The Effect of Piracy Protection in Book Publishing

Description

The E ect of Piracy Protection in Book Publishing Imke Reimers This version: May 18, 2014 Abstract Digitization has impacted media industries by lowering the cost of distributing cultural goods. With the option to download creative works legally, however, comes the possibility of doing so illegally. Some less well-known artists may welcome this online piracy as an opportunity to increase reach, while more established artists do not have to rely on this type of promotion. I examine the e ects of the illegal distribution of books on regular book sales for di erent types of titles and book formats. I exploit a natural experiment where di erent book titles are protected by a book piracy protection company at di erent times. I compare sales of similar book titles with and without piracy protection, before and after titles have been added by the piracy protection company, in a di erence-in-di erences setting. I nd that the e ect of piracy protection varies across formats and titles. E-books, the closest substitute for online piracy, bene t from piracy protection by selling 15.4% more units, while there is no signi cant e ect on other formats. The e ect is more pronounced for titles that have been successful prior to piracy protection, indicating that book piracy has a promotional e ect for lesser known works. NBER and Northeastern University, 304 Lake Hall, 360 Huntington Avenue, Boston, MA 02115, reimers@nber.

Informations

• Publié par: ActuaLitte.com
• Publié le: 09 juin 2014
• Nombre de lectures: 293
• Langue: English

Extrait

The Effect of Piracy Protection in Book Publishing

∗ ImkeReimers

This version:May 18, 2014

Abstract Digitization has impacted media industries by lowering the cost of distributing cultural goods. Withthe option to download creative works legally, however, comes the possibility of doing so illegally.Some less well-known artists may welcome this online piracy as an opportunity to increase reach, while more established artists do not have to rely on this type of promotion. I examine the effects of the illegal distribution of books on regular book sales for different types of titles and book formats.I exploit a natural experiment where different book titles are protected by a book piracy protection company at different times.I compare sales of similar book titles with and without piracy protection, before and after titles have been added by the piracy protection company, in a difference-in-differences setting.I find that the effect of piracy protection varies across formats and titles.E-books, the closest substitute for online piracy, benefit from piracy protection by selling 15.4% more units, while there is no significant effect on other formats.The effect is more pronounced for titles that have been successful prior to piracy protection, indicating that book piracy has a promotional effect for lesser known works.

∗ NBER and Northeastern University, 304 Lake Hall, 360 Huntington Avenue, Boston, MA 02115, [e-mail]

1

1 Introduction

Digitization has significantly decreased the cost of creating and distributing cultural goods in media industries. Consequently,digitization has affected firm and artist profitability in many settings and cases:production and copying of creative works has become cheaper and easier, increasing the variety and quantity of products available for consumption.The wider variety could increase the level of competition and lower prices further.In many cases, these effects shift surplus from producers to consumers.

Digitization has another effect:as thelegaldistribution of creative works has become easier, so has theillegalThus, the question of the optimal intellectualdistribution of those works. property (IP) strategy arises for each artist and distributor.How do copyright protection and illegal distribution of content affect the legal sales of a work?Modern technology has also made monitoring and regulation of illegal activity more feasible.What level of piracy protection is optimal?This paper addresses these questions in the book publishing industry.

A large and growing literature addresses the effect of file sharing in the music and movie 1 industries. Thefindings are divided, with some work indicating that there is no significant effect (e.g. Oberholzer-Gee& Strumpf (2007)), while more recent work has found that regular sales have been significantly displaced by pirated versions (e.g.Zentner (2006), Liebowitz (2008), and Waldfogel (2010)).In the music industry, file sharing seems to have significantly decreased the legal 2 sales of songs and albums.The negative effect can be attributed to the arrival of Napster in 1999, a file sharing website that revolutionized the industry.The first legal option to download music didn’t arrive until two years later, when Apple introduced its music library iTunes in January 2001.

File sharing and online piracy have been similarly prevalent in other media industries. Consumption of pirated content in the movie industry has been facilitated particularly by one website, megaupload.com, a large hosting service of digital content.The website reportedly had 3 over 50 million daily visitors and accounted for four percent of the total internet traffic.The abrupt shutdown of this service in January 2012 had a significantly positive effect on box office 1 See Smith and Telang (2012) for a detailed description of studies on those industries, and Peitz & Waelbroeck (2006) for a review of the theoretical literature. 2 See Rob & Waldfogel (2004); Danaher et al.(2012). 3 See http://www.justice.gov/opa/pr/2012/January/12-crm-074.html.

2

revenues and digital movie sales of popular works, while the shutdown did not seem to have a positive effect on box office revenues of less well-known works (Peukert, Claussen & Kretschmer (2013), Danaher & Smith (2013)). The effect of file sharing and online piracy is less well-studied in book publishing, as digi- tization in this industry has lagged behind other industries.Amazon’s Kindle represented the first viable and widely adopted device for reading an electronic version of a book in November of 2007, with no widespread illegal option to download present at the time.Since the Kindle’s introduction, the popularity of books in electronic formats has skyrocketed.While e-books held a negligible market share among fiction books before 2007, about half of the weekly top 150 bestseller books have been sold primarily as e-books by late 2012 (see Waldfogel & Reimers (2013)). Moreover, there is no one leading file sharing service for books.Instead, there are many cyber lockers, peer-to-peer networks, and other websites on which a savvy consumer can download e-books for free.The differences in the diffusion of both legal and illegal digital products across the different media industries make it difficult to draw inferences on the effect of file sharing on the book industry. This paper is the first to examine the effect of file sharing on the book industry.It uses 4 a novel dataset consisting of detailed physical and electronic book sales from 2010 to 2013.The dataset also includes the level and success of piracy protection over the same time period, through the leading book piracy protection company, Digimarc.It accounts for the fact that some book formats (e.g.e-books) and genres are closer substitutes for pirated versions than others, and it allows for the interaction of a displacement and a promotional effect of online piracy by examining titles of varying consumer awareness. Due to the rather scattered nature of the book piracy landscape, protection from piracy consists of private companies searching websites that offer a title, and sending them take-down notices. Suchpiracy protection agencies started providing piracy protection for different titles at different points in time, thus providing a natural experiment directly relevant to the project.I ex- ploit this differential discontinuity on protection in a difference-in-differences setting.For example, if a particular title moved into piracy protection in June 2011, one would expect to see a decrease in piracy of this title shortly thereafter, while piracy for other titles would remain relatively stable. 4 The sample of titles is taken from the list of books that are published in electronic format by Rosetta Books.

3

A comparison of unit sales before and after a title has changed its piracy protection status provides information on the differential effect of such protection.

By examining titles that have enjoyed different levels of reader attention in the past, the paper analyzes the interaction of two demand-side effects of illegal downloading.First, a free version of the same product (title) competes with the legal nonzero-price options, potentially decreasing 5 legal sales.Second, though, free options have a sampling and advertising effect, and this potentially 6 increases legal sales.Demand for fiction novels is reputational and depends to in part on word-of- mouth advertising.This is also true for nonfiction books, albeit to a lesser degree.Someone who reads a pirated version of a book title might recommend the book to his friends, who then purchase legal editions of the book.I identify the two effects separately, which allows me to make inferences about an optimal piracy protection strategy, from the perspective of the general public, and from the publisher’s perspective.

Not surprisingly, I find that the effect of piracy protection on legal book sales depends on the awareness level of the title, the type of work (fiction or nonfiction), and the format of the edition. Closersubstitutes, such as legally distributed e-books, see a mean differential increase in sales of 15.4%.

This increase is mostly driven by popular works, which increase their sales by 24.4%, while e-book sales of less well-known works do not change significantly as a result of piracy protection. Assuming that works that are already popular do not benefit from an advertising effect while low- popularity works see both the displacement and the promotional effect, my results suggest that the sampling and advertising effect on e-book sales is of a similar magnitude as the displacement effect: around24% of unit sales.The perceived quality of the title does not affect these results significantly.

The effect on physical editions is less clear as large standard errors render the effect in- significant for most formats, although there is a significant positive effect on sales if we combine all physical formats.Much of the effect of file sharing seems to be picked up by its closest substitute: e-books. Thisdoes not mean, however, that there is no effect of piracy on physical formats.I observe the websites that are found and taken down by one (large) piracy protection company, 5 See Smith and Telang (2012). 6 See Peitz and Waelbroeck (2006b).

4

but I do not know how many other websites still offer the same title for free.My estimates can therefore be interpreted as lower bounds of the true effect of online book piracy on sales.

The reputational and displacement effects vary across genres and formats.Demand for nonfiction titles seems to depend less on word-of-mouth advertising than demand for fiction titles. However, it seems that readers of electronic fiction titles are more likely to read pirated content than readers of electronic nonfiction works.Thus, among physical editions, nonfiction titles benefit more from piracy protection than fiction titles (increasing sales of paperback editions by over 20%), while electronic editions of fiction works benefit more from protection than electronic nonfiction editions (sales of electronic fiction editions increase by 17.7%).

Lastly, it is possible that limiting piracy for a title has a direct effect on supply side consid- erations. Oestreicher-Singerand Sundararajan (2013) show that digital rights management has an economically significant effect on pricing decisions of digital content.Preventing piracy effectively limits the level of competition in the market.Theory predicts a resulting increase in prices.I test for this among e-books, but I do not find a significant change in prices as a result of piracy protec- tion. Itseems that book piracy has a larger effect on consumer behavior than on publisher behavior after the book has been written.This result adds to other tests suggesting that endogeneity of piracy protection is not a large issue in my analysis.

The remainder of the paper proceeds as follows.Section 2 provides background on the publishing industry and on the role of piracy.I describe the empirical strategy of the experiment and provide intuition for the identification in section 3.The data are described in detail in section 4. Iproceed with showing results in section 5, and I present robustness checks in section 6.I conclude with some implications in section 7.

2 Backgroundand The Piracy Protection Experiment

Technological change has transformed media industries such as music, newspapers, movies, televi- sion, and books.The recorded music industry faced challenges from digitization when the Napster file-sharing service arrived in 1999.Since then, digitization has posed challenges to other content

5

7 industries as well.Newspaper revenues, for example, have fallen by half since the late 1990s. Some (but not all) of these challenges can be attributed to increased competition through free online versions.

Digitization in the book publishing industry has lagged behind other content industries. While e-books can be read on computers and have therefore been available for over a decade, e- books are most useful to consumers when read on small hand-held devices, such as e-book readers or tablet computers.Sony had released some electronic book readers as early as 2004, but the first widely adopted e-reading device was Amazon’s Kindle, which was introduced in November 2007. The e-reader and e-book markets have grown quickly since then.The share of US adults owning an e-book reader grew from 2 percent in April 2009 to 24 percent in September 2013, and the share of adults owning either an e-reader or a tablet (some viable method for consuming e-books) reached 8 43 percent in September 2013.Electronic books have similarly become increasingly popular in the past years.As of January 2010, none of the weekly top 150 USA Today bestseller books have been sold primarily as e-books.By 2013, over half of the top 150 books sold more e-books than any other format.

With the option to download content online comes the possibility of doing so illegally.In other content industries, online piracy has been relatively concentrated.Although many titles are available on well-known piracy site (for example, piratebay), the book industry does not face one large file sharing service, as was the case with Napster (music) and Megavideo (movies).In the book industry, file-sharing websites and cyberlockers are fairly wide spread, making centralized efforts to stop piracy difficult.Publishing houses have increasingly hired private firms in an attempt to limit the amount of online book piracy.These firms individually target file sharing services for specific book titles by urging them to take down the work (take-down notices), essentially until the work is not available on the site.

This strategy implies that books are protected individually.The leading anti-piracy service, 9 Digimarc Guardian, has developed an algorithm to search for specific book titles.Its piracy protection strategy includes an automated process that finds suspected pirated content, followed 7 See http://stateofthemedia.org/2013/newspapers-stabilizing-but-still-threatened/newspapers-by-the-numbers/ purchases-by-volume/. 8 See http://www.foliomag.com/2013/pew-research-center-reports-increases-tablet-and-e-reader-ownership. 9 The service was introduced by a company called Attributor, which was acquired by Digimarc in December 2012. The service is now known as Digmarc Guardian.

6

by two tiers of human verification of those sites.The sites with confirmed pirated content are then de-listed from search engines, and eventually taken down.After a title has been removed, Digimarc continues to monitor the provider of the pirated content.

Digimarc has added most of the large publishing houses as clients over the past five years. The publishing companies include Hachette (June 2009), HarperCollins (February 2010), Macmillan 10 (October 2010), Simon & Schuster (October 2011), and Random House (June 2013).In addition, 11 the company has added several smaller publishers.

Rosetta Books, which has signed with the company in June 2011, is one of those publishing houses. Thispublisher has secured exclusive rights to publishing electronic versions of over 600 12 titles. Rosetta’slist of titles consists in large part of backlist titles, ranging from well-known classics (including, among others, works by Kurt Vonnegut, Arthur C. Clarke’sSpace Odyssey Books, and Stephen Covey’sThe 7 Habits of Highly Effective People), to works that are less well known today.The publisher also carries some original titles that are available only in electronic format.

Digimarc started searching for, and taking down, different titles in Rosetta’s catalog at different dates after June 2011.When a title moves into piracy protection, Digimarc Guardian searches the internet for copyright infringements for that title, and it sends take-down notices to the infringing sites.According to a Rosetta press release, “If the site doesn’t comply, the hosting provider is contacted, and the site may be delisted from search engines.Finally, payment providers and advertising networks will be notified of the violation as well as domain registrars.In less than 13 five percent of cases, legal action is necessary.”

Digimarc seems to have had some success in taking down infringing websites.Accordingly, when Digimarc starts protecting a title, one would expect the piracy level for that title to decrease, 10 http://www.publishersweekly.com/pw/print/20090622/11099-news-briefs.html, http://www.mediabistro.com/galleycat/harpercollins-and-harvard-business-school-publishing-will-use-attributors- piracy-protection-program b11178, http://www.marketwired.com/press-release/Attributor-Macmillan-Kensington-Publishing-Corp-Lead-Global- Initiative-Educate-1330505.htm, http://www.publishersweekly.com/pw/by-topic/digital/copyright/article/48974-s-s-signs-with-attributor.html, and http://randomnotes.randomhouse.com/anti-piracy-reporting-tool-added-to-author-portal/. 11 See http://www.digimarc.com/guardian/customers for a list of Digimarc’s customers. 12 While electronic editions can only be published by Rosetta, physical editions can be published by other publishers. See http://go-to-hellman.blogspot.com/2011/11/creative-commons-media-neutrality-and.html for more information. 13 See http://www.rosettabooks.com/pressrelease/fighting-digital-piracy-increases-ebook-sales/.

7

while a title that is not added faces an unchanged piracy level.I utilize this differential change in piracy to identify the effect of piracy on regular sales of physical and electronic book formats.

3 EmpiricalStrategy and Identification

Any analysis examining the effect of piracy protection on book sales is faced with a fundamental inference problem.For any given title that is protected from piracy I do not observe the coun- terfactual: whatwould the sales of this title have been if it had not been protected from piracy? However, I can observe and compare similar titles with and without piracy protection.

In an ideal experiment, the econometrician assigns titles randomly across groups and elim- inates piracy for one set of titles.The marginal effect of online piracy on book sales is the change in sales of the treated titles as compared to the group of titles that did not change their level of piracy. Theempirical strategy takes advantage of the change in piracy in a difference-in-differences analysis. Thelog of the (observed) salesq it of titleiin weektare a function of the works’ observable characteristicsX it and its piracy levelpiracy it in that week:

log(q it ) =βX it +α∙piracy it + it .

(1)

This strategy is difficult to implement for two reasons.First, I do not observe actual levels of piracy.Instead I know when a title was “protected” from piracy (that is, piracy sites for the title were taken down) through the Digimarc Guardian service.I use this information as a proxy for a title’s level of piracy.If the company is perfectly effective in preventing piracy, the results will describe the effect of book piracy.If it is not as effective, then my results are a lower bound of the 14 effect of piracy and are interpreted as the effect of piracyprotection.

Second, the identification strategy assumes that the timing of piracy protection is uncor- related with factors that determine the outcome of interest - in this case, log-unit sales.However, the decision to protect a title from piracy may well be endogenous:a publisher may be more in- terested in protecting those titles that have gained popularity lately than in protecting those that 14 I assume here that the sites that Digimarc finds are not different from the sites it misses in terms of their effect on regular sales.

8

do not seem to attract much reader interest at all.For example, the publisher would have liked to increase piracy protection for Richard Matheson’s 1954 novel,I Am Legend, when it was adapted as a blockbuster movie in 2007.The change in regular book sales would then be the result of a combination of two effects:an increase in demand due to the movie’s promotional effect on the novel, and the change in demand from a change in piracy protection.Disentangling these effects is difficult. Iaddress this issue as described below.

The effect of this piracy protection is identified because I observe sales of titles when they are protected and when they are not protected.Since the move into piracy protection happens at different times for different titles, it is unlikely that any changes in sales when a title moves into piracy protection are due to an exogenous time-dependent shock.Moreover, a change in the level of piracy for one title is unlikely to affect legal demand for another title (while it might affect the level of piracy for another title).Accordingly, equation (1) becomes

k q) =α∙protection+δ+ log( it iti µ t + it ,

(2)

k whereqdenotes the unit sales of titleiin weektin formatk, wherek∈ {hardcover, trade it paperback, mass market paperback, audio, and e-books}, andprotection it is a dummy variable 15 that is 1 if the title is under piracy protection in weekt. Theeffect of book piracy on regular α book sales is given byα, where book piracy protection causes a change in regular sales ofe−1 16 percent. Thelog specification assumes that piracy changes sales by a common percentage, rather than by a common number.One would expect that the absolute sales of a title by Stephen King will be affected more strongly by piracy than a more obscure title.

I also make use of the fact that I observe unit sales of over 120 titles (depending on format), over a period of 170 weeks, by controlling for title (δ i ) and week-year fixed effects (µ t title fixed). The effects control for all time-invariant differences between titles.This picks up the overall popularity of the title, as well as genres and the author’s level of popularity.Time fixed effects control for changes over time that affect all titles similarly, such as changes in the economic environment, or 15 I treat a title as protected if a website containing at least one free version of the title has been taken down during time periodtrobustness check shows that the results stand even if I consider a title treated for every time period. A after a website has been taken down.See appendix section 6.3 for more detail. 16k As I observe some periods with zero physical sales for some titles, I use log(q it + 0.00001 as the dependent variable for physical formats.In that case, the marginal effect is interpreted with a bias.This bias understates the true effect 4q and goes to zero quickly (the bias is). q(q+0.0001)

9

17 the release of a new e-reader or tablet.

I address the potential endogeneity of piracy protection in two ways.First, I ignore those titles that never enter piracy protection as those titles may be inherently different from those that move into piracy protection.Second, Digimarc Guardian follows titles with three different priority levels (low, medium, and high).Regarding the urgency with which a title is protected as an indicator of the title’s current market-induced demand, I interact the protection variable with indicators for the different priority levels and compare the coefficients to test for endogeneity of piracy protection.

To capture some of the underlying mechanisms, I further separate the effect of piracy protection on well-known works and less well-known works, on titles that have received good or bad reviews on a large and popular book database (Goodreads), and on fiction and nonfiction titles. While my analysis is restricted to one publisher, the observed titles vary widely in popularity, quality, and genre, making some inference to other publishers and works possible.

Lastly, it is possible that a change in piracy protection is accompanied by supply side reactions. Forinstance, eliminating a zero-price competitor can affect a publisher’s pricing strategy. I include e-book prices in a third estimation of the effect on e-book sales to account for such a supply side adjustment.

4 Data

To examine the effect of book piracy protection on regular book sales I follow the demand for and piracy of a set of book titles over close to four years, from 2010 to 2014.The underlying dataset consists of titles whose electronic versions are exclusively published through RosettaBooks.I obtain 18 a list of 653 titles.Most of the works have been originally published several years ago, going back as far as the first half of the twentieth century.There also are several titles that have been published during the past ten years.The analysis includes the subset of the titles whose piracy protection status has changed between 2011 and 2013. 17 Since the titles in my dataset have originally been published several years ago, I do not include a time trend that follows the work’s time since publication.I do, however, include an indicator variable that equals 1 if a new edition has recently been published in section 6.2 to show that the results are robust in this respect. 18 See http://www.rosettabooks.com/books/books-a-z/ for Rosetta’s catalog.

10

On the demand side, I observe weekly sales data for physical book formats:hardcover, trade paperback, audio versions, and mass market paperback, through the Nielsen BookScan database. These sales date back to 2002, although I focus on the time period from 2010 to 2014.They include all editions of the title that are listed by Nielsen. I also observe monthly e-book unit sales on the title level directly through Rosetta (the exclusive publisher of the titles’ e-books), from July 2011 to December 2013.These editions are sold through Amazon, Apple’s iBooks store, and Barnes and Noble.The publisher carries a few popular titles that have sold far more than 100,000 copies in 2012, and some that have sold less than 1,000.Table 1 shows mean title sales in each format in 2012.Trade paperbacks and e-books 19 seem to be the most popular formats.

Table 1:Annual title sales, 2012 Sales NMean Std.Dev. MinMax Hardcover 126180.746 912.9010 7,954 Trade Paperback126 4,294.55 18,623.550 173,202 Audio 126125.540 1,123.230 12,582 Mass Market126 443.5871,627.57 0 14,358 E-Book 1572,754.27 7,037.65 59,316

On the piracy dimension, I obtain detailed information through Digimarc, the company that protects Rosetta’s works from illegal file sharing.I observe the date that take-down notices were sent to websites offering specific titles, and when these piracy sites were taken down.The titles in my dataset have 32 different protection start dates between November 2010 and January 2014. Ifocus on those titles that have changed their piracy protection status during the observed time period.These titles are protected with different levels of priority, with 1 being the lowest priority, and 3 being the highest.The priority level of a title can vary over time.Digimarc searches the internet for websites, cyberlockers, and peer-to-peer services that make books available illegally for free.The company sends take-down notices to those sites with increasing levels of urgency, until the site is taken down.Table 2 shows summary statistics on the number of sites found, the number of notices sent per site, and the company’s success rate, for those titles that were protected at some point during the observed time period. 19 The number of titles is less than 653 because I only include those titles that changed their piracy protection status during the observed time period.It varies across physical and electronic formats because some titles are only available electronically.

11