Investor inattention to regulatory announcements for new drug applications

(1)

Investor inattention to regulatory announcements for new drug

applications

Redmar de Boer (5732131)

University of Amsterdam, Amsterdam Business School

MSc Business Economics, Finance track Master’s thesis

August 2015 Supervisor: Tolga Caskurlu

(2)

Statement of originality:

Student Redmar de Boer who declares to take full responsibility for the contents of this document writes 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.

(3)

Abstract

This thesis examines the stock price reaction on drug application decisions made by the EMEA and FDA regulatory agencies. It also investigates whether there is inattention in publications made by the regulatory agencies on a Friday. Based on prior research, this thesis expects to find investor inattention for these Friday announcements. This thesis finds no statistically significant effect in Friday announcements and, hence, no evidence for investor inattention.

(4)

For pharmaceuticals, the approval ends a period of uncertainty and a long track of extensive clinical testing. On average, these clinical trials can take six to eleven years and costs of $403 million per drug (Dimasi et al., 2002). With a regulatory approval, either from the FDA in the United States or the European Medicines Agency (EMEA) in Europe, pharmaceuticals can bring the product to the corresponding market. It is possible to market the drug in an other continent as well, but then the corresponding regulator must test the drug application and this again is a expensive and time-consuming period. An approval granted by the FDA (EMEA) does not instantly grant it with an approval by the EMEA (FDA). Without such an approval, the investment is worthless, as the company cannot recover the investment costs of the research and development for the drug on the market. The pharmaceutical company then needs to make adjustments in e.g. the dosage form. However, this depends on the reason why the FDA/EMEA rejected the drug in the first place. And if the company chooses to make adjustments, it again has to engage in expensive clinical trials with the changed dosage. Otherwise, when the pharmaceutical company chooses not to make adjustments, it has to write off this multi-million-research investment.

Not only pharmaceuticals anxiously await the regulatory decisions, investors also await such news as it provides them with a possible trading opportunity. The investors see the approval as a good news event. An approval has, in general, a positive effect on the stock prices and, in some cases, to extreme height (e.g. the quote at the beginning of this thesis).

Knowledge of the price change is important for pharmaceutical managers responsible for new products, who need to generate sensible expectations about new drugs (Sharma and Lacey, 2004). There is a great need for calculating the correct stock price after such regulatory announcements. For managers this price gives insight in how well the research and development department is functioning, which implications it has for the firm’s future cash flows and it can strengthen the position of the company in the market. Little public information is available for research and development activities, but these research developments have the potential to generate great amounts of money in the future. Therefore, it is interesting to examine whether this extreme price change reflects new and unanticipated news or whether the stock price overreacts to such regulatory decisions.

Preceding research on regulatory announcements results in a debate on the Efficient Market Hypothesis (EMH) from Fama (1964). For example, Perez-Rodriguez and Valcarcel (2011) examine the stock prices of the 17 largest pharmaceutical companies in the United States. In their sample, only 6% of the FDA approvals create an abnormal return and there is no violation of the EMH (2011).

Other research, such as the work done by Bosch & Lee (1994) and Sarkar & De Jong (2006), finds abnormal returns and a violation of the EMH. In these papers, most data in the sample consists of FDA

1

(6)

approvals. Because of legal reasons, the FDA does not publish rejection data on their website or database. In some cases there is however, a news announcement available. To find these rejections, one needs to perform extensive research on the Internet by using a set of different keywords. Then examine the validity of the announcements by crosschecking the event dates with the FDA calendar and website information.

This thesis will add to the existing literature by using data of both FDA and EMEA regulatory decisions. Almost all research focuses on the pharmaceutical market in the United States and the FDA drug application announcements. Furthermore, the rejection sample in this thesis is larger than in previous research. Not only does the data consist of FDA rejections, but also EMEA rejections are included in the sample. The bigger sample might help in examining an overreaction effect, if present, in the regulatory decisions and might give more insight on the behaviour of the market in the different outcomes.

The two rejection samples also provide the opportunity to examine whether the market reacts differently to a drug rejection from the FDA or a rejection from the EMEA. Investors might not monitor the European regulator as extensively as the FDA. The regulator in the United States follows a different

procedure for testing and approving drugs than its European counterpart.

There are different explanations for overreaction. For example, the overreaction can be the result of attention-driven buying behaviour. A regulatory announcement receives voluminous media coverage and investors pick this up. This increased coverage can create a situation were investors buy the stock because they do not want to miss out on a great buying opportunity. Overreaction is one way to look at the extreme price changes. In the literature review, more explanations for overreaction are discussed.

This thesis examines investor inattention for Friday announcements, overreaction of the regulatory decision, and tries to explain how one can benefit from these events. The inattention can be the result of investors not monitoring the market extensive on a Friday as compared to the rest of the week. DellaVigna (2009) examines this phenomenon for earnings announcements and finds abnormal returns for the Friday earnings announcements. The FDA and EMEA do not have one distinct day in the week on which they release the data. As a result, these dates make it possible to examine whether there are abnormal returns for regulatory decisions released on Friday. One can find the drugs that are currently under review by the FDA and EMEA in a calendar on the website of the corresponding regulator. If such effect is also present in these announcements, one can benefit from this by trading the Friday announcements.

The regulatory decisions make an interesting field to examine and therefore the research question is: how do stocks on the NYSE/NASDAQ respond to the regulatory agency decisions from the FDA and EMEA for new drug approvals/rejections in the pharmaceutical industry? This question is interesting because it is an up-to-date topic. Pharmaceuticals operate in a multi-billion market and these drug applications are very important for the companies to remain an important player in the sector. The research process is lengthy and there is a high degree of uncertainty in the decision. When the drug application receives a rejection, this has great impact on the company. Not only is there a huge decline in stock price, future investors might also stay away and this makes it harder for the company to rebound from the rejection if the pharmaceutical has no other drug in the research pipeline. For managers in the company, the regulatory decision gives insight on how well the research and development department is functioning in order of productivity and effectiveness.

(7)

The experience gained during the review process can benefit the pharmaceutical company in future drug applications and potentially the company can increases its strategic position in the market.

The approval is an important valuation determinant for companies with high dependency on research and development, as the patent for the drug is an intangible asset, and therefore closely monitored by

investors (Langreth and Herper, 2008). The patent gained from these innovative operations provides the pharmaceutical the ability to stay competitive in the pharmaceutical industry. However, because the decision affects an intangible product, it is difficult to put a price on it. There are different methods to estimate the value of an intangible, such as an option-based method. These different methods all have their advantages and disadvantages. The hardness in determining the value of the patent can result in investors misreading the signal and valuing the drug higher than its actual value. Different drugs target different diseases. This makes it even harder for an investor to make accurate assumptions on the future drug returns. Some drugs are extremely valuable such as drugs that target rare genetic diseases, while others are worth analogously little. Furthermore, Aboody and Lev (2000), point out the uniqueness of the research and development projects of firms. As a consequence, investors can derive little to no information about the value and performance of a new drug just by looking at the R&D performance of other companies (2000). And even it there are benefits, they are more likely to unfold in a later time period.

One who overreacts on the signal can lower his expectations the days after the announcement when the company or public releases more information on the drug and its specifications. One who underreacts can increase his expectations. There is a lot at stake in this multi-billion pharmaceutical market. Understanding the effect of a regulatory decision benefits not only the pharmaceutical company, but also one who is trading the stock.

This thesis adds to preceding research by including announcements made by the EMEA for drug approvals and rejections. And the rejection sample for FDA rejections is larger than in prior research. Furthermore, in addition to the overreaction/underreaction tests, it examines whether investors have limited inattention for Friday announcements. DellaVigna (2009) finds investor inattention in Friday earnings announcements. The research on inattention to Friday regulatory announcements and the bigger rejection sample contributes to this area of study. This will contribute to a better understanding of how the market will react so such drug announcements.

For testing, this thesis performs an ordinary least squares (OLS) regression to estimate abnormal returns by using daily stock price data matched with the event dates from the FDA and EMEA drug approval and rejection announcements. The FDA announcements range from January 1980 till December 2014, while the EMEA sample is more recent with dates ranging from January 2004 till December 2014. EMEA data from before 2004 is not available in the database on the EMEA website.

The following section discusses the prior literature related to this topic. There will be a discussion on the efficient market hypothesis, the decision making process of the FDA and EMEA, and to what extent prices react to news announcement. And then explains how these announcements can cause overreaction and where inattention comes from. After the literature review, this thesis describes the used data set, how the data

(8)

is collected and the method used to perform regressions on this data. Then comes the findings and discussion of these results of this research. The thesis ends with the conclusion.

II Literature review

2.1 FDA

This section elaborates on some of the main tasks of the regulatory agency. The Food and Drugs

Administration (FDA) is an agency within the U.S. Department of Health and Human Services. The FDA is responsible for ensuring the effectiveness and safeness of human and veterinary drugs, vaccines and other biological products. Drug companies seeking FDA approval must first undergo a lengthy process of extensive testing.

The process starts with the Investigational New Drug (IND) application2_{. In this application, the}

pharmaceutical company shows the FDA results of preclinical testing. The FDA, together with an

institutional review board, then decides whether it is safe to proceed with the drug application for testing on human subjects. If the FDA decides the IND is safe or effective for human tests, three phases of clinical testing follow before a company can file for New Drug Application (NDA).

Phase I testing aims to find the drug’s most frequent side effects by testing the drug on 20 to 80 healthy volunteers. According to Dimasi et al. (2002), the probability of a successful Phase I is 30%. When Phase I shows no unacceptable toxicity, the Phase II studies begin. The aim of Phase II is effectiveness and to generate preliminary data on whether the drug works for people with a certain condition or disease. Tests performed by given the drug to a treatment group and a placebo to a control group. Phase II has a probability of success of 14% (Dimasi et al, 2002).

The final phase in clinical testing, Phase III, is the most expensive and time consuming part. A test group, with up to 3000 people from different populations (e.g. race, gender, age), receives the drug in combination with other drugs and different dosages of the drug, in order to gather more information on the effectiveness and safety of the medicine. This phase is the part where most companies do not succeed. Studies show that only 9% of the companies starting the clinical phase pass the Phase III trials (Dimasi et al., 2002).

After roughly 6 to 11 years of extensive clinical testing and with a positive Phase III, the company files for NDA. With this formal step the company asks the FDA to consider approving a new drug for marketing in the United States. The NDA includes all pre-clinical and clinical test results as well as labelling information and manufacturing specifications. After receiving the application, the FDA has 60 days to decide whether to file it. If it does, a team of FDA scientist reviews the safety, effectiveness, labelling and

manufacturing specifications of the drug. There are multiple extensions the FDA tests on and the regulator assigns a specific code, the chemical type, to the drug. For a full list of different chemical types, see table 1

2

(9)

in the appendix. In the method section, more information will be given on the data and how the FDA displays it on their website. What follows after the NDA is or a drug approval and a marketable drug, or the company receives a complete response letter (CLR) and the drug cannot enter the market.

Only companies with a successful Phase III studies will file for a new drug application. Passing the clinical studies is no guarantee for drug approval, but the drugs that do file for NDA are the drugs with success in the clinical trials. Filing for NDA with bad results in the trials is costly and not logical. This explains why there are more approvals than rejections in the sample. The ‘bad’ drugs do not file for NDA because they do not succeed in Phase III. There is already a sort of filter active. There is a probability of 8% that a company starting Phase I reach the market with a drug approval from the FDA (Dimasi et al., 2002).

The explanation in the previous section shows that the process of developing a new drug is not only very costly and time consuming, but that there is also a lot at stake for the companies involved. If the pharmaceutical does not receive the approval to market the drug, it misses on future income as the drug has profit potential and, ultimately, the decrease in operations can decrease the pharmaceuticals’ strategic

position in the market. The marketing approval is always a positive shock to the pharmaceuticals’ future cash flows as it grants the pharmaceutical a real option on the drug’s production (Manela, 2014).

2.2 EMEA

The European Medicines Agency (EMEA) is responsible for the scientific evaluation of applications for European Union marketing authorizations. Similar as with the FDA drug application process, the

pharmaceutical company submits a single marketing authorization to the EMEA. When it receives approval, the market authorization is valid in all European Union member states. An approval from the EMEA does not automatically mean that it will also receive the drug approval from the FDA. Both agencies have there own way of drug testing and regulating. The EMEA does not play a role in the approval process. This is the responsibility of the national competent authorities.

Although both agencies share similar objectives, they differ in structure and evaluating the clinical trial results. The EMEA is more of an administrative framework, and National Agencies do the actual scientific review. This is different from the FDA: FDA reviewers are within the same agency. Same data input will probably result in a different outcome between the agencies. FDA announcements are more examined by prior research than EMEA announcements. This thesis uses EMEA data to make a side-by-side comparison between the two agencies.

2.3 Efficient Market Hypothesis

This section describes the efficient market hypothesis and its implications. According to the efficient market hypothesis, financial markets are informational efficient and all know information is reflected in the stock price (Fama, 1964). There are three forms of market efficiency: weak, semi-strong and strong. For the weak market efficiency, current stock price reflects all historical data. The semi-strong form embeds both

(10)

historical and current public information in the stock price and the strong form adds private information (1964).

The main principle of the efficient market hypothesis means that investors cannot earn excess returns from trading strategies using publicly available information (Fama, 1970, 1991). For the pharmaceutical market, the efficient market hypothesis receives coverage in various papers. However, not all the papers point in the same direction. For example, the research by Perez-Rodriguez and Valcarcel (2011) finds no violation of the efficient market hypothesis and no excess returns in the announcement data. In this research, the R&D process of the 17 largest pharmaceutical companies is examined during a time interval of 19 years. The research shows that only 6% of the abnormal returns can be linked with FDA approvals. They do not find a price backlash, and, therefore no violation of the efficient market hypothesis. The tested backlash in this paper is one day after the news event. In addition, negative news has a bigger impact than positive news (2011).

Sharma and Lacey (2004) find a similar result. In their research, they empirically test the effects of new product development outcomes on overall firm performance. These results are consistent with the efficient market hypothesis and show that the market valuations are strongly responsive (2004). But there are papers that find abnormal returns in news announcements. Liu (2006) finds abnormal returns in news

announcements during 1983-1993 for U.S. biotech firms and the findings show a medium horizon negative drift in the stock price subsequent to the announcement (2006). Interesting in Liu’s research is that R&D or other intangibles provide market relevant value (2006). This might also be the case with regulatory

announcements because they are assigned to intangible patents. Bosch and Lee (1994) investigate the valuation effects of FDA approvals and finds that these regulatory approvals have very large wealth effects.

2.4 Stock reaction to news

According to Chen (2014), fundamentals and emotions are the two main reasons for stock price movement after a news release. With fundamentals, one extracts the qualitative and quantitative information from the news article to make or adjust the investment decision. Emotions, however, in the form of positive and negative mood, influence one’s investment decision (2014).

Gilbert et al. (2010) find that our emotions play a huge role in the decision-making process and influences stock market investments decisions. Uncertainty about the future retains one from starting a position in the company (2010). Other behavioural finance research reveals a relation between emotion and trading. They find that stock returns are influenced by sentiment released with a news announcement or financial report (DeLong et al. (1990), Li (2006), Schumaker (2012)).

These findings from the behavioural finance sector might explain the extreme price changes for pharmaceuticals when the FDA/EMEA releases the drug approval decision. A drug approval releases positive sentiment under traders, while rejections results in negative sentiment. At this point, however, information on future sales and revenues are not available as different drugs target different diseases and therefore have a different revenue scheme. Trading the stock is then mostly based on emotions, rather than on fundamentals.

(11)

2.5 Overreaction

There are multiple explanations for overreaction of stock market prices. One explanation for overreaction, and the one most relevant for this thesis, is uncertainty. The drug approval causes a great deal of wealth effects to the company. The pharmaceutical company can start to produce the drug and launch it on the market as a FDA/EMEA approved drug. Before the FDA/EMEA approval or rejection occurs, there is a great deal of uncertainty under the investors because it is not yet known what the regulatory agency will decide. If the drug receives regulatory rejection, a multi-billion investment is worthless. This uncertainty is resolved by the final approval (Sarkar and De Jong, 2006). Sarkar and De Jong (2006) find that investors react positively to positive signals from the FDA and negative to rejection indicators. They conclude that investor uncertainty provides a base to examine overreaction of the signal (2006). The same conclusion follows from the research of Bosch and Lee (1994). In this research (1994), the researchers find a significant amount of uncertainty in the FDA announcement.

Different from the research of Sarkar and De Jong (2006), Bosch and Lee (1994) and Liu (2006) provide evidence of information leakage preceding the announcement. Sarkar and De Jong do not find this effect and conclude that the market is more efficient now or that the FDA does a better job at controlling for leaks (2006). According to Aboody and Lev (2000), it is reasonable to assume insider trading to be available around the event. The R&D gives insights on the growth potential of a firm and therefore investors and analysts try to acquire private R&D-related information for managers, as insiders are expected to gain from insider trading (2000).

Kaniel (2012) examines earnings announcements and finds that pre-event trading by individuals, with private information, predicts the returns on and after the event. The stocks, accumulated by these private information trading investors the 10-tradings days prior to the event, entails significant abnormal returns and, therefore, the individual trading contains relevant information (2012).

This thesis examines this phenomenon of information leakage in stock trades preceding the event to test whether the market indeed is more efficient, as in the research Sarkar and De Jong (2006), or that information leakage still precedes the announcement, as in the research of Bosch & Lee (1994), Liu (2006) and Kaniel (2012) and entails relevant private information.

A second explanation for overreaction is overconfidence. Overconfidence is related with uncertainty as uncertainty can boost overconfidence and, in turn, can cause an overreaction of information. According to Liu (2006), one can develop wrong ideas about the value of a new technology. As the event time passes by and more information about the technology becomes available, the investor corrects his prior beliefs about price movement. Liu finds that stocks with the highest mispricing tend to have a negative drift following the announcement (2006). This effect can also be visible in the pharmaceutical industry as investors do not yet know the exact details of the approved drug and might overreact the announcement.

If overconfidence occurs, one should be able to observe over- and underreactions before and after the public announcement (Elkemali, 2014). And if uncertainty boosts overconfidence, Elkemali (2014) predicts that these over- and underreactions are stronger when uncertainty is higher and when it concerns a

(12)

high tech company. In addition, Yumei (2011) finds that the misreaction is the result of sentiment-driven mispricing of stocks. According to Yumei, we have to look at the overall market sentiment. In a bullish market, investors tend to have a lower risk-aversion and different assumptions about stocks than in a bearish market (2011). If this is the case, one might see more abnormal returns following drug approvals in a bullish market. This is something for future research to look into.

An investor with private information signals causes the stock to overreact, and, according to Hirshleifer (1998), underreacts to public news. This is the result of one’s own research, which makes the investor overconfident if he puts too much weight his private information signal. If then his signal receives confirmation by a public information release, one’s confidence rises even more. The opposite, however, when the investor does not receive confirmation of his private signal, does not affect his confidence at all or just a little (1998).

The implication of overconfidence is that there are wider price swings, during the event day, away from fundamentals, which causes extreme price volatility (Hirshleifer, 1998). Greater confidence causes a relative underweighting of the public signal, which in turn results lower variance the day after the event. Hirshleifer then states that these wide price swings at the event day need a price correction at t=1 and t=2 so that greater overconfidence can either decrease or increase the volatility around the public announcement (1998).

Looking at the extreme price change effect from a behaviour finance view, one can suggest that such positive news creates momentum caused by overconfidence or over optimism. Investors trade on the positive news and increase demand for the stock, which causes the stock price to rise even more. According to Barber and Odean (2007), individual investors are net buyers of attention grabbing stocks, e.g. stocks in the news such as the companies with an FDA or EMEA announcement. The attention-driven buying is a result of the constraints individual investors face searching the thousands of stocks that are available to buy (2007).

Himmelmann (2012) finds a similar effect: the approval decision of a new drug receives a high degree of public and media attention. This increase in visibility increases the demand for the stock and, in turn, the stock price. In addition, the drug receives regulatory attention as well as corporate press releases, which enlarges the awareness under investors and has a positive effect on the stock price (2012). Mahani and Poteshman (2008) link an overreaction effect to unsophisticated investors. The investor misreads the news announcement and believes the stock will move further away from fundamentals (2008). The following extreme price change can then be the result of overreaction. According to Fama (1998), anomalies are chance results and apparent overreaction is as common as indirection. Consistent with the EMH, the anomalies tend to disappear. Berkman et al. find that the announcement reduces the differences of opinion between investors, and ultimately, reduces the overvaluation of a stock (2009).

Not only does overreaction have a short-term effect, for the long-term there is also an effect visible. Momentum traders can profit from trend chasing but, if they only implement simple trading strategies, their attempts to benefit from these anomalies leads to an overreaction at long horizons (Hong and Stein, 1999).

This thesis examines if it is possible to benefit from the extreme price change by also researching the period after the announcement t=0 for abnormal returns. Another point of interest lies in the period

(13)

preceding the announcement at t=0. Previous research shows the existence of leaks preceding a FDA decision. Bosch and Lee (1994) examine the valuation effects of FDA product approvals, rejections and disciplinary decisions on the firms that operate in the food and drugs industries. Their results provide evidence of information leaks preceding the event date (1994). Aboody and Lev (2000) link these leaks with insider gains. Insiders take advantage of information on changes in R&D. If these leaks are still available around the event day, there remains the possibility to take advantage of this phenomenon by implementing a trading strategy.

III Hypothesis and methodology

3.1 Hypotheses

This part of the thesis describes the hypotheses and methods used to run the regressions. From the literature, the following hypotheses are constructed. The first hypothesis is similar to the work of DellaVigna (2009). This research finds positive abnormal returns because of inattention to Friday earnings announcements (2009). The inattention is not unique to earnings announcements, and therefore provides a base for this hypothesis. To test whether there is inattention in the FDA/EMEA decisions on Friday, hypothesis 1 is: there is inattention in the Friday decision data and one can earn abnormal returns by trading a portfolio consisting of these companies.

In order to determine the day of the announcement, one can use the FDA website. This website provides us with a calendar3, showing the drug application under review and the company holding this new drug application. These dates on the calendar make it possible to determine the event day and the

corresponding trading day. Because of the exogeneity of the release day, the pharmaceutical has no influence on the announcement day and, thus, cannot choose a date to make the news public (Manela, 2014).

The second hypothesis examines whether there is an overreaction in the announcement. Investors might misinterpret the signal and value the stock higher than it should be. This relates to the work of Chan (2003) and Liu (2000). Chan (2003) finds that a stock with extreme price movement, due to a news event, tends to reverse the day after the event. Not only is there a reversal, there is also a slow reaction of investors to bad news (2003). Liu (2000) also finds a negative drift after an event and this is largest for the highest mispriced stocks. The studies of Chan (2003) and Liu (2000) create a basis for hypothesis 2: there is a negative drift following the FDA/EMEA approval announcement.

The third hypothesis tests whether there is information leakage preceding the drug

rejection/approval. Bosch and Lee (1994) find information leakage in their sample, but Sarkar and De Jong (2006) do not share this same conclusion. In their research (2006), they conclude that either the market is more efficient or that the FDA is doing a better job in controlling for leaks. Hypothesis 3 is in line with the Sarkar and De Jong (2006) research: there is no information leakage preceding the FDA/EMEA

3

(14)

announcements and thus markets are more efficient. This hypothesis will also give insight if there is an upside trend in the days preceding the announcement.

Hypothesis 4 tests whether an already approved drug by the FDA or EMEA that is now under review for approval by the other regulator, has the same or different effect after the announcement. Investors might believe that, because one regulator already approves the drug, there is a higher chance that the drug will also receive an approval from the other regulator. Barberis et al. (1998) find that when investors overvalue a company, disappointment follows when the outcome does not match with their expectations. The investor might conclude that because of the approval already in place will result in another approval. For this, I expect the price reaction to be more moderate for a drug with already approval from one regulator that receives approval from the other because it is already priced in. Investors expect a second approval. However, because it is expected to receive an approval, I believe the negative reaction to be more extreme when the drug receives a rejection. Hypothesis 4 is: a drug with FDA (EMEA) approval will show a higher negative price changes when the drug receives a rejection by the EMEA (FDA).

3.2 Methodology

The announcement from the regulatory agency provides a unique date, which makes it possible to perform an event study. The event is the day the announcement is made public (t=0). To research information leakage preceding the announcement day, a trading window of five days before the event examines this effect. Sarkar and De Jong (2006) do not find information leakage, and link this to more efficient markets or better control by the FDA controlling for information leaks (2006). However, Bosch and Lee (1994) find these leaks. By examining the days preceding the event, this thesis tests whether the market indeed is more efficient. In addition, this thesis also tests whether this is the same for EMEA data.

To examine overreaction in the regulatory signal, a period after the event tests whether a price reversal occurs. The econometric model for testing abnormal returns are similar to model used by Sarkar and De Jong (2006). This thesis performs an ordinary least squares (OLS) regression to measure the abnormal returns in news announcements released by the regulatory agency:

Abnormal return= β0+ β1 ln Capitalizationt+ β2 Priority Drugs+ β3 Orphan Drugs+ β4 High Applicant+β5 Low

Applicant+ β6 Chemical Type Dummy+εt

This OLS regression allows for testing a causal effect of FDA approvals on abnormal returns. The FDA assigns a ‘NDA Chemical type’ to every drug applicant and a “Review Classification”. In total, there are 9 distinct chemical types and 3 different types of review classifications. Table 2, in the appendix, shows the possible combinations4_{. For each chemical type, a dummy variable is created. One expects the effect on}

stock price to be highest for a new drug, the new drug entity (NME), as this opens a door to new revenues

(15)

and a lower effect for a change in e.g. manufacturer. Capitalization is defined as the number of shares outstanding multiplied with the share price before the announcement (t=0). This is the same method used by Sarkar and De Jong (2006).

The high applicant group contains companies with the most drug applications in the sample. The high application dummy contains pharmaceutical companies with application equal or above the median of total applications. And the low applicant group contains companies with the fewest drug applications, the companies with applications under the median. There might be a different effect visible between those two groups. If a company has a history of successful applications leading to drug approvals, one can place too much weight on the new approval as one expects the company to receive another approval. According to Barberis et al. (1998), stocks with a consistent record of good news are overvalued, and one can earn

abnormal returns by betting against this overreaction. And the reversed effect for stocks with a record of bad news, these are, according to Barberis et al., undervalued (1998).

When working with EMEA data, there is a minor adjustment needed in the methodology. The EMEA provides no information on the review classification or chemical types. They use the ATC code to label a new drug application. This ATC code provides information on the organ or system the drug works on and how it works. However, it is difficult to use this code because there are five different levels with

subgroups for each level. Most drugs have a unique code and only a handful match with the same ATC code. Therefore, to estimate the abnormal returns for EMEA events, the regression is slightly modified to fit the available variables:

Abnormal Return= β0+ β1 ln Capitalizationt+ β2 Generic+ β3 Orphan Drugs+ β4 Biosimilar Drugs+β5 High

Applicant+εt

The capitalization is defined the same as with FDA approvals. The distinction between drugs for EMEA approvals is between orphan, generic and biosimilar drugs. Orphan drugs target a rare medical condition, where generic drugs are the same as a brand in dosage, safety, strength, and intended use. Biosimilars are biological products, which are approved based on sharing similarities with an EMEA-approved biological product, known as the reference product, and has no meaningful differences in terms of safety and

effectiveness from the reference drug. Dummy variables represent the orphan, generic and biosimilar drugs. High applicant and low applicant are defined the same way as with FDA data. In order to calculate the abnormal returns for the FDA and EMEA sample, the following formula is used:

ARjt=Rjt-(αj+βjRmt)

Where Rjt is the return on security j for day t, αj the OLS estimate for firm j’s market model parameters, the

βjRmt return on the CRSP equally weighted index for day t. In order to measure the abnormal returns, the

adjusted closing price is needed. Then the stock beta times the return on the NYSE/Nasdaq is what the return should be. Subtracting this return with the return on the NYSE/Nasdaq gives the abnormal return for the

(16)

stock for that given day. These calculations helps examining the immediate response of FDA and EMEA announcements on t=0. It is also interesting to research the stock bid and ask prices on the announcement day. These values provide us with the range the stock is trading between on the event day and one might benefit from these ranges.

The 15 days before and 15 days after the event occurs (t=0) are of interest. This timeframe is chosen to limit interference from other events and to focus on the effect of the regulatory announcement. The data section discusses more on the trading window. Rothenstein et al. (2011) use a timeframe of 60-120 days prior to the event to examine information leakage of Phase III trials and a possible prediction of the outcome of these trials. They find the stock to rise in the 60 days before the announcement of positive trials and start to decrease if negative. These findings and the findings of Bosch and Lee (1994), and Sarkar & De Jong (2006) suggest that the days preceding the announcement holds important information on the regulatory decision. Therefore, this thesis uses the time window of -15 and +15 days. A larger timeframe, however, increases the possibility of additional events surrounding the regulatory decision and a less precise estimation of the regulatory announcement effect.

In the Rothenstein et al. (2011) research, one can identify the uncertainty, under investors, days before the release of the Phase III trial results. This decline in stock price is a good entry point and creates the opportunity to benefit if there are abnormal returns surrounding the event.

IV Data

4.1 Data collection

The following section describes the used data, results and discussion of the findings. For the events, this thesis uses the database located on the website of the FDA5 and EMEA6 respectively. The data available in these databases, however, is not the same. The FDA only holds monthly data of drug approvals in their database, while the EMEA allows the user to download not only approval data, but also rejection notes. Due to legal reasons, the rejection data is not available on the FDA website. According to Sarkar and De Jong (2006), these legal notice protect a company who wants to resubmit their application. This thesis collects data on FDA rejections (also know as Complete Response Letters (CLR)) by using various search inquiries on Google. Inquiries such as “Complete letter of Response” combined with a date of interest. In order to ensure the announcement is accurate, the patent number or drug name from that inquiry is then examined for an official FDA document. Collecting FDA rejection data is a manually intensive task. This thesis only uses rejection data with an official FDA document in the rejection sample. For example, keywords such as “Bydureon7_{”, “rejected” and “FDA CLR”, delivers documentation about the rejection of the drug Bydureon}

submitted by the pharmaceuticals Alkmeres8 and Eli Lilly.

5_{http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Reports.ReportsMenu} 6_{http://www.ema.europa.eu/ema}

7

Bydureon is a medicine that helps to lower blood glucose in type 2 diabetes 8_{Alkermes and Eli Lilly collaborated with the Bydureon research}

(17)

For estimation purposes, the data must meet certain conditions. First, the pharmaceutical company must be available on the Center for Research on Security Prices (CRSP) in order to match announcement dates with stock prices. This thesis uses the PERMNO obtained from CRSP to match the announcements with stock prices. This number denotes the current permanent number. The PERMNO then has to be manually added to the announcement files before one can merge the data sets. Second, this thesis needs pharmaceutical companies with information on key specifications, such as volume, return, and share price. This leads to the exclusion of private companies and incomplete observations.

The regressions are done in twofold: one with just single observation per company and regulator and one regression with all observations. This is in order to examine the equal effect of a single drug application. As discussed in the above section, there are companies with multiple applications ranging to 200 filings for one company. Therefore it is interesting to make a distinction between the data. In order to create one single file with all observations, this thesis uses the event method retrieved from the Princeton University Data and Statistical Services homepage9. First, this method counts the number of events per company, the variable ‘eventcount’. Then drops all observations but one observation per company, linked with the number of total event dates. This file is then merged into the CRSP file with daily stock information and expanded with ‘eventcount’ to create the duplicate observations. Finally, after creating the duplicates, the file is merged with the original file containing all announcements.

For each event, this research creates a trading window, or the event window, of -15 days and +15 days and an estimation window of -30 days to +30 days around the event date. This event window is larger than in prior research, but it provides a little more information on the stock price movement further away from the event. The trading window provides enough data to perform calculations. Furthermore, choosing a larger window not only increases the amount of missing stock data, which leads to the exclusion of events due to a lack of information, it also increases the probability of systemic bias. The further one moves away from the event date, the higher the chance that a company releases other price relevant information during this period.

4.2 Summary statistics for the FDA sample

Tables 3 and 4 (appendix) provide summary statistics for the FDA and EMEA announcements respectively. In total, the FDA announcements sums up to 3755 events during January 1, 1980, until December 31, 2014. From these events, 1485 announcements provide usable information for estimation purposes. One can assign these announcements to 164 distinct pharmaceutical companies. Although the time-period is longer than in the Sarkar and De Jong (2006) research, the absolute value of distinct pharmaceutical companies is lower than the 189 firms in the latter. The total events of 1485, however, are larger which means that the amount of distinct companies is decreasing and more applications come from the same pharmaceutical. This can be the result of a merger or collaboration between two or more pharmaceuticals. According to Dimasi et al. (2002), these mergers and consolidation in the pharmaceutical market are the result of the rising cost for drug

(18)

development. Table 3 (appendix) shows that the minimum drug application is one, and the maximum drug applications for a single company are 200. This number includes all chemical types, such as applications related to labelling adjustments and changes in dosage forms. Later on in this thesis, the new drug entity, or chemical type 1, is of interest.

The volume on the event day averages around 3.2 million shares with a maximum of 72 million shares and the average abnormal return, both approval and rejection, yields 0.23% with a minimum value of -6% and a maximum of 6.73%. There appears to be quite some variation in abnormal stock returns with as seen from the standard deviation of 2.45%. In the results section, this thesis discusses the average abnormal returns surrounding the event 15 days prior and 15 days after the announcement with a distinction between regulatory decisions. Market capitalization is defined as the number of shares outstanding multiplied with the stock price. The market capitalization for the FDA announcements yields a mean value of roughly 21 billion dollars with a maximum of 300 billion dollars.

4.3 Summary statistics for the EMEA sample

The EMEA sample, see table 4 in the appendix, consists of a total of 847 announcements during January 1, 2004, until December 31, 2014. From these announcements, only 403 events provide this thesis with usable information. The time-period for the EMEA sample is smaller than the FDA sample because the EMEA started the database somewhere around the year 2000. However, most observation before 2004 are

incomplete and therefore this thesis use data from the year 2004 and up. After breaking down the events, the applications come from 45 distinct companies. The average application amount is 27 with a minimum of 1 and a maximum of 58 applications by one single company. This amount is lower than the 200 applications in the FDA sample. An explanation for this difference is the way both regulators display and label information on their websites. For instance, the EMEA keeps no record on changes in labelling and new dosage forms in their database. They do, however, keep track of the revision number. There is a maximum of 45 revisions and an average of 11.43 revisions.

The abnormal returns, for all EMEA events, averages 0.15% on the event day with a minimum of -6% and a maximum of 6.7%. Also in EMEA events, there is quite some variation in average abnormal returns as seen from the standard deviation of 1.81%. These numbers are very comparable to the FDA minimum and maximum average abnormal return values. On the event day, volume is on average 4.2 million shares with a maximum volume of 56 million shares. The average volume is higher than the average volume for FDA events. An explanation can be that, in the EMEA sample, companies filing drug applications are larger in size. The market capitalization averages around 3.2 billion with a maximum of 288 billion, which implies that, on average, the companies in the EMEA are larger in size and strength than the companies in the FDA sample. Looking from a firm perspective, this seems logical. Mostly companies with the financial strength undergo the review process by the European agency, which is again costly and a lengthy process.

(19)

4.4 Summary statistics on Friday publications

This thesis splits the data in two sets, one containing FDA data and one EMEA data. The reason for this lies in the difference in specifications between the two data samples. For instance, the FDA uses e.g. chemical types to label a certain drug. The EMEA, however, only displays the orphan drug or standard drug label and uses a different indicator, the ATC code, which results in unique drug codes. This ATC code provides, because of its unique code, no usable information for this thesis.

Breaking down all of the announcements, the 1485 FDA events consists of 1422 approvals and 63 rejections. With only a single observation per company, this yields 119 approvals and 45 rejections making a total of applications from 164 pharmaceuticals. Although small, the rejection sample is still lager than the samples used in the research of Bosch and Lee (1994) and Sarkar and De Jong (2006) with 20 and 18 rejections respectively.

Bosch and Lee (1994) use publications from the Wall Street Journal to examine the wealth effects of FDA approvals. These publications have a minimum lag of one day compared to announcements on the Internet. But also in the Wall Street Journal, there is no or limited mentioning of CLR’s. One explanation for the bigger sample in my thesis is that we are in an information era. The Internet is a very popular medium and big data collection is a hot topic. Once released, the information remains available on the Internet. Searching websites is easier, however still a time consuming task, than browsing through dozens of Wall Street Journals.

For the EMEA sample, all of the 403 events consist of 393 approvals and 10 rejections. With only one announcement per company this yields 37 approvals and 8 rejections from a total of 45 distinct pharmaceutical companies. Unfortunately, the rejection sample for EMEA data is quite small compared to the FDA data. Therefore, when making concluding remarks on EMEA rejection dates, one needs to be careful. The EMEA provides one with rejection data on their website. There is, however, one difficulty when downloading this information. They remove the date of announcement in the downloaded Excel file, so one has to manually insert the event dates if one wants to use these decisions. One can do this by matching the event in the Excel file with the corresponding date on the website.

In DellaVigna’s (2009) research, investor inattention leads to abnormal returns in Friday earnings announcements. This phenomenon of investor inattention on Fridays provides the underlying of this section. In order to calculate the abnormal returns corresponding to the trading days, first a variable, which contains the days of the week, is created. After generating this variable with the weekdays, one can distinguish two different sets of data in table with distinct observations and all observation per company. The distinction reason is mainly to ensure there is enough data in the regression since e.g. the EMEA rejection sample only contains a limited amount of observations when dropping duplicate companies, and to examine whether the subset is a good representation of the population.

For examining the inattention for each announcements day, this thesis provides tables with summary statistics; see tables 5 & 6 in the appendix, with summary statistics for Friday and non-Friday publications. Table 5 contains FDA data and table 6 the EMEA data set. One can distinguish the market capitalization in millions of dollars for the Friday announcements and the non-Friday announcements and the volume on the

(20)

event day. The market capitalization is calculated by multiplying the number of outstanding shares with the share price. Tables 7 and 8 (appendix) then provide one with announcements for the FDA events and tables 9 & 10 with the EMEA events. One can see in table 7 that the rejection sample is almost evenly divided. In the approval sample, the FDA releases most publications on Fridays with almost one third of all announcements. Summary statistics of the Friday FDA approvals in table 5 (appendix) show that the mean of average

abnormal returns on Fridays is almost double that of the non-Friday announcements. Friday approval events produces an average abnormal return of 0,15% with a high standard deviation of 2.4% in all observations, while non-Friday publications yield an average abnormal return of 0.08% with a standard deviation of 2.04%. The variance seems to be quite substantial. Despite the inattention on Fridays, as mentioned by DellaVigna (2009), this is the day with the most publications. Volume and market capitalization are also higher than the publications on non-Fridays. With just a single observation per company, the average abnormal return on Friday is 0.84% with a standard deviation of 3.71%. For non-Fridays, 0.37% average abnormal returns with 0.26% standard deviation. Interestingly, here one can distinguish quite a difference between Friday publications and non-Friday publications.

The EMEA approval data shows similar results, and it appears inattention is also not an issue here. With all observations, publications on Friday show 0.09% average abnormal returns with a standard

deviation of 1.82% compared to non-Friday returns of 0.33% with a standard deviation of 1.84%. Volume on Friday is in both sets higher, with 4.9 million and 1.92 million for the set with all observations and set with single observations respectively.

Later on in this thesis, there follows a regression on the weekdays to examine whether this is

significant and calculations of cumulative abnormal returns. Because of the limited amount of rejections, the summary statistics will cover the sample with all observations. However, both samples will be included in the regression.

V Results 5.1 Hypothesis 1

Interestingly, from prior discussion both FDA volume and market capitalization are higher on Fridays than on other weekdays. One can see similar results for the EMEA data. This, of course, has something to do with the amount of publications on Fridays. But, according to the literature, this is supposed to be the day with the highest fraction of distracted investors. The following section will examine the abnormal returns and

cumulative abnormal returns surrounding the event and research whether these returns are statistically significant.

For testing hypothesis 1, whether there is inattention in the Friday publications, this thesis uses winsorising to deal with extreme values in the data and limit the disturbance of these outliers on the

calculated statistics. Table 11 (appendix) shows the average abnormal returns for both FDA and EMEA data as a result of the winsorising measure for the announcements made in the event window without making a distinction between the days of the week on which the announcement has been made. One can distinguish

(21)

average abnormal returns with all announcements used and returns when only one event per company is used. Here, this thesis tests the null hypothesis where one is not able to earn abnormal returns by trading stocks with Friday publications versus the alternative hypothesis that trading these stocks yields abnormal returns. This thesis uses a significance level of 5%, which means the variable is statistically significant if the absolute value is greater than 1.96.

Table 12 shows the regression results and provide information on the cumulative average abnormal returns for Friday publications. In this table, the cumulative abnormal returns are calculated by performing a regression of all Friday publications according to their decision outcome on the cumulative winsorized abnormal returns. Following the table, one can see only significant results for the 31-day event (-15, +15) belonging to the sample with all Friday approval publications. The cumulative average abnormal returns are 1.17% with a t-statistic of 2.39. This means that one can earn abnormal returns by holding the stock from 15 days prior till 15 days after the event. Holding the stock 15 days prior to the event and it turns out to receive a rejection at t=0, will still yield a return of 0.58% after the 31 days event window. However, this return is not significant with a t-statistic of 0.28. In the FDA data set, one can earn abnormal returns by using the inattention of investors. However, these abnormal returns are not statistically significant except for the FDA approvals in the 31-day event window. The regulatory decisions seem to be monitored quite extensively.

The EMEA sample shows no significant cumulative abnormal returns either. One can make a few interesting observations in this data. First, the cumulative average abnormal return is negative for the two-day window (0, +1) and zero in the four-two-day window (0, +3). The Montwo-day following the announcement on Friday does not provide the opportunity to benefit from investor inattention. One might suggest that there is a difference between time zones and that this creates an opportunity to benefit from. The table shows no significant results. For EMEA rejection data, the last column with single rejections is missing in table 12. This is due to a lack of observations. The column with all rejections is already very small, and does not provide much useable information.

This thesis finds no evidence of investor inattention in regulatory publications for drug approvals and rejections. There are cumulative average abnormal returns surrounding the Friday release event but except from the sample with all FDA approvals in the 31-event window, these are not statistically

significant. Testing hypothesis 1 provides no statistical evidence to reject the null hypothesis and is in line with the findings of Langreth and Herper (2008). The approval is an important value determinant for the pharmaceutical company, as the patent for the drug in an intangible asset, and therefore closely monitored by investors (2008).

5.2 Hypothesis 2

This section tests hypothesis 2 whether a reversal occurs after the stock earns abnormal returns due to the publication of the regulatory decision. The null hypothesis tests whether there is no correction after the event date versus the alternative hypothesis with correction after the event date.

The FDA sample in table 11 (appendix), the approved drugs with only one observation per company, shows a statistically significant average abnormal return with a 5% significance level for t=0, the event day. With a

(22)

t-statistic of 1.98, the event associated with an FDA approved drug yields 0.57% average abnormal returns. It appears a reversal occurs, with a significant decrease of -0.91% at t=2. The corresponding t-statistic is -3.02. According to Hirshleifer (1998) wide price swings on the event date need a correction at t=1 or t=2. Such correction takes place at t=2 when examining the distinct FDA approvals. The other regression outputs show no sign of correction or reversal. Although the average abnormal returns are decreasing after the event, these drops are not statistically significant. Cumulative returns, for holding the stock from the event day till day 5, yield -2.5% cumulative average abnormal returns. This thesis only finds a reversal effect for the distinct FDA approvals, yet there is not enough evidence to reject the null hypothesis. In the appendix, graph 1 shows the cumulative side-by-side comparisons for the FDA and EMEA data. This graph uses the average abnormal returns from table 11 to calculate the cumulative average returns. One can see a dip in cumulative returns after the approval. An explanation for this dip can be the profit taking by day traders. After 15-days, the FDA line is at the same point as it was on the event day, t=0. The EMEA cumulative abnormal returns go down the first days after the event. Then, at t=5, the cumulative returns go up again.

When comparing the FDA and EMEA regression lines, one can see that the EMEA approval generates lower cumulative abnormal returns than the FDA approvals. In the chosen event window, FDA approvals yield a cumulative abnormal return of 1% versus a cumulative abnormal return of 0,25% for EMEA approvals. It appears that the investor values a FDA approval higher than an approval made by the EMEA.

For rejection data, both lines follow a similar pattern prior to the event day. The EMEA line,

however, has a lag of one day before the investor reacts to the rejection. On the event day, one can see rising cumulative abnormal returns and on the following days, from t=1 till t=3, loses almost 3%. Interesting is that the EMEA rejections yield a positive cumulative returns of 3% at the end of the event window. For FDA rejections, the cumulative returns show lower returns every day and does not seem to stabilize. On a side note, the FDA rejection sample is larger than the EMEA rejection sample. Therefor, when making

concluding remarks in comparing the two agencies, one must be careful. Future research, when more data on rejection data comes available, is needed to make proper conclusions.

5.3 Hypothesis 3

This section performs a test on hypothesis 3. The null hypothesis tests that there is no information leakage preceding the event versus the alternative hypothesis with information leakage preceding the event. Inspecting the sample with all FDA observations in table 11, there are no significant average abnormal returns on the event day with returns of 0.10%. However, there are significant results in the 5-day trading window. On t=-2, there is a significant abnormal return 0.11% (t-statistic of 2.00) and a similar result on t=-1 with 0.10% (statistic of 1.97). The day after the event yields 0.29% average abnormal returns and a t-statistic of 4.54. It appears that there is information leakage preceding the FDA approval decision, with all observations included, two days before the event occurs.

Bosch and Lee (1994) find average abnormal returns of 1,10% and 0,65% for t=-1 and t=0 (because of the lag due to use of Wall Street Journal articles) respectively and Sarkar and De Jong (2006) 0,54% and

(23)

0,44% for t=0 and t=1 respectively. The abnormal returns for t=0 and t=1, for the single observations, are very comparable to these values. With an average abnormal return of 0,57% on t=0, the return in this thesis is between the findings of the prior research and more in line with the findings of Sarkar and De Jong (2006). Differences are visible in the day following the announcement with an average abnormal return of 0.21%, which is lower than the 0,65% and 0,44% returns. It seems the average abnormal returns are diminishing over time, but one can still earn abnormal returns on the event day. Graph 2, on the next page, compares the empirical results of this thesis with the results of the Sarkar and De Jong (2006) research. In order to make a comparison, the event window is set up 10-days prior to the event till 10-days after the event and average abnormal returns are calculated accordingly. Now the event window is matched, one can see that the line with the FDA approvals follows almost the same pattern and ends with almost the same cumulative average abnormal returns. The line with FDA events prior to t=0, however, is a little more stable than in the Sarkar and De Jong (2006) research, where in the latter one can see a higher shock from the approval. The line with cumulative abnormal returns then stabilizes a bit and ends, after 10-tradings days, at almost the same level as when the event occurs. While in the 2006 research, the cumulative returns are higher after at the end of the event window.

The FDA rejection sample shows a different result than prior research. One might expect a severe crash, due to the loss of the option to market the drug, after a drug receives a rejection, but there is no statistically significant effect in the sample date around the event date in the sample with one observation per company. With an average abnormal return of -0.86% on the day of the rejection, the effect seems lower than the findings of Sarkar and De Jong (2006) with a drop around 4% and the findings in the five days of

abnormal returns in the Bosch and Lee (1994) research. On a side note, the rejection sample in this thesis is larger than in prior literature. This explains some of the difference due to the fact that, in general, more data creates better estimates.

In graph 2, the rejection lines almost follow the same pattern prior to the event. The graph displays all rejection data and one can see that the downfall in cumulative abnormal returns is not as severe as in the Sarkar and De Jong (2006) research. It follows a downtrend, but the movement down is moderate. One explanation for the difference is the rejection sample used. There is no database with rejection data, so data capturing this effect is the result of manually collected events. Comparing the positive event with the

negative event, one can see that a negative event has a bigger impact on investors and a larger negative effect on cumulative abnormal returns.

(24)

Graph 2. Graphical comparison empirical results versus results Sarkar and De Jong (2006)

The following graphics display a comparison of the empirical results of this thesis with the results of the results in the research of Sarkar and De Jong (2006). The graphics use all FDA announcements made in January 1980 – December 2014. The line represents the cumulative abnormal returns in an event window similar to the event in the Sarkar and De Jong (2006) research. The 10-days prior to the event and 10-days after the event are displayed here and use the average abnormal returns of table 11 (appendix).

Table 13 tries to find the Bosch and Lee (1994) information leakage findings. The event windows are set up as in the Bosch and Lee (1994) research, with respect to the one-day lag due to the use of Wall Street Journals in their research. Approval data shows significant cumulative abnormal returns on the cumulative

-. 0 0 5 0 .005 .01 C u mu la ti ve a b n o rma l re tu rn s -10 -5 0 5 10 Event window

Empirical results Results Sarkar and De Jong (2006)

FDA approvals with event window 10 days before and 10 days after event

Empirical results versus results Sarkar and De Jong (2006)

-. 0 6 -. 0 4 -. 0 2 0 .02 .04 C u mu la ti ve a b n o rma l re tu rn s -10 -5 0 5 10 Event window

Empirical results Results Sarkar and De Jong(2006)

FDA rejections with event window 10 days before and 10 days after event

(25)

trading days t=-2 and t=-1 with a return of 1.29%. This suggests that information leakage is still preceding the event. Furthermore, rejection data also shows a similar effect around the event day with significant cumulative abnormal returns of -2% in for the window t=-2 and t=-1.

One can see that the drug rejection has a stronger impact than a drug approval. In the days following the event, the average abnormal returns remain negative. The effect is visible in both rejection samples; the distinct set and set with all observations. This is also what one would expect and mentioned in several prior papers, such as the work of Perez-Rodriguez and Valcarcel (2011). They find that negative news events have a bigger impact than positive news events (2011).

For EMEA data, there is no literature to compare the results with. Prior research focuses on FDA data. Therefore, this thesis uses the results from the FDA as comparison. With single observations, the EMEA data shows no significant abnormal returns around the event for approved drugs (see table 11 in the appendix). With all observations, there are significant returns of 0.28% (t-statistic of 3.01) on the event day.

The rejection sample in table 11, however, does hold interesting results. For t=3 in the set with distinct rejections, there is a significant negative abnormal return of -1.04% with a t-statistic of -2.72. Similar significant abnormal returns are visible when using the sample containing all observations. For t=3 there is an average abnormal return of -0.90% with a t-statistic of -2.65. It appears there exists some lag between the rejection decision and the reaction from the investor. Chan (2003) assigns this lag due to slow reaction to bad news.

This thesis finds significant cumulative returns preceding the event date for the cumulative event t=-2 and t=-1 in the FDA approval sample, which suggests there might still be information leakage and it pays off to take on a long position. Therefore the null hypothesis is rejected for this event window. In the two-days prior to the event t=-2 and t=0, no significant results are present.

5.4 Hypothesis 4

Hypothesis 4 turns out to be difficult to test. When combining all of the drugs, both EMEA and FDA, no match is found. One of the reasons for this is that both regulators have there own way of displaying the information on their website. Also, the label for drug identification is different. One way to overcome this problem is to search the FDA Orangebook for patent numbers. The Orangebook is available on the FDA website. Then match these patent numbers with the EMEA drug patents. Unfortunately, this turns out to be a very difficult job without the knowledge of writing scripts to gather data. Without a script, one must search for patent numbers one by one and then manually search the correct patent numbers. For future research, this could be something to look into. Drugs, with already an approval, might face a different effect when

applying for a second approval. This might provide a better understanding of how the market reacts to such events.

5.5 OLS regression

To measure the abnormal returns, this thesis uses an OLS regression. Table 14 provides the regression results for FDA events. There is a distinction between all events, approved events and rejected events.

(26)

Table 14. Regression result FDA events

This table provides the regression results for three sets of FDA data for applications made in January 1980 – December 2014. The first column uses all applications, the second column only the approved applications, and the third column use rejected applications only. The variable log(capitalization) is the log of the shares outstanding multiplied by the stock price. The variables priority drug, and standard drug are dummies. The high application variable is a dummy variable with value 1 if a company has an application number higher than the mean of total applications in the sample. The new drug entity variable is a dummy variable with value 1 if the drug is first of its kind. The winsorized abnormal returns, the dependent, are the difference between daily returns and the value-weighted market return. The standard errors are in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1%, respectively.

Dependent variable: winsorized abnormal returns

Statistic All FDA applications Approved FDA applications Rejected FDA applications Log(Capitalization) -0.0001 (0.0004) (0.0003) -0.0004 (0.0016) 0.0017 Priority drug -0.0001 (0.0029) (0.0031) -0.0058 Standard drug -0.0012 (0.0025) -0.0071* (0.0027) High application -0.0050*** (0.0011) -0.0049*** (0.0011) 0.0011 (0.0059) New drug entity 0.0048*

(0.0015) 0.0047* (0.0015) Constant 0.0022 (0.0026) 0.0086* (0.0027) -0.0090 (0.0055) R2 0.0227 0.0340 0.0209 N 1401 1339 62

The high application variable contains the companies with the highest amount of drug applications in the sample. The companies above the application median are assigned to the high application variable. This variable is statistically significant in the sample with all observations and approved sample. Companies with a high number of drug applications earn 0.5% less abnormal returns. The companies belonging to this variable are the pharmaceuticals with a higher market capitalization. With a market capitalization of on average 29 billion for high appliers versus on average 17 billion for non-high appliers. The new drug entity, when chemical type equals 1, also earns statistically significant abnormal returns. Approved new drugs earn 0.50% more abnormal returns than drugs with another chemical type. In the rejection data, no significant results are present.

The EMEA regression is available in the appendix under table 15 (appendix). There are no significant results in the EMEA sample regression. When comparing the results with the FDA sample in table 14 (appendix), one can see a similar effect for high application companies. Only in the EMEA regression, this variable is not significant. Orphan drugs perform poorer than non-orphan drugs and biosimilar drugs perform in the best in all the three sets.

Although, one might argue that this is in line with prior findings because of the diminishing

abnormal returns around the event date one can distinguish in the cumulative returns. For the five day period belonging to t=-1 till t=3, Bosch and Lee (1994) report a cumulative average abnormal return of -4,76% and Sarkar and De Jong (2006) report a return of -4,04%. The results of this thesis are close to these findings with a cumulative average abnormal return of -3,62% in the 5-day trading period. There appears to be a diminishing effect and the market processes information quickly.

Investor inattention to regulatory announcements for new drug applications