Anda di halaman 1dari 121

The Economics of Food Safety Communication

Dissertation

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

By Ji Li, M.A. Graduate Program in Agricultural, Environmental, and Development Economics

The Ohio State University 2009

Dissertation Committee: Neal H. Hooker, Advisor Thomas Sporleder Matthew Roberts

Copyright by Ji Li 2009

Abstract

This dissertation is composed of three essays about timely food safety issues in the food and agribusiness marketing. In Chapter 1, responding to increasing customer attention to food attributes agribusinesses are employing novel product differentiation strategies. As an example, the use of food safety claims on new packaged food products are investigated from the food manufacturers perspective. First, using two product innovation databases, we investigate claim use on labels in seven English speaking countries over the period 1980 to 2008. Then, based on manufacturer recommended selling prices and using US data (from 2002 to 2008) we apply parametric and nonparametric hedonic methods to identify supply-side (agribusiness) valuations of chemical and microbiological claims in two food categories. We identify a significant 5 cent premium per ounce for a preservative free claim in spoonable yogurts. We do not find a statistically significant impact for E. coli free messages on meat and poultry products but find a significant price premium ($0.193 and $0.257 per ounce in the two models) for antibiotic free claims in this category. Chapter 2 explores consumer reaction and attempt to personalized food recall notifications. Foodborne illness outbreaks can pose serious threats to consumers. To protect public health, when food companies discover that a food product has been ii

implicated in an outbreak or is found to contain unacceptable levels of contaminants, they usually voluntarily issue a recall notice to recover any unconsumed products thought to pose a risk. While recovering unsold products from retailers is relatively straightforward, reaching consumers has proven more difficult. Through their shopper loyalty card programs, many retailers have the ability to identify and communicate recall information to customers who have purchased foods that become subject to a recall notice. An analysis of nationally representative survey data collected by telephone from 1,101 American adults in August and September 2008 suggests that most consumers (70%) are interested in such a service. Almost fourteen percent would be willing to pay for such a service (mean willingness to pay $67). The results of an ordered logit model suggest that consumers who are younger, who are caregivers for children, have had experience with foodborne illness, share shopping responsibilities with someone else in their household and those who are unemployed appear most interested, and willing to pay, for such a personalized recall notification service. Chapter 3 is centered on pet food recalls. In 2007 the pet food industry encountered a series of recall shocks which attracted considerable attention from the public, media, regulatory bodies and most likely shareholders. Exploring a financial market perspective, this study investigates the influence of recall announcements on three publicly traded pet food companies: Procter & Gamble, Co., Del Monte Foods, Co., and Colgate-Palmolive, Co. An event study, Generalized Autoregressive Conditional Heteroscedasticity (GARCH) and Seemingly Unrelated Regression system (SUR) iii

estimations are conducted and compared using daily and intra-day stock prices to explore recall effects. Findings suggest that the announcement of recalls had a significant effect on Del Montes stock price but no influence on P&G or Colgate. Firms response time and market share appear to be key factors influencing stock price reactions.

iv

Dedicated to my parents and my husband

Acknowledgments I would like to express my sincere gratitude to my advisor, Professor Neal H. Hooker. Throughout my study at the AEDE department, Professor Hooker has given me constant guidance, invaluable advice and warm encouragement. It is his tremendous support that makes this thesis possible. He has devoted great support to my whole Ph.D. study and career development. His admirable personality and professional ability are of immerse benefit to my study and my life in the future. I also want to thank Professor Tom Sporleder and Professor Matthew Roberts. Not only this dissertation benefits greatly from their great help, but also they have offered me a great deal of time and efforts to my career development. I do appreciate their valuable suggestions and comments on this dissertation and on my career development. I would like to express my sincere thanks to Professor Alan Randall, Professor Abdoul Sam, Professor Brian Roe, Professor Stan Thompson, Professor Stan Ernst, Professor Lynn Forster, Professor Carl Zularf, etc. Thank you so much for your time and help. My special thanks go to my parents and husband. Without you, none of my achievements would have been possible. Your trustful and unlimited love has supported me in every step of my life and will always be my light in my life. Thank you all very much.

vi

Vita

2001................................................................B.A. Marketing, Jilin University 2005................................................................M.S. Regional Economics, Zhejiang University of Sciences 2006................................................................M.A. Economics, The Ohio State University

Fields of Study

Major Field: Agricultural, Environmental, and Development Economics

vii

Table of Contents

Abstractii Dedicationv Acknowledgments..vi Vitavii List of Tables...x List of Figuresxii

Chapters: 1. Documenting Food Safety Claims and their Influence on Product Prices...1 1.1 Introduction1 1.2 A Background on Food Labels in Different Countries. 3 1.3 Overall Use of Food Safety Claims.. 6 1.4 The Hedonic Price Model15 1.4.1 Data17 1.4.2 Empirical Estimation and Results..18 1.5 A Hedonic Price Model for E. coli Free..21 1.6 Nonparametric Models 24 1.7 Conclusions..27 1.8 References30

2. Consumer Interest and Engagement Decision in Personalized Food Recall Notification Services..36

viii

2.1 Introduction..36 2.2 Consumer Interest Analysis in Personalized Recall Notification Service..........................................................................................................40 2.2.1 Data40 2.2.2 Methodology..43 2.3 Results and Discussions.. 46 2.4 Conclusions..50 2.5 Consumer Engagement Decision Analysis of Personalized Recall Notification Service..52 2.5.1 Methodology..52 2.5.2 Variables54 2.6 Results and Discussions.. 55 2.6.1 The First Stage Probit Regression Analysis.. 55 2.6.2 The Second Stage Truncated Regression Model58 2.7 Comparison between Derived and Direct Participation Interest60 2.8 Conclusion...62 2.9 Reference.65

3. Stock Price Reactions to Pet Food Recalls... 67 3.1 Introduction..67 3.2 Event Study Methodology...72 3.2.1 Three Publicly Traded Firms.72 3.2.1.1 Del Monte Foods, Co.72 3.2.1.2 Procter & Gamble Co.73 3.2.1.3 Colgate-Palmolive Company.73 3.2.2 Measuring Normal and Abnormal Returns74 3.3 Event Study Results.77 3.4 Publicly Traded Firms not Involved in Pet Food Recalls84 3.5 Nestle Purina PetCare Company Analysis..85 3.6 Discussions and Implications of Event Study.86 3.7 Modeling Intra-day Stock Price Responses.87 3.8 Generalized Autoregressive conditional Heteroscedasticity Method..89 3.9 Seemingly Unrelated Regression System Estimation..91 3.10 Model Comparisons...94 3.11 References..96

Bibliography.99

ix

List of Tables

Table

Page

Table 1.1 Summary of Food Safety Claims 11 Table 1.2 Distribution of Food Safety Claims across Product Categories ...14 Table 1.3 Descriptive statistics of the GNPD data (Spoonable Yoghurt) 19 Table 1.4 Parametric Hedonic Price Model: GNPD Data 20 Table 1.5 Descriptive Statistics of PLA Data (Meat and Poultry Products) 22 Table 1.6 Parametric Hedonic Price Model: PLA Data ...24 Table 1.7 Nonparametric Hedonic Model: GNPD Data.. 25 Table 1.8 Nonparametric Hedonic Model: PLA Data...26 Table 2.1 Summary Statistics for the Explanatory Variables across Dependent Variable Categories..42 Table 2.2 Descriptive Statistics of Survey Data....45 Table 2.3 Estimation Results - Consumer Interest....47 Table 2.4 Predicted Probabilities of Explanatory Variables on Consumers Interest in Notification Service...48 Table 2.5 Descriptive Statistics of Recall Communication Survey Data..57 Table 2.6 Estimation results of 1st stage...58

Table 2.7 Estimation results of 2nd stage..59 Table 2.8 Estimation results of consumer participation interest...61 Table 3.1 Estimated Parameters of Normal Return from the Market Model75 Table 3.2 CAR from Pet Food Recalls for Different Windows78 Table 3.3 Abnormal Returns on the Eleven Days Centering on the Event Announcement Day...79 Table 3.4 CAV from Pet Food Recalls for Different Windows82 Table 3.5 Summary Statistics of Stock Prices...88 Table 3.6 Parameter estimates of GARCH (1, 1) models.90 Table 3.7 Estimation Results of Seemingly Unrelated Regression...92

xi

List of Figures

Figure

Page

Figure 3.1 Pet Food Recall Media Articles from 1998 to 2008.. 69 Figure 3.2 CAR from Pet Food Recalls for Different Windows 78 Figure 3.3 CAV from Pet Food Recalls for Different Windows. 83

xii

Chapter 1 Documenting Food Safety Claims and their Influence on Product Prices

1.1 Introduction Label messages are playing rich and multiple roles in food marketing particularly in light of consumers increasing efforts to enhance health and well being through product choices. Traditionally, food labels provide information required by government agencies such as the US Food and Drug Administration (FDA). This information is intended to provide consumers product selection cues such as nutritional content, serving size and ingredients. Recently, certain food safety claims have emerged on food labels. When effective such label claims transfer salient information from farmers, ranchers, manufacturers, retailers and food service operations to consumers about risk reductions derived from production and processing techniques employed through the supply chain. An effective message is defined as one that provides information sufficient to change consumer behavior. Frequently voluntary, such labeling initiatives may precede government regulation or standardization of claim messages, audit or certification procedures. For example, food labels containing claims such as antibiotic, E. coli and pesticide free, and HACCP are emerging in the US and elsewhere. These labels reflect firms efforts to capture, or to be perceived as possessing, a high degree of food safety

awareness. Such messages also exhibit a firms innovative behavior and product/process standard requirements. As part of a broader analysis of the communication of food attributes through supply chains, this paper explores food manufacturer pricing strategies for new packaged foods in an attempt to determine if safety sells in the final consumer market. While there is an evolving literature exploring consumer stated preferences for food quality (see Melton, Huffman, Shogren, and Fox 1996; McCluskey 2000; Banterle, Baldi, and Stranieri 2008) and safety (See Baker 1999; Onyango, Nayga, Jr., and Govindasamy 2006) attributes, there is limited quantitative analysis of safety messages actually applied on labels.1 In this study we investigate the use of food safety claims on new food and beverage products in seven major English speaking countries. The claims include antibiotic, E. coli, pesticide, salmonella, listeria and preservative free, food safety and HACCP. A comparison is made across two product innovation databases Datamonitors Product Launch Analytics (PLA, formerly ProductScan) and Mintels Global New Products Database (GNPD). This paper examines food manufacturer pricing strategies from a supply side focusing on the introductory price of novel packaged foods. Using hedonic price models applied to U.S. spoonable yogurt products newly launched from January 1st, 2005 to

Kimberlin and Winterstein (2008) is an exception, where the use of pharmaceutical safety messages is

explored. The authors found great variability in the quality of medication information provided.

September 30th, 2008, we investigate perceived price premiums for preservative and pesticide free claims to evaluate chemical risks. To examine the potential for differentiation based on perceptions of reduced microbiological risks, we explore the price premium of E. coli and antibiotic free claims on U.S. meat and poultry products newly launched from January 1st, 2002 to December 31st, 2004. This study explores the pricing behavior of food firms using the manufacturers suggested retail price (MSRP) at the time of launch. Launch MSRP may not be the price offered in all stores at all stages of a product life cycle. Firms may follow price skimming or penetration pricing strategies which create a dynamic price as the level of diffusion/adoption changes. Our hedonic pricing models are supply-side models, and do not analyze consumer response to these pricing strategies. Finally, in order to verify the results estimated by parametric hedonic price models, nonparametric models are applied. Following a labeling background spanning the seven countries the data are introduced, models and results presented leading to conclusions and implications.

1.2 A Background on Food Labels in Different Countries Informational labeling is designed to assist consumers in their product selections. Nutrition information and food safety information as two major types have divergent policy histories in different country. Here we compare the experiences of seven English speaking countries: the United States, United Kingdom, The Republic of Ireland, Canada, New Zealand, Australia, and South Africa. 3

In the US, there are currently two main Federal laws governing food products under FDAs jurisdiction: The Federal Food, Drug, and Cosmetic Act (FD&C Act) and the Fair Packaging and Labeling Act. The Nutrition Labeling and Education Act (NLEA) is an amendment of the FD&C Act. FDA is responsible for ensuring that foods sold in the US are safe, wholesome and properly labeled. They guide the use of terms such as low fat, light, reduced cholesterol, low sodium, low calorie, and fresh. The Nutrition Labeling and Education Act of 1990 mandated detailed nutritional information on most food packages and defined standards for health claims. In September 1994, Center for Food Safety & Applied Nutrition issued A Food Labeling Guide with revisions in June, 1999. The latest version issued in April 2008 detailed claims permitted for food labels. Current Food Labeling Guidance and Regulations include detailed ingredient lists, nutrition labeling and claim use, but do not directly cover issues such as microbial food safety, pesticide residues, and the use of preservatives. Nutrient content claims, health claims, qualified health claims, and structure/function claims are four types of claims with binding and nonbinding recommendations about appropriate usage (see Hooker and Teratanavat, 2008 for a review of the differences across such nutrition marketing claims). In the United Kingdom (UK), the Food Labelling Regulations 1996 require food to be marked or labeled with certain requirements such as name, ingredients and their amounts, the process used in manufacture, and nutrition labeling. Additional labeling requirements were set up for certain categories of food such as pre-packed alcoholic drink and raw milk (The Food Labelling Regulations 1996). The Food Standards Agency (FSA) 4

regularly issues Labelling Guidance notes for food businesses recently clarifying issues such as place of origin, Quantitative Ingredient Declaration on food labeling, what food should carry a use by date and nutrition labeling (FSA 2008). In Ireland the main legislation of food labeling is EU Council Directive 2000/13/EC (referred to as the General Labelling Directive), European Communities (Labelling, Presentation and Advertising of Foodstuffs) Regulations 2002 (SI 483/2002) and EU Directive 2003/89/EC. The legislation covers labels on pre-packaged food, labels on non-packaged food, nutritional labeling, food additives, and food supplements, etc. All food labeling legislation is enforced through the Food Safety Authority of Ireland (FSAI). In April 2008 they published a new guidance leaflet for the food industry to provide clarity on the general and compulsory legal requirements for the labeling of food (FSAI, 2008). In Canada, the 2003 Guide to Food Labelling and Advertising provides information on food labeling and advertising requirements as well as policies which apply to statements and claims made by foods and alcoholic beverages. Food claims adhering to the guidelines set out in this document are considered to comply with the provisions of the Food and Drug Act, the Food and Drug Regulations, the Consumer Packaging and Labelling Act and Regulations and other relevant legislation. Requirements include basic food labeling provisions, food claims whether in advertisements, on food labels, or other displays, nutrition labeling, nutrition content claims, diet-related health claims, etc (Canadian Food Inspection Agency 2008). 5

In South Africa the main food laws are Foodstuffs, Cosmetics and Disinfectants Act, 1972 and its regulations. The Department of Health, Department of Agriculture, and South African Bureau of Standards (SABS) are the key stakeholders in food control. The information required to appear on any label is name and address, product name, list of ingredients, the net mass declaration, storage instructions, grading of the product, and nutrition labeling, etc. New draft Food Labelling Regulations were published in 2007. The regulations made provision for an extensive list of new and amended definitions, mandatory date markings, as well as specific conditions for nutritional information on food labels (South Africa Department of Health, 2008). Australia and New Zealand jointly committed to the development and implementation of a single set of food standards through signing a Joint Food Standards Setting Treaty (the "Food Standards Treaty") in 1995, which led to the joint Australia New Zealand Food Standards Code. There is currently a prohibition on health and related claims under the Australia New Zealand Food Standards Code (The Food Code) with the exception of the folate/neural tube defect health claim. The Food Code applies to packaged food produced in, and imported into Australia and New Zealand (Australia New Zealand Food Authority 1995).

1.3 Overall Use of Food Safety Claims Given the variety of labeling regulations seen across these nations, and the relatively small number of rules specifically addressing food safety (as opposed to nutrition) a 6

comparison of the nature of the claims seen appears to be justified. This section explores the use of food safety claims on products launched in the United States, United Kingdom, The Republic of Ireland, Canada, New Zealand, Australia, and South Africa. We use information from Datamonitors Product Launch Analytics (PLA, formerly ProductScan) and Mintel Global New Products Database (GNPD). These are global databases tracking consumer packaged good innovations. We focus on the food and beverage industries, with searches covering January 1st, 1980 to September 30th, 2008 for PLA and June 1st, 1996 to September 30th, 2008 for GNPD. The following food safety related claims are investigated: antibiotic, E. coli, pesticide, salmonella, listeria, and preservative free, food safety and HACCP. Newly marketed food and beverage products are examined to determine if labels contain the above messages. The eight food safety messages are selected due to their public health significance and appearance on existing product labels. An antibiotic is an antimicrobial drug widely used for treating infections caused by bacteria in medicine, agriculture, and industrial fermentations. Misuse and overuse of these drugs, however, have contributed to public health concerns over antibiotic resistance in certain animal-based products. This resistance develops when potentially harmful bacteria change in a way that reduces or eliminates the effectiveness of human antibiotics. Food-producing animals are given antibiotics for therapeutic, disease prevention or production reasons. However, these drugs can cause microbes to become resistant to drugs used to treat human illness, ultimately making some human diseases harder to treat. Increasing problems with 7

resistant strains call for restrained use and alternative strategies. Evidence that drugs used in food-producing animals can cause antibiotic-resistant infections in consumers spurred the FDAs Center for Veterinary Medicine to take action for example the banning of the antibacterial Baytril for disease treatment in chickens and turkeys (Bren 2001). There have been similar concerns that certain pesticides used on food crops may be dangerous to consumers or to the environment. The increasing number of emerging pathogens is one of the most critical challenges for food safety; since the 1970s, bacteria which were not previously regarded as important causes of foodborne illness became more widespread such as Escherichia (E.) coli O157:H7 and Salmonella enteritidis (FDA 2005). Recent outbreaks of E. coli produced significant public health impact. In 2006, an outbreak associated with E. coli O157:H7 contaminated spinach spread across 26 states. As of January 2007, 205 persons were infected, resulting in 3 deaths; of the 103 people hospitalized, 31 developed a type of kidney failure called hemolytic-uremic syndrome (CDHS-FDA, 2007). Raw meats, poultry, eggs, milk and dairy products, fish, sauces and salad dressing are often associated with Salmonella and E. coli making these products candidates for related food safety claims. The term food safety is a frequently used, inclusive descriptor appearing in all kinds of media communications (Fleming, Thorson and Zhang 2006), often as a catch-all for public concern about food hygiene. It is included as the most general of our search terms. Preservatives are food additives used to inhibit the growth of microorganisms in 8

processed foods and improve their keeping quality or stability. Although preservatives (in the US) are generally recognized as safe (or GRAS) food ingredients consumer risk perceptions and concerns over allergies motivate their inclusion as a search term. Hazard Analysis and Critical Control Point (HACCP) is a state-of-the-art approach to food safety management spanning seven key principles. A number of US food companies already use the system in their manufacturing processes. FDA mandated HACCP for the seafood industry December 18, 1995 and for the juice industry January 19, 2001. HACCP is in use in other countries, including Canada and UK. The Food Safety Enhancement Program is the Canadian Food Inspection Agencys approach to encourage and support the development, implementation and maintenance of HACCP systems in all federally registered establishments. Plant specific HACCP plans are facilitated through the development of various generic models to cover as many processes and products as possible (Canadian Food Inspection Agency 2008). In the UK, from January 1, 2006 Regulation 853/2004 of the European Parliament on the Hygiene of Foodstuffs requires that food business operators shall put into place, implement and maintain a permanent procedure based on the principles of HACCP. The regulation applies to any size of the food business (Food Standards Agency 2008). We searched the two new product databases using inclusive queries (see Table 1 for results of PLA and GNPD searches). All food and beverage products are included in the search process. Few products include a food safety claim. However, there are some interesting similarities and differences in the results across the databases. First, HACCP, 9

E. coli, salmonella and listeria free messages are infrequently used on product labels. This result is consistent across the seven countries, although the appearance of such claims on food and beverage innovations is relatively higher in the US. Regardless, the prevalence of such claims remains very rate. Based on the PLA data, the total number of new food and beverage items launched was: UK 19,500; Canada 7,210; Ireland 965; South Africa 4,098; New Zealand 3,603; Australia 9,807; and USA 115,430. Of these only 0-0.08% contained a food safety claim.

10

UK CANADA Ireland South Africa New Zealand Australia US 11

Antibiotic free 0 5 0 0 0 0 94

E. coli free 1 1 0 0 0 0 14

PLA (January 1st, 1980 to September 30th, 2008) Pesticide Salmonella Food safety HACCP free free 2 0 4 7 10 2 7 1 0 0 0 0 1 0 0 1 1 0 0 0 2 0 0 2 70 18 83 28 GNPD (June 1st 1996 to September 30th, 2008) Pesticide Salmonella Food safety HACCP free free 6 0 38 5 14 0 70 0 1 0 1 0 1 3 18 0 25 0 15 0 10 9 65 0 137 11 424 11

Listeria free 1 0 0 0 0 0 5 Listeria free 0 0 0 0 0 0 1

Preservative free 49 35 2 56 19 45 938 Preservative free 4112 1520 432 831 1034 2384 8585

Antibiotic E. coli free free UK 12 0 CANADA 37 0 Ireland 1 0 South Africa 7 0 New Zealand 1 0 Australia 9 0 US 587 9 Table 1.1: Summary of Food Safety Claims

11

For the following six countries, UK, Ireland, Canada, New Zealand, Australia, and South Africa, new products rarely included any of the first seven claims (antibiotic, E coli, pesticide, salmonella or listeria free, food safety or HACCP). These signals include supply chain wide messages that require information to be shared by the producers and manufacturer before it can be sent to the consumer. In comparison, US innovations appear to more frequently use these terms. This may be contrary to expectations given US product liability laws. Such laws describe the circumstances under which an individual can recover damages for a defective food item and the nature and extent of compensation that may be awarded for injuries or deaths due to contaminated food products (Clark, 2000). When the manufacturer fails to exercise reasonable care in producing, marketing, or selling the implicated food, and because of this failure someone becomes ill, the manufacturer is then may be held liable for any resulting court-awarded compensation (Rasco, Frenzen, and Buzby 2001). If a firm claims a higher level of safety on the food label, but then has a recall, they may be more likely to be held responsible for the personal injuries attributed to the contaminated food.2 The balance of economic incentives for (risk averse) firms may align with caution for food label claims, explaining the relatively low adoption rate for such messages. An alternative strategy has firms engaging in (supply chain wide) efforts to elevate product quality, but not exhibiting them on the food label. If this is the case, firms may be missing a valuable opportunity to attract health-conscious consumers and command a price premium. Certain firms do

Alternatively, the notion of average quality (Rasco 1997) may suggest that once one firm makes such a

claim, it is incumbent upon the entire industry to follow similar practices else they may be found to be supplying dangerous product that breaks implicit warranty provisions.

12

currently present food safety information on their labels to help consumers make product choices and educate consumers about food safety. Antibiotic and pesticide free, and food safety messages are more frequently seen. In GNPD, the number of products carrying these messages reached several hundred in the US and dozens in other countries especially the UK, Canada and Australia. The more frequent presence on food labels is in line with the increasing public attention and preference for natural and organic foods in these markets. For example, according to FDA policy, "natural" means the product does not contain synthetic or artificial ingredients (FDA 2008). A preservative free claim is the mostly frequently seen food safety message. In PLA, there are nearly one thousand new food and beverage products with preservative free messages, in GNPD more than eight thousand. Food safety claims span several food categories (see Table 2). Salmonella free claims mostly appear on dairy foods, but also meat, poultry, meals and entrees. Similarly, E. coli and Listeria free claims are mainly used on meat, certain poultry and dairy products. HACCP claims generally appear on beverages and seldom on food products.

13

Category

dairy food Yes meat, Yes poultry Yes meals and entrees Yes beverage Yes Bread Product Cereals Chips Cookies Sandwiches Pastry & Baked Products Sauces & Gravies Spices, Extracts & Seasonings Sauces, Pizza & Pasta Yes Oil, Shortening & Cooking Yes Sprays Fruits & Fruit Side Dishes Yes Snacks Spices, Extracts & Seasonings Staples Vegetables & Vegetable Side Dishes Note: Yes implies that the number of new food product innovation is at least 2. Table 1.2: Distribution of Food Safety Claims across Product Categories

Salmone lla free Yes Yes Yes Yes

E. coli free Yes Yes Yes

Listeria free Yes Yes Yes

HACCP

Food Safety

Antibiotic free Yes Yes Yes Yes

Pesticide free Yes Yes Yes Yes

Preservative free

Yes Yes Yes Yes Yes Yes Yes Yes Yes

14

Yes Yes Yes Yes Yes Yes

14

1.4 The Hedonic Price Model Turning to the perspective of food manufacturers it is interesting to determine whether they have attempted to incorporate a premium into their pricing strategies for products bearing a claim of some type. Hedonic price models are used to explore pricing strategies. The hedonic model, developed by Rosen (1974) constructs implicit prices of attributes by generating an equilibrium between consumers and producers actions. As stated above, we dont include market prices in this analysis, focusing on the supply-side prices p(z) which we assess for values based on presence or absence of product characteristics. Rosen argued that the hedonic pricing function is completely determined by the supply side in the long run. This long-term equilibrium indicated that the marginal price of a particular attribute should be equal across firms, keeping other attributes constant, and should be the same as the lowest marginal cost of producing that attribute (Nimon and Beghin, 1999). In practice, hedonic pricing models have been widely used to value food attributes employing either demand or supply side data (See Loureiro and McCluskey, 2000; Nimon and Beghin, 1999). Employing the standard hedonic price model setup, the price of a food product P is assumed to be defined by a function, P = P(z ) relating prices and characteristics, where z is a vector of attributes. This z accounts for the possibility that some firms pay better attention to producing a particular bundle of attributes than others. The implicit price of an additional unit of a particular attribute is estimated as the partial derivative of the hedonic price function in terms of that specific characteristic. Each producer chooses an optimal bundle of attributes to produce in order to maximize profit subject to a cost constraint. The coefficient of the attribute suggests the marginal willingness to accept a

15

price for a specific product attribute (Rosen, 1974). A general hedonic model can be written as P ( z ) = p( z1 , z 2 , z3 , L , z n ) , where z = ( z1 , z 2 , z3 , L , z n ) is a vector including a bundle of product attributes. Pzi is the implicit price of the attribute zi from the food manufacturers perspective. Feenstras theoretical framework (1995) suggests that when considering imperfectly competitive markets, price markup above marginal cost should be added as an explanatory variable in the model. However, in empirical work based on Rosens hedonic pricing model perfect competition is widely assumed (yet the approach attempts to value product differentiation strategies). Indeed the literature applying hedonic pricing methods can be categorized into two sets. One set simply does not mention market structure. For example, Loureiro and McCluskey (2000) use a hedonic approach to estimate the price premium of the Protected Geographical Identification label in fresh meat products. Lecocq and Visser (2006) analyzed Bordeaux wine, investigating the objective characteristics presented on the label, finding that market price is influenced by those features. Nimon and Beghin (1999) used apparel catalog data to evaluate price premiums for environmental attributes such as organic-cotton apparel, environmentally friendly dyes, and no-dyes. Lacking production cost, they selected Rosens perfectly competitive model rather than Feenstras imperfect competition framework and argued the virtue of the approach. The other set of articles do have cost information. Osborne and Smith (1997) evaluated the price markup of site-specific environmental amenities for coastal beaches. They explored private firms rental price, using time series data for a six year period. Firm-specific wage indexes and a wage index for the real estate sector were measured

16

cost variables. It seems that in practice most researchers applied the hedonic model without consideration of market structure as few have cost information. In order to consider market structure we incorporate a product innovation level variable. This variable describes if the product is highly innovative or is similar to existing products. This proxy is an indicator of how many similar products there may be in the category and thus is a measure of market structure.

1.4.1 Data This study estimates the implicit initial offer prices of food characteristics. MSRP aims to standardize prices across locations at the time of launch. This price data is generally available in GNPD and PLA. The US spoonable yogurt market is first considered. The entire set of newly marketed yogurts available in the GNPD database, launched between January 1st, 2005 and September 30th, 2008 are explored.

Two possible model specifications can be used (equations 1 and 2). (1)
PPO = 0 + 1 PFREE + 2 KOSHER + 3 BRAND + 4 NEW + 5 CHILD + 6 ORGA + 7 PEST +
LnPPO = 0 + 1 PFREE + 2 KOSHER + 3 BRAND + 4 NEW + 5 CHILD + 6 ORGA + 7 PEST +

(2)

Previous applications have used either linear (see Wilson 1984; Bolan and Schroeder 2002; Maguire, Owens, and Simon 2004; Taylor and Brester 2005) or semi-log functional forms (see Estes and Simith 1996; Steiner 2004) to examine price-quality characteristics. The more flexible functional form of a Box-Cox transformation model has also been used

17

(See Palmquist 1984; Jordan et al. 1985; Loureiro and McCluskey 2000) and is applied here. The power parameter = 0 is found in the estimated confidence interval [-0.5, 0.1], suggesting a semi-log functional form of the hedonic price model is appropriate. Therefore equation 2 is estimated.

1.4.2 Empirical Estimation and Results Inherent in such hedonic approaches concerns can be raised endogenity. A Hausman test confirmed such between preservative free claims and product price. Thus we employed a Two-Stage Least Squares model with an instrumental variable (the new fitted value of the endogeneous variable PFREE) to derive consistent parameter estimates. The price per ounce (PPO) is computed from the MSRP. Variable definitions and summary statistics are presented in table 3. All products which use an antibiotic free message also have a pesticide free claim. In order to avoid multicolinearity only one of the messages can be included in the model. Variable PEST is selected. If variable BIO is used, the results are not significantly different. Spoonable yogurt has an average price per ounce of $0.1245 with a wide range between $0.0309 and $1.165. Table 4 reports the results of the hedonic price model and the estimated coefficients for each included label message. Given the semi-log specification, coefficients can be interpreted as a percent change of the average price. A price premium for each message is calculated using (3) reported in Table 4. (3) Price Premium of Parameter = Average (PPO)*Parameter Estimate.

18

Variable PPO PFREE

Description Price per ounce ($) Preservative free claim

Mean

Std Dev Minimum Maximum 0.0965 0.2495 0.0309 0 1.165 1

181 0.1245 181 0.0663

(yes=1, no=0) Kosher certified (yes=1, KOSHER no=0) =1 if the product is a BRAND national brand, =0 if a private brand =1 if the product is a new NEW formulation or new variety, =0 otherwise Pesticide free claim PEST (yes=1, no=0) =1 if the product is CHILD positioned toward children, =0 otherwise Organic claim (yes=1, ORGA no=0) Table 1.3: Descriptive statistics of the GNPD data (Spoonable Yoghurt) 181 0.0773 0.2679 0 1 181 0.0884 0.2847 0 1 181 0.0055 0.0743 0 1 181 0.4972 0.5014 0 1 181 0.7182 0.4511 0 1 181 0.5028 0.5014 0 1

19

Variable Intercept PFREE KOSHER BRAND NEW CHILD ORGA PEST n=181

Parameter Estimate Standard Error t Value Pr > |t| Estimated Premium -2.4138 0.4090 -0.0297 0.2781 -0.0025 -0.1474 0.3909 -0.2849 F value= 4.52 0.0779 0.1377 0.0665 0.0741 0.0675 0.1188 0.1288 0.4585 Pr > F=0.0001 -31 2.97 -0.45 3.76 -0.04 -1.24 3.04 -0.62 <.0001 0.0034 0.6559 0.0002 0.9706 0.2163 0.0028 0.5353 -0.0509 -0.0037 0.0346 -0.0003 -0.0183 0.0487 -0.0355

Table 1.4: Parametric Hedonic Price Model: GNPD Data

PFREE is positive as expected and significant at the 95% level. The parameter estimate implies a preservative free claim has an average premium of 40.90% (a $0.0509 markup). A Kosher claim, however, has a small negative but statistically insignificant effect on yogurt price. Though the claim has no direct impact on price, it may still have other marketing advantages (promotion, product differentiation, etc.). Compared to a private label, national brands generate an average premium of 27.81%, a $0.0346 markup, significant at the 95% level. NEW is statistically insignificant which implies that innovative products have price advantage over those entering a category with similar existing products. Based on this crude proxy, the effect of market structure is not obvious in spoonable yogurt products.

20

Yogurts with an organic claim have a statistically significant (95% level) price premium, 39.09% ($0.0487). Factors influencing the adoption of an organic claim include that ingredients are grown without using most conventional pesticides, fertilizers made with synthetic ingredients, bioengineering or sewage sludge and that ionizing radiation is not used (USDA 2002). Producers have realized the value of organic claims on product labels, and have an increasing interest in marketing organic yogurt. In comparison antibiotic and pesticide free claims (PEST) are not significant. Notice the number of yogurts carrying such messages is small.

1.5 A Hedonic Price Model for E. coli Free Foodborne illness caused by microorganisms is a large and growing public health problem (World Health Organization 2008). Most countries have documented significant increases over recent decades in the incidence of disease caused by microorganisms in food including pathogens (World Health Organization 2008). Among all microbiological risks, E. coli is chosen as an example in our hedonic analysis due to recent outbreaks. Among the products claiming to be E. coli free, meat and poultry account for 8, the other 6 include dairy and baby food products. Thus meat and poultry products are our focus for the E. coli free claim. The data span January 1, 2002 to December 31, 2004 and were collected from the PLA database. The sample size is 172. Variable PPO is price per ounce calculated as above. Variable definitions and summary statistics are presented in table 5. An average meat and poultry product had a price per ounce of $0.3557 with a wide range between $0.079 and $1.7475. Among the variables explored, product package

21

size may affect price if firms charge lower per unit prices for bulk or family packages (see FAMILY).

Variable PPO

Description Price Per ounce ($)

Mean

Std Dev Minimum Maximum 0.2332 0.1795 0.0790 0 1.7475 1

172 0.3557 181 0.0331

E. coli free claim


EFREE (yes=1, no=0) Natural claim (yes=1, 181 0.0994 NATU no=0) Antibiotic free claim 181 0.0829 ANTIFREE (yes=1, no=0) =1 if the product is 181 0.5856 MEAT meat, =0 if poultry =1 if the product is innovative, =0 NEW otherwise =1 if family package 181 0.0939 FAMILY size, =0 if regular size =1 if cooked or fully 181 0.5249 COOKED cooked, =0 if raw =1 if ground meat 181 0.0276 FREGR products, =0 otherwise 0.1643 0 1 0.5008 0 1 0.2925 0 1 181 0.0523 0.2233 0 1 0.4940 0 1 0.2765 0 1 0.3001 0 1

Table 1.5: Descriptive Statistics of PLA Data (Meat and Poultry Products)

22

Applying a Box-Cox maximum likelihood method again, the optimal is found to be zero. Therefore, a semi-log functional form of the hedonic price model is appropriate - equation 4. (4)
LnPPO = 0 + 1 EFREE + 2 NATU + 3 ANTIFREE + 4 MEAT + 5 NEW + 6 FAMILY + 7 COOKED + 8 FREGR +

Possible endogeneity between antibiotic free and price was again confirmed with a Hausman test. A Two-Stage Least Squares model is used to correct for the endogeneity, table 6 presents the estimated coefficients for label claims. The percent change of the average price and implicit price premiums are also estimated. The sign on EFREE is positive as expected implying products with an E. coli free message command a higher price than those without with an average 14.75% premium ($0.0525). However, the variable is not significant. The total number of products with an E. coli free message is much smaller than the other claims explored above. This may suggest that producers and manufacturers have paid less attention to differentiating products based on E. coli claims, or that concern over liability is more pressing in these products. A natural claim has a positive impact (64.20% premium, $0.2284 markup) which is significant at the 95% confidence level. Factors influencing the use of natural claims include that product does not contain artificial flavors, coloring ingredients, chemical preservatives, or any other artificial or synthetic ingredients, and that the product and its ingredients are not more than minimally processed (USDA 2005). Positive marginal prices for the use of an antibiotic free claim by meat and poultry manufacturers are statistically significant at a 95% level of confidence, $0.5489 per ounce an average 54.32% premium.

23

Variable Intercept EFREE NATU ANTIFREE MEAT NEW FAMILY COOKED FREGR n=172

Estimate -1.1836 0.1475 0.6420 0.5432 -0.1220 0.3271 -0.2556 0.0663 -0.6312

Error 0.0707 0.4584 0.1192 0.1613 0.0717 0.1556 0.1161 0.0705 0.3253

t Value Pr > |t| Estimated Premium -16.74 0.32 5.38 3.37 -1.7 2.1 -2.2 0.94 -1.94 <.0001 0.7481 <.0001 0.0009 0.0909 0.0371 0.029 0.3482 0.054 -0.0525 0.2284 0.1932 -0.0434 0.1163 -0.0909 0.0236 -0.2245

F value= 7.19 Pr > F=0.0001

Table 1.6: Parametric Hedonic Price Model: PLA Data

1.6 Nonparametric Models Rosens theory of hedonic prices (1974) theory has also been tested econometrically using recently developed nonparametric techniques to examine the effects of qualitative factors on product price. Rosen recommended the functional form restrictions should be relaxed. Nonparametric methods place no restrictions on it may provide more reliable information about implicit prices. To verify the estimation results of the parametric hedonic price models in previous sections, we employ generalized nonparametric regression fitting an additive regression model to GNPD and PLA data for spoonable yogurt and meat and poultry products respectively.

24

The generalized additive nonparametric regression model specifies that the average value of y is a sum of separate terms for each predictor, but these terms are merely assumed to be smooth functions of the xs: (5) E ( y x1 , x2 , L xk ) = + f1 ( x1 ) + f 2 ( x2 ) + L + f k ( xk )

Variable Intercept PFREE KOSHER ORGA BRAND NEW CHILD PEST n = 181

Coefficient Estimate Standard Error t Value 0.0852 0.0454 0.0062 0.056 0.037 0.0103 -0.0251 -0.1129 0.0168 0.0296 0.0143 0.0277 0.016 0.0145 0.0255 0.1022 5.082 1.533 0.433 2.021 2.309 0.71 -0.982 -1.105

Pr(>|t|) 9.63E-07 0.127 0.6653 0.0448 0.0221 0.4789 0.3277 0.2707

Table 1.7: Nonparametric Hedonic Model: GNPD Data

As Fox (2005) stated, an advantage of the additive regression model is that it reduces to a series of two-dimensional partial regression problems, which means each f i ( xi ), i = 1, L , k . This is beneficial both in computation and, even more importantly, regarding interpretation. Concerns over dimensionality are avoided through a univariate smoother, and estimates of the individual terms explain how the independent variables

25

affect the dependent variable. The generalized additive nonparametric models for yogurt data and meat and poultry product data are presented in equations (6) and (7). (6)

PPOi = + f 1 ( PFREEi ) + f 2 ( KOSHERi ) + f 3 ( BRANDi ) + f 4 ( NEWi ) + f 5 (CHILDi )


+ f 6 (ORGAi ) + f 7 ( PESTi )

(7)

PPOi = 1 + g1 ( EFREEi ) + g 2 ( NATU i ) + g 3 ( ANTIFREEi ) + g 4 ( MEATi )


+ g 5 ( NEWi ) + g 6 ( FAMILYi ) + g 7 (COOKEDi ) + g 8 ( FREGRi )

where f i ( X ), i = 1, 2, L ,7, and g i ( X ), i = 1, 2, L,8 are smooth functions.

Variable Intercept EFREE NATU ANTIFREE COOKED MEAT NEW FAMILY FREGR

Coefficient Estimate 0.3098 0.1847 0.2237 0.2567 0.0455 -0.015 0.027 -0.0619 -0.3114

Standard Error 0.0322 0.2091 0.0611 0.0732 0.032 0.0326 0.0743 0.0528 0.1491

t Value 9.629 0.883 3.661 3.506 1.425 -0.459 0.362 -1.171 -2.089

Pr(>|t|) < 2e-16 0.378392 0.000339 0.000587 0.156172 0.646634 0.71746 0.243238 0.038273

Table 1.8: Nonparametric Hedonic Model: PLA Data (n = 172)

The yogurt results (see table 7) imply that a preservative free message on average has an implicit price premium of $0.0454. In the parametric model, the implicit price premium was $0.0509. Thus, we are confident that the preservative free claim has a positive premium of 5 cents per ounce. It is significant at an 85% confidence level. If the 26

yoghurt claims to be organic it has an average $0.056 premium compared to the $0.0487 premium in the parametric model (significant at the 95% level). As with the parametric model, pesticide and antibiotic free claims are insignificant. The estimation results of a nonparametric model applied to meat and poultry products (see table 8) suggest that those products with an E. coli free message on average have an implicit price premium of $0.1847 compared to products without such a claim. As with the parametric model, the variable EFREE is insignificant. The number of products using an E. coli free message is so small; firms may also have a limited ability to monitor such a claim through the supply chain (Hooker and Roe 2002). A natural claim has an average $0.2237 premium which is statistically significant at the 95% confidence level. The nonparametric model verifies the significance level of an antibiotic free claim, and indicates an average $0.2567 premium similar to $0.1932 premium in the parametric model.

1.7 Conclusions This study investigates the use of safety messages on food and beverage product labels. By examining recent innovations reported in two tracking databases, we find the US is leading in the still uncommon use of food safety messages among other English speaking countries. Preservative free claims are most frequent among the set of claims explored, followed by antibiotic and pesticide free and food safety claims. E. coli, Salmonella and

Listeria free and HACCP information appear on labels to a less extent.


We develop parametric and nonparametric hedonic price models to investigate potential price premiums from chemical and microbiological risk reductions in two food

27

categories in the US. Nonparametric model results verify the direction of the parameter estimations. We find evidence that a preservative free message adds extra value to yoghurts, on average about 5 cents per ounce. We also found an average premium for an

E. coli free attribute ($0.0525 in parametric and $0.1847 in nonparametric model). The
limited number of products with E. coli free messages may impose restrictions on its significance. Antibiotic free appears to be valued in meat and poultry product price setting. Thus, we find limited support that safety sells, and influences the pricing of yogurt and meat and poultry products. In addition, firms have realized the value-adding benefit of claims such as preservative and antibiotic free. The value of microbiological free claims awaits further recognition and practical labeling, supply chain and liability strategies. The infrequent use of food safety claims may be due to unclear assurances of food from suppliers. While certain claims may be under the direct control of activities conducted within the food manufacturers plant (e.g., pathogen risk reduction through an intervention or kill step) other claims (e.g., hormone free) may rely on information being shared through the supply chain. The price premium findings for food safety claims are specific to spoonable yogurt and meat and poultry products in the US, and they will be enriched from investigation of such claims applied in other food categories and nations. This investigation examines the firms launch (MRSP) pricing behavior, and the analysis of food safety claims could be furthered by incorporating consumer behavior throughout the product life cycle. Sales (e.g., scanner) data combined with this product label information

28

can also contribute to a richer comparison of stated and revealed preferences for food safety claims when linked to consumer estimates of willingness to pay. Claims on food labels provide a (leading) indicator of consumer demand. Their use can provide a strategic advantage for producers and retailers. Full utilization of the value of food safety claims informs researchers about a firms pricing strategy, and provides a better understanding of private incentives to deliver food safety. Food safety risks may be real or perceived, with the perception of risk influenced by uncertainty (Wahlqvist and Ball 2002). This study does not explore this issue in sufficient depth. Further analysis of this dimension is justified.

29

References Australia New Zealand Food Authority. 1995. Code of Practice on Nutrient Claims in Food Labels and in Advertisements. Available at http://www.foodstandards.gov.au/_srcfiles/Code_of_Practice_jan1995.pdf (accessed December 2008). Baker, G.A. 1999. Consumer Preferences for Food Safety Attributes in Fresh Apples: Market Segments, Consumer Characteristics, and Marketing Opportunities.

Journal of Agricultural and Resource Economics 24(1): 80-97.


Banterle, A., L. Baldi, and S. Stranieri. 2008. Do Nutritional Claims Matter to Consumers? An Empirical Analysis Considering European Requirements. Paper presented at the 8th International Conference on Management in AgriFood chains and Networks, Ede-Wageningen, The Netherlands. Boland M., and T. Schroeder. 2002. Marginal Value of Quality Attributes for Natural and Organic Beef. Journal of Agricultural and Applied Economics 34(1): 39-49. Bren, L. 2001. Antibiotic Resistance From Down On the Chicken Farm. FDA Veterinarian Vol. XVI, No.I. U.S. Food and Drug Administration, Washington, D.C. Buzby, J.C., P.D. Frenzen, and B. Rasco. 2001. Product Liability and Microbial Foodborne Illness. Economic Research Service, U.S. Department of Agriculture, Washington, D.C. Canadian Food Inspection Agency. 2008. Acts and Regulations. Available at http://www.inspection.gc.ca/english/reg/rege.shtml (accessed December 2008).

30

CDHS-FDA. 2007. Investigation of an Escherichia coli 0157:H7 Outbreak Associated with Dole Pre-Packaged Spinach. Washington, D.C.: California Food Emergency Response Team, California Dept of Health Services, and FDA-CFSAN. Available at: http://www.marlerclark.com/2006_Spinach_Report_Final_01.pdf (accessed December 9, 2008). Clark, B.T. 2000. Attorney at Marler Clark. Personal Correspondence. Estes E.A., and V.K. Smith. 1996. Price, Quality, and Pesticide Related Health Risk Considerations in Fruit and Vegetable Purchase: An Hedonic Analysis of Tucson, Arizona Supermarkets. Journal of Food Distribution Research 27: 59-76. Fleming, K., E. Thorson & Y. Zhang. 2006. Going Beyond Exposure to Local News Media: An Information-processing Examination of Public Perceptions of Food Safety. Journal of Health Communication, 11(8): 789-806. Food Safety and Inspection Service. 2006. Product Labeling: Definition of the Term Natural. Food Safety and Inspection Service, U.S. Department of Agriculture, Washington, D.C. Available at http://www.fsis.usda.gov/OPPDE/rdad/FRPubs/2006-0040.htm (accessed December 2008). Food Safety Authority of Ireland. 2008. Legislation. Available at http://www.fsai.ie/legislation.html (accessed December 2008). Food Standards Agency. 2008. Labelling Guidance. Available at http://www.food.gov.uk/foodindustry/guidancenotes/labelregsguidance/ (accessed December 2008).

31

Fox, J. 2000. Multiple and Generalized Nonparametric Regression. Sage Publications, Inc. Feenstra, R.C. 1995. Exact Hedonic Price Indexes. The Review of Economics and

Statistics 77(4): 634-653.


Gold, M.V. 2007. Organic Production/Organic Food: Information Access Tools. Alternative Farming Systems Information Center, U.S. Department of Agriculture, Washington, D.C. Available at http://www.nal.usda.gov/afsic/pubs/ofp/ofp.shtml (accessed December 2008). Hooker, N. H. and B. Roe. 2002. The Role of Food Safety Systems in Cattle Marketing. Managing for Todays Cattle Market and Beyond. 2nd Edition. Available online at http://www.ag.arizona.edu/arec/wemc/cattlemarket/CattleMarketing&Food_Safet y.pdf (accessed December 2008). Hooker, Neal H. and R. P. Teratanavat. 2008. Dissecting Qualified Health Claims: Evidence from Experimental Studies. Critical Reviews in Food Science and

Nutrition. 48(2): 160-176.


Huang, C.L., and B.H. Lin. 2007. A Hedonic Analysis of Fresh Tomato Prices among Regional Markets. Review of Agricultural Economics 29(4): 783-800.
Jordan, J. L., R.L. Shewfelt, S.E. Prussia, and W.C. Hurst. 1985. Estimating Implicit Marginal Prices of Quality Characteristics of Tomatoes. Southern Journal of Agricultural Economics. 17: 139-146.

Kimberlin, C.L., and A.G. Winterstein. 2008. Expert and Consumer Evaluation of Consumer Medication Information-2008. Final Report to the U.S. Department of Health and Human Services and the Food and Drug Administration.

32

Lecocq, S. and M. Visser. 2006. What Determines Wine Prices: Objective vs. Sensory Characteristics. Journal of Wine Economics 1(1): 42-56. Loureiro M.L., and J.J. McCluskey. 2000. Assessing Consumer Response to Protected Geographical Identification Labeling. Agribusiness 16 (3): 309-320. Maguire K.B., N. Owens, and N.B. Simon. 2004. The Price Premium for Organic Babyfood: A Hedonic Analysis. Journal of Agricultural and Resource

Economics 29(18): 132-149.


McCluskey, J.J. 2000. A Game Theoretic Approach to Organic Foods: An Analysis of Asymmetric Information and Policy. Agricultural and Resource Economics

Review 29(1): 1-9.


Melton, B.E., W.E. Huffman, J.F. Shogren, and J.A. Fox. 1996. Consumer Preferences for Fresh Food Items with Multiple Quality Attributes: Evidence from an Experimental Auction of Pork Chops. American Journal of Agricultural

Economics 78(4): 916-923.


Nimon, W. and J. Beghin. 1999. Are Eco-Labels Valuable? Evidence from the Apparel Industry. American Journal of Agricultural Economics 81(4): 801-811. Onyango, B., R.M. Nayga, Jr., and R. Govindasamy. 2006. U.S. Consumers Willingness to Pay for Food Labeled Genetically Modified. Agricultural and

Resource Economics Review 35(2): 299310.


Osborne, L. and V.K. Smith. 1997. Environmental Amenities, Product Differentiation, and market Power. Mimeo. Palmquist, R.B. 1984. Estimating the Demand for the Characteristics of Housing.

Review of Economics and Statistics 66(3): 394-404.

33

Rasco, B. 1997. How HACCP Can Help in a Products Liability Action. Food Quality (October): 16-18. Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. The Journal of Political Economy 82(1): 34-55. South Africa, Department of Health. 2008. Legislation-Acts. Available at http://www.doh.gov.za/docs/index.html (accessed December 2008) Steiner B.E. 2004. Australian Wines in the British Wine Market: A Hedonic Price Analysis. Agribusiness 20(3):287-307. Taylor M.R., and G.W. Brester. 2005. Noncash Income Transfers and Agricultural Land Values. Review of Agricultural Economics 27(16): 526-541. The Food Labelling Regulations. 1996. Statutory Instrument 1996 No. 1499 Available at http://www.opsi.gov.uk/si/si1996/Uksi_19961499_en_1.htm (accessed December 2008). U.S. Food and Drug Administration. 2005. Department of Health and Human Services Food and Drug Administration Fiscal Year 2005. FDA, Washington, D.C. Available at: http://www.fda.gov/AboutFDA/ReportsManualsForms/Reports/BudgetReports/20 05FDABudgetSummary/ucm117926.htm (accessed December 9, 2008). ________. 2008. Food Label Helps Consumers Make Healthier Choices. FDA, Washington, D.C. Available at: http://www.fda.gov/ForConsumers/ConsumerUpdates/ucm094536.htm (accessed December 6, 2009).

34

Wahlqvist M.L., and M. Ball. 2002. Food and Nutrition: Australasia, Asia and the

Pacific. Australia: Allen & Unwin.


Wilson, W. 1984. Hedonic prices in the malting barley market. Western Journal of

Agricultural Economics 9(1): 29-40.


World Health Organization. 2002. Foodborne Disease. Document proposed at FAO/WHO Global Forum of Food Safety Regulators, Marrakech, Morocco.

35

Chapter 2 Consumer Interest and Engagement Decision in Personalized Food Recall Notification Services

2.1 Introduction Outbreaks of foodborne illness can pose serious threats to consumers. The Centers for Disease Control and Prevention (CDC) estimates that 76 million Americans are sickened, 325,000 are hospitalized and 5,000 die annually from foodborne illness (Center for Foodborne Illness Research & Prevention, 2009). To protect public health, when food companies discover that a food product has been implicated in an outbreak or is found to contain unacceptable levels of contaminants, they usually voluntarily issue a recall notice to recover any unconsumed products thought to pose a risk. These recall notices are intended to alert distributers and retailers to remove unsold products from their shelves, and to warn consumers to destroy or return such products that they have already purchased. Recalls are designed to prevent public exposure to risks, frequently foodborne pathogens Teratanavat and Hooker, 2004). Therefore, recovery speed and the proportion of product accounted for (recovery rate) are critical to their success. Unfortunately, recovery rates close to 100 percent only occur in about one third of all food recall cases 36

(Teratanavat and Hooker, 2004). Instead, most recalls result in far lower rates of recovery. An analysis of recalls in 2003 by the United States Government Accounting Office (GAO) found that only about 38 percent of recalled food was ultimately recovered in recalls overseen by the USDA (which regulates meat, poultry, and processed egg products), only 36 percent was recovered in recalls overseen by the FDA (which regulates all other food products) (GAO, 2004). Recovery rates can also vary significantly. In a study of meat and poultry recalls from 1994-2002, Teratanavat and Hooker (2004) found that the mean rate of recovery was 49 percent (ranging from 39 to 77 percent). In practice, unsold food products are relatively easy to recover from consignees (Shang and Hooker, 2005). Many major food manufacturers have systems in place to alert their distributers and the retail food sector when a recall notice is issued. In turn, food retail chains are able to quickly contact their stores to pull recalled products from their shelves. In addition, a number of private firms specializing in recovering recalled food products from the marketplace also offer their services such as Stericycle. In contrast, products that have already been purchased by consumers are much more difficult to reclaim. Therefore, efforts designed to understand and address the challenges of recovering recalled products post-purchase, could have significant public health benefits. Unfortunately, there is limited research considering consumer knowledge of and responses to food recalls. A national survey of the public response to the recall of spinach contaminated with E. coli O:157:H7 in 2006, found that 87% of Americans were aware 37

of the recall. However, the details of the recall, including the types of spinach that had been recalled were not clear to the majority. Seventy one percent of respondents who were aware of the recall had learned about it through television coverage (Cuite, et al., 2007). Another study looked at the efficacy of recall notification and advertising based on experiences during a nationwide recall of hot dogs and deli meats. Forty five percent of respondents interviewed in seven states knew about the recall. However, 5% believed the products were safe to eat and 23% were not sure. People over 40 and those interviewed after major newspaper notices had higher knowledge. Seventy percent of respondents who were aware of the recall had learned about it through television coverage. The authors concluded that routine recall notifications failed to reach a large portion of the population and were not well understood (Patrick, et al., 2007). In an assessment of Americans knowledge and actions of the Salmonella Saintpaul outbreak in the summer of 2008, 1,101 respondents were interviewed about their awareness of the status of the outbreak, and the actions they may have taken to avoid becoming ill. The survey found that the majority of Americans (93%) were aware that consumers had been warned not to eat certain types of tomatoes implicated in the outbreak, but only 68% were aware that peppers had been suspected of causing illness. In addition, the study found that less than one-third (31%) of the respondents strongly agreed that they knew which tomatoes consumers were warned not to eat. The study also found that 11% of Americans intentionally disregarded FDAs advice and ate the types of 38

tomatoes they knew they had been warned not to eat. The results imply that while Americans give significant attention to communications about foodborne illness, they often misunderstand or disregard the specifics of the outbreak and recommended actions (Cuite, et al., 2009). In an analysis of how Americans respond to food recalls in general, Hallman, Cuite, and Hooker (2009) concluded that most say they pay a great deal of attention to food recalls and, when they learn about them, they tell many other people. But, they found that 40 percent of consumers think the foods they purchase are less likely to be recalled than those purchased by other Americans, appearing to believe that food recalls just dont apply to them. As a result, only 59% of American consumers say they have

ever looked for a recalled food item, and only 10% say they have found one. The authors
suggest that one way to break through the apparent disregard for recall information would be for retailers to provide more personalized information to their customers when food items they have purchased are the subject of a recall. Acting as the important direct source of food products for most Americans, food retailers play the most immediate role in consumers food purchases and preparation. Thus, food retailers can play a vital role in improving the outcomes of food recalls. In particular, according to the Food Marketing Institute (FMI, 2008), nearly half (48.8%) of food retailers operate frequent shopper loyalty card programs designed to link customers with their purchases, in which 90% of their customers participate. Thus, through their shopper loyalty card programs, many retailers have the ability to identify 39

and communicate recall information to customers who have purchased foods that become subject to a recall notice. Some food retail chains, including Costco, Wegmans, ShopRite, and Kroger have already begun to offer such personalized recall notification services. This paper focuses on American consumers level of interest in participating in personalized recall notification services offered by food retailers, investigating the association of specific demographic and socioeconomic variables with this level of interest.

2.2 Consumer Interest Analysis in Personalized Recall Notification Service 2.2.1 Data The data used in this paper come from a national survey conducted by the Food Policy Institute (FPI) at Rutgers, the State University of New Jersey from August 4 to September 24, 2008. Telephone interviews were conducted with a nationally representative sample of 1,101 American adults. The survey investigated Americans attitudes toward and behavioral responses to food contamination and food recalls, including information about consumers awareness of recent recalls, interest in recall notification services, and prior experience with foodborne illness was collected. Detailed results of the survey are reported in Hallman, Cuite and Hooker (2009). In the survey, respondents were told, Some grocery stores provide personalized

services that alert consumers if a food product that they had already purchased had been recalled. They were then asked, Would you want your grocery store to offer this
40

service?, Would you be willing to pay for this personalized service?, and What is the most youd be willing to pay per year for this service? The responses to these questions result in three discrete levels of interest (LOI) in the service. One group of consumers have no interest in this personalized service (28.4%), another group of consumers does have interest in the service but are not willing to pay for it (57.8%), and another group is interested in the service and is willing to pay for it (13.8%, see table 1). Accordingly, an ordered logit model is appropriate for the required analysis, with corresponding values for the observed variable (LOI) of 1, 2 and 3, respectively.

41

No interest in notification service Frequency N=313 Percentage of sample: 28.4% Variable GENDER PRIM SMELSE ALLER ILL AGE CHILD EDU MARR HISP INCOME EMPLOYF EMPLOYP EMPLOYR 42 Mean 0.521 0.620 0.249 0.163 0.150 54.834 0.313 3.259 0.617 0.077 2.624 0.546 0.077 0.307 S.D 0.500 0.486 0.433 0.370 0.3578 15.878 0.464 7.398 0.487 0.267 3.622 0.499 0.267 0.462 Min 0 0 0 0 0 18 0 0 0 0 1.3 0 0 0 Max 1 1 1 1 1 91 1 20 1 1 12.5 1 1 1

Interested but not willing to pay Frequency N=636 Percentage of sample: 57.8% Mean 0.552 0.602 0.209 0.201 0.226 51.711 0.349 2.893 0.662 0.071 2.233 0.514 0.129 0.256 S.D. 0.498 0.490 0.407 0.401 0.419 15.854 0.477 7.041 0.473 0.257 3.098 0.500 0.335 0.437 Min 0 0 0 0 0 18 0 0 0 0 1.3 0 0 0 Max 1 1 1 1 1 93 1 20 1 1 12.5 1 1 1

Interested and willing to pay Frequency N=152 Percentage of sample: 13.8% Mean 0.553 0.533 0.191 0.243 0.230 0.520 3.158 0.684 0.099 2.479 0.533 0.184 0.164 S.D. 0.499 0.501 0.394 0.434 0.422 0.501 7.317 0.466 0.299 3.449 0.501 0.389 0.372 Min 0 0 0 0 0 18 0 0 0 0 1.3 0 0 0 Max 1 1 1 1 1 75 1 20 1 1 12.5 1 1 1

46.987 14.036

Table 2.1: Summary Statistics for the Explanatory Variables across Dependent Variable Categories

42

2.2.2 Methodology Given the ordinal nature of these differentiated responses, our model is intended to examine whether and how an unobservable level of interest in receiving personalized information about food recalls is associated with a consumers socio-economic characteristics, previous experience of foodborne illness, allergy conditions, and shopping behaviors. The observed dependent variable for this study is LOI, which describes the consumers differentiated engagement decision in the personalized recall notification service. It takes three possible distinct ordered choices as stated previously. The unobservable latent dependent variable LOI* is metrically represented by the observed dependent variable LOI. It is crucial to use a model of this kind because the difference between no interest and having interest but no willingness to pay may not be the same as that between having interest but no willingness to pay and having both interest and willingness to pay. Through the ordered probit, we can explore the different effects on the participation choice and payment choice from various explanatory variables. The ordered probit model is set up as the following form:

i : N (0,1)

LOI* = X i + i , where i = 1,L ,n i

where Xi is the vector of explanatory variables, is a vector of parameters associated with the corresponding explanatory variables and is not containing an intercept. LOI* is unobserved, but the relationship between LOI* and the observed variable LOI is as follows:

LOIi = 1 (or no interest), if LOI* 1 , i

43

LOI i = 2 (or having interest but not willing to pay), if 1 < LOI* 2 , i

LOIi = 3 (or having both interest and willing to pay), if LOI* > 2 . i where the parameters 1 and 2 are known as estimated threshold limit points. They are unknown and determined in the maximum likelihood estimation procedure for the ordered probit model (Daykin and Moffatt, 2002). The estimated parameters of the model are used to calculate the change in probability of a consumers level of interest in a personalized recall information service,
and their link to the independent variables one unit change. The mean X i of LOI*

depends on the explanatory variables contained in the vector X. In terms of explanatory variables, the model considered as the latent regression can be formulated as: LOI i = 0 GENDER + 1 PRIM + 2 SMELSE + 3 ALLER + 4 ILL + 5 AGE + 6 CHILD + 7 EDU + 8 MARR + 9 RACE + 10 INCOME + 11 EMPLOYF The + 12 EMPLOYP + 13 EMPLOYR independent variables explore various characteristics that may explain the different level of interest of consumers in personalized recall communications. Socio-demographic variables include AGE, CHILD, EDU, MARR and RACE. Variable AGE describes the respondents age, variable CHILD indicates whether the respondent is a caregiver for children or not, and is coded as 1 for yes and zero otherwise. Variable EDU records the respondents number of years of formal education. Marital status is categorized into two groups: married and unmarried but living with a partner (coded as 1) and unmarried (coded as 0). The unmarried category includes single, separated, divorced and widowed.

44

Variable GENDER PRIM SMELSE

Description 1=male, 0=female 1=I do most of the food shopping, 0 otherwise 1=someone else does most of the food shopping, 0 otherwise

Mean 0.5431 0.5976

Std Dev 0.4984 0.4906

Min. 0 0

Max. 1 1

0.2180

0.4131

ALLER

Do you or anyone in household have food allergies? 0.1962 0.3973 0 1

ILL AGE CHILD EDU MARR

Have you become ill b/c of eating contaminated food? Respondents age Are you a caregiver for any children? Years of formal education Current marital condition. 1=married or living with a partner, 0=single or living alone

0.2053 51.946 0.3624 3.0336

0.4041 15.794 0.4809 7.1775

0 18 0 0

1 93 1 20

0.6521

0.4765

HISP INCOME EMPLOYF EMPLOYP

1=Hispanic, otherwise 0 Yearly income scaled by 10,000 Full time employment, 1=yes, 0=no Part time employment, 1=yes, 0=no

0.0763 2.3783 0.5259 0.1217 0.2579

0.2656 3.3052 0.4996 0.3271 0.4377

0 1.3 0 0 0

1 12.5 1 1 1

EMPLOYR If the respondent is retired. 1=yes, 0=no

Table 2.2: Descriptive Statistics of Survey Data (sample size N=1,101) 45

Variable HISP is defined as 1 for Hispanic people and 0 otherwise. Economic status variables include INCOME and EMPLOYF, EMPLOYP and EMPLOYR. Unemployed people and those too ill to work are the base group, and EMPLOYF, EMPLOYP and EMPLYR describes people work full-time, part-time and those who have retired. The variable ILL describes prior experience with foodborne illness. The variable ALLER codes whether any person in the household has a food allergy and is set as 1 for yes and 0 otherwise. Table 2 provides an overview of consumers willingness to pay and of the demographic and socioeconomic characteristics.

2.3 Results and Discussions

The estimated results of the ordered logit model are presented in Table 3. Not all variables are significant in explaining the level of interest. The test of overall model fit is the LR statistic calculated by the following equation: LR = 2 * (ln L ln LR ) , where ln L indicates the maximized log-likelihood function given the models estimated parameters, and ln LR is the maximized log-likelihood function given that all of the estimated parameters are equal to zero. The LR statistics is 67.14 and the corresponding p-value is 0.0001 so the estimated model fits the data well. Prior to parameter interpretation, a typical consumer needs to be defined. Such a consumer is measured at the mean values of all explanatory variables. The probability of such a hypothetical consumer choosing to pay for the service is 12.8%, the probability of being interested in the service is 59.9% and the probability of showing no interested is 27.3%. The estimated parameters in the model give the logarithm of the odds of belonging to a particular LOI group, which is inconvenient in interpreting the estimated

46

parameter of each explanatory variable. Predicted probabilities are commonly used for interpretation when binary and continuous explanatory variables change, and thus are also estimated. The probabilities of consumers level of interest in a personalized recall communication service related to changes in the explanatory variables are shown in Table 2.4. Variable GENDER PRIM SMELSE ALLER ILL AGE CHILD EDU MARR HISP INCOME EMPLOYF EMPLOYP EMPLOYR 1 2 Coef. 0.124719 -0.55608 -0.65917 0.179156 0.292133 -0.01749 0.30419 -0.00286 0.114236 -0.03367 -2.45643 -0.50612 0.116571 -0.20361 -2.38915 0.508194 Std. Err. 0.135542 0.172381 0.194245 0.153857 0.151201 0.005378 0.144509 0.008547 0.137309 0.236162 1.904294 0.220266 0.267384 0.258118 0.368808 0.360201 z 0.92 -3.23 -3.39 1.16 1.93 -3.25 2.1 -0.33 0.83 -0.14 -1.29 -2.3 0.44 -0.79 P>z 0.357 0.001*** 0.001*** 0.244 0.053* 0.001*** 0.035** 0.738 0.405 0.887 0.197 0.022** 0.663 0.43 [95% Conf. Interval] -0.14094 -0.89394 -1.03989 -0.1224 -0.00422 -0.02803 0.020958 -0.01961 -0.15489 -0.49654 -6.18878 -0.93783 -0.40749 -0.70951 0.390377 -0.21822 -0.27846 0.48071 0.588481 -0.00695 0.587422 0.013895 0.383356 0.429196 1.275918 -0.0744 0.640633 0.302291

Model fit statistics: LR chi2(14)=67.14; Prob>chi2=0.0000; Log likelihood= -1010.1095. *p<0.1, **p<0.05, ***p<0.01 Table 2.3: Estimation Results of Consumer Interest (sample size N=1,101)

47

No interest in Variable GENDER PRIM SMELSE ALLER ILL AGE (-+1/2) (-+sd/2) CHILD EDU (-+1/2) (-+sd/2) MARR HISP INCOME (-+1/2) (-+sd/2) EMPLOYF EMPLOYP EMPLOYR

Interested but not the service

Interested and willing to pay for the service 1.3862% -6.483%*** -6.429%*** 2.0816% 3.4756%* -0.195%*** -3.085%*** 3.5047%** -0.032% -0.229% 1.2583% -0.372% -9.26% -0.906% -5.721%** 1.3439% -2.191%

notification service willing to pay for -2.483% 0.7463%*** 4.0689%*** -3.469% -5.569%* 0.3473%*** 5.4827%*** -5.922%** 0.0567% 0.4072% -2.286% 0.6731% 6.3031% 1.6123% 9.9718%** -2.268% 4.1315% 1.0964% -4.263%*** -7.64%*** 1.3869% 2.093%* -0.152%*** -2.397%*** 2.417%** -0.025% -0.178% 1.0279% -0.301% -7.043% -0.706% -4.25%** 0.9243% -1.941%

*p<0.1, **p<0.05, ***p<0.01 Table 2.4: Predicted Probabilities of Explanatory Variables on Consumers Interest in Notification Service (sample size N=1,101) Compared to consumers who share primary responsibility for food shopping, the probability that consumers who are the primary food shoppers for the household would be willing to pay for the personalized recall notification service decreases by 6.5%, the probability of simply being interested in this service drops by 4.3% and the likelihood of no interest by this consumer increases by 10.8%.

48

In contrast, compared to consumers who share primary responsibility for food shopping, the probability that consumers who are not the primary food shoppers for the household would be willing to pay for the personalized recall notification service decreases by 7.6%, the probability of simply being interested in this service drops by 6.4%, and the likelihood of no interest by this consumer increases by 14%. The effect of who shops in the household is statistically significant at the 95% confidence level. The consumers age is negatively associated with their interest and willingness to pay for a personalized recall notification service. As a consumers age increases by one year centered on the mean age 52, the probability of willingness to pay for the service is expected to decrease by 0.20 percent. The probability of having an interest in the service but no willingness to pay for it and having no interest in the service are anticipated to decrease by 0.15 percent and increase 0.35 percent, respectively, for an average American adult consumer. The results indicate that whether the consumer is a care-giver for children is positively associated with a greater level of interest in the personalized recall notification service. On average, compared to those who dont take care of any children, consumers who take care of at least one child would have a 3.5% higher probability of being willing to pay for the personalized recall notification service and a 2.4% higher probability of simply being interested in the service. The probability of having no interest falls 5.9% if the consumer doesnt have a child to take care of. Compared to their counterparts, the probability that consumers who have experienced foodborne illness are willing to pay for the notification service is estimated to increase by 3.5%, the probability of simply being interested in the service increases by 2.1% and the likelihood of no interest in this service

49

by this consumer decreases by 5.6%. These impacts are statistically significant at a 90% confidence level. Compared to consumers who are unemployed, the probability that consumers who are employed full-time would be willing to pay for the personalized recall notification service decreases by 5.7%, the probability of simply being interested in this service drops by 4.3% and the likelihood of no interest in the service by this consumer increases by 10%. The impact of being a full-time employee is statistically significant at a 95% level. It is interesting to notice that the unemployed people are more interested in the personalized notification service. It is possible that those unemployed respondents may be housewives who take care of the children and food preparation for the family. Therefore, in order to provide safe food to them, housewives could be more interested in the personalized recall notification service, since this service just offers a protection for the households food safety. So it may be reasonable for the unemployed to be more interested in the notification service. Another reason could be that those unemployed people are also the job seekers, and in order to find a job they need to keep healthy and not get ill. So this is understandable that they would be more interested in this notification service.

2.4 Conclusions

The results of the analyses suggest that the majority of American consumers would be interested in having food retailers notify them if a product they had purchased becomes part of food recall. However, relatively few appear willing to pay for such a service.

50

Consumers who are younger, those who are caregivers of children, have had experience with foodborne illness, share shopping responsibilities with someone else in their household and those who are unemployed appear most interested in, and willing to pay for, such a personalized recall notification service. Children are a key at risk group for foodborne illnesses, their immune systems are underdeveloped, and they can develop mild to severe illness, or even death due to contaminated foods. Prior experience with foodborne illness also appears logical. At present, the retailers who notify their customers that they have purchased a food product that becomes subject to a recall do not charge for such a service. Evidence of consumer interest (and WTP) provides insight into potential welfare gains from a broader adoption of such notification systems. However, we are not able to suggest the likely impact on public health. Questions for future research center on the costs and benefits for retailers and for consumers. Do such programs generate increased loyalty among consumers or added incentive to shop at the food retailers who offer them? What privacy and liability issues associated with using (or failing to use) loyalty card information to warn consumers when they have purchased a recalled product remain to be resolved (Shang and Hooker, 2008)? How could retailers leverage such systems for non-food recalls (e.g., pharmaceuticals or consumer durables)? Which communication channels are most appropriate to deliver such notifications? An analysis of the unintended consequences of such programs also remains to be conducted. If such programs became widespread and consumers began to rely on them for recall information, would they pay attention to more general information concerning food recalls? This is an important question, since some categories of food products,

51

including many fresh produce items are difficult to accurately identify when they are recalled.

2.5 Consumer Engagement Decision Analysis of Personalized Recall Notification Service 2.5.1 Methodology

The dependent variable for this study is the WTP amount for recall risk communications, in particular a personalized notification service that advises the consumers about recalls of previously purchased products. However, the analysis of the consumers willingness to pay for the notification service raises a challenging issue. 86.19% percent of the surveyed Americans reported zero payment for the service. If a direct regression was used to the truncated data, then it would result in biased and inconsistent estimates of the coefficients. In fact, the consumers decision to make a payment for the notification service follows a two step process: (I) the decision is made about whether to participate in the personalized recall notification service, called the participation choice; and (II) the decision of the amount of payment for this service consumption, called the payment level choice. Through two stage analysis, we can explore the different effects on the participation choice and payment level choice from possibly different explanatory variables. A double hurdle model proposed by Heckman (1979) is used to model consumers two stage decision making. This model has two stages of estimation used to characterize a two-step decision making process of consumer response to recall risk communications. This method takes into account the zero expenditure problem inherent in cross sectional data. The first step is to estimate the probability of participation in a

52

recall notification service. If this first step is ignored, only nonzero expenditure observations enter into the parameter estimation. Such sample selection bias leads to inconsistent estimates. By employing a two step procedure so the likelihood of the paying a nonzero amount for this service and the maximum willingness to pay for the service can be determined by different explanatory variables. The double hurdle model setup is as follows. In the first step, a Probit regression is used to estimate the probability of using the service. PAYi* = X i i + i , PAYi* = 1, if PAYi > 0, PAYi * = 0, if PAYi 0, where the dependent variable PAYi*is the latent variable, which is measured by a binary variable reflecting the decision to buy or not to buy the service. PAYi is the observed variable, Xi is the vector of explanatory variables, and i is the vector of coefficients associated with the corresponding explanatory variables. The second step is to estimate WTP using a truncated regression using the inverse Mills ratio to correct for sample selection bias. The model setup is: PAYi* = PAYi , if PAYi * > 0; PAYi * = 0, if PAYi * 0. ln PAYi * = Di i + i M i + i . If only the truncated sample is used for the second stage model, this leads to sample selection bias, and generates an omitted variable problem. Heckman (1979) proposed to use the inverse Mills ratio (IMR) in the second stage to correct sample

53

selection bias. We estimate the inverse of Mills ratio by calculating the ratio of the standard normal density function to the standard normal cumulative distribution function. The IMR (Mi) is used as a regressor in the second step estimation of the expenditure relations, and it is defined as M i = ( Xi) / ( Xi) where and are the standard normal density and cumulative probability functions, respectively. Previous research implied a nonnormal error distribution of the model could exist due to the truncated nature (Mauldin, Mimura, & Lino, 2001). Therefore, the dependent variable was transformed using the natural logarithm. Therefore, we explore the dependent variable through two stages of analysis considering its nature. The dependent variable for the first stage of the double hurdle model Probit model is PAYi*, whether the respondent is willing to pay for this service (coded 1 if yes, 0 if no). This is a latent variable and derived from the observed WTP amount PAYi. The observed dependent variable of the second stage model is the amount they are willing to pay for the service (per year in dollars).

2.5.2 Variables

The independent variables explore various characteristics which may explain the engagement of consumers in recall communications. Socio-demographic variables include AGE, CHILD, EDU, MARR and RACE. Variable AGE describes the respondents age, variable CHILD indicates if the respondent is a caregiver for children or not, and is coded as 1 for yes and zero otherwise. Variable EDU is about the respondents formal education years. Marital status is categorized into two groups:

54

married (coded as 1) and nonmarried (coded as 0). The nonmarried category includes single, separated, divorced and widowed. Variable RACE is defined as 1 for Hispanic people and 0 otherwise. Economic status variables include INCOME and EMPLOYF, EMPLOYP and EMPLOYR. Those employment variables are used to capture employment differences in the decision making process. The unemployed people and those too ill to work are the base group, and EMPLOYF, EMPLOYP and EMPLYR describes people work full time, part time and have retired. Knowledge variable is ILL, which describes prior experience with foodborne illness. Health status variable is regarding if there is any person having allergy in the household and variable ALLER is set as 1 for yes and 0 otherwise. Variables SHOPM and SHOPF capture the different effect from allocation of shopping tasks. Man does most of the food shopping, or woman does most of the food shopping, or the task is equally divided between them are taken into account. Variable SHOPM and SHOPF respectively describe man or woman does the food shopping in the household. Table 5 provides an overview of consumers willingness to pay and of the demographic and socioeconomic characteristics. On a national average, consumers are willing to pay $93 per year for the personalized recall notification service.

2.6 Results and Discussions 2.6.1 The First Stage Probit Regression Analysis

The results of the first stage probit regression model are presented in Table 6. Not all variables are significant in explaining participation decisions. Results indicate that whether the consumer is a care-giver for children is positively associated with

55

participation decision. On average, the consumer who takes care of any children would have a 29% higher probability to engage in the personalized service than the consumer who takes care of no children. Children are among one of the most vulnerable groups to food borne illnesses, as their immune systems are underdeveloped, and they can develop mild to severe illness, or even death due to contaminated foods. Age variable was also a significant factor affecting consumers participation decision. We find a negative relationship between them, indicating the likelihood of taking part in the notification service decreases as the consumers age increases. One possible explanation of this finding is that more spare time is available for the older people, and they can get the recall information through television, newspaper, radio and etc. The youths life is relatively tightly scheduled and notification service could save them more time and efforts to dig such recall information by themselves. The gender of the shopper is also another significant factor for participation decision. In our model, the base group is defined as the shopping task is equally divided by man and woman. Compared to this base group, the group where man is in charge of all shopping is less likely to join in the participation decision. The effect from the group where the female takes care of the shopping is statistically insignificant. These findings suggest that the female are more likely to participate in the notification service. Women take care of the household, and prepare foods at home, so they pay more attention to the quality of foods and food safety related information like recalls and thus it is consistent with the perceptions of the womens role in food shopping and preparation.

56

Variable ALLER ILL AGE CHILD EDU

Description Do you or anyone have food allergies? Have you become ill b/c of eating contaminated food? Respondents age Are you a caregiver for any children? Respondents formal education years Current marital condition. 1=married or living with a partner, 0=single or living alone 1=Hispanic, otherwise 0 Respondents interest to participate in the recall communication service? 1=yes, 0=no The most youd be willing to pay per year for the service Yearly income scaled by 1000,000 1=male does most of the food shopping, otherwise=0 1=woman does most of the food shopping, otherwise=0 Full time employment, 1=yes, 0=no Part time employment, 1=yes, 0=no If the respondent has retired. 1=yes, 0=no 57

N 1101 1101 1101 1101 1101

Mean 0.1962 0.2053 51.946 0.3624 3.0336

Std 0.3973 0.4041 15.794 0.4809 7.1775

Min 0 0 18 0 0

Max 1 1 93 1 20

MARR

1101

0.6521

0.4765

RACE

1101

0.0763

0.2656

PAY*

1101

0.1381

0.34512

PAY INCOME SHOPM SHOPF EMPLOYF EMPLOYP EMPLOYR

151 1101 1101 1101 1101 1101 1101

67.1719 197.0753 0.0238 0.2171 0.1717 0.5259 0.1217 0.2579 0.0331 0.4124 0.3773 0.4996 0.3271 0.4377

2000

0.013 0.125 0 0 0 0 0 1 1 1 1 1

Table 2.5: Descriptive Statistics of Recall Communication Survey Data

Parameter Variable Intercept ALLER ILL AGE CHILD EDU MARR RACE SHOPM SHOPF EMPLOYF EMPLOYP EMPLOYR Log Likelihood Estimate -0.5150 0.1078 0.0694 -0.0101 0.2944 0.0032 -0.0036 0.1002 -0.2301 -0.1748 -0.2658 0.03196 -0.1450

Standard Error 0.2689 0.1186 0.1182 0.0043 0.1107 0.0068 0.1144 0.1759 0.1311 0.1360 0.1675 0.1974 0.2025

Marginal Effect 0.0227 0.0146 -0.0021 0.0621 0.0007 -0.0008 0.0211 -0.0485 -0.0369 -0.0561 0.0067 -0.03059 t Value -1.91 0.91 0.59 -2.35 2.66 0.48 -0.03 0.57 -1.76 -1.29 -1.59 0.16 -0.72 Pr > |t| 0.0555 0.3635 0.557 0.0185 0.0079 0.6333 0.9752 0.5688 0.0792 0.1986 0.1125 0.8714 0.4741

-422.89114

Table 2.6: Estimation results of 1st stage

2.6.2 The Second Stage Truncated Regression Model

Table 7 presents the outcomes of the second stage truncated regression analysis. The IMR (Mi) is significant at the .05 level, indicating that sample selection bias has been solved after we consider the participation choice in the first stage probit regression model. This also indicates IMR is a significant variable in determining the amount of service payment. We notice the significant variables in this second stage are different from the outcomes of the first stage not matter in effect direction or variable names. This informs us that there does exist difference in the consumers two step decision making. The factors effecting

58

their participation decision is not the same as the factors influencing their payment level decision. Parameter Standard Variable Intercept ALLER ILL AGE CHILD EDU MARR RACE INCOME EMPLOYF EMPLOYP EMPLOYR IMR R-Square Estimate 6.4496 0.2231 -0.4633 0.0370 -0.6528 0.0101 -0.1243 -0.0131 -2.3573 1.0019 0.1877 0.4369 -3.3193 0.0809 Error 1.9970 0.2952 0.3021 0.0171 0.5035 0.0170 0.2708 0.4086 3.6903 0.5134 0.4529 0.4916 1.578

t Value 3.23 0.76 -1.53 2.16 -1.3 0.59 -0.46 -0.03 -0.64 1.95 0.41 0.89 -2.1

Pr > |t| 0.0015 0.4512 0.1274 0.0323 0.1969 0.5548 0.647 0.9745 0.524 0.053 0.6792 0.3757 0.0372

Table 2.7: Estimation results of 2nd stage Unlike the probit estimation results, consumer age is now positively associated with the payment level for recall notification service. The consumer pays more for the service for the older people. Although older people seem to have less likelihood to become participants of the personalized service, once they decided to join into it, they are willing to pay more for the service than the young customers. For every 1 year age increase, the consumer is willing to pay 3.7% premium for the service.

59

Those people who have become ill as a result of eating contaminated food are not statistically related with the payment level. In addition, this previous foodborne illness experience is insignificant in explaining the service expenditure; likewise, it is not significant in the first step participation choice. This may suggest that the people who have experienced foodborne illness have paid more attention to the food selection and preparation, and they are now confident that they are capable of dealing with the food recall situation. It is reported that the consumers who had foodborne illness experience would avoid the involved brand products and even the involved food manufacturers other products for a period of time. This may also explain their confidence. Considering the effect of employment on the payment level decision, compared to the base group of unemployed and too ill to work, full time employed people would like to pay more than one time amount for the service. People working full time not only have the ability to pay the service financially, but also have certain motivation to pursue health service leading to better food safety. Other variables seem to not have significant effect on the service payment level choice.

2.7 Comparison between Derived and Direct Participation Interest

In the interview questionnaire, a separate question about consumer participation interest was asked directly. The question is worded as Would you want YOUR grocery store to offer this service?, which captures the consumer interest in participating in grocery stores personalized services that alert consumers if a food product that they had already purchase had been recalled. Considering the possible difference from question wording on the effect of explanatory variables, the stated interest in notification service is used as

60

the dependent variable, named as variable (INTEREST) and code as 1 for interest and 0 otherwise. The explanatory variables keep the same as the first stage in the double hurdle model. Estimation results are presented in Table 8. Parameter Variable Intercept ALLER ILL AGE CHILD EDU MARR RACE SHOPM SHOPF EMPLOYF EMPLOYP EMPLOYR Log Likelihood Estimate 1.4622 0.1322 0.2334 -0.0100 0.0917 -0.0055 -0.0383 -0.0013 -0.2173 -0.1789 -0.3425 0.0448 -0.0692 Standard Error 0.2552 0.1112 0.1106 0.0038 0.1010 0.0058 0.0979 0.1667 0.1063 0.1154 0.1622 0.2030 0.1863 0.0397 0.0701 -0.0030 0.0276 -0.0017 -0.0115 -0.0004 -0.0653 -0.0537 -0.1029 0.0135 -0.0208 Marginal Effect t Value 5.73 1.19 2.11 -2.62 0.91 -0.95 -0.39 -0.01 -2.04 -1.55 -2.11 0.22 -0.37 Pr > |t| <.0001 0.2345 0.0349 0.0087 0.3639 0.3424 0.6959 0.9939 0.041 0.1213 0.0347 0.8253 0.7101

-587.073

Table 2.8: Estimation results of consumer participation interest Compared with the estimation results of the first stage in the double hurdle model, there are two differences. One is that a previous illness experience due to contaminated food is now a significant factor in consumers decision making to partake in the recall notification service. The consumer with such experience is more likely to having an interest in participating in the service. However, in the first stage of the double hurdle model this variable (ILL) is not significant although it has a positive relationship with the 61

dependent variable. Another difference is that those consumers employed full time are less likely to join in the notification service than those unemployed. One possible explanation might be that those consumers think they are capable of obtaining necessary information and sound judgment in dealing with recalled food. In addition, whether having children is not a significant factor in consumers interest in this risk communication service any more, while in the double hurdle model it is. Seeing the different estimation results, we calculated the correlation coefficient between these two dependent variables (INTEREST and WTP), and the coefficient is 0.2236. This implies that the association between consumers stated interest in participating in this notification service and the derived participation decision is not so big. When the consumer answered the question Would you want your grocery store to offer this service?, getting this service for no payment and for payment were two possible categories of offering methods in consumers thinking. Some consumers might think obtaining this service for free, and some might think they need to pay for it. This led to the low correlation between stated interest and derived interest in participation.

2.8 Conclusion

The primary objective of this study is to investigate the relationship of sociodemographic, economic, and health variables to the Americans recall notification service consumption decision trying to obtain an effective evaluation of consumers response to this service. An appropriate double hurdle model is applied to the nationally sampled survey data. The advantage of this methodology is that this model takes into account the inherent two step decision process to correct possible sample selection bias due to a large amount of zero

62

payment for the service. The findings inform us significant factors in consumers decision making. Thus these results have important implications for both the government agencies and food retailing industry. In particular, as women are more conscious of the food safety information and more willing to participate in the service, such service providers need to gear their marketing campaigns towards those female customers to get them more involved into the recall communication. The youth are more readily to accept new things are more likely to accept the innovative recall notification service, the service providers may target them as the main recipients, and the people in the old age range cannot be ignored since once they participate in the service they would pay more than the younger people. Whether the household has children is an important element to consider when popularizing the service. Children play a critical role in determining the parents engagement into the recall communication or not. The households with children need to be emphasized in the employment of this recall notification service, and they are naturally the key targeted receivers. Through analyzing these influencing factors of this recall notification service, when promoting such service, this research provides the food retailing industry and relevant parties with a better understanding of those significant variables, and thus they would deliver the services value more effectively and efficiently. Future research could explore consumers willingness to pay to specific or disaggregated food products. Since some categories of food products are more possible to develop food recall than others, such as salad vegetables, and explore the possibility of attaching a certain expenses of recall service to the retail price to ensure recall

63

information to reach the right customer even if this is only applicable in the group of membership customers. Further research could provide additional insights into the consumers engagement of this valuable service and provide more information to the government inspection agencies and food retail industry.

64

References

Cuite, C. L., Condry, S. C., Nucci, M. L., & Hallman, W. K. (2007). Public Response to the Contaminated Spinach Recall of 2006. (RR-0107-013). Rutgers, the State University of New Jersey, Food Policy Institute. Cuite, C. L., Schefske, S. D., Randolph, E.M., Hooker, N. H., Nucci, M. L., & Hallman, W. K. (2009). Public Response to the Salmonella Saintpaul Outbreak of 2008. (RR-1208-017). Rutgers, the State University of New Jersey, Food Policy Institute. Daykin, A. R. and Peter G. Moffatt (2002). Analyzing Ordered Responses: A Review of the Ordered Probit Model. Understanding Statistics. 1(3), 157-166. Food Marketing Institute (2008). Food Retailing Technology Benchmarks 2008. Food Marketing Institute: Arlington VA. Hallman, W. K., Cuite, C. L., & Hooker, N. H. (2009). Consumer Responses to Food Recalls: 2008 National Survey Report. (RR-0109-018). Rutgers, the State University of New Jersey, Food Policy Institute. Heckman, J. J. (1979). Sample Selection Bias as a Specification Error. Econometrica, 47(1), 153-161. Mauldin, T., Mimura, Y., & Lino, M. (2001). Parental Expenditures on Childrens Education. Journal of Family and Economic Issues, 22, 221-241. Patrick, M.E., Griffin, P.M., Voetsch, A.C., & Mead, P.S. (2007). Effectiveness of Recall Notification: Community Response to a Nationwide Recall of Hot Dogs and Deli Meats. Journal of Food Protection, 70(10), 23732376

65

Shang, W. and Neal H. Hooker, N. H. 2005. Improving Recall Crisis Management: Should Retailer Information be Disclosed? Journal of Public Affairs. 5, AugustNovember: pp. 329-341 Teratanavat, R., & Hooker, N. H. (2004). Understanding the Characteristics of U.S. Meat and Poultry Recalls: 1994-2002. Food Control. 15(5), 359-367. United States Government Accountability Office (2004). Food Safety: USDA and FDA Need to Better Ensure Prompt and Complete Recalls of Potentially Unsafe Food. GAO-05-51. United States Government Accountability Office: Washington DC.

66

Chapter 3 Stock Price Reactions to Pet Food Recalls

3.1 Introduction

The pet food market is a significant outlet for agricultural products in the United States. Pet food manufacturers use a wide range of inputs such as feed grains, oilseed mill products, meat, poultry, seafood and other agricultural and food ingredients. High pet ownership rates drive a large and growing market demand for pet food. Currently, 62% of US households own a pet according to figures from the National Pet Owners Survey conducted by the American Pet Product Manufacturers Association (APPMA). The pet food industry accounted for almost $14.7 billion annual revenue in 2004, and $16.2 billion in 2007 (APPMA 2004, 2008). The US pet food industry is highly concentrated; the fifty largest companies control almost 100 percent of the market (Hoovers, 2008). Indeed, the seven largest companies have a combined 86% market share (Barnes, 2005). Large companies include divisions of Nestle, Procter & Gamble, Colgate-Palmolive, and Del Monte Foods Inc. The top three companies Nestle Purina Petcare Company (Nestle), The Iams Company (Proctor & Gamble) and Hills Pet Nutrition (Colgate-Palmolive) hold a combined 54% market share (Barnes, 2005). Other major companies such as Mars, Inc., Wal-Mart and Del Monte account for 11, 10 and 7 percentage market shares respectively (Barnes, 2005). Regarding the pet foods sales percentage, Proctor & Gambles pet foods sales were 4537 million dollars in 2007, which accounted for 6% of the total sales of 76,476 million 67

dollars. For Del Monte Food Co., the sales of pet products were 1281.9 million dollars in the fiscal year 2007, and it accounts for 38% of the total sales revenues of 3,414.9 million dollars. Colgate Palmolive Co.s pet food sales reached 1859.1 million dollars in the fiscal year 2007, which accounts for 13% of the total sales revenues of 13,789.7 million dollars. In 2007 the pet food industry encountered a series of negative shocks due to product recalls. On March 15, 2007, the Food and Drug Administration (FDA) learned that certain pet foods were associated with illnesses and deaths of cats and dogs across the US. Late that afternoon FDA was first notified by Menu Foods, Inc. of the need for a possible recall. Late on March 16, 2007, Menu Foods voluntarily initiated recalls of cuts and gravy style dog and cat food produced at its facilities in Kansas and New Jersey, because their products were shown to contain tainted wheat gluten. Menu Foods as a contract manufacturer makes cat and dog foods for many other companies such as P&G (P&G Pet Care Press Release, 2007). In the following months, Menu Foods and other companies using common ingredient supply chains recalled additional brands and products. For instance, on March 16, 2007 Nestl Purina Pet Care Company voluntarily withdrew 5.3 ounce Mighty Dog pouch products produced by Menu Foods (Del Monte Press Release, 2007). According to FDAs online reports, these large scale recalls involved 13 pet food companies and more than 100 brands. On March 30, 2007 FDA laboratories found the contaminant melamine in samples of pet foods manufactured by Menu Foods and in the vegetable protein wheat gluten used as an ingredient in pet food (FDA Press Conferences, 2007). On May 25, 2007, FDA released a Safety/Risk Assessment on Interim Melamine and Melamine Analogues (FDA, 2007).

68

In order to assess the impact of recall information, media stories were tracked to determine the degree of coverage of announcements of these recalls. The method used by Piggott and Marsh (2004) to construct a food safety media announcement index was employed. Information was collected by searching for keywords food recall in the US Newspapers and Wires sources of the LexisNexis Academic database. Within the articles collected, the search was narrowed by searching for pet to obtain more precise information from pet food recall articles. Prior to 2007, pet food recall related articles were infrequent. Beginning in 2007, the number of media articles dramatically increased reaching a peak in the summer quarter of that year (see Figure 1). After that, the number of articles decreased but remained at a higher level than prior to 2007. The growth in the number of articles coincides with the time frame of the pet food recall events with media interest continuing several months after the recalls finished.

1200 1000 Number of Articles 800 600 400 200 0 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Year

Figure 3.1: Pet Food Recall Media Articles from 1998 to 2008. Once pet food recalls take place, firms incur economic losses, increased costs and possibly reduced revenues. Removing and disposing of contaminated and potentially 69

contaminated pet foods from distribution, negative media reports and potential reputation damage all may impose significant costs on firms. These may affect stock prices, shareholders perceptions and market valuation of the companies. However, the scale of such responses remains unclear. While several studies have evaluated market reaction to human foodborne illnesses, as yet no research has considered responses to pet food events. It is not clear if market reactions and valuations of risk would be larger or smaller, compared to human food safety recalls. The first purpose of this paper is to study quantitatively how stock prices reacted to these recalls. Based on stock market efficiency theory (Fama, 1970; 1991), prices are argued to quickly reflect all currently available information. If true, stock prices should promptly reflect the influence of pet food recalls. Daily stock prices can be used to quantify the effects of recalls on the value of a firm (MacKinlay, 1997). An event study method is employed to analyze the changes in firm value and shareholder wealth. If investor confidence and earning power of the pet food companies were adversely affected by the pet food recalls, then stock prices should present an efficient market response to such information. While the impact of pet food recalls on implicated companies has not been studied, there are several studies that explore the influence of human foodborne illness recalls. One category of research has investigated price changes of commodities responding to meat recalls. For example, McKenzie and Thomsen (2001) examined the impact of beef recalls due to E. coli O157:H7 contamination on wholesale and farm-level beef prices. They found negative reactions of boneless beef prices to recalls, no reaction in the live cattle prices and very little reaction in boxed beef prices. Lusk and Schroeder

70

(2002) examined the effect of meat recalls on live cattle and lean hog futures prices, and found that the announcement of recalls did not have a strong effect on either price series. Another category of studies, considering a capital market perspective, have examined the costs of food recalls, and indicated share prices react differently depending on the seriousness, size and other factors of the recalls. Salin and Hooker (2001) investigated different food recalls due to microbiological contamination, and show that returns to shareholders reduced in several cases, yet stock market reactions were not discernible in many incidents. Thomsen and McKenzie (2001) analyzed meat and poultry recalls from 1982 to 1998 and found that companies involved in serious food recalls induce shareholder losses whereas less serious recalls had no negative impact. Further, the initial food recall due to bacterial contamination incidents has been shown to be associated with reduced mean returns and higher volatility of the companies studied, yet repeated recalls did not have strong effects on the stock market (Wang et al., 2002). Golub, Wilson and Featherstone (2005) examine the effects of three Starlink corn food safety events on agribusiness firms and found no impact on the magnitude and volatility of the returns and risk of the firms. Additionally, Henson and Mazzocchi (2002) examined the impact on agribusinesses of the United Kingdom governments announcement of a possible link between Bovine Spongiform Encephalopthy and human health. They found significant negative abnormal returns in the beef, pet food, animal feed, and dairy sectors and positive abnormal returns in other meat sectors. Price volatility of stocks or commodities has also been analyzed from the perspective of specific events. Yang, Haigh and Leatham (1996) investigated the relationship between the Federal Agricultural Improvement and Reform (FAIR) Act of

71

1996 and agricultural commodity price volatility. Their findings suggest that FAIR led to an increase in the price volatility for corn, soybeans and wheat, and little change for oats, but a decrease in the volatility of cotton prices. In a study of Ghanas adoption of marketoriented government farm policy reforms, results indicate that in the years following the reforms, a reduction in maize price volatility occurred (Shively, 1996). Crain and Lee (1996) found changing volatility in wheat spot and futures markets was closely associated with changing government farm programs during the 1950-1993 period.

3.2. Event Study Methodology

To examine the market reactions to the 2007 events, this study focuses on those pet food firms that are publicly traded on US stock markets and were involved in the recalls. The pet food industry is highly concentrated, and not all the involved companies are publicly traded. Indeed only three firms meet all requirements. They are Del Monte Foods, Co., Procter & Gamble Co. and Colgate-Palmolive Company. The responses of share prices are first tested using event study methodology. Next, we describe the events of interest.

3.2.1 Three Publicly Traded Firms 3.2.1.1 Del Monte Foods, Co.

Del Monte Foods is one of the largest manufacturers of branded canned fruit, vegetables, tuna, and broths in the US. Its pet food division, Del Monte Pet Products, has a portfolio of pet food and treat brands, including 9Lives, Gravy Train, Milk-Bone, and Meow Mix. On March 31, 2007, immediately after learning that a supply of wheat gluten contained melamine, Del Monte Pet Products undertook a voluntary recall action of select product codes of its pet treat products sold under the Jerky Treats, Gravy Train

72

Beef Sticks and Pounce Meaty Morsels brands as well as select dog snack and wet dog food products sold under private label brands. The amount of affected products was less than one-tenth of one percent of Del Monte Pet Products' annual pet food and treat production. Del Monte Pet Products continued its cooperation with the FDA, and on April 6, 2007 it added additional select code dates in the US market to its voluntary recall list as a precautionary measure. No new brands were included in this second recall (Del Monte Press Release, 2007). Del Monte Foods, Co. is traded on the New York Stock Exchange (NYSE) the stock ticker is DLM. DLM1 denotes the first recall on March 31, 2007; DLM2 the second recall on April 6, 2007.

3.2.1.2 Procter & Gamble Co.

Menu foods, Inc., as a contract manufacturer for P&G Pet Care, produces a small portion of canned and foil pouch wet cat foods for P&Gs Iams and Eukanuba brands. Iams is among P&G's billion-dollar seller brands. On March 16, 2007, Menu Foods began pet food recalls in the United States. Due to the nationwide Menu recall of wet pet foods, on the same day, P&G Pet Care announced a voluntary recall in the United States and Canada of specific canned and small foil pouch wet cat and dog food products manufactured at the Menu Foods Emporia, Kansas plant. Procter & Gamble Co. is traded on the NYSE (PG is the ticker symbol). PG denotes the recall of P&G on March 16, 2007.

3.2.1.3 Colgate-Palmolive Company

Colgate-Palmolive's subsidiary Hill's Pet Nutrition, Inc. makes Science Diet and Prescription Diet pet foods. On March 16, 2007, Hill's Pet Nutrition announced voluntary participation in a nationwide recall of specific canned cat foods. Recalled products

73

included Select Science Diet Savory Cuts Cat Food. On March 30, 2007, Hills Pet Nutrition announced a second product recall. Prescription Diet m/d Feline Dry Food was recalled. Colgate-Palmolive Companys stock ticker is CL (NYSE). CL1 is the recall starting March 16, 2007 and CL2 March 30, 2007.

3.2.2 Measuring Normal and Abnormal Returns

In order to estimate the abnormal returns brought about by the pet food recall events, we first model normal returns without the recall information. There are two approaches to measuring normal performance, a constant mean return model and a market model. The simplest, constant mean return model often produces results similar to the market model while the later possesses an enhanced ability to detect event effects. Therefore, the market model is used in this research. As such the normal return of a stock is estimated by finding the relationship between the returns of the stock and a broad index of the market over the estimation window prior to the recall incidents. For any stock i the market model is (1) Rit = i + i Rmt + it where Rit and Rmt are the actual returns on stock i and market portfolio respectively at day t, and it is a zero mean error term. The parameter i measures the inherent factor from stock i per se, and i measures the influence from the market portfolio index. Among the popular broad based stock indices for the market portfolio, the Standard and Poors (S&P) 500 is used here. The daily closing share prices are downloaded from the yahoo finance website. Based on the adjusted closing prices, the daily stock return of each company is calculated 74

by (2) Rit =
pit pit 1 , pit 1

where pit, pit-1 are the daily closing price of stock i at day t and day t-1 respectively. The base market model is estimated using daily returns over a 75 day pre-event estimation window, omitting the 5 days prior to the event. This avoids the potential for information leakage before the formal announcement of the recall. Thus parameter estimations are not influenced by the returns around the event. i and i are estimated using data from t= - 80 to t= - 6. The estimation results for the five recalls are shown in Table 1.

Recall DLM1 DLM2 PG CL1 CL2

Constant -0.02186 0.01366 (0.09709) -0.01383 (0.0748) 0.04356 (0.06487) 0.03081 (0.05802)

i 0.5728 0.5897 (0.1373) 0.96052 (0.11605) 0.5594 (0.10063) 0.5175 (0.08201)

ti Value Pri>|t| Obs R-Square 4.06 4.3 8.28 5.56 6.31 0.0001 <.0001 <.0001 <.0001 <.0001 75 75 75 75 75 0.1845 0.2017 0.4841 0.321 0.3529

(0.09961)a (0.14094)

Table 3.1: Estimated Parameters of Normal Return from the Market Model

An abnormal return is defined as the difference between the actual return and its expected normal return during the event window. For stock i, the abnormal return is (3) ARit = Rit E [Rit X t ], where ARit is the abnormal return of stock i at day t, the conditional information Xt are 75

obtained by the parameters estimated by equation (1), and t starts from zero (event day). An abnormal return is computed for each recall. To examine the movements of abnormal returns, they are aggregated over different intervals of the event windows. This leads to a cumulative abnormal return (CAR), defined as (4) CARi (t1 , t 2 ) = ARit
t =t1 t2

Abnormal returns are typically aggregated over 5, 10, 20 and 30 days after the recall announcement. In addition to looking at returns, a similar event study process is applied to the volume of shares traded. This measure considers unusual or speculative levels of trading. The numbers of traded shares are examined both in the pre-75 day period and during the various post announcement periods, and are related to the trading volume of the S&P 500 Index in the normal return model. The Abnormal Volume of each stock is obtained by examining the difference between the actual number of shares traded and the predicted ones. To examine the movements of abnormal volumes traded, they are aggregated over different intervals of the event windows. Then a Cumulative Abnormal Volume (CAV) is calculated as the sum of the abnormal volume over different event windows starting from the event announcement day, described in equation 5. Abnormal volumes are aggregated over 5, 10, 20 and 30 days after the recall announcement to demonstrate how each firms number of shares traded change. (5) CAVi (t1 , t 2 ) = AVit
t =t1 t2

76

Although the recall event study literature has no previous application that describes CAVs, its use is common in studies of dividends, tax rates, and trading costs. For example, Koski and Scruggs (1998) investigated the determinants of ex-dividend returns and trading volume, and found a strong abnormal volume prior to the ex-dividend date could be interpreted as dividend-capture trading by taxable corporations. Timing clearly matters in any speculative venture; if there was a lot of speculation during the immediate post-event period, then a large volume of abnormal trading might suggest recall incidents had their impacts on the involved firms stock primarily through speculative activity rather than direct impacts on firm reputation. When combined with abnormal returns, abnormal volumes may provide insight into the price discovery process and thus a more precise measure of market response.

3.3. Event Study Results

Results of CAR are shown in Table 2 and Figure 2. Estimated daily abnormal returns for the total 11-day window centering on the event announcement day are listed in Table 3. Looking at the pre-event abnormal returns when t= - 5 to t= - 1, there was no information leaking for Del Monte Foods, Co.s first recall, possibly because there were no announcements of pet sickness or death linked to eating Del Monte pet foods. On the contrary, for the second recall of Del Monte, there is evidence of information leaking four days prior to the official announcement of the recall. Since the first recall happened on March 31 (Saturday), the stock market was closed. The next trading day was April 2, which was also the day the information leakage happened for the second recall. That is, the actual event day of the first recall for the stock market to react coincided with the information leakage for the second recall. We see a negative abnormal return (-1.943). 77

5-day DLM1 DLM2 PG CL1 CL2 -0.62672 0.45414 0.282596 0.810872 -1.23757

10-day 0.580854 1.290379 0.239104 0.758345 -2.03421

20-day -0.52343 -0.35642 -1.47443 -1.49214 -2.27984

30-day 0.57572 -0.70063 -4.11233 -1.98313 -3.09952

Table 3.2: CAR from Pet Food Recalls for Different Windows

2 1 0 dlm 1 CAR -1 -2 -3 -4 -5 Recall dlm 2 pg cl 1 cl 2 5-day 10-day 20-day 30-day

Figure 3.2: CAR from Pet Food Recalls for Different Windows.

78

Event Day Recall DLM1 DLM2 PG CL1 CL2


79

-5 0.0561 1.3146 -0.2821 0.4792 0.0197

-4 -0.072 -1.943 -0.3593 -0.2832 -0.3408

-3 0.2081 0.7049 0.4814 -0.0416 -0.4158

-2 0.1687 1.5279 -0.3098 -1.2714 0.9988

-1 1.3481 -0.8086 0.5472 -0.4031 -0.2857

0 -1.9031 -0.3136 -0.0996 -0.1245 -0.0468

1 0.7562 -0.2566 -0.4324 -0.2552 -1.2558

2 1.5653 0.019 -0.1139 0.2019 -0.5732

3 -0.7679 -0.4687 -0.5382 0.4853 0.1447

4 -0.2771 1.4741 1.4668 0.2856 0.4935

5 -0.2167 0.2345 -0.1717 -0.0412 0.0619

Table 3.3: Abnormal Returns on the Eleven Days Centering on the Event Announcement Day

79

The normal return estimation results from the market model tell us that all i are positive, statistically significant at the 95% level, and individual stock returns are positively related to the return on the broader market portfolio index. The association between stock value and the market index is greatest for P&G. The estimated values of i are very close to zero and none are statistically significantly different from zero. The R2 values are in acceptable ranges with an average of 0.309. Del Monte Foods, Co. stock experienced negative cumulative abnormal returns during the 5-day event window following the first pet food recall. The returns below the predicted levels indicate that the impact of this first recall went beyond the announcement day and expanded through the fifth day. The first recall announcement took place on March 31, 2007, which was 15 days from the first recall in the series of events. Due to the common contaminated supply source used in pet food manufacturing, other firms initiated recall actions as early as March 16. Shareholders may have had doubts about the safety controls throughout the pet food industry. In its second recall on April 6, 2007 Del Monte stock was accompanied by a positive cumulative abnormal return. This higher than forecast level suggests that the recall did not have a discernible adverse influence on Del Montes stock price. Several factors may explain this reaction. One is that recalled volume was lower than the first recall. The market may have responded with relief. Another factor is that the stock market may have learnt how to respond following Del Montes first recall. As Salin and Hooker (2001) indicated, stock market reaction to second recalls is often lower than that of the first recall. P&G company stock exhibited positive cumulative abnormal returns in the 5- and 10-day event windows. The fact that actual returns were above expected values may

80

suggest that P&G was unaffected by its pet food recall incident. This may be due to the fact that P&G was one of the companies who announced their voluntary recalls on the first day of the recall series. As the world's number one manufacturer of household products with a 12% market share of pet food sales in 2004, P&Gs instant response to potential contamination of a set of their pet foods appears to be information that shareholders appreciated. Further, the firm is highly differentiated and thus risks arising from such individual recalls are diversified. It is important to also note that when P&G announced the recall, P&G Pet Care had received no case reports involving dogs. Additionally, among all the firm press releases, P&Gs was the most detailed and included instructions about how to read can product code numbers. This may help explain why the share price was not impacted by the recall. After Colgate-Palmolive's first recall, its stock price increased (5- and 10-day event windows), just like P&G. These two firms held almost the same market share (12%) of pet food sales, and interestingly, they announced the recalls on the same day (March 16, 2007). When Colgate-Palmolives subsidiary Hill's Pet Nutrition, Inc., released the recall announcement, it had received no reported cases of illnesses for pets. Their announcement also stressed steps to protect the health and well-being of pets much like the P&G communications. Colgate-Palmolive's immediate action in the pet product recall and product diversification (enhancing risk management) may help explain why their stock prices experienced better-than-normal performance. However, unlike P&G, Colgate-Palmolive announced a second recall of Prescription Diet m/d Feline Dry Food on March 30, 2007. Their stocks cumulative abnormal returns exhibited substantial drops compared to the predicted normal returns. This second recall did have an adverse

81

impact on its stock price, and such influences extended more than 10 days after the second recall incident. Table 4 and Figure 3 report the results of cumulative abnormal volume for various event windows after the individual pet food recalls. For Del Monte, Co. CAVs were several million shares less than expected based on normal transaction volume prior to the pet food recall. This suggests little speculative activity during the various post recall periods. In general, the difference in volumes traded before and after a recall may be due to value buyers or over-pessimistic sellers. By comparison, in the 1996 Odwalla apple juice recall, on the event day there was 1.27 million extra shares traded with 1.17 million extra the next day compared to the average 30,000 shares (Salin and Hooker, 2001). Thus negative CAVs may imply that a large number of shareholders held stocks for this much more broadly traded firm. In the immediate 5-day window following the first recall there was a limited pessimistic response by the market.

5-day DLM1 DLM2 PG CL1 CL2 -340711 -1933030 3908650 -1457945 -627680

10-day -2109260 -2251192 243779.1 -3063773 -298755

20-day -3453615 -4136483 5493333 -3755272 2459643

30-day -6118950 -7050271 20048853 -964042 7309912

Table 3.4: CAV from Pet Food Recalls for Different Windows

82

25000000 20000000 15000000 10000000 5000000 0 dlm 1 -5000000 -10000000 Recall 5-day 10-day 20-day 30-day dlm 2 pg cl 1 cl 2

Figure 3.3: CAV from Pet Food Recalls for Different Windows.

CAV

P&Gs positive cumulative abnormal volumes and returns during the first 5 and 10 day windows indicate that its trades were not affected by its recall announcement and shareholders maintained confidence in the company and continued to invest in its stock. This effect cannot be separated from the fact that P&G was among the fastest responding companies who announced pet food recall on the first day of recalls. For Colgate-Palmolive, Co., during the first 5 and 10 day windows, there appeared no negative reactions from shareholders; they maintained normal transactions demonstrating their trust in Colgate-Palmolive possibly due to the release of product recall information. Immediately following the second recall, we did not see much speculation. 10 days later, the negative cumulative abnormal returns were accompanied by positive cumulative abnormal volumes. The second recall brought about a negative impact on its stock price and higher speculative activity.

83

3.4. Publicly Traded Firms not Involved in Pet Food Recalls

As discussed above, the pet food industry is highly concentrated among the leading firms, but otherwise highly fragmented. This suggests there are many small scale and privately owned firms (Hoovers, 2008). It is interesting therefore, to track the reaction of a pet food firm which is public but did not have a recall. United Pet Group (UPG) is one of few such firms. United Pet Group is a manufacturer of premium pet foods and supplies for cats, dogs, fish, and small animals. Pet food brands include Dingo and Eight in One. The company was acquired in August 2004 by United Industries which was bought by Spectrum Brands in early 2005. Thus through acquisitions, Spectrum has moved into pet foods and supplies. Wal-Mart accounts for around 19% of the firms sales. Spectrum gets 60% of its sales outside of the U.S. Hoovers, 2008 . Spectrum Brands, Inc., as UPGs ultimate parent company, is traded on the NYSE with ticker symbol SPC. The same method applied above to the three companies is used to examine Spectrum Brands, Inc.s stock price and trade volumes. Specifically, March 16, 2007 is set to be the event day in order to compare to other recall-involved firms. However, the results show a very different pattern. Its cumulative abnormal return over 5 and 10 days were -22.9 and -10.0, respectively, and -5.2 and -3.1 over 20 and 30 days. Its abnormal returns for a single day are shown in Figure 4. For the first 5 days after the pet food recall announcement, Spectrum Brands, Inc.s stock price experienced much larger decreases than the other three firms. There could be other reasons for such a large movement. For example, negative information was released on March 19, 2007 Spectrum Brands was to refinance a $1.48 billion credit facility underwritten by Goldman Sachs and Bank of America (Stein, 2007). This suggests Spectrum Brands was not performing well and "part 84

of that facility might almost be viewed as a bridge to the asset sale in the next three to six months." (Clouse, 2007) The CAVs reached 40.9 and 71.5 million in the first 5 and 10 day event windows and 0.14 and 0.23 billion in the 20 and 30 day event windows. The CAVs increased 70.7, 96.4 and 62.8 percent respectively between each event window. The large volumes of shares traded indicated there was a lot of speculative activity during the immediate post-event period and longer periods. This speculative activity coincides with the refinancing announcement as well.

3.5. Nestle Purina PetCare Company Analysis

Nestl Purina PetCare Company announced on March 16, 2007 that as a precautionary measure, it was voluntarily withdrawing its 5.3 ounce Mighty Dog brand pouch products produced by Menu Foods, Inc. in response to the recall initiated earlier that day by Menu Foods, a contract manufacturer that does limited business with Purina. Mighty Dog 5.3 ounce pouch products are being withdrawn by Nestl Purina, including those pouches contained in multi-packs. (Nestle Purina, 2007). Nestle Purina PetCare Company is a private firm, and its parental company Nestle S. A. is traded as a pink sheet stock in the U.S. Nestle stock is not available to U.S. investors in any other form. It has to be transacted by the broker-dealers. Its ticker symbol is NSRGY. As Nestle Purina Petcare is the worlds largest pet food company, it would be interesting to look at how its pink sheet returns changed. The same event study methodology is applied to Nestles pink sheet. March 16, 2007 is the event day for its recall announcement. Its cumulative abnormal return over 5 and 10 days were -8.39 and -4.17, respectively and -8.20 and 5.25 over 20 and 30 days. 85

Its abnormal return for March 16 was -8.13. We notice there is an obvious impact of the recall announcement on its pink sheet and the effects lasted for about 20 days. Since pink sheets are the shares of a company who does not meet the minimal criteria for capitalization and number of shareholders required by the NASDAQ and other exchanges, and the quote spreads tend to be steep, the changes of quotes of the pink sheets due to the recall crisis can be expected.

3.6. Discussions and Implications of Event Study

The 2007 series of pet food recalls is regarded as the largest recall of consumer products ever recorded in the United States (Nestle, 2008). During this crisis, shareholders revealed various reactions to different pet food company stocks. Company size and market share appear to influence the impact. Del Monte Foods, Co. is the smallest among the public pet food companies involved in this crisis. Returns to shareholders fell instantly after its first recall, while corresponding returns for P&G and Colgate-Palmolive did not reduce over the same event window. As relatively larger companies, P&G and Colgate-Palmolive have market shares about 70 percent more than Del Monte. In addition, stock markets reacted differently to repeated pet food recalls. The market demonstrated little tolerance for second recalls within a short period, particularly for the relatively large company. The Colgate-Palmolive second recall induced negative abnormal returns and cumulative abnormal returns which extended more than 30 days after the second recall announcement. As a leading company, repeated recalls revealed potential quality control problems, which in turn affected shareholders investment confidence. 86

Reaction time to possible pet food contamination incidents appears to be an important factor influencing firm performance. The fastest responding firms, P&G and Colgate-Palmolive, experienced no adverse impact on stock returns. Immediate attention and reaction from firm management and inclusion of detailed risk communications about pet safety and control measures appeared well received. Following the impact of competitors in the pet food market, financial markets were sensitive to relative performance. P&G and Colgate-Palmolive held similar market shares, their first recalls started on the same day, share prices did not respond to the recall announcements. However, the occurrence of a second recall for Colgate-Palmolive did lead to an adverse market reaction. In general, shareholders maintained confidence in the safety of American pet food products, any loss of public trust only appeared over a limited period. Certainly, contaminated pet food incidents sent out a signal that the pet food industry needs to engage in immediate and additional efforts to correct the undesirable consequences of recalls and prevent possible future contamination events.

3.7. Modeling Intra-day Stock Price Responses

Although daily stock prices reflect the adjustment of prices due to specific events, the use of intra-day prices may uncover additional price volatility. Intra-day stock price adjustments due to food recalls have not been explored to date. It is interesting to explore how the mean price and its volatility are influenced by news of recalls. Therefore, in this section, we examine empirical relationships between recall announcements and intra-day firm stock prices. The same three pet food firms are evaluated Del Monte Foods, Co., Procter & Gamble, Co. and Colgate-Palmolive, Co. 87

Building from results of the previous event study, this section further explores the window of 30 days before and after the 2007 recalls. Hourly Tick Data for the three firms and the hourly S&P 500 index price were collected. Dividends and stock splits have been incorporated in constructing these price series. The S&P 500 index again acts as a proxy for general trends in the stock market. The data cover the dates February 5th, 2007 and May 17th, 2007. Table 5 presents descriptive statistics of the hourly stock prices for the three firms. The stock price of Del Monte Foods, Co. (DLM), Procter & Gamble, Co. (PG) and Colgate-Palmolive, Co. (CL) are denoted as PDLM, PPG and PCL, respectively. The S&P 500 index is denoted as PSP. The mean stock prices for the three series differ, P&Gs price is most volatile in terms of its sample standard deviation, and the Del Monte price is the least volatile. We also calculated the volatility before and after the announcement of the recall with the date two day prior to the announcing date as the cutoff line. The volatilities of Colgate-Palmolive and P&G decreased from 0.81 to 0.58, and from 1.09 to 0.7, respectively comparing the two periods before and after the recall announcement, and the volatility of Del Monte stock prices increased a little bit from 0.2 to 0.21.

Variable PCL PDLM PPG PSP

Mean 66.94 11.54 63.23 1448.9

Std Dev 0.82 0.24 1.02 37.64

Minimum Maximum 64.94 10.93 60.68 1368.62 68.94 11.96 65.44 1515.1

Table 3.5: Summary Statistics of Stock Prices (Sample Size: 504 trading hours)

88

3.8. Generalized Autoregressive conditional Heteroscedasticity Method

A Generalized Autoregressive Conditional Heteroscedasticity GARCH (p, q) framework is applied to the hourly data. This model has been used to model time-varying patterns of stock or commodity prices (See Wang, et al., 2002; Yang, Haigh and Leatham, 2001). Here it is applied to examine the influence of pet food recalls on both mean prices and the volatility of those prices. The GARCH (p, q) model includes a mean equation and a conditional variance equation. The mean equation has added lags of the dependent variable, which account for possible serial correlation in successive stock prices. In the conditional variance equation, risk is captured by the variances of the forecast price errors. In the GARCH (p, q) model, p is the lag length of the squared residuals and q is the lag length for the conditional variance t2. In this work, we also use dummy variables to represent the various recalls, included in the GARCH model to test for firm-level impacts. The GARCH model is specified as the mean equation (6) and conditional variance equation (7), which is capable of capturing the volatility of stock prices of the investigated firms. (6)
pti = a + bmt + cpt 1,i + d 1 Rt1 + d 2 Rt2 + d 3 Rt3 + ti
p

2 (7) ti = 0 + j t2 j + j =1

k =1

k t k ,i
2

where pti is the period t price of stock i, mt is the Standard and Poors (S&P) 500 index, a is a constant, b is the risk of the firms stock, and t is a normally distributed error term with mean zero and variance t2. In addition, p is the lag length of the squared residuals and q is the lag length for the conditional variance t2. The models are estimated by maximizing log-likelihood functions.

89

2 We test for ARCH/GARCH effects. Using ti = 0 + j t2 j , we can test for the j =1

null hypothesis H 0 : There are no ARCH effects. The p-value is less than 5%, so the null can be rejected suggesting a time varying pattern of stock price variability. This justifies the application of a GARCH specification.

Coefficient
a

DLM 0.243 (0.143) 0.912** (0.021) 0.0005 (0.0001) -0.029** (4.98E-08) 0.002** (4.61E-05) 0.092** (0.03) -4.32E-18** (2.73E-13)

PG 0.994* (0.619) 0.984** (0.009) 6.80E-06** (0.0002) 0.068 (32908) 0.005** (0.001369) 0.122** (0.031) 0.787 (0.051)

CL 1.902** (0.790) 0.955** (0.013) 0.0008** (0.0002) -0.053 (1742) 0.029** (0.004) 0.209** (0.048) 0.1375* (0.084)

c b di ARCH0

Table 3.6: Parameter estimates of GARCH (1, 1) models (2.5.2007-5.17.2007, note ** indicates significant at 5%. * indicates significant at 10%.)

Table 6 presents parameter estimates of GARCH (1, 1) models. The Del Monte recall had a negative impact on its stock price, significant at a 5% level. This impact caused the stock price to drop $0.029. However, P&Gs recall announcement had no statistically significant influence on its stock price. The same is true of Colgate-Palmolive.

90

The mean equation of CL tells us that there was a negative but insignificant effect of $0.053 on its stock price. These estimation results are reasonable and consistent with the previous event study analysis. P&G was the first company to respond to the possible contamination at 5:30 PM on March 16, 2007; Colgate-Palmolive was the second company to announce a voluntary recall later that same day (the actual time of the press release was unspecified). Their immediate responses apparently provided investor confidence and the share price was not influenced. However, the announcement of Del Monte was 15 days later on March 31st, 2007 which may be viewed as a slow response and a sign of poor safety management and therefore influenced its stock price. The time-varying pattern of stock price variability was confirmed in the three firms cases. For Del Monte and Colgate-Palmolive both the coefficients of GARCH effects and are significant and for Colgate-Palmolive, is significant. The sum of
and measures the persistence of price volatility. For all the three companies, their

sum is less than one, which implies a finite volatility effect. The sum was 0.908 for P&G, suggesting a relatively slow decay, while Del Monte has a sum of 0.092 a relatively fast decay. This GARCH result is consistent with the findings that P&Gs price is most volatile in terms of its sample standard deviation and that Del Montes price is less volatile.

3.9. Seemingly Unrelated Regression System Estimation

A seemingly unrelated regression (SUR) model is next considered. We are estimating different dependent variables, using a common set of independent variables, thus the error terms may be correlated across a system of equations as the firms are part of the 91

same highly concentrated industry. The equations in a SUR model are related through contemporaneous correlations across the disturbance terms and are estimated jointly.

Variable in RCL equation Intercept RCLt-1 RSP CL CLB CLA EOD EOW

Parameter estimation of RCL equation -0.009 0.014 -0.227** 0.040 0.447** 0.059 0.033 0.296 0.153 0.306 0.108 0.211 0.025 0.041 0.032 0.091

Variable in RDLM equation Intercept RDLMt-1 RSP DLM DLMB DLMA EOD EOW

Parameter estimation of RDLM equation 0.010 0.022 -0.278** 0.042 0.473** 0.092 -0.286 0.462 0.117 0.333 -0.354 0.328 -0.012 0.065 -0.008 0.140

Variable in RPG equation Intercept RPGt-1 RSP PG PGB PGA EOD EOW

Parameter estimation of RPG equation -0.018 0.015 -0.170** 0.040 0.669** 0.061 0.140 0.303 -0.196 0.314 -0.116 0.216 0.035 0.042 0.051 0.093

Table 3.7: Estimation Results of Seemingly Unrelated Regression

This increases the efficiency of the estimations. The model system is set up as follows: (8)

92

RPGt = 0 + 1 RPGt 1 + 2 RSPt + 3 PG + 4 PGB + 5 PGA + 6 EOD + 7 EOW + w pt


RDLM t = 0 + 1 RDLM t 1 + 2 RSPt + 3 DLM + 4 DLMB + 5 DLMA + 6 EOD

(9) + EOW + w dt 7 (10)

RCLt = 0 + 1 RCLt 1 + 2 RSPt + 3 CL + 4 CLB + 5 CLA + 6 EOD + 7 EOW + wct

The dependent variable is the stock return of each firm at time t, RPGt-1 is the one period lag, and RSPt is the return on S&P 500 index, and PG, DLM, and CL are dummy variable representing the various recall announcements. PGB, DLMB and CLB are equal to one for the two hours prior to the recall announcement hour for the three firms. PGA, DLMA and CLA are set to one for the two hours after the announcement hour. Variable EOD measures the end of day effect. If the trading happens at the end of day, then it is coded as 1, otherwise 0. Variable EOW measures the end of week effect. If the trading takes place at the end of a week, then it is coded as 1, otherwise 0. Some literature (Dimson, 1988; Chan, Chui and Knok, 2001; Dicle and Hassan, 2007)indicate that there were end of day and end of week effect, so we include these two dummy variables to examine if there is any effect in this case. The estimation results are presented in Table 7. The results inform us that the S&P 500 index had a significant effect on ColgatePalmolive, Del Monte and P&G stock returns. It appears that the returns of the three firms were influenced by the general performance of the stock market during the time period analyzed. Regarding the effect of the recall announcements, the Del Monte recall had negative effects on their own stock prices. However, P&Gs and Colgates stock prices were not negatively impacted by their recall announcements. Since the P&G recall announcement had the most immediate response, followed by Colgate-Palmolives

93

announcement at the same day, while the Del Monte recall notification was 15 days later after the first voluntary recall, and their stock price dropped -0.030. The dummy variables denoting two hours before and after each recall suggest there was no significant information leakage or delay in the stock market for the series of recall events. The timely publicity about contaminated pet food from P&G left no space for information leakage, the broad media attention ensued that shareholders had sufficient information about the recall. Estimation results indicate that end of day effect and end of week effect were not statistically significant.

3.10. Model Comparisons

The event study methodology, GARCH model and SUR estimation each enrich our understanding of the impact of the pet food recalls on equity valuations. The event study methodology informs us through the cumulative abnormal returns of a broad impact of firm value. Also by examining pre-event abnormal returns, we can determine if there was information leakage. Del Monte stock experienced negative cumulative abnormal returns during the 5-day event window following the first pet food recall and positive cumulative abnormal returns following the second recall. P&G stock was unaffected by its pet food recall. Colgate-Palmolive stock was also not negatively impacted by its first recall, but negatively affected by its second recall. Event study procedures provide a direct reaction of financial market to food recalls, but the aggregated nature of CAR and CAVs may miss immediate price (or quantity) changes and fail to consider the heteroskedasticity properties of the time series of the residuals. Hourly data and the GARCH model help resolves these issues, here the

94

models verify the findings from event study method that the announcement of recalls had a significant negative effect on Del Montes stock price, and had no influence on P&G and Colgate-Palmolive stock price. A short-lived but important pattern of stock price volatility was also confirmed. Finally, as the three firms are considered in the same highly concentrated industry, they are assessed using a simultaneous system to take into account the contemporaneous correlations across the disturbance terms.

95

References

American Pet Products Association (APPA). (2004). 2003-2004 APPMA National Pet Owners Survey. Greenwich, CT. American Pet Products Association (APPA). (2008). 2007-2008 APPMA National Pet Owners Survey. Greenwich, CT. Barnes, N. G. (2005). A market analysis of the US pet food industry to determine new opportunities for the cranberry industry. Center for Business Research, University of Massachusetts Dartmouth. Chan, Y. C., Chui C.W. & Kwok C.Y. (2001). The impact of salient political and economic news on the trading activity. Pacific-Basin Finance Journal, 9, 195-217. Crain, S. J., & Lee J. H. (1996). Volatility in wheat spot and futures markets, 1950-1993: government farm programs, seasonality, and causality. The Journal of Finance, 51, 325-343. Del Monte Press Release. (March 31, 2007). Del Monte Pet Products Voluntarily Withdraws Specific Product Codes of Pet Treats and Wet Dog Food Products. San Francisco. Retrieved November 16, 2007, from http://www.fda.gov/oc/po/firmrecalls/delmonte03_07.html Dicle, M. F. & Hassan M. K. (2007). Day of the week effect in Istanbul stock exchange.
Scientific Journal of Administrative Development, 5, 53-83.

Dimson, E. (1998). Stock market anomalies. Cambridge University Press. Fama, E. F. (1970). Efficient capital markets: a review of theory and empirical work. The
Journal of Finance, 25, 383-417.

___. (1991). Efficient capital markets: II. The Journal of Finance, 46, 1575-1617.

96

Golub, A., Wilson C. A., & Featherstone A. M. (2005). The effects of starlink corn food safety events on returns and risk of agribusiness firms. Working paper, Department of Agricultural Economics, Purdue University. Henson, S. H., & Mazzocchi M. (2002). Impact of bovine spongiform encephalopathy on agribusiness in the United Kingdom: results of an event study of equity prices. American Journal of Agricultural Economics, 84, 370-86. Hoovers Online. (2008). Industry Profile: Pet Food Manufacture. Hoovers North America. Lusk, J. L., & Schroeder T. C. (2002). Effects of meat recalls on futures market prices. Agricultural and Resource Economics Review, 1, 47-58. Mackinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature, XXXV, 13-39. McKenzie, A. M., & Thomsen M. R. (2001). The effect of E. Coli 0157:H7 on beef prices. Journal of Agricultural and Resource Economics, 2, 431-444. Nestle, M. (2008). Pet food policies: the Chihuahua in the coal mine. Berkeley: University of California Press. P&G Pet Care Press Release. (March 16, 2007). P&G pet care announces voluntary participation in Menu Foods nationwide U.S. and Canadian recall of specific canned and small foil pouch wet cat and dog foods. Dayton, OH. Retrieved November 16, 2007, from http://www.fda.gov/oc/po/firmrecalls/pg03_07.html Piggott, N. E., & Marsh T. L. (2004). Does food safety information impact U.S. meat demand? American Journal of Agricultural Economics, 1, 154-174.

97

Salin, V., & Hooker N. H. (2001). Stock market reaction to food recalls. Review of Agricultural Economics, 23, 33-46. Shively, G. E. (1996). Food price variability and economic reform: An ARCH approach for Ghana. American Journal of Agricultural Economics, 78, 126-136. Thomsen, M. R., & McKenzie A.M. (2001). Market incentives for safe foods: An examination of shareholder losses from meat and poultry recalls. American Journal of Agricultural Economics, 3, 526-538. U.S. Food and Drug Administration (FDA). (March 20, 2007). Transcript of FDA press conference on the pet food recall. Maryland. Retrieved November 16, 2007, from http://www.fda.gov/downloads/NewsEvents/Newsroom/MediaTranscripts/ucm12 3638.pdf Wang, Z. J., Salin V., Hooker N. H., & Leatham D. (2002). Stock market reaction to food recalls: A GARCH application. Applied Economics Letters, 9, 979-987. Yang, J., Haigh M. S., & Leatham D. J. (2001). Agricultural liberalization policy and commodity price volatility: A GARCH application. Applied Economics Letters, 8, 593-598.

98

Bibliography

American Pet Products Association (APPA). (2004). 2003-2004 APPMA National Pet Owners Survey. Greenwich, CT. American Pet Products Association (APPA). (2008). 2007-2008 APPMA National Pet Owners Survey. Greenwich, CT. Australia New Zealand Food Authority. 1995. Code of Practice on Nutrient Claims in Food Labels and in Advertisements. Available at http://www.foodstandards.gov.au/_srcfiles/Code_of_Practice_jan1995.pdf (accessed December 2008). Baker, G.A. 1999. Consumer Preferences for Food Safety Attributes in Fresh Apples: Market Segments, Consumer Characteristics, and Marketing Opportunities.
Journal of Agricultural and Resource Economics 24(1): 80-97.

Banterle, A., L. Baldi, and S. Stranieri. 2008. Do Nutritional Claims Matter to Consumers? An Empirical Analysis Considering European Requirements. Paper presented at the 8th International Conference on Management in AgriFood chains and Networks, Ede-Wageningen, The Netherlands. Barnes, N. G. (2005). A market analysis of the US pet food industry to determine new opportunities for the cranberry industry. Center for Business Research, University of Massachusetts Dartmouth.

99

Boland M., and T. Schroeder. 2002. Marginal Value of Quality Attributes for Natural and Organic Beef. Journal of Agricultural and Applied Economics 34(1): 39-49. Bren, L. 2001. Antibiotic Resistance From Down On the Chicken Farm. FDA Veterinarian Vol. XVI, No.I. U.S. Food and Drug Administration, Washington, D.C. Buzby, J.C., P.D. Frenzen, and B. Rasco. 2001. Product Liability and Microbial Foodborne Illness. Economic Research Service, U.S. Department of Agriculture, Washington, D.C. Canadian Food Inspection Agency. 2008. Acts and Regulations. Available at http://www.inspection.gc.ca/english/reg/rege.shtml (accessed December 2008). CDHS-FDA. 2007. Investigation of an Escherichia coli 0157:H7 Outbreak Associated with Dole Pre-Packaged Spinach. Washington, D.C.: California Food Emergency Response Team, California Dept of Health Services, and FDA-CFSAN. Available at: http://www.marlerclark.com/2006_Spinach_Report_Final_01.pdf (accessed December 9, 2008). Chan, Y. C., Chui C.W. & Kwok C.Y. (2001). The impact of salient political and economic news on the trading activity. Pacific-Basin Finance Journal, 9, 195-217. Clark, B.T. 2000. Attorney at Marler Clark. Personal Correspondence. Crain, S. J., & Lee J. H. (1996). Volatility in wheat spot and futures markets, 1950-1993: government farm programs, seasonality, and causality. The Journal of Finance, 51, 325-343.

100

Cuite, C. L., Condry, S. C., Nucci, M. L., & Hallman, W. K. (2007). Public Response to the Contaminated Spinach Recall of 2006. (RR-0107-013). Rutgers, the State University of New Jersey, Food Policy Institute. Cuite, C. L., Schefske, S. D., Randolph, E.M., Hooker, N. H., Nucci, M. L., & Hallman, W. K. (2009). Public Response to the Salmonella Saintpaul Outbreak of 2008. (RR-1208-017). Rutgers, the State University of New Jersey, Food Policy Institute. Daykin, A. R. and Peter G. Moffatt (2002). Analyzing Ordered Responses: A Review of the Ordered Probit Model. Understanding Statistics. 1(3), 157-166. Del Monte Press Release. (March 31, 2007). Del Monte Pet Products Voluntarily Withdraws Specific Product Codes of Pet Treats and Wet Dog Food Products. San Francisco. Retrieved November 16, 2007, from http://www.fda.gov/oc/po/firmrecalls/delmonte03_07.html Dicle, M. F. & Hassan M. K. (2007). Day of the week effect in Istanbul stock exchange.
Scientific Journal of Administrative Development, 5, 53-83.

Dimson, E. (1998). Stock market anomalies. Cambridge University Press. Estes E.A., and V.K. Smith. 1996. Price, Quality, and Pesticide Related Health Risk Considerations in Fruit and Vegetable Purchase: An Hedonic Analysis of Tucson, Arizona Supermarkets. Journal of Food Distribution Research 27: 59-76. Fama, E. F. (1970). Efficient capital markets: a review of theory and empirical work. The
Journal of Finance, 25, 383-417.

___. (1991). Efficient capital markets: II. The Journal of Finance, 46, 1575-1617. Fleming, K., E. Thorson & Y. Zhang. 2006. Going Beyond Exposure to Local News

101

Media: An Information-processing Examination of Public Perceptions of Food Safety. Journal of Health Communication, 11(8): 789-806. Food Safety and Inspection Service. 2006. Product Labeling: Definition of the Term Natural. Food Safety and Inspection Service, U.S. Department of Agriculture, Washington, D.C. Available at http://www.fsis.usda.gov/OPPDE/rdad/FRPubs/2006-0040.htm (accessed December 2008). Food Safety Authority of Ireland. 2008. Legislation. Available at http://www.fsai.ie/legislation.html (accessed December 2008). Food Standards Agency. 2008. Labelling Guidance. Available at http://www.food.gov.uk/foodindustry/guidancenotes/labelregsguidance/ (accessed December 2008). Fox, J. 2000. Multiple and Generalized Nonparametric Regression. Sage Publications, Inc. Feenstra, R.C. 1995. Exact Hedonic Price Indexes. The Review of Economics and
Statistics 77(4): 634-653.

Food Marketing Institute (2008). Food Retailing Technology Benchmarks 2008. Food Marketing Institute: Arlington VA. Gold, M.V. 2007. Organic Production/Organic Food: Information Access Tools. Alternative Farming Systems Information Center, U.S. Department of Agriculture, Washington, D.C. Available at http://www.nal.usda.gov/afsic/pubs/ofp/ofp.shtml (accessed December 2008).

102

Golub, A., Wilson C. A., & Featherstone A. M. (2005). The effects of starlink corn food safety events on returns and risk of agribusiness firms. Working paper, Department of Agricultural Economics, Purdue University. Hallman, W. K., Cuite, C. L., & Hooker, N. H. (2009). Consumer Responses to Food Recalls: 2008 National Survey Report. (RR-0109-018). Rutgers, the State University of New Jersey, Food Policy Institute. Heckman, J. J. (1979). Sample Selection Bias as a Specification Error. Econometrica, 47(1), 153-161. Henson, S. H., & Mazzocchi M. (2002). Impact of bovine spongiform encephalopathy on agribusiness in the United Kingdom: results of an event study of equity prices. American Journal of Agricultural Economics, 84, 370-86. Hooker, N. H. and B. Roe. 2002. The Role of Food Safety Systems in Cattle Marketing. Managing for Todays Cattle Market and Beyond. 2nd Edition. Available online at http://www.ag.arizona.edu/arec/wemc/cattlemarket/CattleMarketing&Food_Safet y.pdf (accessed December 2008). Hooker, Neal H. and R. P. Teratanavat. 2008. Dissecting Qualified Health Claims: Evidence from Experimental Studies. Critical Reviews in Food Science and
Nutrition. 48(2): 160-176.

Hoovers Online. (2008). Industry Profile: Pet Food Manufacture. Hoovers North America. Huang, C.L., and B.H. Lin. 2007. A Hedonic Analysis of Fresh Tomato Prices among Regional Markets. Review of Agricultural Economics 29(4): 783-800.

103

Jordan, J. L., R.L. Shewfelt, S.E. Prussia, and W.C. Hurst. 1985. Estimating Implicit Marginal Prices of Quality Characteristics of Tomatoes. Southern Journal of
Agricultural Economics. 17: 139-146.

Kimberlin, C.L., and A.G. Winterstein. 2008. Expert and Consumer Evaluation of Consumer Medication Information-2008. Final Report to the U.S. Department of Health and Human Services and the Food and Drug Administration. Lecocq, S. and M. Visser. 2006. What Determines Wine Prices: Objective vs. Sensory Characteristics. Journal of Wine Economics 1(1): 42-56. Loureiro M.L., and J.J. McCluskey. 2000. Assessing Consumer Response to Protected Geographical Identification Labeling. Agribusiness 16 (3): 309-320. Lusk, J. L., & Schroeder T. C. (2002). Effects of meat recalls on futures market prices. Agricultural and Resource Economics Review, 1, 47-58. Mackinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature, XXXV, 13-39. Maguire K.B., N. Owens, and N.B. Simon. 2004. The Price Premium for Organic Babyfood: A Hedonic Analysis. Journal of Agricultural and Resource
Economics 29(18): 132-149.

Mauldin, T., Mimura, Y., & Lino, M. (2001). Parental Expenditures on Childrens Education. Journal of Family and Economic Issues, 22, 221-241. McCluskey, J.J. 2000. A Game Theoretic Approach to Organic Foods: An Analysis of Asymmetric Information and Policy. Agricultural and Resource Economics
Review 29(1): 1-9.

104

McKenzie, A. M., & Thomsen M. R. (2001). The effect of E. Coli 0157:H7 on beef prices. Journal of Agricultural and Resource Economics, 2, 431-444. Melton, B.E., W.E. Huffman, J.F. Shogren, and J.A. Fox. 1996. Consumer Preferences for Fresh Food Items with Multiple Quality Attributes: Evidence from an Experimental Auction of Pork Chops. American Journal of Agricultural
Economics 78(4): 916-923.

Nestle, M. (2008). Pet food policies: the Chihuahua in the coal mine. Berkeley: University of California Press. Nimon, W. and J. Beghin. 1999. Are Eco-Labels Valuable? Evidence from the Apparel Industry. American Journal of Agricultural Economics 81(4): 801-811. Onyango, B., R.M. Nayga, Jr., and R. Govindasamy. 2006. U.S. Consumers Willingness to Pay for Food Labeled Genetically Modified. Agricultural and
Resource Economics Review 35(2): 299310.

Osborne, L. and V.K. Smith. 1997. Environmental Amenities, Product Differentiation, and market Power. Mimeo. Palmquist, R.B. 1984. Estimating the Demand for the Characteristics of Housing.
Review of Economics and Statistics 66(3): 394-404.

P&G Pet Care Press Release. (March 16, 2007). P&G pet care announces voluntary participation in Menu Foods nationwide U.S. and Canadian recall of specific canned and small foil pouch wet cat and dog foods. Dayton, OH. Retrieved November 16, 2007, from http://www.fda.gov/oc/po/firmrecalls/pg03_07.html

105

Patrick, M.E., Griffin, P.M., Voetsch, A.C., & Mead, P.S. (2007). Effectiveness of Recall Notification: Community Response to a Nationwide Recall of Hot Dogs and Deli Meats. Journal of Food Protection, 70(10), 23732376 Piggott, N. E., & Marsh T. L. (2004). Does food safety information impact U.S. meat demand? American Journal of Agricultural Economics, 1, 154-174. Rasco, B. 1997. How HACCP Can Help in a Products Liability Action. Food Quality (October): 16-18. Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. The Journal of Political Economy 82(1): 34-55. Salin, V., & Hooker N. H. (2001). Stock market reaction to food recalls. Review of Agricultural Economics, 23, 33-46. Shang, W. and Neal H. Hooker, N. H. 2005. Improving Recall Crisis Management: Should Retailer Information be Disclosed? Journal of Public Affairs. 5, AugustNovember: pp. 329-341 Shively, G. E. (1996). Food price variability and economic reform: An ARCH approach for Ghana. American Journal of Agricultural Economics, 78, 126-136. South Africa, Department of Health. 2008. Legislation-Acts. Available at http://www.doh.gov.za/docs/index.html (accessed December 2008) Steiner B.E. 2004. Australian Wines in the British Wine Market: A Hedonic Price Analysis. Agribusiness 20(3):287-307. Taylor M.R., and G.W. Brester. 2005. Noncash Income Transfers and Agricultural Land Values. Review of Agricultural Economics 27(16): 526-541.

106

Teratanavat, R., & Hooker, N. H. (2004). Understanding the Characteristics of U.S. Meat and Poultry Recalls: 1994-2002. Food Control. 15(5), 359-367. The Food Labelling Regulations. 1996. Statutory Instrument 1996 No. 1499 Available at http://www.opsi.gov.uk/si/si1996/Uksi_19961499_en_1.htm (accessed December 2008). Thomsen, M. R., & McKenzie A.M. (2001). Market incentives for safe foods: An examination of shareholder losses from meat and poultry recalls. American Journal of Agricultural Economics, 3, 526-538. United States Government Accountability Office (2004). Food Safety: USDA and FDA Need to Better Ensure Prompt and Complete Recalls of Potentially Unsafe Food. GAO-05-51. United States Government Accountability Office: Washington DC. U.S. Food and Drug Administration. 2005. Department of Health and Human Services Food and Drug Administration Fiscal Year 2005. FDA, Washington, D.C. Available at: http://www.fda.gov/AboutFDA/ReportsManualsForms/Reports/BudgetReports/20 05FDABudgetSummary/ucm117926.htm (accessed December 9, 2008). U.S. Food and Drug Administration (FDA). (March 20, 2007). Transcript of FDA press conference on the pet food recall. Maryland. Retrieved November 16, 2007, from http://www.fda.gov/downloads/NewsEvents/Newsroom/MediaTranscripts/ucm12 3638.pdf ________. 2008. Food Label Helps Consumers Make Healthier Choices. FDA, Washington, D.C. Available at:

107

http://www.fda.gov/ForConsumers/ConsumerUpdates/ucm094536.htm (accessed December 6, 2009). Wahlqvist M.L., and M. Ball. 2002. Food and Nutrition: Australasia, Asia and the
Pacific. Australia: Allen & Unwin.

Wang, Z. J., Salin V., Hooker N. H., & Leatham D. (2002). Stock market reaction to food recalls: A GARCH application. Applied Economics Letters, 9, 979-987. Wilson, W. 1984. Hedonic prices in the malting barley market. Western Journal of
Agricultural Economics 9(1): 29-40.

World Health Organization. 2002. Foodborne Disease. Document proposed at FAO/WHO Global Forum of Food Safety Regulators, Marrakech, Morocco. Yang, J., Haigh M. S., & Leatham D. J. (2001). Agricultural liberalization policy and commodity price volatility: A GARCH application. Applied Economics Letters, 8, 593-598.

108

Anda mungkin juga menyukai