Putting the Horses Before the Cart: Harnessing the Power
of Partner Brands
Pianpian Kong, Paul B. Ellickson, Mitchell J. Lovett⇤
JOB MARKET PAPER
Preliminary and in Progress
Current Version: 7/11/2016
⇤Pianpian Kong is a doctoral student, Paul Ellickson is Professor of Economics and Marketing, and Mitchell Lovett is Associate Professor of Marketing, all at Simon Business School, University of Rochester. We thank Ron Goettler, Avery Haviv, Yufeng Huang, Guy Arie, and participants at the Simon Marketing group seminar for their comments and suggestions. All errors are our own. Please direct any correspondence to [email protected].
1 Abstract
In many tied good industries, variety of complementary products drives adoption of the platform. In the brewer-coffee industry, as the number of K-Cup coffee brand variety increased from 10 to more than 150 from 2010 to 2015, the installed base of Keurig K-Cup brewers expanded from 11% to 31%. The K-Cup coffee consumables grew from $0.3 billion annual sales to over $3.6 billion, comprising 41% of total dollar sales for the U.S. coffee market. A key aspect of Keurig’s brand growth strategy is aggressively pursuing partnerships and licensing arrangements with well-known brands in the mature ground coffee segment. In this paper, we quantify the importance of brand variety in driving the success of the overall Keurig system.
In so doing, we shed light on the benefits of partnering for high brand equity versus a large number of brand options. We propose and estimate a demand system for the brewer and coffee that endogenizes the consumer’s self-selection into brewer adoption based on their heterogeous tastes for both products. We then perform counterfactuals that demonstrate the role of Keurig’s partnerships with national brands in hastening the growth of the category. We find that K-Cup consumables have a strong influence on K-Cup system adoption. On average, Keurig-owned brands have the largest influence on brewer adoption (5.2%), but partner brands are a close second (4.9%) and by 2015, partner brands are 50% more influential on growth than Keurig- owned brands. Hence, our results both quantify the benefits of brand variety to growth and shed light on the most beneficial types of partnerships. 1 Introduction
In the late 2000s, the previously mature U.S. market for brew-at-home coffee was transformed by the creation of the single-serve coffee pod. Between 2010 and 2015, coffee products employing
Keurig’s newly developed ‘K-Cup’ technology grew from $0.3 billion in yearly sales to over $3.6 billion, with the latter figure comprising 41% of total dollar sales for the U.S. market. Compared to traditional brewing methods, single-serve products represented a mess-free alternative that precisely controls the time and temperature of the brewing process, thereby delivering a standardized, high- quality cup of coffee. Within the single-serve brewer market, Keurig’s positioning strategy was geared around the technological superiority of its brewing system and the wide variety of flavors and brands it offered in this proprietary system. A key aspect of its growth strategy involved aggressively pursuing partnerships and licensing arrangements with brands that already enjoyed strong positions in the ground and whole bean coffee segments.
Keurig’s rapid rise to dominance leveraged the unique role of its proprietary platform, the tied goods aspect of its coffee pods and the indirect network effects arising from its broad inclusion of partner brands and licensees in the K-Cup coffee consumables. Our goal in this paper is to quantify the importance of this brand variety in driving the success of the overall Keurig system.
Through this quantification, we aim to shed light on the nature of brand relationships that are most beneficial, and the magnitude of the benefit from working with national brands that are able to leverage their brand equity in the mature coffee segments into the new K-Cup segment. To do this, we propose and estimate a discrete choice model of K-Cup brewer and coffee demand that endogenizes consumer’s self-selection into brewer adoption based on their heterogeneous tastes for both products and accounts for the availability of partner and licensee brands as they join the
Keurig system. Using these estimates, we then perform counterfactuals that demonstrate the role of Keurig’s partnerships in hastening the growth of the category.
To do so, we adapt the platform and tied goods ‘network’ frameworks developed by Lee (2013) and Derdenger (2014) for the video game and console industry to the coffee-brewer context. While our setting shares some of the aspects of the video game and console market, there are some
1 notable distinctions. First, consumers naturally “multi-home” so that adopting the Keurig system does not preclude consumers from continuing to use traditional drip brewers, it simply expands their choice set of available coffee options. In this sense, the coffee category shares aspects of the
‘razor and blades’ market analyzed by Hartmann and Nair (2010). However, unlike in the razors and blades case, the variety of coffee consumables grows dramatically over time and is important to the consumer’s motivation to purchase a brewer. Second, unlike video games, coffee consumables are purchased repeatedly and consumers select among multiple competing varieties. These features make dynamic considerations less important and choice among consumable brands more central.1
An important aspect of the platform and tied goods frameworks is treating the benefit of adopt- ing the Keurig platform as the change in inclusive value arising from gaining access to the expanded set of products it offers. In contrast to the durable video game (software) context, we do not al- low consumers to forecast future changes to coffee product availability.2 We account for consumer heterogeneity in both brewer and coffee demand by using an aggregate discrete choice random coef-
ficients “BLP”-type framework (Berry 1994; Berry, Levinsohn, and Pakes 1995; See Nevo 2000 for user guide) that allows for flexible substitution patterns. In addition, incorporating heterogeneity allows consumers with high values for coffee and the brewer (i.e., lower price sensitivity) to adopt the system earlier than corresponding low type consumers. In sum, our modeling strategy of using the inclusive value to bridge the demand for the brewer and coffee, and incorporating heterogeneity allows us to endogenize the evolution of the platform’s installed base and capture the time-varying composition of platform users.
The multi-homing aspect of our setting requires a novel modeling approach and estimation strategy. Unlike platform and tied goods studies in the video game and console setting, we do not observe sales of consumables broken out by owners (or non-owners) of the K-Cup brewer. To address this, we construct each consumer’s platform ownership likelihood at each point in time, and estimate
1Although coffee products are clearly not durable in the sense of video games, there is some scope for stockpiling. We ignore stockpiling in the coffee choices for two reasons–K-Cups are relatively large to store and we find no evidence that sales diminish following a price promotion, suggesting intertemporal shifts in demand are limited. 2This appears to be consistent with the consumable purchase behavior in the category. To the extent that expectation formation of consumables would matter to consumer’s brewer decision, the entry of a major brand like Starbucks should see acceleration in brewer sales prior to the entry. In contrast, when Starbucks K-Cups became available brewer sales accelerated only after the entry.
2 the demand for consumables jointly for both platforms, rather than for each platform separately as in other platform and tied good studies. As a result, our estimation strategy uses a modified version of Derdenger’s (2014) nested fixed point routine that iteratively updates the parameters from the hardware and software choice problems until the implied distribution of brewer ownership status converges across both markets. Our modification focuses on addressing the multi-homing aspect.
We estimate the demand system using K-Cup brewer and consumable coffee sales data for 15
US cities from 2010-2015, which covers the period of major brand variety expansion for the K-Cup system triggered by the expiration of an important patent. The number of K-Cup coffee brands increased from 10 to more than 150 during our sample period, which includes the entry of many well-known national coffee brands (e.g., Starbucks, Folgers, Maxwell House, Peet’s). The average installed base of the K-Cup brewer expanded from 11% to 31%.3 Our data provide rich regional variation in demand for both the brewer and coffee consumables.
Our demand estimates indicate that the inclusive value plays a significant role in influencing brewer demand, suggesting that quantifying the indirect network effects for the coffee consumables is important. We also find significant price heterogeneity in the coffee consumable market, but little in the brewer market. Consumers who are less price sensitive to coffee are more likely to adopt
K-Cup brewers earlier, so the average price sensitivity of K-Cup brewer owners increases over time.
We conduct counterfactual experiments based on the structural estimates to shed light on the role of brand variety in the evolution of the Keurig system. In particular, we simulate the K-Cup brewer adoption had there been fewer K-Cup coffee brands available with the following conclusions:
First, K-Cup consumables have a strong influence on K-Cup system adoption. If consumers did not respond to the “software” market, the K-Cup system’s installed base would have been on average
20% lower by the end of 2015 than the actual level it achieved. Second, Keurig-owned K-Cup brands on average have the largest total contribution to the system adoption and competitor brands have the least contribution. In particular, installed base at end of 2015 would have dropped by 5.2% if
Keurig owned brand were not available, 1.5% if competitor brands were not available, and 4.9% if partner brands were not available. The magnitude of the contribution is proportional to the
3Based on data from the top 30 IRI cities.
3 share of each relationship type in K-Cup segment. Third, the total contribution of Keurig’s partner brands increases over time, matching Keurig’s own brands by 2012 when Starbucks enters, and overshadowing Keurig’s own brands by 2015. By then they had almost 50% greater influence on the installed base than Keurig’s own brands.
These counterfactual analyses both provide an estimate of the magnitude of the coffee consum- ables influence on the Keurig system growth, and provide insight about the nature of relationship types are most beneficial. In particular, partnerships with a small number of national brands that have strong brand equity to leverage into the K-Cup segment (ala the partner brands) provide much more system growth than even a large number of smaller brands (ala the competitor or licensed brands). As the system grows, partner brands become more important to growth than owned brands. This finding also indicates that not only would measuring indirect network effects through simple counts of tied good size miss the main influences on platform growth, but also the influence of the software brands can vary as those brands obtain deeper penetration.
We note that these counterfactual estimates are conservative because we only drop brands after
2010 (when our data begins) and because we do not allow for entry of competing brewing systems.
Our positive results for partnering generating system growth suggests that Keurig’s relationships with other brands aren’t purely to foreclose the entry of competing systems. Future research could allow for entry of competing brewing systems in order to calibrate the relative size of the incentive to partner for Keurig system growth versus keeping potential entrants out. For Keurig, this incentive to foreclose has important implications because the software patent expires prior to the hardware patent. Further, knowing which brands are most influential for locking in consumers and growing the system is also helpful. Future research could illustrate the incremental value to the Keurig system of partnering with particular brands that have greater brand equity in the mature ground and whole bean segments that can be leveraged into the K-Cup segment.
Although this paper relates to the broader literature on estimating indirect network effects and consumer demand in markets with platforms and tied goods, it is most closely connected to the aforementioned papers by Lee and Derdenger. However, it also relates to earlier papers on networks and platforms, including those by Gandal, Kende, and Rob (2000), Nair, Chintagunta, and Dubé
4 (2004), Clements and Ohashi (2005), Corts and Lederman (2009), Dubé, Hitsch, and Chintagunta
(2010), and Karaca-Mandic (2011). The papers by Lee and Derdenger were the first to take the heterogeneous quality of the networked products directly into account (the earlier papers followed the Katz and Shapiro 1985 approach of capturing the network benefits exclusively through its size).
In our setting, the quality and positioning of the partner and licensee brands is an important aspect of platform adoption.
The rest of the paper is organized as follows. In the next section we describe the US at- home coffee industry and in particular the single-serve coffee market. In section 3 we describe the data and summary statistics. In section 4 we present a static demand model of brewer and coffee.
Section 5 discusses the estimation strategy and identification. Estimation results and counterfactual experiments are presented in section 6 and 7. Finally section 8 concludes.
2 Industry
2.1 The U.S. At-Home Coffee Market
In the United States, the most common coffee brewer for at-home brewing is the classic auto-drip type. With a standard auto-drip machine, users load coffee grounds into a filter basket and make a pot of coffee in each brewing cycle. These brewers typically range in price from $20 about $40, but have limited ability to chage the brewing amount, lead coffee to be wasted, and are a relative hassle to clean. In contrast, a single-serve type brewer, such as a Keurig K-Cup machine, uses a pre-packaged proprietary portion pack (pod) containing ground coffee and a filter, and brews a single cup of coffee each cycle (more recent models can also brew multiple cups). Thus, the single- serve type brewer provides convenience, while allowing users to brew a wider variety of flavors and brands (due to their single-serve nature). Single-cup brewers are now the fastest growing class of brewer, and the second most popular brewing system overall.4 Fueled by the brewer adoption, the related single-cup coffee products have become the second largest segment (trailing only ground coffee) in the brew-at-home category, with $3.6 billion revenue. This corresponds to 41% of total
4http://www.statista.com/topics/2219/single-serve-coffee-market/
5 coffee sales in the 2015 US market.
The single-serve brewing innovation was first commercialized by Keurig. Founded in Massachu- setts in 1992, Keurig first entered the office coffee service market in 1998 with a highly successful, commercial grade single-cup brewer. In 2003, it introduced the K-Cup brewer for the at-home market.5 Several single-serve brewing systems were launched around the same time, including the
Melitta One:One brewer by Salton, Senseo by Sara Lee, and Home Cafe by P&G. However, por- tion packs developed for a given system were not compatible with the others, creating a platform adoption problem. Keurig positioned its systems as the superior option, primarily due to its higher quality machines and greater variety of flavors and brands.6 By 2010 the K-Cup system dominated the field, and effectively became the proprietary standard of the single-serve market. This domin- ance is reflected in Keurig’s share of coffee sales. In 2010, Keurig accounted for more than 87% of single-serve sales, on a base of $201 million. By 2015, K-Cup sales reached $1.9 billion, while the competing systems as a whole never achieved more than $30 million in sales. By 2015, the revenue share of other systems was smaller than 1%. Because of this dominance, we will treat the Keurig
K-Cup brewing system as a monopoly in the single-serve market, focusing on the adoption of the
K-Cup brewers alone.
2.2 Keurig K-Cup Brewing System Expansion
Keurig strove for further platform growth by increasing number of brands available for the K-
Cup system. In 2006, GMCR, a coffee roaster company, acquired Keurig. At the same time, they brought in-house a number of existing brands who had licensed from Keurig the patents to supply coffee K-cups to grocery stores. The integrated company also worked to develop an active licensing and partnership program to bring existing national brands into the Keurig system (e.g.,
Starbucks and Eight O’Clock). In partner arrangements, the partner brands were able to leverage the manufacturing capabilities of the combined company, thereby eliminating the upfront capital
5Source: Kellogg case “Keurig: From David to Goliath - The Challenge of Gaining and Maintaining Marketplace Leadership” and Wikipedia. 6See Kellogg case “Keurig: From David to Goliath” for detailed discussion on the performance of competing brewers during 2003 to 2008.
6 or leasing costs needed to operate their own production lines. In return, they paid Keurig for manufacturing and occur additional licensing fees. Partners still brought the branded product to market independently. In licensing agreements, Keurig manufactures and takes the K-Cups to market in which case Keurig brings the market itself and pays a licensee fee per cup.
The main patent on K-Cup portion pack expired in September 2012. Before the patent ex- piration, Keurig acquired eight brands (e.g., Tully’s and Donut Shop), licensed six brands (e.g.,
Newman’s Own and Eight O’Clock), and added Starbucks and Folgers as partner into the K-Cup system. A few roasters developed compatible products to be used in the K-Cup brewers including brands manufactured by the Rogers Family Company, but the set of competing brands achieved very limited success prior to the patent expiration. After patent expiration, many more competing brands entered the system since entry was no longer legally restricted.
After patent expiration, Keurig continued to engage with national brands to encourage them to partner rather than compete. Keurig was able to continue to attract major brands including switching from unlicensed/competitor to partner (Peet’s, Maxwell House). Mainly because of com- peting brand entry, the post-patent expiration triggered a rapid increase in the number of brand choices in the K-Cup system compared to prior to expiration, though none were as well-known as the existing national brands.
3Data
Brewer and coffee sales data were obtained from the IRI InfoScan scanner data base. Sales data are available at the city, weekly, and UPC-coded product level for 65 IRI cities from 2010 to 2015. IRI collects average price and cups sold (for K-Cups, 1 portion pack is equivalent to 1 cup; for ground and whole bean coffee, 0.4 oz is equivalent to 1 cup), distribution (average product availability weighted by store all commodity volume in that city), number of items (UPC-coded products) sold per store, and percent of items on feature and display. We aggregate the data by city, month, and product choice, which is defined below. The average prices are defined as monthly dollar sales divided by monthly cup sales. Distribution, items sold per store, and the merchandising variables
7 are the averages for each month.
Installed base, coffee segment penetration rate, and individual demographics were obtained from the IRI panelist database. Data are available at the aggregate panelist, city, and year level from
2012 to 2015. We use the annual K-Cup segment penetration rate (percent of panelists that ever bought a K-Cup product in a given city and year) as a measure of the K-Cup brewer’s installed base. These yearly installed base measures from the panelist data are used to scale the monthly level of brewer sales observed in the point of sale data. This scaling is important because we do not observe the universe of brewer sales (for example, Amazon, Costco, and Bed Bath & Beyond are not in our data). Details of the scaling procedure are given in Appendix A.
For our analysis, we use the IRI city data, which is similar to the data approach taken in Nevo
(2000b). For our purposes, we wish to include the cities with the largest populations in order to ensure sufficient sample size to obtain an accurate estimate fo the installed base. In addition, IRI doesn’t track merchandising variables for all cities and we drop any cities for which this data is missing. As a result, we select the top 15 IRI cities for our analysis. 7
Product choices for coffee are defined as brand-segment combinations. Coffee segments include ground, whole bean, and K-Cups. We use the term K-Cup to refer to both K-Cups and K-Cup brewer compatible single-serve products. Note that Starbucks ground and Starbucks K-Cup are treated as different options. Ground and whole bean products are relevant to our study because they affect the marginal benefit of adopting a K-Cup brewer. This will be explained in detail in the model section below. We exclude instant and ready-to-drink coffee from the analysis because
1) they are relatively small segments and 2) previous research suggests their substitution with K- cups is quite limited (see Ellickson, Kong, and Lovett 2016). We include 10 ground coffee, 3 whole bean coffee, and 50 K-cup brands, which are selected based on median market share among the 15 cities in 2015, for each segment respectively. We include more K-Cup brands since the focus of this paper is on measuring the impact of K-Cup brand variety on K-Cup brewer adoption, and, in doing so, we need to capture as much of the benefits of this variety as possible. The included brands constitute about 67% of total coffee revenue, while the included K-Cup brands capture 98% of total
7We also examined the top 30 IRI cities as compared to the top 15 in order to evaluate whether this selection is likely to lead to bias. We found no meaningful differences in demographic make-up. Details available upon request.
8 K-Cups revenue. Observations with lower than 5% distribution are also excluded because these cases include oddities like new product test markets, initial product roll-outs with reduced pricing
(but low volume), and clearance of discontinued products, none of which are a good characterization of typical demand. In total, these omitted cases comprise about 0.24% of the total inside coffee revenue.
The market size for coffee is defined as 4*days in month*number of coffee buying households in the market. Note that we are assuming an average coffee buying household consumes a maximum of
4 cups of coffee per day. The number of coffee buying households in a market is the projected total households in that city (linearly interpolated to the monthly level based on annual data) times the penetration rate of the coffee category (averaging over 2012 to 2015) that is obtained from the IRI panelist data. The average penetration rate for coffee varies between 58% (San Francisco/Oakland,
CA) to 80% (Orlando, FL) for the sample.
Brewer choice options are defined as manufacturer-product lines. Product line information was collected online.8 We include the top 12 K-Cup brewers in the analysis, which includes six Keurig
K-Cup brewers, three Keurig 2.0 brewers, and three Keurig co-branded brewers (produced by the licensee manufacturer Mr. Coffee). All K-Cup brewers use the same brewing technology and operate similarly. Rather than brewing mechanism, product lines are differentiated more by features such as water reservoir size, cup size and water temperature options, LCD display, programmable on/off switch, and the brewer’s appearance. New models in the same product line typically have updated designs and new or improved features. The 12 brewers capture more than 99% of the total K-Cup brewer units sold. As above, observations with lower than 5% distribution were excluded. These cases are about 6.5% of the total brewer units sold.
General descriptive statistics are provided in figure 1 and table 1. A vertical line indicating the expiration date of K-Cup pod patent is shown in all figures. From the top left panel of figure 1, the differences in average prices across the segments is celar. The average price per K-Cup pod is about $0.60, which is more than six-fold the price of ground coffee per cup (around $0.10), and three times that of whole coffee beans (around $0.20). Despite a slight increase from the beginning
8See, for instance, http://coffeejustright.com/compare-keurig-models/ for a detailed discussion on the K-Cup brewer product line.
9 of the data period to the end of 2011, prices for ground and whole bean are relatively stable over time, while price of K-Cup peaked at more than $0.60 in mid-2012, and graduately went back to
$0.60. The top right panel shows that the price of K-Cup brewers varies from $50 to $175 across product lines. The Mr Coffee co-branded K-Cup brewers is a low end option, with prices at or below $75. The Keurig Platinum is the high end option, with average price around $175. The
Special Edition, and 2.0 series are relatively high end brewers. The middle left panel shows that the K-Cup segment sales have grown significantly over the years. The Keurig partner brands have the largest volume sales among all relationship types at the end of 2015, followed by Keurig’s own brands. Relating the sales to the middle-right panel of figure 1 and table 1, one can see that the
first partner brand, Folgers, entered in late 2010 and that it and Starbucks dominate the partner growth. In contrast, the competitor type has the largest number of brand options, but relatively low total sales. This stark difference in concentration of sales suggests that simply using a count of brands cannot capture the rich heterogeneous role that these brands play in shaping Keurig system growth. The middle and bottom left graphs also indicate that there exists significant seasonality in demand for brewer but not as much in K-Cup consumables. Finally, the bottom right panel demonstrates wide regional variation in both the initial level of the K-Cup brewer installed base and the rate at which it increases.
4 Model
In this section, we develop a static structural demand model for pruchasing the K-Cup system, as well as demand for coffee consumable products themselves (both K-Cups and conventional whole bean and ground coffee). Our overall modelling approach is to adapt the classic hardware-software demand framework to a static choice setting.9 To fix terminology, the K-Cup brewer is the “hard- ware,” and the K-Cup coffee products are the “software” required to utilize the hardware. These terms will be used interchangeably hereafter.
9We will explain our choice of a static framework shortly. For the standard dynamic approach, see the applications to the razor/razor blade by Hartmann and Nair 2010 and to the console/video game settings by Lee 2013 and Derdenger 2014.
10 Figure 1: Brewer and Coffee Statistics
Product Price Brewer Price
0.6
150 Keurig 2.0 K200 Keurig 2.0 K300 Keurig 2.0 K400 Keurig Elite Keurig Mini Plus 0.4 GROUND COFFEE Keurig Other 1 K−CUP Keurig Other 2 WHOLE BEANS Price 100 Keurig Platinum price per cup Keurig Special Edition MR COFFEE − KG1 MR COFFEE − KG2 MR COFFEE − KG5
0.2
50
Jan 2010 Dec 2010 Dec 2011 Dec 2012 Dec 2013 Dec 2014 Dec 2015 Jan 2010 Dec 2010 Dec 2011 Dec 2012 Dec 2013 Dec 2014 Dec 2015 time time
Product Sales Number of K−Cup products 8e+07 16
6e+07 12
COMPETITOR COMPETITOR COMPETITOR/PARTNER COMPETITOR/PARTNER 4e+07 LICENSED LICENSED OWNED 8 OWNED Cupsales PARTNER PARTNER
2e+07 4
0e+00
Jan 2010 Dec 2010 Dec 2011 Dec 2012 Dec 2013 Dec 2014 Dec 2015 Jan 2010 Dec 2010 Dec 2011 Dec 2012 Dec 2013 Dec 2014 Dec 2015 time
Installed Base
30 Baltimore, MD/Washington, DC Brewer Sales Chicago, IL Columbus, OH Dallas/Ft Worth, TX 200 Denver, CO Houston, TX Los Angeles, CA 20 Miami/Ft Lauderdale, FL 150 Minneapolis/St Paul, MN
Percentage New York, NY Orlando, FL Phoenix/Tucson, AZ 100 San Francisco/Oakland, CA St Louis, MO Unit sales (in 1000) Tampa/St Petersburg, FL 10 50
0 Jan 2010 Dec 2010 Dec 2011 Dec 2012 Dec 2013 Dec 2014 Dec 2015 Jan 10 Dec 10 Dec 11 Dec 12 Dec 13 Dec 14 Dec 15 time
11 Table 1: Top 3 Selling K-Cup Brands by Group Group Brand Entry date (m/y) Ave. Share Competitor PRIVATE LABEL* 6/12 5.5% DON FRANCISCOS FAMILY RESERVE 1/13 1.5% BARNIES COFFEE KITCHEN 10/13 1.3% Competitor/Partner PEETS 5/13 3.7% MCCAFE 12/14 3.1% GEVALIA 12/12 2.9% Licensed NEWMANS OWN ORGANICS 1/10 8.2% CARIBOU COFFEE 1/10 6.9% EIGHT O CLOCK 8/12 4.1% Owned GREEN MOUNTAIN 1/10 21.0% TULLYS COFFEE 1/10 6.3% DONUT SHOP 3/10 6.0% Partner STARBUCKS 11/11 17.6% FOLGERS 10/10 12.4% DUNKIN DONUTS 5/15 5.6% Note: Share is average share in K-Cup segment revenue in city month *There are private labels in competitor group as well as partner group
Even though the K-Cup brewer is a durable good, it differs from the video game industry for at least two important reasons. First, while game consoles have declining prices over time and new generations are released every few years, in contrast, brewer prices vary little apart from seasonal changes, and true innovation was limited. Second, buying the K-cup brewer does not “lock-in” consumers to that system. Given the wide distribution and low prices of standard drip brewers, we assume everyone has access to such a brewer. This implies the K-cup brewer simply increses the number of available coffee options for those who choose to adopt. For these reasons, consumers are less likely to expect (and wait for) a price decline or a better brewing technology. As a result, we make the simplifying assumption that consumers believe future prices, varieties and qualities will be the same as today, and maintain this assumption each month.
Note that, similar to the video game literature, coffee quality and availability influences brewer demand through the value of the K-cup coffee options. We formulate this influence as the expected maximum coffee utility, as captured through an inclusive value term. Note that the inclusive value term allows for heterogeneous tastes for coffee that impact hardware adoption. Because consumers leave the hardware market after purchase of a brewer, the composition of consumers in the hardware
12 market changes over time. To account for this endogeneous market evolution, we jointly estimate the demand system for both coffee and brewers. Capturing the endogenous market evolution is central to correctly identifying the size of the impact of K-cup coffee brands on brewer adoption.
On the brewer side, consumers must decide whether to adopt a K-Cup brewer (vs. choosing no K-cup brewer). In each month, if they do not already own a Keurig machine, we assume that consumers considering buying and may purchase one of the available K-Cup brewers, h H . They 2 t exit the brewer market after purchase. Based on the assumption that all consumers are endowed with a drip brewer, non K-Cup brewer owners may purchase any non K-Cup products j J NK 2 t NK at any time, where Jt is the set of available ground and whole bean products at time t; K-Cup brewer owners may purchase any coffee product j J , where J is the extended set of available 2 t t ground, whole bean, and K-Cup coffee at t.
4.1 Demand for Brewers (Hardware)
First consider the hardware decision. In every month t,consumeri decides whether or not to purchase a K-cup brewer h H . Since consumers will only purchase the hardware once and then 2 t exit the hardware market forever (we do not consider replacement purchases), the consumer takes into account the lifetime utility obtained from adopting the K-Cup brewer h at time t:
u˜hw = ↵ K + ↵hw log(phw)+Xhw hw + ⇠hw + ✏hw,h H , iht it i ht ht ht iht 2 t
K where it is the expected present-discounted value (PDV) of choosing the best option among both K the K-Cup (KCt)anddripbrew(NKCt)productsinperpetuity( it will be defined more formally hw hw in section 4.2), pht is the price of brewer h at market t (city-month), ↵i is a heterogeneous hardware price coefficient that is normally distributed with mean ↵¯hw and standard deviation