Are Directors' Dealings Informative? Evidence from European Stock Markets

Transcription

1 Are Directors' Dealings Informative? Evidence from European Stock Markets Kaspar Dardas * European Business School Andre Güttler European Business School Abstract Are directors' dealings reports informative for outside investors? We analyze short-term announcement effects for 2,782 companies from eight European countries between 01/2003 and 12/2009. We find significant announcement effects in four out of eight countries after directors' dealings reports have been disclosed. For most countries the magnitude of the announcement effect depends on transaction size, firm size, book to market ratio, and multiple trades by different insiders on the same trading day. The results are stronger for purchases than for sales. For France, Ireland and Sweden we find tentative evidence that the corporate position of an insider is connected to the size of the announcement effect. JEL classification: G14, G15 Keywords: Directors' dealings, Market Abuse Directive, Insider hierarchy, R&D intensity * Department of Finance, Accounting and Real Estate, EBS Business School, Gustav-Stresemann-Ring 3, Wiesbaden, Germany, kaspar.dardas@students.ebs.edu, (Corresponding author). Department of Finance, Accounting and Real Estate, EBS Business School, Gustav-Stresemann-Ring 3, Wiesbaden, Germany, andre.guettler@ebs.edu We thank an anonymous referee for very helpful comments. We also thank 2IQ Research for providing the data for our analysis. Any remaining errors are our own.

2 In the past decade, the use of directors dealings reports in stock analysis has become common practice among financial professionals. The rationale behind it is that substantial information asymmetries exist between company insiders and outside investors. 1 Since insiders are involved in the day-to-day business of a company, they are expected to be better informed about the true value of their company s assets than any other investor. Thus, whenever insiders trade stocks of their company, they are assumed to be revealing new information. We investigate whether directors dealings reports are informative for outside investors. We analyze short-term announcement effects for 2,782 companies from eight European countries, namely Austria, France, Germany, Ireland, Italy, the Netherlands, Sweden, and the U.K, with an observation period between 01/2003 and 12/2009. Our main findings show that there are significant announcement effects in four out of eight countries after directors dealings reports have been disclosed. We find that the magnitude of the announcement effect is higher for purchases than for sales. We also conclude that the magnitude of the announcement effect depends heavily on transaction size. Factors such as firm size, book to market ratio, or multiple trading by different directors on the same day have an influence on announcement effects. For France, Ireland and Sweden we find tentative evidence that the corporate position of an insider is connected to the size of the announcement effect. Moreover, we show that in some countries the recently implemented Market Abuse Directive had a noticeable effect on the information content of directors' dealings. We further conclude that directors dealings reports lead to largest announcement effects in the healthcare and energy sectors. Previous studies based on directors dealings reports show that company insiders generate abnormal returns by trading stocks of their company (e.g. Seyhun (1986), Rozeff and Zaman (1988), Lin and Howe, (1990), Lakonishok and Lee (2001)). Others focus on the long-term abnormal returns generated by company insiders. For instance, Rozeff and Zaman (1988) examine whether outsiders are able to earn abnormal profits by mimicking insiders. Since the information on the insider trade takes some time to reach the market, they conduct the outsider performance test by imposing a two-month trading lag when creating the outsider portfolios. They find that outsiders are indeed able to generate 1 We use the terms insiders and directors interchangeably unless stated otherwise. Thus, we do not follow the strict U.K. definition of directors but rather refer to all company insiders whenever using the term directors. 1

3 abnormal returns. However, after including transaction costs, the outsider profits disappear entirely. In contrast, by using U.S. data Bettis, Vickrey, and Vickrey (1997) show that outsiders are able to make significant profits. They measure weekly abnormal returns by using large-volume insider trades as event triggers and find that outsiders are able to make statistically and economically significant abnormal returns net of transaction costs for holding periods that are longer than 13 weeks. Specifically, they report an abnormal return of 6% for a holding period of 26 weeks. 2 More-recently published studies focus on the immediate effects of insider trading around announcement and/or trading days. Lakonishok and Lee (2001) examine a five-day period after the announcement as well as after the trading day for U.S. stocks. Only for trades by insiders of small capitalization firms do they detect an abnormal return of 0.93% subsequent to the trading day. Friedrich et al. (2002) examine short-term daily returns following the trading days of insiders for the U.K. They do not find evidence that outsiders are able to earn economically significant returns net of transaction costs. Previous research on directors dealings is mostly concentrated on the U.S. and U.K. This is obviously due to the long history of insider regulation in both countries. The existence of announcement effects is undisputed in both countries. However, research on directors dealings in continental Europe shows mixed results. Klinge et al. (2005), Stotz (2006), Dymke and Walter (2008), and Betzer and Theissen (2009) find significant abnormal returns based on directors dealings in Germany. On the other hand, Eckbo and Smith (1998) do not find any effects in the Norwegian stock market. Del Brio et al. (2002) find that insiders are able to earn abnormal returns in Spain. For the Italian market, Bajo et al. (2006) observe that abnormal market performance occurs after an insider s transaction, usually between the first and the third month after the insider trade. Zingg et al. (2007) find that outsiders who mimic small insider transactions can earn abnormal returns in the Swiss stock market. For the Netherlands, Degryse et al. (2009) find significantly positive abnormal returns after purchases by directors and supervisory board members. 3 Recently, two studies examine insider dealings in a broad selection of countries. Fidrmuc et al. (2010) investigate the dependence of market 2 For long-term studies, which examine portfolios based on insider trades see Jeng et al. (2003) for the U.S. and Giamouridis et al. (2008) for the UK. 3 See also Kallunki et al. (2009), who investigate the motives of Swedish insiders to trade stocks of their firms. 2

4 price reaction to insider transactions and corporate governance in 16 countries. They find that market reaction to insider purchases is larger in countries with good corporate governance. Aussenegg and Ranzi (2008) examine the market reactions to directors dealings reports in seven European countries of German and French legal origin. 4 They show that information asymmetries between insiders and outsiders are stronger in German law than in French law countries. In both of the above cross-country studies, the authors pool countries based on their legal origin and investigated market reactions of these pooled samples. This paper contributes in the following ways to the existing literature. First, with regard to other cross-country studies on legal insider trading we focus our analysis on single countries and not on pooled country samples. This study structure allows us to control for country-specific regulatory changes, which have been implemented during our observation period. Specifically, the E.U. Market Abuse which aims to harmonize financial markets across Europe. Thus for each country from our data sample, we give a short overview on the regulatory changes with respect to directors' dealings and test whether these changes had an effect on the information content of directors' dealings reports. Moreover, we want to show in which countries and under which conditions directors dealings reports are most informative. To give a clear answer as to whether directors dealings announcements are informative or not, we see the need to investigate each country individually. After all, outside investors purchase/sell stocks in single country exchanges and not according to factors such as the legal origin of a country. Our parallel structured cross-country analysis is also unique in its detailed examination of the insider hierarchy hypothesis, which states that the corporate position of an insider causes different announcement effects. This is aided by our data quality, which allows us to identify the insider position for each single transaction. To the best of our knowledge, the insider hierarchy hypothesis is unexplored for most of the countries in our study. 5 Second, we put a strong emphasis on the comparability of results between each country in our analysis. A remarkable aspect of most European studies is that results are often compared to the results from benchmark studies of the U.K. or U.S. (e.g. Dymke and Walter (2008), Degryse et al. (2009)). Yet in the U.K. and U.S. papers, the 4 Aussenegg and Ranzi (2008) classify countries according to the legal origin as defined by La Porta et al. (1998). 5 The insider hierarch hypothesis has been previously examined for the U.K. by Fidrmuc et al. (2006) and for Germany by Dymke and Walter (2008) and Betzer and Theissen (2009). 3

5 event studies are applied over substantially longer and different observation periods than in (Continental) European studies. The comparability of event-study results obtained from different observation periods is problematic because event studies are affected by time-varying idiosyncratic risk. The power and specification of event studies is consequently not stable through time. Therefore, a comparison of event study results from substantially different observation periods might lead to erroneous conclusions. For that reason, our paper analyses announcement effects for different markets with almost identical time periods, identically constructed benchmark indices, and analogous eventstudy variables. Third, due to the quality of our dataset we were able to use only pure buy and sell transactions, which allows an unbiased examination of announcement effects for each single country. For instance, we exclude transactions related to mergers and acquisitions and any other type of transaction that might distort a clear examination of the announcement effect. 6 Finally, we examine which industry sectors show highest announcement returns to directors dealings reports in European stock markets, which is also novel to the literature. 7 The rest of the paper is organized as follows. Section I gives a short overview of the directors dealings regulations in Europe. Section II derives our main hypotheses from previous literature. Section III describes our data and methodology. Section IV analyzes the results and Section V concludes the paper. I. European Regulation on Directors Dealings In the U.S., the reporting duty of directors' dealings has a long tradition with legislation in place since The U.K. was the first country in Europe to implement insider trading laws. Its history of regulating directors dealings goes as far back as 1979, with the implementation of the 1977 Model Code. The 1985 Companies Act, enacted in 1985 extended this regulatory framework. It required the immediate reporting of insider transactions, no later than the fifth business day after an insider trade has been made. In addition, it had a narrow definition of who is considered a company insider. Only 6 Fidrmuc et al. (2006) investigate the impact of news releases prior to directors transactions for the U.K. market. They find that market reactions to insider trades are mostly influenced following the news of mergers and acquisitions. 7 Whenever we use the term insider trading we refer to legal insider trading unless stated otherwise. For a study on illegal insider trading (in credit derivatives markets) see Acharya and Johnson (2007). 4

6 executive board members and non-executive directors were required to report their trades, whereas large shareholders were excluded from reporting. A remarkable characteristic of the U.K. regulation was and still is the existence of extensive blackout periods. In particular, the U.K. regulator requires a two-month trading gap prior to final and interim earnings announcements. 8 Most other European countries did not implement directors dealings regulations before the 1990s. 9 Moreover, all European countries regulated insider trading differently in terms of the definition of an insider and inside information. The implementation of the Council Directive 89/592/EEC in 1992 brought a change to these inconsistent regulations. 10 This was the first adoption of a directive by the European Parliament to control for insider trading across all European capital markets. Nevertheless, this directive gave each member state a wide range of freedom concerning the implementation and enforcement of insider trading laws. In 2004, the more comprehensive Market Abuse Directive 2003/6/EC (2004/72/EC) replaced it. 11 The Market Abuse Directive, which is currently valid for all European Union states, requires all members to establish minimum standards for regulating insider trading. The member states must ensure that directors dealings reports are made within five business days and contain crucial information such as size, price, and the characteristic of the transaction. The directive also defines the people who are obliged to report transactions. In particular, all persons in managerial positions, as well as their families and institutions that are associated with these persons, are required to report their transactions. 12 Furthermore, the E.U. directive requires companies to report the names of all insiders by the issuers and to update this information regularly. All member states can implement a 5,000-euro notification barrier. Thus, small transactions only have to be reported if the accumulated transaction size exceeds 5,000 euros in a calendar year. In 2007 another directive which aims to improve reporting standards and harmonize 8 For a detailed overview of insider trading regulations in the U.K., see Pope et al. (1990) and Fidrmuc et al. (2006). 9 See also Gersbach and Nedwed (1991) who present arguments for and against regulations on insider trading. 10 The release date of 89/592/EEC is 13/11/1989; the effective date is 01/06/ The Commission Directive 2004/72/EC is an extension of the Market Abuse Directive 2003/6/EC. It gives detailed requirements on directors dealings reporting standards. 12 According to 2004/72/EC, art. 1(1), a person discharging managerial responsibilities within an issuer shall mean a person who is (a) a member of the administrative, management or supervisory bodies of the issuer; (b) a senior executive, who is not a member of the bodies as referred to in point (a), having regular access to inside information relating, directly or indirectly, to the issuer, and the power to make managerial decisions affecting the future developments and business prospects of this issuer. 5

7 financial reporting within the E.U. was implemented. This so-called Transparency Directive 2004/109/EG (2007/14/EG) requires each member state to ensure that there is at least one officially appointed mechanism for the central storage of regulated information. Moreover, it states that issuers must ensure appropriate transparency for investors through a regular flow of publicly available information and that corporate fillings have to be reported electronically with an easy access by end users. The Transparency Directive mainly focuses on shareholding disclosures and periodical financial reports, however, it also implies an increased demand for reporting standards on directors' dealings. Table I documents the official implementation dates of the Market Abuse Directive 2003/6/EC as well as the Transparency Directive 2004/109/EG in each country from our data sample. It is noteworthy that for some countries the Market Abuse Directive had a strong impact on the legal framework with regard to directors dealings whilst for others it had almost none. For the U.K. and Ireland, for instance the directives brought only relatively small changes regarding the regulations on directors' dealings because both countries had well-established regulatory frameworks beforehand. However, for countries such as Austria, Germany and France the Market Abuse Directive changed the regulations on directors' dealings fundamentally and improved the reporting standards significantly (see Table I). 13 II. Hypothesis Development Our approach is to examine the short-term abnormal returns within a five-day period subsequent to the announcement of an insider transaction. We expect share prices to adjust within several days after the buying/selling signals have been given to the market. If insiders are indeed able to identify whether their company is over-/undervalued and if insiders give clear signals to the market by buying or selling stocks of their company, then investors should react quickly. Therefore, we examine the following hypotheses. 13 Hypothesis 1: The announcement of purchases (sales) by insiders leads to immediate positive (negative) abnormal returns in the underlying share. See Mazars (2010), the European Commission web page on the Market Abuse Directive: European Commission web page on the Transparency Directive: 6

8 Fidrmuc et al. (2006) report higher abnormal returns for large transactions than for small transactions in the U.K. Betzer and Theissen (2009) obtain similar results for the German market. 14 Following these findings, we assume that announcements of large transactions give stronger signals to the market than announcements of small transactions. Hypothesis 2: The announcement of large purchases (sales) by insiders leads to larger positive (negative) abnormal returns than the announcement of small purchases. We proceed by examining the relation between an insider s position in the company and the size of the announcement effect. It is rational to assume that top-level executives should have an advantage over low-level executives in obtaining relevant information about changes in the firm s value. Market reactions should be stronger following the announcement of trades made by high-level executives than by low-level executives. However, Fidrmuc et al. (2006) show that abnormal returns following the announcements of top-level executives in the U.K. are significantly lower than following the announcements of other directors. 15 Dymke and Walter (2008) obtain similar results for the German market. These, on first sight, counterintuitive results might be explained by the fact that top-level executives are followed more closely by regulatory authorities and therefore trade at less informative moments. To see whether this is also the case for the countries in our data set we test each country for the so-called insider hierarchy hypothesis. Hypothesis 3: The announcement of trades by top-level executives will result in stronger reactions in the market than the announcement of trades by other directors. 14 Fidrmuc et al. (2006) as well as Betzer and Theissen (2009) define large trades when the trading volume exceeds 0.1% of the volume of shares outstanding. 15 See also Jeng et al. (2003). 7

9 In addition, we assume that multiple purchases (sales) by several insiders on the same day will give a stronger signal to the market than a single transaction by one insider. We believe that the announcement effect increases with the number of insider transactions. Hypothesis 4: The announcement of multiple purchases (sales) by different insiders on the same day leads to larger positive (negative) abnormal returns than the announcement of single purchases. Previous studies have shown that firm size in terms of market capitalization has a substantial influence on the announcement effect. Lakonishok and Lee (2001) conclude that directors dealing announcements for small capitalization firms show a significantly larger market reaction than in the case of large firms. Aussenegg and Ranzi (2008) also find a reverse relationship between abnormal returns after trade announcements and firm size for several European countries. Since analysts follow small firms less closely than they follow large firms, the information asymmetry is larger for small firms. Therefore, the possible new information that a directors dealings report contains about a small firm is more significant than in the case of a large firm. We assume the same for our data and therefore we state the hypothesis: Hypothesis 5: The announcement of insider purchases (sales) in small firms leads to larger positive (negative) abnormal returns than the announcement of insider purchases in large firms. A final investigation concentrates on industry sectors. For the U.S., Aboody and Lev (2000) showed that insiders gain higher returns in R&D-intensive firms than in firms with less R&D. The rationale behind it is that information asymmetries are higher in R&D-intensive firms due to the uncertainty of the outcome of R&D projects. This lets us assume that insiders might gain higher returns in R&D-intensive sectors such as healthcare and IT compared to sectors that are less R&Dintensive, such as consumer goods. 8

10 Hypothesis 6: The announcement of insider purchases (sales) in R&D-intense sectors leads to larger positive (negative) abnormal returns than the announcement of insider purchases (sales) in sectors with a low R&D intensity. III. Data Sources, Descriptive Statistics, and Methodology A. Data Sources Our data on directors dealing reports come from 2iQ Research, a data vendor that specializes in European insider trading. Whereas the source data is downloaded from stock exchange internet pages, adhoc news portals or directly from company websites and enhanced with qualitative research, which we will describe below. The original database contains a total of 151,989 insider transactions in exchange-listed companies from Austria, France, Germany, Ireland, Italy, the Netherlands, Sweden, and the U.K. We chose (with a few exceptions) the same observation period for each country, beginning on January 1, 2003 and ending on December 31, This allows us to examine the most extensive data sample, which includes all countries. Each transaction includes information on company name, insider name, insider position, trade date, trade size, announcement date, transaction type (11 different types), and security type (10 different types). In addition, some transaction types are marked with 14 characteristics that further specify each transaction. These characteristics identify whether a buy/sell transaction is option/award/plan related, merger related, tax liability related, capital increase related, a dividend reinvestment, a private placement, or remuneration. For instance, a sell transaction might be marked as tax related, which leads to an exclusion of this transaction from our study. Each single transaction has been researched for relevant company news around the transaction day and marked accordingly. This unique data quality allows us to filter only pure buy and sell transactions for each single country. In particular, Fidrmuc et al. (2006) point out the importance of controlling for merger related transactions, since these transactions decrease the announcement effect almost to zero. The most common transaction types in the original dataset (before filter rules are applied) can be categorized as buys (55.93% of all transactions), sales (33.91%), exercises (5.95%) and subscriptions (1.87%). We applied strict rules to modify this initial dataset. First, we dropped planned, 9

11 private, dividend related, merger related, tax related, neutral, and compensation transactions. This rule mainly eliminated transactions such as exercises, subscriptions, dividend reinvestments, or exercise related sales. Second, we dropped security types such as ADRs, bonds, options, and stock swaps and kept only ordinary shares. Furthermore, we dropped all transactions that are not in the currency of the respective domestic exchange. These filters decreased the sample to 72.46% of the original observations to 110,128 transactions. Moreover, we dropped all observations where the transaction price deviated by more than 20% from the securities closing price on that day. This rule further minimizes the likelihood that private or compensation-based transactions are included in the sample. Furthermore, we dropped all issuer-based transactions and transactions that were reported 20 calendar days past the transaction date, since we believe that these directors dealing reports attract only minimal attention in the markets. 16 These rules decreased the data sample to 58.64% of the original observations (89,125 transactions). We also excluded transactions made by investment firms as well as funds, unless they are controlled by an insider and aggregate multiple transactions by the same insider on the same day. Based on this aggregation we define the total traded volume of an insider as a net purchase or sale. To avoid unnecessary noise we excluded transactions with an aggregated trade volume of less than 1,000 euros. This reduced the dataset to the final sample number of 44,907 observed transactions, which represents 29.52% of the initial sample. Table II gives a detailed overview of this final data set. Daily share returns, book values, and shares outstanding are obtained from Compustat. We also use the Compustat database to track any changes in company names. For each country we use the country-specific (Gross) MSCI Barra Performance Share Indices to calculate risk-adjusted returns. 17 All transactions are converted into euros for non-euro countries using the daily exchange rate of the announcement date from Compustat. 16 Before the implementation of the Market Abuse Directive 2003/6/EC, most European countries did not have any requirements for timely reporting of directors dealings (see also Table I). Even after the implementation of the directive, some transactions are not reported within the required five business day range. Therefore, for about 5% of the transactions described above, the difference between reporting and transaction day is longer than 20 calendar days. 17 Gross refers to the maximum possible dividend reinvestment. The amount reinvested is the entire dividend distributed to individuals resident in the country of the company, but does not include tax credits. 10

12 B. Descriptive Statistics Table III shows the descriptive statistics for our initial data sample, which includes net purchases, sales, option related sales, and exercises. Similar to Lakonishok and Lee (2001) and Fidrmuc et al. (2006), we calculate the (relative) trading frequencies per firm and year for each country. 18 Directors in Germany, Ireland, and the U.K. have the lowest average trading frequency of all countries. On the other hand, Italy, the Netherlands, and Sweden seem to have extremely active insider trading. The biggest difference in (relative) insider trading frequency is between Germany (average 3.60 trades per firm and year) and Italy (average 8.54 trades per firm and year). There are several possible reasons why the average trade numbers vary between countries. First, there is a technical issue concerning how insider trades are reported in certain countries. Sweden and the Netherlands have a detailed accounting style of reporting insider transactions, which artificially inflates the number of transactions. In the Netherlands, for instance, an option exercise followed by the sale of the underlying shares is reported as three transactions in separated reports: the exercise of options, the subsequent purchase of the underlying shares, and the sale of these shares. In Germany, the U.K., and Ireland, such transactions are mostly reported as single sell transactions with a short note in the report that states that the transaction is option related. Second, the legal definitions of insiders vary across countries, resulting in different numbers of people acting as insiders. Countries that have a wide definition of insiders will consequently have more persons acting as insiders and thus have a higher number of insider trades. Italy, for instance, has the broadest definition of insiders among all European countries. This becomes evident when considering that all public officials in Italy are defined as insiders whenever their responsibilities allow them to obtain inside information. This is even the case when the officials do not have a specific relationship to a company. A counterexample of a country with a narrow insider definition is Ireland, where a specific relationship between a public official and a company must exist for the public official to be considered an insider. Furthermore, the Italian legislation does not differentiate between the 18 The (relative) trading frequencies for the UK, reported in Table III are consistent with the trade frequencies reported by Fidrmuc et al. (2006). 11

13 penalties for primary and secondary insiders, which further indicates the wide definition of relevant insiders (Alexander (2007)). 19 Third, the ownership structure of a country might have some relevance for the trading frequency of insiders. A country with an ownership structure that includes many publicly traded firms controlled by the state, families, or other financial firms will consequently have a large circle of people who are defined as insiders. 20 On the other hand, countries with ownership structures that mostly consist of widely held firms must have fewer insiders per firm and therefore less insider trading. According to Faccio and Lang (2002), the highest proportion of widely held firms in Europe is in the U.K. (63.08%) and Ireland (62.32%). Both countries also show the lowest average insider-trade numbers in Table III. Finally, the trading methods carried out by some insiders explain the variation in trading frequencies. This is especially true for Italy, where it can be observed that insiders tend to buy stocks in several tranches within consecutive days. On the other hand, in Germany or the U.K. insiders place their positions in single transactions on a single day. We further examine all transactions in our sample categorized by insider position. Table IV shows the respective descriptive statistics, in which transactions are sorted into three classes according to the corporate position of the insider who performed the transaction. We grouped corporate positions into insider classes based on two criteria: the day-to-day business involvement of an insider and the ability of an insider to make strategic decisions in the company. Insider class A: CEO, Deputy CEO, CFO, COO, President, chairman (also non-executive), and similar corporate positions as well as respective family members. Insider class B: Divisional/regional CEO, CEO of subsidiary, Vice President, non-executive vice-chairman, Chief Information Officer, Chief Scientific Officer, Officer, managing 19 Primary insiders are defined as persons who obtain inside information because they are employees, large shareholders, or the above-mentioned public officials. Secondary insiders are any persons having inside information. For instance, a waitress who overhears a non-public conversation between two directors is already defined as a secondary insider. It must be noted that the legal definition of insiders falls into the discussion on illegal insider trading. However, we believe that a wide definition of insiders such as in Italy will have two consequences. On the one hand, the reported number of trades per firm and year will be higher than in a country that has a narrow definition of insiders. On the other hand, we assume that the information content of directors dealings reports will be lower in countries with a wide insider definition since less informed insiders will report transactions with low relevance to the market. 20 Directive 2003/6/EC, art. 2 (1) (b) defines all persons who are large shareholders as insiders. Therefore, not only persons who have a direct relationship with the company are considered as insiders but all executives of related companies as well. It is important to note that persons must be understood as natural persons as well as legal entities. 12

14 director, executive director, company secretary, group director, upper management, and similar corporate positions as well as respective family members. Insider class C: Non-executive board member, (non-executive) director, supervisory board member, board of auditors, lower class executive (sales director, technical manager etc.), upper management of subsidiaries, former CEO/executive/board member, divisional/regional director and similar corporate positions as well as respective family members. Panel A of Table IV shows the statistics for all transactions, while Panel B refers to large transactions. Similar to Betzer and Theissen (2009) and Fidrmuc et al. (2006), we define large transactions as all net purchases and sales that represent at least 0.1% of shares outstanding. 21 According to Panel A, more purchases (30,878) were made by insiders than sales (14,029). Panel B reports the same with 4,576 purchases versus 4,107 sales. However, in both panels the average mean and median for sales is significantly larger than for purchases. These results are consistent with previous studies (e.g. Betzer and Theissen (2009) and Fidrmuc et al. (2006)). The explanation for the disparity between sales and purchase volumes is that insider sales are often liquidations of large remuneration packages. Purchase transactions, however, are private investments made by the insider, which are only fractions of these remunerations packages. According to Panel B, most active insiders are top executives with 2,373 (1,701) purchases (sales). Panel A reports the same for purchases (12,677); however, for sales all insider classes have about the same trading volumes. C. Methodology We conduct our study by separately analyzing short-term announcement effects in each single country. To assess the information content of insider trade announcements we examine the cumulative average abnormal returns (CAARs) of the announcement day and the subsequent day CAAR(0;1) as well as longer event windows, CAAR(0;4), CAAR(0;20), and CAAR(-20;-1). 21 A threshold relative to a firm s market capitalization has an obvious bias towards small-cap firms since spending on the firm s own shares will not rise evenly with the market capitalization of the firm. Nevertheless, due to cross-country income inequalities and differences in spending we regard a relative threshold as being superior over an absolute threshold. 13

15 To calculate the CAARs and test for their significance, we apply a standard event study approach (MacKinlay (1997)). We define the event date as the announcement date of the transaction and proxy the market return by the corresponding (Gross) MSCI Barra Performance Share Index for each country. The beta is estimated over a period of 200 to 21 trading days prior to the announcement day. We employ a 41-day event window centered on the announcement day. Similar to Fidrmuc et al. (2006), we use three parametric and one non-parametric test statistics to examine whether CAARs are different from zero. In particular, we apply the test statistics t CAAR based on Barber and Lyon (1997), which is Student-t distributed with N-1 degrees of freedom. J 1 and J 2, based on Campbell et al. (1997). The choice between J 1 and J 2 depends on the variance of the securities with higher abnormal returns. J 1 is preferred whenever large abnormal returns occur in securities with high variance. J 2 is preferred whenever abnormal returns are constant across securities. Finally, we apply a non-parametric test statistic t rank based on Corrado (1989). The non-parametric test statistic as proposed by Corrado (1989) has several advantages over parametric tests. For instance, it does not require normally distributed abnormal returns. Furthermore, it is robust in the presence of event clustering or event-induced increases in variance. We use a conservative approach to test the statistical significance of our results and therefore report only the lowest absolute value of the t-statistics described above. Moreover, we apply three other robustness checks to confirm our results and to control for possible thin trading problems in small capitalization stocks. First, we repeat our study by using market-adjusted returns with the respective (Gross) MSCI Barra Performance Share Index as the benchmark index. Second, for France, Germany, Sweden and the U.K. we also calculate CAARs with different all share indices. Specifically, the SBF250 for France, the CDAX for Germany, the OMX Stockholm for Sweden and the FTSE all share index for the U.K. Third, we calculate CAARs from a modified data sample that is free of event window clustering. 22 In none of the modified samples 22 MacKinlay (1997) points out the importance of using non-overlapping event windows to calculate the aggregated cumulative abnormal returns. If event windows overlap, single securities will be correlated, which affects the variance of the aggregated cumulative returns. However, if event windows do not overlap the covariance between securities will be zero and the variance of aggregated cumulative returns can be calculated without any concern about the correlation of securities. The presence of event window clustering might lead to erroneous results in parametric tests. 14

16 described above could we find remarkable differences to our initial results, which are presented in the following section. 23 IV. Results A. Cross-Country Analysis Table V presents the results for each single country from our dataset for all (Panel A) and large transactions (Panel B). Similar to Fidrmuc et al. (2006), we concentrate our basic univariate event study analysis on large transactions. In general, the abnormal returns for large transactions and all transactions are similar but larger in magnitude for large transactions. For large purchases reported in Panel B of Table V, we find that CAARs(0;1) are positive and significantly different from zero in five countries. Furthermore, we found significantly positive results for CAARs(0;4) in all countries besides Austria and the Netherlands. As expected, for most countries the main price adjustments occur within one trading week. After the fifth trading day the CAARs continue to rise, however, at a lower rate. The highest CAARs(0;1) are reported for Ireland (5.03%) and the U.K. (6.12%). However, the results for Ireland have to be regarded with caution since the observation number for large purchases (sales) is as small as 35 (21). For Sweden (1.27%) and Germany (1.11%), the CAARs(0;1) are significantly lower than for the U.K. and Ireland. The lowest significantly positive CAAR(0;1) is reported for Italy (0.49%). For CAARs(0;4), we observe increases relative to CAARs(0;1) in each country except for Ireland. For large sell transactions we observe significantly negative results for CAARs(0;1) and CAARs(0;4) only in the U.K. (-0.41% and -0.88%) and Sweden (-0.82% and -1.39%). The above results indicate clear market reactions following the announcements of insider trades. As hypothesized, we observe positive abnormal returns after purchase announcements and negative abnormal returns after sell announcements. Especially for Germany, Ireland, Sweden, and the U.K. we found significant results. Due to the magnitude of these CAARs, we see our results not only as statistically but also as economically significant. For large transactions, we found higher CAARs than for small transactions. Thus, we conclude that transaction size indeed plays a crucial role in the size of 23 We construct the data sample that is free of event window clustering by eliminating overlapping event windows for single securities. The results are omitted from the paper but are available upon request. 15

17 the announcement effects. 24 Especially for the U.K, the significance of trade size seems to be essential, since the difference between large transaction and all transaction CAARs is 4.28% for CAARs(0;1) and 4.73% for CAARs(0;4). We assume that the results for the U.K. and Ireland are strikingly high because investing based on directors dealings reports is more popular in these countries than in Continental Europe and thus leads to herd-like market reactions after announcements. Furthermore, our results for the U.K. reported in Panel B of Table V are higher compared to Fidrmuc et al. (2006), who report 3.12% for CAAR(0;1) and 4.62% for CAAR(0;4). We explain these differences with our more recent dataset and the increasing popularity of investing based on directors dealings reports. 25 In addition, we see the exclusion of all types of remuneration, private, and merger related transactions as a reason for our high CAARs. We found the weakest evidence of announcement effects in countries with French legal origin. However, Austria, which is a German legal origin country, does not show any announcement effects, which might be due to the relatively small Austrian capital market. Another notable observation in Table V are the results for the pre-event CAARs(-20;-1), which are generally negative (positive) prior to purchase (sell) announcements for all countries except for Austria, Ireland, and the Netherlands. These results demonstrate that insiders tend to make purchase (sell) transactions after the share price has declined (increased). Thus, our univariate results confirm the general opinion from previous research that insiders follow a contrarian investment style (Seyhun (1992), Lakonishok and Lee (2001), Friedrich et al. (2002), Piotroski and Roulstone (2005)). B. Transaction Size Hypothesis 2 states that large insider transactions trigger stronger announcement effects than smaller transactions. Our univariate result also indicates that large transactions are followed by larger announcement effects. However, previous literature also suggests that medium-sized trades are more informative than large trades since insiders tend to make smaller transactions, simply not to attract the attention of other market participants. 26 Moreover, trades which are very large might lead to a 24 We further investigate the connection between announcement effects and transaction size in the following section. 25 Fidrmuc et al. (2006) examined directors dealings for the U.K. between 1991 and Barclay and Warner (1993) also show that in an anonymous market informed insiders tend to hide behind medium-sized trades rather than trading large block transactions. Barclay and Warner (1993) refer to this trading 16

18 substantial decrease in a company's free float and therefore be of minimal interest to other market participants. 27 For the U.K., Friedrich et al. (2002) show that medium-sized trades predict higher returns than large ones. Therefore, we also calculate CAARs(0;1) and CAARs(0;4) for different transaction sizes and examine whether medium-sized transactions generate the highest returns. Specifically, for each country we calculate CAARs for the following net transaction size groups: (i) transaction size < 25,000 euros, (ii) 25,000 euros transaction size < 100,000 euros, (iii) 100,000 euros transaction size < 250,000 euros, (iv) 250,000 euros transaction size < 1,000,000 euros, and (v) 1,000,000 euros transaction size. We also perform a multivariate regression to test for differences between the size groups. In none of the observed countries do we find a clear indication that medium-sized trades generate higher abnormal returns than large ones. For reasons of brevity we do not present these results. 28 Generally, we conclude that (short-term) CAARs rise with the size of the transaction. Similar to Fidrmuc et al. (2006), we find the highest CAARs whenever the trade size surmounts 0.1% of the firm s market capitalization. Thus, in our further study we calculate and mainly discuss results for all transactions as well as transactions that are at least 0.1% of the firm s market capitalization. C. Insider Hierarchy Hypothesis 3 postulates that high-level executives have an information advantage over lowlevel executives. Specifically, we assume that insider transactions by high-level executives should result in stronger announcement effects than those of low-level executives. We use two approaches to test this hypothesis. First, we compute CAARs for all three insider classes (A to C) and compare the results. Second, we perform a multivariate OLS regression with CAAR(0;1) as the dependent variable and insider classes as dummy variables. Table VI reports the results for all transactions from our final data sample. For purchase transactions, only in the U.K. does insider class A (top-level executives) practice as "stealth trading". Since we only observe transactions which are not anonymous, the "stealth trading" hypothesis cannot be fully connected to our study. Nevertheless, we assume that even in a not anonymous market insiders might hide (at least to some extend) behind several medium-size trades rather than executing one large block transaction. For further discussion on the "stealth trading" hypothesis see also Blau et al. (2009) and Lebedeva et al. (2009). 27 See also Giamouridis et al. (2008). 28 All results are available upon request. 17

19 seem to trigger the highest (significant) CAARs. The differences between insider classes are relatively small, however the CAARs show a hierarchical pattern what let us assume that insider class A is indeed best informed in the U.K. For Sweden and Italy, we find the highest CAARs(0;1) for insider class C. In Germany we find about the same results across all insider classes. In Ireland, the announcement effect is largest for insider class B. In France we do find a clear hierarchal pattern for CAAR(0;4), however, none of the results are statistically significant. Altogether, we find only incoherent results from our univariate insider hierarchy test. We further test the hierarchy hypothesis by performing a multivariate regression (Table VII). The OLS regression models include a dummy variable for each insider class, where Insider class A is the reference category. Hence, a positive coefficient for Insider class B or Insider class C indicates that the announcement effect of Insider class A is smaller in magnitude and therefore contradicts the hierarchy hypothesis. A negative coefficient for Insider class B or Insider class C indicates the opposite and consequently supports the hierarchy hypothesis. In conjunction with the hierarchy hypothesis, we also test hypothesis 4, which postulates that multiple trades by directors on the same day will give a stronger signal to the market. In particular, we include the dummy variable Multiple trade, which equals one if more than one insider trades in a firm on a given day, and zero otherwise. Moreover, we include other control variables in our regression model. From the univariate results, it is apparent that (relative) transaction size is crucial for the size of the announcement effect. Thus, we construct Transaction size, which is the number of shares traded, divided by shares outstanding on the day of the announcement. To control for firm size we construct the variable Firm size, which is the natural logarithm of the number of shares outstanding multiplied by the closing price of the underlying share on the trading day. We discuss the relation between firm size and announcement effects separately in the section below. We further include the variable Book to market, which is the book to market value at the beginning of the year when the transaction took place. 29 According to Lakonishok and Lee (2001), insiders tend to purchase stocks that are cheap according to the book to market value. We also believe 29 A high book to market value (above 1) indicates that a stock is undervalued. 18

20 that announcement effects will be larger in undervalued than overvalued stocks. Thus, we expect a positive relationship between book to market ratios and CAARs. Furthermore, we include the variable Momentum, which measures the performance of shares 100 trading days prior to the transaction day. Since company insiders seem to have a contrarian investment style, we expect a negative relationship between Momentum and post-event CAARs. We construct Momentum by calculating the market-adjusted CAARs for 100 days prior to the insider transaction. 30 We also examine whether the frequency of trading in a specific firm has an influence on the announcement effect. We expect a smaller announcement effect in firms with frequent insider trading compared to firms with less insider trading. The rationale behind this is that if insiders, for instance, trade on a monthly basis we expect the information content of these trades to be lower than for one single trade in a year. Therefore, we construct the dummy variable Frequent trading, which is set to one if the average trade frequency in a firm 12 months prior to a transaction is higher than the average trade frequency of all firms in 12 months prior to the transaction in the respective country, and zero otherwise. 31 In addition, we construct two dummy variables Previous buy and Previous sell, which control for previous purchases or sales before the respective announcement. Previous buy is set to one if at least one (large) purchase has been announced 20 days before the respective (large transaction) announcement, and zero otherwise. Previous sell is set to one if at least one (large) sale has been announced 20 days before the respective (large transaction) announcement, and zero otherwise. We also include the variable Day gap, which measures the number of days between trading and announcement days. Due to information leakage we expect directors dealings reports with large differences between transaction and announcement days to have smaller announcement effects. Thus, we expect a negative relation between Day gap and CAARs Given that Momentum controls for the market timing ability of insiders, it is important to note that it is calculated relative to the trading and not the announcement day. 31 The average trade frequency per firm and country is calculated by dividing the number of all trades 12 months prior to the transaction by the number of firms, which reported at least one director s dealings within 12 months prior to a transaction. For each country the dummy variable starts one year after the beginning of the observation period for the respective country. 32 See Betzer and Theissen (2010) for a detailed analysis of reporting delays in the German market. 19