Ufuk Akcigit University of Pennsylvania

August 18, 2011

Abstract To what extent and in what form should the intellectual property rights (IPR) of innovators be protected? Should a company with a large technology lead over its rivals receive the same IPR protection as a company with a more limited advantage? The analysis of these questions necessitates a dynamic framework for the study of the interactions between IPR and competition, in particular to understand the impact of such policies on future incentives. In this paper, we develop such a framework. The economy consists of many industries and …rms engaged in cumulative (step-by-step) innovation. IPR policy regulates whether followers in an industry can copy (or license or build upon) the technology of the leader. With full patent protection, followers can catch up to the leader in their industry only by making the same innovation(s) themselves (or by full licensing). We prove the existence of a steadystate equilibrium in a baseline environment and characterize some of its properties. We then quantitatively investigate the implications of di¤erent types of IPR policy on the equilibrium growth rate and welfare. The most important result from this exercise is that full patent protection is not optimal (welfare maximizing); instead, optimal policy involves state-dependent IPR protection, providing greater protection to technology leaders that are further ahead than those that are close to their followers. This form of the optimal policy results from the impact of policy on dynamic incentives, in particular from a form of “trickle-down” e¤ect: providing greater protection to …rms that are further ahead of their followers than a certain threshold increases the R&D incentives also for all technology leaders that are less advanced than this threshold. Keywords: competition, economic growth, endogenous growth, industry structure, innovation, intellectual property rights, licensing, patents, research and development, trickle-down. JEL classi…cation: O31, O34, O41, L16.

We thank an anonymous referee, conference and seminar participants at the FBBVA Lecture at 2011 ASSA conference in Denver, the Canadian Institute for Advanced Research, EPFL Technology Policy Conference, European Science Days, MIT, National Bureau of Economic Research Economic Growth and Productivity Groups, Toulouse Information Technology Network, University of Pennsylvania, University of Toronto, and Philippe Aghion, Gino Gancia, Bronwyn Hall, Sam Kortum, Suzanne Scotchmer and Fabrizio Zilibotti for useful comments. Financial support from the Toulouse Information Technology Network and from the Canadian Institute for Advanced Research is gratefully acknowledged. An earlier version of this paper was circulated under the title “State-Dependent Intellectual Property Rights Policy”.

1

Introduction

What is the optimal extent and form of intellectual property rights (IPR) protection? Should a …rm with a large technology lead receive the same IPR protection as a company with a more limited technological lead, or should IPR policy be coupled with antitrust and used to limit the monopoly power of technology leaders? Despite broad consensus that innovation is central to the long-run performance of an economy, there is no consensus on the answers to such questions. A large literature on IPR (discussed below) focuses on the static trade-o¤s between the positive incentive bene…ts of IPR protection and its costs in terms of reducing competition and increasing markups. In this paper, we argue that dynamic trade-o¤s between IPR protection and competition, which have so far been overlooked, may be equally or more important for developing answers to these questions. These issues and the importance of these questions are highlighted by several recent highpro…le cases. For example, motivated by antitrust concerns, a recent ruling of the European Commission ordered Microsoft to share secret information about its operating system and products with other software companies (New York Times, December 22, 2004).1 Similar issues were also central to the US Department of Justice (DOJ) case against Microsoft, which started on May 18th, 1998 and ultimately resulted in a ruling against Microsoft. Figure 1 shows the evolution of R&D by Microsoft and by other top 10 publicly traded R&D investors in the IT sector relative to the sector average before and after the start of the DOJ case.2 [Figure 1 here] The relative R&D spending by Microsoft and other industry leaders, which had been steadily— perhaps even exponentially— increasing since the mid-80s, appear to decline after the DOJ action. While one might expect R&D by Microsoft to slow down for a variety of reasons, it is not obvious why there should be a relative decline in the R&D of other top companies, since they partly bene…ted from the weakening of, and the restrictions imposed on Microsoft. This relative decline may have been caused by a combination of a slowdown in the R&D activities of these other companies or an increase in the R&D investments of smaller IT …rms in response to the DOJ action, or by entirely di¤erent and unrelated factors. A sys1

In addition to the Microsoft case, the issue of technological lead has been central in the Department of Justice investigations of Intel (New York Times, May 11, 2009) and the debates about Google’s market share (New York Times, February 21, 2009). 2 All data are from COMPUSTAT. Top 10 …rms is determined by the highest 10 R&D investors (except Microsoft) in 1998. The patterns shown in Figures 1 is very similar if we use the top 10 investors in 2000 or 1990, or if we benchmark it to the median of the industry rather than the mean. Top 10 investors in Figure 1 are: CA Inc, Continuum Inc, Intergraph Corp, Sterling Software Inc, Oracle Inc, Adobe Inc, Symantec Corp, Electronic Arts Inc, Sybase Inc, Intuit Inc.

1

tematic investigation of these issues requires a dynamic equilibrium framework where R&D activities of di¤erent types respond to changes in IPR and competition policy. In this paper, we take a step in this direction. Our framework builds on and extends the step-by-step innovation models of Aghion, Harris and Vickers (1997) and Aghion, Harris, Howitt and Vickers (2001), where a number of (typically two) …rms engage in price competition within an industry and undertake R&D in order to improve their production technology. The technology gap between the …rms determines the extent of the monopoly power of the leader, and hence the price markups and pro…ts. The purpose of R&D by the follower is to catch up and surpass the leader (as in standard Schumpeterian models of innovation, e.g., Reinganum, 1981, 1985, Aghion and Howitt, 1992, Grossman and Helpman, 1991), while the purpose of R&D by the leader is to escape the competition of the follower and increase its markup and pro…ts. Despite the dynamic nature of these models, their policy implications are still mostly based on the same static trade-o¤ mentioned above. For this reason, for example, Aghion, Harris, Howitt and Vickers (2001, p. 481) conjecture that IPR protection should be limited and particularly so for …rms with larger technological leads over their rivals (which face less competition and thus have greater monopoly power). We extend these existing models in several directions. Most importantly, we explicitly introduce state-dependent patent/IPR protection policy, meaning a policy that makes the extent of patent or intellectual property rights protection conditional on the technology gap between di¤erent …rms in the industry. As in racing-type models in general (e.g., Harris and Vickers, 1985, 1987, Budd, Harris and Vickers, 1993), a large gap between the leader and the follower discourages R&D by both. Consequently, overall R&D and technological progress are greater when the technology gap between the leader and the follower is relatively small.3 One may then expect that full patent protection may be suboptimal in a world of step-bystep competition and permitting followers to copy or use the leaders’ technologies would be particularly bene…cial in industries where there is a large technology gap between leaders and followers.4 However, crucially, this reasoning ignores the dynamic incentive e¤ects, which are our main focus in this paper and emerge more clearly when IPR policy is explicitly statedependent. Our analysis establishes that the opposite of the above conjecture is always true in such a dynamic equilibrium framework: optimal IPR policy should provide greater protection to 3

Aghion, Bloom, Blundell, Gri¢ th and Howitt (2005) provide empirical evidence from British industries consistent with the view R&D increases when there is a smaller technological gap between …rms. See also Aghion and Gri¢ th (2007). 4 This is indeed the basis of Aghion, Harris, Howitt and Vickers’conjecture mentioned above.

2

technologically more advanced leaders. Underlying this result is what we refer to as the trickledown of incentives: providing relatively low protection to …rms with limited leads and greater protection to those that have greater leads not only improves the incentives of …rms that are technologically advanced, but also encourages R&D by those that have limited leads because of the prospect of reaching levels of technology gaps associated with greater protection. A corollary of this result is that full IPR protection is not optimal, and there should be limited, but state-dependent, IPR protection for …rms with only limited technology leads over their rivals. More speci…cally, we show that in contrast to the standard disincentive e¤ects of uniform relaxation of IPR policy, state-dependent relaxation that provides greater protection to technologically more advanced …rms creates a positive incentive e¤ ect. This is because when a particular state for the technology leader (say being n steps ahead of the follower) becomes more pro…table, this increases the incentives to perform R&D not only for leaders that are n

1 steps ahead, but for all leaders with a lead of size n

n

1. It is this trickle-down

e¤ect that generates the positive incentive e¤ect and makes state-dependent IPR, with greater protection for …rms that are technologically more advanced than their rivals, preferable to uniform IPR. We start with a partial equilibrium model and provide an explicit characterization of the trickle-down e¤ect. We then develop a richer dynamic general equilibrium framework which allows a systematic analysis of how innovation depends on R&D by technology leaders and followers. Our baseline model focuses on quick catch-up, meaning that a follower can catch up with the technology leader with a single innovation regardless of the size of the gap between them. For this environment, we establish the existence of a stationary equilibrium and characterize some of its properties. We then study the form of optimal (welfare maximizing) IPR and competition policy quantitatively. The same e¤ects as in the partial equilibrium analysis make state-dependent relaxation of IPR optimal. Quantitatively, we …nd that optimal statedependent IPR policy can increase the growth rate of the economy from 1.86% to 2.04%, and does so with fewer workers employed in the R&D sector (because R&D workers are reallocated towards …rms where their e¤orts directly lead to productivity growth). In contrast, uniform relaxation of IPR policy reduces both welfare and growth. These patterns are quite robust to di¤erent parameter values. We next show how the framework can be extended to study these issues when there is slow catch-up, meaning that followers close the gap between themselves and technology leaders only gradually. The presence of slow catch-up also enables us to introduce di¤erent types of R&D e¤orts and di¤erent dimensions of IPR policy, in particular, licensing and patent infringement 3

fees.5 We show that the trickle-down e¤ect and the result that optimal IPR policy should be state-dependent and provide greater protection to technologically more advanced …rms are robust in these alternative environments. In most cases, optimal IPR policy also increases growth by a similar magnitude to our baseline model (though in some cases, it increases welfare but not necessarily growth).6 Our paper is a contribution both to the IPR protection and the endogenous growth literatures. Previous work has focused on the static trade-o¤ between ex-post monopoly rents and ex-ante R&D incentives (e.g., Arrow, 1962, Reinganum, 1981, Tirole, 1988, Romer, 1990, Grossman and Helpman, 1991, Aghion and Howitt, 1992, Green and Scotchmer, 1995, Scotchmer, 1999, Gallini and Scotchmer, 2002, O’Donoghue and Zweimuller, 2004).7 Much of the literature discusses the trade-o¤ between these two forces to determine the optimal length and breadth of patents. For example, Klemperer (1990) and Gilbert and Shapiro (1990) show that optimal patents should have a long duration in order to provide inducement to R&D, but a narrow breadth so as to limit monopoly distortions. A number of other papers, for example, Gallini (1992) and Gallini and Scotchmer (2002), reach opposite conclusions. Another branch of the literature, including the seminal paper by Scotchmer (1999) and the recent interesting papers by Llobet, Hopenhayn and Mitchell (2006) and Hopenhayn and Mitchell (2001, 2011), adopts a mechanism design approach to the determination of the optimal patent and intellectual property rights protection system. For example, Scotchmer (1999) derives the patent renewal system as an optimal mechanism in an environment where the cost and value of di¤erent projects are unobserved and the main problem is to decide which projects should go ahead. Llobet, Hopenhayn and Mitchell (2006) consider optimal patent policy in the context of a model of sequential innovation with heterogeneous quality and private information. They show that allowing for a choice from a menu of patents will be optimal in this context. Hopenhayn and Mitchell (2011) build on an earlier version of our paper, Acemoglu and Akcigit (2006), and derive a form of trickle-down e¤ect using a mechanism design approach in a model with recurring innovations. Our paper builds on and extends Aghion, Harris and Vickers (1997) and Aghion, Harris, 5

In particular, in this regime, we allow …rms to undertake frontier as well as catch-up R&D. With frontier R&D, they can build on the technology leader’s knowledge base, and if successful, they immediately overtake the leader, but might be liable for a patent infringement fee. We also allow followers to license the innovation of the technology leader by paying a prespeci…ed license fee— i.e., a “compulsory licensing” where the license fee is determined by IPR policy. We also show that voluntary licensing agreements would not achieve the same results, so our analysis establishes a potential need for compulsory licensing policy. Previous work emphasizing importance of compulsory licensing includes Tandon (1982), Gilbert and Shapiro (1990), and Kremer (2002). 6 We also show that both licensing and the possibility of frontier R&D (subject to infringement fees) contributes to growth and welfare. 7 Boldrin and Levine (2004, 2008) or Quah (2003) argue that patent systems are not necessary for innovation.

4

Howitt and Vickers (2001).8 Although our model builds on these papers, it also di¤ers from them in a number of signi…cant ways. First, we introduce state-dependent IPR policy. This is crucial for most of the results in the paper, including the trickle-down of incentives and the form of optimal IPR. Second, we also introduce and analyze the slow catch-up regime, and in this context, we allow for compulsory licensing and for leapfrogging, which makes the followers directly contribute to the economic growth. We provide a full quantitative analysis of statedependent IPR policy under these di¤erent scenarios. Third, our economy is a full general equilibrium model with competition between production and R&D for scarce labor.9 Finally, we provide a general existence result and a number of analytical results for the general model (with or without IPR policy), while previous literature has focused on the special cases where innovations are either “drastic”(so that the leader never undertakes R&D) or very small, and has not provided existence or general characterization results for steady-state equilibria. Lastly, our results are also related to the literature on tournaments and races, for example, Fudenberg, Gilbert, Stiglitz and Tirole (1983), Harris and Vickers (1985, 1987), Choi (1991), Budd, Harris and Vickers (1993), Taylor (1995), Fullerton and McAfee (1999), Baye and Hoppe (2003), and Moscarini and Squintani (2004). This literature considers the impact of endogenous or exogenous prizes on e¤ort in tournaments, races or R&D contests. In terms of this literature, state-dependent IPR policy can be thought of as “state-dependent handicapping” of di¤erent players (where the state variable is the gap between the two players in a dynamic tournament). To the best of our knowledge, these types of schemes have not been considered in this literature. The rest of the paper is organized as follows. Section 2 introduces the partial equilibrium model and analytically demonstrates the trickle-down e¤ect. Section 3 presents our baseline environment (where a successful innovation by followers closes the entire gap with technology leaders in one step, i.e., there is quick catch-up). Section 4 proves the existence of a steady-state equilibrium and characterizes some of its key properties under both uniform and state-dependent IPR policy. Section 5 de…nes the social welfare objective and outlines our quantitative methods. Section 6 characterizes the structure of optimal IPR policy quantitatively. Section 7 extends the model to allow for slow catch-up, compulsory license fees and 8

Two other papers are also related, Segal and Whinston (2007) who analyze the impact of anti-trust policy on economic growth in a related model of step-by-step innovation, and Acemoglu, Gancia and Zilibotti (2012) who analyze the role of IPR policy in a model with innovation and standardization based on imitation of new products. 9 This general equilibrium aspect is introduced to be able to close the model economy without unrealistic assumptions and makes our economy more comparable to other growth models (Aghion, Harris, Howit and Vickers, 2001, assume a perfectly elastic supply of labor). We show that the presence of general equilibrium interactions does not signi…cantly complicate the analysis and it is still possible to characterize the steady-state equilibrium. The endogenous allocation of labor between di¤erent …rms and between production and R&D also enables us to show that optimal IPR policy can increase the growth rate of the economy while also reducing the fraction of the workforce employed in R&D (see Section 6 for details).

5

leapfrogging, and quantitatively characterizes the structure of optimal IPR policy under different combinations of these policies. Section 8 concludes. The Appendix, which contains additional results and the proofs of all the results stated in the text, is available online.10

2

A Partial Equilibrium Illustration

We …rst illustrate the main economic force in this paper, the trickle-down e¤ect, using a partial equilibrium model. Consider the following in…nite horizon, step-by-step R&D race between two competing …rms in continuous time. Each …rm maximizes the expected net present discounted value of “net pro…ts,” de…ned as operating pro…t minus R&D cost, Z 1 exp ( r (s t)) [ i (s) Et i (s)] ds; t

where Et denotes expectation at time t, r > 0 is the interest rate, operating pro…t ‡ow and

i (t)

i (t)

is the instantaneous

represents the R&D cost of …rm i at time t. In this game, …rm

i 2 f1; 2g invests in R&D to advance its position relative to its rival i0 6= i. Suppose that the positions of both …rms in this race can be characterized by integer values on the real line, and denote the distance of …rm i from its rival at time t by ni (t). In the partial equilibrium model, we simplify the analysis by following Aghion, Harris, Howitt and Vickers (2001) and Aghion, Bloom, Blundell, Gri¢ th and Howitt (2005) in assuming that the maximum technology gap between a leader and a follower is 2; this assumption is relaxed in the full general equilibrium model analyzed in the rest of the paper. For now it simpli…es the analysis by ensuring that the relative position of …rm i can take …ve possible values, ni (t) 2 NI denote the absolute gap between the two …rms by n (t)

f 2; 1; 0; 1; 2g. Let us

max fni (t) ; n

i (t)g,

and suppress

the time subscripts to simplify notation. The payo¤s in this game are assumed to be stationary and only a function of the relative distance between the …rms, thus represented by In particular,

ni

: NI ! R+ (see equation (20) in Section 3).

0 is the instantaneous payo¤ that …rm i obtains when its distance from its

competitor is ni at time t and is assumed to be a strictly increasing function of ni . To advance its relative position, …rm i invests in R&D, which determines the Poisson rate of arrival of innovation, xi 2 R+ . Let us also assume that the cost of R&D is linear in the arrival rate of innovation, i.e.,

(xi ) = xi ; with

> 0 (again see below for more general formulations).

Each successful innovation is patented and advances …rm i’s state (relative position) by one step, so that following a successful innovation by …rm i at time t we have: ni (t+) = ni (t) + 1 (where ni (t+) stands for ni immediately following time t). 10

See http://econ-www.mit.edu/faculty/acemoglu/paper or http://www.sas.upenn.edu/~uakcigit/web/Research.html.

6

IPR policy governs the expected length of a patent. For simplicity, we model patent length by assuming that it terminates at a Poisson rate. Crucially for our focus, IPR policy is state dependent, and we represent it by the function:

: NI ! R+ : Here

(n)

n

< 1 is the ‡ow

rate at which the patent terminates (patent protection is removed) for a technology leader that is n steps ahead. When

n

= 0, there is full protection at technology gap n, in the sense that

patent protection will never be removed. In contrast,

n

! 1 implies that patent protection

is removed immediately once technology gap n is reached. When the patent protection is removed, the …rm that is behind copies the technology of its competitor and both …rms end up neck-and-neck, i.e., n = 0. Finally, we take the interest rate r as exogenous and assume that it satis…es r < (

n n

n 1 ) =4

for each n 2 NI . This assumption ensures positive R&D by each …rm when

= 0. Throughout we will focus on (stationary) Markov Perfect Equilibria (MPE), where

strategies (R&D decisions) are only functions of the payo¤-relevant state, which is n 2 NI . A more formal de…nition of the MPE in the general equilibrium environment is given below. The MPE can be characterized by writing the value functions of each …rm as a function of the state n 2 NI . These value functions are given by the following recursions: rv2 =

2

+x

2 [v1

v2 ] +

2 [v0

v2 ] ;

(1)

rv1 = max f

1

x1 + x1 [v2

v1 ] + x

rv0 = max f

0

x0 + x0 [v1

v0 ] + x ~0 [v

x1 0

x0 0

rv rv

1 2

= =

max f

1

x

1

+x

1 [v0

max f

2

x

2

+x

2 [v 1

x x

1

2

0 0

v

1 [v0

1]

v

1

v1 ] +

+

v1 ]g ;

(2)

v0 ]g ;

+ x1 [v

2]

1 [v0

(3)

2

v

1]

+

2 [v0

v

2 ]g :

1 [v0

v

1 ]g ;

(4) (5)

In all equations, the …rst term represents current pro…ts. In equations (2)-(5), the second term substracts R&D costs from current pro…ts, the third term represents the fact that the …rm will successfully innovate at the ‡ow rate xn and increase its position by one step. The fourth term incorporates the change in value due to an innovation by the rival …rm. In equations (1) and (2) the last term is the change in value for the leader due to patent expiration, which takes place at the rate

n.

In (4) and (5), the last term is the change in value for the follower.

Finally, equation (3) has the same interpretation except that now n = 0 and the two …rms are neck-and-neck and thus there is no IPR policy (and the ‡ow rate of innovation of the other …rm is denoted by x ~0 , and naturally, in a symmetric equilibrium, we will have x0 = x ~0 ). Note also that in equations (1) and (5), we used the fact that a two-step ahead …rm does not undertake any R&D since it has already achieved the maximum feasible lead. We now characterize the MPE under two di¤erent policy environments: uniform and state7

dependent IPR policy. Uniform IPR policy corresponds to the case where

Uniform IPR Policy.

n

=

0 and/or

> 0, which depends on the social returns from xn ’s. For example, if x1 is socially more

bene…cial than x

1,

1

> 0 will always be preferred. We will next see that this is always the

case in our general equilibrium model.

8

3

General Equilibrium Framework

We now describe our baseline dynamic general equilibrium model. Our baseline model assumes quick catch-up, meaning that one innovation by a follower is su¢ cient to close the gap with the technology leader in the industry. The characterization of the equilibrium in this environment under the di¤erent policy regimes is presented in the next section. Alternative assumptions on the form of catch-up are investigated in Section 7.

3.1

Preferences and Technology

Consider the following continuous time economy with a unique …nal good. The economy is populated by a continuum of 1 individuals, each with 1 unit of labor endowment, which they supply inelastically. Preferences at time t are given by Z 1 Et exp ( (s t)) log C (s) ds;

(6)

t

where Et denotes expectations at time t,

> 0 is the discount rate and C (t) is consumption

at date t. The logarithmic preferences in (6) facilitate the analysis, since they imply a simple relationship between the interest rate, growth rate and the discount rate (see (7) below). Let Y (t) be the total production of the …nal good at time t. We assume that the economy is closed and the …nal good is used only for consumption (i.e., there is no investment), so that C (t) = Y (t). The standard Euler equation from (6) then implies that g (t)

C_ (t) Y_ (t) = = r (t) C (t) Y (t)

;

(7)

where this equation de…nes g (t) as the growth rate of consumption and thus output, and r (t) is the interest rate at date t. The …nal good Y is produced using a continuum 1 of intermediate goods according to the Cobb-Douglas production function ln Y (t) =

Z

1

ln y (j; t) dj;

(8)

0

where y (j; t) is the output of jth intermediate at time t. Throughout, we take the price of the …nal good as the numeraire and denote the price of intermediate j at time t by p (j; t). We also assume that there is free entry into the …nal good production sector. These assumptions, together with the Cobb-Douglas production function (8), imply that the …nal good sector has the following demand for intermediates y (j; t) =

Y (t) ; p (j; t) 9

8j 2 [0; 1] :

(9)

Intermediate j 2 [0; 1] comes in two di¤erent varieties, each produced by one of two in…nitely-lived …rms. We assume that these two varieties are perfect substitutes and these …rms compete a la Bertrand.11 Firm i = 1 or 2 in industry j has the following technology y (j; t) = qi (j; t) li (j; t) ;

(10)

where li (j; t) is the employment level of the …rm and qi (j; t) is its level of technology at time t. Each consumer in the economy holds a balanced portfolio of the shares of all …rms. Consequently, the objective function of each …rm is to maximize expected pro…ts. The production function for intermediate goods, (10), implies that the marginal cost of producing intermediate j for …rm i at time t is M Ci (j; t) =

w (t) ; qi (j; t)

(11)

where w (t) is the wage rate in the economy at time t. When this causes no confusion, we denote the technology leader in each industry by i and the follower by

i, so that we have: qi (j; t)

q

i (j; t) :

Bertrand competition between the two …rms implies that all intermediates will be supplied by the leader at the “limit” price:12 pi (j; t) =

q

w (t) : i (j; t)

(12)

Equation (9) then implies the following demand for intermediates: y (j; t) =

q

i (j; t)

w (t)

Y (t) :

(13)

11

A more general case would involve these two varieties being imperfect substitutes, for example, with the output of intermediate j produced as h i 1 1 1 y (j; t) = 'y1 (j; t) + (1 ') y2 (j; t) ;

with > 1. The model analyzed in the text corresponds to the limiting case where ! 1. Our results can be easily extended to this more general case with any > 1, but at the cost of additional notation. We therefore prefer to focus on the case where the two varieties are perfect substitutes. It is nonetheless useful to bear this formulation with imperfect substitutes in mind, since it facilitates the interpretation of “distinct” innovations by the two …rms (when the follower engages in “catch-up” R&D). 12 If the leader were to charge a higher price, then the market would be captured by the follower earning positive pro…ts. A lower price can always be increased while making sure that all …nal good producers still prefer the intermediate supplied by the leader i rather than that by the follower i, even if the latter were supplied at marginal cost. Since the monopoly price with the unit elastic demand curve is in…nite, the leader always gains by increasing its price, making the price given in (12) the unique equilibrium price.

10

3.2

Technology, R&D and IPR Policy under Quick Catch-up

R&D by the leader or the follower stochastically leads to innovation. We assume that when the leader innovates, its technology improves by a factor

> 1.

The follower, on the other hand, can undertake R&D to catch up with the frontier technology. We will call this type of R&D as catch-up R&D.13 Catch-up R&D can be thought of R&D to discover an alternative way of performing the same task as the current leading-edge technology. Because this innovation applies to the follower’s variant of the product (recall footnote 11) and results from its own R&D e¤orts, we assume in our baseline framework that it does not constitute infringement on the patent of the leader. R&D by the leader and follower may have di¤erent costs and success probabilities. We simplify the analysis by assuming that both types of R&D have the same costs and the same probability of success. In particular, in all cases, we assume that innovations follow a controlled Poisson process, with the arrival rate determined by R&D investments. Each …rm (in every industry) has access to the following R&D technology: xi (j; t) = F (hi (j; t)) ;

(14)

where xi (j; t) is the ‡ow rate of innovation at time t and hi (j; t) is the number of workers hired by …rm i in industry j to work in the R&D process at t. This speci…cation implies that within a time interval of

t, the probability of innovation for this …rm is xi (j; t) t + o ( t).

We assume that F is twice continuously di¤erentiable and satis…es F 0 ( ) > 0; F 00 ( ) < 0, F 0 (0) < 1 and that there exists h 2 (0; 1) such that F 0 (h) = 0 for all h

h. The

assumption that F 0 (0) < 1 implies that there is no Inada condition when hi (j; t) = 0. The last assumption, on the other hand, ensures that there is an upper bound on the ‡ow rate of innovation (which is not essential but simpli…es the proofs). Recalling that the wage rate for labor is w (t), the cost for R&D is therefore w (t) G (xi (j; t)) where G (xi (j; t))

F

1

(xi (j; t)) ;

(15)

and the assumptions on F immediately imply that G is twice continuously di¤erentiable and satis…es G0 ( ) > 0; G00 ( ) > 0, G0 (0) > 0 and limx!x G0 (x) = 1, where x

F h

(16)

is the maximal ‡ow rate of innovation (with h de…ned above). 13 This contrasts with frontier R&D introduced in Section 7, which will allow the follower to leapfrog the leader.

11

We next describe the evolution of technologies within each industry. Suppose that leader i in industry j at time t has a technology level of nij (t)

qi (j; t) = and that the follower

where nij (t)

n

ij

;

(17)

i’s technology at time t is

(t) and nij (t), n

ij (t)

i (j; t)

ij

(t) 2 Z+ denote the technology rungs of the leader and

the follower in industry j. We refer to nj (t)

=

n

q

nij (t) n

;

(18)

ij

(t) as the technology gap in industry

j. If the leader undertakes an innovation within a time interval of increases to qi (j; t +

t) =

nijt +1

and the technology gap rises to nj (t +

(the probability of two or more innovations within the interval represents terms that satisfy lim

t, then its technology

t!0 o (

t) = nj (t) + 1

t will be o ( t), where o ( t)

t) = t).

In our baseline model, we assume that there is quick catch-up between followers and leaders. Namely, when the follower is successful in catch-up R&D within the interval

t, then its

technology improves to q

i (j; t

+

t) =

nijt

;

and thus it catches up with the leader immediately (regardless of how large the technology gap was). In this case, the technology gap variable becomes njt+

t

= 0.

In addition to catching up with the technology frontier with their own R&D, followers can also copy the technology frontier if and when patents expire. In particular, we assume that patents expire at some policy-determined Poisson rate , and after expiration, followers can costlessly copy the frontier technology, jumping to q

i (j; t

+

t) =

nijt 14 .

As in the partial

equilibrium model in Section 2, IPR policy governs the length of the patent and we allow it to be state dependent, so it is represented by the following function: : N ! R+ Here

(n)

n

< 1 is the ‡ow rate at which the patent protection is removed from a

technology leader that is n steps ahead of the follower. When

n

= 0, this implies that there is

full protection at technology gap n, in the sense that patent protection will never be removed. In contrast,

n

! 1 implies that patent protection is removed immediately once technology

gap n is reached. Our formulation imposes that 14

f 1;

2 ; :::g

is time-invariant. Given this

Alternative modeling assumptions on IPR policy, such as a …xed patent length of T > 0 from the time of innovation, are not tractable, since they lead to value functions that take the form of delayed di¤erential equations.

12

speci…cation, we can now write the law of motion of the technology gap in industry j as follows: 8 xi (j; t) t + o ( t) > > nj (t) + 1 with probability > > > > < x i (j; t) + nj (t) t + o ( t) 0 with probability nj (t + t) = : > > > > > > : nj (t) with probability 1 xi (j; t) + x i (j; t) + t o ( t)) nj (t)

(19)

Here o ( t) again represents second-order terms, in particular, the probabilities of more than one innovations within an interval of length

t. The terms xi (j; t) and x

rates of innovation by the leader and the follower; and

nj (t)

i (j; t)

are the ‡ow

is the ‡ow rate at which the

follower is allowed to copy the technology of a leader that is nj (t) steps ahead. Intuitively, the technology gap in industry j increases from nj (t) to nj (t) + 1 if the leader is successful. The …rms become “neck-and-neck” when the follower comes up with an alternative technology to that of the leader (‡ow rate x

3.3

i (j; t))

or the patent expires at the ‡ow rate

nj :

Pro…ts

We next write the instantaneous “operating” pro…ts for the leader (i.e., the pro…ts exclusive of R&D expenditures). Pro…ts of leader i in industry j at time t are i (j; t)

= [pi (j; t) M Ci (j; t)] yi (j; t) w (t) Y (t) w (t) = q i (j; t) qi (j; t) pi (j; t) =

where nj (t)

nij (t) n

ij

1

nj (t)

Y (t)

(20)

(t) is the technology gap in industry j at time t. The …rst line simply

uses the de…nition of operating pro…ts as price minus marginal cost times quantity sold. The second line uses the fact that the equilibrium limit price of …rm i is pi (j; t) = w (t) =q as given by (12), and the …nal equality uses the de…nitions of qi (j; t) and q

i (j; t)

i (j; t)

from (17)

and (18). The expression in (20) also implies that there will be zero pro…ts in neck-and-neck industries, i.e., in those with nj (t) = 0. Also clearly, followers always make zero pro…ts, since they have no sales. The Cobb-Douglas aggregate production function in (8) is responsible for the form of the pro…ts (20), since it implies that pro…ts only depend on the technology gap of the industry and aggregate output. This will simplify the analysis below by making the technology gap in each industry the only industry-speci…c payo¤-relevant state variable. The objective function of each …rm is to maximize the net present discounted value of “net

13

pro…ts” (operating pro…ts minus R&D expenditures). In doing this, each …rm will take the sequence of interest rates, [r (t)]t sequence of wages, [w (t)]t

3.4

0,

0,

the sequence of aggregate output levels, [Y (t)]t

0,

the

the R&D decisions of all other …rms and policies as given.

Equilibrium

Let (t) P1 n=0

f

n (t)

1 n (t)gn=0

denote the distribution of industries over di¤erent technology gaps, with

= 1. For example,

0 (t)

denotes the fraction of industries in which the …rms are

neck-and-neck at time t. Throughout, we focus on Markov Perfect Equilibria (MPE), where strategies are only functions of the payo¤-relevant state variables.15 This allows us to drop the dependence on industry j, thus we refer to R&D decisions by xn for the technology leader that is n steps ahead and by x

n

for a follower that is n steps behind. Let us denote the

list of decisions by the leader and the follower with technology gap n at time t by hxn (t) ; pi (j; t) ; yi (j; t)i and

n (t)

hx

16 n (t)i.

Throughout,

will indicate the whole

(t)

1 n (t)gn= 1 :

We de…ne an allocation as

sequence of decisions at every state, so that follows: De…nition 1 (Allocation) Let

f

be the IPR policy sequence. Then an allocation is a sequence

of decisions for a leader that is n = 0; 1; 2; ::: step ahead, [ decisions for a follower that is n = 1; 2; ::: step behind, [w (t)]t

0,

n (t)

n (t)]t 0 ,

n (t) t 0 ,

a sequence of R&D

a sequence of wage rates

and a sequence of industry distributions over technology gaps [ (t)]t

0.

For given IPR sequence , MPE strategies, which are only functions of the payo¤-relevant state variables, can be represented as follows x:Z

R2+

[0; 1]1 ! R+ :

This mapping represents the R&D decision of a …rm (both when it is the follower and when it is the leader in an industry) as a function of the technology gap, n 2 Z, the aggregate level

of output and the wage, (Y; w) 2 R2+ , and R&D decision of the other …rm in the industry,

x ~ 2 [0; 1]1 . Consequently, we have the following de…nition of equilibrium: 15

MPE is a natural equilibrium concept in this context, since it does not allow for implicit collusive agreements between the follower and the leader. While such collusive agreements may be likely when there are only two …rms in the industry, in most industries there are many more …rms and also many potential entrants, making collusion more di¢ cult. Throughout, we assume that there are only two …rms to keep the model tractable. 16 The price and output decisions, pi (j; t) and yi (j; t), depend not only on the technology gap, aggregate output and the wage rate, but also on the exact technology rung of the leader, nij (t). With a slight abuse of notation, throughout we suppress this dependence, since their product pi (j; t) yi (j; t) and the resulting pro…ts for the …rm, (20), are independent of nij (t), and consequently, only the technology gap, nj (t), matters for pro…ts, R&D, aggregate output and economic growth.

14

De…nition 2 (Equilibrium) Given an IPR policy sequence , a Markov Perfect Equilibrium is given by a sequence [ plied by [

(t)]t

i.e., [x (t)]t [w (t)]t

0,

0

0

(t) ; w (t) ; Y (t)]t

such that (i) [pi (j; t)]t

satisfy (12) and (13); (ii) R&D policy [x (t)]t

0

0

and [yi (j; t)]t

government policy 0

the wage sequence [w (t)]t

0

im-

is a best response to itself,

maximizes the expected pro…ts of …rms taking aggregate output [Y (t)]t

aggregate output [Y (t)]t

3.5

0

and the R&D policies of other …rms [x (t)]t

0

0,

wages

as given; (iii)

is given by (8); and (iv) the labor market clears at all times given 0.

The Labor Market

Since only the technology leader produces, labor demand in industry j with technology gap nj (t) = n can be expressed as n

Y (t) for n 2 Z+ : w (t)

ln (t) =

(21)

In addition, there is demand for labor coming for R&D from both followers and leaders in all industries. Using (14) and the de…nition of the G function, we can express industry demands for R&D labor as hn (t) = G (xn (t)) + G (x where G (xn (t)) and G (x

n (t))

n (t))

for n 2 Z+ ;

(22)

refer to the demand of the leader and the follower in an

industry with a technology gap of n. Note that in this expression, x

n (t)

refers to the R&D

e¤ort of a follower that is n steps behind. The labor market clearing condition can then be expressed as: 1

1 X

n=0

and ! (t)

n (t)

1 ! (t)

n

+ G (xn (t)) + G (x

n (t))

;

(23)

0, with complementary slackness, where w (t) Y (t)

! (t)

(24)

is the labor share at time t. The labor market clearing condition, (23), uses the fact that total supply is equal to 1, and demand cannot exceed this amount. If demand falls short of 1, then the wage rate, w (t), and thus the labor share, ! (t), have to be equal to zero (though this will never be the case in equilibrium). The right-hand side of (23) consists of the demand for production (the terms with ! in the denominator), the demand for R&D workers from the neck-and-neck industries (2G (x0 (t)) when n = 0) and the demand for R&D workers coming from leaders and followers in other industries (G (xn (t)) + G (x

15

n (t))

when n > 0).

De…ning the index of aggregate quality in this economy by the aggregate of the qualities of the leaders in the di¤erent industries, i.e., Z ln Q (t)

1

ln qi (j; t) dj;

(25)

0

the equilibrium wage can be written as:17 P1

w (t) = Q (t)

3.6

n=0

n

n (t)

:

(26)

Steady State and the Value Functions under Quick Catch-up

Let us now focus on steady-state (Markov Perfect) equilibria, where the distribution of industries

(t)

f

1 n (t)gn=0

is stationary, ! (t) de…ned in (24) and g; the growth rate of the

economy, are constant over time. We will establish the existence of such an equilibrium and characterize a number of its properties. If the economy is in steady state at time t = 0, then by de…nition, we have Y (t) = Y0 eg

t

and w (t) = w0 eg t , where g is the steady-state growth

rate. These two equations also imply that ! (t) = ! for all t

0. Throughout, we assume

that the parameters are such that the steady-state growth rate g is positive but not large enough to violate the transversality conditions. This implies that net present values of each …rm at all points in time will be …nite. This enables us to write the maximization problem of a leader that is n > 0 steps ahead recursively. First note that given an optimal policy x ^ for a …rm, the net present discounted value of a leader that is n steps ahead at time t can be written as: Z 1 Vn (t) = Et exp ( r (s t)) [ (s) w (s) G (^ x (s))] ds t

where

(s) is the operating pro…t at time s

ture at time s

t and w (s) G (^ x (s)) denotes the R&D expendi-

t. All variables are stochastic and depend on the evolution of the technology

gap within the industry. Next taking as given the equilibrium R&D policy of other …rms, x interest and wage rates, r (t) and w (t), and equilibrium pro…ts f

n (t), the equilibrium 1 n (t)gn=1 (as a function of

equilibrium aggregate output), this value can be written as (see the Appendix for the derivation i R1h (t) nj ln qi (j; t) l (j; t) dj = 0 ln qi (j; t) + ln Yw(t) dj, where the second equality uses R1 (21). Thus we have ln Y (t) = 0P[ln qi (j; t) + ln Y (t) ln w (t) nj ln ] dj. Rearranging and canceling terms, R 1 n=0 n n (t) , we obtain (26). and writing exp nj ln dj = 17

Note that ln Y (t) =

R1 0

16

of this equation):18 V_ n (t) = max

r (t) Vn (t)

[

n (t)

w (t) G (xn (t))] + xn (t) [Vn+1 (t) + x n (t) + n [V0 (t) Vn (t)]

xn (t) 0

Vn (t)]

; (27)

where V_ n (t) denotes the derivative of Vn (t) with respect to time. The …rst term is current pro…ts minus R&D costs, while the second term captures the fact that the …rm will undertake an innovation at the ‡ow rate xn (t) and increase its technology lead by one step. The remaining terms incorporate changes in value due to quick catch-up by the follower (‡ow rate x

n (t) + n

in the second line). In steady state, the net present value of a …rm that is n steps ahead, Vn (t), will also grow at a constant rate g for all n 2 Z+ . Let us then de…ne the normalized values as Vn (t) Y (t)

vn (t)

(28)

for all n 2 Z, which will be independent of time in steady state, i.e., vn (t) = vn . Using (28) and the fact that from (7), r (t) = g (t)+ , the recursive form of the steady-state value function (27) can be written as: vn = max

1

xn 0

where x

n

n

! G (xn ) + xn [vn+1

vn ] + x

n

+

n

[v0

vn ]

for n 2 N; (29)

is the equilibrium value of R&D by a follower that is n steps behind, and ! is the

steady-state labor share (while xn is now explicitly chosen to maximize vn ). Similarly the value for neck-and-neck …rms is v0 = max f ! G (x0 ) + x0 [v1 x0 0

v0 ] + x0 [v

1

v0 ]g ;

(30)

while the values for followers are given by v

n

= max f ! G (x x

n

0

n)

+ [x

n

+

n ] [v0

v

n]

+ xn [v

n 1

v

n ]g

for n 2 N:

(31)

For neck-and-neck …rms and followers, there are no instantaneous pro…ts, which is re‡ected in (30) and (31). In the former case this is because neck-and-neck …rms sell at marginal cost, and in the latter case, this is because followers have no sales. These normalized value functions emphasize that, because of growth, the e¤ective discount rate is r (t)

g (t) =

rather than

r (t). The maximization problems in (29)-(31) immediately imply that any steady-state equilib18

Clearly, this value function could be written for any arbitrary sequence of R&D policies of other …rms. We set the R&D policies of other …rms to their equilibrium values, x n (t), to reduce notation in the main body of the paper.

17

rium R&D policies, x , must satisfy:

x

xn = max G0

1

= max G0

1

x0 = max G0

1

n

[vn+1 vn ] ;0 ! [v0 v n ] ;0 ! [v1 v0 ] ;0 ; !

(32) (33) (34)

where the normalized value functions, the vs, are evaluated at the equilibrium, and G0

1(

) is

the inverse of the derivative of the G function. Since G is twice continuously di¤erentiable and strictly concave, G0

1

is continuously di¤erentiable and strictly increasing. These equations

therefore imply that innovation rates, the xn s, will increase whenever the incremental value of moving to the next step is greater and when the cost of R&D, as measured by the normalized wage rate, ! , is less. Note also that since G0 (0) > 0, these R&D levels can be equal to zero, which is taken care of by the max operator. The response of innovation rates, xn , to the increments in values, vn+1

vn , is the key

economic force in this model. A policy that reduces the patent protection of leaders that are n+ 1 steps ahead (by increasing vn+1

n+1 )

will make being n+1 steps ahead less pro…table, thus reduce

vn and xn . This corresponds to the standard disincentive e¤ ect of relaxing IPR policy.

This result corresponds to fact (1) in the toy model. In contrast to existing models, however, here relaxing IPR policy can also create a positive incentive e¤ ect. Somewhat paradoxically, lower protection for technology leaders that are n + 1 steps ahead will tend to reduce vn+1 , thus increasing vn+2

vn+1 and xn+1 . This result is very similar to fact (2) in the toy model.

We will see this positive incentive e¤ect plays an important role in the form of optimal statedependent IPR policy. In addition to the incentive e¤ects, relaxing IPR protection may also create a bene…cial composition e¤ ect; this is because, typically, fvn+1 sequence, which implies that xn

1

is higher than xn for n

vn g1 n=0 is a decreasing

1 (see, e.g., Proposition 4).

Weaker patent protection (in the form of shorter patent lengths) will shift more industries into the neck-and-neck state and potentially increase the equilibrium level of R&D in the economy. Finally, weaker patent protection also creates a bene…cial “level e¤ect”by in‡uencing equilibrium markups and prices (as shown in equation (12) above) and by reallocating some of the workers engaged in “duplicative” R&D to production. This level e¤ect will also feature in our welfare computations. The optimal level and structure of IPR policy in this economy will be determined by the interplay of these various forces. Given the equilibrium R&D decisions x , the steady-state distribution of industries across

18

states

has to satisfy the following accounting identities: xn+1 + x

+

n 1

x1 + x 2x0

0

=

n+1

1

+

1 X

n+1

1

1

x

n

= xn = 2x0

+

n

n

for n 2 N;

(35)

0;

(36)

n:

(37)

n=1

The …rst expression equates exit from state n + 1 (which takes the form of the leader going one more step ahead or the follower catching up the leader) to entry into the state (which takes the form of a leader from state n making one more innovation). The second equation, (36), performs the same accounting for state 1, taking into account that entry into this state comes from innovation by either of the two …rms that are competing neck-and-neck. Finally, equation (37) equates exit from state 0 with entry into this state, which comes from innovation by a follower in any industry with n

1.

The labor market clearing condition in steady state can then be written as 1

1 X

1 n

n=0

!

n

+ G (xn ) + G x

and !

n

0,

(38)

with complementary slackness. The next proposition characterizes the steady-state growth rate. As with all the other results in the paper, the proof of this proposition is provided in the Appendix. Proposition 3 Let the steady-state distribution of industries and R&D decisions be given by

, then the steady-state growth rate is " g = ln

2

0 x0

+

1 X

n=1

n xn

#

:

(39)

This proposition clari…es that the steady-state growth rate of the economy is determined by two factors: (1) R&D decisions of industries at di¤erent levels of technology gap, x fxn g1 n=

1;

(2) The distribution of industries across di¤erent technology gaps,

f

1 n gn=0 .

IPR policy a¤ects these two margins in di¤erent directions as illustrated by the discussion above.

4

Existence and Characterization of Steady-State Equilibria

We now de…ne a steady-state equilibrium in a more convenient form, which will be used to establish existence and derive some of the properties of the equilibrium.

19

De…nition 3 (Steady-State Equilibrium) Given an IPR policy rium is a tuple

such that the distribution of industries fvn g1 n= 1

(36) and (37), the values v

satisfy (35),

satisfy (29), (30) and (31), the R&D decision x is

given by (32), (33) and (34), the steady-state labor share ! satis…es (38) and the steady-state growth rate g is given by (39). We next provide a characterization of the steady-state equilibrium, starting …rst with the case in which there is uniform IPR policy.

4.1

Uniform IPR Policy

Let us …rst focus on the case where IPR policy is uniform, i.e. uni .

we denote this by so that v

n

=v

1

n

=

< 1 for all n 2 N and

In this case, (31) implies that the problem is identical for all followers,

for n 2 N. Consequently, (31) can be replaced with the following simpler

equation: v

1

= max f ! G (x x

1

1)

0

+ [x

1

+ ] [v0

v

1 ]g ;

(40)

implying optimal R&D decisions for all followers of the form x

1

= max G0

1

[v0

v

1]

!

;0 :

(41)

Let us denote the sequence of value functions under uniform IPR as fvn g1 n=

1.

We next

establish the existence of a steady-state equilibrium under uniform IPR and characterize some of its most important properties. Establishing the existence of a steady-state equilibrium in this economy is made complicated by the fact that the equilibrium allocation cannot be represented as a solution to a maximization problem. Instead, as emphasized by De…nition 3, each …rm maximizes its value taking the R&D decisions of other …rms as given; thus an equilibrium corresponds to a set of R&D decisions that are best responses to themselves and a labor share (wage rate) ! that clears the labor market. Nevertheless, there is su¢ cient structure in the model to guarantee the existence of a steady-state equilibrium and monotonic behavior of values and R&D decisions. Proposition 4 Consider G0

1

1

1

a

uniform

IPR

policy

uni

= ( + ) > 0. Then a steady-state equilibrium

exists.

Moreover, in any steady-state equilibrium ! < 1. In addition, if either

> 0 or x

1

> 0,

then g > 0. For any steady-state R&D decisions x , the steady-state distribution of industries is uniquely determined. In addition, we have the following results: 20

v

1

v0 and fvn g1 n=0 forms a bounded and strictly increasing sequence converging to

some v1 2 (0; 1). x0 > x1 , x0 provided that

x

1, G0 1

and xn+1 1

1

xn for all n 2 N with xn+1 < xn if xn > 0. Moreover, = ( + ) > 0 and x0 > x

1.

Proof. See the Appendix. Remark 1 The condition that G0

1

1

1

= ( + ) > 0 ensures that there will be posi-

tive R&D in equilibrium. If this condition does not hold, then there exists a trivial steady-state equilibrium in which xn = 0 for all n 2 Z+ , i.e., an equilibrium in which there is no innovation and thus no growth (see the Appendix for more details). Moreover, provided that equilibrium would also involve

0

> 0, this

= 1, so that in every industry two …rms with equal costs

compete a la Bertrand and charge price equal to marginal cost, leading to zero aggregate pro…ts and a labor share of output equal to 1. The assumption that G0 on the other hand, is su¢ cient to rule out

0

1

1

1

= ( + ) > 0,

= 1 and thus ! = 1. If, in addition, the

steady-state equilibrium involves some probability of catch-up or innovation by the followers, i.e., either

> 0 or x

1

> 0, then the growth rate is also strictly positive.

In addition to the existence of a steady-state equilibrium with positive growth, Proposition 4 shows that the sequence of values fvn g1 n=0 is strictly increasing and converges to some v1 , and more importantly that x

fxn g1 n=1 is a decreasing sequence, which implies that technology

leaders that are further ahead undertake less R&D. Intuitively, the bene…ts of further R&D are decreasing in the technology gap, since greater values of the technology gap translate into smaller increases in the equilibrium markup (recall (20)). Moreover, the R&D level of neck and-and-neck …rms, x0 , is greater than both the R&D level of technology leaders that are one step ahead and followers that are one step behind (i.e., x0 > x1 and x0

x

1 ).

This

implies that with uniform policy neck-and-neck industries are “most R&D intensive,” while industries with the largest technology gaps are “least R&D intensive”. This is closely related to the “escape competition” mechanism discussed in Aghion, Harris, Howitt and Vickers (2001), whereby incumbents undertake more R&D when their lead over the followers is more limited. It is also the basis of the conjecture mentioned in the Introduction that reducing protection given to technologically advanced leaders might be useful for increasing R&D by bringing them into the neck-and-neck state.

21

4.2

State-Dependent IPR Policy

We now extend the results from the previous section to the environment with state-dependent IPR policy, though results on monotonicity of values and R&D e¤orts no longer hold.19 Proposition 5 Consider G0 1

1

1

=( +

1)

the

state-dependent

IPR

policy

and

> 0. Then a steady-state equilibrium

exists. Moreover, in any steady-state equilibrium ! < 1. In addition, if either x

1

that

1

> 0 or

> 0, then g > 0.

Proof. See the Appendix. Unfortunately, it is not possible to determine the optimal (welfare- or growth-maximizing) state-dependent IPR policy analytically. For this reason, in Section 5, we undertake a quantitative investigation of the form and structure of optimal state-dependent IPR policy using plausible parameter values.

5

Optimal IPR Policy: Towards A Quantitative Investigation

In the remainder of the paper, we investigate the implications of various di¤erent types of IPR policies on R&D, growth and welfare using numerical computations of the steady-state equilibrium. Our purpose is not to provide a detailed calibration of the model economy but to highlight its qualitative implications for optimal IPR policy under plausible parameter values. We focus on optimal policy, de…ned as steady-state welfare-maximizing choice of policy (growth-maximizing policies give very similar results and are omitted to save space). In this section, we introduce the measure of steady-state welfare and describe our quantitative methodology. Results are reported in the subsequent sections.

5.1

Welfare

Our focus so far has been on steady-state equilibria (mainly because of the very challenging nature of transitional dynamics in this class of models). In our quantitative analysis, we continue to focus on steady states and thus look at steady-state welfare. In a steady-state equilibrium, welfare at time t = 0 can be written as Z 1 Welfare (0) = e t ln Y (0) eg

t

dt

0

=

ln Y (0)

19

+

g

2

;

(42)

This is because IPR policies could be very sharply increasing at some technology gap, making a particular state very unattractive for the leader. For example, we could have n = 0 and n+1 ! 1, which would imply that vn+1 vn is negative.

22

where the …rst-line uses the facts that all output is consumed, utility is logarithmic (recall (6)), output and consumption at date t = 0 are given by Y (0), and in the steady-state equilibrium output grows at the rate g . The second line simply evaluates the integral. Next, note that Z 1 ln Y (t) = ln y (j; t) dj 0 Z 1 q i (j; t) Y (t) dj ln = w (t) 0 Z 1 = ln q i (j; t) dj ln ! (t) 0 ! 1 X = ln Q (t) ln n n (t) ln ! (t) ; (43) n=0

where the …rst line simply uses the de…nition in (8), the second line substitutes for y (j; t) from (13), the third line uses the de…nition of the labor share ! (t), and the …nal line uses the de…nition of Q (t) from (25) together with the fact that in the steady state qi (j; t) = in a fraction

n (t)

n

q

i (j; t)

of industries. The expression in (43) implies that output simply depends

on the quality index, Q (t), the distribution of technology gaps,

(t) (because this determines

markups), and also on the labor share, ! (t). In steady-state equilibrium, the distribution of technology gaps and labor share are constant, while output and the quality index grow at the steady-state rate g . Therefore, for steady-state comparisons of welfare across economies with di¤erent policies, it is su¢ cient to compare two economies with the same level of Q (0), but with di¤erent policies. We can then evaluate steady-state welfare with the distribution of industries given by their steady-state values in the two economies, and output and the quality index growing at the corresponding steady-state growth rates. Expression (43) also makes it clear that only the aggregate quality index Q (0) needs to be taken to be the same in the di¤erent economies. Given Q (0), the dispersion of industries in terms of the quality levels has no e¤ect on output or welfare (though, clearly, the distribution of industries in terms of technology gaps between leaders and followers,

, in‡uences the level of markups and output,

and thus welfare). However, note one di¢ culty with welfare comparisons highlighted by equations (42) and (43); proportional changes in steady-state welfare due to policy changes will depend on the initial level of Q (0), which is an arbitrary number. Therefore, proportional changes in welfare are not informative, though this has no e¤ect on ordinal rankings and thus welfaremaximizing policy is well de…ned and independent of the level of Q (0). Equations (42) and (43) also make it clear that changes in steady-state welfare will be the sum of two components: the …rst is the growth e¤ ect, given by g = 2 , whereas the second is due to changes in

23

P ln ( 1 n=0 n

n) =

ln ! (0). Since changes in the labor share ! (0) are largely driven by the

distribution of industries, we refer to this as the distribution e¤ ect. Policies will typically a¤ect both of these quantities. In what follows, we give the welfare rankings of di¤erent policies and then report the relative magnitudes of the growth and the distribution e¤ects. This will show that the growth e¤ects will be one or two orders of magnitude greater than the distribution e¤ects and dominate welfare comparisons. So if the reader wishes, he or she may think of the magnitudes of the changes in welfare as given by the proportional changes in growth rates. It is useful to note that in general, welfare is not maximized in the equilibrium we have characterized so far for three reasons. First, there are the usual monopoly distortions as prices deviate systematically from marginal costs. Second, only part of the returns from their innovation is appropriated by leaders. For example, if both leaders and followers improve their technologies by one step, consumers are better o¤, but pro…ts remain unchanged. Third, the social bene…t of innovation by followers di¤ers from private bene…ts; socially such innovation is bene…cial because it reduces markups and tends to increase future R&D investments (as it brings the industry to a neck-and-neck state where R&D is greater), while privately it enables followers to become leaders in the future. IPR policy can mostly close the gap between the equilibrium and the social optimum through the second channel, though it indirectly also manipulates the third.

5.2

Quantitative Methods and Parameter Choices

For our quantitative exercise, we take the annual discount rate as 5%, i.e.,

year

= 0:05. In all

our computations, we work with the monthly equivalent of this discount rate in order to increase precision, but throughout the tables, we convert all numbers to their annual counterparts to facilitate interpretation. The theoretical analysis considered a general production function for R&D given by (14). The empirical literature typically assumes a Cobb-Douglas production function. For example, Kortum (1993) considers a function of the form Innovation (t) = B0 exp ( t) (R&D inputs) ;

(44)

where B0 is a constant and exp ( t) is a trend term, which may depend on general technological trends, a drift in technological opportunities, or changes in general equilibrium prices (such as wages of researchers etc.). The advantage of this form is not only its simplicity, but also the fact that most empirical work estimates a single elasticity for the response of innovation rates to R&D inputs. Consequently, they essentially only give information about the parameter in terms of equation (44). A low value of

implies that the R&D production function is more 24

concave. For example, Kortum (1993) reports that estimates of

vary between 0.1 and 0.6

(see also Pakes and Griliches, 1980, or Hall, Hausman and Griliches, 1988). For these reasons, throughout, we adopt a R&D production function similar to (44): x = Bh where B;

(45) 1

> 0. In terms of our previous notation, equation (45) implies that G (x) = [x=B] w,

where w is the wage rate in the economy (thus in terms of the above function, it is captured by the exp ( t) term).20 Equation (45) does not satisfy the boundary conditions we imposed so far and can be easily modi…ed to do so without a¤ecting any of the results, since in all numerical exercises only a …nite number of states are reached.21 Following the estimates reported in Kortum (1993), we start with a benchmark value of checks for

= 0:1 and

= 0:35, and then report sensitivity

= 0:6. The other parameter in (45), B, is chosen so as to ensure an

annual growth rate of approximately 1.9%, i.e., g ' 0:019, in the benchmark economy which features inde…nitely-enforced patents. This growth rate together with

year

= 0:05 also pins

down the annual interest rate as ryear = 0:069 from equation (7). We choose the value of

using a reasoning similar to Stokey (1995). Equation (39) implies

that if the expected duration of time between any two consecutive innovations is about 3 years in an industry, then a growth rate of about 1.9% would require

= 1:05.22 This value is also

consistent with the empirical …ndings of Bloom, Schankerman and Van Reenen (2005).23 We take

= 1:05 as the benchmark value. We then check the robustness of the results to

= 1:01

and

= 1:2 (expected duration of 8 months and 13 years, respectively). Finally, without loss

of generality, we normalize labor supply to 1. This completes the determination of all the parameters in the model except the IPR policy. As noted above, we begin with the full patent protection regime, i.e.,

= f0; 0; :::g. We

then move to a comparison of the optimal (welfare-maximizing) uniform IPR policy 20

uni

to

More speci…cally, (45) can be alternatively written as Innovation(t) = Bw (t) (R&D expenditure) , thus would be equivalent to (44) as long as the growth of w (t) can be approximated by constant rate. 21 For example, we could add a small linear term to the production function for R&D, (45), and also make it ‡at after some level h. For example, the following generalization of (45), x = min Bh + "h; B h + "h for " small and h large, makes no di¤erence to our simulation results. 22 In particular, in our benchmark parameterization with full protection without licensing, 24% of industries are in the neck-and-neck state. This implies that improvements in the technological capability of the economy is driven by the R&D e¤orts of the leaders in 76% of the industries and the R&D e¤orts of both the leaders and the followers in 24% of the industries. Therefore, the growth equation, (39), implies that g ' ln 1:24 x, where x denotes the average frequency of innovation in a given industry. A major innovation on average every three years implies a value of ' 1:05. 23 The production function for the intermediate good, (10), can be written as log (y (j; t)) = n (j; t) log ( ) + log (l (j; t)), where n (j; t) is the number of innovations to date in sector j and represents the “knowledge stock” of this industry. Bloom, Schankerman and Van Reenen (2005) proxy the knowledge stock in an industry by the stock of R&D in that industry and estimate the elasticity of sales with respect to the stock of R&D to be approximately 0.06. In terms of the exercise here, this implies that log ( ) = 0:06, or that 1:06.

25

the optimal state-dependent IPR policy. Since it is computationally impossible to calculate the optimal value of each

n,

we limit our investigation to a particular form of state-dependent

IPR policy, whereby the same

applies to all industries that have a technology gap of n = 5

or more. In other words, the IPR policy can be represented as: IPR policy ! Technology gap: n !

none 0

1

2

3

4

1

2

3

4

z}|{ z}|{ z}|{ z}|{ z

5

}|

{

5 6 7 8 9 10 11 :

We checked and veri…ed that allowing for further ‡exibility (e.g., allowing

5

:1

and

6

to di¤er)

has little e¤ect on our results. The numerical methodology we pursue relies on uniformization and value function iteration. The details of the uniformization technique are described in the proof of Lemma 1 in the Appendix (for details of value function iteration, see Judd, 1999). In particular, we …rst take the IPR policy

as given and make an initial guess for the equilibrium labor share ! . Then

for a given ! , we generate a sequence of values fvn g1 n= policies,

fxn g1 n= 1

1,

and we derive the optimal R&D

and the steady-state distribution of industries, f

1 n gn=0 .

After convergence,

we compute the growth rate g and welfare, and then check for market clearing in the labor market from equation (23). Depending on whether there is excess demand for or supply of labor, ! is varied and the numerical procedure is repeated until the entire steady-state equilibrium for a given IPR policy is computed. The process is then repeated for di¤erent IPR policies. In the state-dependent IPR case, the optimal (welfare-maximizing) IPR policy sequences, ; are computed one element at a time, until we …nd the welfare-maximizing value for that component, for example,

1.

welfare-maximizing value of

We then move the next component, for example, 2

is determined, we go back to optimize over

1

2.

Once the

again, and this

procedure is repeated recursively until convergence.24

6

Optimal IPR Policy

In this section, we present a quantitative analysis of our baseline model.

6.1

Full IPR Protection

We start with the benchmark with full protection, which is the case that the existing literature has considered so far (e.g., Aghion, Harris, Howitt and Vickers, 2001). In terms of our model, 24

After we …nd a maximizer ( to verify the solution.

), we also evaluate several random policy combinations around the maximizer

26

this corresponds to

n

= 0 for all n. We choose the parameter B in terms of (45), so that the

benchmark economy has an annual growth rate of 1.86%. [Figure 2-4 here] The value function for this benchmark case is shown in Figure 2 (solid line). The value function has decreasing di¤erences for n

0, which is consistent with the results in Proposition

4, and features a constant level for all followers (since there is no state dependence in the IPR policy). Figure 3 shows the level of R&D e¤orts for leaders and followers in this benchmark (again solid line). Again consistent with Proposition 4, this …gure also shows that the R&D level of a leader declines as the technology gap increases and that the highest level of R&D is for …rms that are neck-and-neck (i.e., at the technology gap of n = 0). Since there is no state-dependent IPR policy, all followers undertake the same level of R&D e¤ort, which is also shown in the …gure. Figure 4 shows the distribution of industries according to technology gaps (again the solid line refers to the benchmark case). The mode of the distribution is at the technology gap of n = 1, but there is also a signi…cant concentration of industries at technology gap n = 0, because innovations by the followers take them to the “neck-and-neck” state. [Table 1 here] The …rst column of Table 1 also reports the results for this benchmark simulation. As noted above, in each case B is chosen such that the annual growth rate is equal to 0.0186, which is recorded at the bottom of Table 1 together with the initial consumption and welfare levels according to (42) and (43). The table also shows the R&D levels x0 , x

1

and x1 (0.35, 0.22 and

0.29), the frequencies of industries with technology gaps of 0, 1 and 2. The steady-state value of ! is 0.95. Since labor is the only factor of production in the economy, ! should not be thought of as the labor share in GDP. Instead, 1

! measures the share of pure monopoly pro…ts in

value added. In the benchmark parameterization, this corresponds to 5% of GDP, which is reasonable.25 Finally, the table also shows that in this benchmark parameterization 3.2% of the workforce is working as researchers, which is also consistent with US data.26 These results are encouraging for our simple quantitative exercise, since with very few parameter choices, 25 Bureau of Economic Analysis (2004) reports that the ratio of before-tax pro…ts to GDP in the US economy in 2001 was 7% and the after-tax ratio was 5%. 26 According to National Science Foundation (2006), the ratio of scientists and engineers in the US workforce in 2001 is about 4%.

27

the model generates reasonable numbers, especially for the share of the workforce allocated to research.27

6.2

Optimal Uniform IPR Protection

For reference, we now characterize optimal uniform IPR policy, that is, we impose that for all n, and look for values of

=

n

that maximizes the welfare in the economy. Column 2 of

Table 1 shows that the welfare-maximizing value of

is not di¤erent from zero at the three-digit

level. Therefore the results of the full protection case carries over to uniform policy as well. The main reason for this result is the quick catch-up assumption. Recall that the uniform IPR policy discourages innovation, but generates a potential bene…t because of the composition e¤ect (bringing more …rms into neck-and-neck position). In the quick catch-up regime, …rms come into neck-and-neck position at a Poisson rate of 0.22, which results in 35% of sectors being in state 0 and 77% at two-step gap or below. This implies that there are only limited composition gains. In this light, it is not surprising that relaxing the IPR protection uniformly is not bene…cial; it generates a signi…cant disincentive e¤ect and little bene…t. Therefore, optimal IPR policy is to set full protection,

= 0, and thus the value functions, innovation

rates and industry distributions under optimal uniform IPR policy are given by the solid lines in Figures 2-4.

6.3

Optimal State-Dependent IPR

We next turn to our major question; whether state-dependent IPR makes a signi…cant di¤erence relative to the uniform IPR. In particular, we look for the combination of f 1 ; :::;

5g

that

maximizes the welfare. The new value function, innovation rates and industry distribution are plotted in Figure 2-4 and the numerical results are shown in column 3 of Table 1. Two features are worth noting. First, the growth rate increases noticeably relative to column 1; it is now 2.04% instead of 1.86%. Second and more important, we see the key pattern that will be present in all of our quantitative results: optimal state-dependent policy f 1 ; :::;

5g

provides greater protection to technology leaders that are further ahead. In particular, we …nd that the optimal policy involves

1

= 0:71,

2

= 0:08, and

3

=

4

=

5

= 0. This corresponds

to very little patent protection for …rms that are one step ahead of the followers. In particular, since

1

= 0:71 and x

1

= 0:12, in this equilibrium …rms that are one step behind followers are

27 Most endogenous growth models imply that a signi…cantly greater fraction of the labor force should be employed in the research sector and one needs to introduce various additional factors to reduce the pro…tability of research or to make entry into research more di¢ cult. In the current model, the step-by-step nature of innovation and competition plays this role and generates a plausible allocation of workers between research and production.

28

more than six times as likely to catch up with the technology leader because of the expiration of the patent of the leader as they are likely to catch up because of their own successful R&D. Then, there is a steep increase in the protection provided to technology leaders that are two steps ahead, and

2

is 1/12th of

1.

Perhaps even more remarkably, after a technology gap of

three or more steps, optimal IPR involves full protection, and patents never expire. This pattern of greater protection for technology leaders that are further ahead may go against a naïve intuition that state-dependent IPR policy should try to boost the growth rate of the economy by bringing the industries with largest technology gaps (where leaders engage in little R&D) into neck-and-neck competition. This composition e¤ect is present, but dominated by another, more powerful force, the trickle-down e¤ect. The intuition for the trickle-down e¤ect is as follows: by providing secure patent protection to …rms that are three or more steps ahead of their rivals, optimal state-dependent IPR increases the R&D e¤ort of leaders that are one and two steps ahead as well. This is because technology leaders that are only one or two steps ahead now face greater returns to R&D, which will not only increase their pro…ts but also the security of their intellectual property. Mechanically, high levels of

1

and

v1 and v2 , while high IPR protection for more advanced …rms increases vn for n increases the R&D incentives of leaders at n = 1 or at n =

2

reduce

3, and this

2.28

Providing more secure patent protection through less frequent catch-up bene…ts an n-step leader more than (n + 1)-step leader since the preserved pro…t is higher for a more advanced …rm. This results in a steeper value function as illustrated in Figure 2. The slope of the value function is the key determining factor for R&D decisions and this increase in slope re‡ects itself in overall higher R&D e¤ort by the leaders in Figure 3. It is also notable that statedependent IPR introduces positive incentive e¤ect while gaining also from the composition. Figure 4 shows that the mode of the new distribution is at n = 0: The average innovation rate is higher (as re‡ected on a higher growth rate, g = 2:04%) and the average mark-up is lower (C (0) increases by 52%). This pattern of greater R&D investments under state-dependent IPR contrasts with uniform IPR, which always reduces R&D of all …rms. The possibility that imperfect state-dependent IPR protection can increase (rather than reduce) R&D incentives is a novel feature of our approach and has also been shown explicitly in the partial equilibrium model of Section 2. 28

An alternative intuition, suggested by an anonymous referee, is that when the technology gap is greater, leaders will lose more from a relaxation of IPR. However, this intuition can only be partial, since, as shown in Section 2, state-dependent relaxation of IPR in this form creates a positive incentive e¤ect, which is central to our results (and this is independent of how much technology leaders lose as a result of the relaxation of IPR). As a result, we believe that the trickle-down of incentives is the more correct intuition for our results.

29

6.4

Robustness

The patterns shown in Figures 2-4 and Table 1 are quite robust. In Tables A1-A4 in the Appendix, we report results from the same exercise for various di¤erent combinations of values of

and

(in particular, varying

to

= 0:1 and

= 0:6, and

to

= 1:01 and

= 1:2). In

each case, we change the parameter B in equation (45) so that the growth rate of the benchmark economy with full IPR protection without licensing is the same as in Table 1, i.e., g = 1:86%. The overall pattern and in fact the quantitative magnitudes are remarkably similar to the baseline results reported here. We therefore conclude that the trickle-down of incentives and the form of the optimal state-dependent IPR policy apply for a range of plausible parameter values.

7

Optimal IPR Policy in the Slow Catch-up Regime

In this section, we extend our analysis to an environment where followers close the gap with technology leaders also step by step. We then also introduce di¤erent types of R&D e¤orts by followers and study several di¤erent dimensions of IPR policy.

7.1

Value Functions

The environment is the same as in Section 3, except that we now assume that successful R&D by followers close is the gap between themselves and the technology leader by one step. We will allow for di¤erent types of R&D below. The equivalent expressions for the value functions (29)-(31) in this case are vn = max

1

xn 0

n

! G (xn ) + xn [vn+1

vn ] + x

n [vn 1

vn ] +

n [v0

vn ]

for n 2 N; (46)

v0 = max f ! G (x0 ) + x0 [v1 x0 0

v0 ] + x0 [v

1

v0 ]g ;

(47)

and v

n

= max f ! G (x x

n

0

n)

+x

n [v n+1

v

n]

+ xn [v

n 1

v

n]

+

n [v0

v

n ]g

for n 2 N: (48)

These expressions are intuitive in light of those presented in Section 3, in particular, (29)(31). The only di¤erence from equations (29)-(31) is that, when a follower innovates, an n-step leader’s value changes from vn to vn

1

instead of dropping all the way to v0 , since this

innovation closes the technology gap only by one step. Similarly, in this event, the follower’s value changes from v

n

to v

n+1

instead of increasing all the way to v0 . The rest of the analysis

30

mirrors that in Section 4. In particular, existence of stationary equilibria can be proved using an analogous argument to that provided in the Appendix, but we are not able to prove the analogue of the second part of Proposition 4.

7.2

Quantitative Results

We next investigate the form of optimal IPR policy in the baseline slow catch-up regime. Full IPR Protection Under the slow catch-up regime, setting

n

= 0, that is, providing full protection via in…nite

patent length generates too little catch-up by the followers. Consequently, the steady-state distribution has little mass at or around the “neck-and-neck“ state (n = 0). To generate a more plausible distribution with a non-zero share of industries in the neck-and-neck state, we instead impose

n

= 0:02, which implies an expected length of patent protection of 50 years

(as the full protection benchmark) under slow catch-up regime. [Table 2 here] The …rst column of Table 2 reports the results under this scenario. Even with 50 years of protection, the share of industries that are neck-and-neck is only 2%, and the total share of industries that have a gap of less than two steps is only 8%. One implication of this pattern is that a relaxation of IPR policy may now be more powerful because it can a¤ect the composition of industries, reduce the average mark-up in the economy, and perhaps have a large e¤ect on average R&D. Therefore, this is a particularly relevant environment for investigating whether the trickle-down of incentives identi…ed in the previous section is present and robust in di¤erent and perhaps more realistic environments. Optimal Uniform IPR Protection The second column of Table 2 shows optimal uniform IPR policy in this case. Consistent with Proposition 1 in Section 2, relaxing IPR protection creates a powerful disincentive e¤ect. However, it also generates a bene…cial composition e¤ect by bringing more and more …rms into neck-and-neck competition. For this reason, optimal uniform IPR policy is no longer full protection. The results in the table show that the optimal policy reduces patent length from (average protection of 50 years) to

= 0:02

= 0:11 (average protection of 9 years). This involves

a lower innovation rate for technology leaders that are one-step ahead (from 1:1 to 0:15). Similarly average R&D is also reduced and the aggregate growth rate declines from 2:5% to 31

2:3%: However, because of the increase in the share of neck-and-neck industries (from 2% to 16%) and the increase in the total share of industries that are in the …rst 3 states (from 8% to 49%), the average mark-up in the economy decreases. This enables a large (19-fold) increase in initial consumption C (0) (which is the reason why this policy is optimal even though it reduces growth). Optimal State-Dependent IPR Once again, the most interesting case is when IPR policy is state dependent. In this case, the optimal policy not only bene…ts from the composition e¤ect, but can do so without sacri…cing growth (by exploiting the positive incentive and the trickle-down e¤ects highlighted in Proposition 2 in Section 2). The optimal state-dependent policies shown in column 3 of Table 2. Under this optimal policy, the share of the …rst three states increases by an additional 6 percentage point (55%) and the initial consumption further increases relative to the uniform IPR policy by 30%. More interestingly, the innovation rate of a one-step leader now increases from 0:15 to 0:51 (relative to the uniform policy case) and the growth rate increases back to 2:5%. It is noteworthy that these gains are achieved by providing stronger protections to more advanced …rms, and thus exploiting the trickle-down e¤ect. For example, under the optimal policy one-step leader is caught up seven times more frequently than a …ve-step leader due to patent expiration.

7.3

Compulsory Licensing

In this subsection, we introduce (compulsory) licensing. Several recent empirical papers suggest that licensing has a signi…cant positive impact on …rm innovation (e.g., Moser and Voena, 2011, Almeida and Fernandes, 2008). Consistent with these …ndings, we model licensing as a way of generating knowledge spillovers to the licensee. In particular, we assume that in addition to independent R&D to proceed one step in the quality ladder, followers can also close all intervening steps by reverse-engineering the current leading-edge technology. But this is only possible by making use of the knowledge generated by the leading-edge technology, and the follower will have to pay a prespeci…ed license fee ^n (t) 0 to the leader. The licensing decision of the follower

i is denoted by a

Throughout, we allow a

i (j; t)

i (j; t)

= 1 (a

i (j; t)

= 0 corresponds to independent R&D).

2 [0; 1] for mathematical convenience. The fees in question

are compulsory license fees imposed by policy and are state-dependent, and thus we represent them as: ^ (t) : N ! R+ [ f+1g :

32

n

Note that ^ (t)

o ^ (t) ; ^ (t) ; ::: is a function of time. This is natural, since in a growing 1 2

economy, license fees should not remain constant. As in (28), in what follows we assume that ^ (t) =Y (t), to keep the equilibrium license fees are also scaled up by GDP, so that n n stationary.29 Value Functions with Compulsory Licensing With a similar reasoning to before, relevant value functions in this case can be written as vn = max

+a

xn 0

where a n

n

nx

n [v0

1 vn +

n

! G (xn ) + xn [vn+1 ] + 1 a n x n [vn 1 n

vn ] vn ] +

n [v0

vn ]

for n 2 N;

is the equilibrium value of licensing decision by a follower that is n steps behind, and

is the license fee that it has to pay. The value for neck-and-neck …rms remain unchanged

while the values for followers becomes v

n

=

x

max

n

0;a

+ (1

n 2[0;1]

Note that licensing a

n

a

n) x

! G (x n ) + a n x n [v0 v n v n ] + xn [v n 1 v n [v n+1

n] ] n +

n [v0

v

n]

for n 2 N;

2 [0; 1] is the new additional decision variable of the follower.

Full IPR protection in this case corresponds to prohibitively high licensing fees, i.e.,

n

=1

for all n, and as in the previous subsection, patent protection has expected duration of 50 years ( = 0:02). Therefore, the results in this case will be identical to those reported for full protection in the previous subsection (column 1 of Table 2). This is indeed the case; these results are repeated in column 1 of Table 3 for ease of comparison with the remaining results in this table. [Table 3 here] Optimal Uniform IPR Protection Uniform compulsory licensing policy now corresponds to

n

=

0: The results under the

optimal choice of such uniform compulsory licensing policy are reported in the second column of Table 3. This optimal policy involves

= 1:61, which is more than half of the surplus that

a three-step follower generates from licensing, v0

v

3

= 2:9.

Since this type of licensing allows for more frequent catch-up by followers, a greater share of industries are now in tight competition: the total share of industries with one or two step gaps goes up to 66% (this number was 8% under full protection). This again corresponds to a powerful composition e¤ect and generates a signi…cant reduction in the average mark-up and a 29 Voluntary licensing licensing is brie‡y discussed in the Appendix, where we show that it cannot in general achieve the same results as compulsory licensing.

33

corresponding increase in initial consumption. However, consistent with our previous results, this type of uniform licensing again generates a signi…cant disincentive e¤ect on technology leaders. In particular, more frequent catch-up implies a shorter durations of positive pro…ts. As a result, innovation incentives are reduced; the innovation rate of a one-step leader is now 0:43 instead of 1:1 and the average growth rate declines from 2:5% to 2:1%. Optimal State-Dependent IPR As in our previous exercises, the negative incentive e¤ects of uniform relaxations of IPR protection are recti…ed when policy is state dependent. Optimal state-dependent policy has in fact qualitatively very similar pattern to those reported above. Most importantly, column 3 of Table 3 shows that optimal state-dependent policy provides greater protection to technology leaders that are more advanced. For example, while a two-step leader receives a license fee of 2

= 1:5, a …ve-step leader receives more than its double,

5

= 3:3: Given this pattern, the

trickle-down e¤ ect is again at work and generates positive innovation incentives: the innovation rate of a one-step leader increases to x1 = 0:46 and the aggregate growth rate goes back to 2:5% from 2:1%: This positive gain is generated without sacri…cing the composition e¤ect. Under this policy, 50% of total industries operate with a technology gap less than two and the initial consumption C (0) is now even higher than under uniform policy (by 40%).

7.4

Leapfrogging and Infringement under Slow Catch-up

Finally, we allow the follower to engage in frontier R&D and “leapfrog”the technology leader. This exercise is useful for two reasons. First, the models analyzed so far do not allow R&D by followers to directly contribute to aggregate growth. One might conjecture that this feature strengthens the trickle-down e¤ect. Second, frontier R&D and leapfrogging by followers will allow us to introduce another relevant and important dimension of IPR policy, patent infringement fees. Suppose, now, that followers can undertake two types of R&D. The …rst, which is what we have focused on so far, is catch-up R&D, corresponding to R&D directed at discovering an alternative way of performing the same task as the current leading-edge technology. Catch-up R&D improves the technology of the follower by one step as before. The alternative, frontier R&D, involves followers improving the current leading-edge technology. If this type of R&D succeeds, the follower will have improved the leading-edge technology. However, following such an event, the follower will be judged (e.g., by courts) to have infringed the patent of technology leader with probability

2 (0; 1) and will be required to pay a prespeci…ed infringement penalty

34

^n (fee) #

0 to the leader. The infringement fees are also state dependent and represented by: ^ (t) : N ! R+ [ f+1g ; #

and we again adopt the normalization #n

^ n (t) =Y (t), and denote the Poisson arrival rate #

of innovation by catch-up R&D and frontier R&D by xcn and xfn ; respectively. Then the new value of an n-step leader takes the following form: 1

vn = max

+xc

xn 0

n [vn 1

n

! G (xn ) + xn [vn+1 vn ] vn ] + xf n [v 1 vn + #n ] + n [v0

for n 2 N;

vn ]

The main di¤erence in this equation is that the follower has two di¤erent arrival rates of innovation. If the follower is successful with frontier R&D, the current leader falls one step behind the follower. However, in this event, with probability , it receives an infringement fee of #n . With a similar reasoning, the value of an n-step follower now becomes: v

n

=

! G (x

max xc

n

0;x

f

n

+ xc n [v n+1 v n ] + xf n [v1 v +xn [v n 1 v n ] + n [v0 v n ]

n)

0

n

#n ]

for n 2 N:

The value of a neck-and-neck …rm is unchanged. The quantitative analysis requires an empirical estimate for : Lanjouw and Schankerman (2001) report that around 10% of the US utility patents are …led for infringement. We therefore set

= 0:1. Note also that since the followers now improve the technology frontier through frontier

R&D, the aggregate growth rate becomes " g = ln

2

0 x0

+

1 X

f

n

xn + x

n

n=1

#

:

(49)

Full protection in this case corresponds to in…nite patent infringement fees, i.e., #n = 1, and given the same parameter choices as before, will be identical to column 1 of Table 2. We repeat these results in column 1 of Table 4 for ease of comparison with the rest of the table. [Table 4 here] Optimal Uniform IPR Protection In the uniform policy case, we set #n = #

0: Column 2 of Table 4 shows that the optimal

uniform policy in this case is # = 14: Recall that when a follower undertakes frontier innovation, the probability the that it will have to make this payment is expected infringement payment is

= 0:1: Therefore the

# = 1:4 which is more than half of the surplus that a

three-step follower generates out of leapfrogging, v1 35

v

3

= 2:7:

Column 2 also shows that under this policy, followers undertake more frontier R&D (x 0:23) than catch-up R&D x

c 1

f 1

=

= 0:15 . Parallel to the previous uniform policies, the shorter

duration of monopoly position resulting from innovation reduces innovation incentives. For example, one-step leaders now innovate at the rate 0.3 instead of 1.1. However, despite this disincentive e¤ect, the growth rate increases slightly because leapfrogging allows followers to directly contribute to aggregate growth, as shown by equation (49). Column 2 also shows that the share of industries in one-step gap is now much larger, 1

= 0:42. This is because leapfrogging puts the follower one-step ahead of the previous

leader. Thanks to this e¤ect, optimal uniform IPR protection achieves lower average mark-up and higher initial consumption as well as higher growth. Optimal State-Dependent IPR State-dependent IPR policy once again exploits the trickle-down e¤ect and creates positive incentive e¤ects on innovation. The form of state-dependent policy is the same as before: technologically more advanced leaders receive more protection in the form of higher fees when followers infringe their patents. While a two-step leader receives #1 = 18:1 in case of infringement, a …ve-step leader receives more than double of this fee, #5 = 43:7: In expectation, a three-step follower pays almost 3/4th of the surplus that it generates from leapfrogging (

#3 = 3:1 versus v1

v

3

= 4:1). As a result of this pattern, state-dependent policy not

only generates a greater welfare gain in terms of the initial consumption (C (0) is now approximately twice the level under the optimal uniform policy), but it also exploits the trickle-down e¤ect and increases the equilibrium growth rate by an additional 0.5 percentage point relative to the uniform policy. Additional Results and Robustness In the Appendix, we show that the results are similar when all three IPR policies are simultaneously present. In particular, the optimal pattern of R&D involves in…nitely long patents with prohibitively high compulsory license fees. The only dimension in which IPR protection is not full is because of moderate infringement fees, which permit followers to undertake frontier R&D and leapfrog technology leaders. Crucially, this aspect of IPR is state-dependent and exploits the trickle-down e¤ect. We also report robustness checks for di¤erent values of the parameters

and

(again increasing or reducing

to 1.2 or 1.01, and increasing or reducing

to 0.6 or 0.1). In all cases, the pattern of optimal IPR is similar: infringement fees are state-dependent and provide greater protection to technologically more advanced leaders.

36

8

Conclusions

In this paper, we emphasized the importance of dynamic interactions between IPR protection and competition for understanding the structure of optimal IPR policy. Our main result highlights the importance of a new and powerful e¤ect, the trickle-down e¤ ect, which implies that protection given to companies with signi…cant technological leads over their rivals also dynamically incentivizes companies with more limited technological leads— as further innovation will not only increase their productivity but also grant them additional IPR protection. This new e¤ect implies that optimal IPR policy should be state-dependent and provide greater protection to companies with signi…cant technological leads and only limited IPR protection for those without. To systematically investigate these issues, we developed a dynamic general equilibrium framework with cumulative (step-by-step) innovations. In each industry, technology leaders innovate in order to widen the gap between themselves and the followers, which enables them to charge higher markups. Followers innovate to catch up with or surpass the technology leaders in their industry (by undertaking “frontier R&D”), and can also license the technology of leaders. IPR policy regulates the length of patents, whether licensing is possible and the size of patent infringement fees. We provided existence and characterization results, and a quantitative analysis of the form of “optimal” (welfare-maximizing) IPR policy. In several di¤erent environments and under di¤erent parameter values, we consistently found that the trickle-down e¤ect is present and powerful. It implies that optimal IPR should be state-dependent and should provide greater protection to …rms with greater technological lead over their rivals. In our benchmark parameterization, for example, optimal IPR policy increases the growth rate of the economy from 1.86% to 2.04%, and does so while also signi…cantly increasing initial consumption (and in fact reducing the overall amount of resources allocated to the R&D sector). We also showed that similar qualitative and quantitative results are obtained when followers catch up with technology leaders only slowly. In this extended environment, we also investigated the form of optimal compulsory licensing fees and patent infringement fees, and found them to be similarly state-dependent (in a way that provides greater protection to …rms that are technologically more advanced relative to their rivals). These extensions further showed that compulsory licensing, which allows followers to build on the leading-edge technology in return of a license fee, also has a major impact on the equilibrium growth rate. Our main results go against a naïve intuition that providing less protection to technologically more advanced …rms is socially bene…cial because it would exploit a composition

37

e¤ect (bringing …rms that are furthest apart into “neck and neck”competition to both reduce markups and increase R&D which results from tight competition). This naïve intuition is not correct precisely because of the trickle-down e¤ect we emphasized above. The trickle-down e¤ect implies that providing greater protection to su¢ ciently advanced technology leaders not only increases their R&D e¤orts but also raises the R&D e¤orts of all technology leaders that are less advanced than this level. This is because the reward to innovation now includes the greater protection that they will receive once they reach this higher level of technology. Our analysis and results suggest that in addition to the reasoning based on the static trade-o¤ between IPR protection and competition, the trickle-down e¤ect should also be factored into policy analysis, and naturally calls for future empirical work to estimate its empirical magnitude. In this context, it should be emphasized that our objective in this paper has not been to derive practical policy prescriptions. There is little doubt that our model is simpli…ed, excludes a whole host of important factors, and ignores potential limitations on the form and complexity of IPR policies. Nevertheless, we believe that our results demonstrate a range of robust and new e¤ects that should be further investigated in future work. More generally, the analysis in this paper suggests that a move to a richer menu of IPR policies, in particular, a move towards optimal state-dependent policies, may signi…cantly increase innovation, economic growth and welfare. The results also show that the form of optimal IPR policy may depend on the industry structure (and the technology of catch-up within the industry). The next step in this line of research should be to investigate the robustness of these e¤ects in di¤erent models of industry dynamics. It would also be useful to study whether the relationship between the form of optimal IPR policy and industry structure suggested by our analysis also applies when variation in industry structure has other sources (for example, di¤erences in the extent of …xed costs or demand structure causing di¤erential gaps between technology leaders and followers across industries). The most important area for future work is a detailed empirical investigation of the form of optimal IPR policy, using both better estimates of the e¤ects of IPR policy on innovation rates and also structural models that would enable the evaluation of the e¤ects of di¤erent policies on equilibrium growth and welfare.

38

References Acemoglu, Daron and Ufuk Akcigit (2006) “State-Dependent Intellectual Property Right Policy” NBER working paper, No. 12775. Acemoglu, Daron, Gino Gancia and Fabrizio Zilibotti (2012) “Competing Engines of Growth: Innovation and Standardization,” forthcoming Journal of Economic Theory. Aghion, Philippe, Nick Bloom, Richard Blundell, Rachel Gri¢ th and Peter Howitt (2005) “Competition and Innovation: an Inverted-U Relationship,” Quarterly Journal of Economics, 120, pp. 701-728. Aghion, Philippe and Rachel Gri¢ th (2007) Competition and Growth: Reconciling Theory and Evidence, MIT Press, Cambridge. Aghion, Philippe, Christopher Harris, Peter Howitt and John Vickers (2001) “Competition, Imitation and Growth with Step-by-Step Innovation,” Review of Economic Studies, 68, pp. 467-492 . Aghion, Philippe, Christopher Harris and John Vickers (1997) “Step-by-step Competition, Innovation and Growth: An Example,” European Economic Review, 41, pp. 771-782. Aghion, Philippe and Peter Howitt (1992) “A Model of Growth Through Creative Destruction,” Econometrica, 110, 323-351. Aliprantis, Charalambos D. and Kim C. Border (1999) In…nite Dimensional Analysis: Hitchhiker’s Guide, Springer-Verlag, New York. Almeida, Rita and Ana Margarida Fernandes (2008). “Openness and Technological Innovations in Developing Countries: Evidence from Firm-Level Surveys,” The Journal of Development Studies, 44(5), 701-727. Arrow, Kenneth (1962) “ Economic Welfare and the Allocation of Resources for Invention,” in R.R. Nelson, ed. Universities-National Bureau of Economic Research Conference Series; The Rate and Direction of Economic Activities: Economic and Social Factors, Princeton University Press, New York. Baye, Michael R., and Heidrun C. Hoppe (2003) “The Strategic Equivalence of Rent-seeking, Innovation, and Patent-race Games,” Games and Economic Behavior, 44, 217226. Bloom, Nick, Mark Schankerman and John Van Reenen (2005) “Identifying Technology Spillovers and Product Rivalry,” CEP Discussion Paper, No. 675. Blume, Lawrence E. (1982) “New Techniques for the Study of Stochastic Equilibrium Processes,” Journal of Mathematical Economics, 9, 61-70.

39

Boldrin, Michele and David K. Levine (2008) “Perfectly Competitive Innovation,” Journal of Monetary Economics, 55, 435-453. Boldrin, Michele and David K. Levine (2004) “IP and Market Size,” Univesity of Minnesota and UCLA mimeo. Budd, Christopher, Christopher Harris and John Vickers (1993) “A Model of the Evolution of Duopoly: Does the Asymmetry Between Firms Tend to Increase or Decrease?” Review of Economic Studies, 60, 540-573. Bureau of Economic Analysis (2004) “Annual Revision of the National Income and Product Accounts” Bureau of Economic Analysis News.

BEA 04-31.

www.bea.gov/bea/rels.htm Choi, Jay P. (1991) “Dynamic R&D Competition under Hazard Rate Uncertainty,” Rand Journal of Economics, 22, 596-610. Fudenberg, Drew, Richard Gilbert, Joseph Stiglitz and Jean Tirole (1983) “Preemption, leapfrogging and competition in patent races,”European Economic Review 22, 3-31. Fullerton, Robert and Preston McAfee (1999) “Auctioning Entry into Tournaments,” Journal of Political Economy, 107, 573-605. Gallini, Nancy T. (1992) “Patent Policy and Costly Imitation,” RAND Journal of Economics, 23, 52-63. Gallini, Nancy T. and Suzanne Scotchmer (2002) “Intellectual Property; When is it the Best Incentive Mechanism?” Innovation Policy and the Economy, 2, 51-78. Gilbert, Richard. and Carl Shapiro (1990) “Optimal Patent Length and Breadth,” Journal of Industrial Economics, 21(1), 106-112. Green, Jerry and Suzanne Scotchmer (1995) “On the Division of Pro…t in Sequential Innovation,” Journal of Industrial Economics, 26(1), 20-33. Grossman, Gene and Elhanan Helpman (1990) Innovation and Growth in the Global Economy, Cambridge, MA, MIT Press. Hall, Bronwyn H., Jerry Hausman and Zvi Griliches (1988) “Patents and R&D: Is There a Lag?” International Economic Review, 27, pp. 265-83. Harris, Christopher and John Vickers (1985) “Perfect Equilibrium in a Model of Race,” Review of Economic Studies, 52, 193-209. Harris, Christopher and John Vickers (1987) “Racing with Uncertainty,” Review of Economic Studies, 54, 1-21. Hopenhayn, Hugo and Matthew Mitchell (2001) “Innovation, Variety and Patent Breadth,” RAND Journal of Economics, 32, 152-166.

40

Hopenhayn, Hugo and Matthew Mitchell (2011) “Optimal Patent Policy with Recurrent Innovators,” mimeo. Hugo Hopenhayn, Llobet, Gerard and Matthew Mitchell (2006) “Rewarding Sequential Innovators: Prizes, Patents and Buyouts,”Journal of Political Economy, 114, 10411068. Judd, Kenneth (1999) Numerical Methods in Economics, MIT Press, Cambridge. Kelley, John L. (1955) General Topology, Springer-Verlag, New York. Klemperer, Paul (1990) “How Broad Should the Scope of Patent Protection Be?”Journal of Industrial Economics, 21, 113-130. Kremer, Michael (2002). “Pharmaceuticals and the Developing World,” Journal of Economic Perspectives, 16(4), 67-90. Kortum, Samuel (1993) “Equilibrium R&D and the Patent-R&D Ratio: US Evidence,” American Economic Review, 83, pp. 450-457. Lanjouw, Jean O and Mark Schankerman (2001) “Enforcing Intellectual Property Rights,” NBER Working Paper, No. 8656. Moscarini, Giuseppe and Francesco Squintani (2004) “Competitive Experimentation with Private Information,” Cowles Foundation Discussion Paper, No. 1489. Moser, Petra and Alessandra Voena (2011) “Evidence from the Trading with the Enemy Act,” American Economic Review, forthcoming. National Science Foundation (2006) “Industrial R&D Employment in the United States and in U.S. Multinational Corporations” Division of Science Resource Statistics. NSF 05-302. Arlington, VA: National Science Foundation. www.bsf.gov/sbe/srs/stats.htm O’Donoghue, Ted, Suzanne Scotchmer and Jacques François Thisse (1998) “Patent Breadth, Patent Life and the Pace of Technological Progress,” Journal of Economics and Management Strategy, 7, 1-32. O’Donoghue, Ted and Joseph Zweimuller (2004) “Patents in a Model of Endogenous Growth,” Journal of Economic Growth, 9, 81-123. Pakes, Ariel and Zvi Griliches (1980) “Patents and R&D at the Firm Level: A First Look,” Economic Letters, 5, pp. 377-381. Quah, Danny (2004) “24/7 Competitive Innovation,” LSE Mimeo. Reinganum, Jennifer (1981) “Dynamic Games of Innovation,” Journal of Economic Theory, 25, 21-24. Reinganum, Jennifer (1985) “Innovation and Industry Evolution,” Quarterly Journal of Economics, 100, 81-100.

41

Romer, Paul M. (1990) “Endogenous Technological Change,”Journal of Political Economy, 98, S71-S102. Ross, Sheldon (1996) Stochastic Processes, John Wiley & Sons, Inc. Scotchmer, Suzanne (1999) “On the Optimality of the Patent System,” The Rand Journal of Economics, 30, 181-196. Scotchmer, Suzanne (2005) Innovations and Incentives, MIT Press, Cambridge, Massachusetts. Segal, Ilya and Michael Whinston (2007) “Anti-Trust in Innovative Industries,” American Economic Review, 97, 1703-1730. Tandon, Pankaj (1982) “Optimal Patents with Compulsory Licensing,” Journal of Political Economy 90(3), 470-486. Taylor, Curtis (1995) “Digging for Golden Carrots: An Analysis of Research Tournaments,” American Economic Review, 85, 872-890. Tirole, Jean (1988) The Theory of Industrial Organization, MIT Press, Cambridge, Massachusetts. Stokey, Nancy (1995) “R&D and Economic Growth” Review of Economic Studies, 62, pp. 469-489. Stokey, Nancy and Robert Lucas with E. Prescott (1989) Recursive Methods in Economic Dynamics, Harvard University Press.

42

Tables = 0:05 under three di¤erent

Note: Tables 1-4 give the results of the numerical computations with

IPR policy regimes. They consider a di¤erent environment (quick catch-up, slow catch-up, licensing and leapfrogging) at a time. Depending on the applicability and necessity, the tables report the steadystate equilibrium values of the di¤erence in the values v1 v

3

and frontier R&D rates of a follower that is one step behind,

(xc

and v0 v 1;

f

x

1 );

3;

the (annual) catch-up

the (annual) R&D rate of

neck-and-neck competitors, x0 ; the (annual) R&D rate of one-step leader, x1 ; fraction of industries in neck-and-neck competition,

0;

fraction of industries at a technology gap of n = 1; 2; the value of

“labor share,” ! ; the ratio of the labor force working in research; log of initial (annual) consumption,

ln C(0); the annual growth rate, g ; and the welfare level according to equation (42). It also reports the welfare-maximizing uniform and state-dependent IPR policies. Tables A1-A4 in the Appendix contain robustness checks for the benchmark results of Table 1 with alternative step sizes and R&D elasticity parameters. Table A5 combines the three environments (slow catch-up, licensing and leapfrogging). Table A6 reports the robustness checks of the state-dependent results of Table A5 with alternative step sizes and R&D elasticity parameters. See text for details.

= 1:05; = 0:35; B = 0:1 1 2 3 4 5

x 1 x0 x1 0 1 2

! Researcher ratio ln C (0) g Welfare

Full IPR 0 0 0 0 0 0.22 0.35 0.29 0.24 0.33 0.20 0.95 0.032 33.78 0.0186 683.0

Optimal Uniform IPR 0 0 0 0 0 0.22 0.35 0.29 0.24 0.33 0.20 0.95 0.032 33.78 0.0186 683.0

Optimal State-dependent IPR 0.71 0.08 0 0 0 0.12 0.25 0.41 0.46 0.19 0.13 0.96 0.028 34.20 0.0204 692.1

Table 1. Optimal Patent Length in Quick Catch-up Regime

43

= 1:05; = 0:35 B = 0:1; n = 1; #n = 1 1 2 3 4 5

x 1 x0 x1 0 1 2

! Researcher ratio ln C (0) g Welfare

Optimal Uniform IPR 0.11 0.11 0.11 0.11 0.11 0.27 0.14 0.15 0.16 0.19 0.14 0.90 0.055 34.31 0.023 695.3

Full IPR 0.02 0.02 0.02 0.02 0.02 0.75 0.99 1.10 0.02 0.03 0.03 0.56 0.150 31.31 0.025 636.3

Optimal State-dependent IPR 0.69 0.20 0.14 0.12 0.08 0.17 0.32 0.51 0.30 0.15 0.10 0.90 0.059 34.57 0.025 701.2

Table 2. Optimal Patent Length in Slow Catch-up Regime

= 1:05; = 0:35 B = 0:1; n = 0:02; #n = 1 1 2 3 4 5

v0

v

3

x 1 x0 x1 0 1 2

! Researcher ratio ln C (0) g Welfare

Optimal Uniform IPR 1.61 1.61 1.61 1.61 1.61 2.9 0.27 0.39 0.43 0.21 0.25 0.20 0.94 0.043 34.13 0.021 690.9

Full IPR 1 1 1 1 1 10.1 0.75 0.99 1.10 0.02 0.03 0.03 0.56 0.150 31.31 0.025 636.3

Optimal State-dependent IPR 0 1.54 2.45 2.92 3.32 3.2 0.31 0.45 0.46 0.18 0.20 0.12 0.91 0.071 34.47 0.025 699.3

Table 3. Licensing in Slow Catch-up Regime 44

= 1:05; = 0:35 B = 0:1; n = 0:02; n = 1 #1 #2 #3 #4 #5 v1 v 3 xc 1 xf 1 x0 x1 0 1 2

! Researcher ratio ln C (0) g Welfare

Optimal Uniform IPR 14 14 14 14 14 2.7 0.15 0.23 0.30 0.30 0.14 0.42 0.22 0.95 0.028 35.48 0.026 720.0

Full IPR 1 1 1 1 1 21.4 0.75 0 0.99 1.10 0.02 0.03 0.03 0.56 0.150 31.31 0.025 636.3

Optimal State-dependent IPR 0 18.1 31.3 36.6 43.7 4.1 0.14 0.33 0.29 0.39 0.12 0.35 0.17 0.94 0.058 36.17 0.031 735.9

Table 4. Leapfrogging in Slow Catch-up Regime

45

Figures R&D by Microsoft and Top-10 (Except Microsoft)

0

0

Microsoft 40 80

7 14 Top-10 (Except Microsoft)

21

120

Relative to the Sector Average

1985

1990

1995 Fiscal Year

2000

Microsoft

Sector: NAICS-511210 (Software)

2005

Top-10

Figure 1

Firm Values vn 40

30

20 Uniform IPR State Dependent IPR

10

20

10

0

10

20

figure 2. Value Functions

46

Technology Gap 30

Innovation Rates xn 0.35 0.30 0.25 0.20 0.15 0.10

Uniform IPR State Dependent IPR

0.05 20

10

0

10

20

Technology Gap 30

Figure 3. R&D Efforts

Industry Shares

n

0.4 Uniform IPR State Dependent IPR

0.3

0.2

0.1

0.0 0

2

4

6

8

Figure 4. Industry Shares

47

Technology Gap 10