The Song Remains the Same? Technological Change and Search in the Recorded Music Industry

The song remains the same? Technological change and search in the recorded music industry

Mary Benner

3-365 Carlson School of Management University of Minnesota 321-19th Avenue South Minneapolis, MN 55455 [email protected] 612-626-6660

Joel Waldfogel

3-365 Carlson School of Management University of Minnesota 321-19th Avenue South Minneapolis, MN 55455

January 5, 2015

Preliminary draft – please do not quote, cite, or circulate

The song remains the same? Technological change and search in the recorded music industry

Abstract

Technological change has brought fundamental challenges, as well as opportunities, to the recorded music industry. One set of changes – file sharing – has facilitated unpaid music consumption, eroding the abilities of the existing firms to generate revenues and making it difficult for major record labels to continue releasing new music using traditional modes of production, distribution, and promotion. At the same time, other technological changes have reduced the costs of these industry activities, as well as the search for talent. These twin technological changes raise questions about the strategies pursued in response by organizations in the recorded music industry. Using data on over 60,000 albums released in the US 1990-2010, we examine the differing responses of major label and independent label organizations before and after the technological change. We find that major label releases reduce music releases while independent labels increase music releases, and that major labels increasingly select on previously successful artists. This selective approach appears to work as a growing share of major label releases achieve commercial success. Finally, we find that majors also achieve greater current success on debut albums, possibly reflecting greater availability of information about artists even prior to their first releases.

Recorded music revenue has fallen by over half since Napster’s appearance in 1999 (Aguiar, Duch-

Brown, and Waldfogel, 2014), making it difficult for major record labels to continue releasing new music using traditional modes of production, distribution, and promotion: pressing music onto physical media, transporting them to many retail locations, and promoting the new works on terrestrial radio. At the same time, other technological changes have reduced the costs of industry activities, such as producing, distributing, and promoting music products, as well as searching for talent.

These twin technological changes – file sharing along with cost reduction – raise interesting questions about the strategies pursued in response by firms in the recorded music industry. The industry has traditionally been dominated by “major” record labels controlled by a handful of media conglomerates (Universal, Sony, BMG, etc.). These record labels release recordings that account for the vast majority of revenue, using the traditional high-cost, high-promotion approach. Alongside these major labels is a large fringe of “independent” labels releasing the vast majority of new music but accounting for a small share of sales, traditionally about 10-15 percent. Independent labels have also historically employed lower-cost approaches – for example, without radio airplay – that allow them to release music that has lower revenue potential.

Although economists have recently begun to study technological change, ‘digitization,’ and

‘digital disintermediation’ in the music industry (e.g. Liebowitz, 2006; Waldfogel 2012a;b; Aguiar, Duch-

Brown, & Waldfogel, 2014; Smith and Telang, 2010; Zentner, 2006), they have not explored questions at the firm level, i.e. how organizations innovate in response to the radical technological changes, or the associated performance outcomes. Thus, we know little about innovation and change within organizations in the music industry as radical technological change both threatens to destroy the relevance

3 of existing capabilities and business models, but at the same time offers new, lower-cost ways to pursue industry activities.

A rich stream of prior work has explored the challenges for incumbent organizations faced with technological changes in their industries (e.g. Tushman & Anderson, 1986; Utterback, 1994; Christensen

& Bower, 1996; Henderson & Clark, 1990; Tripsas & Gavetti 2000; Benner, 2010; Benner &

Ranganathan, 2012; Sull, Tedlow, and Rosenbloom, 1997). Research in this area has further examined how ‘search’ behaviors within organizations into more distant technological and market domains (i.e. also termed innovation or exploration) can provide firms with the requisite knowledge, capabilities, or absorptive capacity (Cohen & Levinthal, 1990) to initiate – or respond to –such changes in their environments (e.g. Nelson and Winter, 1982; March, 1991; Levinthal & March 1993; Levinthal, 1997).

The work on technological change often makes further distinctions between established incumbents, believed to be rigid and inertial, and the smaller organizations or new entrants that are viewed as more nimble and better able to survive these ‘gales of creative destruction’ (Schumpeter, 1934).

In this study we examine how music organizations (i.e. record labels) have responded to the major technological changes sweeping the recorded music industry. Specifically, we study how search behaviors in record label organizations change in response to the challenges of technological change, as well as the resulting performance outcomes. Our research questions are: How does search for the ‘major label’ incumbent organizations in the recorded music industry change in response to major technological changes, what are the performance outcomes, and how do these behaviors and outcomes differ for the smaller scale ‘independent label’ organizations?

Using a unique, extensive set of data on over 63,000 new music releases (i.e. products) of music producers from 1990 to 2010, we are able to study the influences of changes in both the abilities of producers to appropriate revenues and reductions in costs to produce, distribute, and promote music on both major label and smaller independent label organizations. The advent of Napster and the increase in unpaid file sharing after 1999 provides us with an exogenous shock, or “experiment,” that allows us to study the search behaviors of organizations both before and after the advent of file sharing. Our extensive

4 data on the music releases of both the major labels and smaller independent labels allows us to compare the changes at majors and independents with a research design built on these “differences-in-differences” before and after the technological change.

Next we describe our context of technological change, followed by contextual hypotheses about firms’ search behaviors and the resulting performance outcomes in this setting. We then test the hypotheses and present findings.

Empirical context: Technological change in the music industry

Music is what is termed an “experience good.” Consumers generally cannot determine whether they like it until they have used it – e.g. listened to an album or song – repeatedly. The difficulty that consumers have determining how appealing they find new music has an important analogue on the producer side: the entities releasing music (record labels) find it difficult to predict which of the potential projects they might pursue will be successful with consumers. That is, investments in music products – particularly those by new artists but even those from established bands – are risky. These features of music stemming from the nature of the product affect many of the activities involved in bringing music to market.

Record labels undertake a variety of activities in order to bring music products to market. It is helpful to describe these activities – and the traditional ways in which the activities have been undertaken

– in order to understand how recent technological changes would be expected to influence the activities of various kinds of firms.

Many would-be artists seek to make their music available to consumers. For example, many artists submit “demo tapes” to music labels.1 These possible projects differ substantially in both their ex ante promise (how appealing it seems that they might be if they were produced) as well as their ex post success (how successful they turn out to be, even holding constant how promising they seem at the

1 See Caves (2000) and Vogel (2007) for descriptions of the music industry.

5 outset).2 A very low percentage of the recorded material by traditional record labels has generally been successful, suggested by the industry analyst’s comment that “perhaps as little as 10 percent of new material must make a profit large enough to offset losses on the majority of releases...” (Vogel, 2007).

Record labels produce much more material than would actually succeed and the vast majority of albums and records don’t cover their costs (Caves, 2000). Thus, the traditional discovery process in the music industry was analogous to taking ‘draws from an urn.’

Making music available to consumers has typically required a series of activities. First, music must be recorded and produced, which has traditionally required both expensive recording equipment as well as skilled labor. Once a master copy is recorded, it must be physically produced, i.e. pressed onto some physical medium such as vinyl or, since about 1985, compact disc. Second, physical products have required physical distribution. This, in turn, entails two sorts of activities: physical products needed to be transported from manufacturing facilities to warehouses and then to retail stores. Of course, prior to this, producers also needed to have convinced retailers to carry their releases. A typical physical store in 1990 carried many fewer than the 30,000 albums per year released in the industry. The majority of recordings were therefore not available in most retail outlets. The transitory nature of demand for most successful popular music raised the cost of distribution. Because records needed to be available near consumers during the short window of possible consumer interest in a new album, producers needed to ensure that albums would be in stock in many locations. And because of the difficulty in predicting which releases would be successful, producers needed to undertake the costly step of making many new releases widely available, even though most would not become popular.

Finally, promotion has typically been expensive. Again because music is an experience good, producers find it highly valuable to expose listeners to their new music. The chief means of new music promotion had long been through radio. As Vogel (2007) and Caves (2000) emphasize, the U.S. recording industry has traditionally produced far more new music than can be played on the radio. Hence

2 Aguiar and Waldfogel (2014) present evidence on the distinction between ex ante promise and the ex post success of new music products. A fairly inclusive set of variables can explain less than 40 percent of the variation in commercial success across the tracks released in the US in 2011.

6 there have been strong incentives for ‘payola,’ i.e. bribes from producers to radio stations or their employees to get particular songs played on the radio. As Caves (2000) describes, even after payola was explicitly outlawed, its analog continued to function, such that in 1985, the recording industry was paying radio stations $60-$80 million for airtime at a “time when its pretax profits were at most $200 million.”3

Thus, for the major recording labels, success in the recorded music industry traditionally required four main sets of capabilities: 1) discovering talent/signing new artists, 2) producing albums, both artistically and physically, 3) arranging for physical distribution through retail outlets, and 4) arranging for promotion of music, generally on radio, but also through live concert performances. These activities were both expensive and interrelated. By 2000, music production was concentrated in four large media conglomerate firms that operated the major record labels.

Operating alongside the majors were independent labels unaffiliated with major media firms.

These firms operated on a somewhat different business model. While they too needed to select from the pool of would-be artists, their production, distribution, and promotion activities operated on a smaller scale and typically at lower costs than the majors. Independent music was rarely promoted on terrestrial radio (Thomson 2009).4 Consequently, independent record labels collectively released a large number of albums with modest commercial prospects, while major labels were the outlets of choice for artists with more substantial commercial appeal (Kalmar, 2002).

Search, discovery, and selection in the music industry

Figure 1 gives a stylized description of the selection activities in the industry prior to the technological change (often referred to as ‘digitization.’) To conceptualize the selection of innovations or new product releases for the different types of record producers, below we show how potential projects can be arrayed according to expected revenue, i.e. the expectation that record label executives might have about the appeal or revenue potential of recording artists, if the label were to release the product. To the right are bands perceived to have substantial commercial promise (i.e. high potential revenues from

3 See page 292 in Caves (2000). 4 See http://www.futureofmusic.org/sites/default/files/FMCNYSplaylisttrackingstudy.pdf .

7 record sales), while those to the left are expected to have less commercial promise. In this model, bands that major labels deemed sufficiently promising to warrant their large-scale investment – those with expected revenue above T(major) - would get signed to a major label. Bands likely to generate less revenue than those signed to majors but with promise above T(indep) would be signed to independent labels. Finally, bands with insufficient promise to be signed even by low-cost independent labels would remain frustrated would-be recording artists. Note that the distribution in Figure 1 is expected but not realized revenue. Because record label executives had some ability to predict the success of the bands they signed, expected revenue is correlated with realized revenue; but the revenue realizations contain surprises.

Figure 1

Prior to 1999, these industry activities took place in an environment in which recorded music products were protected by a combination of law and technology, i.e. the ‘appropriability regime’ (Teece,

1986). Under copyright law, the labels producing particular albums had exclusive rights to market these works. Anyone selling copies of the works without agreement with the rights holders would be violating the rights holders’ copyrights. Perhaps equally important, copying – while feasible – was inconvenient relative to the cost of purchasing legal copies. Prior to diffusion of the CD, consumers could copy the music on vinyl recordings onto cassette tapes; but even first-generation copies had poor sound quality.

With the spread of the writeable CD drive in the late 1980s, consumers could make perfect copies of CDs;

8 but the process was somewhat cumbersome and required some technical sophistication. These frictions, along with copyright law, were sufficient to prevent large-scale music file sharing in most advanced economies.

Technological Change

The advent of MP3 technology in the late 1990s along with the diffusion of the Internet triggered several important changes in the music industry. First, with the appearance of Napster in 1999, consumers obtained the ability to access and download high-quality digital recordings via peer-to-peer file sharing – with the click of a mouse, in the privacy of their homes or dorm rooms - without payment to the rights holders. The ease of transferring and downloading music without payment (file sharing) has fundamentally weakened copyright protection and the recorded music industry’s appropriability regime, making it much more difficult for record companies to protect their content. Although academics have debated whether file sharing explains the collapse of recorded music revenue (e.g. Oberholzer-Gee &

Strumpf, 2007; Rob & Waldfogel, 2006; Liebowitz, 2006; Zentner, 2006; Blackburn, 2004), it is now clear that file sharing explains most (if not all) of the reduction of music revenue since 1999. Recorded music revenue fell by more than 50 percent in the US and Europe between 1999 and 2010 (Aguiar, Duch-

Brown, and Waldfogel, 2014).

The ‘appropriability crisis’ triggered by the technological change heightened the difficulty for the traditional major record labels to capture revenue from their music products, and therefore profitably market their recorded music products. With less revenue, record companies have, in turn, found it more challenging to incur the large investments to discover ‘new to the world’ talent in the traditional ways.

Yet, working in opposition to the collapse of revenue, the technological change has also reduced the costs of bringing music to market. Where traditional sound recordings required costly studio equipment, an artist can now create a high-fidelity recording using inexpensive and widely available computers and software (e.g. a Mac with Garageband). The diffusion of the Internet – and digital retailing

– has made distribution trivially inexpensive (Bourreau, et al, 2012). Labels need not produce physical copies, nor do they need to make them ubiquitously available in order for consumers to be able to

9 purchase them if they become popular. Finally, other aspects of digitization have reduced the costs of promotion. The rise of Internet radio, including Pandora, Last.fm, rdio, Spotify, and others has broken the bottleneck of traditional radio as a means for acquainting consumers with new music. Internet radio stations broadcast a much wider variety of music than terrestrial stations, allowing promotion for many more artists. Related, a growing coterie of reviewers makes their views available online. Consumers have access to information on far more new music than they encountered through traditional promotional machinery (Waldfogel 2012a).

Knopper (2009: 246) describes the process and the changes to the process triggered by the new technologies:

An artist who wanted to make a record needed studio time – and that cost money, which meant a sizable loan from the label. An artist who wanted to get a single onto the radio playlist needed connections – and that usually meant a label executive who had the money to hire an independent promoter. An artist who wanted to sell millions of copies of a record needed a big-time distributor with the clout to push CDs into big stores like Best Buy or Target – and that meant one of the major labels’ own subsidiaries like WEA or CEMA. Today it’s not necessary to hook up with a label to do all these things. An artist can make a record cheaply, and professionally, using software like Pro Tools. An artist can forgo the radio, building buzz and exposure online via do-it-yourself websites like MySpace, viral videos on YouTube, or any number of social networking services from Facebook to Garageband.com. As for distribution, who needs crates, trucks, warehouses, stores, or even the discs themselves? Artists can follow Radiohead’s example and simply distribute the music essentially free online.

Waldfogel (2012a) provides the example of Arcade Fire’s album The Suburbs, which won the

2011 Grammy award for best album, as a music release that attained a high level of both commercial and critical success with little traditional radio airplay. The album was released by the independent label

Merge Records, and although the album received critical acclaim (as had their previous albums, Funeral in 2004, and Neon Bible, in 2007), it received little radio airplay. The album received substantial Internet radio airplay, however, and the song “Ready to Start” had over 40,000 weekly listeners at last.fm in 2011.

The album won the Grammy for best album and was subsequently certified Gold by the Recording

Industry Association of America (RIAA), indicating sales of 0.5 million by October, 2011.

Thus, lower-cost technologies make it possible for organizations to bring music to market without possessing all of the capabilities of a traditional major record label. As a result, and despite the

10 appropriation crisis brought on by the technological change and Napster, would-be artists with more modest commercial prospects than before can profitably be brought to market post technological change.

If organizations experience cost reductions that are large in relation to the file sharing-induced reduction in revenue, then organizations taking advantage of new, lower-cost approaches could find that more music products have revenues in excess of their costs to produce, and could find it profitable to release more rather than fewer products.

Our interest is in the effects of these two aspects of technological change on organization search in music organizations as they respond to both the threats and opportunities created by the technological changes. We conceptualize organizational innovation in terms of the music ‘products’ produced by record labels, in this case, in terms of organizational search for – and selection of – the talent for new music releases. Innovation occurs both at the organizational level and in the aggregate at the industry level, as organizations search for, ‘discover,’ and ‘select’ new talent. We propose that the two technological changes in the music industry, the collapse of the appropriability regime due to file-sharing and the reduction in costs to produce and promote music will have different effects on organizations in the industry, and further, that these changes will differentially affect the two main types of organizations in the industry, major record labels and independent record labels. First, the advent of file-sharing reduces the revenue available for any particular recorded music product. Referring again to Figure 1, this could be represented as a leftward shift in the distribution of revenue to potential projects for record labels. We can equivalently represent this as an increase in the cost threshold, as Figure 2 shows.

Figure 2

In the face of decreased appropriability, i.e. the ability of firms to capture the revenues and therefore profit from their product releases or innovations, we would expect music recording organizations to focus on more predictably successful artists to obtain higher revenues for each release, to offset the declines in revenue. Thus generally, as we develop in our hypotheses below, we would expect music labels to focus more on more promising artists, likely to generate higher revenues, i.e. with prospects exceeding the threshold prospects of those previously released. In short, if reduced appropriability were the only effect of technological change, we would expect to see a reduction in the number of products released by both major and independent labels, and further, we would expect both sorts of entities to focus on more promising, higher expected revenue projects.

Although it is clear that the industry technological changes threaten to destroy the value of the interrelated capabilities of the major music labels (cf. Tushman & Anderson, 1986), these complementary assets are also likely to retain their value for the most promising talent. That is, a strategy of high-cost promotion, including radio airplay and an expensive choice of ubiquitous availability of physical product in retail stores, may continue to have high economic returns for the most promising artists. While activities such as traditional radio airplay are expensive and therefore a risky investment in the case of an artist who may have little appeal, it is also unlikely that lower cost approaches, such as discovery via

YouTube or Internet radio are the most effective ways to produce and promote the products with strong

12 commercial prospects. Taylor Swift would expect her music releases to be distributed and promoted in a higher investment way than via Spotify.5

While in past research the existence of complementary assets can strengthen firms’ abilities to appropriate returns from technologies (Teece, 1986) or allow incumbents to withstand technological transitions (Tripsas, 1997; Mitchell, 1989), Wu et al (2014) go further to model the influence of complementary assets as ‘prisms’ or lenses by which firms view choices about the direction and magnitude of investment in particular technological trajectories. They argue that complementary assets can influence these choices, such that we might observe firms making greater investments in technologies that have less market promise, and therefore lower returns, because the trajectory is better aligned with the firm’s complementary assets. Adner and Snow (2010) relatedly show that rather than responding to a new technology, firms may ‘retreat’ to market niches where their particular technological knowledge and capabilities continue to be valuable.

Here we propose that as the major record labels’ complementary assets and proven capabilities continue to be valuable and relevant for producing and appropriating returns from the most promising segment of recording artists, the persistent value of these capabilities prevents or slows adoption of the lower cost sets of activities for producing and promoting music. Prior research in organization learning, capabilities, and cognition suggests further support for our argument. Research suggests that firms have difficulty changing their capabilities and routines, particularly to the extent these routines have led to measurable past success (Levinthal and March, 1993; Levitt and March, 1988). Firms can get stuck in a

‘capability traps,’ i.e. the activities that have led to past success tend to persist (Levinthal and March,

1993). Findings from other research further illuminate this persistence: studies have documented the considerable industry-wide improvements in the performance of soon-to-be-obsolete technologies when faced with radical technological change, as incumbent firms persist and incrementally improve their

5 See Hannah Ellis-Petersen, Taylor Swift takes a stand over Spotify music royalties, The Guardian, November 5, 2014 (http://www.theguardian.com/music/2014/nov/04/taylor-swift-spotify-streaming-album-sales-snub.

13 current capabilities and activities (e.g. Landes, 1983; Utterback, 1994), recently referred to by Furr and

Snow (2012) in their study of carburetor technology as ‘last gasps.’ Further, Leonard-Barton (1992) showed in a study of 20 product development projects that the core capabilities of organizations, i.e. activities that have been the sources of past success, influence the values and norms in organizations and lead to rigidities in the types of innovations and new products that are selected and developed (Leonard-

Barton, 1992).

Prior work in strategy has also shown that extant capabilities and complementary assets can further influence organizational cognition, as choices about which technologies to pursue, how to enter markets, and specific choices of product features and design are perceived in light of existing competencies (e.g. Wu et al, 2014; Tripsas and Gavetti, 2000; Benner & Tripsas, 2012). Anecdotal evidence from our setting comports generally with these ideas from past research, but specifically suggests that in the recorded music industry, the expensive complementary assets held by the major labels are accompanied by a strong industry belief that high-cost production and promotion capabilities are indicative of ‘professionalism,’ while the lower cost approaches for these activities are indicative of

‘amateurism’ (see Lemann, 2006). Together these factors prevent or slow the major labels in the music industry from seizing the opportunity to undertake their activities at lower costs than before.

Thus, as major labels are likely to possess the interrelated complementary assets for production, distribution, and promotion that retain their value for producing and promoting the music of artists with the strongest commercial prospects, they are less likely to change activities in ways that take advantage of the cost reducing technological changes than the independent labels that have not developed these costly capabilities.

Additional anecdotal evidence from industry accounts suggests, in line with our arguments, that while the adoption of lower-cost industry activities seems attractive for all firms, the major labels have not adopted the lower cost activities for bringing products to market. Even after the technological changes widely understood to have enabled low-cost entry, it appears that the major record labels continue to engage in their higher cost activities. As of 2012, the International Federation of the

Phonographic Industry (“IFPI,” the associated representing major record labels internationally) reported that major record labels were spending $1 million per album to release work by a new artist and twice that for an established artist.6 While the majors are very concerned about lost revenue, they make no mention of reduced costs to produce music.

Thus there is a split, suggesting different influences of the technological change and different approaches for different organizations. For the independent labels, both revenues and costs have fallen, while for the major labels, revenues have fallen while costs remain high. We can summarize the effects of technological change on the two main types of record labels in Figure 3. Relative to the pre-Napster status quo, T(major) has risen while T(indep) has fallen. That is, given the lower costs, independent labels are likely to be able to release more music products than they did in the past, i.e. more music products that will now be able to exceed their costs to produce. In contrast, the major labels are likely to release less music, as the appropriation crisis coupled with little reduction in cost results in fewer music releases that will provide revenues in excess of their costs to produce.

Figure 3

Thus, we hypothesize different effects of the technological change on music releases by major versus independent record labels:

6 http://www.ifpi.org/content/library/investing_in_music.pdf

Hypothesis 1a: Subsequent to the technological changes in the music industry, there will be an increase in music releases by independent record labels.

Hypothesis 1b: Subsequent to the technological changes in the music industry, there will be a decrease in the numbers of music releases by major record labels.

We further examine organization search, distinguishing between ‘discovery,’ i.e. releasing music products from new-to-the-world artists, and ‘selection,’ i.e. releasing music products from already- successful artists. Selection can be further conceptualized in two ways: 1) a music label organization can release new music by proven artists that have already had releases/success on its own label, i.e. the exploitation of its own prior success, which we term internal selection, or 2) a music label organization can find and release new music by previously successful artists on another record label, which we term external selection.

As we describe above, the major record labels have developed interrelated capabilities for music production and promotion that were highly valuable in the past and continue to be valuable complementary assets in the industry post-change, but now for a narrowing set of the most commercially promising or highest expected revenue recording artists. These complementary assets are also expensive, and as we note above, the appropriability crisis in the industry is likely to create additional pressure, particularly on these organizations, to release music that is more likely to have higher revenues. This will result in such organizations becoming more selective, specifically selecting a narrower set of more promising music releases than in the past, reducing the overall number of music releases they produce, but potentially increasing the revenue associated with each music release. For reasons we describe above, these major, established organizations are both less likely to adopt and slower to adopt the lower cost activities to produce and promote music, while at the same time, they grow less able to protect content and generate revenues. Thus, they are less able to engage in talent discovery in traditional ways. As a result, we expect that the new products released by these organizations will shift toward greater selection of already-proven talent, entailing an increase in both internal and external selection after the technological change. Thus we hypothesize specifically that after the technological change, there will be

16 a larger increase overall in selection of previously-successful artists by major label organizations than by non-majors, and further, that the major label organizations will increase both their internal and their external selection subsequent to the technological change.

Hypothesis 2: Subsequent to the technological change, there will be a larger increase in selection (music releases that utilize already-successful talent) at major record labels than at the independent labels.

Hypothesis 3: Subsequent to the technological changes in the music industry, the major record labels will increase both internal selection, i.e. talent that has been previously released and has had prior success on the same record label, and external selection, i.e. talent that has previously been released and has had prior success on other record labels.

In Hypotheses 3, our focus is on the major record labels, in line with our argument that the persistent value of their complementary assets in the smaller, higher value niche of the market will drive them to be more selective to offset the weakened appropriability. The implication of our argument is further, that the major labels will increase both internal and external selection on prior success more than the independent labels, a comparison we test below.

Revenue outcomes

As the technological change combined with prior capabilities of the major labels spurs a shift toward selection of previously proven talent in the major record label organizations, i.e. we would expect major labels to focus more on predictably successful artists, and subsequently predict that the average expected revenue of their releases would rise. As prior research suggests, traditionally about 10 percent of new releases cover their costs (Caves, 2000). However, products also vary in the degree to which their appeal is predictable. Sequels to successful movies are less risky than new concepts, and novels by well- known authors (Tom Clancy, John Grisham) are less risky than first novels by unknown writers.

Similarly, follow-up albums by already-successful artists are less risky than albums by untested artists.

Thus, we expect the combination of organizational shifts to a focus on selection of talent with higher expected success combined with the greater predictability of success for the follow-on releases of these successful artists would result in improved revenue performance. Further, this also suggests that the

17 lower bound of expected success for major label releases would rise, and thus, we would expect the current success of the major label’s new releases to rise.

Hypothesis 4: To the extent that music organizations engage in selection (both internal and external selection of previously proven talent, (based on expected value), the current success (realized value) of their music releases will increase.

Data Sources

Our data consist of 63,271 recorded music albums released 1990-2010 and is collected from

Discogs.com. Because our focus is on the creation of new products, we exclude compilations and re- releases. The Discogs data indicate the artist name, album title, release year, and the name of the record label releasing the album. The Discogs data do not include sales measures for the albums. While album sales data exist – and the A.C. Nielsen Company is the main source – these data are prohibitively expensive. However, we were able to get a measure of relative album sales from both the Billboard 200 and the Billboard 50, weekly rankings of the top 200 and top 50 albums by U.S. sales. We collected these data for each week, from 1985 to 2010. We then aggregated the Billboard data by year and recording artist (or group) to create a dataset with the annual number of listings on the Billboard 200 or Billboard top 50 for each artist who appears on these weekly rankings. We then created a mapping to match the artists between Discogs and the same artists’ names in the Billboard rankings, providing us a dataset with ranking based sales information by artist-year. In some cases the mechanical exact mapping was not effective due to typos and other differences in artist names. In those cases, research assistants manually matched as many artists as possible across the two databases.

Our study focus requires us to further classify record labels as major or independent. Because the

Discogs data include 14,982 distinct labels, this is a difficult task to accomplish completely. However, using the label classification in Thomson (2009), we can classify 456 labels as independent and another

137 labels as major.7 This leaves a large number of labels unclassified, but since we observe that the unclassified labels have an average of 3.6 albums per label over the period 1990-2010, in contrast with

7 See Same Old Song (http://futureofmusic.org/article/research/same-old-song ).

15.4 for the classified independent labels and 66.2 for the classified major labels, we conclude that these labels are also independents.

The Discogs data include 63,721 albums, while the Billboard data include certifications for 5,252 artists and 15,917 artist-years. We merge the Discogs and Billboard data, as of each year, including both a measure of past sales for the artist (number of Billboard chart listings prior to this year) and a measure of the sales success for the current release (the number of Billboard listings for this year). We develop measures of the extent to which a record label organization’s releases are debuts of new-to-the-world artists (i.e. discovery), and the extent to which the organization’s releases build on previously established, successful artists (i.e. selection, both internal and external). We are also able to calculate the ‘success’ of the releases of the record label’s albums in each year as a measure of current performance. We use the measure of ‘success’ both to determine past success and thus the organization’s use of talent that previously has been successful, as well as the relative performance outcomes from different search

(variation/discovery versus selection) behaviors.

Measures of Organizational Discovery and Selection

Our first measure is the number of albums released by a music organization, which we calculate by year and by label type (independent versus major). We classify releases according to whether this album is the artist’s first appearance in the dataset8 and if not, according to whether the artist has prior

Billboard chart appearances. Next, if an album is not an artist’s first appearance, we determine whether the album is a re-appearance for that artist on the same label, versus whether the artist is new to a label

(i.e. organization) but not entirely new to the market.

From these initial measures, we create organization/year measures of how selective an organization’s ‘products’ (new music releases) are each year, as well as the performance outcomes in the current year (appearance on the Billboard top 200 or Billboard top 50 rankings each week). Thus, we can

8 Some artists appear ‘new’ to the world in our dataset of music releases because this is their first solo album recorded on a label, but as artists, they may be not entirely new to the world. For example, some recording artists may have had successful releases as part of a previously successful band, although the focal release is their first solo recording. We conducted media searches to better understand how frequently artists might already be ‘known’ from prior music releases.

19 calculate the level of past success of the artists an organization selects for its music releases, as well as the current success of those music releases, for years before and after the technological change.

We measure discovery as the percent of the music releases for an organization/year involving artists that have not been released previously on any label, i.e. this is the first music release for a ‘new to the world’ artist.

We measure our first selection concept, internal selection, by the music releases for an organization/year involving artists that have already been released by the same organization and previously have achieved some success (Billboard top 200 or Billboard top 50).

We measure our second selection concept, external selection, by the music releases for an organization/year involving artists that have already been released on other labels, and previously have achieved some success (Billboard top 200 or Billboard top 50).

We measure the performance of a release (or a set of releases from an organization) using data from the Billboard top 200 and Billboard top 50 rankings. The number of weekly appearances in these rankings provides a basic measure of success. The ranking can also be used along with overall national sales trends to more accurately approximate patterns of sales quantities. Sales ranks, such as the Billboard

200, provide information on relative sales (i.e. that the nth ranked album outsold the (n+1)st), but they provide no information on the absolute level of sales. While some of our hypotheses involve a comparison of sales to major and independent record labels and can therefore be carried out using data based on rankings, it is nevertheless useful to create indices that reflect the overall levels in sales. Figure

4 shows the time pattern of overall US recorded music sales, 1989-2011 (in $2010). As the figure shows, sales rose steadily until 1999 and have since declined substantially.

To construct sales indices reflecting the time pattern of aggregate sales, we do two things. First, we distinguish between the sales level of music releases at different ranks. There is a robust research tradition of translating sales ranks into sales quantities using the following relationship: 푞 = 퐴푟퐵, where q

20 is the quantity sold and r is the sales rank.9 The parameter B reflects how quickly sales fall off at lower ranks (higher values of r). Studies generally find B to be in the neighborhood of -1, so that sales are proportional to (1/r).10 Second, we need a value of A to reflect how particular ranks might map differently into sales across years. For a pass at this we use the data on the overall value of record sales in the US from the Recording Industry Association of America (RIAA) in Figure 4. Define St as sales in year t. We then construct indices of sales as 푞 = 푆 ∑ 1 , where o refers to an organization. From 표푡 푡 푖∈표 ⁄푟푖푡 this we approximate the organization o’s sales in year t as the sum of the reciprocal of its albums’ weekly sales ranks, weighted by the real value of overall sales in that year. The absolute level of the resulting index is not meaningful, but its time pattern will provide a reasonable approximation to organization’s the time pattern of revenue. Note that the organization can literally be an organization such as a particular label, or it can be a group of labels, or even the entire set of labels.

Thus, we can calculate the extent to which an organization’s music releases entail new to the world, unproven talent, versus build on its own proven, successful artists from prior years, versus build on proven successful artists that have been released on other labels in prior years. We are able to study these measures of organization search before and after the technological changes. We are also able to measure the performance outcomes of these different search strategies.

Methods

We test our hypotheses by asking whether various measures of interest – a label’s numbers of releases, the share of a label’s releases that are albums by artists with prior chart success, the share of a label’s releases that achieve chart success – vary over time with the technological change. The reduction in appropriability is easily linked to the appearance of Napster in 1999. While some aspects of the cost reduction occur prior to the Internet (such as the development of the compact disc in 1985), most of the cost reductions occur some time around 2000. Digital distribution became viable with the appearance of the iTunes Music store in 2003. Online media spreading information about new music have grown over

9 See Chevalier and Goolsbee (2003), among other papers. 10 See Chevalier and Goolsbee (2003), Brynjolffson, Hu, and Smith (2003), and Ferreira and Waldfogel (2013).

21 time. Pitchfork appeared in 1995, and Metacritic appeared in 1999.11 There is no single date when cost reduction appeared. Instead, we distinguish between the period since 1999 and the period up to 1999 as a broad distinction between the period before the recorded music industry faced the challenges of technological change.

Accordingly, we illustrate our results in two ways. First, we present figures illustrating and contrasting trends. Second, we test our hypotheses using two kinds of before-and-after models. We test some hypotheses by asking whether the level of a variable of interest changes for a group of labels

(majors or non-majors) following 1999:

푝표푠푡 푦푖푡 = 휇푖 + 훼훿푖푡 + 휀푖푡 ,

푝표푠푡 where yit is an outcome of interest for label i in year t, 훿푖푡 is an indicator that is 0 until 1999 and 1 thereafter, 휇푖 is a label fixed effect, and 휀푖푡is an error term. In this model the coefficient 훼 shows how an outcome at a particular group of labels (major or non-major) varies between the post-change period and the pre-change period.

We also estimate “difference in difference” models that ask whether the outcome of interest varies at major labels relative to its evolution at the non-major labels:

푝표푠푡 푚푎푗표푟 푦푖푡 = 휇푖 + 휃푡 + 훼훿푖푡 훿푖 + 휀푖푡,

푚푎푗표푟 where variables are defined as before, and 훿푖 is an indicator for major labels, and 휃푡is a time effect.

푚푎푗표푟 (Note that we do not include 훿푖 alone as it is subsumed in the label fixed effects, nor do we include

푝표푠푡 훿푖푡 alone as it is subsumed in the year fixed effects). In this model 훼 shows the extent to which the outcome variable at majors deviates from the time pattern at non-major labels.

Findings

Table 1 shows the total number of albums released in the U.S. The rising number of overall releases obscures the difference between the time pattern of releases at major and independent record labels. The second column of Table 1 shows the annual U.S. album releases from known major record

11 See http://en.wikipedia.org/wiki/Pitchfork_Media and http://en.wikipedia.org/wiki/Metacritic .

22 labels. The number of annual releases from this group of labels fluctuates year to year, but it reaches a recent peak of 543 albums in 1998, the last year before Napster. From 1999 through the end of the sample period, reaching 319 in 2007, which is 41 percent below the 1998 release level. Releases from known independent labels, as well as labels whose type is unknown but which are quite likely independent, rise sharply over the same period. Releases from known independents rise from 290 in 1998 to 431 in 2007, while releases from small labels of unknown type rise from 2,253 to 3,486 over the same period.

These results suggest that appropriability challenges brought about by the technological change and file sharing clearly spurred a decrease in music releases at the incumbent major labels, but at the same time, the lower costs to produce music also increased the number of albums released by other labels

(independents). We further tested these hypotheses using a differences-in-differences model, assessing the number of music products released before and after the technological change, and for major labels versus independents. The results are shown in Table 2. Columns (1)-(3) use label type-years as observations, showing that the total annual number of major-label releases fell by about 52 percent after

1999, while independent releases rose about 62 percent. Column (3) includes time dummies to flexibly account for time patterns common across label types. Relative to the independents, the majors fall by 90 percent.12

Columns (4)-(6) use organization-level data, and the results show that music releases from the major labels decreased by 35 percent after the technological change (column 1), while music releases from the non-majors increased by 4 percent after the technological change (column 2). Both of these results are significant at <.01. Column 3 includes year dummies to allow for a flexible time pattern, along with a major x post-1999 interaction, to assess whether the pattern for major and non-major labels diverges after 1999. The results in column 3 show further that the music releases for the major labels decreased by 39 percent per year relative to releases at non-major labels. Thus, these results support hypotheses 1a and 1b.

12 Note that 1-e-2.27 = 0.9.

We further examined whether the major labels shift from discovery of new-to-the-world music to selecting on already proven, less risky artists, and whether they do this more than the non-major labels

(hypothesis 2). Figure 5 shows the new artist share of the releases of the major record label organizations over time. While the new artist share averaged about 35 percent prior to 1999, it has since dropped and averages roughly 28 percent. That is, the new-artist share at majors has fallen about a fifth (7/35). The new-artist share at non-major labels has also fallen, but it remains about double the share at majors. This suggests an overall decrease in the extent to which major label organizations are discovering ‘new to the world’ talent and therefore an increase in the extent to which firms are selecting on already-proven talent.

The left side of Figure 6 shows the share of new music releases by major-label organizations from artists with prior sales success (i.e. already-proven artists). As we note above, whether an artist has already been successful is measured by whether their albums have previously appeared among the

Billboard weekly top 200, or weekly top 50, in the Billboard album sales chart. For major-label releases during the 1990s, the share of releases with artists that had previous top-200 success averages about 17 percent. From 2000 to 2010 the share rises fairly steadily to nearly 40 percent. (The pattern for the top-

50 measure is similar but lower: about 12 percent in the 1990s, rising past 30 percent by 2010).

Some of the increase is inherent in the measure: Because the measure of past success – whether a releasing artist has ever (since 1985) previously appeared in the Billboard 200 – is cumulative, it is by nature increasing. The right side of Figure 6 shows the share of independent releases from already- successful artists. It is clear that the absolute level of the share of releases from already-successful artists is always higher at the majors. The increase for major labels is substantial, a tripling in the share of new releases by already-successful artists, and therefore, a dramatic change in the extent to which the major record label organizations shift toward selecting on already-successful artists versus discovering new artists in their search for talent. While the share of already-successful artists rises at non-majors, it remains low, in 2010 about one tenth the share at majors.

We further tested hypothesis 2 using a differences-in-differences regression, comparing the selection of already successful artists for music releases before and after the technological change, and for

24 major labels versus non-major labels. Our results are shown in Table 3. The results show that after the technological change, both major labels and non-major labels significantly increased their selection of already-successful artists that had previously had music releases on the Billboard top 200 (in columns 1 and 2), and the Billboard top 50 (in columns 4 and 5). These results are significant at <.01. Further, columns 3 and 6 directly show that selection on successful artists was even greater at the major labels after the technological change. These results support hypothesis 2.

We further explore the specific sources of the product releases for the major incumbent organizations, to determine the extent that they reflect internal selection (artists who have previously been successful on the same record label) versus external selection (artists who have previously been successful with releases on other record labels). These are shown in Figure 7.

The chart on the right side in Figure 7 suggests that the major record labels have always relied on internal selection, i.e. releasing follow up music by their own previously released successful artists.

Between 1990 and 2000, 40 percent of major-label follow up albums were by artists whose previous albums had appeared on the Billboard 200. But this has grown between 2000 and 2010, as the already- successful share of major-label internal releases rose to 60 percent. The chart on the left side of Figure 7 shows that major labels have traditionally focused less on selecting previously successful artists externally (i.e. external selection): In the 1990s the already-successful share of external major-label releases stood at just 10 percent. Since 2000, however, the share has risen sharply, reaching 50 percent by

2010. Although in hypotheses 3 we focus mainly on the increase in selectivity of the major labels as they face the issue of weak appropriability and a reluctance to shift to lower cost activities, we also expect this focus on selectivity to be greater for major label organizations than for non-majors. Figure 8 provides the comparison for the non-major labels.

Findings about subsequent revenue outcomes

We documented above that major labels rely more heavily on already-successful artists and that this reliance – for both major and non-major labels – grows over time, although more dramatically for the major labels. Our hypotheses suggest that reliance on already-successful artists is a search strategy

25 arising from the need to make investments with more certain revenues. We can ask whether this approach is successful (i.e. whether selecting on expected revenue results in greater realized revenue) by measuring the sales or revenues of music releases. Because the major labels (e.g. Sony, Universal) are owned by large conglomerates, it is unfortunately not possible to measure the profit performance of their record label businesses. While overall sales in the recorded music industry are falling over time, our sales rank measures are unaffected by changes in sales levels. We can roughly measure the success of current releases according to whether the artist appears on the Billboard album chart, and the relative ranking on the Billboard top 50 or top 200 during the year of the release. From there, we are able to convert the

Billboard rankings into estimates of revenue as we describe in more detail below.

As Figure 9 shows, the albums released by major labels have growing probabilities of ‘success’ over time, here measured as the artist’s appearance and ranking on the weekly Billboard 50 or 200 lists during the current (i.e. release) year. The share of music releases by the major labels that appear in the

Billboard 200 rises from 25 percent in the late 1990s to over 50 percent by 2010. A comparison with the music releases by non-major labels shows that the corresponding probabilities of success for non-major label releases are low – around 5 percent – throughout the period. In Figure 10 we show estimated revenues (‘pseudo-sales’) corresponding to the Billboard rankings. Although the probability of ‘success’ is rising for major label music releases after 2000, because overall national sales are declining, our estimated sales measure in Figure 10 is also declining somewhat after 2000.

We then total the Billboard 200 listings across the music releases for both major labels and non- major labels to develop a measure of the total sales attributable to each label type. This is shown in

Figure 11. The major-label albums account for declining Billboard chart appearances, while the non- major label albums account for a growing number of Billboard chart appearances. In Figure 12 we show estimated revenue (“pseudo-sales”) by year, aggregated up to major labels and non-major labels. Major label revenue is falling overall while non-major revenue is stable. While the major-label strategy of focusing increasingly on predictably successful artists is successful at the level of the individual music

26 release (i.e. major label artists are more likely to be the top selling records that appear on the Billboard lists), major record labels a declining share of total industry revenue.

To examine how successful the increase in ‘external’ and ‘internal’ selection is for the major labels, in Figure 13 we show the current success of music releases for the major labels by their source, internal or external selection (i.e. the share of externally or internally selected music from the major labels that appears on the Billboard top 50 or top 200 listings). The ex post success of the major-label releases confirms the label organization’s likely expectations for increased success from an increase in selection from both internal and external sources. The success rate from internal selection has been stable and high from 1990-2000 at 40 percent, and it has since risen to 60 percent. Thus, efforts to select on more predictably successful internal artists has resulted in greater success in the form of consistently higher revenues for internal selection. Ex post success of external selection (selecting on already successful artists released by other labels) has risen even more sharply. The success rate from external selection stood around 20 percent during the 1990s but has since risen to 50 percent.

Interestingly, the success rate of artist debuts (i.e. discovery of new-to-the-world-artists) for the major labels has also risen. It stood around 10 percent prior to 2000 and has since risen to 30 percent.

This supports the idea that major labels have been selecting a narrower set of more promising artists than in the past, which translates into greater success rates, even in the outcomes of discovery of ‘new-to-the- world’ artists. It is possible that discovery grows more successful in this context as record labels are able to know more about some artists’ prospects even prior to their first recordings. One such source of information is user response to artist content posted at YouTube. It is well known that Justin Bieber was discovered at YouTube.13 To explore this further, we conducted media searches to understand how recording artists may have been ‘discovered’ to assess how many of the ‘new to the world’ artists

13 See Lizzie Widdicombe. “Teen Titan: The man who made Justin Bieber.” The New Yorker, September 3, 2012 ( http://www.newyorker.com/magazine/2012/09/03/teen-titan ).

27 debuting on major labels in 2010 were discovered on YouTube. About a dozen other artists were also discovered at YouTube in 2010 (see Table 6).

Table 4 shows the results of our analysis of the success of releases from major and non-major labels after 1999. The table employs the same structure as Table 3 (with a different dependent variable

(success of music releases rather than selection). As Table 4 shows, there were increases in the successful share of current releases for both major labels and non-major labels after 1999, but the increased success at the major labels was much larger than at the non-majors. Relative to the time pattern at non-major labels (columns 3 and 6), the increase in the successful share at major labels is substantial (about 11 percentage points) and statistically significant.

Table 5 examines whether the labels that exploit more aggressively – as indicated by a higher share of releases from already successful artists – are those with greater current success, i.e. do they put a higher share of their current releases into the Billboard 200 or 50. Columns 1-3 use the Billboard top 200 measure (BB200), while columns 4-6 use the Billboard top 50 measure (BB50). All specifications show a higher success rate in release years in which a higher share of the releases are from artists who have already achieved chart success. Results hold when the sample is restricted to either just major labels, just non-major labels, or all labels together. These findings support hypothesis 4.

Discussion

In this paper, we study the influence of two different aspects of the technological change in the music industry: the difficulty appropriating revenues in the face of file sharing enabled by MP3 technology, and the potential for cost reductions in activities in production, distribution, and promotion of recorded music, also enabled by the technological change. It seems clear that the change in the ability to appropriate revenues and profits spurred by the technological change and Napster has made it difficult for major labels to discover new talent in traditional ways, suggesting both that they are challenged with adaptation to this important technological change, and that we might expect innovation in the industry to decline as these organizations can no longer incur investments to discover new talent. It appears these major challenges spur the outcomes we predict – they have caused major labels to reduce their efforts to

28 discover new talent and to focus instead on selection – identifying and releasing music by more predictably successful artists that have already been released by a record label). While in the past, major labels often released new music from their own roster of previously successful talent, this has increased since the technological change. In addition, the major record labels have also dramatically increased their external selection following the technological change – the search for – and selection of – external talent that has already been successful on other labels.

Our results suggest further that the non-majors, i.e. independent record labels, produce more music products after the technological change. This suggests support for our arguments that they have taken advantage of reduced costs enabled by technological change, and are now able to cover the (now lower) costs to produce, distribute, and promote more music. Our findings showing a reduction in the music releases by major labels since the technological change suggests that in contrast to the independent labels, major labels have not adopted lower cost approaches. The combination of an appropriation crisis that lowers revenue and costs that remain high have led fewer music products to appear promising, i.e. with revenues to cover their costs to produce. This is further supported by anecdotal evidence from the industry that major labels have continued to employ their high cost capabilities for music production, distribution, and promotion. While this appears to be yet another example of the sort of incumbent rigidity and inertia we might expect in the face of new technologies, this persistence in extant activities and capabilities is also likely a result of the continued value of incumbents’ complementary assets specifically for the most profitable segment of the market – the most promising artists are likely to benefit from – and demand – higher cost distribution and promotion capabilities. In turn, the usefulness and value of these interlinked traditional capabilities likely pushes the major labels to continue to select the types of predictably promising talent that will benefit from these relatively expensive and interrelated capabilities.

Thus, the result of the technological change is that discovery of new-to-the-world artists and products is reduced within the major label incumbent organizations. However, at the same time, organizations shift to focus on the most promising talent, increasing their success with each new music

29 release. By shifting their organizational focus to selection rather than discovery of promising products, particularly the dramatic increase in external selection of already-proven talent that has been released on other labels, the relative success of their music releases improves, measured by appearance in the

Billboard top 50 or top 200. However, our estimates of revenue associated with these successes suggest that overall revenue declines.

A well-established idea in strategy research is that at the organization level, although concerted exploitation can lead to better short-term performance, exploration into novel domains is necessary for longer-term success (March 1991, March & Levinthal, 1993). Research in this area has underscored the importance of maintaining a balance between exploration and exploitation (e.g. O’Reilly and Tushman,

2008; Benner & Tushman, 2003), corresponding in our case to the idea of ‘discovery’ versus ‘selection.’

A question raised by our study is the extent to which the hyper-exploitation we observe in this setting as firms resort to selecting on the most promising talent will enable these organizations to succeed long term. It is important to understand how longer term outcomes compare for firms adopting this focus-on- selection strategy and firms that continue to pursue discovery of new-to-the-world artists.

The outcomes for the smaller independent record labels contrasts markedly with the story that emerges from our results for the major labels. It is apparent that the independent labels adopt lower cost ways to produce, distribute, and promote new music releases. Thus, despite the change in appropriability, adopting lower cost activities spurs them to produce more music after the technological change. This is an important underlying mechanism in Waldfogel’s (2012a) finding that the quality or variety of music for consumers has not declined post-Napster, despite the warnings from the executives of major record labels that an inability to appropriate profits would lead to less music discovery. Within the industry generally, the influence of the technological change has been to shift the locus of innovation away from the major incumbent organizations to the smaller independents.

Although our study has specifically focused on an important technological change and outcomes for organizations in the music industry, the findings from our rich set of data on new product introductions for multiple organizations in an industry are generalizable to other settings of technological

30 change. The changes in appropriability and cost reductions are features of other shifts in technology, particularly those associated with digital convergence or digitization. Our findings uncover an important mechanism driving incumbents’ responses to new technologies that has not been explored in prior work.

Here, incumbents continue to retain their high cost capabilities, partly for cognitive reasons – the belief that entities adopting lower cost capabilities are lower-quality ‘amateurs,’ while the traditional large incumbents are ‘professionals,’ but also for a reason more directly related to short term financial performance: Although the technological change appears to be competence destroying on its face, the interrelated set of expensive capabilities that incumbents possess actually retains its value for a very small set of new product releases – those of the most promising, highest value products. Thus, incumbents appear to have incentives not to reduce costs, and the continued use of their capabilities and focus on greater selection appear to be complementary.

Limitations and opportunities for future research

Our findings raise several questions for future research. One question is the impact of this external stress on the industry as a whole: do risky artists continue to get released, even if not by major labels? (That question, and the concomitant impact on consumers, are the topics of research elsewhere).

A second question that is not clear from our study is the long term result of the reduction in variation within-firm. Specifically it is not clear whether selecting on the variation produced by external sources substitutes for variation-creation in organizations for longer-term performance.

We are not able to study costs directly in this study, although our findings are consistent with major labels failing to adopt lower cost approaches. One aspect of cost that would be useful to study in subsequent work pertains especially to the increase in external selection. While the traditional approaches for internal discovery by the major labels had characteristics of sustainable competitive advantage

(discovering talent that others do not yet see, suggesting that organizations might be able to make relatively small investments and benefit from relatively large gains later), acquiring already-successful artists potentially subjects interested major labels to higher prices and bidding wars. Thus, although we

31 can observe higher average revenues for the major labels, in part due to the dramatic increase in external selection or exploitation, we cannot directly assess the profit associated with these acquisitions.

Figure 4

Value of US Recorded Music Shipments ($2010) 20000 18000 16000 14000 12000 10000 8000 6000 4000 2000 0 1985 1990 1995 2000 2005 2010 2015

Figure 5

New Artist Share of Releases major non-major

1990 1995 2000 2005 20101990 1995 2000 2005 2010 year Median bands new share

Graphs by type1

Figure 6

Prob of Past Success by Release Type major non-major

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

mean of dpast50 mean of dpast200

Graphs by type1

Figure 7

Past Success of Major Releases by Search Source external internal

meanof dpast200

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by stype

Figure 8

Past Success of non-Major Releases by Search Source external internal

.08

.06

.04

meanof dpast200

.02

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by stype

Figure 9

Prob of Current Success by Release Type major non-major

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

mean of d50 mean of d200

Graphs by type1

Figure 10

Average Level of Current Pseudo-Sales by Release Type major non-major

4,000

3,000

2,000

meanof ps1

1,000

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by type1

Figure 11

Total Level of Current Success by Release Type major non-major

4,000

3,000

2,000

1,000

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

sum of n50 sum of n200

Graphs by type1

Figure 12

Total Level of Current Pseudo-Sales by Release Type major non-major

2.0e+06

1.5e+06

1.0e+06

sumof ps

500000

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by type1

Figure 13

Successful Share of Major Releases by Search Source external internal

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

new

meanof d200

199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by stype

Figure 14

Successful Share of non-Major Releases by Search Source external internal

.15

.05

199019911992199319941995199619971998199920002001200220032004200520062007200820092010 199019911992199319941995199619971998199920002001200220032004200520062007200820092010

new

.15

meanof d200

.05

199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Graphs by stype

Table 1: Releases by Type

year known known unknown Total independent majors 1990 145 785 754 1,684 1991 177 627 927 1,731 1992 188 551 1,034 1,773 1993 231 581 1,335 2,147 1994 251 581 1,579 2,411 1995 262 525 1,943 2,730 1996 219 458 1,613 2,290 1997 258 540 2,236 3,034 1998 290 543 2,253 3,086 1999 327 451 2,489 3,267 2000 314 450 2,465 3,229 2001 363 413 2,421 3,197 2002 365 413 2,536 3,314 2003 370 372 2,647 3,389 2004 423 336 2,855 3,614 2005 435 351 3,053 3,839 2006 447 326 3,302 4,075 2007 431 319 3,486 4,236 2008 379 273 3,381 4,033 2009 349 263 3,053 3,665 2010 274 200 2,503 2,977 Note: Labels in Discogs are matched with the label list in Thompson (20??) to produce our list of known major and known independent labels.

Table 2: Releases by Majors and Non-Majors before and after Technological Change

(1) (2) (3) (4) (5) (6) log number of log number of log number of log number of log number of log number of major non-major releases major firm indep. firm firm releases releases releases releases releases post '99 -0.5181 0.6202 -0.3515 0.0414 (0.0899)** (0.1237)** (0.0357)** (0.0113)** post '99 x -2.2735 -0.3922 major (0.2716)** (0.0323)** Constant 6.3249 7.4600 6.7335 1.1385 0.3026 0.3890 (0.0651)** (0.0895)** (0.4505)** (0.0234)** (0.0081)** (0.0301)** Observations 21 21 42 1836 29482 31318 R-squared 0.64 0.57 0.78 0.06 0.00 0.01 Label FE No No No Yes Yes yes Year no no yes no no yes dummies Notes: * significant at 5%; ** significant at 1%. The dependent variable is the log of the number of releases. Columns (1) – (3) use the label type (major vs non-major)-year as an observation. Columns (4) – (6) use the organization –year as an observation. Columns (3) and (6) include year dummies to flexibly account for common time patterns.

Table 3: Selection by Majors and non-Majors before and after Technological Change (1) (2) (3) (4) (5) (6) share in past share in past share in past share in past share in past share in past BB200 - BB200- BB200- BB50- BB50 –non- BB50 - major non-major all major major all post '99 0.1235 0.0411 0.1098 0.0241 (0.0171)** (0.0036)** (0.0151)** (0.0026)** post '99 x 0.0681 0.0770 major (0.0111)** (0.0084)** Constant 0.1877 0.0225 0.0032 0.1135 0.0081 0.0001 (0.0112)** (0.0026)** (0.0103) (0.0099)** (0.0018)** (0.0078) Observations 1836 29482 31318 1836 29482 31318 Number of 416 14566 14982 416 14566 14982 labels Year No No Yes No No yes dummies R-squared 0.04 0.01 0.03 0.04 0.01 0.03 Notes: Regressions of share of annual label releases by artists previously appearing in the Billboard 200 or Billboard 50. Standard errors in parentheses. * significant at 5%; ** significant at 1% . Record label fixed effects are included.

Table 4: Success by Majors and non-Majors before and after Technological Change (1) (2) (3) (4) (5) (6) share in share in share in share in share in share in BB200 this BB200 this BB200 this BB50 this yr BB50 this yr BB50 this yr yr yr yr major non-major all major non-major all post '99 0.1533 0.0320 0.1403 0.0157 (0.0167)** (0.0031)** (0.0147)** (0.0021)** post '99 x 0.1109 0.1180 major (0.0098)** (0.0074)** Constant 0.1998 0.0118 0.0040 0.0942 0.0039 -0.0094 (0.0110)** (0.0022)** (0.0091) (0.0096)** (0.0015)* (0.0069) Observations 1836 29482 31318 1836 29482 31318 Number of 416 14566 14982 416 14566 14982 labels Year No No Yes No No yes dummies R-squared 0.06 0.01 0.03 0.06 0.00 0.04 Notes: Regressions of share of annual label releases appearing in the Billboard 200 or Billboard 50. Standard errors in parentheses. * significant at 5%; ** significant at 1%. Record label fixed effects are included.

Table 5: Selection and Success (1) (2) (3) (4) (5) (6) share in share in share in share in share in share in BB200 this BB200 this BB200 this BB50 this BB50 this BB50 this yr yr yr yr yr yr major Non-major all major Non-major all share of past 0.5255 0.3637 0.4062 releases in BB200 (0.0217)** (0.0064)** (0.0062)** number of -0.0020 0.0003 -0.0013 -0.0019 0.0001 -0.0016 releases (0.0014) (0.0005) (0.0004)** (0.0013) (0.0003) (0.0003)** share of past 0.4607 0.3345 0.3851 releases in BB50 (0.0224)** (0.0062)** (0.0062)** Constant 0.0997 0.0031 -0.0020 0.0196 -0.0035 -0.0138 (0.0349)** (0.0081) (0.0082) (0.0316) (0.0057) (0.0062)* Observations 1836 29482 31318 1836 29482 31318 Number of labels 416 14566 14982 416 14566 14982 R-squared 0.36 0.19 0.23 0.33 0.17 0.21 Notes: All regressions include year effects. Standard errors in parentheses. * significant at 5%; ** significant at 1%. Record label fixed effects are included.

Table 6: YouTube Discoveries by Major Labels, 2009-2010 artist debut album label Straight No Chaser With A Twist Atlantic Zee Avi Zee Avi Brushfire Records Priscilla Renea Jukebox Capitol Records Cee Lo Green The Lady Killer Elektra Little Boots Hands Elektra Chiddy Bang The Preview EMI Fisher True North Interscope Records Samantha Mumba Gotta Tell You Interscope Records Justin Bieber My World Island Records Kristinia DeBarge Exposed Island Records Mike Posner 31 Minutes To Takeoff J Records Don Philip Don Philip Jive Teeh Manibusan Lesson Learned Pacific Island Entertainment Guy Sebastian Like It Like That Sony Music Sick Puppies Tri - Polar Virgin

REFERENCES

Adner, R. and Snow, D. 2010. Old technology responses to new technology threats: Demand heterogeneity and technology retreats. Industrial and Corporate Change 19(5): 1655-1675.

Aguiar, L. and J. Waldfogel, 2014. Panning for gold: The random long tail in music production. Working paper.

Benner, M. J. 2010. Securities Analysts and Incumbent Response to Radical Technological Change: Evidence from Digital Photography and Internet Telephony. Organization Science, 21(1): 42-62

Benner, M.J. and Ranganathan, R. 2012. Offsetting illegitimacy? How pressures from securities analysts influence incumbents in the face of new technologies. Academy of Management Journal.

Benner, M.J. and Tripsas, M. 2012. The influence of prior industry affiliation on framing in nascent industries: the evolution of digital cameras. Strategic Management Journal 33(3): 277-302.

Blackburn, D. 2004. Online piracy and recorded music sales. Working paper, Harvard University.

Bourreau, M., M. Gensollen, F. Moreau, 2012. The impact of a radical innovation on business models: Incremental adjustments or big bang? Industry and Innovation, 19:5: 415-435.

Brynjolfsson, E., Hu, Y. and Smith, M.D., 2003. "Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Booksellers," Management Science, INFORMS, vol. 49(11), pages 1580-1596, November.

Caves, R. 2000. Creative industries: Contracts between art and commerce. Harvard University Press

Christensen, C. 1997. The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail. Harvard Business School Press: Boston, MA

Cohen, W. and Levinthal, D.A. 1990. Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly 35(1) 128-152.

Chevalier, J. and Goolsbee, A. 2003. "Measuring Prices and Price Competition Online: Amazon.com and BarnesandNoble.com," Quantitative Marketing and Economics, Springer, vol. 1(2), pages 203-222, June.

Ferreira, F. and Waldfogel, J. 2013. "Pop Internationalism: Has Half a Century of World Music Trade Displaced Local Culture?," Economic Journal, Royal Economic Society, vol. 123, pages 634-664, 06.

Furr, N. and Snow, D. 2012. Threat rigidity or threat action? Rethinking threat response in light of technology’s last gasp. BYU working paper.

Handke, C. 2012. Digital copying and the supply of sound recordings. Information Economics and Policy, 24: 15-29.

Henderson, R. 1993. Underinvestment and incompetence as responses to radical innovation. Rand Journal of Economics, 24: 248-269.

Henderson, R.M. & Clark, K. B. 1990. Architectural innovation: The reconfiguration of existing product technologies and the failure of established firms. Administrative Science Quarterly, 35: 9-30.

Kalmar, V. 2002. Label Launch: A Guide to Independent Record Recording, Promotion, and Distribution. St Martin’s: New York.

Landes, D. 1983. Revolution in Time: Clocks and the Making of the Modern World. Cambridge: Harvard University Press.

Leeds, J. 2005. The Net is a boon for indie labels. New York Times, December 27.

Lemann, Nicholas, 2006. Amateur Hour: Journalism without journalists. New Yorker, August 2006.

Leonard-Barton, D. 1992. Core capabilities and core rigidities: A paradox in managing new product development. Strategic Management Journal, 13: 111-125.

Levinthal, D., 1997, Adaptation on Rugged Landscapes, Management Science, 43(7): 934-950.

Levinthal, D. & March, J.G. 1993. The myopia of learning. Strategic Management Journal, 14: 95-112.

Liebowitz, S. J. 2011. The metric is the message: How much of the decline in sound recording sales is due to file sharing? Available at ssrn.

Liebowitz, S. 2006. File sharing: Creative destruction or just plain destruction? Journal of Law and Economics, 49(1): 1-28.

March, J. 1991. Exploration and exploitation in organizational learning. Organization Science, 2: 71-87.

Oberholzer-Gee, F. and K. Strumpf, 2007. The effect of file sharing on record sales: An empirical analysis. Journal of Political Economy, 115: 1-42.

Mitchell, W. 1989. Whether and when? Probability and timing of incumbents’ entry into emerging industrial subfields. Administrative Science Quarterly, 34(2): 208-320.

Nelson, R. and Winter, S. 1982. An evolutionary theory of economic change. Belknap Press: Cambridge, MA.

Rob, R. and J. Waldfogel, 2006. Piracy on the High C’s: Music downloading, sales displacement, and social welfare in a sample of college students. Journal of Law and Economics, 49: 29-62.

Smith, M.D., R. Telang. 2010. Piracy or Promotion? The Impact of Broadband Internet Penetration on DVD sales. Information Economics and Policy, Special Issue on the Economics of Digital Piracy, 21 289-298.

Sull, D., Tedlow, R., & Rosenbloom, R. 1997. Managerial commitments and technology change in the U.S. tire industry. Industrial and Corporate Change, 6: 461-500.

Teece, D. 1986. Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy.” Research Policy, 15: 285-305.

Tripsas, M. 1997. Unraveling the Process of Creative Destruction: Complementary Assets and Incumbent Survival in the Typesetter Industry. Strategic Management Journal 18 119-142.

Tripsas, M. and Gavetti, G. 2000. Capabilities, cognition, and inertia: Evidence from Digital Imaging. Strategic Management Journal, 21: 1147-1161.

Tushman, M., Anderson, P. 1986. Technological discontinuities and organizational environments. Administrative Science Quarterly, 31: 439-465.

Utterback, J. 1994. Mastering the dynamics of innovation. Boston: Harvard Business School Press. Vogel, H. 2007. Entertainment Industry Economics, 7th edition. Cambridge, Cambridge University Press.

Waldfogel, J. 2012a. And the bands played on: Digital disintermediation and the quality of new recorded music. NBER working paper.

Waldfogel, J. 2012b. Copyright protection, technological change, and the quality of new products: Evidence from recorded music since Napster. Journal of Law and Economics 55(4) 715-740.

Waldfogel, J. 2011. Bye, bye, Miss American Pie? The supply of new recorded music since Napster. NBER Working Papers 16882, National Bureau of Economic Research, Inc.

Wu, B. Wan, Z, and Levinthal, D.A. 2014. Complementary assets as pipes and prisms: Innovation incentives and trajectory choices. Strategic Management Journal, 35: 1257-1278.

Zentner, A. 2006. Measuring the effect of file sharing on music purchases. Journal of Law and Economics, 49: 63-90.