence adjustment method of equation (6). Furthermore, the cumulative abnormal returns are tested using equation (7). Both these two methods should indicate if there are abnormal returns observed for two different groups of firms: firms with an exposure to foreign EU countries and firms without that same exposure. Note that to control for event clustering the crude depe ndence adjustment method is performed for two subsamples separately. There are two ways to estimate abnormal return performance and take into account event clustering. The first one is using multivariate regressions and event time dummies. The second method is performed in this research and first estimates abnormal returns and subsequently tests for abnormal performance using the crude dependence adjustment method. Note that in both methods the abnormal returns are the same and thus can be used in the regressions of this research.

After estimating the expectation that both kind of firms are affected significantly negative by the Referendum outcome, the difference in abnormal return magnitude is tested. First, the different (cumulative) abnormal returns are evaluated using the Wilcoxon rank-sum test. This test is a nonparametric test which estimates the hypothesis that two subsamples are from the same distribution. The Wilcoxon rank-sum test looks at independent samples, which is true for the three samples of foreign exposure. However, when using paired data a Wilcoxon signed rank-sum test is more appropriate. Corrado (1989, pp. 394-395) concluded that this nonparametric rank test is better described under the null hypothesis than the signed rank and sign tests. In evaluating daily stock returns the rank-sum test has more statistical power due to the high non-normal distribution of daily stock returns.

Moreover, to further test the significance of the difference in abnormal stock returns several regressions are performed using heteroskedastic standard errors. First, the three different foreign exposure indicators are regressed on the cumulative abnormal returns of 24-27 June:

18 The reason for taking the cumulative abnormal return of this particular event window is the fact that on both these two days there were significant negative abnormal returns.

Since these first regressions lack the use control variables the results still have omitted variable bias. After including the natural logarithm of the market capitalization additional variation in abnormal returns is explained. The reason for only including the natural logarithm of the market capitalization as company size characteristic is the high correlation of this variable with weekly Brexit sales, operating revenue and total assets. Including multiple of those variables will cause multicollinearity:

𝐶𝐴𝑅𝑖= 𝛽0+ 𝛽1 ∗ 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 − 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖+ 𝛽2∗ 𝑀𝑎𝑟𝑘𝑒𝑡 − 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖+ 𝜀𝑖 (11) Moreover, also the market-to-book ratio is included since it represents the number of growth options. The underlying theory is that for firms with a higher book-to-market ratio their value depends especially on growth options and not on assets in place. Those firms are less affected by political risk since those firms can diversify away risk easily by transferring their research and development operations (Beaulieu, Cosset, & Essaddam, 2006, p. 625). Furthermore, the profitability of a company is used as a control variable. This control variable is added to estimate if more profitable firms reacted different to the Referendum outcome and since more profitable firms are more likely to be able to afford business abroad. The profitability of each firm is measured as operating revenue divided by total assets to take into account size differences. In the end, also industry dummies are included since different industry sectors might have a different magnitude of foreign exposure and reacted different to the Referendum:

𝐶𝐴𝑅𝑖= 𝛽0+ 𝛽1 ∗ 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 − 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖+ 𝛽2∗ 𝑀𝑎𝑟𝑘𝑒𝑡 − 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖+ 𝛽3∗

𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽4∗ 𝑀𝑎𝑟𝑘𝑒𝑡 − 𝑡𝑜 − 𝑏𝑜𝑜𝑘 𝑟𝑎𝑡𝑖𝑜𝑖+ 𝛽𝑗∗ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑠𝑒𝑐𝑡𝑜𝑟𝑠𝑖+ 𝜀𝑖 (12) The regressions are first performed looking at foreign exposure dummy-variables for each of the three different measures separately: cross-listing, foreign subsidiaries and foreign segmentation. Subsequently also the level of foreign exposure is used as explanatory variable. These levels are estimated for the EU-segmentation and the EU-subsidiaries sample. These two additional foreign exposure variables are described in the Data-section and give this study the ability to see the magnitude of foreign exposure for British firms.

Finally, for every of the three foreign exposure indicators firms are matched by market capitalization. For example in the foreign segmentation sample the firms with foreign segmentation in the EU are matched with firms without this same segmentation by market size. Then all the firms that are not matched are dropped out of the sample such that two equal subsamples remain. These two subsamples consist of firms with similar market capitalization and in this way are able to take away the effect of firms size on abnormal return differences. Then for these matched firms again the Wilcoxon rank-sum tests and the same set of regressions are performed using the matching weights.

19

3.3 Hypothesis III – Firms from different British countries reacted significantly different to

the Brexit Referendum

The belief that firms from different British countries are affected significantly different by the Brexit is based on the current frictions between especially Scotland and England. Also the dataset contains too few firms from Wales and Northern-Ireland and therefore looking at firms from those countries lacks statistical power. In Scotland the majority of the voters wanted to stay within the EU. In England the majority of voters wanted to leave the EU. Currently the Scottish Parliament is seriously considering to join the European Free Trade Association (EFTA). In this EFTA Scotland does not lose the position as single market in the EU like Iceland, Liechtenstein, Norway, and Switzerland. Also before the Brexit Referendum there were already frictions visible between Scotland and England. In 2014 there was a Referendum in which the Scottish population chose to remain within the UK with a majority of only 55.3% of the votes. A new Referendum might show that currently the majority of Scottish people wants to leave the UK. A new Referendum is on the other hand unlikely at this moment since Spain uses it voting rights to prevent it. Spain is afraid of losing Catalonia when Scotland is able to leave the UK. For this reason a Referendum in Scotland is not supported unanimously in the EU (Financieel Dagblad, 2017).

Since the attitude towards the European Union is different in Scotland and the majority of voters in Scotland wanted to remain within the EU the hypothesis is that for Scottish f irms the Referendum outcome came more as a surprise. Therefore the hypothesis is that Scottish firms are affected significantly worse by the Referendum outcome. This hypothesis is tested using again the matching by market size procedure since the dataset contains much more firms from England (913) than from Scotland (57). After this matching the different returns in both countries are evaluated using the Wilcoxon rank-sum test and subsequently again regressions are performed to estimate the significance of the Scotland dummy variable. Again the same control variables as in hypothesis two are used:

𝐶𝐴𝑅𝑖= 𝛽0+ 𝛽1∗ 𝑆𝑐𝑜𝑡𝑙𝑎𝑛𝑑𝑖+ 𝛽2∗ 𝑀𝑎𝑟𝑘𝑒𝑡 − 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖+ 𝛽3∗ 𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽4∗ 𝑀𝑎𝑟𝑘𝑒𝑡 − 𝑡𝑜 − 𝑏𝑜𝑜𝑘 𝑟𝑎𝑡𝑖𝑜𝑖+ 𝛽𝑗∗ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑠𝑒𝑐𝑡𝑜𝑟𝑠𝑖+ 𝜀𝑖 (13)

20

4. Data

For stock return data the Orbis dataset of Bureau Van Dijk provides daily stock prices for 2016. In the Orbis dataset 1,140 publicly listed firms located in the United Kingdom are selected. In addition, the Orbis dataset provides information on market capitalization, revenues and other values based on the firm’s performance and assets. Moreover, the dataset has data available on foreign exposure and the ability to generate country and even city location dummies. This makes the dataset the most comprehensive database for this study. DataStream provides the market indexes: the FTSE 100, FTSE 250, FTSE 350 and FTSE All-Share Index. The most comprehensive will be to use the FTSE All-Share Index since this captures about 98% of the UK market capitalization and it is the aggregation of the FTSE 100, FTSE 250 and the FTSE SmallCap Index. For this reason it is a better benchmark when using almost all listed UK firms.

First all companies publicly listed in the UK are generated from the Orbis dataset. Then these 1,140 firms are reduced by deleting all firms with no market capitalization data. Subsequently all firms with a market capitalization of below 10,000 are dropped. The dataset is further reduced with firms having no stock data available in the estimation and event window periods which provides a final set of firms equal to 981. These firms form the basis of the analyses in this research. In total 913 of all the firms are located in England and 57 firms are located in Scotland. Only 10 firms are located in Wales and even only one firm is located in Northern-Ireland. This indicates that these countries are not useful in estimating the third hypothesis.

Table 1 shows summary statistics for the 981 firms in the dataset. The total assets variable demonstrates that the dataset consists of firms with high asset values. The lower quartile has a value of 25.336 million pounds and the maximum total assets is even around 1,600 billion pounds. In addition, the mean operating revenue of the firms is 944 million pounds. However, the median operating revenue is only 29.251 million pounds showing that big firms pushing up the mean operating revenue. The same effect is visible for the total assets, shares outstanding, trading volumes and market capitalization. The average market capitalization is nearly 2 billion pounds with an minimum and maximum of 256 and 1,955 million pounds respectively. The price over book value ratio has a median of 1.91 and the mean is even 2.52 indicating that most firms in the dataset have a high market-to-book ratio and thus more growth options. The median profitability measure is equal to 44.66%, while the mean is equal to 69.06% which is due to the presence of firms with a profitability measure of above 100%. The number of subsidiaries shows that most UK firms have only one or two subsidiaries in the EU, the median is equal to 2 firms and the mean is equal to 4.38. The maximum is equal to 50 subsidiaries. The median share of EU-segmentation sales or profits in total sales or profits i s equal to 23.81% while the mean is slightly higher and equal to 34.87%.

21 Finally, the average trading volume variables show that there are still some firms in the current dataset which are very infrequently traded when looking at the minimum trading volume per month or week. However, the 25% of the firms with the lowest monthly trading volume already have a maximum of 515,715 trades. Also around the Brexit Referendum there are some firms not being traded more frequently. This might indicate the existence of firms not reacting strongly to the Referendum outcome. For the weekly average trading volume around the Brexit Referendum the 5% percentile was only 2,548 and the 10% percentile was still only 5,222. This shows that the current dataset presents not only very big capitalized firms, but in addition contains listed firms which are traded infrequently even in unique settings. For this reasons trading volume i s an important robustness check.

For the second hypothesis data for three foreign exposure characteristics is collected in order to estimate if companies with a higher exposure towards foreign markets experience different abnormal returns in reaction to the Brexit compared to firms without foreign exposure. Firstly, all companies located and listed in the United Kingdom which have a foreign subsidiary in the European Union are selected. The criteria in this selection process is that the UK firms need to have an 51% ownership in the corresponding EU firm, and thus a majority of shares. Furthermore, the number of foreign subsidiaries in the EU is also collected to see the level of foreign exposure.

Secondly, cross-listing indicates that beyond listed on the London Stock Exchange the UK firm is also listed on a stock exchange abroad. Most UK firms are cross-listed on German stock exchanges. All firms only listed at the London Stock Exchange are labell ed as domestic firms. Not all European exchanges were taken into account, but only the biggest stock exchange per European country. Most British firms were crosslisted in Frankfurt, other firms were listed on the EuronextAmsterdam, -Brussel, -Paris, Warsaw Stock Exchange, Prague Stock Exchange, Irish Stock Exchange, Nasdaq-Copenhagen or -Stockholm.

Finally, the foreign segment sales and profits shows firms which have had non -negative sales or profits obtained in the EU somewhere in the years 2015-2016. Also the sales and profits need to be higher than 100,000 pounds to satisfy the criteria. For this variable also the magnitude of the exposure is estimated. For the last two years the total sales or profit in the EU is divided by the total sales or profits of the company. Then among those four numbers the maximum is taken and used as index. However, all indexes that are higher than 1 and below zero are dropped out of the final analysis.

In total 257 firms located in the UK have a foreign subsidiary in the EU, this represents 26.20% of all the firms. A total of 724 firms in the dataset do not have a foreign subsidiary in the EU. For the cross-listing variable the share of firms with a foreign exposure is similar and equal to 28.33% which are 278 firms. 703 UK firms are not cross-listed. For foreign segment sales and profits the share of firms with a foreign exposure is 32.31% and thus the highest among the three measures. 317 UK firms obtained foreign segment sales or profits in the EU in the last years, while 664 UK firms did not.

22 Table 2 demonstrates that firms with a foreign subsidiary have a higher market capitalization. About half of the firms with a foreign subsidiary are in the upper quartile while only 17.27% of the firms without a foreign subsidiary are in the upper quartile. For cross-listing the concentration in the upper quartile of the market size is even bigger. More than 60.08% of the cross-listed firms are in upper quartile of market capitalization while only 11.10% of the firms not cross-listed are in the upper quartile. Finally, for the foreign segmentation sales and profits there is a slightly smaller difference looking at the upper quartile of market capitalization. 35.65% of the UK firms with foreign exposure are in the upper quartile and 19.88% of the firms without foreign exposure are in the upper quartile. In other words, firm size tends to be a significant estimator if a firm has foreign exposure or only domestic exposure.

In total there are 47 Scottish firms without a foreign subsidiary compared to 10 firms with a foreign subsidiary. For listing the difference is equal to 46 Scottish firms which are not cross-listed compared to 11 firms which are cross-cross-listed. For foreign segmentation this difference is the biggest with only 4 Scottish firms having foreign segment sales or profits in the EU. These differences are less pronounced looking at the relative percentages, though it demonstrates that most firms located in Scotland do not have an exposure to foreign countries and this observati on can play a key role in the third hypothesis.

Table 3 shows the distribution of firms based on industry sectors and shows that there seems to be industry concentration for firms with financial and insurance activities. Most firms with financial and insurance activities do not have a foreign subsidiary in the EU. On the other hand, in the manufacturing-sector most firms do have a foreign subsidiary. This observation is the same for the cross-listing variable where also 37.55% of the firms with financial and insurance activities are not cross-listed. The industry segmentation for financial and insurance firms is most pronounced for the foreign segmentation variable. Less than 2% of those firm have foreign segmentation sales or profits, while 43.37% of the firms have not. Also the manufacturing industry-concentration is more present in the foreign segmentation sample. This industry concentration in the two sectors shows the potential of industry being a significant control variable in abnormal returns differences. For the other industry sectors there are also differences in shares but the difference is le ss pronounced than for the financial and insurance and the manufacturing sector.

23 Table 1 – Summary statistics for 981 publicly listed firms in the United Kingdom

This table shows summary statistics for publicly listed firms located in the United Kingdom generated from the Orbis dataset. Firms with no stock data and market capitalization below 10,000 are dropped. Also the book -to-market ratios are winsorized with 5% and operating revenue is winsorized with 0.5%. In the end there remain 981 firms in this dataset. Total assets and operating revenue are measured in units of 1,000 pounds and represent year-end 2015 values. This is because data for year-end 2016 is not available for more than half of the companies yet. Shares outstanding is also based on year-end 2015 and is measured in units. The market capitalization is the most recent value available (27th _{of April 2017) and this measure is available for each firm in units of 1,000} pounds. The average monthly trading volume is measured in units and represent the average monthly volume traded per firm in 2016. The average weekly trading volume represents the average volume traded per week in the three weeks surrounding the Brexit Referendum (13 June – 1 July). Profitability is equal to operating revenue over total assets and one outlier is dropped out of the sample. The amount of subsidiaries counts the number of subsidiaries a particular UK firm owns in the EU. EU-segmentation measures the sales or profits of a British firm in the EU over the total sales or profits of the same firm in 2015 or 2016. Shares below zero or above one are not taken into account.

Table 2 – Distribution of 981 firms based on foreign exposure and market capitalization quartiles This table shows the distribution of the total 981 firms based on market capitalization (measured in quartiles) for the three different foreign exposure characteristics. Market capitalization is measured in units of thousand pounds.

Market Capitalization Foreign Subsidiaries Cross-listing Foreign Segmentation

YES NO YES NO YES NO

256.000 - 30,171,000 16.34% 28.18% 9.35% 31.29% 20.50% 27.26% 30,171,000 – 136,411,000 15.56% 28.31% 11.15% 30.44% 22.40% 26.20% 136,411,000 – 673,483,000 21.40% 26.24% 19.42% 27.17% 21.45% 26.66% 673,483,000 - 128,454,640,000 46.70% 17.27% 60.08% 11.10% 35.65% 19.88% Total 100% 100% 100% 100% 100% 100% Variable N Minimum Lower Quartile Median Upper

Quartile Maximum Mean St. Dev

Total assets 966 174 25,336 110,768 576,100 1,626,058,496 8,845,610 80,006,926 Operating Revenue 930 0.00 4,400 29,251 270,539 39,137,076 943,699 3,607,129 Market capitalization 981 256 30,171 136,411 673,483 128,454,640 1,955,487 8,750,317 Shares outstanding 761 480 44,414 117,791 355,678 71,373,736 600,836 3,109,811 Market-to-book ratio 684 0.34 1.02 1.91 3.27 10.43 2.52 2.06 Profitability 929 0 0.0408 0.4466 0.9871 5.6516 0.6906 0.8447 Amount of subsidiaries 257 1 1 2 5 50 4.38 6.18 EU-segmentation sales or profits over total

sales or profits 317 0.0038 0.1174 0.2381 0.4734 1 0.3487 0.3082

Average monthly

trading volume 979 323 515,715 2,268,687 11,590,516 3,938,863,104 32,343,786 167,629,394

Average weekly

24 Table 3 – Distribution of 981 firms based on foreign exposure and industry sectors

This table shows the distribution of the total 981 firms based on the industry -sector of the firm for the three different exposure characteristics. Note that wholesale and retail trade represents the business of repairing motor vehicles and motorcycles. The res t of the industry sectors is self-explaining.

Industry Foreign Subsidiaries Cross-listing Foreign Segmentation

YES NO YES NO YES NO

Mining and quarrying 5.84% 6.08% 10.43% 4.27% 3.47% 7.23%

Manufacturing 37.74% 18.23% 30.22% 20.63% 42.27% 14.31%

Wholesale and retail trade 8.56% 5.25% 9.35% 4.84% 8.83% 4.82%

Transportation and storage 1.56% 1.38% 2.16% 1.14% 2.52% 0.90%

Information and communication 10.89% 8.01% 9.71% 8.39% 12.93% 6.78%

Financial and insurance activities 6.23% 38.26% 10.43% 37.55% 1.58% 43.37%

Real estate activities 2.33% 3.59% 3.96% 2.99% 0.95% 4.37%

Professional, scientific and technical activities 7.78% 5.80% 5.40% 6.69% 11.04% 4.07% Administrative and support service activities 10.51% 2.35% 5.76% 3.98% 6.94% 3.31%

Other activities 8.56% 11.05% 12.58% 9.52% 9.47% 10.84%

25

5. Results

5.1 Hypothesis 1

The results for the crude dependence adjustment method are presented in Table 4. This table summarizes the results after using the crude dependence adjustment method and is performed for the 981 selected UK firms and the FTSE All-Share Index is used as benchmark. The abnormal returns are estimated and subsequently used in estimating the test-statistics. In Table 4 the average abnormal returns per date are provided together with two test-statistics. The first student-t statistic represents the crude dependence adjustment method and is equal to equation (6). The second student -t statistic is the basic t-test of an event study and is equal to equation (8). This second statistic is added to see the statistical effect of taking into account event clustering.

The average abnormal return columns indicates that on the 24th _{and 27}th _{of June 2016 higher} abnormal returns were observed compared to the other event dates. The average abnormal return is -2.61% on the 24th _{of June and -2.43% on the 27}th _{of June. In both cases the average abnormal returns} are statistically significant at the 1% level with high student-t statistics of -8.596 and -8.000 respectively (TS1). These statistically significant test-statistics support the first hypothesis that there are abnormal stock returns observed after the Brexit.

The TS3-statictics show that when there is not controlled for event clustering, there is overestimation in the test-statistics. For the 24th _{of June the student-t statistic using the basic t-test is} -15.250 and for the 27th _{of June this statistic is equal to -13.817. This supports the importance of using} the crude dependence adjustment method and indicates that the TS1-estimations are robust to event clustering. In addition, there are several other days with significant test-statistics which indicates that on more than two days there was abnormal return performance. This is not supported by the TS1-statistics and proofs the overestimation of the standard errors when not using the crude dependence adjustment method.

Figure 1 shows the cumulative average abnormal returns per date in the four weeks surrounding the event date using the FTSE All-Share Index. The figure shows that on the 24th _{of June} there was a significant drop in the cumulative abnormal returns moving from -1.80% to -4.81%. This drop was even heightened over the weekend on the 27th _{of June when the cumulative abnormal} returns increased further to -7.66%. The cumulative returns slightly recovered in the next week, but decreased again with 1.75% in the days 4 until 6 July. To test for cumulative abnormal return performance two additional tests are performed, again using the crude dependence adjustment method, and the outcome is summarized in Table 5. The first test looks at the whole event window, whereas the second test looks at the cumulative abnormal returns for the two days experiencing significant abnormal return performance (June 24 and June 27) . The basis of the test-statistics are

26 equations (3) and (7). Over the whole event window (10 working days) the ave rage cumulative abnormal return is -4.35% and this is -5.03% for the two-day event window. The cumulative returns are significant at the 1% level for both event windows and therefore cumulative returns have statistical power in demonstrating abnormal return performance surrounding the Brexit Referendum.

To conclude, on average the abnormal returns were negative and statistically significant at the 1% level using the crude depended adjustment method on the 24th _{and 27}th _{of June. This proofs} abnormal return performance around the Brexit Referendum outcome and that financial market were unable to anticipate the short-run effect of the Brexit. Stock markets were thus interrupted by an event of political nature. In addition, performing basic t-tests for the same sample shows that the results are robust to event clustering since the crude dependence adjustment method significantly decreases the overestimation. Finally, also looking at cumulative abnormal returns shows powerful evidence for abnormal return performance using the crude dependence adjustment method.

Table 4 – Test for average abnormal returns in the event window for all firms

In this table the first hypothesis 𝐻0: 𝐸(𝐴𝑅𝑖𝑡) = 0; 𝐻1: 𝐸(𝐴𝑅𝑖𝑡) ≠ 0 is tested against three different significance levels. The estimations are performed with the FTSE All-Share Index as benchmark. 𝐴𝐴𝑅𝑡 is the average abnormal return of all 981 firms in the dataset observed per date t. TS1 is the student-t statistic for the crude dependence adjustment method and equal to equation (6). TS2 represents the normal student-t statistic for abnormal returns and is added to evaluate the performance of the crude dependence adjustment method (equation s (8) and (9)).

FTSE All-Share Index

Date 𝑨𝑨𝑹𝒕 TS1 TS3 20 June 2016 0.0011466 0.378 0.971 21 June 2016 -0.0011328 -0.374 -1.371 22 June 2016 0.0000906 0.030 0.096 23 June 2016 0.0008681 0.286 1.366 24 June 2016 -0.0260567 -8.596*** -15.250*** 27 June 2016 -0.0242504 -8.000*** -13.817*** 28 June 2016 0.0001157 0.038 0.093 29 June 2016 0.0022756 0.751 2.334** 30 June 2016 -0.0002437 -0.080 -0.225 1 July 2016 0.0037223 1.228 4.268*** Significance levels: *** p<0.01, ** p<0.05, * p<0.1

27 Table 5 – Test for cumulative average abnormal returns in the event window for all firms

This table shows the statistical test for cumulative abnormal returns using the crude dependence adjustment method. CAAR is the average of equation (3) and thus the average of the cumulative abnormal returns of all 981 firms over the event window. TS2 is defined in equation (7) and tested against three different significance levels. The tests are performed with the FTSE All-Share Index as benchmark.

FTSE All-Share

Event window CAAR TS2

10-day event window -0.0434647 -3.819***

2-day event window -0.0503071 -4.439***

Significance levels: *** p<0.01, ** p<0.05, * p<0.1

Figure 1 – Cumulative abnormal returns over time for all 981 firms

The graph below shows the average cumulative abnormal returns per date for the whole sample. The figure contains the event window plus one week before and after.

5.2 Hypothesis 2

The second hypothesis states that there exist abnormal return differences for different UK firms listed on the British stock market surrounding the Brexit Referendum. The different samples of firms are selected based on three different indicators of foreign exposure. For the evaluation of this hypothesis first the same steps as in Hypothesis 1 are followed, however now the tests are performed for two different subsamples of firms separately. These two subsamples are determined by the magnitude of foreign exposure each firms has. In addition, a Wilcoxon rank-sum test is performed to estimate if the firms experience statistically significant different abnormal returns. The nonparametric Wilcoxon

rank--.09 -.08 -.07 -.06 -.05 -.04 -.03 -.02 -.01 0 C u mu la tive Ab n o rma l R e tu rn 1 3 ju n 2 0 1 6 1 4 ju n 2 0 1 6 1 5 ju n 2 0 1 6 1 6 ju n 2 0 1 6 1 7 ju n 2 0 1 6 1 8 ju n 2 0 1 6 1 9 ju n 2 0 1 6 2 0 ju n 2 0 1 6 2 1 ju n 2 0 1 6 2 2 ju n 2 0 1 6 2 3 ju n 2 0 1 6 2 4 ju n 2 0 1 6 2 5 ju n 2 0 1 6 2 6 ju n 2 0 1 6 2 7 ju n 2 0 1 6 2 8 ju n 2 0 1 6 2 9 ju n 2 0 1 6 3 0 ju n 2 0 1 6 0 1 ju l2 0 1 6 0 2 ju l2 0 1 6 0 3 ju l2 0 1 6 0 4 ju l2 0 1 6 0 5 ju l2 0 1 6 0 6 ju l2 0 1 6 0 7 ju l2 0 1 6 0 8 ju l2 0 1 6 Date