Empirical evidence on shadow insider trading in the US

Empirical Evidence on Shadow Insider Trading in the US

PREFACE AND ACKNOWLEDGEMENTS To my family, Tessa and my friends I wish to express my gratitude to my Thesis Advisor, Professor Stefan Arping. During the last months I received his generous advice, purposeful guidance and encouragement. Statement of Originality This document is written by Student Giuseppe Lecca who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

ABSTRACT The thesis investigates a phenomenon related to insider trading called shadow insider trading proposed by Metha Reeb and Zhao (2014). It has been implemented a model partially derived from the one applied by Metha Reeb and Zhao (2014) to a dataset of daily observations from January 2013 to December 2015 of all firms listed in the United States. The outcome shows some evidences that may be connected to the presence of shadow insider trading among the dataset, however the model needs further improvements to increase the predictive power and reduce the endogeneity.

TABLE OF CONTENTS PREFACE AND ACKNOWLEDGEMENTS ... 2 ABSTRACT ... 3 TABLE OF CONTENTS ... 4 LIST OF TABLES AND FIGURES ... 5 CHAPTER 1: INTRODUCTION ... 6 CHAPTER 2: LITERATURE REVIEW ... 9 Section 2.1: Law Enforcement and the Effect of Corporate Governance on Insider Trading ... 9 Section 2.2: Relation between Informational Asymmetry and Informed Trading. ... 13 CHAPTER 3: DATA SAMPLE AND METHODOLOGY ... 16 Section 3.1: Data Sample ... 16 Section 3.2: Methodology ... 17 Subsection 3.2.1: Standard Approach to Measure Insider Trading ... 17 Subsection 3.2.2: Methodology to Measure Shadow Insider Trading ... 19 Subsection 3.2.3: Original Contribution Measuring Shadow Insider Trading. ... 21 Subsection 3.2.4: Possible Sources of Endogeneity ... 22 Subsection 3.2.5: Model Specification ... 23 Subsection 3.2.6: Controls ... 27 CHAPTER 4: RESULTS OF THE RESEARCH ... 29 Section 4.1: Summary Statistics ... 29 Section 4.2: Data Analysis ... 31 Subsection 4.2.1: First Research Question ... 31 Subsection 4.2.2: Second Research Question ... 35 Subsection 4.2.3: Third Research Question ... 37 CHAPTER 5: CONCLUSION ... 40 REFERENCES ... 41 APPENDIX ... 43

LIST OF TABLES AND FIGURES Figure 1 ... 18 Table 1 Example of the final Dataset ... 25 Table 2 Summary Statistics of all the variables ... 29 Table 3 List of the most frequent Sectors in the Dataset ... 30 Table 4 Outcome of the first research question ... 32 Table 5 Outcome of the second research question ... 36 Table 6 Outcome of the third research question ... 43

CHAPTER 1: INTRODUCTION Recently, several academic studies have proposed new forms of investigation on phenomenon related to the illegal trade of private information. This thesis investigates on one of them called shadow insider trading, proposing a particular methodology to measure it. Shadow insider trading has been theorized and measured in a recent study by Metha Reeb and Zhao (2014). The authors obtained significant outcomes over its presence in the US market by applying a proper methodology to detect it. By shadow insider trading, it is referred to the illegal exploitation of undisclosed firm’s information to realize a secure profit on the stock of another company. Alongside the insider trading, the phenomenon starts from a subject holding undisclosed information about relevant aspects of a company’s activity. The information can be gathered by exploiting insider’s position as employee of the firm or by using other illegal sources such as computer hacking or blackmailing. The difference between the two phenomena is that whereas under the insider trading model the undisclosed information is exploited on the same firm’s stock, under the shadow insider trading model the target of the trade is another company. Such target companies are highly positively or negatively correlated with the firms where the information is provided. The authors identified the target companies as business partners or business competitors. For example, assuming that the undisclosed information is collected in the firm X, under the shadow insider trading theory the insider makes a profitable trade on the firms Y and Z highly correlated with X. Metha Reeb and Zhao (2014) call the company X ‘source firm’, since the information starts from them, and companies Z and Y as ‘target firms’ since the trade takes place on them. The ratio behind the phenomenon of shadow insider trading is that it allows insiders to realize an arbitrage is reducing the risk to be liable for illegal insider trading from the authorities. In fact, even though shadow trading is assimilated to an illegal form of tipping and trading of undisclosed information Metha Reeb and Zhao (2014) stated that at the moment there is not enough control from the authorities on this phenomenon. Thus, it is still unknown its magnitude and the potential bias it may produce in the American market. This thesis will provide a modification of the model proposed by Metha Reeb and Zhao (2014), and an application to a dataset of American firms from January 2013 to December

2015. Shadow insider trading produces a set of negative externalities in the markets that are the same created by the illegal insider trading. As widely reported in the financial literature over the last decades, the phenomenon of illegal insider trading represents a major concern among stock market investors worldwide. One of the effects most recognized in several studies is that the presence of insider trading is directly related to an increase of the firm’s cost of capital (Bhattacharya and Daouk, 2002). From a financial viewpoint such consequence is negative for all the stakeholders as it undermines the firm’s activity and future development. Furthermore, as proven in recent studies a strong presence of highly profitable insider trades is linked to a weak system of corporate governance from the firm’s side and poor law enforcement of the market authority. Another significant effect that results from the phenomenon is a generalized as a loss of investor’s confidence, and that might prefer not to invest in a business environment with a high presence of information asymmetry. In fact, the models that describe the liquidity drivers in the financial market’s doctrine consider the presence of two different types of investors: informed and uninformed. Informed traders make their trades based on an informational advantage on the real fundamental value of a stock on a future time, and their payoff is expected to be positive. Conversely, uninformed or noise traders make their trades for liquidity reasons without a knowledge of the fundamental value of the stock, and their payoff follows a random variable. Given this framework, the subjects that have access to company’s classified information and exploit their advantage are the informed trades. The notion of insiders may also refer to external subjects that obtained, usually through illegal practices, undisclosed information from a company. In both cases, such investors exploit an informational asymmetry against the other pool of investors, and they have the possibility to set arbitrary trades to gain a secure profit. At this stage, insider trades may develop in two specular ways based on the signal they have a certain security. If the signal is a stock value higher than its fundamentals, they will take a short position. Conversely, a stock value under the fundamental will induce them to take a long position. In the first case informed trade will have a detrimental effect for uninformed traders as they will buy when the value is high, while in the second case they won’t realize a profit by selling a stock that is underpriced. Given this condition uninformed traders are supposed to bear all the costs of insider’s trades regarding missed earning in case the

fundamental price is higher and they sell at low price, or a loss in case the fundamental price is low and they buy high. This is one of the reasons why market authorities have tried to minimize the phenomenon. Under these conditions, it is assumed that insiders exploit their informational advantage in the proximity of periods immediately before events where the information themselves are disclosed to all investors. Several studies on insider trading and the study on shadow trading generally recognize such events as the company’s quarterly earning announcements. Beside the direct connection with increases in firm’s cost of capital and loss of confidence of uninformed investors, insider trading may also produce beneficial effects for the market. However, such effects are not fully accepted by the financial community and nowadays they are part of the debate around the role of informed trading. It is hypothesized that there may be an effect of a faster price discovery due to the presence of more informed traders in the market. Such effect is reported as having a positive impact in the market as it keeps prices closer to their fundamental values. In fact, when insiders do their informed trades against uninformed traders security prices will embed insider’s signal, and they will be closer to their real values. Given all the previous hypotheses, high levels of insider trading are recognized as a feature that generates an overall loss of confidence in the firms, in the markets and the regulatory institutions. For this reason, it is of primary interest for regulators worldwide to control and minimize the phenomenon. The first rule on insider trading is dated 1934, when the US authority for the market control, the Security and Exchange Commission (SEC), decided to regulate and prosecute the phenomenon. Nowadays, a law framework in the US and Europe are highly developed and it is proven that they contribute to significantly control the phenomenon. However, several recent studies are proposing different methodologies to measure insider trading and are exploring the presence and the magnitude of different forms of trades of private and undisclosed public information. In fact, the technological improvements on the IT sector over the last decade determined new possibilities for subjects internal and external to a given firm to have access to undisclosed classified information and to set more complex trading strategies in order to make a secure gain.

CHAPTER 2: LITERATURE REVIEW Section 2.1: Law Enforcement and the Effect of Corporate Governance on Insider Trading The first traces of regulatory production to limit and control insider trading are dated back to 1934, in the USA. Over that period, the Security and Exchange Commission promulgated a first set of rules aiming at seizing the phenomenon of insider trading. At that time, the American market authorities recognized the potential damage caused by insider trading to the investors concerning value losses. Since then, the major regulation improvements in the US has been the Insider Trading and Securities Fraud Enforcement Act in 1988 and the Stock Enforcement Remedies and Penny Stock Reform Act in 1990 (Lee at al., 2014) However, several studies debated also over the possible beneficial effect informed trading may deliver to the market in terms of increasing information in the securities due to a fastest and more accurate price discovery. Finally, in recent year several studies focused the attention on the corporate governance measures that companies set to minimize and regulate insider trading. All the different studies listed provide an overview of the regulation of informed trades from inside the firms and from the outside, from the authority’s perspective. This outline allows to understand how the phenomenon of shadow trading is important and represents a step beyond in the research. A recent paper by Ventoruzzo (2014) provides an interesting overview of the discipline over insider trading, providing similarities and differences between the EU and the US laws patterns. The regulation under the Common law system of the United States relies on a certain number of cases and by the SEC regulation under the Section 10(b) of the 1934 Exchange Act and the Rule 10b-5. Conversely, in the European Union, the discipline derives from the “Market abuse directive” approved in 2003. The core of both systems consists in imposing the obligation on the side of the firm’s insiders to either report the trade or abstain from it. According to this rule, all the insider’s trades not disclosed are declared illegals and compared to market frauds regarding liability and prosecution. As reported by the authors, the US and the EU regulations follow the same outcome previously described. However, the two disciplines differ in the concept of informed trade. In the EU, the pillar of the discipline is represented by the so-called “Parity of Information Theory”. Under such theory, all investors are supposed to have access to the same degree

of information. Thus, any investor who holds undisclosed information concerning a company should either disclose the information letting the pool of investors be aware of the trade or not exploit its informational advantage by simply not trading. Conversely, under the US rules it is required the violation of the fiduciary duty that any insider bear on the company. The ratio of such duty consists in the obligation of all firm’s insiders to not disclose private information or to make the use of them to realize a gain. Another fundamental aspect of the American doctrine consists in the so called “misappropriation theory”. It consists in the exploitation of the position of insiders to make a secure profitable gain from the informed trade. Under this framework, it can be said that the authority’s notion of insider trading results limited to its 1934 definition, with few improvements in the doctrine. In the case of the shadow trading theory, Metha Reeb and Zhao (2014) confirmed that the activity itself in the US could be assimilated to a breach of the fiduciary duty since there is a use of private information to realize a secure profit. However, since the American authorities rely only upon sentence (Dolgopolov, 2012) the phenomenon of shadow trading is still underestimated and there is not enough law enforcement against it. In perspective, it is a challenge for legislators and firms to broaden the notion of insider trading to cope with the new phenomenon of shadow trading with an adequate law enforcement on the side of the market authority and internal corporate governance measures on the side of the firms. Over the last decades, the financial literature has debated over the need to either introduce deeper rules and strengthen the law enforcement against insider trading or to let insider trades to be freer to do their trades. It is although widely recognized that the evolution of the Information Technology across the last decades allowed to new and more complex trades. As a result, new forms of private informative trades are now possible. Several financial papers starting from the late 80s up until these years provided an overview over the possible reaction to new forms of law enforcement against insider trading. Leland (1992) in his work inspected over the need for a total ban on insider trading. His study is developed over the different impact insider trades may have on liquidity in the market and the uninformed –noise- trader’s returns. Specifically, under his assumptions market suffers a higher degree of illiquidity when informed trades takes place because price shocks are interpreted by the pool of uninformed investors as a signal coming from the

leakage of undisclosed information. The combined effects for liquidity traders in this case is expected to be both negative and positive. In fact, liquidity traders will suffer insider trading regarding lower returns when they trade against an informed trader. On the other hand, liquidity traders will benefit a lower volatility of their returns because in this environment prices embed information about their real value, leading to a better choice for them. The author finds another subject namely firm’s possible gainers from insider trades that may benefit from an increase in prices due to informed trades. In another study dated 1983, Fischel and Carlton provides an overview of the effectiveness of the law pattern on insider trading. Firstly, the author criticized the way the law on insider trading compares firm’s insiders and outsiders. Such distinction is claimed to minimize the application of insider trading laws only to the insiders, excluding possible informed trades by outsiders that are involved at several degrees in the firms. As a result, only a small fraction of ‘real’ informative trading undergoes to the insider trading law definition. Secondly, it stated that –at the time- there was no interested from firms to develop internal due diligence structures to control for insider trading. In this case, several studies show that recently firms adopted forms of internal governance to reduce the phenomenon, in the attempt to reduce the cost of capital and became more attractive to external investments. Finally, the author focused on the allocation of property rights in valuable information. Such forms of tipping that involve the highest branches of company’s stakeholders are linked to insider trading and are due to a lack of regulation. Since 1983 the phenomenon of insider trading and its regulation and enforcement has changed. Nowadays, company’s efforts in controlling undisclosed information led to a new set of improvements in internal governance. Several papers analyzed the current condition on internal forms of corporate governance and provided an interesting viewpoint. In a 2012 paper, Skaife, Veenman and Wangerin analyzed the degree of insider trading in companies showing ineffective internal control over financial reporting (ICFR), according to the SEC. “Tone at the top” refers to the ethical culture in the workplace carried out by the management, rated in several studies as an essential feature to guarantee the functioning of internal governance rules. From Section 404 of the SOX under the so-called “tone at the top”, the authors derived the variable ICFR. Author’s major finding is an empirical confirmation of the hypotheses, in fact insider trading results more profitable when a firm

experience an ineffective ICFR. Secondly they found out that when a company manage to solve its inefficiency the profitability of insider trades drops. Finally, it is demonstrated that in case of a lack of “tone at the top”, the profits realized from insider trades soars. At this stage it is clear that the corporate policies are of a central role in the minimization of insider trading. In a 2000 paper, Bettis Coles and Lemmon provided an overview on the internal policies used by firms to control for insider trading. Specifically, the authors investigated over the 92% of the companies that since 1996 have adopted internal governance policies of different kind. Among all different policies, it is found out from the data panel that it is of major use the method of the so-called black-out periods. Among companies that adopted such measure, these periods are characterized by the forbidding for insiders to make any trade on the company. Conversely, insider trades are allowed over certain periods usually immediately after the quarterly earning announcements. Through the analysis of the data, the authors found out that such measure is useful for companies to significantly reduce the bid-ask spread and keep the stock liquid. However, insider trades are not wholly blocked during the blackout periods which indicates that there is still the need to develop further policies. Another study by Lee, Lemmon, Li and Sequeira (2014) investigates the effectiveness of the corporate governance rules adopted by a panel of companies aiming at reducing the presence of insider trading. The authors address also the hypotheses that under the presence of effective restrictions, corporate insiders bypass them by executing their informed trades in a particular way. Finally, the authors measured the correlation between the presence of restrictions and the level of information asymmetry among their dataset. An interesting outcome is derived from the dataset used; that covered the period 1986 to 2010. In 2010 the 70% of the firms adopted restrictions toward insider trading. About firms with restrictions, the authors found out that they suffered a lower level of information asymmetry. Information asymmetry has been calculated by combining the variables of analyst forecast dispersion, the idiosyncratic volatility and the probability of information-based trading. On the first hypotheses the empirical evidence showed that the restrictions help to significantly reduce insider trading, only on the side of the negative information. In fact, the insider’s trade of positive information is not reported to change. This outcome

result in line with the one theorized: insiders trade on positive information because it doesn’t cause loss of value for the uninformed investors. A study by Fidrmuc, A. Korczak and P. Korczak (2012), investigates over the degree of reaction of the investors to insider trading among companies with different levels of shareholder’s protection. It is hypothesized that among countries with more protection, insider trades results in a stronger reaction of the market. In fact, under the so-called information-content theory lower is the amount of informed trading in the market, higher will be the impact of insider trading. The dataset included a large number of insider’s sales and purchases across the US and 15 European countries, to control for different regulatory background. The results of the empirical analysis show that shareholder’s protection is positively correlated to the market reaction, confirming the information content hypotheses. The market reaction has been measured by the Cumulative Abnormal Return after insider’s purchase and sells. At this stage the previous studies showed that alongside the law pattern that seems to not have significantly changed over the decades, firms are incentivized by the market to lower informed trades for two major reasons: lowering the cost of capital and keep the stock liquid. Section 2.2: Relation between Informational Asymmetry and Informed Trading. The literature on insider trading has evolved over the last decades on the side of its measurements through the development of new methodologies, and on the side of the doctrine through the improvements of new definitions of the phenomenon. In particular, it can be said that a common path for many papers since the 2000s consisted in the study from different perspective the causality between the informational asymmetry and its potential profitability, and the presence of certain phenomenon within firms that might trigger them. Furthermore, a series of scandals connected to insider trading raised the attention of the public opinion on how companies themselves and the market authority should protect investors and stakeholders. Lastly, a new branch of the literature began studying new forms of insider trading derived from a different use of the information. It is the case of the shadow trading theory, described firstly by Metha Reeb and Zhao (2014) and

object of this thesis. The following papers provide an overview of the relations between the relevance of insider trading within companies characterized by particular characteristics. A relevant study that inspected over possible factors correlated to the presence of informed trading has been provided by Joseph and Wintoki (2013). The authors investigate on the link between companies characterized by a high amount of investments in advertising and the profits of insiders gains. Their methodology consisted in an OLS regression based on the Fama and French 3-factors model, where it is added the ratio between the advertising investment and sales, and the most common controls for the presence of informed trading namely the firm size, the Market-to-book ratio, the analyst following and others. More specifically, they constructed a zero-cost portfolio of firms with net insider purchases, where a long position is taken on companies with significant investments in advertising and a short position in firms without such investments. The portfolio obtained yielded an annual abnormal return of 5.5%. Such result confirms the author’s hypotheses. In fact, among companies with a level of investments in advertising, there is a positive correlation with a higher degree of informed trading. Following the results of the first assumption, the authors also hypothesized the presence of information asymmetry among the panel of companies characterized by a high level of advertising expenditures. Information asymmetry takes place in a company between the insiders and outsiders and the authors link its presence as a principal trigger for the high profitability of insider trading. This paper showed that there are certain characteristics within firms that may trigger the information asymmetry and consequently the profitability of insider trading. Other studies around the informative asymmetry are more focused on the relation it has compared to the cost of capital of the firms. In a study by Barth, Konchitchki and Landsman (2013), it is investigated the relation between corporate transparency and the cost of capital of a panel of firms. The authors implemented to the dataset an empirical measure of the corporate transparency, finding out a significant relation with the expected cost of capital. Specifically, their transparency variable is calculated following a two steps analysis: first, it is calculated and 𝑅" _derived from the annual relation between returns and earnings, selected by industry. The second 𝑅" is derived again from the annual relation between returns and earnings, selected by portfolio.

The final variable is constructed by adding up the two 𝑅"_{. The ratio of the model is that the} transparency is supposed to take place in companies where the explanatory power of a specification that includes the returns and earnings has a high value. Through the application of a regression specification that includes the Fama and French three factors model, the major finding is that there is a negative correlation between the transparency and the cost of capital. Such conclusion provides another firm’s feature that presented significant correlation with the profitability of insider trading. Beside a large number of studies focused on features that may cause information asymmetry and informed trading, another branch of the literature is focused on the consequences of insider trading regarding increase in the cost of capital. A paper by He, Lepone and Leung (2013) provides an interesting viewpoint in this perspective, investigating the effects of information asymmetry in the cost of equity capital for a firm. The hypothesis of the model is that following previous studies; there is a significant relation between the two factors. According to the author’s model, as a proxy for the information asymmetry it has been used the adverse selection cost, while for the Cost of Equity Capital the average estimated cost of capital. Consequently, they implemented an empirical model where it is tested the null hypotheses that the information asymmetry does not affect the Cost of Equity Capital. The test consists in a regression where the dependent variable is the COEC, and the independent variables are the number of analyst following, their forecast dispersion, the adverse selection cost measured as a component of the Bid-Ask spread and other controls. The outcome of the research shows that the information asymmetry is positively correlated with the Cost of Equity Capital, at a high degree of significance. In a paper by Aboody and Lev (2000), it is demonstrated that the magnitude of insider’s gains depends on the R&D expenditures. The authors found out empirical evidences of the relation between intensive investments in R&D and the level asymmetry of information within the firm. The level of asymmetry directly connected to the presence of insider trading, resulting in more profitable insider gains. Finally, it is suggested that corporate governance policies such as disclosure on the R&D expenditures and timely releases of information are able to reduce the phenomenon.

CHAPTER 3: DATA SAMPLE AND METHODOLOGY This chapter provides a description of the whole methodology applied to the dataset to address the research questions. The next section provides a description of the different sources used to collect the data and all the variables used in the final dataset that has been treated in the methodology session. Section 3.1: Data Sample The analysis developed in the following chapters has been realized on a data sample consisting of a set of daily financial observations of all firms publicly listed in the United States from January 1st _{2013 to December 31}st _2015. All data have been provided by the site WRDS, more specifically from the datasets CRSP, S&P CAPITAL IQ and I/B/E/S. The final database results from a process of merging of the three unprocessed databases, using the common variable Ticker code of each public firm and the data variable the daily trading days from January 1st, _{2013 up to December 31}st, 2015. The data source of CRSP provides the following variables: • Price: daily closing price. • Share Volume: daily number of shared traded each day. • Ask: highest daily stock price. • Bid: lowest daily stock price. • Shares outstanding: number of each company’s shares publicly held. • Return on Standard & Poor’s Composite Index: simple daily return of the S&P Composite Index. The database of S&P Capital IQ provided the following variables (quarterly updated): • Net Income • Market Value • GIC Sub-Industry Code • Total Asset • Long Term Liabilities

• Quarterly Earnings The I/B/E/S database provided the following set of variables (quarterly updated): • Return on Asset • Return on Equity • Analyst codes • Net Income Finally, as risk-free rate used for the calculation of the Capital Asset Pricing Model, it has been used the Daily Treasury Yield Curve Rates (maturity of 10 years) provided by the US Department of the Treasury. The next section provides all the information regarding the methodology applied to the database just described. Section 3.2: Methodology The methodology used in this thesis is partially derived from the previous studies on shadow trading by Metha Reeb and Zhao (2014), but it also includes a set of original features that follows some assumptions. The methodological analysis object of the previous shadow trading study is itself derived from a development of the most common approaches measuring insider trading. The next subsection provides a brief description of the method of the Abnormal Returns, which has been used in several studies to provide consistent analysis on insider trading. Subsection 3.2.1: Standard Approach to Measure Insider Trading A high number of studies focused on the measure of insider trading, used the method of the Abnormal Returns (ARs) as principal driver to quantify the presence of informed trades. For example, Aktas et al. (2008) and Fidrmuc et al. (2012) used such method to study the magnitude of the market reaction to insiders purchase and sell of stocks. ARs are themselves derived from an elaboration of the Capital Asset Pricing Model (CAPM) developed by William Sharpe in 1965. Abnormal Returns can be defined as the difference between the price of a security and its CAPM return. The ratio of the model is to estimate

on a daily basis how a public company’s stock price is drifting away from its CAPM forecasted value. The fundamental drivers of the CAPM value are the beta, namely the sensitivity of the stock price to fluctuations of the market, the risk-free rate and the market returns. If the difference between the CAPM forecasted price and the real one if significantly large, it can be interpreted as a possible proof that informed trading has taken place. Moreover, the ARs can be sum up in order to verify the persistence of a certain event in the security analyzed. In this case, we refer to as the Cumulated Abnormal Returns (CARs). According to the insider trading theory, ARs and CARs can be calculated ex-post in proximity of important announcement events such as corporate’s quarterly earnings or just before any possible event arising from the firms that may drive significantly the stock price away from its fundamental value. Figure 1 Figure1 shows how the insider trading detection model works. The time periods goes from a period of some days or a month before and after the quarterly earning announcements (EA). The CAR variable is measured few days before the EA, when according to the insider trading theory the insiders places their trades in order to obtain the maximum gain from the informational asymmetry. The AR in Figure 1 indicates the Abnormal Return in correspondence of the earning announcement and it is used to measure how the market behave when the information is disclosed to all investors. A strong correlation between the CAR and the AR is interpreted as a sign of the presence of informed trading. However, despite its fundamental role just described, the Abnormal Returns can’t be directly linked in a causality relation with the presence of informed trading. In fact, as shown in the literature review a common approach to investigate the phenomenon consists in the development of a model that hypothesizes a causality connection between a certain factor and its positive or negative correlation with insider trading. However, there is a wide EA CAR AR Time

range of phenomenon that may cause heavy fluctuations in the security value and produce a biased outcome of the research. Periods of market distress or strong signal around the financial situation of the firms may cause significant shifts in the prices regardless the presence of informed trading. Therefore, a challenge for the researchers consists of finding an appropriate methodology to control as much as possible for all such event. The aim is to obtain a model where the fluctuations in the CAR can be explained with a high degree of significance by the informed trading. Subsection 3.2.2: Methodology to Measure Shadow Insider Trading Despite the theory of insider trading accounts for a wide number of methodologies, the shadow insider trading theory relies only on the model provided by the paper of Metha, Reeb and Zhao (2014). Under the shadow insider trading theory proposed by the authors, it is hypothesized that an insider who holds an undisclosed information on his company exploits his advantage by doing a profitable trade in another company. The firms where the information is taken and the firm object of the informed trade are each other related as business partners or business competitors. At this stage, the firms from whom the information is leaked are referred to as ‘source’ firms since it is hypothesized that the informational advantage arises from them. Consequently, companies where the profitable trade of the source firm information takes place -business partners and competitors- are called ‘target’. The aim of the research of Metha Reeb and Zhao (2014) is to find a significant causal relation between the Cumulated Abnormal Returns of target and source firms in a period before the source firms quarterly earning announcements. Such significance is interpreted as a confirmation that the theory of the presence of shadow insider trading among the US market holds. The presence of shadow insider trading has been measured by Metha Reeb and Zhao (2014) using an OLS regression as follows: 𝑆ℎ𝑎𝑑𝑜𝑤 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 = 𝛽₁+ 𝛽_"𝑃 𝐶𝐴𝑅 + 𝛽₆𝐶 𝐶𝐴𝑅 + 𝛽₇𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝜀

Where: • 𝑆ℎ𝑎𝑑𝑜𝑤 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 Is the dependent variable. It has been calculated as the ratio of the average daily short sales of the target firm in a time window of 30 days before the source firm quarterly earning announcements and the average short sales over the rest of the year outside that period. • 𝑃 𝐶𝐴𝑅 The first regressor is the source firm Cumulative Abnormal Returns on a time window of 3 days before and after the t quarterly earning announcements (from t-1 to t+1) when the company is a business partner. • 𝐶 𝐶𝐴𝑅 The second regressor is the source firm CAR on a time window of 3 days before and after their quarterly earning announcements (from t-1 to t+1) when the company is a business competitor. Metha Reeb and Zhao (2014) model has been run on a panel data of US companies. They found a significant causality between the dependent variables Shadow Trading and the two regressors. The source firm’s CARs used in the model cover a time interval of three days around the quarterly earning announcements because according to the doctrine on insider trading, it is the period where all the undisclosed information became common to all investors. Therefore, the causality between the CARs and the Shadow Trading variable indicates that the information disclosed on the source firms during their earning announcements are able to describe part of the abnormal short selling of the target company. This evidence according to the authors is the proof that a certain amount of source firm’s insiders exploits their informational advantage by realizing profitable short selling trades on the target firm. This interpretation results in accord with the theoretical of shadow insider trading. The nature itself of the model specifications requires a high number of controls. In fact, the regression may be affected by a set of endogenic variables correlated with the independent variables that might bias the estimation. The variables included in the controls are similar to the ones used in several OLS regression model to measure insider trading. Specifically, they can be sorted out in controls of the target firm’s financial condition such as firm size and Return on Asset, controls of the analyst activity such as the forecast dispersion and the number of analyst following and controls on market factors such as the price volatility. By

including these controls the outcome of the regression should return an unbiased estimation. Subsection 3.2.3: Original Contribution Measuring Shadow Insider Trading. The methodology applied in this thesis is inspired by the one proposed by Metha Reeb and Zhao (2014), although some important modification has been done. The first and most important modification regard the hypotheses of the distinction between business partners and competitors. The authors created two variables that divided all source firms into two groups of business partners and business competitors with respect of the target companies. To do so, Metha Reeb and Zhao (2014) used a dataset and a manual of manual picking of the partners and competitors. Such methodology causes a limitation in the size of the dataset and may carry errors in the case of companies that can be considered at the same time business partners and competitors, for example Samsung and Apple. In this thesis it has been decided not to create two separated variables, rather a Dummy variable that takes the value 1 in case target and source are considered business partners, 0 if competitors. The Dummy will be described in detail in the Subsection 3.1.5. The second significant difference regards the variable that links in each regression the target and source firms. Specifically, the variable used in this thesis is the Sub-Industry Global Industry Classification code (GIC Sub-Industry). Since shadow insider trading is hypothesized as a leakage of private information on a company, such condition will involve companies operating in the same sectors. In this thesis it is assumed that both competitors and business partners belong to the same sub-sector. In fact, it is hypothesized that stock prices of companies in the same sectors are more likely to be object of shadow insider trading. For example, under the shadow insider trading theory an insider holding private information on Apple is supposed to trade on business competitors or partners that operate in the same sector. This methodology allowed to couple automatically source and target firms, resulting in a large and more consistent dataset. Another significant difference in the methodology of this thesis regards the dependent variable. In this thesis it has been used the CARs of the target firms instead of the index ‘Shadow Trading’ used by Metha Reeb and Zhao (2014) to measure directly the reaction of the source firm CAR on the target.

Furthermore, the time horizon for the calculation of the Cumulative Abnormal Returns has been modified from the Metha Reeb and Zhao (2014) model. In fact, they built their variable using CARs of a window from the day before the quarterly earning announcements of the source firm up to one day after them. In this thesis it has been decided to use a time window starting from the 4th day up to one day before the quarterly earning announcements of the source firm. There are two major reasons for this choice. The first one is to reduce the cases of overlapping of the CARs in correspondence of other companies earning announcements. If so, the regression cannot be taken into account because the outcome of the target firm is biassed by other factors, endogenous to the model. Such phenomenon should be minimized in order avoid source company’s CAR to be influenced by external events. By clustering firms into groups with the same GIC Sub-Industry Code, it is significantly reduced the case that the quarterly earning announcements of source and target firms happen on the same time. Secondly, in this thesis it has been decided to create one independent variable that accounts for the Abnormal Returns from day t-4 to t-1 and another variable that accounts for the Cumulated Abnormal Returns from t to t+1 of the source firm quarterly earning announcements. In this way it is possible to measure the impact on the dependent variable of the information when it is disclosed. In fact, in correspondence of the Cumulated Abnormal Returns from t to t+1 all investors are supposed to have access to the information that is revealed and translated into price. Finally, it has been created an interaction variable calculated as the product between the CAR from t-4 to t-1 and the Dummy variable. Such variable allows to estimate the relationship between target and source firm’s CARs in the case of a partnership relation. Subsection 3.2.4: Possible Sources of Endogeneity This subsection describes the possible sources of endogeneity of the regression model implemented in this thesis. The major concern for potential sources of endogeneity may arise from particular conditions of market distress where systemic events take place. Under this condition, a certain sector or the whole market may receive a signal, driving informed and uninformed investor’s trades in the same direction resulting in a significant price

change. In such cases, prices may drift away from fundamental values, and extreme ARs may be registered even though there is not a clear sign of illegal insider trading. The most common cases of market turmoil are spillover effects resulting from distress in one or more major sectors or systemic events whose effects are spread out the markets worldwide. Examples of such cases are represented by periods of market distress caused by a negative event such as the burst of a stock market bubble, or bearish periods caused by external events such as an economic crisis or political instabilities. Sectorial crises such as fall in energy commodity’s prices are another example of external –in this case involving one or more sectors- shocks. Under this example, the ARs may assume high levels resulting from the shocks and the outcomes of the models would produce a false positive case of shadow insider trading. For this reason, the model implemented in this thesis uses several controls and in the last research question it has been implemented a regression on different sectors to inspect over possible significant differences caused by endogenic drivers. Subsection 3.2.5: Model Specification This subsection explains how Cumulative Abnormal Returns (CARs) are computed for each public firm in the data panel. Starting from the daily closing price of the panel of public companies listed in the US, in the first steps the Capital Asset Pricing Model (CAPM) is calculated, and the returns obtained are used in the estimation of the Abnormal Returns. In the first step are calculated the daily logarithmic returns from each company’s stock closing price. The formula applied is the following:

LogReturn_E,G = ln (𝑃𝑟𝑖𝑐𝑒_E,G 𝑃𝑟𝑖𝑐𝑒_E,GL1)

Where ‘i’ is the entity (each public company) and ‘t’ is the time variable (daily). Once the calculation is done the dataset has been set as a panel data, using as entity variable the company’s TIC code and as time variable the daily trading day for each year from January 2013 up to December 2015.

Secondly, the dataset has been expanded by adding a risk-free rate namely the yearly daily average of the 10-years maturity US Bond. At this point, it has been calculated the CAPM Beta of each company. To obtain a beta coherent as much as possible to the fluctuation of the stock, it has been calculated for each firm a set of rolling regressions using a three months time window. The linear regression follows the model:

𝐿𝑜𝑔𝑅𝑒𝑡𝑢𝑟𝑛E,G = 𝛽P+ 𝛽1𝑆&𝑃𝑅𝑒𝑡𝑢𝑟𝑛G+ 𝜀E,G The 𝛽1 resulting from the computation of the regression has been saved and they’re used in the next step for the calculation of the CAPM returns. Finally, CAPM returns are calculated as follows: 𝑅_RSTU,E,G = 𝑟_V+ 𝛽_W(𝑅_X− 𝑟_V)

Where 𝑅X represents the Standard and Poor’s daily returns and 𝑟V are the 10 years maturity bonds. 𝑅RSTU,E,G are the CAPM returns for the i company at time t.

Next to the calculation of the CAPM returns, it is derived from each company in the data panel, the ARs and CARs.

ARs are calculated following the formula:

𝐴𝑅E,G = 𝑅E,G − 𝑅RSTU,E,G

Where 𝑅E,G are the daily returns for company ‘i’ at time t, and 𝐴𝑅E,G are its correspondent daily Abnormal Returns. The CARs are calculated as the sum of the ARs over a time window of four days. Even in this case, the result obtained is referred to each company for each trading day. The formula is as follows: 𝐶𝐴𝑅_E,G = 𝐴𝑅_E,G GL1 GLZ

The first data panel calculated so far consists of a set of public company each one having for each trading day its logarithmic price returns, its CAPM returns and Abnormal Return and Cumulated Abnormal Return from January 2013 up to December 2015. Furthermore, for each company of the panel it has been included the reported date of quarterly earnings. In fact, according to the shadow trading theory around earning announcement dates firm’s insiders are keen to exploit their informational advantage by setting profitable trades on target companies, categorized as business partners or competitors. For example, if an insider holds information over a drop in its firm’s revenues it is assumed he would set up two different strategies: a short position on a business partner stock, or a long position on a business competitor. The ratio is that business partner’s stock is supposed to be positively correlated with the firm whose information is exploited namely the source firm, and negatively correlated with the business competitor’s stock. The final specification takes into account all the assumptions made so far, and the final OLS regression is set up by adding the assumptions on the target and source firms. In fact, it is necessary to link to each company in the panel a source company with its own ARs and CARs. The following example in Table 1 shows a partial version of the final dataset obtained. Table 1 In this case target and source firms belong to the GIC Sub-Industry sector of ‘Technology Hardware, Storage and Peripherals’. In the left the Date variable indicates the date of source firm’s quarterly earning announcement. Continuing from left to right, the variable ‘ticx’ is Variables

date ticx CARx ARx tic AR CAR

19-Mar-13 AAPL .0681322 -.0099302 ALOT .0017281 -.0311361 22-Mar-13 AAPL .0374549 .0222854 DRAM .0577795 174.357 21-May-13 AAPL -.0298562 -.0135024 NTAP -.0023508 .0165553 21-May-13 AAPL -.0298562 -.0135024 ALOT -.0144698 -.0148993 22-May-13 AAPL -.0016916 -.0027066 HPQ .1653431 -.0093793 07-Jun-13 AAPL -.0481142 -.0155394 CRDS -.0047467 -.0891907 09-Jul-13 AAPL -.0165376 -.0016098 NCR .0121638 -.0112106

the ticker code of the target company in this case Apple. The next two variables refer to the CAR and the AR of the target company, in this case Apple in correspondence of the dates of source firms earning announcements. The following variable indicates the GIC Sub-Industry code, used as criteria to merge in each regression target and source firms. Finally, the variable ‘tic’ is the ticker of the source firm and the last two variables indicate its CAR and AR in correspondence of its earning announcement. As an example of the outcome, the source firm HPQ made on May the 22nd _{2013 an earning announcement. CARs and ARs of} target and source firm are displayed in Table1. The new dataset obtained allows calculating through an OLS regression model, the presence of shadow trading in the market. Specifically, the regression for each target firm X and source firm Y used in this thesis and partially derived from the methodology of Metha Reeb and Zhao (2014) is as follows: 𝐶𝐴𝑅_{7,[W\]_W]](_)} = 𝛽_P+ 𝛽₁𝐶𝐴𝑅_{_,[W\]_W]](_)}+ 𝛽_"𝐴𝑅_{_,[W\]_W]](_)}+ 𝛽₆𝐷𝑢𝑚𝑚𝑦_{_,7} + 𝛽Z𝐷𝑢𝑚𝑚𝑦_,7 ∗ 𝐶𝐴𝑅_,[W\]_{dee f} + 𝛽E𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠7,[W\]_W]](_)+ 𝜀E Where: • 𝐶𝐴𝑅_{7,[W\]_W]](_)}: target firm X Cumulative Abnormal Returns from day t-4 to day t-1 before firm Y quarterly earning announcement. It is the dependent variable of the model. • 𝐶𝐴𝑅_{_,[W\]_W]](_)}: source firm Y Cumulative Abnormal Returns from day t-4 to day t-1 before its own quarterly earning announcement. • 𝐴𝑅_{_,[W\]_W]](_)}: source firm Y Cumulated Abnormal Return from the day of its own quarterly earning announcement, to day t+1. • 𝐷𝑢𝑚𝑚𝑦_{7,_}: This variable has been included in order to hypothesize the relation between the target and source firms. Specifically, it indicates if the companies are considered business partners or competitors. The variable takes the value 1 if the product between the source firm’s 𝐴𝑅_,[W\]_W]](_) and target firm’s 𝐴𝑅7,[W\]_W]](_) is positive. Conversely, it takes value 0 if the same product is negative and the firms

are hypothesized to be competitors. It is assumed that if from day t of the earning announcement until day t+1 firm’s X and Y CARs have the same sign, they are considered as business partners (Dummy equal to one). Conversely, if the two companies present CARs of different sign, they will be considered as business competitors (Dummy equal to zero). • 𝐷𝑢𝑚𝑚𝑦_{7,_} ∗ 𝐶𝐴𝑅_{_,[W\]dee f} : This interaction variable indicates the impact of source firm’s CAR in the dependent variable, when the companies are business partner (Dummy equal to 1). This independent variable is very important for the model because it addresses the hypotheses that business partner’s CAR and target CAR are positively correlated. More specifically, it allows estimating if the signal of the source firm’s informed trader can be exploited on the target firm. • 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_{7,[W\]_W]](_)}: It has been added a set of controls of the target firm X in correspondence of source firm Y quarterly earning announcements. Subsection 3.2.6: Controls In this subsection is described the set of controls that has been included in the final OLS specification. As previously discussed, there is some controls that need to be added in order to minimize the possible endogeneity in the OLS regression. In fact, it is a high degree of likelihood that the dependent variable may be driven by external phenomenon that may cause endogeneity and produce a biased outcome. Most of the controls included in the specification are taken from the Metha Reeb and Zhao (2014) model. The following list shows all the controls added to the final dataset. • Firm size: it is calculated as the logarithm of the average yearly Total Asset. • Return on Asset (ROA): Net Income divided by the Total Asset (yearly average). • Leverage: Long Term Liabilities divided by the Total Asset (yearly average). • Bid Ask spread: measured as the difference between the daily Ask price and Bid price.

• Amihud illiquidity ratio: LogReturnE,G

• Analyst Following: measured as the average per year of analyst following target the firm X (yearly average).

CHAPTER 4: RESULTS OF THE RESEARCH Section 4.1: Summary Statistics This section provides a brief overview of the descriptive statistics of the final dataset object of the study in this thesis.In the following table are displayed the summary statistic of all the variables included in the OLS regression. The column to the left indicates the variable of the specification, while the upper row indicates the descriptive statistic tool. The final dataset contains 1660990 observations in the dependent and independent variables and a lower number of the controls caused by missing values in the data for all the years considered. The dataset covers the interval from January the 1st _{2013 to December} the 31st _{2015. About the population, the dataset accounts 4121 publicly listed American} firms. Table 2 A remarkable feature of Table 2 regards the value of the of the source firm’s CARs and ARs. They are all negative, indicating that on average from 2013 to 2015 companies had underperformed on their CAPM returns. This outcome can be interpreted as the result of an overall bearish period of the market across the period, and as the behavior on average for Stats

Variables N mean variance min max p50

CARj,klmn_lnn(o) 1660990 -0.0023 0.0074 -2.272 5.653 -0.0023

CARo,klmn_lnn(o) 1660990 -0.0009 0.0055 -1.100 1.743 -0.0017

ARo,klmn_lnn(o) 1660990 -0.0067 0.0101 -1.811 2.160 -0.0048

Dummyo,j 1660990 0.5376 0.2485 0 1 1

Dummyo,j∗ CARo,klmnstt u 1660990 -0.0004 0.0029 -1.100 1.7435 0 Log Amihud Illiquidity ratio 1654801 -10.41 401.072 -17.10 -0.3384 -10.51 Log Asset Total 1659189 1.711 0.1853 -6.685 2.557 1.742 Analyst Following 1522209 72.86 9368 0 817 37 Bid/Ask spread 1660990 1.014 5.933 0 442.03 0.53 Leverage 1660990 0.27356 0.2235 0 16.08 0.1787 ROA 1660990 -0.0286 1.215 -9.071 56.62 -0.0064

the investors to sell their stocks before and during the firm’s quarterly earning

announcements. Furthermore, the variables 𝐶𝐴𝑅_,[W\]_W]](_) and 𝐴𝑅_,[W\]_W]](_) result on average positively correlated indicating that the Abnormal Returns from t-4 to t-1 bears a signal about the information undisclosed in the quarterly earning announcements. It is also noticeable that the average CAR of the source firm has a lower variance than its equivalent of the target firms. Such stability in the CAR of source companies in a proximity of their own earning announcements is caused by less variation between the log returns of each source company stock and the CAPM returns. It may be interpreted as the tendency for the stocks to be less sensitive to the normal fluctuation of the market few days before its earning announcements. Conversely, the larger variance showed by the target firm’s CAR variable may be interpreted as a more volatile variation in correspondence of other firm’s earning announcements. Table 3 shows the 20 largest sectors in the sample by a number of firms, with their correspondent percentage of the total. Table 3

GIC- Sub Industry Code Number of firms Percentage

Biotechnology 334 14,19% Internet Software & Services 158 6,71% Pharmaceuticals 139 5,90% Oil & Gas Exploration & Production 137 5,82% Health Care Equipment 136 5,78% Application Software 118 5,01% Semiconductors 110 4,67% Communications Equipment 94 3,99% Oil & Gas Storage & Transportation 91 3,87% Industrial Machinery 88 3,74% Restaurants 65 2,76% Oil & Gas Equipment & Services 63 2,68% Electronic Equipment & Instruments 62 2,63% Aerospace & Defense 59 2,51% Systems Software 51 2,17% IT Consulting & Other Services 50 2,12% Electrical Components & Equipment 49 2,08% Health Care Services 48 2,04% Gold 45 1,91% Semiconductor Equipment 44 1,87%

The sector of Biotechnology results the most populated one by a large margin (14.19%), followed by Internet Software & Services (6.71%), Pharmaceuticals (5.90%) and Oil & Gas Exploration & Production (5.82%). This outcome indicates that the Biotech industry has the largest number of public firms in the US on the total. However, this result doesn’t indicate that it is the most important sector because it hasn’t been considered the Market capitalization of such firms on other sectors. Furthermore, the GIC sub industry classification applies a criterion that produce more division in other sectors (for example the Oil & Gas is split into 3 groups), and remain more generalist in other cases (for example Pharmaceuticals and Biotech). Section 4.2: Data Analysis The data analysis in this thesis has been developed by using the Ordinary Least Squares Regression methodology. Specifically, a set of regressions has been implemented to the final dataset from 2013 to 2015 to address all the research questions. The dataset is not a data panel because as shown in Table1, there is more observation over the same date in case more source firms have their earning announcements in the same dates. Furthermore, to avoid biassed outcomes, from the final dataset it has been deleted all the dates where the earning announcements of target and source firm happened on the same date or in the days used to calculate the dependent variable CAR. Subsection 4.2.1: First Research Question The first research question inspects over the possible relation of causality between the CAR of the target firms as dependent variable, and the CAR of the source firms earning announcements. The regression also includes as independent variables the Cumulated Abnormal Returns of the day of the source firm earning announcements and the day after, a Dummy that indicates whether the two firms considered are supposed to be partners of competitors followed by an interaction variable calculated as the product of the Dummy and the source firm’s CAR. Finally, a set of controls has been added to reduce the degree of

endogeneity. Table 4 shows the outcome of the first regression specification, applied to the dataset from 2013 to 2015. The specification of the regression is the same as page 25. Dependent Variable: 𝐶𝐴𝑅_,[W\]_W]](7) Coefficients: CARo,klmn_lnn(o) 0.0879*** (51.92) ARo,klmn_lnn(o) 0.0013* (1.97) Dummyo,j -0.0005*** (-4.23)

Dummyo,j∗ CARo,klmn_lnn(o) 0.0331***

(13.43) Log Amihud Illiquidity ratio -0.0011*** (-18.20) Log AssetTotal 0.0071*** (22.33) Analyst Following -0.0001*** (-42.35) Bid/Ask Spread 0.0036*** (27.87) Leverage -0.0038*** (-22.60) ROA 0.0007*** (11.65) Constant -0.0248 (-50.26) t-statistics in parentheses * p<0.05, ** p<0.01, *** p<0.001 Table 4

The OLS regression’s coefficients in Table 4 are robust to heteroscedasticity. The regression accounts for roughly 1.500.000 observations. The F-stat that measures the hypotheses that all the independent variables are jointly zero has a value of 1128.37, which indicates that the set of regressors are together strongly significant in the regression. The 𝑅" _{of the} regression has a value of 1.57%, which indicates that most of the variability of the dependent variable is not explained in the regression. A plausible reason for such outcome is that target firm’s CARs are mainly driven by other market factors (such as bullish or bearish periods) rather than from the independent variables included. The analysis of the independent variable’s coefficients is as follows: 𝐶𝐴𝑅_{_,[W\]_W]](_)}: Source firm’s CARs in correspondence of their own earning announcements shows a positive coefficient of 0.0879 and its p-value indicates a strong degree of significance of the regressor. Therefore, an increase of one point the Cumulated Abnormal Returns of the source firm produces a 0.0879 increase of the source firm’s CAR, with a high degree of significance. 𝐴𝑅_{_,[W\]_W]](_)}: The Cumulated Abnormal returns in correspondence of source firm’s earning announcements has a beta of 0.0013 and it is significant at 5% level. This variable has been included in the regression to estimate how the value of the signal when disclosed to the market influences the dependent variable. The significance of a lower degree of this regressor indicates that the trades when the information of the earning announcement is disclosed to the market have a low but significant power to forecast target firm’s CAR. Under the assumptions of the thesis, this outcome represents a confirmation that with a 5% level of significance, target firm’s CARs are influenced ex-post by the abnormal returns in correspondence of the days of source firm’s earning announcement. 𝐷𝑢𝑚𝑚𝑦: The Dummy variable indicates that if the companies are considered business partners, it influences negatively the dependent variable with a high degree of significance. In this case, the outcome provides a conclusion different from the one hypothesized, despite the lower value of the regressor (-0.0005). As described in the variables description under the Methodology Section, it is assumed that each regression would either involve firms considered as business partners or competitors. Target and source firms were selected as business partners if their ARs in correspondence of source firm’s earning announcements drifted in the same direction. In this case, the product between the variable would have

been always positive. Conversely, business competitors were selected in the opposite case namely when the variable ARs drifted apart, and in this case the product of the two ARs would have assumed an opposite sign. Under the assumption of shadow trading, a source firm marked as partner would be positively correlated to the target’s CAR. The reason is that insiders would set arbitrary trades by assuming the same position they would have on the source company, toward the target company. Conversely, if source and target firms were rated as business competitors the trades would take opposite signs. In fact, in this case a good news for the source is bad news for the target competitor and vice versa. For this reasons, the negative value of the Dummy indicates that there is room for improvement in the model to separate partner and competitor companies in each regression. 𝐶𝐴𝑅_{_,[W\]_W]](_)} ∗ 𝐷𝑢𝑚𝑚𝑦: Such interaction variable indicates the value of the source firm’s CAR in case the companies are partners, zero otherwise. The value of 0.033 indicates a positive impact when the companies are partners, and it is strongly significant. This coefficient is of remarkable importance because it shows that partner companies are positively correlated to each other. Controls: All the controls are strongly significant. Specifically, we find a negative relation between the dependent variable with the logarithm of the Amihud illiquidity ratio, the Analyst Following, and the Leverage. It is possible to deduce that more stock is illiquid, lower is its CAR. Analyst following also contribute negatively to the target’s CAR, but with a small impact. Conversely, a positive relation is found with the independent control variables of the Bid Ask spread, ROA and Asset Total. In this case the outcome seems to confirm the general economic theory, to the extent that the variables are positively correlated to with the target’s CAR. Finally, this first specification model provides some empirical evidence that can be connected to the presence of shadow insider trading. On the first stage, all the independent variables are jointly and separated strongly significant with the only exception of the 𝐴𝑅_{_,[W\]_W]](_)}. This outcome indicates that the independent variables are useful predictors of the dependent variable, even though the 𝑅" _{shows that a major part of the variability is} explained by factors not included in the specification. The value and significance of the interaction independent variable 𝐶𝐴𝑅_,[W\]_W]](_) ∗ 𝐷𝑢𝑚𝑚𝑦 provides the most important proof that confirms the theory of shadow insider trading of this model. In fact, insiders

holding positive private information about the source firm and trades it on the target firm’s stock from day t-4 to the dayt-1 from the source firm earning announcement, will be able to generate an arbitrage profit. This arbitrage model is confirmed by the positive and significant coefficient of the variable 𝐶𝐴𝑅_,[W\]_W]](_) ∗ 𝐷𝑢𝑚𝑚𝑦, and possible sources of endogeneity are minimized by the use of a wide set of controls included in the specification. At this stage, the investigation needs to address the second research question, which investigates more in detail the role of the variable 𝐷𝑢𝑚𝑚𝑦. Subsection 4.2.2: Second Research Question The first research question has demonstrated that target company’s CAR variable is influenced by the source firm’s CAR and when it holds a relation of partnership, such relation is positive. This outcome is in accord with the hypotheses that informed traders may place their trades on a target company related to its own (source) company. The specifications of this Subsection inspect over possible differences between the outcomes across two OLS regressions, the first one obtained by filtering the companies considered business partners and the second one when they are considered as business competitors. The specification applied is as follows: 𝐶𝐴𝑅_{7,[W\]_W]](_)} = 𝛽_P+ 𝛽₁𝐶𝐴𝑅_{_,[W\]_W]](_)} + 𝛽_"𝐴𝑅_{_,[W\]_W]](_)} + 𝛽_E𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_{7,[W\]_W]](_)} + 𝜀E Regression 1 has been run with the condition of the Dummy equal to zero –indicating that the companies are business competitors-, while in regression 2 the Dummy is equal to 1 –in this case business partners-. Table 5 shows the outcomes of the model.

Table 5 All the coefficients of Table 5 are robust to heteroscedasticity. The regressions of Table 5 display similar outcomes, and all variables are significant at 1%. The regressions provide some interesting outcomes. Firstly, the CAR of source firms impacts more companies that are rated as business partners and it is significantly larger than the case of competitors companies, with values of 0.1212 in the first case and 0.0878 in the second. This outcome seems to confirm that target companies rated as business partners are more influenced in a positive relation with source firms CARs before its own earning announcements. Conversely, in the case target and source firm are business competitors, the impact is lower. It is expected that if the companies are competitors, insiders would place trades of an Dependent Variable: 𝐶𝐴𝑅7,[W\]_W]](_) Regression 1, Dummy=0 Regression 2, Dummy=1 Independent Variables: Coefficients: Coefficients:

𝐶𝐴𝑅_,[W\]_W]](_) 0.0878*** 0.1212*** (51.78) (67.85 𝐴𝑅_,[W\]_W]](_) -0.0166*** 0.0170*** (-16.26) (18.20 Log Amihud Illiquidity ratio -0.0009*** -0.0011*** (-11.09) (-14.43) Log AssetTotal 0.0078*** 0.0061*** (16.26) (14.74) Analyst Following -0.00003*** -0.00002*** (-29.05) (-30.50) Bid/Ask Spread 0.0040*** 0.0032*** (18.35) (21.58) Leverage -0.0039*** -0.0037*** (-14.83) (-16.92) ROA 0.0007*** 0.0006*** (8.04) (7.81) Constant -0.0255 -0.0242 (-35.21) (-36.78) Number of Observations 699574 817612 * p<0.05, ** p<0.01, *** p<0.001

opposite sign on the target companies due to the negative correlation between a couple of firms considered. The second independent variable, the Abnormal Returns from day zero to day one from the earning announcements, present an output coherent with the assumption of shadow insider trading. In fact, when source and target firms are competitors the 𝐴𝑅_,[W\]_W]](_) variable displays a negative relation while in the partners case, the relation is positive. On the controls there are no significant differences between the two regressions. The reason is that the division between competitor and target firms do not produce any difference in the background characteristics of the two samples. This outcome provided another important point in support of the theory of shadow insider trading. The coefficient of the 𝐴𝑅_,[W\]_W]](_) variable in case the companies are business competitors, shows that the signal when the information is disclosed is negatively related to the dependent variable. In this case holds the theory that an informed investor that knows a negative (positive) signal on the source firm company would place a trade in a competitor target firm positive (negative) to realize a profit. The ratio is that a negative (positive) signal from the source firm can be profitably traded with a long (short) position on the competitor target firm’s stock. The first two research questions provided consistent proofs on the presence of shadow insider trading in the market and the specifications included a wide number of controls to reduce the endogeneity. However, it has to be taken into consideration that a large percentage of the observation used in the OLS regressions regards companies of the Biotechnology sector, causing an outcome unbalanced toward it. The next research question aims at providing different regressions for the largest sector. Subsection 4.2.3: Third Research Question The next step consists in the application of the model with respect to different sectors. It is of primary importance to investigate whether or not there are significantly different outcomes in the application of the model to different sectors. This question helps providing answers on a possible unbalanced outcome in the first specification and studying an hypotheses of corporate finance. In fact, as reported by several corporate finance studies