Innovation and Dynamic Efficiencies in Merger Review

Prepared By:
Andrew Tepperman and Margaret Sanderson
CRA International

Date: April 9, 2007
CRA Project No. D09208-00

3. Innovation in a Competitive Effects Analysis: Threshold Issues

To usefully incorporate innovation and dynamic efficiency into merger review we need to recognize several distinguishing features of innovative competition.  First, there is no robust model of innovation that links the rate of innovation to industry concentration.  Second, there is far more uncertainty surrounding innovation than is normally encountered in a typical merger review focused solely on price and output.  Third, associated with innovation are complicating measurement problems that do not ordinarily exist with more conventional merger analysis. 

These issues do not prevent us from taking innovation and dynamic efficiency into account when assessing mergers.  But they do mean that we cannot rely on simple measures to quantify levels of innovative activity or to confidently predict how mergers will impact on innovative activity.  Instead, as we discuss in the subsequent section, a more fact-intensive, case-by-case approach is required.

3.1. No Predictive Model Relating Innovation to Concentration

When considering the potential price effects of a merger in the short run, economists typically make reference to a number of accepted models of competitive firm interaction.²⁷  Such models provide guidance on the relationship between firm concentration and price, once other variables such as the demand elasticity and ease of entry are defined.  We would like to have a similar model that relates innovation to industry concentration.  In particular, we would like to know whether the rate at which firms pursue a particular innovation (and accordingly their likelihood of success) increases with the number of firms involved in the pursuit.²⁸  Unfortunately, a robust framework that predicts the amount by which innovation will increase for a given increase in the number of firms has proven elusive. 

One model that is frequently invoked (to varying degrees of explicitness) to address this issue is the “patent race” model.  Firms in patent race models interact under the following specific set of assumptions: (i) all firms are assumed to be pursuing the same (known) prize; (ii) the number of firms is fixed at the outset; (iii) the time needed to innovate successfully is uncertain; (iv) only the first firm to attain the innovation will be able to profit from it; and (v) firms that spend more are more likely to innovate first.  This model leads to some interesting conclusions, a fundamental one being that R&D efforts are “strategic complements”: the optimal strategy in response to an increase in innovative effort by a rival is to increase one’s own R&D.²⁹  The rate of R&D spending by each firm, and thus the speed at which the innovation is realized, depend positively on the number of participants.  As an obvious application, a merger between two participants in the patent race will decelerate the pace of innovation and so there might be some justification for challenging the merger.

But patent race models, and the conclusions reached from these models, are highly dependent on the underlying assumptions.  Consider, for example, the repercussions of changing the “winner take all” assumption.  Instead, make the assumption that an R&D success by one firm leads to a phase of product market competition against other participating firms.  This might arise if success by one firm provides important benefits to other R&D competitors, such that they are then able to complete their own projects and sell products in competition with the successful innovator.³⁰  Models in which research projects are assumed to be substitutes and in which there can be only one “winner” will therefore describe this structure very poorly.³¹  Here, it is theoretically possible for R&D effort to be negatively related to the number of competitors, so that a decrease in the number of firms involved in a race will tend to increase each remaining firm’s R&D expenditure.³²  A merger between R&D competitors would then have an ambiguous effect, as there would be fewer participants each performing more R&D.  This innocuous-seeming change in assumptions completely eliminates the “classic” result.  Careful consideration of market conditions is generally necessary before the patent race model can be used.

Empirical studies of the broader relationship between concentration and innovation have also failed to arrive at robust results.  Cohen and Levin’s comprehensive review of the early empirical literature finds little compelling evidence of a systematic relationship between concentration and innovation in cross-industry studies, with much of the variation in R&D intensity explained by industry-specific effects and technological opportunities.³³  A more promising model has recently been advanced that relates innovation to competition in an “inverted U” pattern: innovation increases with competition up to some critical point after which the relationship is reversed.³⁴  Empirical research seems to bear this out.  If industrial sectors are ranked according to a measure of competitive intensity based on profit margins, an increase in competition seems to promote innovation in less competitive sectors and stifle it in sectors that are already highly competitive.³⁵

The results of economic research on competition, firm size and innovation generally are sufficiently mixed that the Advisory Panel on Efficiencies advised against either a deliberate strategy of increasing competition, or a deliberate strategy of encouraging market concentration as a means by which Canada’s innovative capacity could be predictably and reliably improved upon.³⁶

3.2. Uncertainty

In the typical merger analysis, we can observe what products the merging parties are selling and where they are selling them.  Customers’ choices are also observable, as is the existence of rival firms.³⁷  In many cases, potential competitors can also be identified by observing which firms have the necessary production capacity to sell the relevant product in the relevant geographic market.

Matters are different where mergers among innovating firms are concerned.  In the case of product innovations, we are dealing with products that likely do not yet exist.  Two subsidiary issues flow from this. 

First, innovation is highly uncertain.  When firms set out to do R&D, they often do not know whether their investment will lead to a product that works from a technical perspective and that would be desired by consumers.³⁸  In many innovative industries, failure is more common than success for any particular inventive path.  As a result, firms will pursue numerous possibilities, particularly at the early stages of R&D in order to increase the probability that one of these will be successful.  This form of R&D is tremendously costly and exposes companies to a great deal of risk, since even if a product is approved for sale it may not ultimately be successful in the market. 

Second, innovation takes time.  In some cases there may be years between the R&D stage and the point at which a product can be brought to market.  During this intervening time period, innovating firms receive information about the technical characteristics of an innovation and its likely market prospects, and make decisions based on this new information as it arises.  As a result, a product may be quite different by the time the end of the innovation process is reached than was envisioned earlier on.  The time it takes to complete R&D may also make it difficult to identify other firms that would be positioned to undertake similar types of innovation.  Suppose a firm estimates that an R&D project is likely to take five years to complete, and that at the end of that time period the firm expects to have a successful product.  It is conceivable that two years into this time period, as a result of general technological progress or specialized knowledge, another firm may be able to complete a similar project within three years, such that both resulting products arrive on the market simultaneously.  The entry of this second firm would have been completely unanticipated at the outset.  Carlton and Gertner argue that since many innovations arrive from unexpected sources, such situations are relatively common, and as a result it is difficult to anticipate the firms that will offer competition in the future to firms that are innovating today.³⁹

3.3. Measurement Problems

Associated with uncertainty are measurement issues.  First, consider price in the context of innovation.  It would be desirable to know the price consumers would be willing to pay for an innovation, since this would be the first step toward being able to determine whether a proposed transaction would likely increase this price.  For innovations that have not yet been realized, price is unknown.  The value that consumers are expected to place on an innovative product, net of the value placed on other non-innovative aspects of the product, is unobservable prior to the product appearing on the market.  Even once a product is sold it may be difficult to disentangle the innovation itself from other aspects of the product contributing to its value.  Economists have estimated what are referred to as “hedonic” pricing models in an effort to measure the effect of individual product characteristics on value.  These empirical models allow for the measurement of quality-adjusted prices, facilitating a comparison across products that would be uninformative were only nominal prices to be used.  Where an innovation relates to one identifiable product characteristic, and where sufficient data exists to be able to isolate the effect of the innovation from everything else, it may be possible to estimate the price effectively paid by consumers for that innovation using these methods.⁴⁰  Data requirements normally make this very difficult.  In addition, except for perhaps minor incremental innovations, it will be risky to attempt to use a model based on existing data to predict the value placed by consumers on future improvements. 

This brings up a related complication.  Many products are made up of numerous components.  Each of these components may be a distinct product which the firm “bundles” with other components before making it available to consumers.  Consider as an example the personal computer, or PC.  The PC can be crudely described as a collection of components: microprocessor, memory, storage devices such as hard drives, interface devices, and so on.  Each one of these components is available on a standalone “unbundled” basis, the price of which can be observed on the market.  If an innovation improves one of these components, its value could (in theory) be inferred by observing the price of the component before and after the new technology was introduced.  Yet many innovations occur at much finer levels.  A microprocessor firm might develop a new means by which logical operations could be performed more efficiently on a chip.  This innovation might be incorporated into the next generation of chips sold on the market, along with numerous other improvements and established technologies.  In such cases, which are very common, the value of that one logic improvement cannot be isolated because it is not sold on its own, and many other factors contribute to chip performance and thus to the value placed on the product by consumers. 

Similar reservations apply to measurement of “quantity” in the context of innovation.⁴¹  In general, no completely adequate measures of quantity exist in the context of innovation, although the level of R&D expenditure, and indicators of innovation output such as patents, have sometimes been used. 
R&D spending is a useful indicator of a firm’s overall involvement in innovation at a point in time, but it is problematic to use to represent quantity on at least two counts.  First, it is a measure of input rather than output.  While greater R&D expenditure toward a particular end may represent a greater chance of eventual success, it is unknown precisely how this relationship works at the individual firm level.  If one firm spends twice as much on R&D as another in pursuit of an innovation, it is probably inappropriate to say that the first firm has twice the competitive significance as the second firm.  If we were to suggest the first firm is twice as big as the second based on this information we do not even know the extent to which we have over- or underestimated the firms’ relative position.  Second, being a measure of the flow of dollars, R&D spending may not provide a sufficiently precise indicator of innovative rivalry with respect to any one particular target.  R&D may be shared across numerous projects if a company’s scientists or equipment tend to be engaged in various different pursuits.  Economics does not help in determining how shared R&D should be allocated across various projects. 

Patents have been suggested as a potential alternative to R&D as a measure of innovation quantity.  In contrast to R&D spending, patents have the benefit that they are a unit of R&D “output” that is correlated with success, at least as measured along some technical dimension.  Moreover, firms often patent relatively early in the R&D process, long before they are ready to sell related products, so patenting might be observed early enough for the information to be useful in an analysis of a merger between innovating firms.  Finally, enough information is contained in patent documents to be able to associate each patent with a particular R&D program, at least in principle. 

Unfortunately, there are also serious problems with the use of patents as a measure of innovation quantity.  The main problem is that not all innovations are patented.  Patents appear to be much more useful as a means of protecting returns to innovation in some industries than in others.⁴²  In many industries, patents do not appear to be used for this purpose to any great degree.  Therefore, in industries like semiconductors, attempting to infer innovation quantity by observing the number of patents would be misleading.  Even in industries where firms actively use patents to protect their innovations, it is widely recognized that simply counting patents gives a relatively “noisy” indication of research output.⁴³  If patents are to be used, it is preferable to find some way of indexing them for their importance.  Weighting patents by the number of citations each has subsequently received from later patents has been used with some success.⁴⁴
We are therefore left with no single good measure of innovation quantity.  Instead, several imperfect indicators, such as R&D spending and patents, must typically be relied upon.

Previous | Table of Contents | Next

---

27 If firms are assumed to make strategic decisions concerning output, the Cournot model of oligopoly is commonly used.  If instead firms are assumed to compete on price, and in particular if products are differentiated, the Bertrand model is used.  Both of these models yield the sensible predictions that pricing depends on demand elasticity and marginal cost, and that prices tend to increase as industry concentration increases holding all else constant.  These types of static models have also performed well in empirical tests.  As a result, we have few reservations about applying standard economic models to evaluate short-term competitive effects. 

²⁸ This question is of obvious interest in the merger context.  For example, it was at the core of the Federal Trade Commission’s recent challenge to Genzyme’s acquisition of Novozyme, and the Commission’s subsequent decision to clear the merger.  In this action, the Federal Trade Commission (“FTC”) initially expressed a concern that since Genzyme and Novozyme were in 2001 the only firms conducting R&D toward a treatment for Pompe disease, the merger would potentially have had a negative impact on the pace of innovation.  The FTC ultimately concluded (over dissenting views from other Commissioners) that innovation efforts were not likely to be harmed by the transaction.  See “FTC Closes its Investigation of Genzyme Corporation’s 2001 Acquisition of Novozyme Pharmaceuticals, Inc.,” January 13, 2004, available at http://www.ftc.gov/opa/2004/01/genzyme.htm.

29 See Jean Tirole, The Theory of Industrial Organization, MIT Press, 1988, pp. 394-396.

30 In a sample of drug research projects in 10 pharmaceutical firms observed over a 17 year period, Cockburn and Henderson find a very low degree of correlation of R&D investment levels across firms, and in fact find evidence of a positive correlation in research output—success by one firm raises the likelihood of success by another.  See Iain Cockburn and Rebecca Henderson, “Racing to Invest?  The Dynamics of Competition in Ethical Drug Discovery,” Journal of Economics and Management Strategy, Vol. 3, 1994.

31 Cockburn and Henderson (1994) conclude (p. 484): “Our finding that the modern game theoretic literature is of only limited usefulness as an empirical guide points to the need not only to model R&D as a race with multiple prizes, but also to develop theories that incorporate richer models of adjustment costs and firm heterogeneity and to collect appropriately comprehensive and detailed data.”

32 Jennifer F. Reinganum, “The Timing of Innovation: Research, Development, and Diffusion,” Chapter 14 in R. Schmalensee and R.D. Willig (eds.), Handbook of Industrial Organization, Volume I, Elsevier, pp. 865-866.  Note also that in many cases no clear analytical result is available.

33 Wesley M. Cohen and Richard C. Levin, “Empirical Studies of Innovation and Market Structure,” Chapter 18 in R. Schmalensee and R.D. Willig (eds.), Handbook of Industrial Organization, Volume II, Elsevier, pp. 1077-1078.

34 Philippe Aghion, Nick Bloom, Richard Blundell, Rachel Griffith, and Peter Howitt, “Competition and Innovation: An Inverted-U Relationship,” Quarterly Journal of Economics, Vol. 120, 2005.

35 Aghion et al. (2005), pp. 705-710.

36 Report of the Advisory Panel on Efficiencies, August 2005, p. 43.

37 Of course there may be issues around whether a particular rival not selling the product in a particular location today is able to do so post-merger, or whether a particular rival selling another product would be able to sell the relevant product post-merger.

38 The pharmaceutical industry provides what is probably the starkest example of this phenomenon.  Drug companies must screen an estimated 5,000 chemical compounds for each product approved for commercial sale.  See PhRMA (Pharmaceutical Research and Manufacturers of America), “Why do Prescription Drugs Cost so Much and Other Questions About Your Medicines,” June 2000, p. 2.

39 Dennis W. Carlton and Robert H. Gertner, “Intellectual Property, Antitrust, and Strategic Behavior,” in A.B. Jaffe, J. Lerner, and S. Stern (eds.), Innovation Policy and the Economy, MIT Press, 2003, pp. 42-43.

40 For example Berndt et al. have estimated a hedonic model for antiulcer drugs sold in the U.S. and have found that prices are strongly dependent on factors such as convenient dosing regimens, the absence of significant negative drug interactions, and the availability of approved treatment for different sets of symptoms (Ernst R. Berndt, Robert S. Pindyck, and Pierre Azoulay, “Consumption Externalities and Diffusion in Pharmaceutical Markets: Antiulcer Drugs,” Journal of Industrial Economics, Vol. LI, 2003).  Therefore, their model would, for example, allow one to estimate the price premium consumers would be willing to pay over and above existing therapies for a new symptom to be treated.  There are also numerous examples of economists attempting to adjust prices within the computer industry, controlling for changes in quality using hedonic pricing models.  See, for example, Kenneth H. Brown, “Hedonic price indexes and the distribution of buyers across the product space: an application to mainframe computers,” Applied Economics, Vol. 32, 2000; and Ariel Pakes, “A Reconsideration of Hedonic Price Indexes with an Application to PCs,” American Economic Review, Vol. 93, 2003.

41 By “quantity” we do not necessarily mean the number of individual innovations made by each possible firm, since this would tend to implicate numerous different “markets”—instead we use the term as shorthand for an indicator of competitive significance in innovation, as output quantity is used in conventional merger analysis. 

42 In a classic survey, R&D managers in most manufacturing industries reported that patents and other formal IP rights were only one of several means for earning returns on R&D (Richard C. Levin, Alvin K. Klevorick, Richard B. Nelson and Sidney G. Winter, “Appropriating the Returns from Industrial Research and Development,” Brookings Papers on Economic Activity, Vol. 3, 1987).  Respondents to a survey of 650 R&D managers ranked lead time, moving down the learning curve, and complementary sales and service efforts as more important than patents for new product inventions.  For new processes, the picture was even more striking, with patents ranked last out of the various means for earning returns, behind lead time, learning curve advantages, secrecy, and sales and service efforts.  The results of an updated survey of 1,478 R&D laboratories confirmed many of the same findings (Wesley M. Cohen, Richard R. Nelson and John P. Walsh, “Protecting Their Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not),” National Bureau of Economic Research Working Paper 7552, 2000).  However, both surveys show substantial variation across industries in the ways in which firms seek returns on R&D expenditures, with secrecy and lead time generally important for product innovations, and secrecy alone the most important mechanism for process innovations.

43 Griliches notes that while patents are strongly associated with R&D in cross-sectional data, the relationship between patents as measures of inventive output and R&D in within-firm time-series data is much weaker.  He states: “Because the bulk of R&D expenditures are spent on development, most of the time-series variance in this variable must come from the differential success in the further development of existing projects rather than from the initiation of new ones.  The relatively low correlations in the time dimension should, therefore, not be all that surprising, but they imply that patent numbers are a much poorer indicator of short-term changes in the output of inventive activity or the ‘fecundity’ of R&D.”  See Zvi Griliches, “Patent Statistics as Economic Indicators: A Survey,” Journal of Economic Literature, Vol. XXVIII, 1990, p. 1574.

44 See e.g., Dietmar Harhoff, Francis Narin, F.M. Scherer, and Katrin Vopel, “Citation Frequency and the Value of Patented Inventions,” Review of Economics and Statistics, Vol. 81, 1999; Bronwyn H. Hall, Adam Jaffe, and Manuel Trajtenberg, “Market Value and Patent Citations,” RAND Journal of Economics, Vol. 36, 2005.