Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/statedependentinOOacem DEWEY HB3i .M415 3M Massachusetts Institute of Technology Department of Economics Working Paper Series STATE-DEPENDENT INTELLECTUAL PROPERTY RIGHTS POLICY Daron Acemoglu Ufuk Akcigit Working Paper 06-34 December 11, 2006 Room E52-251 50 Memorial Drive Cambridge, MA 021 42 This paper can be downloaded without charge from the Social Science Research Network Paper Collection at http://ssrn.com/abstract=951 788 MASSACHUSETTS INSTITUTE OF TECHNOLOGY JAN C 8 2007 LIBRARIES State-Dependent Intellectual Property Rights Policy" Daron Acemoglu Ufuk Akcigit MIT MIT December 2006 Abstract What form economic growth? Should technological followers be able to license the products of technological leaders? Should of intellectual property rights (IPR) policy contributes to a company with a large technological lead receive the same IPR protection as a company with a more limited lead? We develop a general equilibrium framework to investigate these The economy many and firms engaged in cumulative (step-bystep) innovation. IPR policy regulates whether followers in an industry can copy the technology of the leader and also how much they have to pay to license past innovations. With full patent protection, followers can catch up to the leader in their industry either by making the same innovation(s) themselves or by making some pre-specified payments to the technological questions. consists of industries leaders. We prove the existence of a steady-state We equilibrium and characterize some of then quantitatively investigate the implications of different types of equilibrium growth rate. growth rate in the The two major IPR its properties. policy on the results of this exercise are as follows. First, the standard models used in the (growth) literature can be improved significantly by introducing a simple form of licensing. Second, we show that full patent protection is not optimal from the viewpoint of maximizing the growth rate of the economy and that the growth- maximizing policy involves state-dependent IPR protection, providing greater protection to technological leaders that are further ahead than those that are close to their followers. This form of the growth-maximizing policy is a result of the "trickle-down" effect, which implies that providing greater protection to firms that are further ahead of their followers than a certain threshold increases the R&D incentives also for all technological leaders that are less advanced than this threshold. Keywords: competition, economic growth, endogenous growth, industry structure, inno- vation, intellectual property rights, licensing, patents, research JEL and development, trickle-down. classification: 031, 034, 041, L16. *We thank Science Days, conference and seminar participants at the Canadian Institute of Advanced Research, European MIT, National Bureau of Economic Research, Toulouse Information Technology Network, and Philippe Aghion and particularly Suzanne Scotchmer for useful comments. Financial support from the Toulouse Information Technology Network is gratefully acknowledged. 1 Introduction How should the intellectual property rights of a company be protected? Should a firm with a large technological lead receive the same intellectual property rights (IPR) protection as a company with a more limited technological lead? These questions are central to sions of patent policy. A recent ruling of the Microsoft to share secret information about York Times, December 22, 2004). There is European Commission, its many discus- example, has required for products with other software companies (New a similar debate about whether Apple should make iPod's code available to competitors that are producing complementary products. Central to the debate about Microsoft and iPod the fact that they have a substantial technological is lead over their rivals. This implies that the analysis of emic questions requires a framework By policy. state-dependent IPR many for the analysis of state-dependent policy, we mean a policy that patent/IPR protection makes the extent or intellectual property rights protection conditional on the technology firms in the industry. Existing first of patent gap between different work has investigated the optimal length and breadth of patents assuming an IPR policy that does not allow a and acad- of the relevant policy for licensing and is uniform. In this paper, we make attempt to develop a framework that incorporates both licensing of existing patents and state-dependent Our IPR policies. basic 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 (typically two) firms number of R&D in engage in price competition within an industry and undertake order to improve the quality of their product. The technology gap between the firms determines the extent of the monopoly power of the leader, and hence the price markups and profits. The purpose of R&D by the follower Schumpeterian models of innovation, Grossman and Helpman, is e.g., (e.g., up and surpass the leader standard (as in Reinganum, 1981, 1985, Aghion and Howitt, 1992, 1991), while the purpose of competition of the follower and increase general to catch its R&D markup and by the leader profits. As is to escape the in racing-type models in 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. 1 One may expect that full patent protection may be suboptimal in a world of step-by-step competition; by stochastically or deterministically allowing the follower Aghion, Bloom, Blundell, Griffith and Howitt (2005) provide empirical evidence that there is greater R&D where there is a smaller technological gap between firms. See O'Donoghue, Scotchmer and in British industries Thisse (1998) for a discussion of how patent life may come to an end because of related innovations. to use the innovations of the technological leader, the likelihood of relatively small gap between and leaders may less and thus the amount of R&D may be followers, further conjecture that state-dependent IPR policy raised. may also 2 Based on this intuition, one be useful and should provide protection to firms that are technologically more advanced relative to their competitors. There are two problems with this intuition. First, existing growth models (including those mentioned above) do not allow licensing of the leading-edge technology and the presence of licensing changes the trade-offs underlying the above intuition. derived from models with uniform of state-dependent To IPR policy on IPR policy, R&D 3 Second, the above intuition without a systematic study of the implications incentives. we construct a general equilibrium model with investigate these issues systematically, step-by-step innovation, potential licensing of patents and state-dependent model economy, each firm can climb the technology ladder via three by "catch-up R&D," that leader; (ii) by "frontier R&D is, R&D," that leader for a pre-specified license technological leader. but of, is is, policy. In our different methods: (i) building on the patented innovations of the technological and feature essential for understanding The combination IPR investments applied to a variant of the technology of the fee; The second is (iii) is as a result of the expiration of the patent of the not present in any growth model that we are aware how IPR policy works in practice. of these policies allows a variety of different policy regimes. The first is full patent protection with no licensing, which corresponds to the environment assumed in existing growth models (e.g., Aghion, Harris, Howitt and Vickers, 2001) and provides full (indefinite) patent protection to technological leaders, but does not allow any licensing agreements the license fees to infinity). The second is full patent protection with licensing, to regime as pay a patent "full The third regime 2 3 than the We is from the use of the leading- uniform imperfect patent protection, which deviates from the previous two benchmarks by allowing either expiration of patents and/or fees that are less fee. patent protection" because patents never expire and followers have fee equal to the gain in net present value resulting edge technology. 4 sets which allows technological followers to build on the leading-edge technology in return for a license refer to this (it full benefit to the follower. The license adjective uniform indicates that is conjectured, for example, in Aghion, Harris, Howitt and Vickers (2001). See Scotchmer (2005) for the importance of incorporating these types of licensing agreements into models This of innovation. 4 The alternative would be for the followers to pay a fee equal to the damage to the technological leader, and further innovation. In practice, licensing fees or patent infringement fees reflect both the benefits to the firm using the knowledge and the damage to the original inventor (see, e.g., Scotchmer, 2005). In our analysis, we allow the license fees to be set at any level, thus incorporating both possibilities. resulting from the loss of the leader's monopoly power after their licensing in this policy environment industries are treated identically irrespective of the technology all gap between the leader and the The follower. state- dependent imperfect patent protection, and most interesting policy regime final which deviates from patent protection as a full function of the technology gap between the leader and the follower in the industry allows technologically more advanced more or firms to receive policy regimes captures a different conceptualization of Note however that the right. We first last regime is IPR less protection) policy and is is . Each (i.e., it of these interesting in its own general enough to nest the other three. prove the existence of a stationary (steady-state) equilibrium under any of these policy regimes and characterize a number example, we prove that with uniform IPR For of features of the equilibrium analytically. R&D policy, investments decline when the gap between the leader and the follower increases. We then turn to a quantitative investigation of growth-maximizing by providing a number of simulations different forms of 1. IPR policies. 5 Our of the equilibrium for plausible ( "optimal" ) IPR policy, parameter values and for quantitative investigation leads to two major results: Allowing for licensing of patents increases the equilibrium growth rate of the economy significantly. Intuitively, without such licensing, a large part of the R&D effort goes to R&D does not contribute to the growth rate of the economy. Licensing implies that R&D by all firms — not just the leaders — contributes to growth and also increases the R&D incentives of followers. In our benchmark parameterization, duplication, and followers' allowing for licensing increases the steady-state equilibrium growth rate of the economy from 1.86% to 2.58% per annum. 2. Growth-maximizing IPR policy centive effect of relaxing IPR is state dependent. protection on R&D, In particular, because of the disin- uniform IPR policy (either by ma- nipulating license fees or the duration of patents) has minimal effect on the equilibrium growth tion rate. In contrast, state-dependent and growth state-dependent in the IPR economy. For example, " policy can significantly increase innova- in our baseline parameterization, optimal policy increases the growth rate to 2.96% relative to the growth rate of 2.63% under uniform of IPR IPR policy. More important than growth-maximizing state-dependent IPR policy is its form. the quantitative effect We find in all cases that Throughout, we simplify terminology by using the terms "optimal" and "growth-maximizing" interchangeis well known that welfare-maximizing and growth-maximizing policies need not coincide, and in fact, in models with quality competition, the equilibrium may involve excessive innovation and growth (e.g., Aghion and Howitt, 1992). Nevertheless, when the discount rate is small, welfare and growth-maximizing policies will be close, which justifies our use of the term "optimal" in this context. ably. It optimal state-dependent IPR policy provides greater protection to technological leaders that are further ahead than those that have only a small lead relative to their followers. The reason for this result is the trickle-down of R&D incentives. In particular, particular state for the leader (say being n* steps ahead of the follower) perform this increases the incentives to ahead, but for all leaders with a lead of size can be powerful enough to increase nological leaders relative to the the IPR R&D is very profitable, not only for leaders that are n* n < n* — 1. when a — 1 steps Notably, the trickle-down effect —rather than reduce—the R&D investments by tech- benchmark with full patent protection. Thus, reducing protection of leaders that are few steps ahead while offering a high degree of protection to firms that are sufficiently advanced relative to their competitors increases R&D incentives. Consequently, in contrast to existing models, imperfect state-dependent IPR can increase R&D investments relative to full protection. Moreover, contrary to the IPR conjecture above, the trickle-down effect implies that the optimal state-dependent policy should provide greater protection to firms that have a larger technological lead. Overall, our analysis illustrates the different effects of IPR policy on the equilibrium growth rate of an economy. It shows that allowing for licensing can increase growth significantly that optimal state-dependent relative to uniform policies could IPR IPR policy. policy can also increase the growth rate of an and economy Consequently, the benefits of considering a rich set of IPR be substantial in practice. Our analysis also suggests that the structure of optimal IPR may depend on the equilibrium industry structure (which determines the technology gap between leaders and followers as a function of the underlying technology of the industry). more IPR detailed analysis of the relationship between industry structure policy is Our paper atures. The an area is and the optimal form of for future research. a contribution both to the industrial organization IPR and growth protection and the endogenous growth literatures though monopoly power 1981, Tirole, 1988, liter- emphasize that ex-post monopoly rents and thus patents are central to generate the ex-ante investments in progress, even A also creates distortions (e.g., R&D and technological Arrow, 1962, Reinganum, 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). 6 Much of the literature discusses the trade-off between these two forces to determine the optimal length and breadth of patents. For example, Klemperer (1990) and Gilbert and Shapiro J Boldrin and Levine (2001, 2004) or Quah (2003) argue that patent systems are not necessary for innovation. . in order to provide (1990) show that optimal patents should have a long duration R&D, but a narrow breadth so as to limit monopoly distortions. A number inducement to of other papers, for example, Gallini (1992) and Gallini and Scotchmer (2002), reach opposite conclusions. Another literature, including the seminal branch of the interesting papers (2001), adopts a paper by Scotchmer (1999) and the recent by Llobet, Hopenhayn and Mitchell (2001) and Hopenhayn and Mitchell mechanism design approach to the determination of the optimal patent intellectual property rights protection system. and For example, Scotchmer (1999) derives the patent renewal system as an optimal mechanism in an environment where the cost and value of different projects are unobserved and the main problem is to decide which projects should go ahead. To the best of our knowledge, no other paper in the literature has considered state- dependent patent or IPR policy, which is the focus of the current paper. As a first attempt, we only look at state-dependent patent length and license fees (though similar ideas can be applied to an investigation of the gains from Our paper is most making the breadth closely related to and extends the of patent awards state-dependent) results of Aghion, Harris and Vickers (1997) and Aghion, Harris, Howitt and Vickers (2001) on endogenous growth with step-by step innovation. ' Although our model builds on these papers, of significant ways. we First, specified license fee for building show that such is from them they both compete for scarce labor. 8 IPR firms. We on the equilibrium growth rate of the economy. IPR policy that can be state dependent. Third, a general equilibrium interaction between production and general model (with or without a number in on the leading-edge technology developed by other licensing has significant effects economy there also differs allow licensing agreements whereby followers can pay a pre- Second, our economy incorporates a general in our it R&D, since we provide a number of analytical results for the policy), while Aghion, Harris, Howitt and Vickers (2001) Finally, focus on the case where innovations are either "drastic" (so that the leader never undertakes R&D) or very small. They also do not prove the existence of a steady state or characterize the properties of the equilibrium in this class of economies. Finally, our results are also related to the literature Fudenberg, Gilbert, Stiglitz on tournaments and example, and Tirole (1983), Harris and Vickers (1985, 1987), Choi (1991), Budd, Harris and Vickers (1993), Taylor (1995), Fullerton and McAfee 7 races, for (1999), Baye and Hoppe Segal and Whinston (2005) analyze the impact of anti-trust policy on economic growth in a related model of step-by-step innovation. 8 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 significantly complicate the analysis and it is still possible to characterize the steady-state equilibrium. and Moscarini and Squintani (2003). (2004). This literature considers the impact of or exogenous prizes on effort in tournaments, races or state-dependent IPR The In terms of this literature, policy can be thought of as state-dependent handicapping of different players (where the state variable To the best R&D contests. endogenous is the gap between the two players in a dynamic tournament). of our knowledge, these types of schemes have not been considered in this literature. rest of the paper organized as follows. is Section 2 presents the basic environment. IPR Section 3 characterizes the equilibrium under uniform results to the case in which IPR policy is state dependent. Section 5 quantitatively evaluates the implications of various different types of IPR policy regimes characterizes the growth-maximizing state-dependent the Appendix contains the proofs of Section 4 extends these policy. IPR on economic growth and policies. Section 6 concludes, while the results stated in the text. all Model 2 We now The describe the basic environment. different policy regimes is characterization of the equilibrium under the presented in the next section. Preferences and Technology 2.1 Consider the following continuous time economy with a unique populated by a continuum of supply individuals, each with 1 unit of labor 1 inelastically. Preferences at time t are given good. final The economy is endowment, which they by /oo E where Ej denotes expectations at date t. The exp (-p(s-t)) log C{s)ds, t at time t, p > the discount rate and is is Y (t) be the total production of the closed and the final C (t) = Y (t). good is , is good final The standard Euler equation from this equation defines g the interest rate at date t. and the discount rate at time used only for consumption . ^*)3 where C (t) is consumption logarithmic preferences in (1) facilitate the analysis, since they imply a simple relationship between the interest rate, growth rate Let (1) (t) C(t) cw as the = (1) Y(t) yW t. (i.e., We there (see (2) below). assume that the economy is no investment), so that then implies that = r (t)_p ' ' growth rate of consumption and thus output, and r (2) (t) . The final Y good produced using a continuum is 1 of intermediate goods according to the Cobb-Douglas production function \nY(t)= f Jo where y the output of jth intermediate at time assume that there is free entry into the final have the following demand Each intermediate infinitely-lived firms. compete a firms at time j 6 [0, 1] We Y (t) y^ = p-{jjy la Bertrand. where k(j,t) is comes two in difference (j, t) We . These assumptions, final good producer Firm i = 1 (4) different varieties, each produced by one of two varieties are perfect substitutes and these or 2 in industry j has the following technology = qi {j,t)h(j,t) (5) the employment level of the firm and qi(j,t) The only t. imply that each by p ^e[0,l). assume that these two 9 (3), sector. t for intermediates y(j,t) time Throughout, we take the price of the t. good production together with the Cobb-Douglas production function will (3) good as the numeraire and denote the price of intermediate j final also (j, t) is \ny(j,t)dj, between two firms is their technology, is its level of which will technology at be determined endogenously. Each consumer in the economy holds a balanced sequently, the objective function of each firm The production function producing intermediate j is for intermediate for firm i to portfolio of the shares of maximize expected goods, and industry j (5), at time all firms. Con- profits. implies that the marginal cost of t is MGiU,t)-=^r where w (t) When is the wage rate in the economy at time this causes (6) t. no confusion, we denote the technological leader in each industry by i and the follower by —i, so that we have: Qi {j, t) > q-i (j, t) Alternatively, we can assume that these two varieties are imperfect substitutes, output of intermediate j is given by y with a > 1. (j, t) = [<pyi ti, t) " + (i - <p) yz (j, t) for example, so that the • This generalization has no effect on any of our qualitative results, but increases notation. therefore prefer to focus on the case where the two varieties are perfect substitutes. We Bertrand competition between the two firms implies that all intermediates will be supplied by 10 the leader at the "limit" price: w &(3,t)= Equation (4) (t) );.y n (7) then implies the following demand for intermediates: v(M-.*ffiY®. R&D (8) and IPR Policy 2.2 Technology, R&D by the leader or the follower stochastically leads to innovation. the leader innovates, The follower, its > technology improves by a factor A on the other hand, can undertake R&D nology or to improve over the frontier technology. 11 The We assume that when 1. to catch up with the first possibility, frontier tech- catch- up R&D, be thought of as discovering an alternative way of performing the same task as the We leading-edge technology (applied to the follower's variant of the product). because this innovation R&D efforts, it is for the follower's variant of the can .current assume that product and results from its own does not constitute infringement of the patent of the leader, and the follower does not have to make any payments to the technological leader in the industry. Therefore, the follower chooses the first possibility, it will leader (as applied to own patent fee. its The have to retrace the steps of the technological variant of the product), but in return, For follower firm —i in industry j at time a -i U, alternative, frontier leading-edge technology. R&D, If this t) = t, we denote R&D it will not have to pay a this type of R&D by 0. involves followers building type of if on and improving the current succeeds, the follower will have improved the leading-edge technology using the patented knowledge of the technological leader, and thus have to pay a pre-specified patent will 10 If fee to the leader. 12 We the leader were to charge a higher price, then the market would be captured by the follower earning positive profits. A lower price can always be increased while making sure that prefer the intermediate supplied by the leader i always gains by increasing A it third possibility is first making the for the follower to is at good producers even demand curve if infinite, is still the latter were the leader price given in (7) the unique equilibrium price. climb the technology ladder step- by-step, meaning that, for example, some technology rung discover technology n-i 3 "slow catch-up" in —i and the follower itself is at n- j (t) < n l} (t) — 1, have investigated this type of environment with a previous version of the paper. Since the general results are similar, we do not discuss this the current leader must its price, all final rather than that by the follower supplied at marginal cost. Since the monopoly price with the unit elastic when specify the patent fees below. (t) + 1, n;, (t) et cetera. % We variation to save space. 12 Clearly, the follower will never license the technology of the leader for production purposes, since this lead to Bertrand competition and zero ex post profits for both parties. would This strategy denoted by is = a-i(jyt) Throughout, we allow a_; G (j, t) [0, 1] for !• mathematical convenience, thus a should be R&D by the follower. R&D by the follower, inter- preted as the probability of frontier R&D by the leader, catch-up and frontier R&D by the follower have different costs and success probabilities. Nevertheless, we simplify the analysis by may assuming that all In particular, in three types of cases, all R&D have the same costs and the same probability of success. we assume that innovations follow a controlled Poisson process, with R&D investments. Each firm (in every industry) has access to the arrival rate determined by the following R&D technology: XiO,t) = F{hi(j,t)), where x, (j, t) hired by firm is i the flow rate of innovation at time in industry j to work in the R&D (9) and t hi process at t. (j, t) is assume that < F'(O) < (0) (0, labor oo implies that there w is (t), is R&D the cost for and the assumptions on satisfies G' (•) > 0, x* is = (•) > for all no Inada condition when is (j, t) At + 0, F" h > r hi (j, t) o (At). () h. — 0. < 0, The The an upper bound on the flow rate of not essential but simplifies the proofs). Recalling that the wage rate for is therefore G(i G" (•) F > (j\t)) i w (t) G (xj 0, G" (0) > (j,i)) where = F-\( ? .0",*)), immediately imply that G is and lim x _ s G' x is oo) such that F' (h) assumption, on the other hand, ensures that there innovation (which is twice continuously differentiable and satisfies F' is oo and that there exists h G assumption that F' last F of workers This specification implies that within a time interval of At, the probability of innovation for this firm We number the (10) twice continuously differentiable and (x) = = F(h) oo, where (11) the maximal flow rate of innovation (with h defined above). We next describe the evolution of technologies within each industry. Suppose that leader in industry j at time t i has a technology level of qi (j,t) and that the follower — i's technology at time q- i (j,t) = t = ^, (12) n \ -<^, (13) X n is where mj > (t) n-ij and (t) the follower in industry If j. n-ij riij (i), (t) We refer to rij j. £ Z+ denote the technology rungs = riij (t) —n^ij (t) (t) of the leader as the technology gap in industry the leader undertakes an innovation within a time interval of At, then increases to qi (j, t + At) A nyt+1 and technology gap — more innovations within the probability of two or rises to interval and + rij (t At = At) its technology rij (t) + 1 (the be o(At), where o(At) will represents terms that satisfy lim& t ^o o (At) /At). When At, then the follower it catches successful in catch-up is up with the and leader R&D R&D technology, so and pays the license fee (i.e., a_j — X ^, In contrast, 0. (j, t) = 1), then if At) = rijt+At = (j,t and the technology gap variable becomes is successful surpasses the leading-edge it + X 1 n ( ^ +1 an d from this point and —i are swapped, since the previous follower now becomes the In addition to catching R&D, the follower we have q. l i within the interval 0) n + At) = and the technology gap variable becomes njt +At in frontier = a_, (j,t) technology improves to its q^(j,t (i.e., followers onwards, the labels leader). up with or surpassing the technology frontier with their can also copy the technology frontier because IPR policy is own such that some patents expire. In particular, we assume that patents expire at some policy-determined Poisson rate q- l 77, (j,t and + after expiration followers At) n = \ »K This description makes the patent; it license fees. (ii) can costlessly copy the frontier technology, jumping to 13 clear that there are In our two aspects to IPR 7] for all t > r\ (n) = r\ n < 00 technological leader that is full 13 the length of two functions: : N -> R+ 0, C(t) Here (i) most general policy regime, we allow both of these to be state dependent, so they are represented by the following and policy: is is : N -» R+ U {+00} . the flow rate at which the patent protection n steps ahead of the follower. When 7? n = is 0, this removed from a implies that there protection at technology gap n, in the sense that patent protection will never be removed. Alternative modeling assumptions on IPR policy, such as a fixed patent length of T> innovation, are not tractable, since they introduce a form of delayed-differential equations. 10 from the time of In contrast, gap n r\ —» n oo implies that patent protection reached. Similarly, £ (n, is = t) Cn {t) removed immediately once technology is denotes the patent fee that a follower has to pay in order to build upon the innovation of the technological leader, industry £ (£) is = {C n steps. x (t) , (2 14 (*) i Our formulation imposes that ••} license fees should not is This a function of time. remain constant. Below, we when the technology gap ={77^ 772,'...} 77 in the time-invariant, while is natural, since in a growing economy, is £ grows at the same rate will require that as aggregate output in the economy. When £n (t) — 0, there technology at zero cost. is prohibitively costly. protection is imperfect. 15 is no protection because followers can license the leading-edge In contrast, when Note however that (n (t) = n < 00 does not necessarily imply that patent C, what In particular, in follows is and-neck being at a technology gap of 0) as regime as uniform IPR intellectual property protection law treats all if both firms also n > n. assume that there Given this specification, exists 77 "full and £ will interpret (i.e., rj uni protection". (t) and industries gap between the leader and the follower We we a situation in which equal to the net extra gain from surpassing the leader rather than being neck- the license fee (i.e., 00, licensing the leading-edge technology 16 We also refer to a policy are constant functions, meaning that identically irrespective of the technology = {77, 77, ...} some n < 00 such that we can now write the law r) of n and C um = "q n (t) and ( n = {( (t) (t) = , (n £ (t) (t) , ...}). for all motion of the technology gap in 14 Throughout, we assume that £ is a policy choice and firms cannot contract around it. An alternative approach would be to allow firms to bargain over the level of license fees. In this case, it is plausible to presume that the legally-specified infringement penalties or license fees will affect the equilibrium in the bargaining game, so the effect of policies we investigate would still be present. We do not allow bargaining between firms over the license fees in order to simplify the analysis. 11 Throughout, we interpret f„ (t) = as £n (t) '= e with e J. 0, so that followers continue not to license the new technology without innovation (recall the comment in footnote 12). 16 In other words, we interpret "full protection" to correspond to a situation in which £n (t) = V\ (t) — Vo (t), one step ahead of its rival and Vo is the value of a firm that is neck-and-neck with its rival (naturally, £ n (t) > Vi (t) — Vo (t) would also correspond to full protection, since in this case no follower would choose to license). Alternatively, full protection could be interpreted as corresponding to the case in which the follower pays a license fee equal to the loss of profits that it causes for the technology leader (see Scotchmer, 2005). In our model, this would correspond to £n (t) = Vo (t) — V-i (t), where V-i is the net present value of a firm that is one step behind the technology leader. This expression applies, since without licensing the follower would have made its own innovation, in which case the leader would have fallen to the state of a neck-and-neck firm rather than being surpassed by one step. In all equilibria we compute below, we find that the second amount is significantly less than the first, thus our interpretation of full protection licensing fee is large enough to cover both possibilities. In any case, what value of £ is designated as "full protection" does not have any bearing on our formal analysis, since we characterize the equilibrium for any £ and then find the growth-maximizing policy sequence. where Vi refers to the net present value of a firm that is 11 industry j as follows: nj n-j (t + At) = (t) + < nj (t) with probability ((1 with probability i -<*-* _ fx + a-i At Xi (j, t) with probability 1 (jY*)) ^ . t) x_, (j, t) x -i -Oi*) + x _. ( j, : ^ t) t)At + Vn + j (i)) ^ + o (At) At+ °(At) Ai _ N ( Ai )) (14) Here o (At) again represents second-order terms, in particular, the probabilities of one innovations within an interval of length At. The terms rates of innovation is is by the leader and the follower; a_i up with a leader or surpass trying to catch and rj allowed to copy the technology of a leader that a-i (j,t) — leader is successful (flow rate Xj (j,t)) or 2.3 Profits We of it; (j, t) n ./ is t \ € is Xj (j,t) [0, 1] if the follower rij (t) is expenditures and license fees). Profits of leader w(t) i steps ahead. rij (t) to Intuitively, rij (t) + rij (t) = n^ (£)— n_jj (t) is the 1 if either (i.e., the profits exclusive in industry j at time t are Y(t) qi(j,t)J Piti't) l-A-M^F^) where when successful (flow rate x_j (j,t)). w(t) \ q-i(j,t) the flow denotes whether the follower next write the instantaneous "operating" profits for the leader R&D (j,t) are the flow rate at which the follower the technology gap in industry j increases from 1, and X-i more than ( the technology gap in industry j at time t. The first line 15 ) simply uses the definition of operating profits as price minus marginal cost times quantity sold. The = w (t) /q-i (j, t) second line uses the fact that the equilibrium limit price of firm as given and is (13). by (7), The and the is p final equality uses the definitions of qi (j,t) 7; (j, t) and q-i (J,t) from (12) expression in (15) also implies that there will be zero profits in an industry that neck-and-neck, profits, since i i.e., in industries with rij (t) = 0. Also clearly, followers always make zero they have no sales. The Cobb-Douglas aggregate production function profits (15), since it implies that profits only and aggregate output. This in (3) is responsible for the form of the depend on the technology gap of the industry will simplify the analysis below by making the technology gap each industry the only industry-specific payoff-relevant state variable. 12 in . The objective function of each firm profits (operating profits this, each firm output (t)] to maximize the net present discounted value of net expenditures and plus or minus patent take the sequence of interest rates, will [Y levels, R&D minus is sequence of wages, [w t>0 the (t)] , and policies as given. 2.4 Equilibrium t > [r(i)] , t>0 , R&D the fees). In doing the sequence of aggregate decisions of all other firms = {/i n (£)}^Lo denote the distribution of industries over different technology gaps, with (£) denotes the fraction of industries in which the firms are S^Lo^n (*) = 1- For example, Let fi (t) /j, neck-and-neck at time t. Throughout, we focus on Markov Perfect Equilibria (MPE), where strategies are only functions of the payoff-relevant state variables. the dependence on industry leader that n is us denote the by fn(*) = ( thus j, refer to R&D a_„ and x_„ steps ahead and by list we of decisions by the leader x n{t),Pi{j,t),yi{j,t)) and £_„ for 17 decisions by This allows us to drop xn a follower that is for the technological n steps behind. Let and follower with technology gap n (t) = {a- n ,X- n {t))P (t) indicate the whole sequence of decisions at every state, £ (£) = at time Throughout, £ {£„ {t)}'^= _ 00 We t will define an allocation as follows: Definition 1 (Allocation) Let cation is (rj, [C(*)]t>o) ^e a sequence of decisions for a leader that a sequence of R&D decisions for a follower that sequence of wage rates [w (t)] t>0 , is ^ e ^^ is n = n — P°^ C V sequences. 0,1,.. .,oo step ahead, l,...,oo step behind, and a sequence of industry Then an allo- [£ n (t)] t>0 [£-7iW]t>o j , a distributions over technology gaps [MOW For given IPR sequences r\ payoff- relevant state variables, and [£ (i).]t>o> ^PE strategies, which are only a function of the can be represented as follows x a : : Zx R2_-> R+, Z_\ {0} x R2.-» [0, 1] 17 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 firms in the industry, in most industries there are many more firms and also many potential entrants, making collusion more difficult. Throughout, we assume that there are only two firms to keep the model tractable. 1 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, riij (t). With a slight abuse of notation, throughout we suppress this dependence, since their product p; (j,t)yi {j,t) and the resulting profits for the firm, (15), are independent of n^ (t), and only the technology gap, n 3 (£), matters for profits, R&D, aggregate output and economic growth. 13 R&D The first it is a leader in an industry) with an function represents the when decision of a firm (both n it is a follower and when step technology gap (n e Z), and the aggregate level of output and the wage ((Y,w) £ K+). The second function represents the decision of a follower a function of the (as R&D output and the wage in the economy) of whether to direct level of its to reaching or surpassing the leading-edge technology (or more precisely, the probability with which the follower will choose to undertake R&D to surpass the leading edge technology) . it represents Consequently, we have the following definition of equilibrium (below we will define a steady-state equilibrium in terms of a solution to maximization problems): Definition 2 (Equilibrium) Given an IPR policy sequence by a sequence fect Equilibrium, is given and [y* (j,t)] [£* t>0 implied by taking aggregate output [Y* the R&D policies given by (3); and (t)] ,x* wages [w* , of other firms [a* (iv) the labor [a* (t) i.e., t>0 w* , (i) ,x* (t)] (t)] (t)] t>0 market clears , {t)] t>0 such that R&D policies (i) Markov Per\pl {j, t)] t>0 [a* (t) x* , {t)] t>0 maximizes the expected profits of firms government policy [C(i)] t > (77, as given; (Hi) aggregate output [Y* at all times given the Since only the technological leader produces, labor (t) = n can be expressed ) (i)] wage sequence [w* an d t>0 (t)] is t>0 . demand X = n (t) ^ neZ + for w(t) In addition, there industries. R&D Using is (9) demand for labor and the coming for definition of the K G R&D (16) from both followers and leaders function, Note that refers to the (t)) + G(x- n (t)) if neN if n= its X- n (t)) of a follower in an industry with a technology gap of n. (t) refers to the optimal choice of a_ n (t) R&D 6 effort of [0, 1]). The for R&D coming from the two a follower that is n steps Moreover, this expression takes into account that in an industry with neck-and-neck competition, be twice the demand (17) , 2G{x demand in this expression, behind (conditional on G(x n in all we can express industry demands (t)={ { < with technology gap . labor as f where n in industry j as In (t) for Y* (8); (ii) t>0 t>0 , [C(0]t>o)> a The Labor Market 2.5 nj (t) and t>0 satisfy (7) {t)] are best responses to themselves, [£* (t) (77, i.e., with n = 0, there will firms. labor market clearing condition can then be expressed as: 1 !>£>„(*) u (t) X- + G(x n (t)) + G(x- n (t)) n=0 14 (18) and is u> (t) > 0, with complementary slackness, where the labor share at time supply is equal to 1, then the wage rate, this will never The t. and the demand cannot exceed w (£), be the case this and thus the labor share, w in equilibrium). production (the terms with for labor market clearing condition, (18), uses the fact that total neck-and-neck industries (2G (xq from leaders and followers in the co demand If falls short of 1, have to be equal to zero (though (t), The right-hand side of (18) consists of the demand R&D workers from the for R&D workers coming denominator), the demand for when n = (t)) amount. in other industries and the demand 0) (G (x n + G (x- n (£)) (£)) when n > 0). Defining the index of aggregate quality in this economy by the aggregate of the qualities in the different industries, i.e., lnQ(t)= [ \nq(j,t)dj, (20) Jo 19 the equilibrium wage can be written as: w{t) ^\ t (21) Steady State and the Value Functions 2.6 Let us now focus on steady-state (Markov Perfect) equilibria, where the distribution of in- dustries n(t) = {fi n (£)}^L *s stationary, economy, are constant over time. characterize a by = Q{t)\-^=° n definition, rate. number we have of its will establish the existence of properties. If the Y (t) — The two equations We Y^e 9 * 1 defined in (19) and g, the growth rate of the oj (t) economy in steady state at w (t) = woe 9 where u (t) — oj* for all t > 0. and also imply that is such an equilibrium and ', the parameters are such that the steady-state growth rate g* is g* is time t = 0, then the steady-state growth Throughout, we assume that positive but not large enough to violate the transversality conditions. This implies that net present values of each firm at all point in time will be finite and enable us to write the maximization problem of a leader that is n > steps ahead recursively. First note that given leader that is n an optimal policy x steps ahead at time exp 199 T Note that (16). Thus we t for a firm, the net present discounted value of a can be written (-r- (s - t)) [II (s) as: +Z(s)-w (s) G (x (a))] ds Y (t) = JQ \nq(j,t) {j,t)dj = L lniji (j,t) + In ^JA nj have In Y (t) = J [In qt (j, t) + In Y (t) — In w (t) — rij In A] dj. In and writing exp / n 3 I ; In Xdj = A _ ^"=° nM " (t) we obtain (21). , 15 dj, where the second equality uses Rearranging and canceling terms, where the operating profit at time IT (s) is paid) by a firm which s > the leader and is All variables are stochastic t. s > t, Z (s) the patent license fees received (or is w(s)G(x(s)) denotes the R&D expenditure at time and depend on the evolution of the technology gap within the industry. Next taking the equilibrium wage rate, w* (t), R&D policy of other firms, as given, this value can be written as: x*_ n (t) and and equilibrium a*_ n (t), 20 Vn (t) = max {[nn (i) - w*(t)G(x n (t))] At +o (At) (22) x n (t) {x n + -exp(-r(t + (r) part of this expression interval of length At. Vn +i (t) and Vo (t) ahead and a firm (l - Vn+1 At a*_ n (t)) x*_ n (t) (t + At) + o (At)) V The second is (l {a*_ n (t) x*_ n (t) - xn (t) At - rj n At the flow profits minus part is At + - o (At)) (T/_a + At) + At) + CJ At-o (At)) Vn (t + At) x*_ n (t) R&D (t expenditures during a time the continuation value after this interval has elapsed. are defined as net present discounted values for a leader that an industry that in {t > + first + o {At)) At) At) + The n At At + {t) is neck-and-neck (i.e., n = is The second 0). n + 1 steps part of the expression uses the fact that in a short time interval At, the probability of innovation by the leader is first line xn (t) At + o (At), where o (At) again denotes second-order terms. This explains the of the continuation value. For the remainder of the continuation value, note that the probability that the follower will catch up with the leader in particular, undertaking if a*_ n (t) R&D 1, this eventually will never (l a*_ n (t) = 1. in the third line, conditional — a*_ n (t)) x*_ n (t) fee (n . There are two differences between the second and third on success by the follower, a leader Finally, the last line applies when no R&D 20 Vn (t) from both sides, divide Clearly, this value function could be written for set the R&D main body line, lines; moves to the position (ii) patents continue to be enforced, so that the technology gap remains at Next, subtract + o (At); This explains the third of a follower rather than a neck-and-neck firm (V-\ instead of Vq); dependent patent At happen, since the follower would be not to catch up but to surpass the leader. which applies when (i) = is it receives the state- effort is successful n and steps. everything by At, and take the limit as At — » 0. any arbitrary sequence of R&D policies of other firms. We and ol„ (t), to reduce notation in the policies of other firms to their equilibrium values, x*_ n (t) of the paper. 16 . This yields: r(t)Vn (t)-Vn = (t) maxIp^-^WG^^+XnWlK+iW-KW] (23) x n (t) where V^ (i) . ((1 + (a*_„ (t) x*_ n (t) a*_ n (i)) x*_ n (t) denotes the derivative of present value of a firm that neZ + - + is n Vn (t) what n ) [V + [C„ - Vn (t) T/ (*) - V-i will also grow [*)] (t)]}, In steady state, the net at a constant rate g* for all Let us then define the normalized values as n € = YV (24) Z, which will be independent of time in steady state, follows we assume that will ensure the existence of (t) — vn . Similarly, in Cn(*) = tn vn i.e., up by GDP, so that license fees are also scaled f. which r, with respect to time. Vn{t) for all + Vn ) Vn (i), steps ahead, + Y ^ - a (stationary) steady-state equilibrium (the use of £ n (t) for the original patent fees and C n f° r the scaled-up fees should not cause any confusion). Using (24) and the fact that from value function (23) can be written pvn = max{ (l + where x*_ n is [(1 - (2), r (t) + rj the equilibrium value of steady-state labor share (while x n is n] + (x n ) [up R&D now xn + p, = max {-oj*G (x \v n+1 - v^ + the recursive form of the steady-state - vn ] a*_ n x*_ n [y-i explicitly chosen to ) (25) , by a follower that Similarly the value for neck-and-neck firms pvQ (t) g as: A" n ) - u*G a*_ n ) x*_ n — is -v n + C n ]} n for steps behind, neN, and ui* is the maximize vn ). is +x [v\ - v + x* [v-\ ] - v XQ ]} (26) , while the values for followers are given by pv- n = max {-ui*G (x- n + ) C — n )0> — n +a_„x_ n which takes into account that it will become the new if [vi - V- n - [(1 Cn ] - a_ n x- n + rjn + xn ) [V- n -l the follower decides to build leader, but will - [v - v- n U~n]} for (27) ] 71 <E N, upon the leading-edge technology have to pay the patent 17 ] fee Q n . , For neck-and-neck firms and followers, there are no instantaneous in (26) and and (27). In the in the latter case, former case this because neck-and-neck firms is because followers have no emphasize that, because of growth, the profits, sales. which is reflected marginal cost, sell at These normalized value functions effective discount rate is r(t) — g (t) = p rather than r(t). The maximization problems R&D rium policies, (o*,x*), immediately imply that any steady-state equilib- in (25)- (26) must satisfy: = € 1 if vi [0, 1] if vi = if Vl - (n > Cn = vo Cn < Vq vq (28) and < = m ax {G'-^ [Vn+1 Vn] u: 1 - a*_ n ),o\ [vq ) - (29) v- n ] + a*_ n [vi - v^ n - („] ,0 ur where G~ (•) is differentiable [Vl ~ l± = max^g- 1 x* Vq] UJ ,o|, ,0 \ G'~ l strictly concave, incremental value of moving to the next step by the normalized wage rate, u*, is less. be equal to is which The response economic force in leaders that are n + 1 1 here relaxing IPR greater max form of lower and when the > 0, whenever the R&D, these as R&D measured levels can operator. — vn , is the key For example, a policy that reduces the patent protection of r/ effect of relaxing IPR policy. n+1 or reducing Cn+i) — v n and x*n Q may encourage directly contributing to aggregate growth. x* +1 This corresponds to the effect. This positive incentive . further frontier R&D by the followers, Second and perhaps somewhat more paradoxically, lower protection for technological leaders that are — v n+ \ and being equation (30) shows, weaker patent protection in the First, as license fees (lower . wm make In contrast to existing models, however, policy can also create a positive incentive thus increasing v n+ 2 cost of to the increments in Values, v n+ \ steps ahead (by increasing has two components. G is twice continuously rates, the rc*'s, will increase steps ahead less profitable, thus reduce v n+ \ standard disincentive effect , and since also that since G' (0) taken care of by the this model. n + is Note of innovation rates, x* G function, continuously differentiable and strictly increasing. is These equations therefore imply that innovation zero, (31) , the inverse of the derivative of the and (30) We n+1 steps ahead will tend to reduce v n+ \ will see that this latter effect plays 18 an important role in the may IPR form of optimal state-dependent also create a beneficial composition effect; this decreasing sequence, which implies that x*_j 4 and 5 below). Weaker patent protection more industries into the neck-and-neck state R&D in the economy. 21 The optimal determined by the interplay of these three Given the equilibrium across states /j.* R&D ir* for n > 1 — protection v n }^Lo *s a (see Propositions the form of shorter patent lengths) will shift and potentially increase the equilibrium and structure level IPR because, typically, {v n+ i is higher than is (in Finally, relaxing policy. of IPR policy in this economy level of be will forces. decisions (a*, x*), the steady-state distribution of industries has to satisfy the following accounting identities: «+l + Z-n-l + Vn+l) Mn+l = <Mn for n e N ( ' 32 ) oo + llj + T]^ {x\ = 2x^0 + Yl a -n x -nVn, Ml (33) . n=l oo 2sSM0 = E ( i 1 - a ~n) X ~n + In) < (34) n.=l The first expression equates exit from state more step ahead or the follower catching n + 1 (which up surpassing the leader) to entry into the state for n making one more (which takes the form of a leader from the state equation, (33), performs the this state same accounting comes from innovation by 1, innovation) . The second taking into account that entry into two firms that are competing neck-and-neck. either of the innovation by a follower in any industry with The labor market for state with entry into this state, which comes from from state Finally, equation (34) equates exit takes the form of the leader going one n> 1. clearing condition in steady state can then be written as J_ + G{x 1>EX A n=0 k T * n) +G ( x *_ n ) and u* > (35) 0, with complementary slackness. The next proposition characterizes the steady-state growth rate. results in the paper, the proof of this proposition Proposition (/x*, 1 is As with all the other provided in the Appendix. Let the steady-state distribution of industries and R&D decisions be given by a*, x*), then the steady-state growth rate is g* = \n\ 2 V0 X + Yl & ( X« + a -n X -n) (36) 71=1 21 Weaker patent protection also creates a beneficial "level effect" by affecting equilibrium prices (as shown in equation (7) above) and by reallocating some of the workers engaged in "duplicative" R&D to production. 19 This proposition by three that the steady-state growth rate of the economy is determined factors: 1. R&D 2. The 3. Whether decisions of industries at different levels of technology gap, x* distribution of industries across different technology gaps, followers are undertaking = frontier, a* IPR clarifies R&D to catch up with the fj,* = {^^j^L.^. = {f^^^Lo- frontier or to surpass the {a*}"!.^. policy affects these three margins in different directions as illustrated by the discussion above. We now define a steady-state equilibrium in a more convenient form, which and derive some of the properties of the equilibrium. establish existence Definition 3 (Steady-State Equilibrium) Given an IPR policy librium is a tuple (32), (33) and be used to will v, (fx*, £), a steady-state equi- (77, a*, x*, u*, g*) such that the distribution of industries ll* satisfy (34), the values v= {v n }n=-oo satisfv ffi), (%6) and (2 V> th ^ R& D decisions a* and x* are given by (28), (29), (30) and (31), the steady-state labor share u* satisfies (35) and the steady-state growth We is given by (36). next provide a characterization of the steady-state equilibrium, starting case in which there is uniform IPR first with the policy. Uniform IPR Policy 3 Let us r) rate g* < first IPR policy, whereby um and £ um In this which we denote by r] assume that the only available policy option = 00 and Cn C (27) implies that the < n E N, °° f°r an problem is is uniform . identical for all v- n followers, so that = nn = case, V-\ for n G N. Consequently, (27) can be replaced with the following simpler equation: pv-i = max {-u*G{x-i) + X— ,a_ 1 [(l-a-i)x-i+ii\[vo-v-i] + a-iX-i[vi-v-i-£]}, (37) 1 implying optimal R&D decisions for xU = max JG'- all followers of the form m&X ([t = '° 1 ( V ~ l] J Vl - V~ l ~ Let us denote the sequence of value functions under uniform C]) ) IPR , 0] (38) . as {v n }'^L_ 1 . with a number of results that characterize the form of a steady-state equilibrium in omy. 20 We start this econ- Proposition 2 Consider a uniform v, a!_ l5 x*, u>*, g*) <(^z*, bounded and The next is n> (fi*-, (r) un1 uni £ , and suppose ^ and {vn }^=0 forms a vq shows that we can a finite ^ and a corresponding steady-state restrict attention to policy (r) un% c\ , um N Then, there exists n* £ v, a!_ 1; x*, u*, g*\. sequence of values: such that x* = for all n*. The next = {x^}^=1 proposition shows that x* a decreasing sequence, which implies that is technological leaders that are further ahead undertake less further R&D Proposition 4 Consider a uniform IPR librium with uniform IPR, x^ +1 -"A R&D and final R&D. markup comparison < R&D um and um £ rj N x*n for alln £ between the is policy (recall (15)). Then . in and moreover, £* +1 any steady-state < £* if x*n > The next proposition shows that neck-and-neck industries are "most R&D Proposition 5 Consider a uniform IPR Moreover, The if G'~ l ((l - A" 1 ) condition G'~ l ((l / {p Xq / (p in the steady-state equilibrium (see + > x\ and x^ policy r) and C > rj)) Remark um and um £ > A steady state with with there are no profits, thus it ui* < would then imply that there is 1, x*_j below and 1 1 is Then Xq . > x*_ > x\ and Xq > Lemma in the 1 (//*, (36) together no economic growth, = 1 (so with i.e., jig v, a*_ 1} x*, g* = R&D u>* , g*) with when uj* = 1, firms are neck-and-neck that all = and x*n 1 . Appendix). economically more interesting, since (j,q x*_ 1 v here ensures that there will be positive must be the case that and there are no markups). Equation = x\, so that without licensing, then x* 0, Next, we prove the existence of a steady-state equilibrium 1. 0. intensive". + r])) > — A -1 ) > equi- neck and-and-neck firms, Xq, and the levels of investments of followers and leaders in industries with a technology gap of n x\. < Intuitively, the benefits of are decreasing in the technology gap, since greater values of the technology gap translate into smaller increases in the equilibrium ui* < Then, v-\ that sequence converging to some positive value v^. Proposition 3 Consider a uniform IPR equilibrium policy a steady-state equilibrium. strictly increasing result IPR — for all n £ Z+ 0. Establishing the existence of a steady-state equilibrium in this economy is made compli- cated by the fact that the equilibrium allocation cannot be represented as a solution to a maximization problem. Instead, as emphasized by Definition taking the R&D 3, each firm maximizes its value decisions of other firms as given; thus an equilibrium corresponds to a set of 21 R&D and a labor share (wage decisions that are best responses to themselves clears the labor market. We rate) lu* that establish the existence of a steady-state equilibrium in a number of steps. First, note that each x n belongs to a compact interval Now of innovation defined in (11) above. (Markovian) steady-state strategies for all that given while its some z optimal we denote = {Co, R&D R&D result characterizes this [z], xn (z) Xn e letters and refer to their policy (r) um £ , policies of all other firms are given by z R&D uniquely determined, policies convex-valued for each z G A-\ [z] C what follows, elements by lowercase [z]. um = ^ and suppose that the labor share Then (<D,a_i,x). mization problem of an individual firm leads to a unique value function and optimal is problem and shows choices are given by a convex-valued correspondence. In Proposition 6 Consider a uniform IPR and the the maximal flow rate and a sequence (a_i,x) of a_i,x), the value function of an individual firm a_i (z) G A-\ letters, e.g., [0, 1] is other firms in the economy, and consider the dynamic and correspondences by uppercase sets where x u G a labor share fix Our next optimization problem of a single firm. [0, x], and [0,1] X [z] : { Z and upper hemi- continuous — 1} U Z+ v =1 in z (where [z] [0, v the [z] : x] = dynamic opti- {-1}UZ + — R + » are compact and {v n [z]}^__ 1 and X [] = {*«[«]}"_!;}. Now z 6, this xn [z] E let is us start with an arbitrary z mapped Xn [z]. into optimal From R&D = {Co, R&D decision sets we policies (d_i,x), equations (32), (33) and (34). Then we can Z = [0, a_i, x) 6 A-i calculate /x X [0, x]°°. From Proposition [z], where a_i 6 A-\ [a_i,x] = {p, n [a-i,x],}°L and [z] x 1] [z] and using rewrite the labor market clearing condition (35) as ,,im OjAl yT + G (x n u + G {x- n ) ) (39) where due to uniform IPR, <!> [z] = ((p (z) , A^i [z] , X x*_ n = [z]), which maps Z into x*_ 1 for all n > $: That $ maps Z into itself follows since z G Z Z 0. Next, define the mapping (correspondence) itself, that is, =J Z. (40) consists of a_i and the image of z under $ consists of a_i G [0, 1] and x G clearly in [0,1] (since the right-hand side is nonnegative 22 G [0, 1], x G [0,x]°°, [0, x]°° and w G and moreover, and bounded above by 1). [0, 1], (39) is Finally, from Proposition A-\ 6, [z] upper hemi-continuous in and z, Proposition 7 Consider G" _1 ) IPR Then a 0. > > then g* 0, Remark policy a trivial equilibrium in which x* which imply that Xq = T, jjLq — _1 ((l for all > x*n for all n^O). of output equal to — x*_ x x*_i < -1 A / (p ) n G Z+, > > n > n* out /j,q — = > 77 is oj* is = — A if -1 3, (since there is either n in the that G'" ((l no innovation -A ) in industries all > C) / (p ) A 0. compete a + 77)) > followers, industries 22 To R&D /j,* why see implies v-i = or is no 5, 0, Bertrand la and a labor share on the other hand, either i.e., > 77 or sufficient condition to ensure that (32) is that pn = for all with technology gap greater than n*). Thus jump a*_ x = 1), this Markov chain. Moreover, to the neck-and-neck state Markov chain is (when irreducible (and This is stated next proposition. policy (r) uni uni , £ ^ and a steady-state equilibrium Then, there exists a unique steady-state distribution of decisions (a*_ ly x*). . this condition and is sufficient (25) implies vi > (l l 1 to positive growth is suppose that 7? = and also that zLj =0. Then (37) immediately -1 A ) /p. Moreover, from (38) and the fact that u>* < 1, we have — -1 > G'- (ui - v-i - C) > G'~ ((1 - A hypothesis that ilj = 0, and implies xlj > x-i > 22 combined with industries technology gap of one (when Proposition 8 Consider a uniform IPR sequence of g*) relaxed, then there exists aperiodic), thus converges to a unique steady-state distribution of industries. and proved to*, then this equilibrium would also 0, -1 also strictly positive. immediate consequence of Proposition 0) or to the al l5 x*, that addition the steady-state equilibrium If in 1. - is 1 because after an innovation by a follower, = suppose and from Propositions 4 and 1 the law of motion of an industry can be represented by a finite a!_! establish: an equilibrium in which there i.e., if ((l /p 77 and ^ cost, leading to zero aggregate profits and thus 1 then the growth rate when An + 7?)) > some probability of catch- up or innovation by the 0, UTU In addition, 1. Lemma Moreover, The assumption that G'~ l 1. sufficient to rule involves £ , so that in every industry two firms with equal costs and charge price equal to marginal is un% steady-state equilibrium (/n*, v, innovation and thus no growth (this follows from involve (r) we can 0. assumption that G' 1 If the each z 6 Z, and also for continuous. Using this construction, uniform a + 77)) > / (p <p is compact and convex-valued are [z] Moreover, in any steady-state equilibrium u* exists. x*_ t —A ((l -1 Xn and ) 0. /p'-t). Therefore, G'' The reason why 77 > related to the composition effect discussed above. 23 1 ((l - A" can, under 1 ) /p - C) > contradicts the some circumstances, contribute . IPR State-Dependent 4 We now IPR Policy extend the results from the previous section to the environment with state-dependent The main policy. from the previous section generalize, but the argument results is slightly modified. It is IPR no longer necessarily the leader. that the sequence of values {^nj^L-oo is increasing, since be very sharply increasing, making a particular state very unattractive policies could 23 the. case For this reason, we do not have the equivalent of Proposition can be established that only a finite number for Nevertheless, 2. it of states will have positive weight in the steady- state distribution: Proposition 9 Consider (fi*, u* v, a*_i, x*, such that jj,* n — for <?*) , all policy (77, Then a steady-state equilibrium. is n> IPR state-dependent the and 0, suppose that G there exists a state n* N n* Because of the reasons highlighted in footnote 23, Proposition 4 may not necessarily hold with state-dependent IPR. Nevertheless, the proofs of these propositions make as long as (77, £) not too far from uniform IPR, the conclusions in these propositions will is continue to hold. In fact, our numerical results with optimal state-dependent IPR the conclusions of Proposition 4 (but interestingly not those of Proposition 5; Section clear that it always verify see Figure 5 in 5). Our next result is a generalization of Proposition maximization problem is R&D which shows that each individual firm's well-behaved with state-dependent IPR. Proposition 10 Consider share and the 6, the state- dependent IPR policy (77, policies of all other firms are given by z £) = and suppose (<I>, Then a, x). optimization problem of an individual firm leads to a unique value function A convex-valued for each zgZ A = {A n [z]};Loo ^d X and optimal R&D policies [z] Z^\{0} =4 [0,1] and X and upper hemi- continuous [z] [z] : = {Xn [*]}« _, [z] : Z =J [0, in z (where that the labor x] v [z] v the [z] : dynamic Z — R+ are compact = {v n [z]}^L_ 1 r) n — an ^ Vn+i ~* °°> which would imply that v„+i 24 , and establishes the existence of a steady- state equilibrium with positive growth. we could have and ). Finally, the next result generalizes Proposition 7 'For example, > — v„ is negative. Proposition 11 Consider G'~ l ((l > ) + Tyj)) > / (p IPR state- dependent the Then 0. > then g* 0, policy (77, a steady-state equilibrium Moreover, in any steady-state equilibrium exists. x*_ x -A -1 ui* < 1. and £) v, a*_ 1 (/j,*, In addition, if suppose , x*, that g*) oj*, either n^ > or 0. Although the analysis so and characterized some of dependent IPR policy far has established the existence of a steady-state equilibrium its properties, it is not possible to determine the optimal state- analytically. For this reason, in Section 5, we undertake a quantitative and structure of optimal state-dependent IPR policy using plausible investigation of the form parameter values. A Optimal IPR Policy: 5 In this section, we investigate the implications of various different types of R&D the equilibrium growth and Our purpose librium. Quantitative Investigation policy under plausible parameter values. values we 5.1 Methodology We As we take the annual discount rate as 5%, R&D given by (9). function. For example, 1 The The . its implications for optimal will see, the structure of optimal relatively invariant to the range of IPR pol- parameter is i.e., p year — 0.05 and without loss of generality, we theoretical analysis considered a general production function empirical literature typically assumes a Cobb-Douglas production Kortum (1993) considers a function of the form Innovation where Bq model and consider. normalize labor supply to for on not to provide a detailed calibration of the model economy, but to is and the innovation gains from such policy are icy policies rates using numerical computations of the steady-state equi- highlight the broad quantitative characteristics of the IPR IPR a constant and exp (nt) (t) is = Bq exp (nt) (R&D a trend term, which inputs) 7 (41) , may depend on general technological trends, a drift in technological opportunities, or changes in general equilibrium prices (such as wages of researchers fact that to R&D etc.). most empirical work estimates a inputs. of this form is not only its simplicity, but also the single elasticity for the response of innovation rates Consequently, they essentially only give information about the parameter in terms of equation (41). concave. The advantage For example, A low value of 7 implies that the Kortum R&D production function is 7 more (1993) reports that estimates of 7 vary between 0.1 and 0.6 25 , Pakes and Griliches, 1980, or Hall, Hausman and Griliches, 1988). For these reasons, (see also throughout, we adopt a R&D production function similar to (41): x where J3, where w the exp 7 > term). (tit) and can be 24 = checks for 7 number we = and 7 0.1 0.6. benchmark value The 0-05 also pins We that if down 7 g* i.e., as the = and then report 0.35, ~ licensing. sensitivity chosen so as to ensure benchmark economy This growth rate together with = 0.069 from equation the expected duration of time between any two consecutive innovations benchmark value, 1 and check the robustness of the = (2). ...} , C, = {00, 00, tection and licensing, More specifically, (42) results to A i.e., is about 3 years = 1.05 A = 1.2 26 We = 1.01 all of the parameters 1.05. year and 12 years, respectively). This completes model except the IPR = {0, 0, 24 is 0.019, in the the annual interest rate as r year take A and policy. As noted above, we begin with the (77 numerical choose the value of A using a reasoning similar to Stokey (1995). Equation (36) implies (expected duration of of the ...}). (77 We full patent protection regime without licensing, then compare this to an economy with = {0, 0, ...} ,£ = {CiCj •••})> where ( = v\ — full vq. patent pro- 27 After (t) = Bui (t) " (R&D some level h. ' and h large, it = min {Bh? + eh; Bh 1 + eh} makes no difference to our simulation results. benchmark parameterization with full protection without In particular, in our licensing, 24% of industries are in the neck-and-neck state. This implies that improvements in the technological capability of the R&D make For example, the following generalization of (42), x for e small this, 1 expenditure)' thus would be equivalent to (41) as long as the growth of w (i) can be approximated by constant rate. 25 For example, we could add a small linear term to the production function for R&D, (42), and also flat after i.e., can be alternatively written as Innovation 26 far Following the estimates reported in an industry, then a growth rate of about 1.9% would require A in captured by results, since in all other parameter in (42), B, which features indefinitely-enforced patents and no = 25 of it is [x/B]~i w, boundary conditions we imposed so satisfy the of states are reached. start with a G (x) — (thus in terms of the above function, an annual growth rate of approximately 1.9%, Pyear (42) do so without affecting any of the easily modified to (1993), economy Equation (42) does not exercises only a finite Kortum Bh"< In terms of our previous notation, equation (42) implies that 0. the wage rate in the is = 76% R&D economy both the leaders and the followers in 24% of the industries. Therefore, the growth equation implies that we have g ~ In A x 1.24 x 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 A ~ 1.05. 27 For the interpretation of full patent protection as C = vi — vo, recall the discussion in footnote 16. Note also is driven by the efforts of the leaders in of the industries 26 and the efforts of > we move IPR to the optimal state-dependent optimal value of each late the state-dependent IPR technology gap of n IPR = r\ Since policy. and ( n we n 5 or more. In other words, the policy (Vl, Cl) — n — ^^ none s- little effect IPR and make an growth rate g* , and then check Depending on whether there process 56789 flexibility (e.g., allowing relies process and is is then repeated we ( £, are demand excess IPR take the We now and or £ 5 7] e iteration. R&D policies, is r)y. On We — vo and £ conditional on w*, }^ = After convergence, _ OQ , is state- {a n }nL-<x> we compute the market from equation in the labor is (18). varied and the whole IPR policy we find the is computed. IPR policy sequences, growth-maximizing value then move the next component, for example, determined, we go back to optimize over present our simulation results. IPR rj-y T] 2 - for Once again, and results We start with the parameter values A = and 1.05 change when we vary these parameters. Protection Without Licensing benchmark with that at a license fee of v1 77 repeated recursively until convergence. start with the in this case policies policies. at a time, until 772 is and then show how the Full {x* or supply of labor, w* IPR IPR on whether there full protection and no licensing, which existing literature has considered so far (e.g., Aghion, Harris, Howitt = and 775 Then uj*. case, the optimal (growth-maximizing) computed one element procedure 5.2 market clearing for the growth-maximizing value of 0.35, first or { un}^l_i depending of industries, {/^}^Lo- for different that component, for example, C 10 11 on uniformization and value function repeated until the entire steady-state equilibrium for a given is In the state-dependent We "> results. of values {^nl^L-oo and the steady-state distribution = (V5> Cs) , 4 dependent IPR policy or not), and we derive the optimal 7 form of industries that have a all C4) ,-^, guess for the equilibrium labor share initial we generate a sequence this >-um £" , details of the uniformization technique are described in the proof of Proposition 6. as given 77 ('74. 3 value function iteration, see Judd (1999). In particular, The um policy matrix takes the form: ^s^ 2 The numerical methodology we pursue The r] computationally impossible to calcu- C2) 073> C3) ,—^~, (172. 1 on our policy limit our investigation to a particular , checked and verified that allowing for further C 6 to differ) has it is whereby the same V and C applies to policy, Technology gap: We IPR to a comparison of the optimal (growth-maximizing) uniform £, followers are indifferent between a we always suppose — e for e 0. that they choose o = J. 27 1. = and a Thus = 1, and in is the case that the and Vickers, 2001). In computing the equilibrium alternatively one might wish to think that terms of our model, this corresponds to = (28) implies that a*_ n for all n. benchmark has an annual growth The value function for this We r] = n B co for terms of in benchmark case shown is n > 0, in Figure which R&D (with the solid 1 for firms that are no state-dependent IPR also shown neck-and-neck and efforts for leaders R&D followers in this policy, all followers The undertake the same benchmark a leader declines as the level of at the technology (i.e., line). no state dependence in the IPR technology gap increases. Moreover, consistent with Propositions 4 and is (42), so that the consistent with Proposition 4, is level for all followers (since there is (again with the solid lines). This figure shows that the R&D Equation all n. rate of 1.86%. Figure 2 shows the level of policy). — n and £n choose the parameter value function has decreasing differences for and features a constant for all 5, the highest level of — gap of n 0). R&D level of Since there is which is effort, in the figure. Figure 3 shows the distribution of industries according to technology gaps (again the solid benchmark line refers to the n = 1, but there The mode case). also a concentration of industries at technology is them implies that innovations by the followers take The B above, column of Table first of the distribution 1 The the bottom row. is reasonable. 28 the workforce Full 5.3 We In the GDP. Instead, next turn to 1 — u* shows that = technology C is — vi ~ a*_ n = equal to 0.0186, which is recorded at 1 and 2. The IPR is all this in this lo* u> is should not be thought 5% in GDP, which is benchmark parameterization 3.2% of corresponds to also consistent protection with licensing. v o for steady-state value of with US of data. 29 With Licensing protection with licensing as corresponding to and Cn 0, benchmark parameterization, Protection full because measures the share of pure monopoly profits working as researchers, which IPR is the only factor of production in the economy, Finally, the table also is 0, table also shows the gap between x*Q and x*_ l (0.029 versus 0.018), the of as the labor share in value added. = gap n gap of benchmark simulation. As noted also reports the results for this frequencies of industries with technology gaps of 0.952. Since labor at the technology to the "neck-and-neck" state. chosen such that the annual growth rate is is n r] n = As specified above, for all n (so that the license fee for we think of full IPR (so that patents never expire) making use of a leading-edge equal to the net present discounted value gap between being a one step ahead Bureau of Economic Analysis (2004) reports that the ratio of before-tax profits to GDP in the US economy was 7% and the after-tax ratio was 5%. 29 According to National Science Foundation (2006), the ratio of scientists and engineers in the US workforce in 2001 is about 4%. 28 in 2001 28 and a neck-and-neck leader effort levels and distribution of industries is pay the no longer a spike license R&D is effort at similar to that in the n = 0, however, since column (in fact, as Table 2 of 1 shows considerably higher than in the benchmark case. In particular, The resulting growth rate is growth comes not from the increase now of the followers Table 2.58% instead of 1.86% in the R&D shows that 1 economy without now is no licensing. firms always prefer to level of xl x in the effort overall, this considerably higher is R&D now growth rate by followers is 0.021 rather than but from the fact that the economy equation (36)). In (recall be no benchmark. The boost to also advances the technological frontier of the and build on the leading-edge technology license Since there . in equilibrium there will More importantly, the industries in the neck-and-neck state). 0.018. lines) and jump ahead of the leading-edge technology. This makes the neck-and-neck no longer special state in dashed for this case (with the state-dependent policy the general pattern There R&D Figures 1-3 show the corresponding value functions, firm). R&D since they can fact, column 2 of achieved with a lower fraction of is the workforce, only 2.6%, working in the research sector. This which result, is robust across different parameterizations of the model, is the first important implication of our analysis. Relative to existing models of step-by-step innovation, such as Aghion, Harris, Howitt and Vickers (2001), which do not allow for the possibility of' licensing, here the R&D effort by followers can directly contribute to economic growth and this increases the equilibrium growth rate of the economy. Optimal Uniform IPR Protection 5.4 We next turn to optimal (growth-maximizing) that r) life = rj and Cn Column rate. to n = C f° r au 3 of Table 1 n and look > IPR policy with licensing. for values of 77 That is, and £ that maximize the growth shows that the growth-maximizing values of 77 and £ are both equal the benchmark parameterization. This corresponds to zero license fees and indefinite in of patents, so that followers can never copy the leading-edge technology without they can always advance one step ahead of the leader efforts (without paying any license The resulting value function, when they The figures R&D effort levels and column 3 of Table 1 are successful in their IPR R&D and industry distributions according to lines). — vq declines significantly), but encourages by the followers, since when successful they do not have to pay the optimal uniform but show that the growth-maximizing IPR policy discour- ages leaders (which can be seen from the fact that v\ effort R&D, fees). technology gaps are shown in Figures 4-6 (with the solid R&D we impose increases the growth rate 29 by only a little, license fee. The however. While the growth economy with rate of the increase full IPR protection with licensing was 2.58%, it is now 2.63%. This associated with a very modest rise in the share of the labor force working in research is (from 2.6% to 2.7%). Optimal State-Dependent IPR Without Licensing 5.5 We next turn to our second major question; whether state-dependent difference relative the uniform IPR. the comparison is benchmark to the and look for the shown column 4 of Table in Two column We first now 2.03% column 1). we In particular, set ( n — oo for all n The results are growth rate increases noticeably relative to that maximizes the growth rate. ...,775} 1. features are worth noting. 1; it is a significant investigate this question without licensing (so that case in combination of {77^ IPR makes First, the instead of 1.86%. Nevertheless, this increase IPR to the benefits of licensing. Therefore, state-dependent is quite small relative policy with no licensing is not a substitute for licensing. Second, we see an interesting pattern (which investigations). The optimal state-dependent policy technological leaders that are further ahead. involves = rj^ 0.054, 7? = 2 0.005, and = 773 77 4 in fact quite general in all of is {77 l7 ..., T75} In particular, = t] 5 = 0. our quantitative provides greater protection to we find that the optimal policy This corresponds to very little patent protection for firms that are one step ahead of the followers, which can be seen by comparing 77 ! = 0.054 to followers are x*_ .j = 0.010. more than This comparison implies that firms that are one step behind five times as likely to catch up with the technological leader because of the expiration of the patent of the leader as they are likely to catch successful R&D. Then, there is leaders that are two steps ahead, and ?7 2 This pattern of greater protection is it is economy by bringing more little R&D) is full or more is IPR policy should try to boost the growth into neck-and-neck competition. This composition effect as follows: force, the trickle-down effect. by providing secure patent protection steps ahead of their rivals, optimal state-dependent secure protection of their own ahead may industries with large technology gaps (where leaders leaders that are one and two steps ahead as well less protection and patents never expire. for technological leaders that are further dominated by another, more powerful trickle-down effect own l/10th of 77^ Perhaps even more remarkably, after go against a naive intuition that state-dependent engage in their a steep increase in the protection provided to technological a technology gap of three or more steps, there rate of the up because of —despite the intellectual property. 30 IPR The is present, but intuition for the to firms that are three increases the R&D fact that these firms In fact, this is effort of now face precisely because of low protection for technological leaders that are one or two steps ahead combined with the promise of much greater protection once they reach a technology gap of three steps or more. Mechanically, high levels of ^ advanced firms increases vn for and r\ n > 2 IPR reduce v\ and V2, while high Consequently, 3. that are one or two steps ahead to undertake R&D it becomes more protection for more beneficial for leaders to reach the higher level of IPR protection. R&D for leaders under state-dependent IPR contrasts with uniform IPR, which always reduces R&D by all technological leaders. The possibility that imperfect state-dependent IPR protection can increase R&D incentives is a novel feature of our approach, This pattern of increased IPR since previous models have not considered state-dependent Optimal State-Dependent IPR With Licensing 5.6 we turn Finally, to the most general policy choice within the context of our approach and investigate which combinations of {r7 l5 ..., ?? 5 and {Cii—jCs} maximize the growth rate and } the contribution of the optimal state-dependent shown of this exercise are uniform to policies. IPR in column 5 of column policy with licensing in maximize the aggregate growth Table rate. 1. 3, IPR policies to Now the economic growth. The results comparison should be to the optimal where uniform IPR The value functions, policies R&D economy distribution over different levels of technology gaps in this r\ efforts and C were chosen and the industry shown are in Figures 4-6 (with the dashed lines). We see in column 5 that patents never expire. growth-maximizing IPR policy involves Nevertheless, IPR paying any license From fees. is = 0, of the economy policy, 2.63% rate of the is in full there on, £ increases to £ 2 column economy 3. is for all n, so that is not In full. one step ahead of their own knowledge without patent protection (since v\ 2.96%, which = which implies that followers can = significantly higher — vq = which 0.82, patent protection. After three or more steps, however, we have £ 3 which are very close to n protection for technological leaders particular, the growth- maximizing policy involves £i build on the leading-edge technology that rj = 1.96). is still less 1.94 The and C4 — than Cs — full 1-95, resulting growth rate than the growth rate under uniform IPR This shows that state-dependent policies can increase the growth significantly. State-dependent policies again achieve this superior growth performance by exploiting the trickle-down effect, which we already saw in the case without licensing. In particular, £n is an increasing sequence, which implies that technological leaders that are further ahead receive greater protection. boosting the R&D As in the previous subsection, this pattern of effort of technological leaders 31 IPR is used as a way of that are one or two steps ahead of their rivals (see Figure 5) Since these leaders receive . increase both their profits and their IPR little protection and understand that they can protection by undertaking further innovations, they have relatively strong innovation incentives and undertake high the fraction of the labor force working in in Table 1) R&D and again illustrates how R&D to 3.9% (which is levels of R&D. greater than IPR imperfect state-dependent This increases the other cases all protection can increase incentives relative to full protection. The growth ers that are rate of the economy also receives a boost one step behind the leaders, from the R&D effort of the follow- R&D thanks to licensing, followers' since, directly contributes to the advancing the technological frontier of the economy. Figure 5 shows that followers that are one step behind the frontier case with growth-maximizing uniform this pattern of R&D efforts is IPR now have R&D a higher (which involved (n that the growth-maximizing — IPR effort than even for all n). The reason more than once step behind followers that are Overall, the results consider the show that state-dependent IPR economy without R&D level of lower than in the economy with uniform IPR. is can increase the equilibrium policies growth rate of the economy substantially and that the trickle-down when we for and the trickle-down policy have increased the value of being a technological leader. In contrast, the effect in the licensing, but also in the effect is powerful, not only presence of licensing. Robustness 5.7 Tables 2-5 show the robustness of the patterns documented in Figures 1-6 and in Table 1. In particular, each of these tables changes one of the two parameters A and 7 (increasing or reducing A to 1.2 or 1.01, and increasing or reducing 7 to 0.6 or 0.1) and show the results corresponding to each one of the we change the parameter case, economy with full IPR B regimes and discussed so five different policy in all cases we protection without licensing see a significant increase in the smallest increase R&D, is when 7 = 0.6, In each equation (42) so that the growth rate of the benchmark is the same as in Table Notably, the qualitative, and even the quantitative, patterns in Table In far. 1 1, 1.86%. are relatively robust. growth rate when we allow licensing. The presumably because with limited diminishing returns to incentives were already sufficiently strong without licensing. the growth rate increases only from 1.86% to 1.98%. In increases the growth rate to above 2.5%, which is all As a result, in this case, other cases, allowing for licensing a very large increase relative to the baseline of 1.86%. Moreover, in further, all cases, moving to state-dependent IPR policy increases the growth rate though the extent of the increase varies depending on parameters. The most modest 32 when 7 = gain comes because in this case the economy with 0.1, already achieves a very high growth rate of 2.78%, and thus there an increase in growth rate from state-dependent IPR. In growth rate is is the fact that in shaped by the trickle-down is considered, there ahead, but IPR is little or only little room left for other cases, the increase in the In effect. all all cases, growth-maximizing state-dependent of the various parameterizations no protection provided we have to technological leaders that are one step protection grows as the technology gap increases. This implied by the trickle-down We is quite substantial. Perhaps, more noteworthy IPR all protection and licensing full is the typical pattern effect. therefore conclude that both the substantial benefits of licensing in models of tech- and the benefits of state-dependent nological competition policies, mostly coming from the trickle-down effect, are robust across different specifications. Conclusions 6 In this paper, we developed a general equilibrium framework to investigate the impact of the extent and form of intellectual property rights (IPR) policy on economic growth. major questions we focused on are whether leading-edge technology in return of a license rate licensing, fee, The two which allows followers to build on the has a major impact on the equilibrium growth and whether the same degree of patent protection should be given to companies that are further ahead of their competitors as those that are technologically close to their competitors. In our model economy, firms engage in cumulative (step-by-step) innovation. Leaders can innovate in order to widen the technology gap between themselves and the followers, which enables them to charge higher markups. On the other hand, followers innovate to catch up with or surpass the technological leaders in their industry. Followers can advance in three different ways. First, the patent of the technological leader industry to copy the leading-edge technology. R&D to improve its own when successful, In the model economy, it will be far followers, expire, allowing the follower in the Second, each follower can undertake catch-up variant of the product to catch follower can undertake frontier in this case, may and R&D, it IPR will up with the Third, each building on and improving the leading-edge technology; have to pay a license fee to the technological leader. policy regulates whether licensing also the length of the patents. steady-state equilibrium and proved leader. its We existence under general is possible and how costly characterized the form of the IPR policies. We then used this framework to investigate the form of "optimal" (growth-maximizing) IPR policy quantitatively. 33 The major 1. A move findings of this quantitative exercise are as follows: from an IPR policy without licensing to one that allows on the equilibrium growth icant effect for licensing has a signif- For the benchmark parameterization of our rate. model, licensing increases the growth rate from 1.86% to 2.58% per annum. This substantial increase in 2. State-dependent rate. In growth rate IPR is robust to a large range of variation in the parameters. also leads to a significant improvement in the equilibrium growth our benchmark parameterization, the growth rate of the economy increases from 2.57% under the growth-maximizing uniform IPR policy to 2.93% under state-dependent IPR Perhaps more interesting than policy. growth rate intuition, is we the form of the optimal state-dependent find that the impact on the equilibrium this substantial IPR policy. Contrary to a naive growth-maximizing IPR policy provides greater protection to firms that are further ahead of their rivals than those that are technologically close to their competitors. The effect. Underlying this form of the optimal IPR policy the trickle-down trickle-down effect implies that providing greater protection to sufficiently advanced technological leaders not only increases their R&D is efforts of all technological leaders that are less the reward to innovation now R&D but also raises the efforts, advanced than this level, because includes the greater protection that they will receive once they reach this higher level of technology. Our results suggest that the trickle-down is powerful both with and without licensing, and insensitive to the exact The policy industry). move move towards optimal state-dependent increase innovation and economic growth. IPR may depend on form and magnitude are relatively parameter values used in the quantitative investigation. analysis in this paper suggests that a particular, a its effect The to a richer menu of policies with licensing results also IPR may policies, in significantly show that the form of optimal the industry structure (and the technology of catch-up within the Nevertheless, these conclusions are based on a quantitative evaluation of a rather simple model. It would be interesting to investigate the robustness of these results to different models of industry dynamics and 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, differences in the extent of fixed costs). most important area IPR policy, using also structural for future work is a detailed empirical investigation of the form of optimal both better estimates of the models where the The effects of IPR effect of different policies 34 policy on innovation rates and on equilibrium can be evaluated. , , Appendix: Proofs Proof of Proposition Equations (19) and (21) imply 1. Q(i)A-£-° n/i " (1) u (t) «>(t) Y(t) (t) ' Since = u> (t) w* and {/^}^L Y (t) are constant in steady state, grows at the same rate as Q(t). Therefore, , = q Now + At)- In <?(£ ,. lim . At-o At At (i) in the fraction /i* of the industries with £*At + o(At); (ii) in the same industries, the note the following: during an interval of length technology gap n > the leaders innovate at a rate 1 Q(i) In followers innovate at the rate a*_ n x*_ n At+o {At); (iii) in the fraction Pq of the industries with technology both firms innovate, so that the total innovation rate is 2x$ At + o (At)); and (iv) each innovation increase productivity by a factor A. Combining these observations, we have gap of n = 0, \nQ(t Subtracting + = At) \nQ (t), ln<2(£) +lnA dividing by 2p* x* At &< A + J2 At and taking the limit * + At 9 (At) — + JT a*_ n x*_ n At + o (At) gives (36). > Proof of Proposition 2. Let (a_i, {x n }^°__ 1 } be the R&D decisions of the firm and {v n }'^'__ l be the sequence of values, taking the decisions of other firms and the industry distributions, {xn}™=-i> {/OnL-i' w * anc^ 5' as gi ven By choosing x n = for all n > —1, the firm guarantees v n > for all - n > — 1. Moreover, that {w n }^= _ 1 since flow profit satisfy Suppose, first, Proof of v\ > vq problem in equilibrium (26) implies vq : (25) implies that v\ > v when since 1 for all x*_ x = 0, V\ > i>i < ^o, = v\ = (l — A -1 n > — 1, i>„ < 1/p n > — 1. n > — 1, establishing then (31) implies Xq = 0, and by the symmetry of the As a result, from (30) we obtain xlj = 0. Equation 0. ) / (p + rj) > v\ — 77/ (p C, < < 0, yielding a contradiction and proving that Suppose, to obtain a contradiction, that u_i > vq. — 0. This implies i>_i = 771J0/ (p + 77), which contradicts v-i Y Thus we must have vi — £ > vq, which implies that a*_ x = 1. Imposing a*_ x vq ' vq, (30) yields x*_ + 77) < 1. the value function of a neck-and-neck firm can be written pv = max{-u*G(x Xq > ma,x {-gj* XQ > -u*G = pV-l, which contradicts the hypothesis that Proof of v n (25) < vn+ i : ) + + x*_ (xlj) tj_i +x ) max{-w*G(x > ) G (x + x XQ and for all 1/p] for all [0, . Proof of v-i If nn < a bounded sequence, with v n £ is > vq l x \vi = Vq 1, as: {vi-v ]+x °[v^i [vi - v [vi - v-\ - - C] - > V-l }} -v (43) }} , £]} + 7? , [v - V-i] and establishes the claim. Suppose, to obtain a contradiction, that v n > v n +\. - f„] Now (29) implies x*n = 0, becomes pv n = (1 - A" n ) + xl a [a*_! (u-i + C) 35 + (1 - all) ^o + V bo - t>n] (44) . + Also from (25), the value for state n pvn+i > (1 1 satisfies - A""- 1 ) + x*_ x [a'_ (vj-i 1 + C) + (1 - ali) ^o - + ^n+i] - vn+ i] [vo t? (45) Combining the two previous expressions, we obtain - A" n ) + as!.! ~ - \- n + xl [a* (1 l > _n_1 1 < A~ n j + C) + (l - all) v - v n + [v - vn (v-i + C) + (1 - all) ^o - «h+i] + V N - «*ti] (u_! [at! 1 r\ ] ] • this implies u n < n n +i, contradicting the hypothesis that v n > vn +i, and establishing the desired result, v n < v n +i- Consequently, {v n }^__ 1 is nondecreasing and {vn}^L * s (strictly) increasing. Since a nondecreasing sequence in a compact set must converge, {v n }'^'__ con1 Since A verges to its limit has a nonnegative , point, v^o, which initial value. Proof of Proposition must be strictly positive, since {v n }^L strictly increasing is and This completes the proof. The 3. maximization of the value function first-order condition of the (25) implies: G' — > (x-n) and x n > 0, with complementary slackness. G' (0) is strictly positive by assumption. If (v n +\ — v n ) /ui* < G' (0), then x n = 0. Proposition 2 implies that {v n }'^__ l is a convergent and thus a Cauchy sequence, which implies that there exists 3n* € N such that v n+ i — v n < u>*G' (0) for all n > n* u . Proof of Proposition From equation 4. 5 n+ i is sufficient to establish that = x n+l < x*n v n+x (29), -v n <v n - v n _i = 6n (46) . Let us write: pv n = max{(l - A~ n ) - u>*G(x n + ) x* ,-[v n +i ~'Vn] +^-i [all (.V-i + + (l - a*_ x ) v ] +tjv } , (47) where p = p + x*_ x + r\. Since x n+1 , x* and x*n _ Y are maximizers of the value functions v n+ i, v n and v n -!, (47) implies: - ™" 1 pu n+ l = 1 -A pv n > 1 - A" n - u'G pv n > 1 pVn-x = 1 - W*G « +1 ) + <+i Kn-2 - Un+l] + X*-\ [«-l (^-1 + '+ (l ~ a -l) ^o] + TO), (48) Now + x*n+l [v n+ - Vn] + x*_! [all («-i + C) + (l - all) vQ + tjvq, n - A" - uj*G «_x) + <_i K+i - v n + x*_i [al (v_i + C) + (l - a*-i) v + TO), - A" n+1 - LJ*G «_ + X*n _! [Vn - Vn-i] + itj. [alj {V-i + C) + (1 - ali) Wb] + TO(x* +1 ) i ] ] ] x x ) taking differences with pv n and using the definitions of n - -1 <5 n 's, we obtain - pSn+i < A~ P5n > A-^^l-A-^+^.i^n+l-^)- (l A ) + x n+1 (<5 n+2 S n +i) Therefore, (P + <-i) (<Wi - 5n ) < -k n + xn+1 {5 n+2 where kn = (A-l) 2 A- n_1 >0. 36 - S n+ i) , (49) . Now — to obtain a contradiction, suppose that <5„+i 5n > 0. . From (49), this implies <5 n+ 2 — 6 n+ i > Repeating this argument successively, we have that if 5 n + \ — 8 n > 0, However, we know from Proposition 2 that {v n }^L Q is strictly then 5 n+ i — 5 n > for all n > n' increasing and converges to a constant v x This implies that 5 n 0, which contradicts the hypothesis that <5 n+ i — S n > for all n > n' > 0, and establishes that x* +1 < x^- To see that the inequality is strict when x* > 0, it suffices to note that we have already established (46), i.e., 6 n+ i — 5 n < 0, thus if equation (29) has a positive solution, then we necessarily have x* +1 < x*n since k n is strictly positive. i ' . . J. . > Proof of Proposition al x =0, and (26) 5. Proof of Xq can be written as = pv We have vq > from Proposition 2. -lj*G x*_ 1 (xq) Suppose + Xq > i>o Suppose : > > -v V\ This inequality combined with Inequality (51) strictly We now a"_ 1 = and (31) 0, Then 0. V-i+vi-2vq +vi- [v-i that C first 2v > > [v ~ > and > (38) }delds Xq > = 0. G () is Suppose next that vq x"L±. x*_ x x*_ is Moreover, since . x =0. an optimal policy, so that -uj*G(xI)+Xq[vi-vo]+Xo[v-i-vo\ -uj*G{xq)+x*q [vi-V-i -C} + v[vo -v-i}. Subtracting the second expression from the P (28) implies V-i. . pv-i Then (51) Vq- given by (31), (50) then implies Xq = and thus next show that when £ < v\ — vo, Xq > x'L 1 In this case, a*_ 1 = 1 = vq. (50) is pv - ) (50) implies Xq holds as a weak inequality and implies that Xq convex and Xq v\ U_i] < we obtain first, + X*Q [V-i C - Vq] + {X* + Tj) [V-l ~ Vq] , and therefore -v-i] < [vq +C-t'o]- [v-i < vo, and therefore V-i + C > vq. Next observe that with a*_ l = 1, > imply that x*_ if and only if v\ — Vq > V\ — V-\ — Q, or equivalently if and only if Xq (38) x > we vq. Thus have established that Xq > x*_ x both when £ > v\ — vq and when C < vi — vq. C Proposition 2 implies that v_\ (31) u_i and + Lemma Lemma 1 1 and proof: We next first prove a lemma, which will be used in the rest of the Appendix. 1 Suppose that G'^ 1 ((l - A^ ) / (p + Proof of Lemma: Suppose, to obtain a x*_ 1 = 0- Then (25) implies -q)) > 0, > then Xq contradiction, that Xq then implies that pv\ Equation (26) together with Xq = > gives vq v1 Combined with 1 = 0, A + [vq j] - vi] and hence -v > 1-A" P + V 1 this inequality, (31) implies naxfG'- 1 A 1 ( maxJG'-M 37 . r - . {p i + rf) A A" - P+V ) n .n and v > = 0. 0. The first part of the proof , . < 1. where the second inequality follows from the fact that ui* The assumption that -1 then implies Xq > 0, thus leading to a contradiction and establishing G'~ ((l — A ) / (p + r/)) > l > that x^ > G'" when G'~ l - 0M > then implies vq £ x*_! ((l A -1 1 ((l when > together with Xq and C > 0: Given Lemma 1, + w)) > ) / (p - A" ) / {p + r))) > implies that Xq > 0. Then the first part of the proof implies that > vi — vo, Xq > x'L 1 Next suppose that < C < v i — vo- Then the same argument as above that Xq > x*_ 1 if and only if vi — vq > vi — V-i — £, or equivalently if and only if v^i + £ > vq. Proof of Xq 1 G () Strict convexity of 0. . implies Suppose this thus v-x = Then from the first part of the proof, we have that Xq = x*_ x = 0, and which implies u_i + > vq and thus Xq > x*_ 1 This yields a contradiction and not the case. is = vq 0, (, > completes the proof that Xq when x*L x Proof of Xq > x\ To prove that Xq proof of Proposition 4: > : . > £ G/_1 and — A" 1 ) ((l + / (p > 77)) x\, let us write the value functions v 2 0. and t>i , vq as in (48) in the Now 4, 1- \~ 2 -u*G{x*2 )+X2lv3 -v 2 }+xt = pv 2 ~pv x > pv x > pv = [a*_ 1 + (l-a*_ )vo}+r)v + (l -a*-i)v Q +V v o, {v-i+C,) 1 1 - X" -oj*G{x 2 + x*2 1 \v 2 -v{\ + x*_ 1 [alj (v-i + 1- A" -uj*G{x* )+x*q[v 2 -v }+x*_ [al^-i + C) + -LJ*G(xq) + Xq [Vl - Vq] +nVQ + X*_ Vo + Xq [V-i - Vq] 1 ) ] 1 1 1 , (l - alj v ] + nvo, . x taking differences with pv n and using the definitions of <5 n 's as in (46) in the proof of Proposition we obtain p5 2 ph pSi p5i < X- 1 (I - - X-1 1 ) + + xl(5 3 - 52 ) - + xl > (1 > > (l-X- )+x -1 (l - A ) +Xq A ) xq {5 2 1 '(5 2 (6 2 b~i) -5 - 1 ) 5i) (52) x*_ 1 a*-i(+ + - (x*_ — of the proof has established that x*_ x — xo\ [v-i — Vq] > 0, and the 1 tu.i combining + (p — last inequality with the this inequality 1 < Xq pSi> (1- Now a*_ first - x* ) (6 2 <5i) + (u-i l + Next note that Proposition 2 implies that [x*_± .[al l + C) (1 -v v ] -x^[v-i-.v ], {x*-ia*-i-xZ)[v-i-vo\, - Xq) [u_i < vq v ] Moreover, the 0. Combining 0. this with first part of the a*_ x < 1 first part establishes that then implies A" 1 +x* ) {5 2 - <5i). inequality of (52), < - - a*_ x ) - A" (1 2 1 ) we obtain + x2 {S 3 - 62 ) (53) . Proposition 2 has established 62 > S3, so that the right-hand side is strictly negative, therefore, we must have 5 2 — 5\ < 0, which implies that Xq > x\ and completes the proof. Proof of Proposition problem of a representative pv n = max Fix z 6. = (uj, {.T n }^ = _1 , {a n }~= _ 00 ), and consider the optimization firm, written recursively as: { (1 - A~ n ) - uG {x n ) + xn [v n+1 - vn ) x„e[o,x] + x-i (a_i Pvq [v-i - vn + C] + (1 = max {-loG (xq) +x - a_i) [vi [v - + v[ v o [v-i -v v n ]) -v ]+ x -vj) for n 6 N }} io£|0,i] pv-i = max {—u>G(xo) + x-.i(a-i[vi—v-i-Q + (l-a-i)[vo-v-i]) i_i€[0,x], a_i£[0,l] +t][vq -v-i]}. 38 We now transform this dynamic optimization problem into a form that can be represented as a mapping using the method of "uniformization" (see, for example, Ross, 1996, Chapter contraction 5). Let I = ({x n }~ _ 1 ,{a n } n= _ c and be n' starting with state n when the firm and the R&D policy of other firms = policy, (x- n ,a- n ) (i_i,a_i) (l-Q-l)z_i+7) X-1+I7 p-lo €l€ po.ikk, Pn,-1 (£ — Pn,0 £J Xn+ i_,+ n £ , {a n } n _ X_!+7) (l-q-i)x_i+7; x n +£_i+77 I Pn,n+1 Uniformization involves adding fictitious transitions from a state into itself, = £ I Hx-i+t; which do not change the value of the program, but allow us to represent the optimization problem as a contraction. purpose, define the transition rates tp n These transition rates are (* + X-i+T] for n e + for n = — 1 for n = x-i 2x n t) I For this as xn 4'n IPR Xo + Xq £ I ({x n )^__ l transition probabilities can P-a,iUI£ = =j^ = Using the fact that, because of uniform be written as: £. n £ N, these po,-i«K I I) be the probability that the next state will I in question chooses policies £ given by is for all Pn,n> (« finite since ip n I £ £ | {1,2, ...} r\ < ) = ip + 2x < rj and 00 for all n, where x is the maximal by assumption). Now following equation (5.8.3) in Ross (1996), we can use these transition rates and define the new transition probabilities (including the fictitious transitions from a state to itself) as: flow rate of innovation defined in (11) in the text (both x fw are finite 77 V'nQsie) Pn,n' (t I) I 1 if n 7^ n' if n = 1 n' This yields equivalent transition probabilities P_ 1 ,_ 1 (^IC) = P0.-1 (€ \l) Pn,0 (€ 1 g) = = 1-^ 2§bj -a (1 and also defines an ^ P-1,0 (« 1 P~n,n (« ||) 1+ " effective discount factor 1 /3 given by + V" p the per period return function (profit net of n f T l- J + + 77 + R&D ' , , 1 ( ?"\ I 1 tj ' o_-is_i 2x+r, i) = ^ £ £ — I J Tv n , I Il n for all (x„) + P^2p n ,n' n£Z. 39 Pn,-1 (fi I 1) = 77 u"^ 1 (54) otherwise Using these transformations, the dynamic optimization problem can be written max 2J+r) expenditures) be P+2S+7, — (€ Pn,n+1 2x 25 ip = po,i ^fe^ - (e „ P-1,1 (s a£,- = p let -B Po.o(«l€)=l-^ 1+T; /3 Also (1 = e) Un \ I) *V > , for all as: n G Z, (55) 5-ii-i 22+7) v={v n }^L_ where V = {v functions = z (<I>, : and the second 1 {^1} U Z + — M + } » defines line into the operator T, T itself. a_i, {in}^L_i), P ossesses a unique fixed point v* clearly is = a mapping from the space contraction, Q {v*l } ^= _ l thus, for of given Stokey, Lucas and Prescott, (e.g., 1989). Moreover, x n G [0,5], a^\ G [0, 1], and v n for each n = -1, 0, 1, ... given by the right-hand side of is continuous in a n and x n (a n applying only for n = —1), so Berge's Maximum Theorem (Aliprantis and Border, 1999, Theorem 16.31, p. 539) implies that the set of maximizers (Al x {X^}'^'__ 1 ) exists, (55) , is nonempty and compact-valued for each z and is upper hemi-continuous in z = (w,a_i, {x n }^L_- i Moreover, concavity of v n in a n and x n for each convex- valued for each z, completing the proof. n = —1, 0, 1, ... implies that (A*_ 1; Z4 Z constructed in 7. We first show that the mapping $: -1 _1 and then establish that when G' ((l - A + ??)) > ) / (p Proof of Proposition fixed point, to a steady state with u>* First, in the it < {X*}^L_ 1 \ is ). also (40) has a this fixed point corresponds 1. 3> maps Z into itself. We next show that Z is compact a subset of a locally convex Hausdorff space. The first part follows has already been established that product topology and is Z can be written as the Cartesian product of compact subsets, Z = [0, 1] x [0, 1] x Then by Tychonoff's Theorem (e.g., Aliprantis and Border, 1999, Theorem 2.57, p. 52; Kelley, 1955, p. 143), Z is compact in the product topology. Moreover, Z is clearly nonempty and also convex, since for any z, z' £ Z and A G [0, 1], we have Az+ (1 — A) z' € Z. Finally, since Z is a product from the n^L-i fact that [OjS]. on the of intervals Lemma 1999, Next, n G { if is real line, 5.54, p. it is a subset of a locally convex Hausdorff space (see Aliprantis and Border, 192). a continuous function from — 1} U Z+ Z into are upper hemi-continuous in z. [0, 1] and from Proposition Consequently, <& = (p [z] , 6, A_i A-i [z] (z) , X and Xn (z) [z]) has closed for X graph in z in the product topology. Moreover, each one of <p{z), A-i (z) and n (z) for n = —1,0, ... is nonempty, compact and convex-valued. Therefore, the image of the mapping $ is nonempty, compact and convex-valued for each z G Z. The Kakutani-Fan-Glicksberg Fixed Point Theorem implies that if the function $ maps a convex, compact and nonempty subset of a locally convex Hausdorff space into itself and has closed graph and is nonempty, compact and convex- valued z, then it possesses a fixed point z* G $(z*) (see Aliprantis and Border, 1999, Theorem 16.50 and Corollary 16.51, pp. 549-550). This establishes the existence of a fixed point z* of $. To complete the proof, we need to show that the fixed point, z*, corresponds to a steady state equilibrium. First, since a n (ui* a*_ l {i* ,}^L_ 1 ) = a*-\ and x n (ui*, alj, {x*n }'^'= _ 1 ) = x*n for n G , , ^' _ ) constitutes an R&D policy vector = 1 by steady-state equilibrium (Definition 3). Next, we need to prove that the implied labor share ui* leads to labor market clearing. This follows from the fact that the fixed point involves w* < 1, since in this case (39) will have an interior solution, ensuring labor market clearing. Suppose, to obtain a contradiction, that u>* = 1. Then, as noted in the text, we must have for n € { — 1} U Z + However, Lemma 1 implies Hq = 1. From (32), (33) and (34), this implies x*n = 1 that this is not possible when G'~ l ((l — A" ) / (p + r])) > 0. Consequently, (39) cannot be satisfied at ui* = 1, implying that w* < 1. When u* < 1, the labor market clearing condition (35) is satisfied at ui* as an equality, so ui* is an equilibrium given {x* }'^L_ 1 and thus z* = (ui*, a*_ lt {a;* }^__ 1 ) is a { — 1} U Z + we have , that is that given a labor share of best response to itself, c ui* , (a!_j, {x* l } as required . l , steady-state equilibrium as desired. Finally, g* > 0. if rj > 0, from Lemma 1, equation (36) implies then (34) implies that p,*, > 0. Since Xq > if x*_ > 0, then g* > follows from (36), completing the proof. l Alternatively, and a;* > for Proof of Proposition 8. We will establish that there exists n* such that £*. = n > n* which will imply that the states n < n* are transient and can be ignored, and that {Pn) n =o forms a finite and irreducible Markov chain over the states n = 0, 1, ..., n*. is strictly increasing. Then it follows First, recall that Proposition 2 has -established that {v n } n€ z for all n > from the proof of Proposition 3 that there exists a state n** G N such that a;** = all , 40 , n S N:u n+ i — v n < oj*G' (0)}. Such an n* exists, since the set { ,...,n--} { -1 and nonempty because of the assumption that G'~ l ((l — A ) / (p + 77)) > 0. Then by construction x* > for all n < n* and x*. = as desired. Now denoting the probability of T T for being in state n starting in state n after r periods by P (n, n), we have that lim T _>00 P (n, n) = finite Markov chain over the states n = 0, 1, ...,n* all n > n* and for all n. Thus we can focus on the Now n**. *s {Mn)n=o n= finite is and {Mn}"=o ^s the limiting (invariant) distribution of this un iq ue ly 0, 1, ...,n* in Stokey, = min ne n* let {0, ...,n**} — so that 1, Markov chain. Given a*_ x and {x*}™__ 1 Markov chain is irreducible (since x* > communicate with n = or n = 1). Therefore, by Theorem defined. Moreover, the underlying all states Lucas and Prescott (1989, Proof of Proposition Then it such that x* = for all n > p. 62) there exists a unique stationary distribution , for 11.2 {aC}^1 . There are two cases to consider. First, suppose that {v n } n£Z is from the proof of Proposition 3 that there exists a state n* g N n* and as in the proof of Proposition 8, states n > n* are transient (i.e., for all n>n*. for all n > n* and for all n), so /a* = lim T _ 00 P T (n, n) = > Second, in contrast to the first case, suppose that there exists some n** 6 Z + such that v n —+\. Then, let n* = min n6 {o,...,n"} { n £ N fn+i — v n < ui*G' (0)}, which is again well defined. for all states with n < n*, and Then, optimal R&D decision (29) immediately implies that x* > T for all n > n* and for all n, since x* = 0, all states n > n* are transient and lmv^oo P (n, h) = strictly increasing. 9. follows , iv : . completing the proof. Proof of Proposition The proof 10. follows closely that of Proposition 6. In particular, again using uniformization, the maximization problem of an individual firm can be written as a contraction mapping tp = max n 2x {r) similar to (55) there. + max n {rj n} < n } is finite, since The 00 (this each rj finiteness of the transition probabilities follows, since is n 6 a consequence of the fact that x defined in (11) is and by assumption, there exists n < 00 such that K+ Z — M+ Theorem 16.31, p. This contraction mapping uniquely determines the value function v Maximum Theorem ip n [z] : > (£ | £j < finite and = r\^). r\ n . and Border, 1999, 539) again implies that each of A n (z) for n € Z_\ {0} and n (z) for n G Z is upper hemi-continuous in z = (w, a, 5c), and moreover, since v n for n € Z is concave in a n and x n the maximizers of v [z], A = {^4- n }^Li and X = {^n}^^^) are nonempty, compact and convex- valued. Berge's (Aliprantis X , Proof of Proposition KKLoo 11. The proof follows that of Proposition 7 closely. Fix (*, {«n}i-oo>. and define Z = [0, 1] x IT=-i [0, 1] x ir=-oo [0,x]. Again by Tychonoff's Theorem, Z is compact in the product topology. Then consider the mapping $: Z =3 Z are defined in Proposition 10. constructed as $ = (v?, A, X), where ^ is given by (39) and and Clearly $ maps Z into itself. Moreover, as in the proof of Proposition 7, Z is nonempty, convex, and a subset of a locally convex Hausdorff space. The proof of Proposition 10 then implies that $ has closed graph in the product topology and is nonempty, compact and convex-valued in z. Consequently, the Kakutani-Fan-Glicksberg Fixed Point Theorem again applies and implies that $ has a fixed point z* € $ (z*). The argument that the fixed point z* corresponds to a steady-state equilibrium is identical to that in Proposition 7, and follows from the fact that within argument identical to that of Lemma 1, -1 implies that Xq > 0. The result that u* < 1 then follows immediately. G'" 1 ((l — A ) / (p + ??i)) > z = , A Finally, as in the proof of Proposition 7, either rj 1 > 41 or x!_ x > X is sufficient for g* > 0. References Aghion, Philippe, Nick Bloom, Richard Blundell, Rachel Griffith and Peter Howitt (2005) "Competition and Innovation: an Inverted-U Relationship" Quarterly Journal of Economics, 120, pp. 701-728. 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, sis: Charalambos D. and Kim C. Border (1999) New Hitchhiker's Guide, Springer- Ver lag, Infinite Dimensional Analy- York. Arrow, Kenneth (1962) ''Economic Welfare and the Allocation of Resources for In- vention" in R.R. Nelson, ed. Universities-National Bureau of Economic Research Conference Series; The Rate and Direction of Economic ton University Press, New Economic and Social Factors, Prince- York. Baye, Michael R., and Heidrun C. seeking, innovation, Activities: Hoppe (2003) "The strategic equivalence of rent- and patent-race games," Games and Economic Behavior, 44, 217-226. Boldrin, Michele and David K. Levine (2001) "Perfectly Competitive Innovation," Univesity of Minnesota and UCLA mimeo. Boldrin, Michele and David K. Levine (2004) "Innovation and the Size ket," Univesity of Minnesota and UCLA of the mimeo. Budd, Christopher, Christopher Harris and John Vickers (1993) "A Model Evolution of Duopoly: Does the Review of Economic Studies, Asymmetry Between Firms Tend and Product of the to Increase or Decrease?" 60, 540-573. Bureau of Economic Analysis (2004) come Mar- Accounts" Bureau of "Annual Revision of the National Economic Analysis News. BEA In- 04-31. www.bea.gov/bea/rels.htm Choi, Jay P. (1991) "Dynamic Rand Journal R&D Competition under Hazard Rate Uncertainty," of Economics, 22, 596-610. Fudenberg, Drew, Richard Gilbert, Joseph 42 Stiglitz and Jean Tirole (1983) "Pre- emption, 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, it Nancy T. and Suzanne Scotchmer (2002) "Intellectual Property; the Best Incentive Mechanism?" Innovation Policy and the Economy, Gilbert, Richard, 2, When is 51-78. 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 Profit in Sequential Innovation," Journal of Industrial Economics, 26(1), 20-33. Grossman, Gene and Elhanan Helpman (1990) Innovation and Growth Economy, Cambridge, MA, Hall, Is Bronwyn There a Lag?" MIT H., Jerry International Harris, Christopher in the Global Press. Hausman and Zvi Griliches (1988) "Patents and Economic Review, R&D: 27, pp. 265-83. and John Vickers (1985) "Perfect Equilibrium in a Model of Race" Review of Economic Studies, 52, 193-209. Harris, Christopher and Economic John Vickers (1987) "Racing with Uncertainty" Review of Studies, 54, 1-21. Hopenhayn, Hugo and Matthew Mitchell (2001) Breadth" RAND Journal of Economics, 32, 152-166. Judd, Kenneth (1999) Numerical Methods Kelley, "Innovation, Variety and Patent John in Economics, MIT L. (1955) General Topology, Springer- Verlag, Klemperer, Paul (1990) "How Broad Should the Scope New Press, Cambridge. York. of Patent Protection Be?" Jour- nal of Industrial Economics, 21, 113-130. Kortum, Samuel (1993). "Equilibrium American Economic Review, Llobet, Gerard, Sequential Innovators: R&D and the Patent-R&D Ratio: US Evidence" 83, pp. 450-457. Hugo Hopenhayn and Matthew Prizes, Patents Mitchell (2001) "Rewarding and Buyouts," Federal Reserve Bank of Minneapolis Research Department Staff Report 273, mimeo. Moscarini, Giuseppe and Francesco Squintani (2004) "Competitive Experimentation with Private Information," Cowles Foundation Discussion Paper, No. 1489. National Science Foundation (2006) "Industrial R&D Employment in the United States and in U.S. Multinational Corporations" Division of Science Resource Statistics. 43 NSF 05-302. Arlington, VA: National Science Foundation, www.bsf.gov/sbe/srs/stats.htm O'Donoghue, Ted, Suzanne Scotchmer and Jacques Frangois 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) Growth" Journal of Economic Growth, 9, "Patents in a Model of Endogenous 81-123. Pakes, Ariel and Zvi Griliches (1980) "Patents and Look." Economic Letters, at the Firm Level: A First pp. 377-381. 5, Quah, Danny (2004) R&D "24/7 Competitive Innovation," Reinganum, Jennifer (1981) "Dynamic Games LSE Mimeo. of Innovation," Journal of Economic Theory, 25, 21-24. Reinganum, Jennifer (1985) "Innovation and Industry Evolution," Quarterly Journal of Economics, 100, 81-100. Romer, Paul M. (1990) "Endogenous omy, Technological Change" Journal of Political Econ- 98, S71-S102. Ross, Sheldon (1996) Stochastic Processes, John Wiley Scotchmer, Suzanne (1999) "On the Optimality & Sons, Inc. of the Patent System," The Rand Journal of Economics, 30, 181-196. Scotchmer, Suzanne (2005) Innovations and Incentives, MIT Press, Cambridge, Massa- chusetts. Segal, Ilya tional Bureau of and Michael Whinston (2005) Economic Research Working Paper. Taylor, Curtis (1995) "Digging American Economic Review, Tirole, "Anti-Trust in Innovative Industries," Na- golden carrots: an analysis of research tournaments," for 85, 872-890. 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 Economic Dynamics, Harvard University Press. 44 E. Prescott (1989) Recursive Methods in Table Full IPR 1. Full Protection Benchmark Results IPR Optimal Uniform Optimal State State IPR Dependent Dependent Protection without with licensing licensing Optimal with without with licensing licensing licensing 1.05 1.05 0.35 0.35 A 1.05 1.05 1.05 7 0.35 0.35 0.35 Vi 0.054 V2 0.005 Vz Vi V5 CO 3.52 CO C2 CO 3.52 CO 0.82 C3 CO 3.52 CO 1.94 C4 CO 3.52 CO 1.95 Cs CO 3.52 CO 1.95 - 2.70 3.52 1.71 1.59 1.96 zil x 0.018 0.021 0.023 0.010 0.026 0.029 0.033 0.023 0.022 0.024 Mo 0.238 Ml 0.326 0.481 0.506 0.195 0.446 M2 0.189 0.253 0.253 0.135 0.232 W* 0.952 0.931 0.936 0.961 0.937 0.032 0.026 0.027 0.028 0.039 0.0186 0.0258 0.0263 0.0203 0.0296 V\ ' Ci Do 0.440 Researcher ratio 9* Note: This table gives the results of the benchmark numerical computations with p 7 = 0.35 under five different the difference in the values V\ R&D IPR — policy regimes. vq\ the rate of neck-and-neck competitors, fraction of industries at a technology force working in research; state-dependent IPR xj$; gap of n and the growth policies R&D It = 0.05, A = 1.05, reports the steady-state equilibrium values of rate of a follower that is one step behind, X*_±; the fraction of industries in neck-and-neck competition, — 1; the value of "labor share," rate, g*. It also reports the with or without licensing. See text for 45 U)*; ^ij-J; the ratio of the labor growth-maximizing uniform and details. Table Full IPR Optimal State State IPR Dependent Dependent with without with licensing licensing licensing Protection without with licensing licensing 1.01 Optimal Uniform IPR Full Protection A= 2. Optimal A 1.01 1.01 1.01 1.01 1.01 7 0.35 0.35 0.35 0.35 0.35 m 0.272 % 0.017 V3 V4 % Ci oo 0.19 C2 CO 0.19 CO 0.04 C3 CO 0.19 CO 0.10 C4 CO 0.19 CO 0.10 Cs CO 0.19 CO 0.10 -v *U CO 0.14 0.19 0.09 0.08 0.10 0.090 0.106 0.107 0.053 0.126 x 0.139 0.162 0.107 0.103 0.115 Mo 0.245 Mi 0.330 0.496 M2 0.186 CJ* 0.991 v1 0.452 0.501 0.191 0.455 0.251 0.251 0.137 0.232 0.987 0.987 0.993 0.988 0.008 0.007 0.007 0.007 0.010 0.0186 0.0257 0.0257 0.0203 0.0293 Researcher ratio 5* Note: This table gives the results of the benchmark numerical computations with p 7 = 0.35 under five different the difference in the values V\ R&D IPR — policy regimes. vq; the R&D It = 0.05, A — 1.01, reports the steady-state equilibrium values of rate of a follower that is one step behind, x*_^\ the rate of neck-and-neck competitors, Xq\ fraction of industries in neck-and-neck competition, fraction of industries at a technology force working in research; state-dependent IPR gap of and the growth policies n = 1; rate, g* . the value of "labor share," It also the ratio of the labor reports the growth-maximizing uniform and with or without licensing. See text for 46 ui*\ //j-j; details. Table Full IPR 3. A= IPR Full Optimal Uniform Optimal State State IPR Dependent Dependent with without with Protection Protection 1.2 Optimal without with licensing licensing licensing licensing licensing A 1.20 1.20 1.20 1.20 1.20 7 0.35 0.35 0.35 0.35 0.35 Vi 0.014 V2 0.002 m 0.001 V4 V5 Ci CO 24.85 CO C2 oo 24.85 CO 5.69 C3 CO 24.85 CO 12.25 C4 CO 24.85 CO 14.75 C5 oo 24.85 CO 14.98 -V xU 20.11 24.85 13.69 12.40 14.99 0.005 0.006 0.007 0.003 0.008 * 0.008 0.010 0.007 0.006 0.007 Mo 0.221 Ml 0.318 0.452 0.523 0.206 0.445 fJ-*2 0.199 0.260 0.261 0.142 0.236 U* 0.806 0.743 0.783 0.847 0.760 0.068 0.056 0.070 0.058 0.090 0.0186 0.0261 0.0284 0.0200 0.0306 Vi x 0.418 Researcher ratio 9* Note: This table gives the results of the benchmark numerical computations with p 7 = 0.35 under five different the difference in the values v\ R&D IPR — policy regimes. vq; the R&D It — 0.05, A — 1.2, reports the steady-state equilibrium values of rate of a follower that is one step behind, x*_i\ the rate of neck-and-neck competitors, Xq\ fraction of industries in neck-and-neck competition, /ig; fraction of industries at a technology force working in research; state-dependent IPR gap of and the growth policies n = 1; rate, g* . the value of "labor share," It also the ratio of the labor reports the growth-maximizing uniform and with or without licensing. See text for 47 uj*; details. Table Full IPR Full 4. IPR Protection Protection without with licensing licensing 7= 0.1 Optimal Uniform Optimal State State IPR Dependent Dependent with without with licensing licensing licensing Optimal A 1.05 1.05 1.05 1.05 1.05 7 0.1 0.1 0.1 0.1 0.1 0.291 ^71 V2 V3 V4 V5 V\ Cl 00 3.11 00 C2 00 3.11 00 0.77 C3 00 3.11 00 1.90 C4 00 3.11 00 1.90 C5 00 3.11 00 1.90 2.20 3.11 1.64 0.48 1.90 . -Vq 0.022 0.023 0.024 0.016 0.025 x 0.024 0.026 0.024 0.021 0.024 Mo 0.314 Ml 0.333 0.497 0.501 0.096 0.492 M2 0.172 0.251 0.251 0.058 0.248 W* 0.944 0.916 0.917 0.981 0.917 0.008 0.007 0.008 0.003 0.010 0.0186 0.0278 0.0280 0.0220 0.0286 »i'i 0.777 Researcher ratio 9* Note: This table gives the results of the benchmark numerical computations with p 7 = 0.1 under five different difference in the values v\ IPR — — 0.05, force working in research; state-dependent IPR vq; the R&D gap of n = and the growth policies = 1.05, policy regimes. It reports the steady-state equilibrium values of the rate of a follower that is one step behind, x*_^\ the rate of neck-and-neck competitors, Xq\ fraction of industries in neck-and-neck competition, of industries at a technology A 1; the value of "labor share," rate, g*. It also reports the with or without licensing. See text for 48 to*; /1q; R&D fraction the ratio of the labor growth-maximizing uniform and details.. Table Full IPR Full Protection 7= 5. IPR Protection without with licensing licensing 0.6 Optimal Uniform Optimal State State IPR Dependent Dependent Optimal with without with licensing licensing licensing A 1.05 1.05 1.05 1.05 1.05 7 0.6 0.6 0.6 0.6 0.6 Vi 0.070 0.007 V2 0.019 Vz 0.007 V4 0.003 V5 Ci 00 5.24 5.24 00 C2 00 5.24 5.24 00 Cs 00 5.24 5.24 00 1.69 C4 00 5.24 5.24 00 2.13 Cs 00 5.24 5.24 00 2.35 - 0.59 4.38 5.24 5.24 2.25 s-1 0.011 0.015 0.015 0.002 0.027 x* 0.053 0.070 0.070 0.020 0.021 Mo 0.092 0.327 0.037 Mi 0.257 0.413 0.413 0.094 0.242 M2 0.193 0.252 0.252 0.074 0.128 w* 0.942 0.937 0.937 0.9320 0.9307 0.073 0.044 0.044 0.087 0.010 0.0186 0.0198 0.0198 0.0230 0.0303 Vl v . 2.35 Researcher ratio 5* Note: This table gives the results of the benchmark numerical computations with p 7 — 0.6 under five different difference in the values V\ — IPR = 0.05, vq\ the of industries at a technology gap of state-dependent IPR R&D n — and the growth policies = 1.05, policy regimes. It reports the steady-state equilibrium values of the rate of a follower that is one step behind, x*_^\ the rate of neck-and-neck competitors, Xq\ fraction of industries in neck-and-neck competition, force working in research; A 1; the value of "labor share," rate, g* . It also fraction the ratio of the labor reports the growth-maximizing uniform and with or without licensing. See text for 49 oj*; /j,q\ R&D details. R&D Efforts Optimal Uniform IPR Optimal State-Dependent IPR Technology Gap Figure 5. R&D Efforts. Industry Shares 0.5 Optimal Uniform IPR Optimal State-Dependent IPR Technology Gap Figure 600k 6. Industry Shares. 52 36