The effect of a changing shareholder investment horizon and deviation from one share-one vote : deviation to double voting right by law, a natural experiment in France

MSc. Thesis

Title

The effect of a changing shareholder investment horizon and deviation from

one share-one vote

Deviation to double voting right by law, a natural experiment in France

A

This thesis brings together two discussions in the field of Finance and analyzes the claims in a natural experiment in France where in 2014 a new law was implemented. The first discussion is about short-termism, where the debate is about if a short-term horizon is value destroying. The second discussion is about the deviation of the one share-one vote principle. In this case, the deviation is about a reallocation of voting rights among shareholders, moving from one share-one vote to a double voting right after holding the share for a certain amount of time. Three hypotheses are formulated to analyze the effect of moving from a one share-one vote regime in firms in three French indices (CAC40, CACMID60, and CACSMALL). The first hypothesis analyzes firm value, and finds that in the beginning of the sample period (announcement of the law) firm value went up (2%), and at the end of the sample period (implementation of the law) it went down (4%). The second hypothesis analyzes shareholder value, and finds a negative cumulative abnormal return (up to -1,7%) at all events in the legislative process. The third hypothesis analyzes the liquidity and the results show increased liquidity on events throughout the sample period, however overall the liquidity in the sample period seems unaffected.

Maurice Bernards

July, 2016

Name

Date

University of Amsterdam (Amsterdam Business School) 10243801

dr. T. (Torsten) Jochem

MSc. Business Economics (Finance) 20.265

Long-term, Voting rights, Disproportionate, France, Tobin’s Q, Excess return, Amihud measure of Illiquidity

University Student number Supervisor Program Word count Key words

Page 2 of 73

S

O

This document is written by Maurice Bernards 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.

A

I am grateful to Rients Abma, Executive Director at Eumedion, for his invaluable comments, suggestions and insights.

Page 3 of 73

T

ABLE OF

C

ONTENTS

2.2 LONG VS. SHORT TERM PERSPECTIVE 9

2.3 CONTROL ENHANCING MECHANISMS 10

2.4 DISPROPORTIONAL VOTING RIGHTS 11

2.5 OVERVIEW RESEARCH MAIN VARIABLES (IN RELATION TO TOPIC) 11

3 INSTITUTIONAL BACKGROUND AND RESEARCH QUESTION 11

3.1 FLORANGE ACT 12 3.2 RESEARCH QUESTION 15 4 DATA 16 4.1 SITUATION 2012 16 4.2 SUMMARY STATISTICS 17 4.3 MAIN VARIABLES 18 5 METHODOLOGY 19 5.1 REGRESSION SETUP 19 5.2 VALIDITY ISSUES 23 5.3 FINDINGS 24 5.4 ROBUSTNESS TESTS 28

Page 4 of 73 6 CONCLUSION 30 7 REFERENCES 33 8 TABLES 37 9 GRAPHS 50 10 APPENDICES 57 10.1 APPENDIX 1 57 10.2 APPENDIX 2 58 10.3 APPENDIX 3 59 10.4 APPENDIX 4 60 10.5 APPENDIX 5–I 61 10.6 APPENDIX 5–II 62 10.7 APPENDIX 5–III 63 10.8 APPENDIX 5–IV 64 10.9 APPENDIX 6 65 10.10 APPENDIX 7-I 66 10.11 APPENDIX 7-II 67 10.12 APPENDIX 7–III 68 10.13 APPENDIX 8-I 69 10.14 APPENDIX 8-II 70 10.15 APPENDIX 8–III 71 10.16 APPENDIX 8-IV 72 10.17 APPENDIX 8-V 73

T

G

Graph 1 - Google Trends for law _____________________________________________________________________ 50 Graph 2 - Government shares & voting rights ___________________________________________________________ 51 Graph 3 - Adjusted Tobin's Q _______________________________________________________________________ 52 Graph 4 - Cumulative Abnormal Return _______________________________________________________________ 53 Graph 5 - Amihud Measure of Illiquidity _______________________________________________________________ 54 Graph 6 - CAR and AI: Air France - KLM _____________________________________________________________ 55

Page 5 of 73

Graph 7 - CAR and AI: Renault ______________________________________________________________ 56

T

T

Table 1 – Research overview main variables ____________________________________________________ 37 Table 2 - Overview of the Florange Law (2014-384) ______________________________________________ 38 Table 3 - Legislative timeline of the Florange Law ______________________________________________ 39 Table 4 – Situation 2012 ____________________________________________________________________ 40 Table 5 - Summary statistics key variables _____________________________________________________ 41 Table 6 - Summary Statistics Q ______________________________________________________________ 42 Table 7 - Summary Statistics AI______________________________________________________________ 43 Table 8 - Variable definitions of the D-in-D regressions ___________________________________________ 44 Table 9 - Regression Q _____________________________________________________________________ 45 Table 10 - Regression CAR _________________________________________________________________ 46 Table 11 - Regression AI (1) ________________________________________________________________ 47 Table 12 - Regression AI (2) ________________________________________________________________ 48 Table 13 - Regression AI (3) ________________________________________________________________ 49 Table 14 - Correlation Q / AI and controls ______________________________________________________ 57 Table 15 - Summary statistics _______________________________________________________________ 58 Table 16 - Overview of share capital __________________________________________________________ 59 Table 17 - Summary statistics _______________________________________________________________ 60 Table 18 - Summary statistics (total and per index, for 2012) _______________________________________ 61 Table 19 - Summary statistics (total and per index, for 2013) _______________________________________ 62 Table 20 - Summary statistics (total and per index, for 2014) _______________________________________ 63 Table 21 - Summary statistics (total and per index, for 2015) _______________________________________ 64 Table 22 - Robustness Test Q _______________________________________________________________ 65 Table 23 - Robustness Test CAR(I) __________________________________________________________ 66 Table 24 - Robustness Test CAR(II) __________________________________________________________ 67 Table 25 - Robustness Test CAR(III) _________________________________________________________ 68 Table 26 - Robustness Test AI(I) ____________________________________________________________ 69 Table 27 - Robustness Test AI(II) ____________________________________________________________ 70 Table 28 - Robustness Test AI(III) ___________________________________________________________ 71 Table 29 - Robustness Test AI(IV) ___________________________________________________________ 72 Table 30 - Robustness Test AI(V) ____________________________________________________________ 73

Page 6 of 73

1

I

NTRODUCTION

Ever since the 2008 financial crises and the following global recession, there has been a growing concern about companies being too short-term focused (Barton and Wiseman, 2014). The short-term focus emerges from the “quarterly capitalism”, which is the pressure from the financial markets on public companies to maximize term profits. Attempts to make corporate leaders to abandon their short-term perspective during the last decade have been fruitless. Barton and Wiseman (2014) also argue that if the short-termism continues, it would harm economic growth and lower returns for savers.

Bruno le Roux, leader of the Socialist Group in the National Assembly in France, also shares this view. He introduced a law, named “Loi Florange”, which translates to Florange law, to (amongst others) counter short-term perspective among shareholder in France. The French investors are used to being able to invest on the long term, it has been possible ever since dual-class shares had been prohibited in 1933. French firms are able to include a clause in their bylaws that enables long-term investors to be granted a double voting right after a certain period of time (typically 2 years). Le Roux, however, intensified the fight against short-termism. From April 2014 onwards, all investors whom have held their shares (in registered form) for more than 2 years are granted a double vote per share they own. Only if two-thirds of the investors agree to go back to the old situation (one share-one vote) is it possible for firms to opt-out of the provision. The situation changed from opt-in to opt-opt-out.

There is a discussion, however, that debates the very existence of short-termism, as in this setting it can have one of two meanings. One is shareholder myopia, stating that shareholders only look at the short-term movement of share prices. This is a very plausible consideration as the average holding period of stock has dropped from just over 2 year in 1990 to half a year in 20101. The other view is management myopia, which entails that, in decision making, managers have a cognitive inability to incorporate the temporal dimension of decision making. Counterarguments, however, state that the efficiency of markets is optimizing the trade-off between short-term and long-term investments. Additionally, empirical evidence challenges several claims made for short-termism.

The aim of this thesis is to investigate whether the claims that come forward from the abovementioned debate can be observed in a natural experiment in France. Moreover, an extra dimension is added in the form of another debate in corporate governance, the deviation from the one share-one vote regime. As already mentioned, the French government introduced a law, creating a situation that allows for these two debates to be analyzed together. The main research question in this thesis is: Does the sudden switch

Page 7 of 73

to a long-term perspective among shareholders affect firm value, shareholder value, and liquidity in France due to the implementation of the law? It is not new for French firms to adopt an alternative to the one-share one-vote policy, around two-third of the firms have already incorporated this in their bylaws, far before the Florange Act was established. This change could be done to increase the long-term perspective among shareholders, giving them one more vote per share owned after a certain amount of time, in the French case it is at least two years. The concept behind the deviation from a one-share one-vote to create a long-term perspective is called loyalty shares. The in 2014 established Florange Act changes the default situation of firms in a way that instead of opting in, firms now have to opt out of the deviation from one-share one-vote policy. Due to this sudden change, the expectation is a shift in time horizon among investors to a more long-term perspective, and hence affect firm value, shareholder value, and decrease liquidity (change in holding period).

In order to answer the research question, three hypotheses are made. The first hypothesis, firm value hypothesis, states that firm value decreases due to the implementation of the law. Here, the Adjusted Tobin’s Q (Q) is used to approximate firm value. The second hypothesis, shareholder hypothesis, states that shareholder value increases due to the implementation of the law. Here, the cumulative abnormal return (CAR) is used to measure shareholder value. The third hypothesis, liquidity hypothesis, states that the liquidity will decrease due to the implementation of the law. Here, the Amihud measure of illiquidity (AI) is used as an approximation of liquidity. The hypotheses take a stance on the direction of the effect based on expectations from previous research done in one (or both) the debates mentioned above. However, there is little research performed with both deviations from one share-one vote and short-termism, and existing research shows mixed results, making ex-ante expectations hard to formulate. Both firm value and shareholder value are considered in answering the research question. Even though both are strongly related to each other, firm value also takes book values into account, whereas shareholder value only looks at the changes in market value.

Firms are split in two groups, the control and treatment group. The control group consists of firms on which the law should have no effect (double voting right already in place), the treatment group consists of firms with a one share-one vote policy at the beginning of the sample period. As expected, the outcomes of this thesis are mixed. Firm value for the treatment group over the whole sample period seemed unaffected. However, when dividing the sample period into two sub periods, firm value increase in the beginning (2%) and then decreases at the end (4%) of the sample period. Around the events considered in the legislative process there are negative cumulative abnormal returns, indicating that around the releases of new information considering the law, firms in the treatment group showed a

Page 8 of 73

decrease in shareholder value (up to 1,7%). Liquidity over the whole sample period seems unaffected for the treatment group, however in the beginning of the legislative process the firms in the treatment group experienced a negative liquidity shock (up to 45% less liquid). To check for robustness, firms in the treatment group are again split according to the situation in 2016 (opt-out of the law in between 2014 and 2016 versus holding on to the double voting right). The results for firm value and liquidity do not alter greatly. Shareholder value (CAR) does change, firms that do hold their double voting provision (enforced on them by the law) experience a positive CAR around the events in the legislative timeline, versus a negative CAR for the overall treatment group.

This thesis starts with an overview of the literature in section 2. The institutional background and the research question are elaborated on in section 3 and the data used is discussed in section 4. The methodology and results are presented in section 5. Section 6 ends with the conclusion and is followed by the references to literature in Section 7. Tables and graphs are found in Section 8 and 9, respectively. Appendices are in section 10.

2

L

ITERATURE

To better understand the outcome of the research and to place this thesis among existing literature, various subjects covered in (finance) literature that are touching the subject covered in this thesis are elaborated upon. This thesis is a natural experiment of the decoupling of voting rights and economic interest.

2.1 S

The one share-one vote debate, and corporate governance in general, is focused on a fundamental dilemma, namely whether entrenched owners or contestable managers are more prone to run firms efficiently (Becht et al., 2003). Large owners have the incentive to maximize value (due to a large share in the firm’s equity), but they are more immune to hostile takeovers. Contestable (professional) managers are more exposed to hostile takeovers but also have less economic interest in the firm.

Shareholder empowerment in this case focuses on the deviation from one share-one vote, and the consequences thereof. In essence, a one share-one vote means aligning voting power with economic incentives. There are benefits to the one share-one vote principle. According to Burkart and Lee (2008), it is an enabler for good governance as it makes a level playing field for takeover contests, which ensures control is allocated to the most efficient party. They continue with arguing that the allocation of voting right across shares matter, as it determines the balance of power among shareholders and their leverage over management.

Page 9 of 73

There are, however, also benefits to deviating from the one share-one vote regime. When there are block holders, for example, leveraging the voting power improves its ability to monitor and intervene in management. At the same time it also carries a disadvantage, namely it also allows for it to take self-serving actions (Burkart and Lee, 2008).

2.2 L

.

Besides the one share-one vote debate there is also an ongoing debate about short-termism (versus a long-term horizon). An important issue with short-termism is that it is (largely) inconsistent with the efficient market hypothesis (EMH). If indeed all new information is incorporated in the share price without delay and hence no above-average returns can be realized (Malkiel, 2005), then there would be no discussion of short-termism. Today’s prices should reflect long-term performance and as a result investors should encourage long-term development of a firm as it increases the share price today. Moreover, there is empirical evidence challenging the claims made from short-termism (Hansen & Hill, 1991).

Many claim that short-termism does indeed exists (example: Fried, 2014 or Miller, 2006) whereas others say the opposite (Example: Woolridge, 1988 or Jensen, 1988). The extent to which short-termism is problematic is how choosing for short-term profit happens at the expense of the long term. In other words, according to Laverty (1996), the intertemporal tradeoff is suboptimal. Conversely, taking actions in favor of the long-term and at the expense of the short-term is also problematic (Marginson and McAulay, 2008). However, this will not be dealt with in this thesis.

An often used term to describe short-termism is myopia. Myopia can refer to both management and shareholders. Samuel (2000) describes shareholder myopia as “the tendency of shareholders to focus on the behavior of stock prices in the short term as opposed to the long term”. Miller (2002) describes managerial myopia as the cognitive inability to see the relation between short term and long term in decision making. Also, managers opting for short-term profits instead of long-term profits is also a problem of short-termism. Compensation packages can be focused on short-term profits, or pressure from shareholders can cause this behavior. Therefore, the latter can also be seen in relation to the shareholder view of myopia. In this thesis, both types are used when talking about short-termism.

A company’s long-term value is determined by its ability to generate cash flows to fund value-creating growth and pay dividends (Rappaport, 2005). However, investment decisions are commonly based on short-term metrics, such as earnings, instead of a long-term valuation such as a DCF valuation. One reason why earnings are a commonly used metric for valuation is that estimating cash flows in the future

Page 10 of 73

is costly, speculative, and time-consuming. Due to this informational imperfection (Dallas, 2011), short-term metrics such as earnings are used, in favor of long-short-term models such as a DCF valuation.

There are advantages and disadvantages of using short-term or long-term models. One advantage of using earnings as a valuation is that earnings follow the market’s pricing. When using a DCF model for valuation, the current market price can deviate substantially from the long-term value of the firm (Rappaport, 2005), which does not happen when the valuation is based on earnings. Another disadvantage of using a model with a long-term horizon is that stock typically have a short average holding period, and in the short run shares can deviate substantially from their long-term value. Today, a professionally managed fund has an average holding period of less than a year (Rappaport, 2005).

There are also advantages of investing in the long term. First, managers serving short-term investors may destroy economic value (Fried, 2014). Also, in a study among family businesses, Brenton-Miller and Miller (2006) found that family businesses out-compete other firms due to their long-term perspective.

However, it is not clear that a long-term perspective actually creates value in terms of a higher share price. Bushee (2004) finds that in companies with a high percentage of long-term investors, volatility goes down, but does not look at share prices. Fried (2014) finds that, for US firms, managers serving long-term investors do not necessarily create more value than managers serving short-term investors. Nonetheless, a long-term perspective is considered better than a short-term perspective, a view held by leading academics, executives, lawyers and judges (Fried, 2014).

2.3 C

E

M

Control enhancing mechanisms (CEMs) allow for a deviation of the one share-one vote principle by disproportionate ownership (Adams and Ferreira, 2008). Deviations from the one share-one vote principle has been researched at various levels, at the micro level it can lead to distorted investment decisions and at macro level it can lead to underdeveloped capital markets and retardation of growth (Morck et al., 2005). However, there are benefits. Too much control in the hands of managers may lead to entrenchment (and deterioration of value), but some control may have an advantage because it can shield management from the appropriation of the returns to organizational-specific investment by other management teams (Fama, 1980).

There are many forms of disproportional ownership. Besides shares with differential voting rights, there are voting rules and caps, voting agreements, pyramid control structures, cross-ownership of

Page 11 of 73

shares, fiduciary voting, owner dispersion, etc. The CEM analyzed in this thesis are the so-called loyalty shares.

The loyalty share is one of many control enhancing mechanisms available to attract more long-term investors. In essence, the loyalty share entitles the owner of a share more rights (cash flow or voting) after a predetermined period of time (Fried, 2014) in the form of, for example, a call-warrant (Bolton and Samama, 2014). Using the loyalty share, the investor would only gain the extra right after, for example, 2 years. This way, investors are incentivized to hold the stock for longer periods of time, turning short-term into a long(er)-term perspective. An important aspect of the loyalty share is that only the behavior of the investor ultimately determines the ownership of the loyalty part of the share. The Florange act is based on disproportionate voting rights, allowing for a double voting right for shareholders who have had the shares in registered form for (at least) two years.

2.4 D

Deviating from the one-share one-vote principle disturbs the free market for takeovers (long-term shareholders have a more decisive role in determining whether the company will engage in M&As). This results in a disadvantage, the cost of capital will rise due to the added protection. This fact has also been proven shown by Himmelberg, Hubbard and Love (2004). They state that the risk premium for shares with added takeover protection is zero to five percent. More generally, they state that the higher the concentration of inside ownership (i.e. large shareholders), the higher the cost of capital. While their research focused on firms with a one-share one-vote policy, the intuition can be extended to the case where there are disproportionate voting rights.

Effectively, under the double voting rights in France, new shareholders are stripped of their voting rights. One share (held for less than two years) has proportionally less voting right compared to the old situation of one-share one-vote. This translates in the share price, non-voting shares (in general, not only in France) often trade at a 5-10% discount (The Economist, 2015). The literature that also covers this is discussed in Table 1.

2.5 O

(

)

[Insert Table 1 here]

3

I

NSTITUTIONAL BACKGROUND AND RESEARCH QUESTION

The aspect of corporate governance in France that is of interest in this thesis is the double-voting right. The provisions for the double voting right dates back to 1933 as a reaction to the prohibition of

dual-Page 12 of 73

class shares (Conac, 2005). The double-voting right allows for an additional vote to be granted after a typical holding period of 2 years, but it can be longer. A very important aspect is that the double vote is not attached to the share, but only granted to the holder and hence it cannot be transferred. Consequently, the shares are ordinary shares, without any special classifications.

In essence, the double voting right system is similar to a dual-class structure, consisting of ordinary shares with each one vote, and shares with two votes each. Also, like dual-class shares, the double-voting shares can be used to consolidate an incumbent’s control and leverages the incumbent in control contests (Lannoo, 1999). However, even if the incumbent is willing to let go of its control, the double-voting rights may impair takeovers. As the double votes are lost in a transaction, the bidder may not have the same amount of votes that was paid for, leaving the incumbent not able to ensure a successful takeover. Additionally, in the case of France, where there is a mandatory bid rule2, the bidder has to make an offer for all shares. However, price-discrimination is not allowed as both shares with a single and double vote are the same class of shares. This forces the bidder to offer a control premium to small shareholders (Burkart and Lee, 2008). In effect, the mandatory bid rule enlarges the entrenchment effect of the double-voting right shares.

French publicly listed firms could opt-in the double voting right to keep the influence of institutional investors to a minimum. Following Burkart and Lee (2008), this category of investors were thought to a have short holding periods, and hence pressuring managers to focus on the short-term profits, presumably at the cost of long-term profitability. Various research have made such allegations against activist hedge funds (example: Becht et al. 2010; Kahan and Rock, 2006), however the opposite is shown in other studies (example: Edmans, 2007).

3.1 F

A

The official name of the Florange Act is 2014-384. The law aims to reclaim the real economy (translation from: “Loi visant à reconquérir l'économie réelle”). It was designed as an anti-hostile-takeover measure and consists out of three levers. The first lever makes it obligatory for an acquirer to seek a buyer for a factory before deciding to shut it down. This way, the workers of a target firm have more protection. The second lever is a provision for employee activism. These two levers do not affect the control and treatment group differently of the firms investigated in this thesis. The third level consists of the provisions for the double voting right. Table 2 provides a more comprehensive description of the law and its levers.

Page 13 of 73

[Inset Table 2 here]

The law was spurred by the hostile takeover of Arcelor by Mittal Steel in 2006, which subsequently closed a plant and laid off 600 workers. According to Benatia and Gomez (2015), the government claimed the law was a response to the deindustrialization of the economy (in benefit of financial interests) in the last decade, during which the real economy suffered. For this thesis, the third lever is of interest, as it states that double voting rights are attributed to shareholders that hold their shares in registered form for at least two years. This policy is not new in the French legislation, however the Florange act reverses the system from opt-in to opt-out. The exception has become the rule. Firms who wish not to have the double voting-right can opt-out by having at least two-third of the shareholders vote against the policy (and the one share-one vote policy remains). An overview of important dates regarding the legislative process is given in Table 3. Along with the dates, the description, source and probability of the law passing is included. In case of good news related to the law, the probability of the law passing increased. In case of bad news, the probability of the law passing decreased.

[Insert Table 3 here]

To include the search trends on the internet for the law, input from Google3 trends indicates that the month March in 2014 and April in 2015 are of importance. Graph 1 indicates the search tends on Google. Interestingly, before the end of 2013, there was limited search activity for the law. This can indicate that the law was not yet widely known and/or not covered much in the news.

[Insert Graph 1 here]

According to The Economist (“Short-term or Short-changed?”, 2015), 22 out the 40 firms on the CAC already have a clause to indicate that they deviate from the one-share one-vote policy before the 1st _of April, 2014. However, the law applies to all French publicly listed firms, therefore not limited to the 40 in the CAC-40 index. According to Institutional Shareholder Service (2007), in a sample of 40 firms (see Appendix 3), 58% of the French firms use multiple voting rights (23 out of 40).

In France, there is a discussion as to whether this approach by the government will have the intended effect, namely to promote the long-term perspective and hence increase firm value. Gaffard and Iacopetta (2015) argue that this intended effect carries another advantage, yet only for the government. They say that, due to this law, the French government can reduce its public debt without losing influence in the companies in which it is shareholder. See Appendix 2 for an overview of the share and voting right the

3 _{According to Netmarketshare Google global marketshare (Netmarketshare, 2016), Google is the largest search engine in}

Page 14 of 73

government had in the firms it invested in. Moreover, they argue that the law will have no impact in reviving the French economy.

The Economist (2015) also highlights two disadvantages of the law. The first disadvantage of disproportional voting rights is that the law will add more power to an elite of insiders who are known for a mediocre record (in France). On the one hand, a very powerful investor allows for a focus on the long-term and monitoring of the management. However, as stated in the same article and shown by Maury and Pajuste (2005), firms can underperform if insulated from outside threats. Large shareholders can engage in the pursuit of private goals that differ from profit maximization or they can reduce valuable management incentives.

Many companies have announced to opt-out of the law immediately (in the two-year transition phase). As stated in a publication by Benatia and Gomez (2015), 11 firms have filed motions to maintain the one-share one-vote policy (as of 2015). Moreover, they say that the French government uses its share in various public firms to promote the introduction of the double voting rights, even increasing its share to make sure the double voting policy is implemented.

The effect of the law are examined using the three main variables found in section 2.5 (Table 1). The effect on firm value and shareholder value follow from the debate on the implementation of the law (reclaiming the real economy, as stated at the beginning of this section). Real-economy measures are not discussed in this thesis, though, financial measures are affected by the economic outlook of the country. One reason for investigating liquidity is that a long-term horizon among shareholders will result in less trading (shareholders want to keep their shares at least 2 years). However, there is another reason why liquidity might decrease as a result of the law implemented in France. According to Attig et al. (2006), a larger shareholder (or block holder) can benefit from insider information and thus trade accordingly and extract the private benefits of control. This will also lead to lower trading by other shareholders, knowing that there is a block holder with more information. This results in a wider bid-ask spread and thus implies a lower liquidity. This effect can be amplified by the implementation of the law as a large and long-term shareholder or block holder gains more control, resulting in less demand from smaller investors, and hence lower liquidity. Generally (and keeping other factors constant), a larger deviation of control (voting right) and ownership (shares) results in a lower liquidity (Ginglinger & Hamon, 2012). However, according to Ginglinger & Hamon (2012), there are reasons for the liquidity to increase. Insiders in the firm (often long-term shareholders) want to hold on to their superior voting right and therefore are less likely to trade. Consequently, the probability of an uninformed trader to encounter an informed trader decreases, which, in turn, should increase liquidity. In the same article, in a sample of

Page 15 of 73

French firms and in a sample taken from 1998-2002, they found that a double voting right decreases the spread. Though, the effect is limited to small family-firms.

3.2 R

Q

This thesis attempts to analyze the situation in France along the following research question:

Does the Florange Act in France decrease firm value, increase shareholder value, and reduce liquidity due to a default-situation switch to a more long-term perspective among shareholders? To answer this question and based on the literature discussed above, three hypotheses have been formulated. Each are briefly discussed below.

3.2.1 Firm value hypothesis

Based on the research done by Maury and Pajuste (2005), the expectation is that the firm value will decrease due to the new allocation of voting right among the shareholders. When voting power is distributed more equally, firm value (measured as Tobin’s Q) increases. For firms with one or more block holders, a separation of votes and equity increases the extraction of private benefits compared to similar one-share one-vote firms. La Porte et al. (2002) found that there is a higher valuation of firms in countries with better minority shareholder protection and higher cash flow ownership by controlling shareholders. The first is not the case in France, as the law does the exact opposite. It protects the majority shareholder by giving them more voting right. Lastly, Cronqvist & Nilsson (2003) show that there is a negative relationship between vote ownership by controlling owners and firm value (estimated agency cost of control is 6-25% of firm value). Based on the above, the hypothesis is formulated as: Firm value will decrease due to the (announcement of the) implementation of the law.

3.2.2 Shareholder value hypothesis

Based on the research done by Smith and Amoako-Adu (1995), superior voting shares sell at a premium relative to their counterpart restricted (normal) shares. Also, Kunz & Angel show that shares with superior voting rights trade at significantly higher prices. Additionally, control-block votes are valuable in countries of French legal origin. Large shareholders are willing to pay for votes due to expected benefit of control (Nenova, 2003). In contrast to the literature discussed above, King and Santor (2008) argue that dual-class shares perform worse compared to firms with a one share- one vote policy (17% lower valuations). As the findings out of other literature are ambiguous, an empirical analysis will be needed to answer this question. Due to more arguments why it should increase, the hypothesis will be stated as follows: Shareholder value will increase due to the (anticipation of the) implementation of the law.

Page 16 of 73

3.2.3 Liquidity hypothesis

Intuitively, if the law is successful at creating a long-term horizon among investors, it will result in less trading and therefore the liquidity will decrease. A similar conclusion is also reached in other literature, such as the research done by Becht (1999), he finds that dual class shares with multiple voting rights have a small negative impact on liquidity. Attig et al. (2006) shows that shares with greater deviations between ultimate control and ownership have a larger information asymmetry component of the bid-ask spread. In other words, due to a deviation from the one share-one vote regime the liquidity decreases. Based on this, the following hypothesis is formulated: Liquidity will decrease due to the (announcement of the) implementation of the law. For interpretation purposes, it is important to note that liquidity is measure using a measure for illiquidity, therefore an increase in illiquidity means a decrease in liquidity.

4

D

ATA

The sample consists of firms from the three indices on the Paris stock exchange (CAC40, CACMID60 and CACSMALL). In total, there are 309 firms in the index. Share information is taken from Compustat (common shares and treasury shares) and the ownership data is hand collected from annual reports and for state ownership the data is retrieved from official state documents. Information about the voting results from the firms in the treatment group is also hand collected from the voting disclosures and minutes of the General Meetings (GMs) during 2014 and 2015. Firms in the indices are excluded that have had any corporate activity during the sample period (IPO, M&A, delisting), are listed outside of France (Brussels and Amsterdam), or have no publicly available information about share ownership in the sample period. After the exclusion restrictions the sample remains with 238 firms.

4.1 S

2012

Table 4 sketches an initial picture of the situation in France before there was any mentioning of the law (both in the media and through official channels). As there was a first mentioning of the law in 2013, the table shows the situation at year-end of 2012. One variable has been created to describe the data, the Double-voting ratio, which is the ratio of double voting rights to the single voting rights. The total number of single voting right shares is calculated by subtracting the treasury shares (no voting right) from the common shares outstanding. The years after which a double voting right is granted is called the Loyalty Years. See Appendix 4 for an extensive overview of these summary statistics per index. In total, 67% of the firms have a double voting right clause in 2012 (159 out of 238 firms in the sample), before the law was implemented.

Page 17 of 73

The highest amount of firms with a double voting right clause is in the CACSMALL, where 72% of the firms have double voting right clause, compared to only 51% in the CACMID60 (for detailed overview, see Appendix 4). The findings in this sample are in line with findings in other literature. The double voting ratio also shows an interesting pattern. The biggest listed firms in France have a relatively low double voting ratio, meaning that the amount of double voting rights compared to single voting rights is relatively low. The smaller the firms, the higher the ratio. A possible reason for this is that smaller firms have a higher concentration of shareholder (more large shareholders), and therefore have more long-term shareholders.

[Insert Graph 2 here]

In the sample of 238 firms, there were 7 firms that had a high government stake (+5%). The French government has stakes in firms for social and national strategic reasons. Graph 2 shows the share the government has and the voting right that is associated with that share (the dashed line shows the equal distribution of voting rights). The graph shows that the government’s unequal voting right is distributed around the dashed line, with the majority just above the line. However, especially above the line, the deviation from the line is not that great. See Appendix 4 for an overview of the government stakes throughout the sample period.

No claims from the literature can be confirmed just by looking at these statistics. In the transition period (between 2014 and 2016), the French government was not active in some companies, whereas they were very active in others, increasing the government’s stake to ensure the disproportional voting right4. See section 5.5 for a brief case study of the companies Air France-KLM and Renault around the GM in 2015 concerning the CAR and AI. During the GMs, a proposition was brought forward by the board to continue under the one share-one vote regime, but due to an increased share in equity and opposing views by the government, this proposition was blocked.

4.2 S

S

The summary statistics for the key variables in the sample is given in Table 5. See Appendix 5 for a more detailed summary statistics overview for the entire sample period.

[Insert Table 5 here]

Among the variables not displayed above, closing price and trading volume have been winsorized (both at percentile 10 and 90).

4 _{For example Renault and Air France-KLM, see (Meichtry and Landauro, 2015) for a more detailed case analysis. Also, see}

Page 18 of 73

4.3 M

4.3.1 Adjusted Tobin’s Q

[Insert Table 6 here]

The dependent variables Q summarized in Table 6 and is displayed graphically in Graph 3, where per date the cross-sectional mean of the one-year moving average is taken. Throughout the sample, the average overall Q has a positive trend, starting at 1,29 in 2012 and rising to around 1,43 at the end of 2015. A remarkable finding appears when the sample is split in the treatment and control group. As can be seen in the graph, the treatment group (no double voting right clause in 2012) has a higher Q compared to the control group. At the end of the sample period, the two converge slightly. However, the Q of the treatment group shows a decreasing trend from April 2014 onward, while the Q of the control group shows a slight positive trend. For this reason, two additional ranges are included in the regression specifications, to investigate whether these movements are not based merely on firm specific characteristics. These ranges are specified in Section 5.3.1.

[Insert Graph 3 here] 4.3.2 Cumulative Abnormal Return

[Insert Graph 4 here]

Graph 4 shows the cumulative abnormal return (CAR) for every event date (daily events) as described in Table 2. First, the event where P[Law Pass] is expected to increase, and hence, the cumulative abnormal return for the treatment group is expected to be positive, only shows on the event on 24th of February 2014. Conversely, where the P[Law Pass] is expected to decrease, and hence, the CAR for the treatment group to be negative, also is not visible in the graphs. When the law is implemented (1st of April, 2014), there is a positive CAR for both groups. The day before the definitive compliance date (April 1st, 2016) there is a large positive CAR, which decreases in the days thereafter. An interesting find in the Graph 4 is that both groups seem to move parallel. Only on the definitive compliance date do the two lines move apart for a prolonged period in the event window.

The cumulative abnormal returns in Graph 4 are without any controls. Expectations are that the excess return of the treatment group on and after day 0 would be positive (in case of a positive event) or negative (in case of a negative event). As indicated by other literature, and at first glance, there is no clear link between (an expected) switch to a long-term perspective among shareholders and the CAR.

4.3.3 Amihud Illiquidity

Page 19 of 73

Table 7 summarizes the liquidity in the sample period, Graph 5 presents it graphically. Intuitively, and based on previous research, the liquidity of firms in the treatment group should be lower compared to the control group. Throughout the whole sample, the illiquidity of the control group (with the double voting right in place in 2012) is higher compared to the treatment group (i.e. liquidity is lower). This confirms that firms with a double voting right clause have lower liquidity compared to firms without. Also, throughout the sample, the two groups grew closer together. There is a distance that remains at the end that can be explained by, in the treatment group, various firms have opted-out of the double voting right and remained in the treatment group as a single voting right company (more on this in section 5.4). Moreover, there is a negative trend in the liquidity for both groups in the sample period.

[Insert Graph 5 here]

5

M

ETHODOLOGY

5.1 R

To capture the market reaction, an event study is used. This event study takes the form of a difference-in-difference (D-in-D) specification, with either the cumulative abnormal return (CAR), Adjusted Tobin’s Q (Q) or the Amihud measure if illiquidity (AI) as a dependent variable. A D-in-D method is generally used in event studies to examine the effect of an event (or treatment) in the period following the event. An important reason for using the D-in-D method is it bypasses some of the endogeneity problems often encountered when comparing two heterogeneous groups (Meyer, 1995).

5.1.1 Cumulative Abnormal Return (CAR)

The following formula shows the computation of the CAR.

CARit = ∑(Rit - R̅i) (1)

The above calculations are in the event window, ranging from day -2 through 2 around an event date. The entire window has 261 days (one year), separated in the estimation window (-256 through -3) and event window (-2 through +2). The average return is calculated in the estimation window.

R̅_{i =} 1

254∑ Rit (2)

The cumulative abnormal return is calculated in the event window. The excess return is not calculated using market-based variables because this will generate biased values of excess return. The sample consists of the three main market indices in France, therefore the excess return moves with the market return, creating unreliable and biased excess return values.

Page 20 of 73

5.1.2 (Approximation of) Adjusted Tobin’s Q (Q)

Q_it=BVit Assets + MVit Common Stock - BVit Common Stock - Deferred Taxesit

BVit Assets (3)

Where BV is “book value” and MV is “Market Value”. The book value for a given date t is the year-end value of the prior fiscal year. The market value of common stock can be computed daily.

For testing the hypothesis, the one-year moving average is created: Q̅

i = 1

261∑ Qit (4) 5.1.3 Amihud Measure of Illiquidity (AI)

The Amihud Measure of illiquidity is calculated using the following formula. AIit=

|r_it|

volit (5)

Where 𝑟𝑖𝑡 is the daily return of share 𝑖 at time t. The currency denoted volume is volit, measured in euros. To smoothen the values, a moving average of the illiquidity measure is made for testing the hypotheses. AI for testing the hypothesis for entire sample period, using a one-year moving average:

AI ̅̅̅_{i =} 1

261∑ AIit (6) AI for testing the hypothesis for a monthly date range, using a one-month moving average:

AI ̅̅̅_{i =} 1

20∑ AIit (7) AI for testing the hypothesis for daily events, using a one-week moving average:

AI ̅̅̅_{i =} 1

5∑ AIit (8)

Defining the dates for the event study might seem obvious, however it is not. The issue here is not whether the law was implemented, but when the market could have anticipated the news (Henderson 1990). Critical moments have been identified in the event timeline to show the anticipation of news, and therefore when the probability of acceptance of the law P[law pass] would increase or decrease, see Table 2. As these dates should have some an impact on investors’ view on firms, they have been marked as critical dates for the event study. For each of these dates, a D-in-D regression will be done to analyze the effect, with date ‘0’ being the critical date. The set-up of the event study will follow the methodology described in Brown and Warner (1985).

Different approaches for the event study have been chosen to test each hypothesis. To test the firm value hypothesis, the one-year moving average of Q has been created. For the event, and for each firm in the sample, two values of Q are used, one for the ex-ante and one for the ex-post date. Three regression will be done, the first one covers the whole sample period, the ex-ante date is the first business day of

Page 21 of 73

May 2012, the ex-post day is the last business day in December 2012. This will be split up, creating one regression that covers period May 2012 until Sept 2014, and another on that covers Sept 2014 until Dec 2015. The last two periods have been chosen based on the visual representation of Q in Graph 3. In total three regressions will be done to test these hypotheses.

The test for shareholder value is done using estimation windows and event windows (Henderson 1990). In the estimation window, the average return in the year leading up to the event (day -256 through -3) is estimated. The abnormal return is calculated using the return in the event window (-2 through +2) minus the average return from the estimation window. Finally, the return are aggregated per firm to form the cumulative abnormal return. The cumulative abnormal return in the regression is the sum of the abnormal returns in the event window. For each firm, there is one ex-ante and one ex-post observation. In total seven regressions will be done to test this hypothesis.

Lastly, the test for liquidity is separated into three parts. For each event, two values of AI per firm are entered into the regression. The value of AI for the first category, the five-day moving average, one value is at day -1 and the other is at day +4. This way, the average AI before the event is compared to the average after the event. The second category is the monthly range, which is a 20-day moving average, where the date ex-ante is -1 and ex-post is 19. One separate regression is done for the entire sample period which uses a 261-day moving average for AI. In this regression, the ex-ante date is May 1st, 2012 and the ex-post date is May 1st, 2016. In total 11 regressions will be done to test this hypothesis.

Control variables have to be entered for each group of regressions. Different control variables are identified for each dependent variable. For the Q, control variables include controls for firm size, leverage, industry, growth prospects (Black et al., 2005), risk (Dowell et al., 2002) and a dummy for an indication of the index (CAC40, CACMID60 or CACSMALL) to control for visibility. For the CAR, control variables include variables that control for firm characteristics (leverage, Q, market-to-book, and return on assets), a dummy for an indication of the index. Variations among companies, such as size, shares outstanding and stock levels are controlled for by the mean adjusted returns used to calculate the CAR (Charles, 2010). For the AI, control variables are market-to-book ratio, a dummy for an indication of the index, and size. Additionally, as AI is correlated with previous values of illiquidity, as lagged variable of AI is included as a control. For each specification, another control variable has to be entered if the French state has an interest in a company, due to the alternative interests of the state (percentage of total equity owned by the state). One of these alternative interests might be the pursuit of social value, instead of shareholder value, at the cost of other existing shareholders. All controls have been checked for correlation, see Appendix 1 for an overview for all three dependent variables. To allow for cross

Page 22 of 73

correlation on the same date, but independence by date, robust standard errors are clustered by date for Q, CAR and AI, as done by Lacker et al. (2011).

Below are the regression specifications. The regression method used is a D-in-D analysis. The D-in-D looks at the change between ex-ante and the ex-post variables and between the treatment and control group. The ex-ante values are values before an event and the ex-post values are after an event. The event differs per regression, each event is specified in the regression output tables. The treatment group consists of the firms that have not adopted a double-voting clause in 2012, and the control group are the firms that already have a double-voting clause. The difference-in-difference estimator (δ) is calculated as follows:

δ = (y̅_T,2 - y̅_T,1) - (y̅_C,2 - y̅_C,1) (9)

δ = ∆y̅_T - ∆y̅_C (1)

Here, y is the independent variable (in this thesis it will be Q, CAR, AI), T is the treatment group, C is the control group, 1 is an ex-ante observation and 2 is a ex-post observation. The D-in-D estimator will be found by using the regression models described below. One advantage of using the D-in-D method is that D-in-D controls for time invariant factors and unobservable factors that affect all firms. This way, controls only have to be specified for firm-specific effects. One assumption for the use of the D-in-D method is that unobserved differences between the firms are constant over time, also known as the common trend assumption (Stock & Watson, 2010).

The following D-in-D regressions will be used: 5.1.4 Firm value hypothesis

Q_ist= α + γGroup_s+ λEvent_t+ δ(Group_s×Event_t) + cW_it + e_it (10)

As the book values for 2016 are not readily available, this period is left out of consideration in the sample for Q.

5.1.5 Shareholder value hypothesis

CAR_ist= α + γGroup_s+ λEvent_t+ δ(Group_s×Event_t) + cW_it + e_it (11)

For testing this hypothesis, all the dates are considered as discussed in Table 3. The date variable (Event_t) will be 0 before the event date, and one on and after the event date. The event window is 5 days (day -2 through +2).

5.1.6 Liquidity hypothesis

Page 23 of 73

For testing this hypothesis, all daily dates are considered as discussed in Table 3 and, additionally, three longer periods. Table 8 summarizes the main variables in the regression specification, including description and source.

[Insert Table 8 here]

Based on other literature there are various expectations from the regressions. The coefficient of interest is the D-in-D coefficient (δ). For firm value (Q), the coefficient is expected to be negative (when P[law pass] increases), indicating that firm value went down for the treatment group at the end of 2015 compared to May 2012. For shareholder value (CAR), the coefficient is expected to be positive (when P[law pass] increases), as according to other research there should be no abnormal return due to the law change. However, this setting has not been tested before. Therefore, an alternative expectation would be that the coefficient is positive if P[law pass] increased, and negative when P[law pass] decreased. For liquidity (AI), the coefficient is expected to be positive (i.e. liquidity decreases) if P[law pass] increased, and negative if P[law pass] decreased.

5.2 V

I

With the above specified regressions, there is possibly a case of endogeneity. One source of endogeneity, the omitted variables, is possible. The legislative process has been mapped and important dates have been identified. However, there may be an unobserved date that has been missed in the mapping and thus contributes to this source of endogeneity. As a D-in-D method is used, for firm-specific unobserved effects should not cause a problem as long as they remain constant over time (common trend assumption). Simultaneity as a source of endogeneity can be ruled out, as the Q, CAR or AI cannot have caused the Law and control variables. The last source of endogeneity, measurement error, is very unlikely for the CAR and AI as they are calculated using data that is readily available and does is not prone to much interpretation or assumptions. Moreover, the dates could have been specified incorrectly. The Q, however, is subject to measurement error. In the denominator of the equation of Q, see section 5.1.2, the book value of assets contains intangible assets, where book value can differ from the “true” value. Due to this measurement error, the errors in Q are right-skewed. These measurement errors do not create a bias in the Q, but does inflate the residuals and standard errors (Gompers et al., 2010).

Another important issue concerning the validity of a D-in-D model, according to Bertrand et al. (2004), is autocorrelation, and the failure to correct for it. The result of failing to correct for autocorrelation results in large overestimation of t-statistics and significance levels. By disregarding the time-series

Page 24 of 73

aspect of the data, Bertrand et al. (2004) suggest that the t-statistics and significance levels will not be overestimated. By simply having one observation ex-ante and one ex-post, there will be two groups of observations and therefore the time-series information is ignored. As discussed in the beginning of section 5.1, the dependent variables are adjusted for this.

Another problem with the D-in-D model when it comes to the validity is the heterogeneity in the treatment effect. For example, it is possible to opt-out of the law and returning to the one share-one vote regime. However, there were various reasons why some firms were able to do this and some were not. Reasons for opting out would be resistance from an active state as shareholder, or the amount of free float or block holders there were. Firms could also behave differently from receiving one additional anti-takeover provision.

One of the issues concerning internal validity of event studies is the market anticipating the event, therefore the event becomes a non-event. This issue is tackled by examining the legislative process to test multiple events leading up to the implementation of the law. Another issue is that there is not more than one event on the event date. As described at the beginning of section 4, firms that had any form of corporate activity are excluded from the data set.

5.2.2 External validity

As the findings of this thesis can be used as input to assess the effectiveness of public policy, it is important to see how and where the findings from the sample used will apply. There are two fronts to where the external validity is important. First, as the sample consists of three French indices, containing large and small publicly listed firms, it provides a representative sample of the French publicly listed firms. Therefore, it can be said that the findings are valid for the population of French publicly listed firms. Second, the degree to which the findings from this sample of French firms is applicable to firms in other countries. This will depend, amongst others, on the legal system used and the degree of other measures in place to protect investors.

5.3 F

5.3.1 Firm value

Table 9 shows the output of the regression. The event is the total sample period (May 1st, 2012 until May 1st, 2016). Two additional ranges are included, as specified in Section 4.3.1, two investigate whether the observations from Graph 3 can be explained in more detail.

Page 25 of 73

The table shows the regression output to test the firm value hypothesis with the Q as a dependent variable. The coefficient of interest, (δ), is not significant, indicating that the Q did not change for the treatment firms when comparing 2012 with 2015. Without controls, this effect becomes negative and significant, indicating that the treatment firms, without controlling for firm specific characteristics, declined in value.

The other two date ranges explain the finding from Graph 3 in more detail. Conform the findings prima facie, in the first range (May 2012 – Sept 2014) the coefficient (δ) has a positive significant value of 0,023, indicating that in this range the firm value increased by 2% (using the median value of the treatment group). For the second range (Sept 2014 – Dec 2015), the same coefficient has significant value of -0,058, indicating a percentage decrease of 4%.

Another finding from the regression is the coefficient of Event (λ), which in expectation should have no significant value. However, the value of the coefficients of Event is positive and significant, indicating all the firms have a higher Q (increase by 0,162, or (using the control group median), 13%) at the end of 2015 compared to May 2012. This also holds in the other two date ranges. The positive trend can also be seen in Graph 3. A reasons for this finding could be an event that affects all firms, such as an economic boom.

To the extent the implementation of the law has an effect on the Q can be seen by comparing regression 1 and 2. Without any controls for firm specific, the results are all strongly significant. However, when controlling for a firm-specific characteristics, the results become not or less significant. This indicates that the increase in Q is due to the firm characteristics, and not due to the law change.

5.3.2 Shareholder value

Table 10 shows the output of the regression. There are seven events in the regression, with the date of the event stated in the table below.

[Insert Table 10 here]

The table shows the regression output to test the shareholder value hypothesis with dependent variable CAR. An (expected) change from short term to long term in the shareholder horizon, according to previous research, should yield no clear result (either positive or negative) for the (cumulative) abnormal return. The regressions provide clear results, as for all the events the coefficient of interest (δ) is negative (in a range of 0% to -1,7%), regardless of the probability of the law being implanted (or actually being implemented). The latter, especially the last event, shows a large negative effect on the CAR. Consequently, it can be said by looking at the results from the regression, that an expected switch to a long-term perspective among shareholders decreases shareholder value.

Page 26 of 73

When no controls are entered for firm specific effects, the results alter slightly in strength of the significance, not in the direction of the coefficient. This finding suggests that firm specific characteristics also play a role in determining the CAR around the event date, but the law also played a part.

5.3.3 Liquidity

Table 11 shows the output of the regression. There are seven events in the regression, with the date of the event stated in the table below.

[Insert Table 11 here]

The table shows the regression output to test the liquidity hypothesis with dependent variable AI. For the coefficient of Event×Group (δ), the results are ambiguous. For regression 1, the (unofficial) announcement of the law, the value of the coefficient is the most positive and strongly significant, indicating that the firms in the treatment group became less liquid compared to the days before the event. For a firm in the treatment group, this means that the illiquidity rose with 45% around the announcement of the law (based on the mean value of the treatment group). The event after, the publication of a negative letter to the French ministry (regression 2), shows the same trend, however the effect is slightly smaller. These two events combined create a liquidity negative liquidity shock which can also be seen in Graph 5, where just after the first quarter of 2015 illiquidity seemed to rise (however, it was only short lived). Of the other events, only 3 (out of 5) show significant outcomes, all negative. An interesting find here is that these 3 event represent an increase of the probability of the implementation of the law. For these events, the relative change in illiquidity lies in the range of -24% to -3%.

In order to check the impact of the law on the liquidity, the regressions are done again without controls. The outcome of these regressions do not differ much from the regression in Table 11 (direction of the coefficient), only the strength of the significance of the coefficients. The R-squared values do change substantially (from a range of 66,65%-74,81% in Table 11 to a range of 2,80%-4,92% in the regressions without controls).

Not surprisingly, some of the other coefficients are not significant. For the Event coefficient (λ), the coefficients are mostly (weakly) significant, which means that the firms in the control group are affected by the law. However, the directions of the coefficients are mixed. For the Group coefficient (γ), only regression 7 shows a strong difference between the treatment and control group before the event.

Table 12 shows the regression output for the events that do not have a specific date, rather a date range (a month). The months correspond to the official publication of a letter to the French ministry (regression 2) and trends visible in Google trends (regression 1 and 3).

Page 27 of 73

Regression 1 and 3 do not have a negative or positive expectation on the coefficient of interest (δ) , as Google trends does not show why certain words have been searched, only the trend. There is a significant value of the coefficient visible in both regression 1 and 3, indicating that a search trend for the law has a strong impact on the liquidity of the treatment group in the event month, compared to the month before. In regression 1, the coefficient is negative and significant, meaning the liquidity increased. Regression 3 shows the exact opposite, a positive strongly significant value. Of course, the trends on Google do not stand alone, they are a result of certain news. For example, the trend observed in March 2014 can be the results of a turbulent month in February (with respect to the probability of the law passing). This can cause uncertainty among investors, and hence, the treatment firms’ shares becoming more liquid. The event of regression 2 is the publication of an official letter to the French ministry from a French asset management association (negative news and P[law pass] decreasing). This can be seen in the output, the coefficient of interest (δ) has a significant positive value. A positive value is contrary to what the expectations are for this event.

Table 13 shows the regression output for the AI for the whole sample (May 1st, 2012 until May 1st, 2016).

[Insert Table 13 here]

The table shows the regression output for the total sample period to test the liquidity hypothesis. For the coefficient of Event×Group (δ), the value is negative and weakly significant. Percentage-wise, firms in the treatment group saw an increase of their liquidity due to the implementation of 9% (using the mean value of the treatment group). This indicates that over the whole sample period the liquidity went up for the treatment group. This does not confirm the expectations as a long-term perspective among shareholders should result in less trading (i.e. liquidity decreases).

In contrast to the date events, the Event (λ) coefficient is not significant. This indicates that for the control group, the liquidity did not change due to the implementation of the law. This was also expected. The impact of the law on the illiquidity of firms in the sample can be seen by comparing regression 1 and 2 in Table 13. In regression 2, where there are no controls, all the coefficients are strongly significant. However, when controls for firm specific characteristics are entered, much of the significance disappears. This shows that throughout the whole sample, the law had a minor effect on the liquidity of the firms

Combining the findings Table 12 and Table 13, there is evidence that the law had an impact on the liquidity of the firms in the sample. Over the whole sample period, there is weak evidence of decreased

Page 28 of 73

liquidity, and the earlier dates provide strong evidence of a decrease in liquidity due to information about the law becoming public.

5.4 R

The composition of the treatment group has changed throughout the sample period. From April 1st, 2014 until March 31st, 2016 firms had the chance to present a plan to counter the law’s provisions in the Annual General meeting. If more than two-third of the shareholders would vote against the double-voting regime (i.e. keep the one share-one vote), then the law would not apply. To test whether the findings from the above regression hold, the regressions are rerun with an adjusted treatment group. Two adjusted treatment groups are made, one group which successfully voted against the law’s provisions during the two years from 2014 to 2016 (Group 1). Another group which includes firms which did either nothing or did not pass the two-third quorum necessary (Group 2). The total treatment group consists of 79 firms, which is split up in 55 firms in Group 1 and 24 firms in Group 2.

First, the regressions for the firm value are rerun. Refer to Appendix 6 for the details of these regression. When the original treatment group is replaced by Group 1 or Group 2 for the overall period, no big changes are visible. One small difference is that the coefficient of interest (δ) is weakly significant for Group 1. Looking at the first sub-period of May 2012 – Sept 2014 it becomes evident that there is a difference when separating the two groups. In the original treatment group, the coefficient of interest was positive and significant. When separating the two groups, the significance is lost for both groups. When looking at the second sub-period, from Sept 2014 – Dec 2015, the coefficient of interest does not change greatly. The magnitude does not change greatly, only the significance changes slightly, it is strongly significant for Group 1 and weakly significant for Group 2. In short, there are slight differences when separating the two groups.

Second, the regressions for shareholder value are rerun. Refer to Appendix 7 for the details of these regressions. Interestingly, and intuitively sound, the effect on CAR is the exact opposite for every event (except the last event) when separating the two groups. However, whatever the nature is of the event (increased or decreased probability of the law passing), Group 1 has a negative return and Group 2 has a positive return. The last event has a large negative effect for both of the groups. These test portrait a different picture than the original specification as the two groups behave differently throughout the sample.

Third, the regressions for liquidity are rerun. Refer to Appendix 8 for the details of these regressions. Similar to the original regression specification, the two different groups show significant effects in the