The Informative Role of Advertising and Experience in Dynamic Brand Choice: An Application to the Ready-to-Eat Cereal Market

Abstract

We study how consumers make brand choices when they have limited information. In a market of experience goods with frequent product entry and exit, consumers face two types of information problems: first, they have limited information about a product’s existence; second, even if they know a product exists, they do not have full information about its quality until they purchase and consume it. In this chapter, we incorporate purchase experience and brand advertising as two sources of information and examine how consumers use them in a dynamic process. The model is estimated using the Nielsen Homescan data in Los Angeles, which consist of grocery shopping history for 1,402 households over 6 years. Taking ready-to-eat cereal as an example, we find that consumers learn about new products quickly and form strong habits. More specifically, advertising has a significant effect in informing consumers of a product’s existence and signaling product quality. However, advertising’s prestige effect is not significant. We also find that incorporating limited information about a product’s existence leads to larger estimates of the price elasticity. Based on the structural estimates, we simulate consumer choices under three counterfactual experiments to evaluate brand marketing strategies and a policy on banning children-oriented cereal advertising. Simulation suggests that the advertising ban encourages consumers to consume less sugar and more fiber, but their expenditures are also higher because they switch to family and adult brands, which are more expensive.

This paper is based on Yan Chen’s dissertation, “Information, Consumer Choice and Firm Strategy in an Experience Good Market” (August 2008). We are deeply grateful to Mark Denbaly and Ephraim Leibtag at the U.S. Department of Agriculture’s Economic Research Service for providing access to the Nielsen Homescan data. We have also benefited from the insightful comments of John Rust, Roger Betancourt, Erkut Ozbay, Pallassana Kannan, and seminars participants at the University of Maryland and the Economic Research Service. All errors are ours.

Appendices

5. Appendixes

5.1.1 5.1.Controlling for Unobserved Consumer Heterogeneity

We introduce consumer-brand random effects to capture the unobserved consumer heterogeneity in brand preferences. Specifically, the utility function can be written as

$$ {U_{ijt}} = {Z_{ijt}} \bullet \Phi + {\nu_{ij}} + {\varepsilon_{ijt}} $$

where Z _ijt represents the vector of explanatory variables, Φ represents the vector of coefficients corresponding to Z _ijt , and ${\nu_{ij}}$represents consumer i’s unobserved preference for brand j, which is independent from ${Z_{ijt}}$ and${\varepsilon_{ijt}}$.

Let ${\nu_{ij}} = {\mu_{ij}} + {\omega_j}$. ${\mu_{ij}} : N(0,\varsigma_{ij}^2)$, and ${\omega_j} = E({\nu_{ij}})$ is a constant. Assuming ${\varepsilon_{ijt}}$ has a generalized extreme value distribution, then we can write the probability that consumer i will choose j conditional on${\mu_{i1}},\;{\mu_{i2}}, \ldots {\mu_{i51}}$, and choice set ${C_{it}}$as

$$ \begin{array}{llll} P(j|{\mu_{i1}},{\mu_{i2}},...,{\mu_{i51}},{C_{it}}) = & \frac{{\exp (({Z_{ijt}} - {Z_{i51t}}) \bullet \Phi + {\mu_{ij}} + {\omega_j} - {\omega_{51}})}}{{\sum\limits_{l = 1}^{51} {\exp (({Z_{ilt}} - {Z_{i51t}}) \bullet \Phi + {\mu_{il}} + {\omega_l} - {\omega_{51}})} }} \\= & \frac{{\exp ({z_{ijt}} \bullet \Phi + {\mu_{ij}} + {\xi_j})}}{\sum\limits_{l = 1}^{51} {\exp ({z_{ilt}} \bullet \Phi + {\mu_{il}} + {\xi_l})}} \end{array} $$

where for the second equal sign we use ${z_{ijt}} = {Z_{ijt}} - {Z_{i51t}}$ and ${\xi_j} = {\omega_j} - {\omega_{51}}$.

$p(j|{C_{it}})$is equal to $P(j|{\mu_{i1}},{\mu_{i2}}, \ldots, {\mu_{i51}},{C_{it}})$integrated over the marginal distribution of the ${\mu_{ij}}$’s. Specifically, it is equal to

$$ \int_{ - \infty }^\infty {\int_{ - \infty }^\infty {...\int_{ - \infty }^\infty {\frac{{\exp ({z_{ijt}} \bullet \Phi + {\mu_{ij}} + {\xi_j})}}{{\sum\limits_{l = 1}^{51} {\exp (} {z_{ilt}} \bullet \Phi + {\mu_{il}} + {\xi_l})}}f(} } } {\mu_{i1}})f({\mu_{i2}}) \ldots f({\mu_{i51}})d{\mu_{i1}}d{\mu_{i2}} \ldots d{\mu_{i51}} $$

It is hard to compute $p(j|{C_{it}})$ analytically, and we simulate it by taking S draws from the distribution of ${\mu_{ij}}$, for all j. The simulator for$p(j|{C_{it}})$ is

$$ \hat{p}(j|{C_{it}}) = \frac{1}{S}\sum\limits_{s = 1}^S {\frac{{\exp ({z_{ijt}} \bullet \Phi + \mu_{ij}^s + {\xi_j})}}{{\sum\limits_{l = 1}^{51} {\exp ({z_{ilt}} \bullet \Phi + \mu_{il}^s + {\xi_l})} }}} $$

To reduce the number of parameters to be estimated, we allow ${\omega_j}$ to vary across brand segment, and $\varsigma_{ij}^2$ to vary across both brand segment and whether the household has children. There are a total of eight parameters to estimate for unobserved consumer-brand preferences, of which six are scale parameters: $\varsigma_{FK}^2,\varsigma_{FN}^2,\varsigma_{AK}^2,\varsigma_{AN}^2,\varsigma_{KK}^2,\varsigma_{KN}^2$, where the first subscript denotes whether the brand belongs to the family, adult, or kid segment, and the second subscript denotes whether there are any children in the household; two are location parameters: ${\omega_A}$ and ${\omega_K}$, where the subscript denotes whether the brand belongs to the adult or kid segment. ${\omega_F}$ is normalized to zero.

5.1.2 5.2.Choice Set Simulation Details

In the simulation, we assume that choice set is a function of brand advertising and purchase experience, as shown in Eqs. (5.1) and (5.2). The specific choice set simulation process is outlined as follows.

• Step 1. Calculate ${q_{ijt}}$(ϕ) for each consumer, each brand, and each time, where ϕ = (ϕ_0, ϕ_1, ϕ₂).

• Step 2. For each consumer-time-brand combination, draw a random number $u_{ijt}^r$from the uniform distribution between 0 and 1.

• Step 3. If $u_{ijt}^r < {q_{ijt}}$, then brand j is included in consumer i’s choice set at time t; otherwise it is not. This defines the choice set in the rth simulation $C_{it}^r$. After simulating the choice set, we can calculate simulated brand choice probabilities for each consumer.

• Step 4. Calculate ${P^r}(j|{C_{it}})$, consumer i’s probability of choosing brand j conditional on $C_{it}^r$ _. (The formula for calculating ${P^r}(j|{C_{it}})$ depends on the distributional assumption on the error term in the utility function).

• Step 5. Calculate $p_{ijt}^r = \prod\limits_{j \in C_{it}^r} {{q_{ijt}}\prod\limits_{k \notin C_{it}^r} {(1 - {q_{ikt}})} } \times {P^r}(j|{C_{it}})$, consumer i’s unconditional probability of choosing brand j at time t in the rth simulation.

• Step 6. Draw the random numbers $u_{ijt}^r$ repeatedly for R times, and each time repeat steps 2–5.

• Step 7. Calculate the simulated choice probability ${\hat{p}_{ijt}} = \frac{1}{R}\sum\limits_{r = 1}^R {p_{ijt}^r} .$.

5.1.3 5.3.Contraction Mapping Details

In the instrumental variable estimation, we need to find the δ that makes predicted market shares based on the model equal to the observed market shares. Given an initial guess of δ, $\Pi$, and $\Sigma$, the predicted market share for brand j, ${\sigma_j}(\delta^h,\Pi, \Sigma, \kappa, \lambda, \gamma )$, is calculated as follows.

First, based on advertising data and household characteristics, simulate choice sets for each consumer on each shopping occasion.

Second, given δ, $\Pi$,$\Sigma$, κ, λ, and γ, a consumer compares the utility levels of all brands in his choice set on the shopping occasion and chooses the one that yields the highest utility.

Third, sum the consumer brand choices in a year to get predicted brand market shares.

To obtain the values of δ that solve ${\sigma_j}(\delta^h,\Pi, \Sigma, \kappa, \lambda, \gamma ) = {S_j}$, we use the iteration

$\delta_j^{h + 1} = \delta_j^h + \ln ({S_j}) - \ln ({\sigma_j}(\delta^h,\Pi, \Sigma, \kappa, \lambda, \gamma ))$. The proof that the iteration is a contraction mapping follows Goeree (2008).

Define $f({\delta_j}) = {\delta_j} + \ln ({S_j}) - \ln ({\sigma_j}(\delta^h,\Pi, \Sigma, \kappa, \lambda, \gamma ))$. To show that f is a contraction mapping, we need to show that $\forall$j and m, $\partial f({\delta_j})/\partial {\delta_m} \ge 0$, and $\sum\limits_{m = 1}^J {\partial f({\delta_j}} )/\partial {\delta_m} < 1$.

We can write${\sigma_j} = \int {\sum\limits_{{C_i} \in {\Omega_j}} {\prod\limits_{l \in {C_i}} {{q_{ilt}}} \prod\limits_{k \not in {C_i}} {(1 - {q_{ikt}}} )} } P(j|{C_i})f(v)dv$,where, and Ω_j denotes the set of choice sets that include j.

$$ \partial f({\delta_j})/\partial {\delta_m} = \frac{1}{{{\sigma_j}}}\int {\sum\limits_{{C_i} \in {\Omega_j}} {\prod\limits_{l \in {C_i}} {{q_{ilt}}} \prod\limits_{k \notin {C_i}} {(1 - {q_{ikt}}} )} } P(j|{C_i})Q_j^mf(v)dv, $$

where

$$ p(j|{C_i}) = \int\limits_v {\frac{{\exp ({\delta_j} + {\chi_j}\bullet \Pi \bullet {D_i} + {\chi_j}\bullet \Sigma \bullet v + \kappa \cdot unuse{d_{ij}} + {\lambda_i} \cdot unuse{d_{ij}} \cdot ad{v_j} + pastchoic{e_{ij}} \bullet \gamma )}}{{\sum\limits_{k = 1}^{51} {\exp ({\delta_k} + {\chi_k}\bullet \Pi \bullet {D_i} + {\chi_k}\bullet \Sigma \bullet v\kappa \cdot unuse{d_{ij}} + {\lambda_i} \cdot unuse{d_{ij}} \cdot ad{v_j} + pastchoic{e_{ij}} \bullet \gamma )} }}} f(v)d(v)$$

$$ \begin{array}{llll} Q_j^m = \frac{{\exp ({\delta_m} + {\chi_m}\bullet \Pi \bullet {D_i} + {\chi_m}\bullet \Sigma \bullet v + \kappa \cdot unuse{d_{im}} + {\lambda_i} \cdot unuse{d_{im}} \cdot ad{v_m} + pastchoic{e_{im}} \bullet \gamma )}}{{\sum\limits_{l \in {C_i}} {\exp ({\delta_l} + {\chi_l}\bullet \Pi \bullet {D_i} + {\chi_l}\bullet \Sigma \bullet v + \kappa \cdot unuse{d_{il}} + {\lambda_i} \cdot unuse{d_{il}} \cdot ad{v_l} + pastchoic{e_{il}} \bullet \gamma )} }},if\begin{array}{*{20}{c}} {m \in {\Omega_j}} & {} \\ \end{array} \hfill \\ \begin{array}{*{20}{c}} {} & { = 0,if\begin{array}{*{20}{c}} {m \notin {\Omega_j}} & {} \\ \end{array} } \\ \end{array} \hfill \\ \end{array} $$

Note that for m = j, $Q_j^m = P(j|{C_i})$

Since all elements in the integral are nonnegative, we have $\partial f({\delta_j})/\partial {\delta_m} \ge 0$.

Moreover, $\sum\limits_{m \in {\Omega_j},m \ne 51} {Q_j^m} < 1$, therefore $\sum\limits_{m \in {\Omega_j},m \ne 51} {\partial f({\delta_j}} )/\partial {\delta_m} \,< \,1$ is satisfied.

5.1.4 5.4.Price Elasticity Calculation

Suppressing the time subscript, we can write the consumer utility function as

$$ {U_{ij}} = {\alpha_i}{p_j} + {\Upsilon_j}g {\beta_\Upsilon }_i + {\varepsilon_{ij}} $$

where ${\alpha_i} = \alpha + {\Pi_3}g {D_i} + {\Sigma_{33}} \cdot {v_3}$, ${\Upsilon_j}$ represents the vector of variables other than price, and ${\beta_{\Upsilon i}}$ the vector of coefficients for ${\Upsilon_j}$.

The formula for price elasticity is given by

$$ {\rho_{jk}} = \frac{{\partial {s_j}}}{{\partial {p_k}}} \cdot \frac{{{p_k}}}{{{s_j}}} = \left\{ {\begin{array}{ll} {\frac{{{p_j}}}{{{s_j}}}\frac{1}{N}\sum\limits_{i = 1}^N {\int {{\alpha_i}{{\hat{p}}_{ij}}(1 - {{\hat{p}}_{ij}})f(v)dv,\begin{array}{ll} {j = k} & {} \\\end{array} } } } \\{ - \frac{{{p_k}}}{{{s_j}}}\frac{1}{N}\sum\limits_{i = 1}^N {\int {{\alpha_i}{{\hat{p}}_{ij}}{{\hat{p}}_{ik}}f(v)dv,\begin{array}{ll} {j \ne k} & {} \\\end{array} } } } \\\end{array} } \right. $$

where p _ij represents the probability that consumer i will choose brand j.

In the estimation, we take NR random draws of v from f(v) to get α_i and compute ρ _jk using the formula

$$ {\hat{\rho }_{jk}} = \left\{ {\begin{array}{lll} {\frac{{{p_j}}}{{{s_j}}}\frac{1}{{N*NR}}\sum\limits_{i = 1}^N {\sum\limits_{nr = 1}^{NR} {\alpha_{_i}^{nr}{{\hat{p}}_{ij}}(1 - {{\hat{p}}_{ij}})} }, j = k} \\{ - \frac{{{p_k}}}{{{s_j}}}\frac{1}{{N*NR}}\sum\limits_{i = 1}^N {\sum\limits_{nr = 1}^{NR} {\alpha_{_i}^{nr}{{\hat{p}}_{ij}}{{\hat{p}}_{ik}},j \ne k} } } \\\end{array} } \right. $$

5. Commentary: Explaining Market Dynamics: Information Versus Prestige

• Mead Over 

1. Center for Global Development, Washington, DC, USA

   Mead Over

Information is valuable to cereal manufacturers, who pay for advertising. Information is valuable to consumers, who reveal by their expenditure response that they attend to advertising. Information is valuable to nutrition activists, as a policy instrument to manipulate in the paternalistic hope that consumers deprived of advertising for sugary cereals will feed their children less sugar. And finally, information is valuable to the authors of the chapter, because using more of it enables them to explain more of the variation in market shares across the cereal brands and to predict more plausibly the reaction of consumers to price or advertising interventions for an individual brand or by a government consumer protection agency

Advertising is one of the industries whose business model involves the packaging and delivery of information. In contrast to the commercial publishing industry, wherein the author and originator of the information profits when the consumer values the information enough to buy the book, profits of the advertising industry derive from the advertiser’s willingness to pay to subsidize information provision to the consumer. The distinction is due to the fact that consumers of books value them for their own sake, whereas consumers of information about advertised products use that information to inform their expenditures on those products. In an imperfectly competitive market for ready-to-eat cereals, cereal manufacturers are willing to subsidize consumers’ information acquisition in order to differentiate brands from one another and reduce consumers’ price elasticity of demand for their own brands.

The chapter deploys a variety of interesting microeconomic modeling and computationally intense econometric techniques to exploit a large data set on consumer purchases of ready-to-eat cereals and estimate the potential effect of a specific type of government intervention in this market: a ban on the advertising of children’s cereal. The authors conclude that such a ban would indeed be effective in reallocating consumer expenditure away from the least healthful types of cereals and toward more healthful, more expensive brands, but it would induce consumers to spend more on cereal than they would without the ban. But one wonders whether the extraordinarily complex econometric paraphernalia the authors would really be required to show these impacts of advertising.

Since the authors generously provide the market shares of the top 50 brands as well as their average prices, brand-specific monthly advertising expenses, and market segment (in Table 5.2), one can calculate a descriptive ordinary least squares regression of (the logit of) market share on this grouped data. The results of this “naïve” regression are presented here in Table 5.C.1.

Table 5.C.1 Ordinary least squares regression of logit of average market share on log price and advertising expenditures, by market segment

Although requiring very little effort beyond the tabulation of the average market shares, prices, and advertising expenditures for the 50 top brands, these results seem somewhat informative. The point estimates of the three estimated price coefficients, one for each of the three market segments, are all negative, as expected, with the one for family cereals being large (>2 in absolute magnitude) and statistically significant. Furthermore, all three advertising coefficients are highly statistically significant, suggesting that an extra million dollars of advertising increases market share by 0.86% for adult cereals, 0.92% for family cereals, and 1.05% for kid cereals. The category of kid cereals seems to respond more to advertising expenditures than the other two.

So why do more? What have the authors’ prodigious efforts added to our knowledge of the ready-to-eat cereal market?

This chapter supports the proposition that “information is valuable to economic researchers” in three ways. First, by exploiting detailed information on the thousands of individual consumer transactions summarized in Table 5.2, the authors are able to relax several of the assumptions that are maintained by the above naïve analysis. In so doing, they demonstrate the value of that detailed information to the understanding of this complex market. Second, by bringing to bear an economic theory of decision making, the authors demonstrate that this theory itself has information content—because it helps explain the market data. Third, by combining the unusually detailed and granular data with this powerful theory, the authors are able to distinguish the two channels by which advertising hypothetically affects consumer behavior, the “information” channel and the “prestige” channel, and to demonstrate that it’s the information that influences the consumer’s behavior—not the prestige. Fourth, by using information from the supply side of the cereal market, the authors are able to reject some types of endogeneity that would cast doubt not only on my naïve model, but also on their three principal models.

Consider the estimated price elasticities. Figure 5.C.1 displays for each of the three market segments the confidence intervals for my naively estimated price elasticities from Table 5.C.1 and the range of estimated elasticities for the top 10 cereal brands presented by the authors in their Table 5.8. There are two adult cereal brands in the top 10, six family brands and two kid brands. Note the extremely wide confidence intervals from my naïve estimates. Next to those confidence intervals (in green), my Fig. 5.C.1 displays the range of estimated price elasticities for each of the authors’ three estimated models. Although the authors do not report confidence intervals, the point estimates of the brand-specific coefficient estimates from which these elasticities are derived (the first row of Table 5.5) are from 3 to 50 times larger than their estimated standard errors, suggesting tight confidence intervals for the elasticities. And the range of these reported estimates is also relatively tight within each market segment. Thus, one benefit of the information in the granular data appears to be tighter estimates of the brand-specific price elasticities.

Fig. 5.C.1

Adding information either with more granular data or more theory-constrained economic structure increases both the precision and heterogeneity of estimated price elasticities across brands

The authors’ basic model is a random-coefficients logit model (RCL) structured to assume that the choice sets for all consumers include all 50 brands (plus a 51st composite of all other brands) and characteristics of all brands are known. Figure 5.C.1 shows that the estimated elasticities for this model are roughly the same across the three market segments. (See the orange boxes in Fig. 5.C.1.) The authors’ second model, whose elasticity estimates are represented by the blue boxes labeled “RCL + Learning,” relaxes the assumption that all consumers know the characteristics of all brands. In this model the consumers again choose among all brands but only know the qualities of brands previously purchased. Advertising directly influences a brands market share. Thus, the impact of the economic theory on the estimated elasticities is to differentiate the three theoretically distinct markets, information that is useful to students of this ready-to-eat cereal market. Finally when the authors use an elaborate simulation model to require advertising to inform consumers of an unused brand’s existence before it can affect their purchases (the assumption of heterogeneous choice sets), the estimated elasticities diverge even more across the three market segments (pink boxes) and also increase substantially in absolute magnitude. In the words of the authors, “[t]he estimated price elasticities in the … specification [allowing a heterogeneous choice set] are more plausible, since their absolute values are all bigger than 1, which is consistent with the fact that profit-maximizing firms should be operating at the elastic part of the demand curve.” Once more, economic theory has improved the fit of the model and contributed insight on the cereal market.

Variation in observed market shares, the naïve model contains substantial information. Its prediction error (defined by the authors as the square root of the sum of squared differences between the actual market share of Table 5.6 and the predicted share) equals 6.5, which is actually less than the 7.26 scored by the authors’ random-coefficients model (bottom row of Table 5.6). However, both of the authors’ more sophisticated models do better than my naïve model, scoring 5.28 and 3.81 respectively, and thus can be said to contain more valuable information.

Because they are able to simulate the consumers’ choice sets each time on each visit to the grocery store, the authors can distinguish the two possible channels by which advertising might induce people to spend more on cereal—the information channel and the prestige channel. It’s interesting that for this market, the authors find no support for the hypothesis that advertising persuades consumers to increase their consumption of ready-to-eat cereals that are familiar to them—which would be a prestige effect of advertising. Instead, advertising’s role seems to be to induce consumers to try cereals that are unfamiliar. When they model this effect, the authors estimate much larger price elasticities (the pink boxes in Fig. 5.C.1). Since consumers have many choices in the cereal market, evidence that price elasticities are large in the children’s cereal market and small in the adult cereal market suggests that the adults who purchase cereal for children see them as highly substitutable for one another, whereas they are loath to substitute one adult cereal for another. Adult cereal brands thus have more market power than children’s brands.³⁰

The authors’ simulations of a ban on advertising for children’s cereal and of a “pulsed” advertising strategy both raise the issue of the potential value to the public of government use of advertising. Using their third model, which incorporates consumer learning and heterogeneous choice sets, and assuming that affected cereal manufacturers hold constant the prices of their brands, the authors simulate a ban on advertising and conclude that “the total market share of kid brands goes down by about 6%, of which 2% goes to the adult brands and 4% goes to the family brands.” It’s possible to perform this same experiment with the naïve model, by first computing the fitted market shares from the OLS regression in the children’s market and then computing them a second time after the value of advertising has been set to zero. The result from the naïve model is that the total market share of children’s brands would decline from 17.7 to 9.5% of the market, a reduction of about 8.2%. Under the assumption of the independence of irrelevant alternatives (the well-known IIA assumption typically maintained in multinomial logit models), about 2.2 percentage points of this decline would be reflected by an increase in the adult segment and about 5.8% age points would go to the family segment. Despite the simplicity of the naïve model, these results are remarkably similar to those obtained by the authors.

In contrast to the ban on advertising of children’s cereals, the possible effects on the market of pulsed advertising could not be analyzed with the naïve model. The authors have used their heterogeneous choice set model to show that spreading the same advertising dollars smoothly is less effective at increasing market share than would be a strategy of bunching the advertising in specific months. The superior effectiveness of pulsing seems to be due to the lack of a prestige effect of advertising in this market. The implication is that government public awareness campaigns that intend to improve people’s awareness of alternatives—and subsequently depend on their good experience with these alternatives to motivate behavior—could also benefit from pulse advertising. Whether the reverse is true for public awareness campaigns that intend to enhance the prestige of certain behavior remains to be determined.

The authors allude in passing to monopolistic pricing strategies when they point out that a monopolist operates in the elastic portion of its demand curve. Under certain conditions one could go further and assert that a profit-maximizing firm in a monopolistic or monopolistically competitive market will set its price-cost margin equal to the inverse of the elasticity of demand. According to the authors’ heterogeneous choice set model, the median elasticities in the adult, family, and children’s market segments are about −1.5, −2.3, and −2.8, respectively. This suggests that typical markups of price over marginal cost in these three segments are 65, 49, and 38 % of marginal costs, respectively. Furthermore, markups on individual brands vary from 34 to 72% of marginal costs. This information is of only academic interest in the market for ready-to-eat cereals, imagine if a similar analysis of the pharmaceutical market revealed such information about the prices of pharmaceutical brands. Views on pharmaceutical pricing range from the idea that monopoly profits in the pharmaceutical market are unproductive “rent” gained from branding products that largely result from government-subsidized research to the position that these profits are a just return on pharmaceutical firms’ own research investments and motivate their future research. An objective observer would grant that both views have some legitimacy in various parts of the market. But policy intervention on the prices of individual drugs is hampered by the secrecy with which pharmaceutical firms guard their cost information. To the extent that the techniques employed in this chapter could be used to reveal the apparent markups of pharmaceutical prices over costs, regulators would value this information as an input to the regulation of the monopoly prices of individual pharmaceutical products.