`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 1 of 120
`
`EXHIBIT 6
`EXHIBIT 6
`PUBLIC REDACTED VERSION
`PUBLIC REDACTED VERSION
`
`
`
`
`
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 2 of 120
`
`
`
`
`
`UNITED STATES DISTRICT COURT
`NORTHERN DISTRICT OF CALIFORNIA
`SAN FRANCISCO DIVISION
`
`
`IN RE GOOGLE PLAY
`ANTITRUST LITIGATION
`
`STORE
`
`THIS DOCUMENT RELATES TO:
`
`In re Google Play Consumer Antitrust
`Litigation, Case No. 3:20-cv-05761-JD
`
`State of Utah et al. v. Google LLC et al., Case
`No. 3:21-cv-05227-JD
`
`No. 3:21-md-02981-JD
`
`MERITS REPLY REPORT OF
`
`HAL J. SINGER, PH.D.
`
`Judge: Hon. James Donato
`
`
`
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`
`
`
`
`
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 3 of 120
`
`
`
`-2-
`
`
`Introduction and Summary of Conclusions .................................................................................... 5
`Qualifications .................................................................................................................................. 5
`I.
`Dr. Leonard Fails to Undermine My Conclusions .............................................................. 5
`Dr. Leonard Fails to Undermine the Standard Economics and Rigorous Empirical
`A.
`Analysis That I Use to Establish Pass-Through ...................................................... 6
`Dr. Leonard Baselessly Rejects Standard Economic Methods Deriving
`1.
`Pass-Through From a Market-Wide Change in Costs ................................ 8
`Dr. Leonard Baselessly Rejects Standard Econometric Methods
`Demonstrating Developer Pass-Through of Ad Valorem Sales Taxes ....... 9
`Dr. Leonard Baselessly Rejects Standard Economic Principles of Pass-
`Through ..................................................................................................... 10
`Dr. Leonard’s Empirical Estimates of Pass-Through Are Biased and
`Unreliable .................................................................................................. 11
`a. Dr. Leonard’s “Real-World Examples” Do Not Reliably Estimate Pass-
`Through In the But-For World ...................................................... 13
`b. Dr. Leonard’s “Synthetic Control” Regressions Do Not Reliably
`Estimate Pass-Through In the But-For World .............................. 19
`Dr. Leonard Fails to Undermine the Well-Established Economic Models and
`Methods That I Use to Estimate Take Rates in a More Competitive But-For
`World .................................................................................................................... 21
`Dr. Leonard’s Critiques of My Direct Consumer Discount Model Are Without
`Merit ...................................................................................................................... 25
`The Claims of Dr. Leonard and the Other Google Experts Regarding Markets In
`China Do Not Undermine my Conclusions .......................................................... 26
`Dr. Leonard’s Remaining Critiques Are Without Merit ....................................... 28
`E.
`Dr. Gentzkow Fails to Undermine My Conclusions ......................................................... 31
`Dr. Gentzkow Does Not Demonstrate That Google Lacked Monopoly Power in
`A.
`the Relevant Antitrust Markets At Issue During the Class Period ........................ 32
`Dr. Gentzkow Fails To Demonstrate the Challenged Conduct Is Procompetitive 34
`Dr. Gentzkow Fails to Demonstrate that the Restrictions on Creating Even
`1.
`One Non-Google Version of Android in the AFA Are Procompetitive ... 35
`Dr. Gentzkow Fails to Demonstrate that the MADAs Are Procompetitive
`................................................................................................................... 36
`Dr. Gentzkow Fails
`to Demonstrate
`that
`the Early RSAs Were
`Procompetitive .......................................................................................... 40
`Dr. Gentzkow Fails
`to Demonstrate
`that
`the Later RSAs Were
`Procompetitive .......................................................................................... 43
`Dr. Gentzkow Fails to Demonstrate that Project Hug Is Procompetitive . 44
`Dr. Gentzkow Fails to Demonstrate that Google’s Unnecessary Technical
`Barriers Are Procompetitive ..................................................................... 46
`
`2.
`
`3.
`
`4.
`
`2.
`
`3.
`
`4.
`
`5.
`6.
`
`B.
`
`C.
`
`D.
`
`B.
`
`II.
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 4 of 120
`
`
`
`-3-
`
`C.
`
`D.
`
`B.
`
`III.
`
`7.
`
`8.
`
`9.
`
`2.
`
`3.
`4.
`
`2.
`
`3.
`
`4.
`
`2.
`
`3.
`
`4.
`5.
`
`Dr. Gentzkow Fails to Demonstrate that Google’s Developer Distribution
`Agreements Are Procompetitive ............................................................... 47
`Dr. Gentzkow Fails to Demonstrate that the Aftermarket Tie-In Is
`Procompetitive .......................................................................................... 48
`Dr. Gentzkow Fails to Demonstrate Other Challenged Conduct Is
`Procompetitive .......................................................................................... 52
`a. Project Banyan ...................................................................................... 52
`b. App Campaigns ..................................................................................... 52
`c. Project Alley-Oop ................................................................................. 53
`Dr. Gentzkow Ignores Standard Antitrust Principles ........................................... 56
`Dr. Gentzkow’s Conclusion That “Competition Has Not Been Foreclosed”
`1.
`Does Not Follow From Standard Antitrust Principles .............................. 56
`Dr. Gentzkow Ignores the Coordinated Nature and Effects of the
`Challenged Conduct .................................................................................. 66
`Dr. Gentzkow’s “Fragmentation” Defense Is Without Merit ................... 67
`Dr. Gentzkow Incorrectly Assumes That All Improvements in Output and
`Quality Are Attributable to the Challenged Conduct ............................... 69
`Dr. Gentzkow Fails to Demonstrate That Consumer Plaintiffs Have Not Suffered
`Antitrust Injury Resulting From Google’s Anticompetitive Conduct, or that
`Consumer Plaintiffs “Would Likely Be Worse Off” Absent The Challenged
`Conduct ................................................................................................................. 71
`Dr. Tucker Fails to Undermine My Conclusions .............................................................. 72
`Dr. Tucker’s Proposed Megamarket Disregards Elementary Principles of Antitrust
`A.
`Economics ............................................................................................................. 75
`A Relevant Antitrust Market Is Comprised of Substitutes, Not
`1.
`Complements ............................................................................................ 75
`A Relevant Antitrust Market Should Contain the Smallest Set of Substitutes
`Necessary To Exercise Market Power ...................................................... 77
`The Relevant Antitrust Market Is Defined Solely Based on Demand-Side
`Substitution ............................................................................................... 78
`Dr. Tucker’s Proposed Megamarket Suffers from the Cellophane Fallacy
`................................................................................................................... 78
`Dr. Tucker’s Critiques of My Market Definitions Are Without Merit ................. 79
`The Market For Licensable Mobile Operating Systems Is A Relevant
`1.
`Antitrust Product Market .......................................................................... 79
`The Android App Distribution Market Is Separate From the In-App
`Aftermarket ............................................................................................... 84
`The Android App Distribution Market Is A Relevant Antitrust Product
`Market ....................................................................................................... 86
`The In-App Aftermarket Is A Relevant Antitrust Product Market ........... 89
`Dr. Tucker’s Claims On The Relevant Geographic Market Do Not
`Undermine My Conclusions ..................................................................... 90
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 5 of 120
`
`-4-
`
`
`C.
`
`Dr. Tucker’s Report Fails To Show That Google Lacks Monopoly Power In The
`Relevant Markets At Issue .................................................................................... 90
`Dr. Skinner Fails to Undermine My Conclusions ............................................................. 96
`IV.
`Conclusion .................................................................................................................................... 97
`Appendix 1: Materials Relied Upon ............................................................................................. 98
`Appendix 2: Dr. Leonard’s “Real World Examples” Adjusted for Inflation ............................. 105
`Appendix 3: Linear Demand Does Not Fit the Data Well .......................................................... 113
`Appendix 4: Tax-Rate Regressions Collapsed Nationwide ........................................................ 115
`Appendix 5: Flaws In Google Experts’ Take Rate Benchmarks ................................................ 116
`
`
`
`
`
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 6 of 120
`
`
`
`-5-
`
`INTRODUCTION AND SUMMARY OF CONCLUSIONS
`1.
`I have been asked by counsel for Consumer Plaintiffs to respond to the expert
`reports of Drs. Gregory Leonard,1 Matthew Gentzkow,2 Catherine Tucker,3 and Douglas Skinner4
`(collectively, the “Google Experts”). As detailed below, having carefully considered the reports of
`the Google Experts, I am not inclined to materially alter the opinions expressed in my report on
`the merits of this case (“Merits Report”), or those in my prior class certification reports.5
`
`QUALIFICATIONS
`My qualifications are provided in my prior reports.
`
`2.
`
`I. DR. LEONARD FAILS TO UNDERMINE MY CONCLUSIONS
`3.
`Below I explain why Dr. Leonard’s critiques of my analysis do not undermine my
`conclusions. As a preliminary matter, I note that Dr. Leonard and other Google economists claim
`that modeling a competitive but-for world in this case requires a degree of specificity that would
`be impractical or impossible to satisfy in any antitrust case.6 None of the Google economists
`explain what economic models might possibly satisfy the conditions that they articulate or the data
`
`
`1. Expert Report of Dr. Gregory K. Leonard (November 18, 2022) [hereafter, Leonard Report].
`2. Expert Report of Matthew Gentzkow (November 18, 2022) [hereafter, Gentzkow Report].
`3. Expert Report of Catherine E. Tucker (November 18, 2022) [hereafter, Tucker Report].
`4. Expert Report of Douglas J. Skinner (November 18, 2022) [hereafter, Skinner Report].
`5. Merits Report of Hal J. Singer, Ph.D. (October 3, 2022) [hereafter, Singer Merits Report]. Unless otherwise
`defined, capitalized terms herein are defined the same as they are in the Singer Merits Report, and in my class
`certification reports. See Class Certification Report of Hal J. Singer, PhD (February 28, 2022) [hereafter, “Singer Class
`Cert Report”]; see also Class Certification Reply Report of Hal J. Singer, PhD (April 25, 2022) [hereafter, “Singer
`Class Cert Reply”]; see also Class Certification Reply Report of Hal J. Singer, PhD (Errata) (May 10, 2022) [hereafter,
`“Singer Class Cert Reply Errata”]. The materials I relied upon in forming my opinions are noted in the footnotes
`throughout this report or otherwise listed in Appendix 1, or in my prior reports. I reserve the right to supplement,
`expand, or amend my opinions. All of my economic models use Play Store transaction data produced by Google
`(“Google Transactional Data”). The Google Transactional Data includes billions of records, and was produced in two
`batches. The first batch (GOOG-PLAY-007203251) was produced on July 27, 2021, and includes U.S. transactions
`from November 2010 through July 3, 2021. The second batch was produced on August 17, 2022, and includes U.S.
`transactions between July 4, 2021 through May 31, 2022. As of the filing date of my Merits Report (October 3, 2022),
`portions of the second batch had not been fully incorporated into my analysis due to the substantial computational
`burden of processing and analyzing billions of records. On October 19, 2022, I produced an amended version of my
`Merits Report reflecting the fully processed transaction data. This update did not materially alter my conclusions.
`6. Leonard Report ¶¶27-29 (“Specifying the but-for world is a complex undertaking…In this case, the important
`aspects of the but-for world that are relevant for Plaintiffs’ experts damages calculations, but that they either did not
`address at all or did not address clearly include (1) what app stores or types of app stores would have entered, when
`would they have entered, what devices would they be on, and which apps would have been available in those stores;
`(2) how Google would have changed its level of investment in Android, Google Play and app developer support; (3)
`if, how, and why Google would have changed its monetization strategy; (4) how consumers would have been affected
`by the existence of additional app stores (e.g., greater search costs or increased malware on Android devices); (5) how
`developers would have been affected by the existence of additional app stores and/or multiple Android-based OSs
`(e.g., greater distribution costs required with multiple stores or multiple OSs); (6) to the extent there is a claim that
`there would have been additional apps in the but-for world, the identity, attributes, and quality of those apps; and so
`on.”). See also Gentzkow Report ¶636.
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 7 of 120
`
`
`that would be required to implement them, because such models do not exist, and the data
`requirements would likely be impossible to satisfy in any event.
`
`-6-
`
`A.
`
`Dr. Leonard Fails to Undermine the Standard Economics and Rigorous Empirical
`Analysis That I Use to Establish Pass-Through
`
`4.
`In my Merits Report, I explained why, as a matter of standard economic principles,
`profit-maximizing firms engage in pass-through: When their costs increase, firms have a clear
`economic incentive to charge higher prices; conversely, when costs decrease, the incentive is to
`reduce prices to increase volume of sales, thus maximizing profit.7 The “pass-through rate” gives
`the change in a firm’s profit-maximizing price resulting from a given change in marginal cost.8 To
`assess the pass-through rate in this case, I applied standard economic models and standard
`econometric methods to Google’s voluminous transaction data.
`
`5.
`I have previously explained why Google’s limited take rate reductions in the actual
`world do not provide a reliable basis for estimating the market-wide pass-through that would have
`occurred in a more competitive but-for world.9 Nevertheless, Dr. Leonard attempts to estimate
`pass-through using flawed, SKU-level comparisons similar to those employed by Google’s class
`certification expert (Dr. Burtis).10 I explain why Dr. Leonard’s analysis is flawed and unreliable in
`Part I.A.4 below.
`
`6.
`Instead of Dr. Leonard’s unreliable approach to estimating pass-through, I used
`standard econometric methods to measure the shape of the demand curve facing app developers,
`which determines the pass-through rate, and concluded that the standard logit model fits the data
`well.11 Dr. Leonard claims that “today it is uncommon for an empirical economics research study
`to use this [logit] model.”12 Dr. Leonard is wrong. My Merits Report cites published, peer-
`
`
`7. Singer Merits Report Parts VI.D.1-2.
`8. For example, if a firm’s costs increase by $10 per unit, and the firm responds by raising its price by $5, the
`pass-through rate is $5/$10 = 50 percent. Dr. Leonard agrees with this definition. Leonard Report ¶30.
`9.
`In a more competitive but-for world, all or almost all developers would enjoy substantially and permanently
`lower take rates. See, e.g., Singer Class Reply ¶10. In contrast, Google’s take rate reductions have been generally
`limited to narrowly defined SKUs comprising a small share of developer revenue, and/or short time horizons, making
`it unlikely that any change take rates would have had a material effect on developer finances or pricing. See, e.g.,
`Singer Class Reply ¶103; ¶126. In addition, Google’s 2018 reduction in subscription take rates applied only to users
`that had maintained their subscriptions for at least one year. If a consumer has subscribed to a product for a year (or
`more) and paid the same monthly price, that consumer has revealed a strong willingness to pay for the subscription
`offering, and it would make little economic sense for a developer to target price cuts to its least price-sensitive
`customers. Id. at ¶119. Moreover, the pricing rules in the Play Store’s developer interface likely made it difficult or
`impossible for most developers to drop prices to subscribers after the first year, even if they had wanted to do so. Id.
`at ¶122. In addition, developers would be incentivized to pass on savings from a lower take rate via steering and
`discounting, to induce consumers to switch to the low-cost provider. These incentives are absent in the actual world.
`Developers that enjoyed Google’s limited take-rate decreases in the actual world did not have to share any of the
`savings with their customers in order to realize the cost savings. Singer Merits Report ¶369.
`10. Singer Class Reply ¶¶123-133.
`11. Singer Merits Report Part VI.D.3.
`12. Leonard Report ¶72. Dr. Leonard claims that the logit model is “highly restrictive.” Id. In fact, economists
`recognize that “the logit model is the ideal rather than a restriction.” Kenneth Train, Logit, in DISCRETE CHOICE
`METHODS WITH SIMULATION, 36 (Cambridge University Press 2009).
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 8 of 120
`
`-7-
`
`
`reviewed research demonstrating otherwise.13 Among many other things, I cite to a peer-reviewed
`2018 article authored by academics and DOJ economists explaining that logit is one of the primary
`models included in the antitrust software package developed by DOJ economists.14 This article
`and others that I cite were published more recently than all the papers that Dr. Leonard relies upon
`to support his incorrect claim, most of which were published in the 1980s or 1990s.15
`
`7.
`Dr. Leonard does not dispute that the standard logit demand systems that I used to
`calculate pass-through explain over 95 percent of the variation in consumer demand in the
`voluminous Google transaction data.16 Nor does he dispute that, consistent with economic
`expectations, the logit regression results confirm a negative and highly statistically significant
`relationship between demand and price.17 Instead, Dr. Leonard claims erroneously that I did not
`provide sufficient empirical justification for the standard logit demand systems that I employ
`here.18 I follow standard practice in empirical antitrust work, in which the form of the demand
`curve is assessed based on “how well the model fits the observable data.”19
`
`8.
`In my Merits Report, I also provided corroborative empirical evidence of pass-
`through by demonstrating that higher tax rates (which as I discuss below are economically
`analogous to higher take rates) are systematically passed on by developers to consumers in the
`form of higher prices.20 The combination of standard economic principles and my empirical
`analysis allowed me to conclude that lower take rates would have resulted in lower prices for
`consumers in a more competitive but-for world and to quantify the resulting overcharge for
`purposes of determining impact and calculating damages to consumers in the United States who
`purchased Apps or In-App Content.21
`
`
`13. Singer Merits Report ¶348, n. 809; ¶356, n. 835.
`14. Singer Merits Report ¶348, n. 809, citing Luke Froeb et al., Economics at the Antitrust Division: 2017–2018,
`53 REVIEW OF INDUSTRIAL ORGANIZATION 637, 639-642 (2018).
`15. Leonard Report ¶72, n. 76. Dr. Leonard relies upon papers that advocate “random coefficient” logit models
`(or “mixed logit” models), which are sometimes used by academics, but seldom by antitrust practitioners. As I
`explained in my Merits Report, in practice, these techniques suffer from well-documented computational problems,
`which can severely limit their applicability and accuracy when applied to real-world data sets. Singer Merits Report
`¶348, n. 810.
`16. Singer Merits Report ¶354.
`17. Id.
`18. Leonard Report ¶¶68-72.
`19. Luke Froeb et. al., Economics at the Antitrust Division: 2017–2018, 53 REVIEW OF INDUSTRIAL
`ORGANIZATION 637, 640 (2018). The approach I adopt is more rigorous than assuming that the demand curve has a
`standard form (such as logit) and then calibrating the demand system to the data based on that assumption. See, e.g.,
`Nathan Miller, Marc Remer, & Gloria Sheu, Using cost pass-through to calibrate demand, 118 ECONOMICS LETTERS
`451 (2013) (“Researchers in industrial economics frequently conduct counter-factual experiments based on
`parameterized systems of consumer demand. The functional form of demand is assumed and the structural parameters
`are either estimated from data or calibrated.”) (emphasis added). See also Froeb et. al. (2018), supra, at 640-643. As
`shown in Appendix 3, linear demand does not fit the data well. Contrary to economics, the price coefficients are
`statistically insignificant in 12 out of the 33 regressions. By contrast, the price coefficients are statistically significant
`for each and every app category for the logit model. Singer Merits Report Table 12.
`20. Singer Merits Report Part VI.D.4.b.
`21. In my Merits Report, I demonstrated empirically that my damages models can accommodate focal-point
`pricing, and that at most a de minimis share of apps would not lower price due to focal-point pricing in the but-for
`
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 9 of 120
`
`
`1.
`
`-8-
`
`Dr. Leonard Baselessly Rejects Standard Economic Methods Deriving Pass-
`Through From a Market-Wide Change in Costs
`
`9.
`In a more competitive but-for world, all or almost all developers would benefit from
`substantially and permanently lower costs, owing to substantially and permanently lower take
`rates. Accordingly, in my Merits Report, I estimated the pass-through rate by applying standard
`economic calculations of market-wide cost-pass-through.22 These calculations solve for each
`developer’s profit-maximizing price decrease to consumers, given what other developers are
`charging in light of lower market-wide costs.23 This provides the correct pass-through rate—
`namely, the extent to which prices for App and In-App Content would decline in the but-for world,
`given the market-wide decrease in cost that would occur, pushing the market to a new, more
`competitive equilibrium. Consistent with standard practice, I calculated the change in equilibrium
`prices resulting from a lower take rate by multiplying (1) the decrease in cost resulting from a
`lower take rate by (2) the pass-through rate.24
`
`10. Without citation to any authority, Dr. Leonard erroneously claims that the “correct
`way” to solve for the pass-through rate “is to ask the question, ‘[H]ow the service fee rate change
`would affect the profit maximizing price?’”25 In making this unsupported claim, Dr. Leonard
`ignores the elementary economic fact that developers adjust their prices based on changes in their
`actual costs (in dollars), not in the “service fee rate” per se (which is not a dollar amount). Dr.
`Leonard’s calculations in Appendix D of his report proceed from this flawed premise26 despite the
`fact that what matters for pass-through is the change in costs resulting from a change in the take
`rate, which is what I have calculated.27
`
`11.
`Dr. Leonard’s error is compounded by the fact that his calculations do not account
`for the market-wide nature of the cost decrease that would occur in a more competitive but-for
`world; Dr. Leonard’s calculations in Appendix D of his report are limited to how a single developer
`reacts to a change in its own take rate. In other words, Dr. Leonard’s calculations in Appendix D
`
`
`world. Singer Merits Report ¶¶408-413. Dr. Leonard ignores this quantitative analysis. Leonard Report ¶31, n. 5. Dr.
`Leonard claims incorrectly that the lack of steering in the actual world contributes to focal-point pricing. Leonard
`Report ¶32, n. 7. In fact, developers in the but-for world would have economic incentives to depart from focal-point
`pricing in increments of $1; moreover, in the few episodes where we do observe steering in the actual world,
`developers have been observed to deviate from $1 pricing increments. Singer Merits Report ¶405. In addition, Apple
`recently announced new price points in App Store. In addition to allowing ten-cent intervals, Apple also lowered its
`minimum price from $0.99 to $0.29. This provides additional evidence that focal-point pricing in $1 increments is far
`from economically inevitable. See Apple, Apple announces biggest upgrade to App Store pricing, adding 700 new
`price points (Dec. 6, 2022), https://www.apple.com/newsroom/2022/12/apple-announces-biggest-upgrade-to-app-
`store-pricing-adding-700-new-price-points/.
`22. Singer Merits Report ¶¶358-360.
`23. Singer Merits Report ¶¶358-360; ¶337, n. 795.
`24. Singer Merits Report ¶363; ¶337 n. 795.
`25. Leonard Report Appendix D, ¶2 (emphasis in original).
`
`26. Leonard Report Appendix D, ¶2 (“That is, the correct calculation would be based on ∂P/∂t, where 𝑃 is the
`app price set by an app developer and 𝑡 is the service fee rate.”)
`
`27. Singer Merits Report ¶363; ¶337 n. 795.
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 10 of 120
`
`
`do not account for the fact that developers would decrease their prices in response to other
`developers’ price cuts.28
`
`-9-
`
`12.
`Despite the voluminous empirical econometric literature on pass-through, Dr.
`Leonard does not (and cannot) cite to a single instance of his formula ever being used in a peer-
`reviewed research article (or anywhere else) for any purpose, let alone to calculate pass-through.
`
`2.
`
`Dr. Leonard Baselessly Rejects Standard Econometric Methods
`Demonstrating Developer Pass-Through of Ad Valorem Sales Taxes
`
`13.
`In Table 15 of my Merits Report, I applied standard regression methods to Google’s
`transaction data to demonstrate empirically that developers systematically pass on higher tax rates
`imposed by state or local authorities (that is, higher ad valorem costs) in the form of higher prices
`paid by Consumer Plaintiffs.29 Dr. Leonard mistakenly claims that these regressions are
`“mathematically guaranteed” to show 100 percent pass-through.30 To support this claim, Dr.
`Leonard misleadingly focuses on one of the regressions (column (1) of Table 15 of my Merits
`Report).31 Dr. Leonard conveniently ignores the very next column of Table 15, which shows pass-
`through less than 100 percent: According to column (2) of Table 15, a one percentage-point
`increase in the sales tax rate leads to a 0.7 percent increase in the post-tax price. To illustrate,
`suppose that the post-tax price of an App is initially $2.00, and that the tax rate is five percent, so
`that the pre-tax price is $2.00/(1.05) = $1.90. If the tax rate increases by five percentage points (to
`ten percent), the post-tax price will increase by [0.7 percent x 5] = 3.5 percent, resulting in a new
`post-tax price of [$2.00 x 1.035] = $2.07. The new pre-tax price will be $2.07/1.10 = $1.88. The
`pass-through rate in this example is 75 percent (equal to the change in the post-tax price divided
`by the change in the tax amount, or [$2.07 - $2.00]/[($2.07 -$1.88) - ($2.00 -$1.90)]). The pass-
`through rate is below 100 percent because the developer decreases its pre-tax price in response to
`an increase in the tax rate.
`
`14.
`Dr. Leonard claims my regressions in Table 15 do not measure pass-through
`because developers lack the “ability to adjust pre-tax prices in response to tax rate variation as a
`way to absorb taxes.”32 This is incorrect; developers are free to adjust the pre-tax price in response
`to anything they choose, including sales taxes. More fundamentally, Dr. Leonard ignores the
`elementary economic principle that, when a tax is levied on a product or service, the pass-through
`rate determines the extent to which a consumer bears the cost of the tax; this is true regardless of
`whether the tax is levied on the firm or on the consumer.33 Therefore, the extent to which the
`
`28. Leonard Report Appendix D, ¶¶5-6 (Equations (A.2) - (A.9) are solved individually, instead of on a
`marketwide basis).
`29. Singer Merits Report ¶368.
`30. Leonard Report ¶82.
`31. Leonard Report ¶82.
`32. Leonard Report ¶81.
`33. Singer Merits Report ¶367, citing N. GREGORY MANKIW, PRINCIPLES OF MICROECONOMICS 120-127
`(Cengage Learning 8th ed. 2018) [hereafter, MANKIW]. MANKIW at 123 (“Taxes levied on sellers and taxes levied on
`buyers are equivalent. In both cases, the tax places a wedge between the price that buyers pay and the price that sellers
`receive. The wedge between the buyers’ price and the sellers’ price is the same, regardless of whether the tax is levied
`on buyers or sellers…The only difference between a tax levied on sellers and a tax levied on buyers is who sends the
`
`
`
`PARTY AND NON-PARTY HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY
`
`
`
`Case 3:21-md-02981-JD Document 487-7 Filed 04/20/23 Page 11 of 120
`
`-10-
`
`
`before-tax price does or does not adjust in response to a change in the tax rate is directly
`informative of the pass-through rate. The regressions in Table 15 of my Merits Report are grounded
`in these standard principles.
`
`15.
`Dr. Leonard claims that “Google’s system does not allow developers to
`systematically set different pre-tax prices for different states in the US,” but my regressions do not
`rely on “tax variation across states.”34 Dr. Leonard ignores that my regressions control for state
`fixed effects.35 The inclusion of state fixed effects means that my regressions do not measure pass-
`through by comparing tax rates and prices across different states, as Dr. Leonard mistakenly
`suggests.36
`
`3.
`
`Dr. Leonard Baselessly Rejects Standard Economic Principles of Pass-
`Through
`
`16.
`In the past in his own published work, Dr. Leonard has acknowledged the standard
`economic principle that firms charge higher prices when their costs are higher and lower prices
`when costs are lower. He has written: “Economic theory makes a straightforward prediction: The
`decrease in cost will lead to a decrease in price, with the relationship between the decreases in cost
`and price depending on the shape of the demand curve.”37 As Dr. Leonard makes clear in his
`article, these fundamental conclusions regarding pass-through are not just abstract theoretical
`results; they apply directly to real-world economic outcomes.38 Yet Dr. Leonard critiques me for
`estimating pass-through based on the shape of the demand curve in this case.39
`
`17.
`In contrast, in his expert report, Dr. Leonard distances his opinions from these
`fundamental economic principles. Dr. Leonard claims baselessly that pass-through could be
`“negative” in the context of App pricing. In other words, Dr. Leonard speculates that developers
`
`money to the government.”) (emphasis in original). Dr. Leonard ignores this basic principle when he claims that my
`regression is “uninformative about the pass-through of service fees,” because “while the service fees are levied directly
`on the developers, sales taxes are levied directly on consumers.” Leonard Report