Faculty of Economics and Business – Master of Finance
Product Recall:
The effect of increased consumer awareness on stock price
Name: Daniël H. van de Luijtgaarden Student Number: 0223611
Study: Business Economics Specialization: Finance
Supervisor: Dr. Philippe J.P.M. Versijp
1
Product Recall:
The effect of increased consumer awareness on stock price
*Daniël H. van de Luijtgaarden – Student Number 0223611
University of Amsterdam, Faculty of Economics and Business – Master of Finance
Master’s thesis, November 19th 2014
Abstract
This thesis researches 156 consumer product recall announcements announced between 1987 till and including 2013 to test whether the announcement caused a significantly more negative abnormal stock return when the recall announcement was made in the period 2001 till and including 2013. Furthermore this thesis sets out to test if various other variables had influence on the impact of a consumer product recall announcement. To test the hypotheses the randomly picked events were placed into two sample groups which were subsequently compared with each other. The research finds that the market’s reaction is significantly more negative to recall announcements made between 2001 till and including 2013 than recall announcements made between 1987 till and including 2000. During this research it was also found that if the retail value of the recall estimated approximately 0,1% of the recalling company’s market value it also contributed significantly negative to the abnormal stock return.
*
I would like to take this opportunity to thank several people for their help and support. Firstly, I would like to thank Philippe Versijp for his helpful comments, insight and guidance while supervising this thesis. Secondly, my thanks go to Barend van Drooge for his tips and the useful discussions on how to approach various aspects of this study. Finally, my thanks go out to my family and friends for their endless support.
2
Table of Contents
1. Introduction 04
1.1 Rationale and Contribution 05
1.2 Research Question, Objectives and Hypotheses 05
1.3 Thesis Structure 08
2. Literature Review 09
2.1 A Product Recall Announcement 09
2.1.1 Consumer Product Safety Commission 10 2.2 Previous research on the financial impact of a product recall 10 2.2.1 Product recalls in the automobile industry 11 2.2.2 Product recalls in the pharmaceutical industry 12 2.2.3 Product recalls consumer products 13
2.3 Incidents, Injuries and Remedies 14
2.3.1 Media Attention 14
2.3.2 Recall Remedies 14
2.4 Consumer Awareness 15
3. Research Method 16
3.1 Product Recall Event Sample 16
3.2 Additional Data 18
3.2.1 Share Price Return 18
3.2.2 Market Index 18
3.2.3 Estimated retail Value 19
3.2.4 Market Value 19
3.2.5 Incidents and Injuries 19
3.2.6 Remedy 20
3.2.7 Country of Manufacture 20
4. Methodology 20
4.1 Event Window 21
4.2 Abnormal Returns 21
3
4.2.2 Ordinary Least Squares Market Model 22
4.2.3 Measuring Abnormal Returns 23
4.3 Ordinary Least Squares Regression 23
5. Descriptive Statistics 25
5.1 Event Sample Characteristics 25
5.1.1 Event Sample Categorization 25
5.1.2 Event Sample Value Distribution 26 5.1.3 Event Sample Remainder Variables 27
6. Results 28
6.1 Event comparison group A and B 28
6.2 The first group of hypotheses 29
6.3 Impact of the variables 30
6.3.1 Impact RVV 30
6.3.2 Impact Incidents and Injuries 30
6.3.3 Impact Remedies 31
6.3.4 Impact Country of Manufacture 31
6.4 The second group of hypotheses 31
7. Conclusion 34 7.1 Discussion 35 8. Bibliography 36 8.1 Consulted websites 38 Table 1 39 Table 2 44 Table 3 45 Table 4 46 Table 5 47
Appendix – Event sample composition 48
4
1.
Introduction
The aim of this thesis is to examine the impact of increased consumer awareness on the stock market’s response to a product recall announcement. This research assumes that nowadays consumers are better informed about the products they buy due to the readily availability of new media like the Internet. Therefore, this research hypothesizes that this reflects on the market by increased negative abnormal stock return after a product recall announcement.
By using this new media and with only a few mouse clicks away, a quick look will learn that during the writing of this thesis the Consumer Product Safety Commission (CPSC) issued nineteen different consumer product recalls. These recalls totaled roughly 3,1 million units with a retail value of no less than $640 million. Keeping in mind that in the United States the CPSC is only one of six federal agencies who regulate various product markets and that recalls made in the rest of the world aren’t even considered for this example which shows the considerable magnitude of consumer product recalls.
In 1972 the U.S. Congress enacted the Consumer Product Safety Act. It was believed that an unacceptable amount of faulty consumer products was in circulation. These presented an unreasonable risk of injury and the Congress felt that the general public should be protected against them. As a consequence the CPSC was founded to develop new and uniform safety regulations, assist consumers in evaluating products and promote research directed at the prevention of product related deaths and injuries (CPSC, 2014). Later, with the introduction of the Electronic Freedom of Information Act (FOIA) Amendments of 1996, all federal agencies were required to use electronic technology to better serve the general public, including the CPSC. Which means that today one can subscribe to the Twitter feed of the CPSC, their Youtube channel, Google Plus and their Blog and almost instantly be informed about new consumer product recalls. If one wants to stay informed, no recall can ever escape attention. Therefore, a poorly executed product recall may lead to the loss of an entire product line, the loss of a company’s market share, reputational damage or even to the demise of a company (Chen, 2009)(Van Heerde, 2007). Since the consumer market has become more aware companies should be extra alert. Hora (2011, p6) put it nicely and said: ‘the good news is that an effective recall can minimize short-term damage and guarantee long-term survival’.
5
The numerous consumer product recalls are being issued for various reasons and under different circumstances. Some recalls are made out of precaution while others are the result of an incident. Therefore, this study will also examine the effect of a variety of variables on the impact of a consumer product recall announcement. These results will give some extra insight in answering the main research question.
1.1
Rationale and Contribution
This thesis aims to explore the role of increased consumer awareness on the market’s reaction on consumer product recall announcements, as was mentioned in the introduction. The idea for this topic arose from all the product recall announcements that seemed to show up time and time again. After being inspired by this topic, the literature seemed mostly to describe recall announcements made in the automobile and pharmaceutical industry. Especially for literature that described the impact of a product recall on the underlying security. However, there are few studies done on the impact of consumer product recalls. The studies that do exist only targeted very specific companies or limits itself to a short time window. None of those studies tried to examine if there was a significant difference within their sample, by splitting their sample into two separate time windows. This thesis attempts to contribute to the discussion by examining if increased consumer awareness, and therefore the passage of time, has an effect on the market’s response to consumer product recalls.
1.2
Research Question, Objectives and Hypotheses
As mentioned in the previous paragraph, the aim of this research is to determine what the effect of a consumer product recall announcement is on the underlying security and if this effect has changed over time. Since the consumer market has changed due to new media and consumers generally being better informed, this research question can be divided into two separate objectives:
1. Is there a significant difference in the market’s response to a consumer product recall announcement after the start of the internet revolution in the year 2000? 2. What is the impact of various variables on the market’s response?
6
These research objectives can subsequently be divided into two groups of hypotheses. From the first objective follows the first group of hypotheses:
Hypothesis 1(0) Consumer product recall announcements will cause no significant effect in abnormal stock returns.
Hypothesis 1(1) Consumer product recall announcements will cause significant effect in abnormal stock returns.
Hypothesis 2(0) Consumer Product recall announcements will cause no significant effect in abnormal stock returns from 2001 to and including the year 2013.
Hypothesis 2(1): Consumer Product recall announcements will cause a significant effect in abnormal stock returns from 2001 to and including the year 2013..
These hypotheses will be answered by comparing consumer product recall announcements that were issued from the year 1987 up to and including the year 2013.
The second group of hypotheses will consist of several hypotheses who each address a specific variable. In this research the following variables will be tested: recall value, incidents, injuries, recall remedy and country of manufacture. The first hypothesis in this subset will target the variable “product volume”:
Hypothesis 3(0) A high recall value with respect to the market value of the recalling company has no significant negative effect.
Hypothesis 3(1) A high recall value with respect to the market value of the recalling company has a significant negative effect.
The next hypotheses will target the variables “incident” and “injury”. The thought behind these hypotheses is that before new media became relevant the consumer had to rely on conventional media to get informed. From there follow the next hypotheses for the variable “incident”:
Hypothesis 4(0) Incidents have no significant negative effect on a consumer product recall announcement.
7
Hypothesis 4(1) Incidents have a significant negative effect on a consumer product recall announcement.
Hypothesis 5(0) Incidents don’t have more negative effect on a consumer product recall announcement when the recall occurred before the year 2001.
Hypothesis 5(1) Incidents have a more negative effect on a consumer product recall announcement when the recall occurred before the year 2001.
and the hypotheses for the variable “injury”:
Hypothesis 6(0) Injuries have no significant negative effect on a consumer product recall announcement.
Hypothesis 6(1) Injuries have a significant negative effect on a consumer product recall announcement.
Hypothesis 7(0) Injuries don’t have a more negative effect on a consumer product recall announcement when the recall occurred before the year 2001.
Hypothesis 7(1) Injuries have a more negative effect on a consumer product recall announcement when the recall occurred before the year 2001.
The following hypotheses target the variable “recall remedy”. This variable is a group of four variables: “refund”, “replacement”, “repair” and “discard”. The thought behind this hypotheses is that a cash refund to the consumer is more costly than a replacement of the product. Which in turn is assumed to be more costly than an offered repair. The variable “discard” would bear no direct recall cost. The hypothesis will be:
Hypotheses 8(0) A product refund will not cause the most negative reaction of all the remedies that are being offered.
Hypothesis 8(1a) A product refund will cause the most negative reaction of all the remedies that are being offered.
Hypothesis 8(1b) A product replacement will cause the most negative reaction of all the remedies that are being offered.
Hypothesis 8(1c) A product repair will cause the most negative reaction of all the remedies that are being offered.
8
Hypothesis 8(1d) A discard request will cause the most negative reaction of all the remedies that are being offered.
The final hypothesis in the second group of hypotheses targets the variable “country of manufacture”. The different countries that were featured in the event sample are each allocated to a subset of either “advanced economy” or “non-advanced economy” as will be discussed at a later moment. The hypothesis assumes that products manufactured in advanced economies tend to be of more quality and therefore are less likely to default. The hypothesis that follows is:
Hypothesis 9(0) The market doesn’t react significantly more negative when a product that is made in a developed country is being recalled.
Hypothesis 9(1) The market will react significantly more negative when a product that is made in a developed country is being recalled.
1.3
Thesis Structure
This thesis is written in a way so that it provides a logical structure that will ultimately lead to answering the raised hypotheses. Chapter two will start off with giving a short but concise overview of relevant literature. That chapter will mainly discuss literature of previous findings of the market’s reaction on product recall announcements. For the most part this literature can be classified into three different groups of industries being: automobile, pharmaceutical and consumer products industries. To measure the effect of increased consumer awareness on product recalls this study will focus on the consumer product industries since product recalls issued by the automobile and pharmaceutical industries have generally been covered by the conventional media. The next chapters, chapter three, four and five will discuss the various steps that were conducted for this research. Chapter three will elaborate on the research plan and explain how the data for product recall event sample was collected. This is then followed by an explanation of the used methodology for finding the results in chapter four. In chapter five the characteristics and statistics of the event sample will be discussed. Finally, the results will be presented in chapter six together with a discussion about the impact of a consumer product recall announcement on shareholder value. This thesis will conclude with a conclusion on the results and give some ideas for future research on this topic.
9
2.
Literature Review
In this chapter various academic articles will be discussed concerning product recalls. First a general outline of a product recall will be given. This is then followed by an overview of previous studies done on the effect of a product recall announcement on company value. Next a description will be given of various variables that might influence the reception of a recall announcement by the general public. Finally this chapter will discuss the changes in consumer awareness.
2.1
A Product Recall Announcement
When a company or federal agency officially announce that a product is unsafe for usage or consumption and demand the consumer to stop using the product, is called a product recall. This announcement is usually written as a short formal statement wherein the consumer is informed about safety issues concerning the product. This announcement gives a short and concise description of the reasoning behind the recall. It also contains some general information about the product so the public can identify it and instructions for the consumer on how to proceed with the unsafe product (Lyles et al., 2008) (Beamish and Bapuji, 2008). Recent history shows that that companies recall their products with an increasing regularity (Desai, 2014). As cited by Berman (1999, p69), a study done by A.T. Kearney over 500 consumer good producing companies indicated that almost 25 percent of those companies had to deal with a product recall (Richardson, 1992). However, at present the exact numbers of product recalls are unknown. This is mainly due to the disjunction of the consumer market and governmental regulations. Beside the (inter)national product recalls there are also a numerous amount of product recalls made locally which are hard to measure (Desai, 2014). The study done by Desai (2014, p6) also shows that Gibson tried to estimate the amount of units being recalled in the years 1996 and 1997. Gibson concluded that the number of units that were recalled reached almost 2 billion and almost half a billion in 1996 and 1997 respectively (Gibson, 1998). Because of the tragic consequences faulty products might cause Gibson argues that product recall regulation is of increased importance. Especially since globalization means that more and more stakeholders get involved with a product recall (Gibson, 1995).
10
2.1.1 Consumer Safety Product Commission
Most countries have laws concerning the safety of products and appointed agencies who carry them out. In the United States six federal agencies are concerned with product recalls. In this research paper those six agencies will be briefly addressed in the next chapter. Since this research paper researches the recall announcements of consumer products in the U.S, which are under legislation of the Consumer Safety Product Commission (CSPC), these other agencies are left out for further consideration. As summarized by Etayankara (2009), the CSPC coordinates the consumer product recalls in the U.S. Companies that come to know of a defect or safety risk concerning their product are obliged by law to report to the CSPC within twenty four hours. However, also the consumers themselves have the possibility to report a defect or unsafe product directly to the CSPC. This is done through a form on their homepage or by calling the CSPC directly. Lastly the CSPC keeps surveys among the U.S. population to keep statistics. The results of these surveys are then used to help guide companies, impose specific remedies or fines and for the development of new laws and regulations concerning safety of consumer products (CPSC, 2014). Following the Consumer Product Safety Act (CPSA) of 1972, which essentially is the founding stone of the CSPC, the Consumer Product Safety Act (CPSIA) was accepted in 2008. Despite the CPSA another law was required to give the CPSC the necessary legal authority. Under CPSA alone the agency lacked the required power and was therefore not able to impose product recalls without a company’s cooperation (Felcher, 2001) (Beamish and Bapuji, 2008). Nowadays during a product recall crisis, companies cooperate willingly with the CPSC to recall their faulty product (Ni et al, 2014).
2.2
Previous research on the financial impact of a product recall
With the ever increasing amount of product recalls one can wonder what effect this might have on the companies recalling the product, the investors and consumer confidence. Despite all the different layers of regulation like internal quality control and legal regulations, products still reach the market while being unsafe for use. Which will result in negative publicity, facing possible lawsuits and expensive recall operations. In this paragraph a brief overview will be given of prior research done on the financial impact of product recalls on the recalling company.
11
2.2.1 Product recalls in the automobile industry
The prior researches done on the effect of a product recall announcement focused mainly on two big well defined industries, these being the automobile and the pharmaceutical industry. One of these early studies that concentrated on the effect of a product recall was done by Jarrell and Peltzman (1985). They focused solely on recalls made in the automobile and pharmaceutical branch. The choice for these specific industries was made because of the strict regulations on product quality and safety before release to the consumer market. They found that recalls made in these specific industries had similar effect on shareholder wealth. Namely that the loss of shareholder wealth was a lot greater than the direct costs for recalling the product. This was mainly caused by loss in company goodwill. Another interesting result from their research was that for both the automobile and pharmaceutical industry the financial gain for competitors was overshadowed by the general impact of these recalls on these specific markets. Finally they determined that per product recall the loss for the pharmaceutical industry was far greater than on the automobile industry, 6 percent versus 1,5 percent respectively (Jarrel and Peltzman, 1985, p536). However, they found that recalls in the automobile industry happened almost twice as often and involved less companies compared to the recalls in the pharmaceutical industry. This suggests to certain extent that the market already incorporates product recalls in their valuation (Jarrel and Peltzman, 1985). This result was later supported by the findings of a cross-industry event study by Chu et al. (2005, p47). Their results revealed that out of the three different industry groups, being pharmaceutical, toy / appliances and automobile the pharmaceutical industry suffered most from a recall announcement. This was followed by toy and appliances and with the least effect occurring in the automobile industry (Chu et al., 2005, p47). In response to the research by Jarrel and Peltzman, Hoffer et al. (1988) redid the research with some alterations to the research done by Jarrel and Peltzman. Hoffer et al. altered their research model with some changes and focused his research only on the automobile industry. Hoffer et al. found that no significant evidence remained to support the idea that product recalls in the automobile industry have an effect on share prices (Hoffer et al., 1988). This research was subsequently followed in 1989 by Bromiley and Marcus (1989). They also did research on the effect of abnormal stock return around the production of defect automobiles. Bromiley and Marcus hypothesized that the effect a product recall has on the stock market could function in a way to control companies’ behavior. Their research also indicated that there was not enough evidence to support this. They did find however that the market’s response varied considerably between companies and over time. Suggesting that this might be due to the
12
various stages a company might be in. For example that the market punishes a company more for recalling a product when it is facing bankruptcy (Bromiley and Marcus, 1989, p248). In contradiction to Jarrel and Peltzman (1985) they found evidence supporting that the reaction of the stock market was not greater than the direct cost of a specific product recall would indicate (Bromiley and Marcus, 1989, p248). Yet another study on the effect of a product recall in the automobile industry contradicts both Hoffer and Bromiley and Marcus. This study, done by Barber and Darrough (1996), supports the earlier findings of Jarrel and Peltzman. They conclude their research by saying that it would be beneficial to a companies’ shareholders when product reliability is improved in the automobile industry (Barber and Darrough,1996, p1098).
2.2.2 Product recalls in the pharmaceutical industry
The findings of the discussed studies mentioned above show that there is not an unequivocal answer on the effect of product recalls made in the automobile industry. Compared to the automobile industry, prior research on the effect of product recalls in the pharmaceutical industry prove to be more consistent. Also with their study on pharmaceuticals recalls Jarrel and Peltzman (1985) where one of the first to research the effect on company value, as was already described in the previous paragraph. Even though their findings concerning the automobile industry were answered with contradicting results by others, this was not the case for their research on the pharmaceutical industry. The results found by Dowdell et al. (1992) support the earlier results of Jarrel and Peltzman (1985). Dowdell et al. did a research on a pharmaceutical incident. During their study they also encountered a significant negative stock return after new federal regulations imposed by the Food and Drug Administration (FDA). However, the actual incident that eventually resulted in those new regulations, caused significant negative abnormal returns for the involved company. Dowdell et al. also found that initially there was no negative effect for the recalling company’s competitors. When the new FDA packaging regulations were accepted however, also the competitors suffered negative effects. These effects were roughly the same between all direct competitors and the original company. Furthermore the research by Dowdell et al. supports Jarrel and Peltzman finding and with their argument that the market punishes a company far beyond the direct costs involving the product recall (Dowdell, 1992, p298-299).
Another study supporting the findings in the pharmaceutical industry as mentioned above is the study done by Dranove and Olsen (1994). In their study Dranove and Olsen tried to explain the spillover effect between pharmaceutical companies during a product recall. Even
13
though they found enough evidence that the recalling company suffered significant loss on the stock market, spillover effects only occurred when the market anticipated on possible regulatory changes due to that specific recall which corresponds with the findings of Dowdell et.al.
2.2.3 Product recalls for consumer products
With the results from the automobile industry being inconclusive and the pharmaceutical industry being consistent, some research was done on the effect of a product recall involving consumer products. In their study Chen et al. (2009, p219) found not enough significant evidence to support the existence of negative abnormal returns around a product recall announcement. However, the aim of their study was not per se to find a negative effect in general but rather how companies should deal with a product recall, pro-active or passive. They did find significant evidence indicating that companies which use a pro-active product recall strategy. These companies were punished more than companies which used a passive strategy. Another interesting result was that companies with less reputation would choose more often to opt for a pro-active approach. The explanation that was given by Chen et al. was that the results suggest that during a recall crisis companies are more concerned by the response of the financial market rather than their social responsibility (Chen et al., 2009, p224-225). A more recent study by Ni et al. (2014) also found no significant evidence indicating that a consumer product recall announcement resulted in negative abnormal returns (Ni et al., 2014, p316-317). But also for them applied that the research aim was more specific. Namely, to research if there was a difference in market reaction when a company recalled one of their own brands or a national brand. To be able to measure this Ni et al. only studied product recalls made by the top ten U.S. retailers. The results supported their hypothesis that there would be a more significant loss in shareholder value when a company recalled one of their own labels instead of a national brand (Ni et al., 2014, p316-317). This conclusion does not necessarily contradict the earlier findings of Chen et al. Since the effect due to a product recall with a national brand would lie primarily with the manufacturing company and not the retailer. Which in turn can choose its own recall strategy.
Ni et al. (2014) also studied the effect of the offered remediation strategy and the level of injury on the market’s reaction. There proved to be enough evidence to assume that the most severe injuries had a significant negative effect on the shareholders’ value. Their findings also supported that a refund remediation had a more negative impact than a replacement.
14
2.3
Incidents, Injuries and Remedies
In this section some light will be shed on the role of incidents and injuries that precede a product recall and their role in the event of a product recall. In the second part product recall remedies will be reviewed.
2.3.1 Media Attention
In the study done by Ni et al (2014, p313-314) the role and extent of media attention is explained as a major factor for the influence incidents and injuries have on the market’s response to recall announcements. They come to the conclusion that loss aversion coupled with heightened media attention, caused by severe safety issues, go hand in hand. In an article by Barber and Odean (2006), as referred to by Ni et al., was found that the newsworthiness to report a product recall depends on the severity of prior incidents leading to the recall. Which in turn only strengthens the effect of loss aversion and resulting in a more negative market reaction (Ni et al., 2014). However, most recalls do not occur due to prior incidents and injuries. They are far more often being recalled because of the product not conforming to safety standards and regulations and therefore possessing a potential risk for the consumer (Hora et al. 2011)(Berman,1999). In his article, where Berman warns manufacturers for inevitable precautious recalls, he wrote: “Unfortunately, too often test marketing activities are more involved with assessing a product’s sales performance than with safety considerations” Berman (1999, p72).
2.3.2 Recall Remedies
The process on how a company handles the event of a product recall can be of considerable value. The speed at which the company operates during the product recall and the kind of remedy that is being offered, is of determining value to the consumers’ satisfaction (Berman,1999) (Bapuji, 2011). In general there are three different types of remedies that are being offered to consumers who own an unsafe or defect product that is being recalled. The three remedies are: a refund of the purchase price, a replacement product or a repair. In their study, Davidson and Worrell (1992) hypothesized and found evidence to conclude that in the event when a full refund or a replacement was offered the abnormal returns proved to be more negative than when it was solely a repair or checkup. However, there is also the psychological aspect of the remedy being offered. Ni et al. therefore hypothesized, supported by the findings
15
of Van Heerde et al. (2007) and Siomkos and Kurzbad (1994), that the negative experience with the recalled product can leave its mark on consumers. For instance, when it gets repaired, consumers might argue that the company chose a cheap solution to deal with the defect product. Which in turn can leave the consumer wondering if the defect was really dealt with or that it was merely a patch-up. Therefore a refund seems to be the most effective remedy to offer to preserve consumer confidence (Ni et al., 2014).
2.4
Consumer Awareness
In the previous paragraph the topic of media attention was already touched upon. As was already referred to, the article by Barber and Odean (2006) showed that media attention is invoked by the severity of incidents concerning and leading up to a product recall. However, with increased accessibility and availability of the internet media attention may also result from communities on the internet being heard. The recent “ALS ice bucket challenge”, a typical internet meme, that spread on the internet and reached millions of people before the media actively reported about it is a good example. It also shows that nowadays companies, or institutions in the case of the ALS-challenge, do not have to depend on conventional media to get media attention. Another example can be internet communities consisting of fans that are being created to support a specific brand or product. For instance the website Appleinsider, a website that reports on anything about the Apple brand and hosts a forum for fans to discuss Apple products. A last example would be a large, if not the largest, social network site of the world: Facebook. Here consumers can gather and find each other easily by joining certain “Facebook communities” to exchange their experiences and discuss countless different topics, including brands and consumer products.
In 1981 Mowen et al. (1981) found that the consumers view and feel for a company or its products after a product recall greatly depends on, to cite Mowen et al.;” the knowledge that a recall had been made, the perceived danger of the defective product, the perceived corporate responsibility of the company, the knowledge of recalls by other companies, and the perceived responsibility of the company for the defect.” Mowen et al. (1981, p405). Later was also found that the involvement a company has in social responsibility significantly explained the judgment and perception of the consumers after a product recall (Matos and Rossi, 2007). After the start of the internet revolution in 2000 (Hilbert and Lopez, 2011, p63) it became clear that the internet was becoming a big and powerful tool for consumers. In their study
16
Kucuk and Krishnamurthy (2007) therefore conclude that the power of consumers has increased significantly since the arrival of the internet. The new and various ways consumers have access to information and are able to share it can change markets (Kucuk and Krishnamurthy, 2007, p55). Because of these findings, in conjunction with shorter lines of communication, consumers are better informed. Which in turn can reflect on the effect of the market’s response on a product recall announcement.
3.
Research Method
Before it is possible to test the research hypotheses it is necessary to collect a large enough sample of companies who recalled one of their products during the period 1988 till 2014. To make further distinctions possible and to be able to explain possible outcomes additional information has to be collected. In the following paragraphs a detailed description of this process will be given.
3.1
Product Recall Event Sample
A product recall is a somewhat unusual event. Every week somewhere around the globe one or more product recalls are made. The recalled products vary immensely and cover all kinds of different retail branches, reaching from food to automobiles to children’s toys. Not only the recalled products vary but also the economic environment wherein the recall happens. Another issue associated with different economies is that they are most likely not regulated in the same way. Where in one economy someone has to suffer a major injury before a recall is announced, in another economy not meeting standardized safety requirements is already enough reason for a precautious product recall.
To overcome the aforementioned problems this study is limited to, product recalls made in, the United States market. The U.S. market is a highly regulated market and is extensively covered by all well-known financial databases. Limiting to one economy makes financial data more easily comparable and makes sure that the analyzed product recalls are made under the same regulations and are based on similar quality standards.
In the U.S. six federal agencies with each a different jurisdiction regulate the safety of products on the American market. The six agencies are: The U.S. Consumer Product Safety
17
Commission, The National Highway Traffic Safety Administration, U.S. Coast Guard, Foodsafety.gov (a combination of another group of federal agencies with the Food Safety and Inspection Service being the first mentioned), Food and Drug Administration and The Environmental Protection Agency (Recall.gov, 2014).
For the selection of this event sample a further limitation is made by limiting the product recalls to recalls announced by the Consumer Product Safety Commission (CPSC). The CPSC is a U.S. federal agency assigned with the task of protecting the American public from the numerous different types of consumer products which may cause unjustified risk of injuries or even death. The CPSC has jurisdiction over more than fifteen thousand different consumer products including: appliances, furniture, lightning, clothing, household products, outdoor, electronics, toys and sports (CPSC, 2014).
Whereas the other government agencies involved in product recalls only target a very specific product group, for example automobiles or cosmetics, the CPSC covers a broad range of different consumer products across industries and product groups. Therefore the database of this federal agency is chosen for selecting the samples of this event study.
Over more than 40 years the CPSC keeps track of and officially announces product recalls concerning the U.S. market. Their database of past product recalls is easily accessible and very extensive. Today their database exceeds more than six thousand product recalls.
For the construction of this event sample only product recalls announced within a certain period are needed. Therefore only recalls made between 1987 and 2014 are considered. This period is chosen because of the main hypothesis, were increased consumer awareness has a significant effect on the market reaction after a product recall. The events found in this time period are divided into two groups with the year 2000 as being it center.
However this first limitation still results in over five thousand eligible product recalls. To make the selection more feasible within the constraints of this research paper, roughly one eighth is picked randomly for further analyzing. This is done by placing all five thousand product recalls in consecutive order followed by selecting one sample every other eight. It could happen that the selected sample is not an actual product recall but a general announcement made by the CPSC. When this happens the selected sample is discarded and the selection process is continued as previously stated. After this initial selection process 704 samples remain.
However, the selection process is not done. To be able to answer the research question the event sample requires the announcing companies to be listed on the U.S. stock market during the event window. Not all product recalls are necessarily done by listed companies and
18
therefore the selected samples have to be checked for the possibility that the recall announcement is done by a private company. In the event that the recalling company is a subsidiary of a public company, the listed parent company is used for further analysis. In the situation where the company is a public company but not on the U.S. market it also doesn’t qualify and has to be left out. These criteria result in an additional 450 dropouts due to the recall being done by private companies and another 44 due the company not being listed in the U.S. In 11 cases the data was not clear or missing essential information about the product recall. These were also dropped, leaving 199 product recall events in the sample.
The last step in selecting the final event sample is finding out if there are any overlapping events. There might arise some data issues and biases in the estimations when event windows overlap. For this reason all recall announcements done by the same company within a year of each other are excluded from the sample. This results in a last elimination of 43 events. The final product recall sample totals 156 events which will be used in this research paper.
3.2
Additional Data
With the final event sample known a subset of additional information is required for each individual product recall. In this paragraph the reason and thought behind the necessity of this data will be discussed and the source of the data will be given. The additional collected data that will be discussed are: share price return, market index, estimated retail value, market value, incidents and injuries, remedies and country of manufacture.
3.2.1 Share Price Return
To be able to calculate the returns during the event window, share price data is needed for all the companies in the event sample. For all the companies in the event sample the Return Index (RIt) data was collected on a daily basis for a period of 321 trading days prior to the
official recall announcement. This data was retrieved from Thomson Reuters Financial’s Datastream database.
3.2.2 Market Index
For this research abnormal returns will be estimated by using the OLS market model as will be elaborated in the next chapter. However, in order to use this model correctly also market return (Rmt) data is needed. Since the event sample covers a broad scala of different type of
19
products and standard market indexes like the NASDAQ, S&P500 or NYSE might be biased towards certain industries or company sizes, the market index as described by Fama and French (Fama and French, 1993) is chosen as a market index estimator for this research. The market return derived from this model is a better indicator since it is constructed from the value-weight return of specified companies listed on NASDAQ, NYSE and AMEX. On the official website of K.R. French a data file is being updated on a regular basis which contains the market return for the U.S. (French, 2014).
3.2.3 Estimated Retail Value
The constructed event sample limits the product recalls to a certain branch of consumer products as is described earlier in this chapter. However this does not mean that all recalled product are of the same size, have the same value or are produced in the same quantity. To overcome this issue and be able to compare or group these event samples in a later stage, an estimation is made of the total retail value of the recalled product. With every recall announcement the CSPC also publishes some extra details about the product. Among these details they give the exact retail price or a price range for which the recalled product has been sold. The CSPC also provides an approximation of the recalled volume. The estimated retail value is calculated by multiplying both factors with each other. In those cases where CSPS lists a price range an average of the price range was used in order to calculate the retail value.
3.2.4 Market Value
Not only the recalled products vary in size, price or form but also the companies recalling the products do. Meaning that for one company a specific value of a recall is hardly noticeable whereas another might go bankrupt if they have to issue a recall with that same value. For this reason also data is collected of the company market value which will be used in conjunction with the estimated retail value. The market value for all the companies in the event were found through the usage of the Thomson Reuters Financial’s Datastream database. For each company the market value of the first trading day in the month in which the recall occurred was selected.
3.2.5 Incidents and Injuries
Beside the aforementioned details the CSPC also informs the general public with some other interesting information about the recalled products. One of these other details describes if the recalled product is being recalled out of precaution or due to product failure, injuries caused
20
by product failure or even death. In order to be able to test if incidents and injuries have a significant effect on the behavior of stock prices this data is included in the event sample for every event.
3.2.6 Remedy
The CSPC also keeps track of the remedy offered by the company whose product is being recalled. In general those companies choose between four recurring solutions. These solutions are: a refund of the purchase price, a replacement by a similar product, a reparation or adjustment of the failing product and lastly the company sometimes just asks to discard the complete product without any form of compensation. Since the selection of a specific remedy might reveal some interesting insight in the market behavior, this data is also selected for the event sample. In some cases the recalling company offered more than one solution to the customer. When this occurred a refund was chosen over a replacement, a replacement over a repair and a repair over a discard.
3.2.7 Country of Manufacture
The last variable that will be included in the event sample is the country of manufacture, also provided in the dataset of the CSPC. However the countries will be grouped into two groups to make this a more manageable variable. One group for advanced economies as is categorized by the International Monetary Fund (IMF, 2014) and one group for those countries which are not categorized as such. The inclusion of these two groups is done mainly because of the assumption that the market might expect that products manufactured in non advanced markets have a higher tendency to fail than products made in advanced markets.
4.
Methodology
In this part the used methodology will be discussed. The commonly used methodology for testing the markets response to recall announcements is the event study methodology (e.g. Dowell et al. (1992), Davidson and Worrell (1992), Govindaraj et al. (2004). The purpose of an event study is to test the impact of a particular event which is assumed to be unknown to the market prior to its announcement. Therefore the event and in this research, a product recall, can be viewed as new information to which the market will immediately react. This
21
also explains why event studies are a common practice in finance as pointed out by Kothari and Warner (2004). Hence, the event study methodology will be used to measure the short term impact of a recall announcement on the share price return of the recalling company. The event date (t=0) will be the date of the official announcement as published on the website of the CSPC. The day before the announcement will therefore be: t=-1. Unless otherwise stated the event study methodology and literature of Brown and Warner (1980, 1985), Dodd and Warner, 1983 and Kothari and Warner (2004) will be used in this chapter.
4.1
Event Window
In order to perform the calculations necessary within the event study methodology two event windows are required. The first window is an estimation window. This estimation window is used to calculate the predictors of the ordinary least squares (OLS) market model. For this research an estimation window of 321 trading days till 60 trading days before the event date (t=0) is used, representing a full year of 260 trading days stopping approximately three months before the official recall announcement. The three month period prior to the event date is chosen to make sure that there is no negative bias on the estimation of the parameters. For example, incidents that eventually lead to a product recall might have an influence on the company’s share price in the period before the event. The estimators alpha αi and beta βi,
which are derived from this time window, will be discussed in the next paragraph.
The second window is the actual event window. It is over this window that the abnormal returns will be estimated. However, using a broad event window, or horizon, greatly reduces the power of the test and therefore a short horizon is recommended when an exact event date is known (Kothari and Warner, 2004). Thus for this research a short 2-day horizon was taken, [0,1]. This is consistent with the method of other similar researches done by Ni. et al (2014), Chen et al. (2009) and Zhao et al. (2013).
4.2
Abnormal Returns
Given the collected data for this event study some additional calculations are required in order to use the formulas in a good manner.
22
4.2.1 Daily Return Data
The collected financial data for every event in the sample was the Return Index (RIt), as was
already specified in the previous chapter. However, in order to proceed it is necessary to turn the daily (RIt) into (Rt). This can be done by using the following formula:
=
( )
As shown, (Rt) is calculated by calculating the percentage change relative to t-1. After this
required computation the next step could be taken.
The data used to calculate market return (Rmt) was found by using the published market
return data published by K.R. French (2014), which was already addressed to in the previous chapter.
4.2.2 Ordinary Least Squares Market Model
In this research the expected returns will be predicted based on the OLS market model. As already briefly hinted in the previous paragraph the OLS model uses past security (i) performance relative to the market (m) performance as a predictor for future returns. The formula that follows is to calculate the expected return:
= + +
In this formula Ritis the expected market return composed of Kitand eit. Kit would then be the
normal or expected market return and eitthe abnormal or unexpected return. Therefore is also
true:
= + this is followed by a rewritten form: = +
This rewritten formula estimates Kit,the expected market return of the security at time t. Here Rmis the return of the market and alpha (αi) and beta (
β
i) are estimators where alpha is theintercept and beta the slope of the OLS regression between Ri and Rm over a specified event
window as was already explained in the previous paragraph.
The formulas for calculating beta (
β
i) and alpha (αi) are respectively:=(,)
23
As the formulas show and within the specified event window beta is found by dividing the covariance between the return of the security Ri and the return of the market Rm by the
variance of the market returns. The means of the security
x
i and the marketx
m are alsocalculated over the same event window.
Finally all variables for computing the abnormal return for the security i at time t are known. The abnormal returns are then calculated by using the following formula:
= − ( + )
4.2.3 Measuring Abnormal Returns
In order to proceed and to be able to tell if the measured abnormal returns are of significant value within the event window, the outcome of each individual security has to be summed and an average has to be taken. This commonly used method for measuring abnormal returns over a short horizon [t1,t2], in this study [0,1], is the Cumulative Abnormal Return (CAR) method.
The formula is written as:
( !, ") = ∑$
%
This formula gives us the cumulative abnormal return for each individual security i. However, in this research we want to prove if a group of securities shows a significant reaction after the event. Therefore we have to compute the CAAR, the Cumulative Average Abnormal Return, and is written as:
( !, ") = !
&∑ ( &
%! !, ")
The acquired result will then be tested for significance.
4.3
Ordinary Least Squares Regression
To determine the impact of a product recall announcement on the return of the recalling company a linear multivariate Ordinary Least Squares (OLS) regression will be executed. With this multivariate OLS regression the relationship of other variables in the event of a product recall will be determined. In this regression the cumulative abnormal return for security i will be used as a dependent variable. For the independent variables multiple dummies will be used; Group, Relative Recall Value (RRV), Incident, Injuries, Remedy and
24
Manufactured (Made). The primary independent variables will be Group and RRV the other mentioned independent variables will be controls. Group will distinguish the event into two separate groups for a specified time period. RRV will tell if the size in value of the recall relative to the company’s market value has an effect on the dependent variable. The regression model will therefore become:
= '+ !()*+, + "- + ./01230 + 4/05+)26 + 7839 + :;<3 + =
Where:
CARi is the dependent variable for the cumulative abnormal return for each company i Groupi is the indication if the underlying company i belongs to group A (1) or Group
B (0). A company belongs to group A when the product recall occurred in the period of the year 2001 to and including 2013, and group B when the recall occurred in the period of year 1987 until and excluding 2001.
RRVi is an indicator for the relative recall value. For this research all events were split
into two groups by using the following formula with ERVi being the Estimated Retail
Value and MVi being the market value of company i at time t=0:
-- =>
? ∗ 100
A company was classified for either group “High”(1) when the RVVi was higher than
0,1% or for group “Low” (0) when the result was lower than 0,1%.
Incidenti is a dummy variable that is 1 when an incident occurred with the recalled
product or 0 when the recall was announced out of precaution by company i.
Injuriesiis a dummy variable that is 1 when an injury occurred during the incident or 0
when it did not.
Remedyi is a set of four dummy variables:
1) That is 1 if Company i offered a refund, 0 if not. 2) That is 1 if Company i offered a replacement, 0 if not. 3) That is 1 if Company i offered a repair, 0 if not.
4) That is 1 if Company i instructed to discard the failing product, 0 if not.
Madei is a dummy variable that is 1 when the recalled product was made in an
advanced market and 0 when it was manufactured in an non advanced market.
The parameters ß0 and ε are respectively the constant term of the regression and the
25
In order to execute the regression STATA will be used. The regression will be performed “robust” so that the results will be corrected for heteroskedasticity. The t-test results by STATA are two-sided. However, this research tests the negative impact of various variables. Therefore, the two-sided p-values are converted into left-sided p-values. The formula is written as:
[
two-sided p-value
] = 2 ∗ [*0-sided p-value
]5.
Descriptive Statistics
This chapter will give a brief overview of the event sample characteristics. The tables and appendixes which are revered to in this chapter can be found at the back of this research paper.
5.1
Event Sample Characteristics
As already has been mentioned in a previous chapter, the event sample consists of 156 events. This was the result of a thorough shake down from initially over five thousand recall events that occurred during the timeframe of 1987 till 2014. Appendix 1 gives a full overview of all the product recalls that were used in the event sample, sorted in alphabetical order. To get a better understanding on how a product recall announced by the CPSC looks like an example is provided and can be found in Appendix 2. The given example gives a clear view on how the relevant information concerning a product recall is presented to the public. A summarize of all the variables that were used in this event study can be found in table 1A. This table also gives a quick overview of the means, standard deviations and correlations of the variables.
5.1.1 Event Sample Categorization
The selected events samples were sorted into two different event groups, group A and group B. The categorization was based on the event date. Product recalls made in the period 2001 till 2014, after the start of the “internet revolution” as mentioned by (Hilbert and Lopez, 2011, p63), where placed in group A. The events that occurred before 2001 where placed in group B. Table 1B shows the event sample distribution arranged by year. From this table follows that the categorization of the events resulted in 72 percent that fell into group A and 28% into
26
group B. These numbers also show that there is an increasing number of reported product recalls over the years. Coupled with the random selection method used for compiling the event sample this implies that recall announcements happen more frequently. Which in turn is consistent with the observations made by Desai (2014).
5.1.2 Event Sample Value Distribution
In order to estimate if the size in retail value of the recalled product had any effect on the valuation of the security, every sample in the event was given a “high” or “low” label. This label depended on a straightforward calculation where the relative size of the product recall was compared to the size of the company. In this research paper this is called: Relative Retail Value (RVV). Table 1C shows the distribution and composition between groups A and B of the event sample. Shown are four smaller tables, the Estimated Retail Value (ERV) of the recalled products, the Market Value (MV) of the companies recalling the products, Relative Retail Value (RVV) and RVV Classification. At first glance it seems that the values in group A are far larger than group B. Keep in mind that these numbers are not corrected for inflation and therefore partially explain why the means for group B are lower than group A. The mean of the complete sample however, proves to fall somewhere in the middle of group A and B for
ERV and MV. In order to make a comparison between all product recalls possible RVV is calculated. Table 1C “Relative retail Value” therefore shows an interesting result. For both group A and B the RVV value appears to be around 0,27 percent. These numbers are calculated by dividing the mean of ERV by the mean of MV, e.g. Group A ERV $38 million divided by Group A MV $14.175 million. This result ensures that RVV is comparable between the different product recalls regardless of the group classification the event sample has. The “RVV Classification” table shows that the events themselves are also categorized evenly, with 47% of the vents belonging to “High” and 53% belonging to “Low”.
On final remark on table 1C is that the standard deviation for category “High” seems a bit out of place. This is caused by one single event with a extreme RVV value of 2,2. This indicates that the retail value relating to this event was more than twice the size of the market value of the recalling company. If this event were to be removed from the whole sample the standard deviation would drop towards a more reasonable 0,09 and RVV MAX would drop to 0,67. Since this outlier doesn’t have a direct impact on the classification criteria it is kept in the event sample.
27
5.1.3 Event Sample Remainder Variables
The composition and distribution of the last variables in this event sample, “Made”, “Incidents and Injuries” and “Recall Remedies” are presented in tables 1D, 1E and 1F respectively.
As can be seen in table 1D the categories “Advanced” and “Non Advanced” economies consist primarily of two countries. These countries are the United States and China. This table also shows that for almost a third of the product recall announcements no origin of manufacture was mentioned in the actual announcement. In order to label the acquired events as much as possible for this variable, it was chosen to group the samples in a category for “Advanced” and “Non Advanced” economy instead of United States and China. However, the formulated hypothesis wants to test the effect of the stock market’s reaction when the recalled product was made in an advanced economy. Therefore, all the events without a mention of country of origin were also grouped under “Non Advanced”.
The data that was collected concerning incidents and injuries is presented in table 1E. The percentage of incidents and of injuries is given for group A and B. For this variable it is also true that the ratio is evenly distributed between A and B. Both groups contain roughly 65 percent events wherein an incident occurred. The amount of injuries resulting from those incidents seems to be around 50% which is also consistent between group A and B. However, this indicates a high correlation between incidents and injuries. This can also be noticed and confirms the results shown in table 1A. The existence of this high correlation makes sense since injuries always happen from incidents. This will be considered when the results are being discussed.
Finally, table 1F presents the distribution of the remedies that were being offered to consumers whose product had been recalled. The table shows that an actual “remedy” was offered in almost all product recalls. In only two events the recalling company asked consumers to discard the unsafe product. The three remedies are spread almost evenly among themselves and among the groups A and B. A product replacement was the most offered with 34,62 percent, followed by a refund with 30,13 percent and a repair with 33,97 percent.
28
6.
Results
This chapter will discuss the outcomes of the used models and regression which were used to test whether a consumer product recall announcement led to a more significant negative abnormal stock return. The first part of this chapter will start off with the interpretation of the results that relate to the event and the group tests. In the second part of this chapter the results and impact of the various variables will be discussed.
6.1
Event comparison group A and B
Table 2 shows the daily average abnormal returns around the consumer product recall announcement date, beginning ten days prior and ending ten days after the announcement. The average abnormal returns show a significant negative effect for the whole event sample and for group A (post 2001) around the announcement date. For group B (pre 2001) no significant negative average abnormal return is measured. The results for the cumulative average abnormal returns are summarized in Table 3A and Table 3B. The tables show the
CAAR for the whole event, including all events, for group A, with recalls made in the period 2001-2013, and group B, with recalls made in the period 1987-2000. From Tables 3A and 3B follows that a recall announcement for the whole event is related with a significant negative abnormal stock return of -0.67%. Moreover, for group A the market’s reaction is associated with an even stronger significant negative effect of -0.95%. However, this is in contrast with group B where no significant negative effect is measured with an average stock return of 0.04%. Therefore, a consumer product recall announcement is associated with a negative abnormal stock return. To measure if there is a significant difference between group A and B a two sample t-test was performed. The result of this test can be seen in Table 3C. This table shows the comparison between the CAAR of group A and B. The abnormal stock return for group A is, with a difference of -0.99%, with a p-value of less than 0.1 significantly more negative than the abnormal stock return for group B. Since these results provide an indication that the abnormal returns for group A are significantly more negative than group B an ordinary least squares regression is performed on the whole event. The results of the OLS-regression can be found in Table 4. The results shown in Table 4 are corrected in STATA for heteroskedasticity and are robust. The p-values presented in table 4 are the corrected left-sided p-values. The first regression, OLS regression 1, is the regression for the whole sample
29
period. In this regression it was tested if the variable Group contributed significantly to the abnormal stock return. Table 4 shows that Group not only contributes negatively but also shows that with a coefficient of -0.0103 it contributes with a p-value of less than 0.1 significantly negative to the prediction of the abnormal stock return. Therefore, these results provide consistent evidence that when a consumer product recall was announced during the years 2001-2013, Group A, the abnormal stock return were significantly more negative.
6.2
The first group of hypotheses
In the first chapter two groups of hypotheses were given. The first group of hypotheses were formulated to answer the first research question: “Is there a significant difference in the market’s response to a consumer product recall announcement after the start of the internet revolution in the year 2000?”. The hypotheses that were formulated around this research question were:
Hypothesis 1(0) Consumer product recall announcements will cause no significant effect in abnormal stock returns.
Hypothesis 1(1) Consumer product recall announcements will cause a significant effect in abnormal stock returns.
Hypothesis 2(0) Consumer Product recall announcements will cause no significant effect in abnormal stock returns from 2001 to and including the year 2013.
Hypothesis 2(1): Consumer Product recall announcements will cause a significant effect in abnormal stock returns from 2001 to and including the year 2013.
The results from the previous paragraphs indicate that for the whole event, thus all the events in the sample, there was evidence found that a consumer product recall announcement is associated with a significant negative abnormal stock return. Therefore the null hypothesis of hypothesis 1 is rejected in favor of the alternative hypothesis. Since recall announcements made in group A (post 2001) are associated with a significant negative abnormal stock return of approximately -1% and significantly differs to group B (pre 2001), the alternative hypothesis of hypothesis 2 is supported and the null hypothesis is rejected.
