New Product Announcements and Shareholder Value

(1)

New Product Announcements

and Shareholder Value

A Study in the Pharmaceutical Industry

Stijn Broekema (2002507) University of Groningen

(2)

2

New Product Announcements

and Shareholder Value

A Study in the Pharmaceutical Industry

S.P.M. Broekema University of Groningen Faculty of Economics & Business Master thesis MSc. Marketing Intelligence

12-01-2016 Van Sijsenstraat 14 9724 NN Groningen 0629092535 s.p.m.broekema@student.rug.nl S2002507

(3)

3

Abstract

Marketing executives are under increasing pressure to be accountable for their investments. Marketing academics have responded to this call in recent years by focusing on the link between marketing instruments and shareholder value, called the marketing-finance interface. One of the marketing instruments that managers have at their disposal is to innovate and introduce new products. This study adds to the literature on the marketing-finance interface by studying the effects of new product announcements on three different parts of shareholder value: returns, systematic risk, and idiosyncratic risk, using a sample of 113 new product announcements in the U.S. pharmaceutical industry. Applying event study methodology, the results indicate that product announcements are accompanied by significantly positive abnormal returns. The abnormal returns are not significantly different for incremental and radical innovations. Using GARCH-modeling, systematic risk appears to increase after the announcement for incremental innovations. Furthermore, product announcements appear to have no effect on idiosyncratic risk.

Keywords: New product announcements, shareholder value, stock returns, systematic risk,

(4)

4

Preface

With a background in both marketing and finance, I am proud to have finalized this thesis on the marketing-finance interface. Even though the process was quite difficult at times, I am looking back on it with a smile. I strongly believe that the time and energy I have invested in this thesis have paid off. Particularly, I feel that I have had the opportunity to show my abilities to independently carry out an academic research project.

Firstly, I would like to thank my supervisor Jaap Wieringa for his support and helpful feedback during the process. Secondly, I would like to thank my fellow students for their insights. Furthermore, I want to thank my family and friends for giving me the opportunity to find a good balance between working on the thesis and doing other things, particularly during the final (busy) weeks. Lastly, but most importantly, I would like to thank my girlfriend Maaike, who supported and motivated me throughout the entire process.

For many students, finishing their thesis will mean that an end has come to their time as a student. Yet, next semester, I will write my MSc. Finance thesis to finish that program as well. I am certain that the knowledge and experiences that I have gained during the process of writing this thesis will help me in writing the next.

I hope that you enjoy reading the rest of the thesis.

Stijn Broekema

(5)

5

1. Introduction

In recent years, there has been increasing attention for accountability in the marketing field. Senior managers tend to see marketing purely as an expense (instead of an investment in future profitability), since it is difficult to accurately determine and communicate the impact of marketing activities on different firm metrics (e.g., sales, margins or firm value). Several studies indicate that firms cut back on their marketing expenditures during difficult economic times (e.g., Deleersnyder et al., 2009), even though this may be harmful for firms in the long run. To support decision-makers in coming to more optimal budget allocations, it is important to understand the effectiveness of different marketing instruments.

In the marketing literature, many studies have been conducted to study the impact of different marketing mix instruments. These studies focus primarily on determining marketing’s impact on metrics such as sales (see e.g., Ataman et al., 2010). The major disadvantage of such an approach is that it looks at top-line results (i.e., sales), but not at bottom-line results (i.e., profits) or stock returns, which means that it provides an incomplete picture of the value added (Mizik & Jacobson, 2003). This means that it is difficult to determine whether the marketing mix instruments add value to the firm, and if so, how large the added value is.

To this end, researchers started to look at the effects of marketing efforts on shareholder value (Srinivasan and Hanssens, 2009), which is also stressed by the Marketing Science Institute (MSI). The third top research priority for the period 2014-2016 is listed by MSI as “Measuring and Communicating the Value of Marketing Activities and Investments”, since firms are under increasing pressure to “make every dollar count” (MSI, 2013). Srinivasan and Hanssens (2009) state something similar: “The marketing profession is being challenged to assess and communicate the value created by its actions on shareholder value. These demands create a need to translate marketing resource allocations and their performance consequences into financial and firm value effects”. Clearly, this shows that the research field linking marketing and finance, called the marketing-finance interface, is becoming more and more important.

(7)

7 One marketing instrument that managers have at their disposal is the possibility to innovate and introduce new products, which is regarded as one of the most important marketing instruments. For example, Ataman et al. (2010) show that the relative impact on sales of line length (defined as number of SKUs) is twice, ten times, and thirty times as large as the relative impact of distribution, advertising, and price discounts, respectively. Yet, what are the effects of new products on shareholder value? Particularly, do product announcements increase stock returns? And what is its effect on the stock’s risk?

The inherent risk that comes with a new product, namely that the product will fail in the marketplace, has been of interest of marketing researchers for a long time. For instance, in 1977, Crawford carried out a review of several studies and statistics provided by market research firms and found that about 50-80% of new products fail. These numbers have not changed significantly over time according to the Nielsen Breakthrough Innovation Report of 2014, since product failure rates are still around 75% on average across industries. The risk of failing to launch the product as announced, or more specifically, the risk of the product failing altogether in the marketplace, is found to be related to the degree to which the new product provides unique benefits over already existing and competing products (Cooper & Kleinschmidt, 1987). Hence, this yields the intuition that product announcements of radical innovations may have a different effect on returns and risk than announcements of incremental innovations.

When studying the effects of new products on returns and risk, one can study at least three different events. To avoid confusion, it is important to immediately be clear about the distinctions between them, which is sometimes overlooked in the literature (Koku et al., 1997). Firstly, it is possible to look at the effects of the actual introduction of the product in the market (i.e., the moment at which the product is firstly available). Secondly, it is possible to study the effect of product preannouncements, i.e., announcements that are made well in advance of the product introduction (see e.g., Sorescu et al, 2007). Thirdly, one may consider the effects of announcements made close to the time of the actual introduction, called product announcements.

(8)

8 measures of risk, namely systematic (the part of total risk that can be diversified away by holding a well-diversified portfolio) and idiosyncratic risk (the part of total risk that cannot be diversified away)? And if so, what drives these differences?

A market well-suited to study the effects of product announcements on risk and return is the pharmaceutical market. Firstly, as stated by Sorescu et al. (2003), new products are common in pharmaceuticals, since the industry is driven by innovation. Secondly, the Food and Drugs Administration (FDA) provides an objective measure of the degree of innovativeness (FDA, 2013). Before a drug is approved, it goes through an extensive review process by the FDA in which the drug will be classified as “standard” (i.e., incremental) or “priority” (i.e., radical).

Next to adding insights to literature on the marketing-finance interface, the results are relevant for executives as well. The results can lead to improved budget allocations which in turn can have a positive effect on financial performance of the firm. Particularly, it will provide senior managers with insights into how the market values new products on average. Large financial returns may trigger executives into investing more in new product development. It will also provide executives with insights into what happens to systematic and idiosyncratic risk.

Applying event study methodology, the results show that product announcements are associated with significantly positive abnormal stock returns, and that the abnormal returns are not significantly different between radical and incremental innovations. Furthermore, using GARCH-modeling, systematic risk appears to increase on average after a product announcement for incremental innovations, whereas no impact on idiosyncratic risk is found (neither for incremental nor for radical innovations).

In the following chapter a review of relevant literature will be presented from which hypotheses will be derived. In the third chapter the methodology and data will be discussed. The fourth chapter provides the results. The final chapter contains a discussion, managerial implications, limitations, directions for future research, and a conclusion.

2. Theoretical Framework

(9)

9 2.1 The Marketing Mix and Shareholder Value

Portfolio theory as introduced by Markowitz in 1952 and later work on the Capital Asset Pricing Model (CAPM) (see e.g., Sharpe, 1964) show that investors are mean-variance optimizers. In other words, rational investors only care about risk and return. Investors are willing to take on extra risk, as long as it is compensated by a higher expected return. Hence, the combination of risk and return determines shareholder value.

Generally, risk is split up in two parts. The first part is called systematic (or market) risk, which can be defined as a stock’s sensitivity to the stock market as a whole. The measure used for systematic risk is beta (β) and is defined as:

𝛽_𝑖 =𝐶𝑜𝑣(𝑅𝑖,𝑅𝑀)

𝜎_𝑀2 (1)

where 𝛽_𝑖 represents the level of systematic risk for stock i, 𝐶𝑜𝑣(𝑅_𝑖, 𝑅_𝑀) represents the covariance between the return of stock i (𝑅_𝑖) and the market return (𝑅_𝑀), and 𝜎_𝑀2 is the variance of the market return. Systematic risk measures the part of total risk that cannot be diversified away, which implies that investors need to be compensated for borne systematic risk.

The second part of the total risk is called idiosyncratic (or firm-specific) risk, which represents the risk that is unique to a firm. By holding a well-diversified portfolio of stocks, this type of risk can be diversified away. Therefore, investors should in principal not be compensated for idiosyncratic risk.

Return can be defined as the percentual change in the stock price. According to the efficient market hypothesis (EMH), security (e.g., stock) prices fully incorporate all available information (and expectations) and are thus an unbiased valuation of the security (Fama, 1965 & 1970). This implies that stock prices react to changes in information when these changes are unexpected. For example, if the introduction of a new product changes investors’ expectations of (discounted) future cash flows, the stock price will respond. It will go up when the expected change in (discounted) future cash flows is positive, or down when the expected change is negative (given a constant discount rate).

(10)

10 movements by lowering systematic risk. There also exists a negative relationship between customer satisfaction and idiosyncratic risk. Interestingly, these results are not in line with the positive risk-return relationship stressed by asset pricing models like the CAPM, since customer satisfaction has a positive effect on stock returns (see, e.g., Aksoy et al., 2008).

Furthermore, Bharadwaj et al. (2011) find that a firm’s (perceived) brand quality is positively related to stock returns (which Srinivasan et al. (2009) find as well), while idiosyncratic risk shows a negative relationship with brand quality. Yet, even though these two findings enhance shareholder value, increases in brand quality are also accompanied by increases in systematic risk. This might be because consumers do not have constant price sensitivity over the business cycle (e.g., price sensitivity during economic downturns is higher than during economic booms), and since price is positively associated with quality, increases in brand quality may make the firm’s stock more vulnerable to market movements.

The second stream of research focuses on the link between marketing actions, like advertising and price promotions, and shareholder value. An example of a study related to this second stream of research is the study by Homburg et al. (2014), where the authors find that announcing the use of new distribution channels generates positive abnormal returns. However, increasing the distribution intensity (e.g., increasing the number of entities within a certain channel) can have both positive and negative effects on stock returns. For example, increasing the distribution intensity in a highly competitive market is likely to result in a negative response in the return, since the channel entities are not dependent on the specific firm, and since the firm must pay higher costs to ensure that the entity complies with the firm.

A study in the automobile industry by Srinivasan et al. (2009) shows that new product introductions lead to positive abnormal returns, particularly when the new product is supported by significant advertising expenditures. On the other hand, promotional incentives are value-destroying in the long run, since it signals to the market that there is weak demand for the product.

(11)

11 returns. Interestingly, they also find that negative chatter (i.e., negative chatter) has a stronger effect on returns than positive chatter, and that negative chatter increases idiosyncratic risk significantly, while no such effect is found for positive chatter.

Table 1 provides an overview of the previously discussed studies and some other studies on the marketing-finance interface. In the following sections, some of these studies will be used to derive hypotheses on the relationship between product announcements and returns and risk. Please note that table 1 does not provide a complete overview of the literature on the marketing-finance interface. For a broader overview, please see Srinivasan and Hanssens (2009).

2.2 Effects of Product Announcements on Returns and Risk

New products are believed to be vital to a company’s long-term success (Bayus et al., 2003). Yet, relatively many studies on new products and innovation have focused on antecedents of product performance, rather than measuring effects on firm value. The following sections

2.2.1 Returns

Clearly, a new product announcement is accompanied by uncertainty, since the ultimate success of the product in the marketplace is contingent on many factors. As long as investors believe that the new product will provide more benefits than costs to the firm, and thus believe that the net present value (NPV) of the project is positive, an increase in the stock price will result.

Still, even though the total innovation project may increase firm value, a new product announcement does not necessarily do so, since information conveyed in a product announcement may already be incorporated in the stock price. This could be the case when, for example, the product is preannounced, when information about the product has leaked to the market or when investors expect the firm to introduce new products frequently (Pauwels et al., 2004). Therefore, a priori, it is unclear what the effect of new product announcements will be. Yet, previous findings in the literature on product (pre)announcements provide some directions.

(12)

12

TABLE 1

Representative Studies on the Marketing-Finance Interface

Authors (Year) Marketing Assets/Actions Returns Systematic Risk

Idiosyncratic

Risk Method Industry Main Findings Sorescu, Shankar, and

Kushwaha (2007)

Product preannouncements Yes No No Event study and calendar-time portfolio approach

Hardware and software industries

The returns from product preannouncements are positive in the long run. For the short run, only firms that offer specific information about the preannounced product generate positive abnormal returns.

Tuli and Bharadwaj (2009) Customer satisfaction No Yes Yes Fama-French three-factor model

Multiple industries Changes in customer satisfaction have a negative effect on both systematic and idiosyncratic risk.

Rego, Billet, and Morgan (2009)

Brand equity No Yes Yes CAPM Multiple industries Brand equity has a negative relation with systematic and idiosyncratic risk. The effect on idiosyncratic risk is stronger. They also find that the effects are magnified during economic downturns.

Srinivasan, Pauwels, Silva-Risso, and Hanssens (2009)

New products, advertising, promotions, perceived quality

Yes No No Adapted Carhart four-factor model

Automobile industry New product introductions are associated with positive effects on stock returns. The effect is greater when there is substantial advertising support and when the perceived quality of the new product is high. Price promotions have a negative effect on stock returns.

Osinga, Leeflang, Srinivasan, and Wieringa (2011)

Advertising Yes Yes Yes Adapted Carhart four-factor model with Kalman filtering

Pharmaceutical industry Direct-to-consumer advertising (DTCA) increases stock returns, lowers systematic risk and increases idiosyncratic risk. Direct-to-physician (DTP) expenditures have modest positive effects on returns and idiosyncratic risk.

Bharadwaj, Tuli, and Bonfrer (2011)

Brand quality Yes Yes Yes Adapted Fama-French three-factor model

Multiple industries Brand quality is positively related to stock returns and systematic risk, and negatively related to idiosyncratic risk.

Tirunillai and Tellis (2012) User-generated content Yes No Yes Carhart four-factor model, VAR

Multiple industries Chatter (i.e., user-generated content) has a positive effect on stock returns. The relationship is assymetric, with negative chatter having a stronger effect than positive chatter. Idiosyncratic risk increases significantly due to negative chatter.

Homburg, Vollmayr, and Hahn (2014)

Distribution channels Yes No No Event study Multiple industries Channel expansions through new distribution channels have positive effects on stock returns. Increasing the distribution intensity can lead to both positive and negative effects on stock returns, depending on certain contingency factors like competitive intensity.

Thomaz and Swaminathan (2015)

This study (2015)

Strategic marketing alliances

New product announcements

No Yes Yes Yes Yes Yes Event study Event study, GARCH Multiple industries Pharmaceutical industry

The announcement of an alliance decreases systematic and idiosyncratic risk. However, for high levels of network density, the effects are reversed.

(13)

13 from 1975-1988, the average daily abnormal return over a three-day event period is 0.20%.

Sorescu et al. (2007) study product preannouncements in the hard- and software industries. Applying event study methodology, they find that product preannouncements have a positive effect in the short run, but only if they contain specific information (e.g., a price and a launch date) about the product. They argue that the specific information is perceived by the market as a credible signal of the “unobserved product development process”, which leads to an increase in firm value. Similar results are found by Lee et al. (2015), and Mishra and Bhabra (2001). For the long run, Sorescu et al. (2007) find that preannouncements lead to abnormal returns. This is particularly the case when firms keep updating the market (after the preannouncement) with new information about the product.

The effect of new products has been studied by Pauwels et al. (2004) as well. They study the effect on firm value of new products and sales promotions in the automobile industry. Using VAR-methodology, they show that new product introductions increase earnings and firm value, whereas promotions do not. In fact, promotions decrease long term firm value, even though they have positive effects on revenues and short run profits. The results imply that executives should not rely heavily on promotions and instead should focus their efforts on introducing new products.

Based on the previous discussion, the first hypothesis is:

H1: Product announcements increase stock returns.

2.2.2 Systematic Risk

Previous studies have found that marketing can be used to limit the sensitivity of a firm’s cash flows to market movements. For example, referring back to table 1, advertising (Osinga et al., 2011), customer satisfaction (Tuli & Bharadwaj, 2009), and brand equity (Rego et al., 2009) have all been found to lower systematic risk.

(14)

14 In their study on the effect of advertising and R&D on systematic risk, McAlister et al. (2007) argue that R&D can create strategic differentiation, and that it yields the firm greater dynamic efficiency and flexibility compared to competitors, which makes it easier to adapt to changes in the firm’s environment. All these effects are likely to make the firm less sensitive to market movements. Their results show that firms with higher levels of R&D indeed tend to have lower systematic risk. Interestingly however, Luo & Bhattacharya (2009) find that this affect disappears when taking the effect of corporate social responsibility (CSR) into account. In contrast to McAlister et al. (2007), Ho et al. (2004) argue that R&D-intensive firms are likely to be actually more sensitive to market movements than less R&D-intensive firms, since firms generally exploit their R&D opportunities during economic expansions rather than during economic contractions (i.e., R&D expenditures tend to follow a pro-cyclical pattern). Indeed, they find that more R&D-intensive firms tend to have higher systematic risk.

Literature in the field of behavioral finance shows that private investors tend to buy attention-grabbing stocks (Odean, 1999; Barber & Odean, 2008). An explanation for this is that people are unable to consider each and every stock available, because of cognitive and temporal limitations. Generally, when faced with a selection decision, people first select a limited number of alternatives (the consideration set) from all alternatives possible. Next, one or more alternatives in the consideration set are actually chosen. It is argued that attention-grabbing alternatives are more likely to end up in the consideration set and are therefore more likely to ultimately be chosen. Product announcements are likely to be events that draw the attention of individual investors. Consequently, a larger part of a firm’s stock ownership will be held by individual investors, whose buy and sell decisions are likely to be less coordinated than those of institutional investors (Xu & Malkiel, 2003), which should therefore lower systematic risk (Osinga et al., 2011).1

Despite some contradictory evidence, the following is hypothesized:

H2: Product announcements decrease systematic risk.

2.2.3 Idiosyncratic Risk

Firstly, one might wonder why it would be interesting from a shareholder perspective to study idiosyncratic risk, since it should not impact the valuation of securities if investors hold well-diversified portfolios (Osinga et al., 2011). Yet, evidence shows that idiosyncratic volatility

1_{Even if trades of individual investors are (to some extent) coordinated (Barber et al., 2009), the effect on stock prices of}

(15)

15 actually does have an impact, since the average investor does not hold a well-diversified portfolio: Goetzmann and Kumar (2008) show that about 75% of investor portfolios contain no more than five stocks, while at least thirty stocks are needed to have a well-diversified portfolio (Statman, 1987). If investors are unable to hold well-diversified portfolios, they will demand a premium for bearing idiosyncratic risk (Fu, 2009). This implies that idiosyncratic risk will be priced in. Furthermore, idiosyncratic risk represents about 80% of total risk (Goyal & Santa-Clara, 2003). Clearly, these arguments imply that idiosyncratic risk is of interest to both investors and managers (Luo & Bhattacharya, 2009).

As in the case of systematic risk, literature on the relationship of product announcements and idiosyncratic risk is scarce. Yet, product announcements are inherently surrounded by risk, since at the time of the announcement, it is uncertain whether the product will eventually succeed in the market. This may lead to an increase in idiosyncratic risk, since investors may have differing expectations about the future success of the product (and thus differing valuations of the stock). Kothari et al. (2002) show that this argument holds for R&D: firms with higher R&D investments tend to have greater uncertainty about the future benefits of those investments.

Furthermore, Osinga et al. (2011) argue that advertising raises investor awareness. Consequently, investors are likely to react more strongly to company news, which increases idiosyncratic risk. This reasoning could be applied to product announcements as well, since these are also likely to attract investor attention and raise investor awareness about the firm.

Based on the foregoing, the third hypothesis is:

H3: Product announcements increase idiosyncratic risk.

2.3 The Moderating Effect of Innovativeness

Innovativeness can be defined as (1) the degree to which the new product incorporates new technologies and (2) the extent to which it is better able to fulfill customer needs than existing products (Chandy & Tellis, 1998; Sorescu et al., 2003). New products that score high on both dimensions are called radical innovations, and products that score low on both dimensions are called incremental innovations.

2.3.1 Returns

(16)

16 the consumer packaged goods industry. They show that breakthrough innovations have a positive effect on stock returns, but they do not find this effect for incremental innovations.

Srinivasan et al. (2009) study the effects of innovativeness on stock returns in the automobile industry. They find that new product introductions have a positive effect on stock returns. More so, they show that radical (“new-to-the-market”) innovations have a seven times as large impact on stock returns than incremental (“new-to-the-company”) innovations.

In the studies by Chaney et al. (1991), and Chaney and Devinney (1992), the authors split the sample, after the initial analyses, in announcements of truly new products and announcements of products that are merely updates of existing ones. They find that new product announcements are associated with a 0.20% average daily excess return, whereas announcements of product updates have a non-significant 0.11% average daily excess return.

Consequently, the fourth hypothesis is:

H4: Product announcements of radical innovations have a stronger effect on stock returns than

incremental innovations.

2.3.2 Systematic Risk

Firstly, an argument regarding investor attention could be made for the effect of innovativeness on systematic risk as well (see the discussion in section 2.2.2). By their nature, radical innovations are more likely to attract investor attention than incremental innovations. In light of this argument, it can be expected that the effect of product announcements on systematic risk becomes stronger when the innovation is radical compared to incremental. Yet, Sorescu & Spanjol (2008) do not find evidence that radical innovations have a stronger impact on systematic risk than incremental innovations. Particularly, they find that neither has a significant impact on systematic risk. They do find however that radical innovations increase total risk.

Despite lacking empirical evidence, but in line with previous arguments:

H5: Product announcements of radical innovations have a stronger effect on systematic risk

than incremental innovations.

2.3.3 Idiosyncratic Risk

(17)

17 H1 (+) H2 (-) H3 (+) H4 (+) H6 (+) H5 (+)

Spanjol (2008) find, radical innovations are associated with increases in idiosyncratic risk. Furthermore, similar to the arguments presented earlier, radical innovations are likely to raise investor awareness more than incremental innovations. This implies that investors will subsequently react more strongly to company news, making the idiosyncratic part of the stock return more volatile.

Hence, the final hypothesis is:

H6: Product announcements of radical innovations have a stronger effect on idiosyncratic risk

than incremental innovations.

2.4 Research Framework

Figure 1 provides a graphical summary of the hypothesized relationships. The next chapter discusses the empirical context, data, and methodologies.

FIGURE 1 Research Framework

New Product Announcements

Idiosyncratic Risk Systematic Risk

Stock Returns

(18)

18

3. Research Design

This chapter starts with introducing the empirical context. Next, the process undertaken to get the relevant data is outlined and an introduction to the data is given. Finally, the methodologies are discussed.

3.1 Empirical Context

A suitable industry for testing the hypotheses related to product announcements is the (U.S.) pharmaceutical industry. Firstly, new products are common, since the industry is driven by innovations (Sorescu et al., 2003). However, new drugs can only be marketed if the U.S. Food and Drug Administration (FDA) has given its approval. To get FDA approval, an extensive review process has to be undertaken. A brief summary of the review process is as follows (for a more elaborate overview, please visit www.fda.gov):

1. If a firm wants to market a (new) drug in the U.S., it submits a New Drug Application (NDA) to the FDA. The NDA should include data of clinical trials along with analyses of the data and information about the behavior of the drugs and how it is manufactured; 2. After submitting the NDA, the FDA has 60 days to decide whether to file the NDA for review. If the FDA decides not to file the NDA, the firm can submit an NDA in the future with more complete information and analyses. If the NDA is filed by the FDA, the NDA will be reviewed;

3. If the review is favorable, the drug will be approved and marketing of the drug is allowed.

The pharmaceutical industry is suitable for a second reason, since an objective measure of the degree of innovativeness can be used (Sorescu et al., 2003). Specifically, the FDA classifies each NDA as “Standard Review” or “Priority Review” before the NDA is filed. “Priority Review” will be given to NDAs of drugs “…that, if approved, would be significant improvements in the safety or effectiveness of the treatment, diagnosis, or prevention of serious conditions when compared to standard applications” (www.fda.gov).

3.2 Data

(19)

19

TABLE 2

Overview of Data and Sources

Data Source

New product announcements Drugs@FDA database Company websites LexisNexis

Stock prices Datastream

Innovativeness Drugs@FDA database

Firm size Datastream

Carhart factors Kenneth French's website

3.2.1 Data on New Product Announcements

Data on new product announcements is obtained by using a census of the FDA database of new drug approvals from 2011-2015. To get to the final dataset of a list of new product announcements, the following steps are taken:

1. Drug approvals for which no standard or priority review is present are deleted from the dataset (this is the case for approvals of medical gases like oxygen and helium);

2. The remaining observations are checked for whether the firm that gets the FDA approval is publicly listed in the U.S. at the time of approval, and if not, whether it is a division of a (larger) listed firm or whether it is a subsidiary of a listed firm. Furthermore, it is checked whether it concerns a joint venture, partnership or licensing agreement. All observations of non-listed (i.e., private) firms, listed firms in other countries than the U.S., joint ventures, partnerships, licensing agreements, and divisions or subsidiaries of large conglomerates are deleted from the dataset;

3. Next, for every observation, the date for which the approval firstly became public knowledge is determined, since the actual approval may not always be communicated to the public on the same day. The dates are found by consulting the companies’ websites. The dates are checked (and if necessary adapted) by investigating the FDA website and LexisNexis (focusing on PRNewswire and Business Wire). If the approval is in no way accompanied by an announcement (e.g., since it is merely a new dosage form for an already approved drug), the observation is deleted from the dataset;

(20)

20 announcements, dividend announcements, merger and acquisition announcements and approval announcements of other products.

The final product announcement dataset consists of 113 announcements from 2011-2015 by 63 firms. Table 3 shows that most of the firms in the dataset only have one announcement (45 firms), and that there are just two firms with more than five announcements. Not surprisingly, the group of firms that introduced more than one new product tend to be the “big” pharmaceutical companies, like Pfizer, Johnson & Johnson, Novartis and AstraZeneca. This appears to be in line with Sorescu et al. (2003), who argue that these kind of firms are better able to introduce innovations.

TABLE 3

Distribution of Number of Firms per Number of Announcements Nr. of Announcements Nr. of Firms 1 45 2 7 3 2 4 4 5 3 More than 5 2

Total number of firms 63

Table 4 splits the sample according to the degree of innovativeness. 79 of the 113 new products received a standard review (i.e., incremental innovation), whereas 34 of the products were approved under a priority review (i.e., radical innovation). Table 4 also provides the average firm size (in millions) per type of innovation. Interestingly, radical innovations tend to be introduced by relatively large firms, which is again in line with Sorescu et al. (2003).

TABLE 4

Distribution of Standard and Priority Review Review Type

Nr. of Announcements

Average Firm Size (in Millions)

Standard 79 37.6

Priority 34 52.3

Total number of

(21)

21

3.2.2 Return Data

The stock prices of the firms in the dataset are retrieved from Thomson Reuters’ Datastream. Datastream is also used to obtain total assets as a proxy for firm size (which will be used as a control variable). Furthermore, expected returns will be calculated based on the Carhart four-factor model (Carhart, 1997). To this end, Kenneth French’s website is used (http://mba.tuck.dartmouth.edu).

3.3 Event Study

To study the effect of product announcements on stock returns, an event study is a suitable method, since it can be used to assess the impact of firm-specific events on stock returns (Brown and Warner, 1985). The basic idea behind the method is that, according to the efficient market hypothesis, new information should be immediately incorporated into the stock price (Srinivasan and Hanssens, 2009). Consequently, one can study the impact of the event by comparing the return given the event (i.e., the actual return) to the ‘normal’ return that would be expected without the event. The methodology has been widely applied, particularly in the finance and economics literature. Yet, event studies have been relatively frequently applied in the marketing field as well. For example, event study methodology was used by:

 Agrawal and Kamakura (1995), who find that the announcement of a celebrity endorsement contract is positively received by the market;

 Sorescu et al. (2007), who find that product preannouncements can lead to abnormal stock returns, particularly if the firm offers specific information about the product in the preannouncement;

 Wiles and Danielova (2009), showing that product placements in successful films have a positive relationship with stock returns;

 Homburg et al. (2014), who find that announcements of distribution channel expansions have a positive effect on stock returns.

In the following section, the event study methodology will be outlined.

3.3.1 Methodology

As was already mentioned, an event study compares the return given the event with the expected return without the event. The difference is called the abnormal return (𝐴𝑅). More formally,

𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡− 𝐸(𝑅𝑖𝑡) (2)

where 𝐴𝑅𝑖𝑡 is the abnormal return of stock i at time t, 𝑅𝑖𝑡 is the realized return of stock i at time

(22)

22 𝑅_𝑖𝑡 can be calculated as

𝑅𝑖𝑡 = 𝑃𝑖𝑡

𝑃𝑖,𝑡−1− 1 (3)

where 𝑃𝑖𝑡 is the adjusted closing price of stock i at time t (the adjusted closing price has to be used since it is adjusted for stock splits and dividends) and 𝑃𝑖,𝑡−1 is the adjusted closing price of stock i at time t-1.

To calculate expected returns, several different models are often applied in the literature (Kolari and Pynnönen, 2010), like the CAPM, the Fama-French three-factor model, or the Carhart four-factor model. In line with previous studies in marketing (see e.g., Osinga et al., 2011, and Sorescu et al., 2007), the Carhart four-factor model will be used to calculate expected returns. Other specifications are possible, but with short event windows (e.g., a couple of days), the chosen model to calculate expected returns is relatively unimportant, since expected returns are likely to be close to zero over short horizons (Brooks, 2014: 636).

Expected returns according to the Carhart four-factor model (Carhart, 1997) can be calculated as

𝐸(𝑅_𝑖𝑡) = 𝑟𝑓_𝑡+ 𝛽_𝑖,0+ 𝛽_𝑖,1𝐸𝑅𝑀𝐾𝑇_𝑡+ 𝛽_𝑖,2𝑆𝑀𝐵_𝑡+ 𝛽_𝑖,3𝐻𝑀𝐿_𝑡+ 𝛽_𝑖,4𝑀𝑂𝑀_𝑡 (4)

where (1) 𝐸(𝑅_𝑖𝑡) is as defined before, (2) 𝑟𝑓_𝑡 is the risk-free rate at time t, (3) 𝐸𝑅𝑀𝐾𝑇_𝑡 is the excess market return (i.e., the market return minus the risk-free rate) at time t, (4) 𝑆𝑀𝐵𝑡 (“small minus big”) is the Fama-French size factor calculated as the difference in returns between a low market capitalization and high market capitalization portfolio at time t, (5) 𝐻𝑀𝐿_𝑡 (“high minus low”) is the Fama-French market-to-book factor calculated as the difference in returns between a portfolio of high market-to-book (equity) firms and a portfolio of low market-to-book firms at time t, and (6) 𝑀𝑂𝑀_𝑡 is the Carhart momentum factor calculated as the return difference between a portfolio of previous winners and a portfolio of previous losers at time t.

To estimate the parameters of the Carhart four-factor model, the following regression is run for each stock separately

𝐸𝑅_𝑖𝑡 = 𝛽_𝑖,0+ 𝛽_𝑖,1𝐸𝑅𝑀𝐾𝑇_𝑡+ 𝛽_𝑖,2𝑆𝑀𝐵_𝑡+ 𝛽_𝑖,3𝐻𝑀𝐿_𝑡+ 𝛽_𝑖,4𝑀𝑂𝑀_𝑡+ 𝜀_𝑖𝑡 (5)

(23)

23 When the abnormal returns are determined, the cumulative abnormal return (𝐶𝐴𝑅) over the entire event window (this concept will be defined later), starting at 𝑡1 and ending at 𝑡2, can be calculated as

𝐶𝐴𝑅_𝑖(𝑡₁, 𝑡₂) = ∑𝑡2 𝐴𝑅_𝑖𝑡

𝑡=𝑡1 (6)

where 𝐶𝐴𝑅_𝑖(𝑡₁, 𝑡₂) is the cumulative abnormal return of stock i from time 𝑡₁ to 𝑡₂, and 𝐴𝑅_𝑖𝑡 is as defined before.

Finally, these cumulative abnormal returns can be averaged over all N stocks to find the cumulative average abnormal return (𝐶𝐴𝐴𝑅)

𝐶𝐴𝐴𝑅(𝑡₁, 𝑡₂) = 1

𝑁∑ 𝐶𝐴𝑅𝑖(𝑡1, 𝑡2) 𝑁

𝑖=1 (7)

where 𝐶𝐴𝐴𝑅(𝑡1, 𝑡2) is the cumulative average abnormal return, and 𝐶𝐴𝑅𝑖(𝑡1, 𝑡2) is as defined before. This final measure can be used in a t-test for the hypothesis testing, where the standard error of the 𝐶𝐴𝐴𝑅(𝑡₁, 𝑡₂) is based on the cross-sectional volatility of returns in the estimation window.

3.3.2 Determining the Event and Estimation Window

In an event study, it is important to precisely define the event date. This seems like a straightforward task, but it may actually be quite difficult (Henderson, 1989). The difficulty is that, even if there is no uncertainty concerning when the event took place, it may be that the market is unable to respond immediately. For example, it could be the case that a certain event took place after the stock market had closed. Consequently, the first day the market can respond to the event is the next trading day. If the time of the event is known, the event date can be adjusted for this fact. Yet, as in the case of this study, the actual time an announcement became public knowledge may not be easy to determine, simply because the (press) announcements are not accompanied by a time stamp. A second problem is that it might be that information about the event has already leaked to the market before the actual event took place, called information leakage (Binder, 1998).

(24)

24 To be able to calculate the expected returns in the event window, the parameters of the Carhart four-factor model have to be estimated in the so-called estimation window. The estimation window generally does not overlap with the event window, so that the parameter estimates are not influenced by the event. Determining an appropriate length of the estimation window is up to the researcher. Since using an estimation window of more than 100 days leads to problems in the current study, an estimation window of day -110 (compared to the event date) up to and including day -11 is used. Extending the estimation window has the consequence that some observations drop out, because not for all firms stock prices are available for the entire corresponding estimation window (i.e., because the firms were not publicly listed yet at the start of the extended estimation window).2

Figure 2 provides a graphical overview of the estimation and event windows.

3.3.2 Cross-Sectional Regression

To test whether radical innovations have a stronger effect on returns than incremental innovations, a cross-sectional regression will be carried out. The regression takes the following form

𝐶𝐴𝑅𝑖(𝑡1, 𝑡2) = 𝛿0+ 𝛿1𝐼𝑁𝑉𝑖+ 𝛿2𝑆𝐼𝑍𝐸𝑖 + 𝜀𝑖 (8) where 𝐶𝐴𝑅𝑖(𝑡1, 𝑡2) is as defined before, 𝐼𝑁𝑉𝑖 represents the degree of innovativeness for stock

i which equals 0 for a standard review and 1 for a priority review, and 𝑆𝐼𝑍𝐸_𝑖 is defined as the size (in millions) of the firm of stock i, as proxied by total assets, at the time of the announcement. The 𝐶𝐴𝑅_𝑖(𝑡₁, 𝑡₂)-measure used as dependent variable will be based on the event window which yields the most significant results in the event study.

2_{The results are not significantly changed when extending the estimation window and dropping the observations}

that do not have complete data for the entire extended estimation window. The same is true if the observations are retained and estimates for these observations are based on the maximum possible estimation window, and the parameter estimates for the other observations are based on the entire extended estimation window.

Estimation Window Event

Window

-110 -11 t1 t2

FIGURE 2

(25)

25 3.3 Risk and GARCH-modeling

To study the effect of product announcements on systematic and idiosyncratic risk, in light of parsimoniousness, a model is needed that is able to model both effects simultaneously. To this end, GARCH-models can be used. These models simultaneously estimate a mean equation and a variance equation. The effect of product announcements on systematic risk can be captured in the mean equation, and the effect on idiosyncratic risk can be modeled in the variance equation. Furthermore, return series are generally characterized by volatility clustering, which can be accurately incorporated using GARCH-like models. GARCH has not been often applied in the marketing field, but it has received significant attention in the financial literature.

In the context of this study, the model is based on the Carhart four-factor model

𝐸𝑅_𝑖𝑡 = 𝛽_𝑖,0+ 𝛽_𝑖,1𝐸𝑅𝑀𝐾𝑇_𝑡+ 𝛽_𝑖,2𝑆𝑀𝐵_𝑡+ 𝛽_𝑖,3𝐻𝑀𝐿_𝑡+ 𝛽_𝑖,4𝑀𝑂𝑀_𝑡+ 𝜀_𝑖𝑡 (9)

where the variables are as defined before.

To incorporate the effects on systematic risk, equation (9) is altered as follows

𝐸𝑅_𝑖𝑡 = 𝛽_𝑖,0+ 𝛽_𝑖,1𝐸𝑅𝑀𝐾𝑇_𝑡+ 𝛽_𝑖,2𝑆𝑀𝐵_𝑡+ 𝛽_𝑖,3𝐻𝑀𝐿_𝑡+ 𝛽_𝑖,4𝑀𝑂𝑀_𝑡+ 𝛽_𝑖,5𝐷_1𝑡+ 𝛽_𝑖,0∆𝑃𝐷_2𝑡+ 𝛽_𝑖,1∆𝐸𝐸𝑅𝑀𝐾𝑇_𝑡𝐷_1𝑡+ 𝛽_𝑖,1∆𝑃𝐸𝑅𝑀𝐾𝑇_𝑡𝐷_2𝑡+ 𝜀_𝑖𝑡 (10)

where 𝐷1𝑡 is a dummy variable that equals one for observations in the event window, and zero for observations outside the event window, and 𝐷_2𝑡 is a dummy variable that is equal to one for observations after the event window, and zero elsewhere. 𝛽_𝑖,1∆𝐸 (‘E’ stands for ‘event window’) then captures the effect on systematic risk during the event window, and 𝛽_𝑖,1∆𝑃 (‘P’ stands for ‘post-event’) captures the effect after the event window. To take into account that abnormal returns may be earned in the event window, the main effect of 𝐷_1𝑡 is included. Furthermore, the main effect of 𝐷_2𝑡 is included as well, since the constant (as represented by 𝛽_𝑖,0) may shift as well because of the product announcement.

(26)

26 For this second model, following Cyree and Degenarro (2002), and Uhde and Michalak (2010), 𝛽_𝑖𝑡,1∆𝐸 and 𝛽_𝑖,1∆𝑃 are assumed to take the following forms

𝛽_𝑖𝑡,1∆𝐸 = 𝛽_𝑖,6(𝑡1− 𝑡)(𝑡 − 𝑡2) + 𝛽𝑖,7(𝑡 − 𝑡1) (11) and

𝛽_𝑖,1∆𝑃 = 𝛽_𝑖,7(𝑡₂− 𝑡₁) (12)

where 𝑡₁ and 𝑡₂ represent the start and the end of the event window, respectively.

These specifications allow systematic risk to change during the event window in a linear or non-linear way (since 𝑡2 enters equation (11)), and allow systematic risk to exit the event window at a permanently different level. Figure 3 provides a graphical example of how systematic risk can change according to the specifications: before the start of the event window (assumed to be day -5 in the graph) systematic risk is constant, during the event window (from day -5 to day 5) the change in systematic risk can be concave, convex or linear, and systematic risk may exit the event window at a permanently different level (in the graphical example of figure 3, systematic risk exits the event window at a lower level compared to before the event window).

FIGURE 3

Graphical Example of Linear and Non-Linear Changes in Systematic Risk

(27)

27 If there is a permanent shift in systematic risk, this will be captured by a significant coefficient for 𝛽_𝑖,7. Depending on the sign of 𝛽_𝑖,6, the change in systematic risk during the event window can either be convex (if 𝛽_𝑖,6 is negative), concave (if 𝛽_𝑖,6 is positive) or linear (if 𝛽_𝑖,6= 0).

For both models, a GARCH(1,1)-specification is chosen for the variance equation, since it is generally sufficient to capture the volatility process in the data (Brooks, 2014: 430). The variance equation takes the following form

𝜎_𝑖𝑡2 = 𝛼_𝑖,0+ 𝛼_𝑖,1𝜀_{𝑖,𝑡−1}2 + 𝛼_𝑖,2𝜎_{𝑖,𝑡−1}2 + 𝛼_𝑖,3𝐷_1𝑡+ 𝛼_𝑖,4𝐷_2𝑡 (14)

where 𝜎_𝑖𝑡2 represents the conditional variance for stock i at time t, 𝜀_{𝑖,𝑡−1}2 is the lagged squared residual for stock i, 𝜎_{𝑖,𝑡−1}2 is the lagged conditional variance for stock i, and the dummy variables 𝐷_1𝑡 and 𝐷_2𝑡 are as defined previously.

The mean equations (10) and (13) can be estimated simultaneously with the variance equation (14) using Maximum Likelihood Estimation (MLE).

The estimation will be done per event over a period from 110 days before the event to 110 days after the event, which yields a total of 221 trading days (including the event day) for each observation, which is in line with Uhde and Michalak (2010). For some stocks, prices are unavailable for the entire 221 days. These stocks will be left out of the analyses (N = 101 after the deletion). To take into account the fact that stock-specific event windows might differ from the average most significant event window, the estimates will be obtained for the event window [-5, 5]. The results will be aggregated over the full sample and per type of innovation. Significance will be tested using a simple t-test and a Wilcoxon signed-rank test. Furthermore, the average results of the two groups will be compared using t-tests. Alternatively, to take into account the uncertainty with which the individual parameters are estimated, each parameter estimate will be multiplied by a weighting factor (Wittink, 1977). This weighting factor is calculated as the ratio of the (absolute) parameter estimate to its standard error.

4. Results

The following chapter provides the results of the different analyses. It starts with the results of the event study and the cross-sectional regression. Next, the results of the GARCH-modeling will be presented.

4.1 The Effect of Product Announcements on Returns

(28)

28 and the subsample of radical innovations. For each event window, the cumulative average abnormal return along with the corresponding t-value is presented. Furthermore, the total number of positive and negative abnormal returns are provided for each event window.

As table 5 shows, in line with hypothesis 1, product announcements are generally accompanied by significantly positive abnormal returns surrounding the announcement day. Particularly, for the full sample, the abnormal returns are significant (p at least smaller than 0.05 for a one-tailed test) for all event windows except for [-10, 10]. Interestingly, even though on average the abnormal return associated with a product announcement is significantly positive, some firms experienced a negative abnormal return in the days surrounding the announcement. This implies that the market generally reacts favorably to new products, but that this is not always the case.

The most significant event window includes the announcement day and the day following the announcement, where the total abnormal return is 3.10% (t = 6.95, p < 0.01) over the two-day event window. That the most significant event window includes the day after the announcement, might be an indication that the market does not immediately incorporate the new information appropriately in the stock price, which is in contrast with the efficient market hypothesis. Yet, a more likely explanation is that it is caused by the fact that the event dates are probably not perfectly determined. Particularly, as was already mentioned before, the actual times the (press) announcements were made public are impossible to determine, since this information is not available. It is likely that some of the announcements were made after the stock markets had closed, implying that the day after the announcement is actually the event day.

Figure 4 shows graphically how the average abnormal return cumulates over the event window [-10, 10]. To interpret this graph, please note that, for example, the value at t = -5 represents the cumulative average abnormal return for the event window [-10, -5]. Similarly, for t = 2, the graph shows the cumulative average abnormal return for [-10, 2]. Consequently, at t = 10, the cumulative average abnormal return for [-10, 10] is shown, which corresponds to the 1.18% found in table 5.

(29)

29

TABLE 5

CAAR for Different Event Windows and Samples Carhart Four-Factor Model

Full Sample (n=113) Incremental (n=79) Radical (n=34)

Event Window CAAR

(t-value) Pos./Neg. CAAR (t-value) Pos./Neg. CAAR (t-value) Pos./Neg. [-10, 10] 1.18% 58/55 0.56% 40/39 2.61% 18/16 (0.82) (0.33) (0.93) [-5, 5] 2.09% ** _60/53 _1.79%* _39/40 _2.79%* _21/13 (2.00) (1.47) (1.38) [-3, 3] 2.46% *** _67/46 _2.11%** _47/32 _3.27%** _20/14 (2.95) (2.18) (2.02) [-2, 2] 2.37% *** _63/50 _2.14%*** _41/38 _2.91%** _22/12 (3.36) (2.61) (2.13) [-1, 1] 3.11% *** _74/39 _3.08%*** _52/27 _3.17%*** _22/12 (5.69) (4.86) (2.99) [0, 0] 1.95% *** _67/46 _1.74%*** _45/34 _2.43%*** _22/12 (6.18) (4.76) (3.98) [0, 1] 3.10% *** _75/38 _3.06%*** _51/28 _3.18%*** _24/10 (6.95) (5.90) (3.68) [0, 2] 2.54% *** _75/38 _2.40%*** _53/26 _2.84%*** _22/12 (4.65) (3.79) (2.68)

*_{p < 0.10 (one-tailed t-test),}**_{p < 0.05 (one-tailed t-test),}*** _{p < 0.01 (one-tailed t-test)}

In figure 4, a clear spike can be observed during the announcement day (day 0) and the day following the announcement (day 1). Interestingly and surprisingly, after day 1, there appears to be downward pressure on the stock price. This negative abnormal return is significant over the event window [2, 10] (t = -1.98, p < 0.05). This could be an indication that the stock market generally overreacts to new product announcements.

(30)

30

FIGURE 4

Evolvement of Cumulative Average Abnormal Return (in %) over Event Window [-10, 10], Carhart Model

Another issue that is worthwhile investigating with respect to the observed effect is whether it is caused by including a constant in the calculations of the expected returns. Theoretically, the constant should be equal to zero. Still, to be able to calculate the expected returns in the event window, the constant is estimated in the estimation window. Yet, one needs to be careful when including the estimated constant in the expected return calculations, since the constant may be relatively high (low) during the estimation window because of some unrelated event or in anticipation of the actual event (Brooks, 2014: 637). Since one is interested in what the effect of the event is compared to a situation without the event, it may be preferable to exclude the constant to calculate the expected returns. Particularly, since the average estimated constant in the sample (0.00118) is significantly positive (t = 4.96, p < 0.01), this pushes the expected returns up and thus the abnormal returns down. This becomes more pronounced when the abnormal returns are cumulated over a certain time interval.

The expected returns were recalculated by excluding the constant for each observation. The corresponding graph can be found in figure 6. When comparing figure 6 with figure 4, including the constants clearly pushes the abnormal returns down.

The question is which of the two specifications is the most reliable in the current setting. Clearly, including the constants will be more in line with the actual data. Yet, intuitively, figure 6 is more in line with what could reasonably be expected: there is some upward pressure on the stock price in the days before the event in anticipation of the announcement, during the event day (and the day after the announcement) a jump in the stock price can be observed, and

(31)

31 thereafter there is no significant abnormal return (the decrease that can be observed from day 2 to 10 is not significant in figure 6: t = -0.85, p > 0.10).3

FIGURE 5

Evolvement of Cumulative Average Abnormal Return (in %) over Event Window [-10, 10], CAPM (Left) and Mean Return (Right)

Furthermore, new products are often preannounced in the pharmaceutical industry (Sorescu et al., 2007). More specifically, pharmaceutical companies generally send out a press announcement when their new drug has been sent to the FDA for review. Since the FDA review process can take considerable time, the estimation window may already include effects caused by the actual approval announcement (i.e., the product announcement) that happens after the estimation window.

Based on the arguments discussed above, it may be more appropriate in the setting of this study to exclude the constants from the expected return calculations. Table 6 provides the corresponding results for different event windows. Results are provided for the full sample, the subsample of incremental innovations and the subsample of radical innovations. For each event window, the cumulative average abnormal return along with the corresponding t-value is presented. Furthermore, the total number of positive and negative abnormal returns are provided for each event window.

For the full sample, the cumulative average abnormal return is significantly positive for all event windows considered (p < 0.01 for all event windows). The widest event window [-10, 10] shows a cumulative average abnormal return of 3.66% (t = 2.52, p < 0.01), which corresponds to an average daily abnormal return of 0.17% over the 21 days in the event window. Again, the

3_{The observed decrease in figure 6 seems to be driven by three announcements that were accompanied by large}

(32)

32

FIGURE 6

Evolvement of Cumulative Average Abnormal Return (in %) over Event Window [-10, 10], Excluding Constants

TABLE 6

CAAR for Different Event Windows and Samples Carhart Four-Factor Model, Excluding Constants

Full Sample (n=113) Incremental (n=79) Radical (n=34)

Event Window CAAR

(t-value) Pos./Neg. CAAR (t-value) Pos./Neg. CAAR (t-value) Pos./Neg. [-10, 10] 3.66% *** _66/47 _3.74%** _47/32 _3.47% _19/15 (2.52) (2.22) (1.24) [-5, 5] 3.39% *** _63/50 _3.45%*** _44/35 _3.24%* _19/15 (3.23) (2.83) (1.60) [-3, 3] 3.29% *** _69/44 _3.17%*** _51/28 _3.56%** _18/16 (3.93) (3.26) (2.20) [-2, 2] 2.96% *** _71/42 _2.90%*** _48/31 _3.11%** _23/11 (4.19) (3.52) (2.27) [-1, 1] 3.46% *** _78/35 _3.53%*** _55/24 _3.29%*** _23/11 (6.31) (5.55) (3.10) [0, 0] 2.07% *** _69/44 _1.89%*** _46/33 _2.47%*** _23/11 (6.53) (5.15) (4.03) [0, 1] 3.33% *** _81/32 _3.36%*** _57/22 _3.27%*** _24/10 (7.45) (6.46) (3.77) [0, 2] 2.89% *** _80/33 _2.86%*** _59/20 _2.96%*** _21/13 (5.27) (4.49) (2.80)

*_{p < 0.10 (one-tailed t-test),}**_{p < 0.05 (one-tailed t-test),}*** _{p < 0.01 (one-tailed t-test)}

most significant event window includes the announcement day and the day after the announcement, with a cumulative average abnormal return of 3.33% (t = 7.45, p < 0.01) over the two-day event window. More specifically, the average abnormal return on day 0 is 2.07%

(33)

33 (t = 6.53, p < 0.01) and 1.26% on day 1 (t = 4.00, p < 0.01). Furthermore, during the two-day event window, 81 announcements were accompanied by a positive stock market reaction, whereas the remaining 32 announcements were negatively received by the market. Perhaps, for the latter announcements, the terms of the FDA approval were worse than expected (e.g., if unexpectedly the new drug is not approved for a particular age group). Interestingly, the cumulative average abnormal return for the [0, 2] event window is lower than that for the [0, 1] event window, which means that there is a negative average abnormal return on day 2, equal to -0.44% (t = -1.40, p < 0.10). This indicates that there appears to be, on average, a small correction in the stock price two days after the announcement. Yet, taking into account that this result is only marginally significant, it would be inappropriate to claim that there is strong evidence that the stock market generally overreacts to a product announcement.

For the subsample of incremental innovations, all the considered event windows show a significantly positive cumulative average abnormal return (p < 0.05 for event window [-10, 10] and p < 0.01 for all other event windows). The most significant event window is [0, 1] with a cumulative average abnormal return of 3.36% (t = 6.46, p < 0.01). The results are quantitatively and qualitatively similar to the results of the full sample.

For the radical innovations, the event window [-10, 10] is not significant (t = 1.24, p > 0.10). The other event windows are at least significant at the 10%-level. The most significant event window only includes the day of the announcement with a cumulative average abnormal return of 2.47% (t = 4.03, p < 0.01).

As a robustness check, the t-statistics are recalculated by estimating the standard error of the cumulative average abnormal return by the cross-sectional volatility during the event window, rather than during the estimation window, since the event may induce an increase in the variance (Brooks, 2014: 641). The results (not reported) indicate that for the full sample and the subsample of incremental innovations, the cumulative average abnormal return is significant at the 1%-level for all event windows. For the subsample of radical innovations, the event windows [-10, 10], [-5, 5], and [-2, 2] are significant at the 10%-level. The other event windows are significant at the 5%-level.

(34)

34 event window [0, 1] yielded the most significant results in the event study on the full sample, this event window is used.4_{Table 7 provides the coefficients for, respectively, the constant, the} type of innovation, and firm size, along with the corresponding t-values.

The regression model as a whole is significant (F = 3.20, p < 0.05) with an adjusted R-squared of 0.038. In line with previous research (e.g., Sorescu et al., 2007), firm size has a negative impact on the cumulative abnormal return earned. Since returns represent percentual changes in firm equity value, larger firms will show a lower return for a certain innovation project than a smaller firm with a similar innovation project. For every million increase in firm size (as measured by total assets), the cumulative abnormal return is about 0.04% lower.

Interestingly, the type of innovation does not appear to have a significant impact on the cumulative abnormal return earned (p > 0.10), even though the effect is in the hypothesized direction. This implies that the market does not appear to react significantly different to radical innovations than to incremental innovations, even if one controls for firm size.

TABLE 7

Cross-Sectional Regression of CARi for Event Window [0, 1] Variable Coefficient (t-value)

Constant 5.00 *** (3.98) INVi 0.55 (0.28) SIZEi -0.04 ** (-2.53) F 3.20** Adj. R2 _0.038 *_{p < 0.10 (two-tailed),}**_{p < 0.05 (two-tailed),} ***_{p < 0.01 (two-tailed)}

4.2 The Effect of Product Announcements on Systematic and Idiosyncratic Risk Table 8 provides the results of the GARCH-modeling. The results are shown for the event window [-5, 5]. The table provides the mean estimates for the coefficients for different samples and models, along with the corresponding t-values. The different samples considered are the full sample, the sample of incremental innovations and the sample of radical innovations. The two models are the model as presented in equation (10) (called “Linear” in table 8) and the

(35)

35 model that allows for non-linear changes in systematic risk (equation (13), called “Non-linear” in table 8). Significant results according to t-test and Wilcoxon signed-rank test respectively are marked. If the results according to the tests are similar, only the results of the t-test will be reported in the text. If they differ, results of both tests will be presented. The interpretation of the results will focus mainly on the effects related to the hypotheses. Furthermore, to avoid repetition (since the results of the two different models are similar), interpretation will only be done for the “linear” model. That the results are similar shows that the findings are robust to using different specifications with respect to how systematic risk changes.

The results of the full sample indicate that even in this more complex model (as compared to the model used in the event study), an average abnormal return during the event window is still detected (t = 2.18, p < 0.05, and p > 0.10 according to the Wilcoxon signed-rank test). Please note that this result and the results of the event study cannot be easily compared. Specifically, the GARCH-models include just as much trading days after the event as before the event. This is in contrast with the event study, where the expected returns are calculated based on parameter estimates obtained using only trading days before the event. Furthermore, the parameters may change because of the event (in this study the change in systematic risk is accounted for, but not the change in other factors), which can lead to different parameter estimates in the GARCH-models as compared to the estimates obtained for the event study calculations. Still, the results in table 8 provide support for the results of the event study.

There is weak evidence that the constant changes after the event (p < 0.10 according to the Wilcoxon signed-rank test). Interestingly, the constant is significantly positive before the event (t = 1.78, p < 0.10, and p < 0.01 according to the Wilcoxon singed-rank test), which provides evidence that there may be some anticipation of the event. Then, after the event, this effect disappears and the constant is close to zero (i.e., 𝛽0 + 𝛽0𝛥𝑃 ≈ 0).

During the event window, there appears to be no significant change in systematic risk (t = 0.44,

p > 0.10). Yet, after the event window, systematic risk seems to increase on average by 0.12 (t

New Product Announcements and Shareholder Value

New Product Announcements

and Shareholder Value

A Study in the Pharmaceutical Industry

New Product Announcements

and Shareholder Value

A Study in the Pharmaceutical Industry

Abstract

Preface

Table of contents

1. Introduction

2. Theoretical Framework

3.

Research Design

4.

Results