Abstract
Digital products are differentiated from online and offline physical products in important ways. This paper studies the influence of digital products on existing channels and the optimal multichannel management strategy in the context of the book industry. Using individual-level online transaction data and county-level offline bookstore data, I estimate a demand model of book format and retailer choices across genres. I use the estimates to solve for publishers’ optimal wholesale pricing strategy across channels. The demand-side estimates reveal that e-books and offline bookstores appear to compete head-to-head in book genres that serve casual reading purposes such as fiction, science fiction, humor, and biographies, which I categorize as “casual” books. The supply-side results suggest that as local bookstore availability increases, publishers should charge higher wholesale prices in the offline print channel, especially for “casual” books. I find that the e-book channel does not always hurt print channels but can serve as a strategic complement and enhance the pricing power of print channels in some markets and genres; this complementarity does not rely on branding or marketing communication and crucially depends on the relative strength of the channels. Specifically, a new channel can help an existing channel when two conditions hold: first, the new channel is not too weak and can generate enough market expansion effect; second, the existing channel is not too strong and can avoid too much cannibalization from the new channel. I use counterfactual analysis to illustrate the mechanism behind this result and how a multichannel management strategy should account for relative strength across channels.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Hardcovers account for only 5% of the transactions.
I use the sales-unweighted prices in the estimation, as the sales-weighted and unweighted prices differ by less than 2%.
Esri Demographics and Business Database contains county-level data on number of bookstores and sales (http://www.esri.com/data/esri_data/business-overview/business). County Business Patterns data (CBP) contain county-level data on total number of establishments by industry (https://www.census.gov/programs-surveys/cbp.html). Economic Census data (ECN) contain county-level data on total number of establishments and sales by industry (https://www.census.gov/programs-surveys/economic-census.html). The Esri dataset is available for 2010 and is the most comprehensive among the three data sets: it covers the largest number of counties and covers all the counties in the Comscore online panel. CBP is available for years between 2008 and 2013. ECN is available for 2007 and 2012 because the census is conducted every 5 years. To obtain bookstore number and sales data between 2008 and 2012, I use the Esri data as the baseline and use the CBP and ECN data to obtain the year-to-year percentage changes. For 2010, I directly use the Esri data on bookstore number and sales; for the rest of the years, I impute bookstore number and sales such that the year-to-year percentage changes in bookstore number and per-store sales relatively to 2010 match those in the CBP and ECN data.
Specifically, I first need to scale down the population size in the true population to in-sample in order to estimate the demand model. I determine the size of the full sample as N = Non/γon, where Non is the total number of consumers who have bought any kind of products online in the Comscore sample and γon is the fraction of consumers who have bought any kind of products online from Nielsen Online Shopping Trend report (2012). This definition allows me to include both book buyers and nonbuyers in the full sample. Given the size of the full sample, I determine the county-level number of consumers in the full sample as Nc = γcN, where γc is county c’s population as a fraction of the total population in the true population; this approach ensures that the relative sizes of the counties remain the same in the full sample as in the true population. Third, within each county, I determine the number of consumers in each age-income group in the full sample as \({N_{c}^{k}}={\gamma _{c}^{k}}N_{c}\), where \({\gamma _{c}^{k}}\) is the fraction of consumers in age-income group k in county c in the true population; this approach ensures that the demographic distribution within each county remains the same in the full sample as in the true population.
Book Industry Study Group survey (2011) indicates that consumers dominantly use e-readers as the e-reading device. Consumers using dedicated e-readers are also the largest contributor to e-book sales.
This blog (http://ilmk.wordpress.com/category/analysis/snapshots/) takes monthly snapshots of Amazon.
The local rent information comes from the county-level median gross rent in the American Community Survey (https://www.census.gov/programs-surveys/acs). The local wage information comes from the county-level wage in the “Sporting goods, hobby, book and music stores” industry from the Quarterly Census of Employment and Wages program, collected by the Bureau of Labor and Statistics (https://www.bls.gov/cew/datatoc.htm#NAICS_BASED).
The results are robust when I use logged number of bookstores and logged ratio of number of bookstores to number of coffee shops, as discussed in Section 4.2.1.
I model book demand at the genre level, instead of at the book title level, for the following reasons: 1) aggregate book sales are more relevant in the pricing problem than single-title sales; 2) modeling at the title level would require strong assumptions about the books that enter consumers’ choice set. It is not appealing to assume that consumers must decide from the millions of books that are available or from best sellers only, as 99.94% of the titles were purchased fewer than 10 times in the data. 3) modeling at the title level also requires estimating title fixed effects to account for price endogeneity issues. I do not have title-level aggregate book sales data and cannot estimate such fixed effects.
I focus on Amazon as the main e-reader retailer because Kindle is the dominant e-reader during my sample period. I conduct a robustness check by allowing consumers to buy other reading devices after 2010 in the demand estimation. I find that the key demand-side results are qualitatively robust, with the estimated Kindle qualities being smaller.
The e-book availability is not genre-specific. If there is any difference in e-book availability across genres, it will be captured by the genre fixed effects \({\theta _{g}^{E}}\).
In the data there exists a mild correlation in book purchases across genres. Conditional on buying books in any of the three genres, the correlation between the number of books bought is 0.22 for “lifestyle” and “practical” genres, 0.19 for “lifestyle” and “casual” genres, and 0.10 for “casual” and “practical” genres. This data pattern can be generated by either allowing for interaction among genre-specific utilities (i.e., substitution across genres) or a positive correlation in genre-specific preferences. As the data cannot separately identify the two, I allow for the correlation in the genre-specific preferences 𝜃ig in 𝜃igt of Eq. 2 and do not allow for interaction among genre-specific utilities. As shown in the estimation results in Section 5.2, the data reveal four heterogeneous preference segments. One of them prefers both “lifestyle” and “practical” books. One of them prefers all three genres. Both segments represent a positive correlation in genre preferences and can generate a positive correlation in genre consumption.
In this setup, Kindle qualities do not interact with book utilities so that consumers’ book utilities are not affected by the types of Kindles they use. The reason is that I cannot empirically identify such a relationship. I add Kindle quality dummies to book utilities in a robustness check and find that the estimated dummies are insignificant.
In a robustness check I allow consumer expectations to follow an AR(1) process and empirically estimate the coefficients of the AR(1) model. The results are robust. I keep the perfect foresight assumption because Kindle prices changed annually during the five-year data period; the short panel makes it less attractive to estimate an AR(1) process.
Numerically, the integration is conducted over \(H_{igt}=1,2,...,\bar {H}\) instead of from 1 to infinity. I choose a large enough \(\bar {H}\) such that \(\Pr \left (H_{igt}=\bar {H}\right )\) is a very small value (in the magnitude of 10− 3).
Due to yearly resampling, I cannot always observe the book purchase transactions for consumers for whom I observe Kindle purchase transactions. This issue is minor because I do not need to observe consumers’ actual book purchases to calculate their Kindle purchase probability; the indirect book utilities in the Kindle purchase utilities are calculated by taking the expectation over the error terms in the book choices. The log likelihood function of the Kindle purchase data thus contains only Kindle adoption probabilities: \({\sum }_{t}\left \{ n_{i1t}\log \left [\Pr \left (d_{it}=1\right )\right ]+n_{i0t}\log \left [1-\Pr \left (d_{it}=1\right )\right ]\right \}\). Here, ni1t is the observed Kindle sales, \(n_{i0t}=N_{i0}-{\sum }_{\tau =1}^{t}n_{i1\tau }\) is the number of decisions to not buy, and Ni0 is the initial market size of type i consumers. Yearly resampling also means that I cannot always observe Kindle adoption decisions for consumers for whom I observe book transactions if their Kindle adoption occurred in a different year. I take a probabilistic view of consumers’ device ownership status: for consumers who bought e-books, I assume that they own Kindles; for consumers who did not buy e-books, similar to Gowrisankaran and Rysman (2012), I allow them to have Kindles with the model-predicted device ownership probability \({\varPsi }_{it}^{1}\). Li (2019) uses similar individual-level online transaction data and faces a similar resampling problem. I refer to Li (2019) for more details on how the log likelihood of the book purchase data is constructed.
The first-stage regression coefficients are negative (-0.0013 for local rent and -0.0037 for local wage) and statistically significant at 1% level.
Using local wage as an instrument can be problematic when it also affects book demand through affecting the income consumers use to purchase books. Rysman (2004) faces a similar situation when he uses local wage to instrument for the provision of advertising in yellow pages. He addresses this issue by including county-level income as controls in the demand model. Similarly, I address this issue by including household income in the demand model.
92.82% of the book titles had only one purchase record per year, 5.53% had two purchases, and 99.94% had fewer than 10 purchases.
I determine the number of segments by incrementally adding segments until one of the segment sizes is not statistically different from zero. For each genre, I are able to identify two levels of occasion parameters and taste parameters, high (\({\lambda _{g}^{H}},{\theta _{g}^{H}}\)) and low (\({\lambda _{g}^{L}},{\theta _{g}^{L}}\)). The four consumer segments have genre-specific occasion parameters \(\left \{ \lambda _{ig}\right \}_{g=1,2,3}=\left \{ {\lambda _{1}^{L}},{\lambda _{2}^{H}},{\lambda _{3}^{L}}\right \} ,\left \{ {\lambda _{1}^{H}},{\lambda _{2}^{L}},{\lambda _{3}^{H}}\right \} ,\left \{ {\lambda _{1}^{H}},{\lambda _{2}^{H}},{\lambda _{3}^{H}}\right \} ,\left \{ {\lambda _{1}^{L}},{\lambda _{2}^{L}},{\lambda _{3}^{L}}\right \} \), baseline taste parameters \(\left \{ \theta _{ig}\right \}_{g=1,2,3}=\left \{ {\theta _{1}^{L}},{\theta _{2}^{H}},{\theta _{3}^{L}}\right \} ,\left \{ {\theta _{1}^{H}},{\theta _{2}^{L}},{\theta _{3}^{H}}\right \} ,\left \{ {\theta _{1}^{H}},{\theta _{2}^{H}},{\theta _{3}^{H}}\right \} ,\left \{ {\theta _{1}^{L}},{\theta _{2}^{L}},{\theta _{3}^{L}}\right \} \) and population mass \(\left \{ m_{1},m_{2},m_{3},1-m_{1}-m_{2}-m_{3}\right \} \), respectively.
The reason is that the two types have the same e-format taste, so they are equally likely to prefer e-books to print books. However, avid readers have higher baseline tastes and are more likely to buy print books in the absence of e-books.
I obtain the observed wholesale prices from the observed Amazon retail prices. The wholesale prices of both formats are 50% of the list price on average. The list price of e-books is 80% of the list price of its print book counterpart. Amazon’s retail price is 60% of the list price.
The dynamic aspect of the demand model is that Kindle penetration increases over time. The static problem here assumes that the market starts with a zero Kindle install base. It is challenging to solve for a dynamic pricing problem with three channels. I start with the static problem to illustrate the key trade-offs in the setting with three channels.
For print books, the printing cost is a function of physical characteristics of the book such as page count, cover, binding, and size. The average printing cost is $4.25 (see https://www.dogearpublishing.net/ak-author-purchase-prices.php, https://kdp.amazon.com/en_US/help/topic/G8BKPU9AGVZSF9QF, and https://www.theguardian.com/books/booksblog/2011/aug/04/price-publishing-ebooks). The royalty rate is typically 8% of the retail price for print books and 25% of the wholesale price for e-books (see http://inkandquills.com/2016/10/08/traditional-publishing-royalties/, https://www.alanjacobson.com/writers-toolkit/the-business-of-publishing/, and https://www.thebinderyagency.com/blog/howdopublisherspayauthors). I convert the royalty rate for print books as a fraction of retail price to as a fraction of wholesale price based on the wholesale-to-retail price ratios discussed earlier.
Figure 6 contains the per-consumer profit change for a typical consumer. I first compute the per-consumer profit change for each heterogeneous consumer types. I then integrate over the distribution of the heterogeneous consumer types to obtain the profit change for a typical consumer. Figure 7 contains the per-consumer profit change for a typical consumer, which is computed in a similar way.
One caveat is that Fig. 6 plots the per-consumer profit change in markets with a particular level of bookstore availability, while the optimal strategy in Fig. 5 is a result of balancing across markets with different levels of bookstore availability. Therefore, a positive profit change in Fig. 6 does not necessarily leads to a higher optimal price in Fig. 5. However, the overall pattern holds: a profit gain in Fig. 6 suggests a higher optimal price in Fig. 5.
Li and Srinivasan (2019) use a similar simulation to assess the influence of within-tier hotel competition on equilibrium outcomes.
References
Ansari, A., Mela, C.F., Neslin, S.A. (2008). Customer channel migration. Journal of Marketing Research, 45(1), 60–76.
Balasubramanian, S. (1998). Mail versus mall: a strategic analysis of competition between direct marketers and conventional retailers. Marketing Science, 17, 181–195.
Bell, D., Gallino, S., Moreno, A. (2017). Offline showrooms in omnichannel retail: demand and operational benefits. forthcoming, Management Science.
Biyalogorsky, E., & Naik, P. (2003). Clicks and mortar: the effect of online activities on offline sales. Marketing Letters, 14(1), 21–32.
Brynjolfsson, E., Hu, Y., Rahman, M. (2009). Battle of retail channels: how product selection and geography drive cross- channel competition. Management Science, 55(11), 1755–65.
Chen, H., Hu, Y.J., Smith, M.D. (2018). The impact of e-book distribution on print sales: analysis of a natural experiment. forthcoming, Management Science.
Chiang, W.K., Chhajed, D., Hess, J.D. (2003). Direct marketing, indirect profits: a strategic analysis of dual-channel supply-chain design. Management Science, 49(1), 1–20.
Ching, A., & Ishihara, M. (2012). Measuring the informative and persuasive roles of detailing on prescribing decisions. Management Science, 58(7), 1374–87.
Choi, J., & Bell, D. (2011). Preference minorities and the internet. Journal of Marketing Research, 58(4), 670–82.
Dubé, J.P. (2004). Multiple discreteness and product differentiation: demand for carbonated soft drinks. Marketing Science, 23(1), 66–81.
Forman, C., Ghose, A., Goldfarb, A. (2009). Competition between local and electronic markets: how the benefit of buying online depends on where you live. Management Science, 55(1), 47–57.
Gentzkow, M. (2007). Valuing new goods in a model with complementarity: online newspapers. American Economic Review, 97(3), 713–44.
Ghose, A., Smith, M., Telang, R. (2006). Internet exchanges for used books: an empirical analysis of product cannibalization and welfare impact. Information Systems Research, 17(1), 3–19.
Gilbert, R. (2015). E-books: a tale of digital disruption. Journal of Economic Perspectives, 29(3), 165–84.
Gordon, B. (2009). A dynamic model of consumer replacement cycles in the PC processor industry. Marketing Science, 28(5), 846–867.
Gowrisankaran, G., & Rysman, M. (2012). Dynamics of consumer demand for new durable goods. Journal of Political Economy, 120, 1173–1219.
Hendel, I. (1999). Estimating multiple-discrete choice models: an application to computerization returns. The Review of Economic Studies, 66(2), 423–446.
Lal, R., & Sarvary, M. (1999). When and how is the Internet likely to decrease price competition? Marketing Science, 18(4), 485–503.
Li, H. (2019). Intertemporal price discrimination with complementary products: E-Books and E-Readers. Management Science, 65(6), 2665–2694.
Li, H., & Srinivasan, K. (2019). Competitive dynamics in the sharing economy: an analysis in the context of airbnb and hotels. Marketing Science, 38 (3), 365–391.
Kannan, P.K., Pope, B., Jain, S. (2009). Pricing digital content product lines: a model and application for the national academies press. Marketing Science, 28(4), 620–636.
Kumar, N., & Ruan, R. (2006). On manufacturers complementing the traditional retail channel with a direct online channel. Quant. Market Economics, 4(3), 289–323.
Narayanan, S., Manchanda, P., Chintagunta, P. (2005). Temporal differences in the role of marketing communication in new product categories. Journal of Marketing Research, 42(3), 278–90.
Neslin, S.A., & Shankar, V. (2009). Key issues in multichannel customer management: current knowledge and future directions. Journal of Interactive Marketing, 23(1), 70–81.
Oberholzer-Gee, F., & Strumpf, K. (2007). The effect of file sharing on record sales: an empirical analysis. Journal of Political Economy, 115(1), 1–42.
Ofek, E., Sarvary, M., Katona, Z. (2011). Bricks and clicks: the impact of product returns on the strategies of multichannel retailers. Marketing Science, 30(1), 42–60.
Pauwels, K., & Neslin, S. (2015). Building with bricks and mortar: the revenue impact of opening physical stores in a multichannel environment. Journal of Retailing, 91(2), 182–197.
Petrin, A., & Train, K. (2010). A control function approach to endogeneity in consumer choice models. Journal of Marketing Research, 47(1), 3–13.
Reimers, I., & Waldfogel, J. (2017). Throwing the books at them: amazon’s puzzling long run pricing strategy. Southern Economic Journal, 83(4), 869–885.
Rysman, M. (2004). Competition between networks: a study of the market for yellow pages. The Review of Economic Studies, 71(2), 483–512.
Rust, J. (1987). Optimal replacement of GMC bus engines: an empirical model of harold zurcher. Econometrica: Journal of the Econometric Society, 999–1033.
Sinai, T., & Waldfogel, J. (2004). Geography and the internet: is the internet a substitute or a complement for cities? Journal of Urban Economics, 56 (1), 1–24.
Verhoef, P.C., Kannan, P.K., Inman, J.J. (2015). From multi-channel retailing to omni-channel retailing: introduction to the special issue on multi-channel retailing. Journal of Retailing, 91(2), 174–181.
Wang, K., & Goldfarb, A. (2017). Can offline stores drive online sales? Journal of Marketing Research, 54(5), 706–719.
Zettlemeyer, F. (2000). Expanding to the internet: pricing and communications strategies when firms compete on multiple channels. J. Marketing Res., 37(3), 292–308.
Zhang, X. (2009). Retailers’ multichannel and price advertising strategies. Marketing Science, 28(6), 1080–1094.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Monte Carlo study: identification using aggregate offline sales data
An identification challenge is that the offline book sales data are at the county level. Following Gordon (2009), I conduct a simulation analysis to assess whether the model can use aggregate offline sales data to identify substitution patterns and consumption occasions. I conduct the simulation for 4,500 consumers in two consumer segments. Given the true parameters, I simulate consumers’ book and Kindle purchases at the individual level and aggregate over the offline purchases at the county level. I combine the aggregate offline data with the individual-level online data to estimate the model. I generate 50 sets of simulation datasets for the same set of true parameter values. I then obtain the estimates for each simulated dataset to calculate the standard errors of the estimates.
In Table 7, I present the true parameter values in Column 1 and the estimated parameter values in Column 2. The standard errors are based on the estimates from the 50 simulations. The true parameter values are within the 95% confidence intervals of the estimates. The results suggest that the model is able to recover consumer heterogeneity in book tastes and channel preferences, which are central to generating the substitution patterns across channels.
Appendix B: Robustness check: varying retailer pricing rule and kindle price
The supply-side pricing problem abstracts from retailers’ pricing decisions and assume that retailers following the same pricing rules as the observed ones. I conduct robustness checks by varying the values of \(\left \{ \gamma _{j},\gamma _{e}\right \} \) and Kindle prices and allow them to be different from the observed values. Without loss of generality, I consider the case in which the new wholesale pricing strategy triggers more competition among retailers so that retailers set lower prices; the case with less retailer competition can be similarly derived. In particular, I re-solve the publishers’ wholesale pricing strategy in four scenarios: 1) \(\tilde {\gamma }_{j}=0.9\gamma _{j}\) for j = A, B, which represents a 10% price cut in online print channel, while all other values remain the same; 2) \(\tilde {\gamma }_{j}=0.9\gamma _{j}\) for j = off, which represents a 10% price cut in the offline print channel, while all other values remain the same; 3) \(\tilde {\gamma }_{e}=0.9\gamma _{e}\), which represents a 10% price cut in the e-channel, while all other values remain the same; and 4) Kindle prices drop by 20% while all other values remain the same. Figure 9 and Table 8 compare the new optimal pricing solutions with the original solutions in Section 6.2. The main results on the multichannel pricing strategy are robust. The only difference is that when retailers cut prices in a specific channel, publishers charge higher wholesale prices in all channels. The intuition is that greater retailer competition allows publishers to charge higher wholesale prices without an excessive increase in retail prices to consumers. Similarly, when Kindle prices drop, publishers charge higher wholesale prices in all channels.
Robustness Check: Vary Retailer Pricing Rule and Kindle Prices. a Scenario 1: Online Print Book Retail Price Cut by 10%. b Scenario 2: Offline Print Book Retail Price Cut by 10%. c Scenario 3: E-Book Retail Price Cut by 10%. d Scenario 4: Kindle Price Cut by 20%. These figures present the solutions to the optimal publisher pricing problem when the retailers follow different pricing rules from the observed ones or when Kindle prices are different from the observed ones. The solid lines represent the optimal pricing strategies under alternative scenarios. The dashed lines represent the optimal pricing strategies under the original scenario. Compared to the original scenario, the results on the multichannel pricing strategy under alternative scenarios are robust
Appendix C: Robustness check: duopoly competition
A caveat in the duopoly competition analysis is that doubling the number of publishers may artificially inflate the channel sales. I need to adjust the model so that the channel-specific sales remain the same as the sales in the monopoly case given the same prices. This can be achieved by letting the two publishers split the original sales in each channel. Let ψjgt denote the original probability of choosing channel j in the monopoly case. Let \(\tilde {\psi }_{jgt}\) denote the new probability of choosing channel j from one of the two publishers in the duopoly case. The probabilities can be calculated as follows (subscript i is removed for illustrative purpose):
Given the same observed prices and characteristics, I adjust the channel-specific taste to \(\tilde {\delta }_{jgt}=\delta _{jgt}+\log \frac {1}{2}\) while keeping the rest of the parameters the same so that condition \(\tilde {\psi }_{jgt}=\frac {1}{2}\psi {}_{jgt}\) holds. Li and Srinivasan (2019) use a similar simulation with the same adjustment when assessing the influence of competition on pricing strategies.s
Rights and permissions
About this article
Cite this article
Li, H. Are e-books a different channel? Multichannel management of digital products. Quant Mark Econ 19, 179–225 (2021). https://doi.org/10.1007/s11129-021-09235-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11129-021-09235-0