World Development Vol. 46, pp. 66–78, 2013� 2013 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevhttp://dx.doi.org/10.1016/j.worlddev.2013.01.023

Innovation, Intellectual Property Rights, and Economic

Development: A Unified Empirical Investigation

JOHN HUDSONUniversity of Bath, UK

and

ALEXANDRU MINEA *

University of Auvergne, France

Summary. — Two important strands of literature investigate the way the effect of intellectual property rights (IPR) on innovation de-pends on either the initial IPR level or the level of economic development. We expand on this by studying their joint effect, in a single,unified, empirical framework. We find that the effect of IPR on innovation is more complex than previously thought, displaying impor-tant nonlinearities depending on the initial levels of both IPR and per capita GDP. The policy implications of this are examined andinclude the conclusion that a single global level of IPR is in general sub-optimal.� 2013 Elsevier Ltd. All rights reserved.

Key words — intellectual property rights, innovation, economic development, nonlinearities

* We gratefully acknowledge the comments of two anonymous referees

and the editor (O. Coomes), which have substantially improved the paper,

as well as N. Apergis, N. Ary Tanimoune, G. Colletaz and C. Hurlin for

comments on a previous version of the manuscript. We thank the FERDI

(Fondation pour les Etudes et Recherches sur le Developpement Inter-

national) and the ANR (Agence Nationale pour la Recherche) for their

financial support through the “Grand Emprunt” and the LABEX IDG-

M+ mechanism. Of course, any errors are our responsibility alone. Finalrevision accepted: January 16, 2013.

1. INTRODUCTION

Among the different engines of economic growth, none hasreceived as much attention as innovation. In recognition of theimportance of innovation for promoting economic develop-ment, the World Trade Organization (WTO), along with otherinternational institutions, emphasized several decades ago thecrucial role of intellectual property rights (IPR) for enhancinginnovation worldwide. Since then, the study of the relation be-tween IPR and innovation has become a prominent topic ineconomic research. In an excellent recent survey, Park(2008a) identifies three main questions that have been ad-dressed in the literature: (i) how do IPR affect the compositionof technology transfer by mode of entry, (ii) does the impact ofIPR on innovation vary by the stage of economic develop-ment, and (iii) do stronger IPR stimulate innovation?

The goal of the present contribution is to expand on theselatter two issues in a single, unified, empirical analysis linkinginnovation, IPR and the level of economic development. Tothis end, we provide econometric estimations based on panelsmooth threshold regressions (PSTR), a technique recentlydeveloped by Gonzalez, Terasvirta, and van Dijk (2005). Spe-cifically we employ an identification procedure to search forendogenously estimated threshold effects of IPR on innova-tion, on a sample of both developed and developing countries.Contrary to previous studies that constrained the shape of theimpact of IPR (and of the level of economic development) oninnovation to simple polynomial forms, the identification pro-cedure allows for a high complexity of the nonlinearities. In par-ticular, since the transition between regimes is smooth, theinnovation/IPR elasticity differs across countries and time-peri-ods, with respect to the levels of both IPR and economic devel-opment. This is particularly important given the numerousstructural reforms in many countries during the 90s. 1

We find firstly that the influence of IPR on innovation isnonlinear, depending on the IPR level. We emphasize twotypes of nonlinearity. On the one hand, stronger IPR wouldincrease innovation in countries with either relatively low orrelatively high initial IPR, and decrease it in other countries.

66

Secondly, we enlarge this analysis to allow for the presenceof a measure for economic development, namely the per capitaGDP level. The tests used in the identification procedure con-firm that the level of per capita GDP also exerts a nonlinearinfluence on the innovation/IPR relationship.

Failing to account for both variables (IPR and per capitaGDP) could produce biased evaluations of the effect of aworldwide IPR strengthening of innovation. For countrieswith low IPR levels, we find a family of curves, dependanton the per capita GDP level, between IPR and innovation.Such a result complicates the task of policymakers that aimat maximizing innovation, since the optimal IPR level is con-tingent on the level of economic development, which is ofcourse endogenous. For countries with high IPR levels, weshow that stronger IPR increase innovation, provided thatthe per capita GDP level is above a certain threshold. Thishas important policy implications. Firstly, the same level ofIPR has a different impact on richer countries than poorerones, as does strengthening IPR, and in addition, given posi-tive adjustment costs a single minimum standard for IPR asin TRIPS is unlikely to be optimal either globally or for indi-vidual countries. Similarly fixed time adjustment periods arealso unlikely to be satisfactory.

The paper is organized as follows. Section 2 reviews the lit-erature. We then describe the PSTR methodology and presentthe dataset. In Sections 5 and 6, we present the results, thenconsider those results and finally conclude the paper.

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 67

2. THE LITERATURE

Early influential theoretical contributions in the “optimalpatent literature” (Cadot & Lippman, 1995; Horowitz &Lai, 1996; O’Donoghue & Zweimuller, 2004; Scotchmer &Green, 1990) find the existence of an inverted-U curve betweenthe level of IPR and innovation, a result supported by morerecent theoretical analysis (see, among others, Furukawa,2007, 2010; Futagami & Iwaisako, 2007; Horii & Iwaisako,2007). 2 Thus, the optimal, innovation-maximizing, IPR levelshould solve a trade-off between the positive (a higher abilityto appropriate R&D investment-based profits, larger markupsthat are an incentive for further innovation, etc.) and negative(reduced competition from the higher blocking of rival entry,higher transaction costs for licensing, etc.) effects of tighterIPR. Based on these results, an important empirical literaturehas investigated the existence of an inverted-U curve betweenIPR and innovation. Kanwar and Evenson (2003) show notonly that tighter IPR always enhance innovation (as in Schnei-der, 2005; Varsakelis, 2001), but also that the increase in inno-vation is amplified as the level of IPR increases, a resultconfirmed by Kanwar (2007). Allred and Park (2007) add tothe complexity of the nonlinear relation between IPR andinnovation by emphasizing a U-shaped curve. Hence, thereis some disagreement on the nature, if any, of the relation-ship. 3

We turn now to the second point, namely the possible influ-ence of the level of economic development on the relationshipbetween IPR and innovation. Compared to the optimal patentliterature, this point has generated even more controversy,especially since the adoption of the TRIPS agreement in themid 90’s, which has been criticized for favoring developed overdeveloping countries, by implementing stronger IPR world-wide. The potential conflicting effects of tighter IPR for theindustrialized (“North”), compared to developing (“South”)countries, go back at least to a strand of several influential the-oretical papers from the beginning of the 1990’s (see, Chin &Grossman, 1991; Deardorff, 1992; Diwan & Rodrik, 1991;Helpman, 1993). 4 For example, Chin and Grossman (1991)show that, while industrialized countries benefit from risingIPR, developing countries may find themselves penalized,especially since higher IPR may result in higher monopolyprices and lower welfare. Consequently, they conclude that aworldwide IPR tightening should go along with some compen-sation mechanism, in the form of a transfer from North toSouth countries. Similar conclusions are reached by theabove-mentioned papers, 5 although the underlying mecha-nisms delimitating “good” versus “bad” effects of tighter IPRin developing countries may be different, and depend on: theimitation cost of innovations by the South countries for Dear-dorff (1991) or Glass and Saggi (2002), trade (linked to FDI)and absorption capacities (particularly regarding the existenceof a sufficient stock of human capital) for Maskus and Penubarti(1995), Lai (1998) and Grossman and Lai (2004), or a Chamber-lain-based horizontal differentiation in the variety of productsfor Diwan and Rodrik (1991) and Lai and Qiu (2003). 6

The empirical attempts to estimate the effect of the level ofeconomic development on the relation between IPR and inno-vation can be classified into roughly three categories. First,several studies estimate the innovation/IPR elasticity ingroups of countries defined using an ex-ante classification.Schneider (2005) shows that IPR have a positive effect oninnovation in developed countries and a negative effect indeveloping countries. This result was partially questioned byAllred and Park (2007) who fail to find any significant effectfor developing countries, but a U-shaped curve in developed

countries. However, these papers can be criticized (a) for usingan ex-ante distinction between developing and developedcountries, and (b) for not allowing a direct role of the levelof economic development on the innovation/IPR elasticity. 7

A second group of papers deals precisely with the first of theseproblems. Ginarte and Park (1997) and Park and Ginarte(1997) highlight different effects of IPR on innovation forcountries below and respectively above the median value ofthe GDP level. However, there are no tests that discuss therobustness of this exogenously determined threshold, 8 andthe innovation/IPR elasticity is constant within groups (no ex-plicit influence of the level of economic development).

Finally, several papers aim at illustrating a direct effect ofthe level of economic development on the link between IPRand innovation (the critique (b) above). Among them, Chenand Puttitanun (2005) outline a nonlinear dependence of theinnovation/IPR relation on the GDP level, within a sampleof developing countries. However, the authors use a simplefirst-order polynomial interaction term (between IPR andthe GDP level) to account for the nonlinear influence ofGDP, which suffers from the usual limitations (for example,an ex-ante specification of the shape of the relation, no testsfor higher-order polynomials, 9 etc.), and their analysis focusesexclusively on developing countries.

3. METHODOLOGY

The Panel Smooth Threshold Regression (PSTR) model, re-cently developed by Gonzalez et al. (2005), can be seen as anupgrading of two existing techniques. Firstly, a generalizationto panel data of thresholds with smooth transition used in timeseries (Chan & Tong, 1986). Secondly, a generalization tosmooth transitions of panel threshold models (PTR) with bru-tal transitions. In the following, we present the PSTR methodfrom the latter perspective. PTR models were introduced byHansen (1999), as a tool to estimate threshold effects on paneldata. Assuming a panel model with i ¼ 1;N countries andt ¼ 1; T years, the simplest PTR model with one threshold is

INNOV it ¼ ai þ b1IPRit þ b2IPRitCðQit; QÞ þPJ

j¼1/jZj

it þ eit

Cð:Þ ¼ 0; if Qit < Q

1; if Qit P Q

(8>><>>: ;

ð1Þ

where ai are country fixed effects, Zjit stands for additional

explanatory variables and eit is an error term. The impact ofIPR on innovation (INNOV) depends nonlinearly on the val-ues of the transition variable Q with respect to the threshold Q:dINNOV/dIPR = b1 if Qit < Q (regime 1) and dINNOV/dIPR = b1 + b2 if Qit P Q (regime 2).

The PTR model suffers from two important shortcomings.Firstly, the presence of a brutal transition between regimes.This is rather problematic in general, and particularly forour analysis, since it would be hard to justify such large struc-tural differences in the impact of IPR on innovation in coun-tries with fairly close IPR or per capita GDP levels.Secondly, even in the presence of multiple thresholds, thePTR model still allows for only a limited number of regimes,which may be considered as unrealistic when one deals withpanel data with (i) an important time dimension (in our anal-ysis, 30 years), and (ii) potential heterogeneity among coun-tries. The PSTR technique overcomes these problems.

We assume the following basic PSTR model

68 WORLD DEVELOPMENT

INNOV it ¼ ai þ b1IPRit þ b2IPRitCðQitÞ þPJ

j¼1/jZj

it þ eit

CðQit; c;QhÞ ¼ 1þ exp �cQ3

h¼1ðQit � QhÞ� �h i�1

8<: :

ð2ÞThis allows for just one transition function, although in prac-tice we will potentially allow for Ck (k ¼ 1;K) transition func-tions, each dependant on the transition variable Qit, on up toh ¼ 1; 3 (as suggested by Gonzalez et al., 2005) endogenouslyestimated thresholds, and on a smoothing parameter (c P 0).

The properties of the transition function, and of the PSTRmodel, crucially depend upon the (nonnegative) transitionparameter c. When c = 0, the transition function is constant(it equals 0.5), and the PSTR model collapses to a linear mod-el. In the case where c ! 1 the transition function takes onlytwo independent values, and the PSTR model collapses to aPTR model with brutal transition between regimes. 10 Theintermediate case (0 < c < 1) represents a PSTR model withsmooth transition between regimes (the lower the smoothingparameter, the smoother the transition).

As usual with nonlinearities, the first operation consists oftesting for their existence. This can be done by computing afirst-order Taylor-linearization of the transition functionaround the smoothing parameter c ! 0. For simplicity, con-sider a simple formulation of (2) with one transition functionand one threshold. Using the first-order linearization, namelyCð:; c! 0Þ ¼ 0:5þ cðQit � QÞ=4, we find

INNOV it ¼ ai þ b�1IPRit þ b�2IPRitQit þXJ

j¼1

/jZjit þ eit

b�1 � b1 þ 0:5b2 � cb2Q=4; b�2 � cb2=4

8><>: : ð3Þ

Under this transformation, the PSTR model collapses to afirst-order polynomial nonlinear panel model. Moreover, sinceb�2 is a multiple of c, we can test the absence of nonlinearities inthe PSTR model (H0:c = 0) using the test H 0 : b�2 ¼ 0. This re-sult also holds in the presence of two thresholds. 11 We followthe related literature and use two tests, namely the LagrangeMultiplier-based (LM) test and its Fisher (F) version. 12

To identify a PSTR model, we use the following cascadeprocedure. In the first phase, we consider a PSTR model withone transition function and up to three thresholds, as sug-gested by Gonzalez et al. (2005). Based on the Eqn. (3), wecompute linearity tests, for each number of thresholds. Fol-lowing Terasvirta (1994), the number of thresholds for the firsttransition function is chosen according to the strongest rejec-tion of the null hypothesis (linear model). Once the numberof thresholds for the first transition function is identified, wecheck for the presence of a second transition function againwith up to three thresholds. The procedure stops when the nullhypothesis of linear effects can no longer be rejected. Conse-quently, contrary to previous papers, which employ polyno-mial forms to account for nonlinearities, our approachincluding the allowance for multiple thresholds and transitionsis less restrictive.

Next, we proceed to the estimation of the PSTR model (i.e.,the threshold(s), the transition parameter(s), and the slopes),in a three-step procedure. The first step eliminates fixed effects,an operation which can lead to some complications. 13 The sec-ond step identifies starting values for the parameters c and Qh

for use in the convergence algorithm in the last step. For eachvector, we use the traditional panel least squares technique toestimate the coefficients b1;2 and /j. The vector of initial valuesfor ðc; QhÞ, is chosen to minimize the sum of squared residualsusing a “grid search” technique. The third and last step of the

algorithm estimates the PSTR model using nonlinear leastsquares; the final estimators for the parameters ðc; QhÞ are thenused in the panel least squares regression to estimate the slopesb1;2 and /j.

4. THE DATASET

We conduct the analysis on a dataset of 62 developed anddeveloping countries (see Appendix 1 for the list of countries),for the 1980–2009 period. To capture medium-term effects andaccount for the fact that changes in IPR take time, we use5-year averages, leading to six observations for each country.In this way our results are easily comparable to other studiesthat also use 5-year averages. 14 We divide variables into pri-mary and control. In the primary group, we consider innova-tion, intellectual property rights, and economic development.Concerning innovation, there are mainly two ways to measureit: (i) R&D expenditure (an input) and (ii) the number of pat-ents (an output). Since the present empirical study requires alarge time-series dataset that includes both developing anddeveloped countries, we use the number of US patents per ca-pita granted to residents of a given country each year. Toavoid a selection bias in favor of US innovations, we followprevious studies (Porter & Stern, 2000) and omit the UnitedStates of America from the dataset (Appendix 2 provides a de-tailed description of all variables and sources, and Appendix 3illustrates descriptive statistics). This does result in a numberof zero values for this variable, although the use of 5-years’averages largely mitigates this problem to less than 10% ofthe observations. In doing this we are being consistent withprevious work. Ideally we should use the Tobit estimator,but this is not yet possible in the technique we are using. Thisneeds to be borne in mind when interpreting the results.

In general, we made every effort to use the same data as pre-vious studies in the literature, to ensure a degree of compara-bility regarding the choice and definition of control variables.This includes the two unscaled variables FDI and INFRA. 15

Concerning the other primary variables, we again follow theliterature and measure intellectual property rights (IPR) bythe index developed by Ginarte and Park (1997), and updatedby Park (2008b), while we proxy the level of economic devel-opment using real per-capita GDP (GDPCAP). It is importantto emphasize that the IPR index is a constructed not a “mea-sured variable”, and it is based, as previously stated, on Gin-arte and Park’s approach to the assessment of the strength ofpatent regimes. We believe it is the best measure available, butnonetheless it is important to understand its subjective nature.

The choice of control variables is largely determined by ref-erence to the existing literature. In a recent survey, Benhabiband Spiegel (2005) observe that human capital is a key factorinfluencing innovation. Using the Barro and Lee (2010) data-set, we define human capital as the percentage of the totalenrollment among the school-aged population over 15 at thetertiary level (EDUC). Next, openness is found to be a sourceof knowledge and technology transfers (Porter & Stern, 2000);openness (OPEN) is defined as the total trade (exports plusimports) as a percentage of GDP. We consider two measuresof investment. Firstly, inflows of foreign investment enhancethe possibility of technology transfer, which seems to signifi-cantly impact on innovation (see, for example, Grossman &Helpman, 1995), an effect we capture using foreign directinvestment stocks (FDI). Secondly, given that innovation ismore likely to emerge in the presence of good infrastructure(Bardhan, 1995), we include the variable INFRA, defined asthe production of electricity 16. We note that this same proxy

Table 1. Identification of the PSTR model: nonlinearities in the IPR level

First transition function Second transition function (first transition function has two thresholds)

One threshold LM test 20.7*** (5.24e�006) 0.05 (0.827)F test 18.1*** (2.74e�005) 0.04 (0.827)

Two thresholds LM test 59.1*** (9.31e�013) 0.33 (0.848)F test 19.2*** (2.00e�011) 0.13 (0.876)

Three thresholds LM test 52.2*** (4.64e�012) 0.62 (0.891)F test 25.0*** (9.02e�011) 0.17 (0.919)

Note: The tests are based on the linearized form of regression [A6] in Table 2 below. Emboldened values signal the highest rejection of the null hypothesis(namely, a linear panel). p-Values are reproduced in brackets.*** Significance at the 1% level.

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 69

for infrastructure is used by Schneider (2005). Finally, all thesevariables are in logarithms.

5. RESULTS. ESTIMATING THE (NONLINEAR)IMPACT OF IPR ON INNOVATION

In this section, we focus on the effect of IPR on innovation,depending on the initial IPR level. To this end, we assume thatQit = IPRit in the model (2); Table 1 presents identificationtests.

The first column of Table 1 depicts the values of the tests,discussed above, for the first transition function dependingon the IPR level, with respectively one, two, and three thresh-olds. 17 The low p-values associated to each of the tests confirmthe existence of strong nonlinearities, between IPR and inno-vation. Regarding the first transition function, the tests show

Table 2. The impact of IPR on innov

Innovation [A1] [A2] [A3] [A4]

IPR �0.0059** �0.0120*** �0.0107*** �0.0129***

(0.003) (0.0035) (0.0037) (0.0041)IPR2

IPR* CðIPRÞ 0.0579*** 0.0460*** 0.0474*** 0.0490***

(0.0055) (0.0055) (0.0057) (0.006)GDPCAP 0.0186*** 0.0136*** 0.0143*** 0.0144***

(0.0034) (0.0036) (0.0037) (0.0037)EDUC 0.0479*** 0.0489*** 0.0472***

(0.0131) (0.0131) (0.0132)OPEN �0.0035 �0.0039

(0.0030) (0.0030)FDI 0.0015

(0.0014)INFRA

cIPR 9.984 11.17 11.05 10.1IPR1 (in levels) 3.076 3.071 3.063 3.059IPR2 (in levels) 3.076 3.071 3.063 3.059

N/countries 372/62 372/62 372/62 372/62Sum of sq. resid. 0.0472 0.0456 0.0454 0.0453

Note: Regressions [A1]–[A6] were estimated following the algorithm presentestandard errors. N denotes the number of observations. As described in the texttransition between regimes.* Denotes significance at the 10% levels.** Denotes significance at the 5% levels.*** Denotes significance at the 1% levels.

that the highest rejection (i.e., the lowest p-value) is associatedwith two thresholds for both LM and F tests. Thus, we set thenumber of thresholds to two in this first transition function, 18

and search next for a second transition function. Irrespectiveof the number of thresholds considered, all tests accept thenull hypothesis (i.e., the absence of a second transition func-tion). Consequently, the PSTR model that comes out fromthe identification procedure consists of one transition functionwith two thresholds

INNOV it ¼ ai þ b1IPRit þ b2IPRitCðIPRitÞ þPJ

j¼1/jZj

it þ eit

CðIPRit; c; IPRhÞ ¼ 1þ exp �cQ2

h¼1ðIPRit � IPRhÞ� �h i�1

8<: :

ð4ÞWe present in Table 2 different estimations of the model (4).The regressions confirm that the level of IPR plays a key role

ation, subjected to the IPR level

[A5] [A6] [A7] [A8] [A9]

�0.0126*** �0.0155*** �0.0071 �0.0174*** �0.1131***

(0.0042) (0.0047) (0.0051) (0.0056) (0.0237)0.0441***

(0.0106)0.0483*** 0.0553** 0.0580***

(0.0062) (0.0074) (0.0013)0.0147*** 0.0154*** 0.0164*** 0.0218*** 0.0139***

(0.0041) (0.0040) (0.0042) (0.0047) (0.0049)0.0479*** 0.0533*** 0.0732*** 0.0910*** 0.0803***

(0.0138) (0.0151) (0.0149) (0.0168) (0.0166)�0.0039 �0.0036 �0.002 0.0011 �0.002(0.0030) (0.0031) (0.0032) (0.0036) (0.0036)0.0015 0.0013 0.0015 0.0028 0.0024

(0.0014) (0.0015) (0.0016) (0.0018) (0.0017)0.0005 0.0003 �0.0032 �0.0066 �0.0024

(0.0026) (0.0030) (0.0030) (0.0043) (0.0034)

10.26 8.046 13.063 3.067 2.3673.063 3.067 3.726

372/62 372/62 372/62 372/62 372/620.0452 0.0435 0.0457 0.0503 0.0476

d above. Regressions [A6]–[A9] include also period-dummies. (.) Denotes, the two transition points are identical. [A7] shows an equation with brutal

Table 3. Identification of the PSTR model: nonlinearities in the GDP level, conditional on the presence of nonlinearities in the IPR level

Conditional to one transitionfunction in the IPR level

First transition function(in the GDP level)

Second transition function(in the GDP level)

(first transition functionhas two thresholds)

One threshold LM test 11.3*** (7.63e�004) 0.29 (0.593)F Test 12.7*** (4.34e�004) 0.32 (0.573)

Two thresholds LM test 35.7*** (1.77e�008) 0.49 (0.781)F test 16.4*** (1.73e�007) 0.81 (0.446)

Three thresholds LM test 37.3*** (3.98e�008) 1.59 (0.662)F test 11.7*** (2.91e�007) 0.99 (0.400)

Note: The tests, based on the linearized form of regression [B6] in Table 4 below, are performed assuming the presence of nonlinear effects in the IPR level,according to Section 5 above. Emboldened values signal the highest rejection of the null hypothesis (namely, a linear panel). p-Values are reproduced inbrackets.*** Significant at the 1% level.

70 WORLD DEVELOPMENT

in explaining the influence of IPR on innovation. This result isfound to be robust, since considering alternative specificationswith different control variables (see regressions [A1]–[A5] inTable 1), has little effect on both the sign and the significanceof the main variables. In addition, regression [A6] shows thatresults are equally robust to the presence of period dummies.Moreover, the effect of control variables is similar to what isusually found in the related literature; in particular, educationand the GDP level exert a positive and strongly significantinfluence on innovation. 19 Neither FDI nor INFRA are sig-nificant, but this is not unusual in this literature, as for exam-ple in the final two columns of Table 3 in Schneider (2005),who, as we do, allowed for country fixed effects and countryand time fixed effects, respectively. In regression [A7] we show,for purposes of comparison, the use of a model with brutaltransitions, in [A8] and [A9] regressions with a linear and qua-dratic relationship between innovation and IPR. In all threecases the fit of the preceding equations is substantially betterthan any of these alternatives. 20

Thus, our results unveil the presence of positive and negativeeffects of IPR on innovation, where models with brutal transi-tions fail to identify any significant effect linked to just IPR. Ofcourse, the complex nonlinearities we capture between IPRand innovation cannot be captured by simply includingIPR as can be seen from regressions [A8] and [A9]. BothIPR and IPR2 are significant indicating a symmetric U shapedrelationship with innovation, which increases as either IPRstrengthen or weakens. If we exclude IPR2 then IPR is nega-tively significant suggesting that the dominant effect is thattighter IPR has a negative impact on innovation.

How to interpret the result that, although two thresholds areidentified, these turn out to be identical? Going back to Eqn.(2), observe that in estimating two thresholds we are both esti-mating two potentially distinct thresholds for Q1 and Q2, andalso specifying two terms multiplying each other. The term inthe exponential expression is cðQit � Q1ÞðQit � Q2Þ; if the twothresholds are identical, this simplifies to cðQit � QÞ2, resultingin a single minimum point in Figure 1. With two separatethresholds this would become more complex, but still with aunique minimum, located between the two thresholds. Butarguably the more important issue is that in estimating twothresholds we are also specifying a function which can be Uor inverted-U shaped around the threshold point(s). Witheither one or three threshold points this would not be the case.Thus, in contrast to a linear function in the traditional ap-proach, our approach allows for such nonlinearities, and incontrast to using a quadratic in the traditional approach our

functional firm flattens out the marginal impact of IPR at highand low values for IPR as in Figure 1.

We focus on the effect of IPR on innovation:

d INNOVd IPR

¼ b1 þ b2CðIPRÞ þ b2IPRcC2ðC�1 � 1Þð2IPR

� IPR1 � IPR2Þ: ð5ÞThis effect, illustrated in Figure 1, should be distinguishedfrom previous findings. In particular, the U-curve derived byAllred and Park (2007) and the U-curve displayed by Figure 1are very different because the latter concerns the effect of theIPR level on the innovation/IPR derivative, and not on the le-vel of innovation.

Figure 1 emphasizes that IPR have a positive marginal im-pact on innovation at both relatively low and high IPR levels.In between, the marginal impact is negative and the differencesin terms of magnitude are substantial. The presence of a broadrange of nonlinearities leads to interesting policy implications,which we examine later. According to Figure 1, the first turn-ing point corresponds to an IPR level of 1.5. To the best of ourknowledge, this is the first paper emphasizing an inverted Ushaped curve for countries with low to medium IPR, with a lo-cal maximum at 1.5, linking the impact of IPR on innovation,in an empirical analysis performed using an aggregated mea-sure of innovation and at an international level. Later on weshow that the nature of this inverted U curve varies with acountries level of GDP per capita. Moreover, these resultsdo confirm recent findings obtained in a significantly differentsetup based on event analysis (see, for example, Lerner, 2002,2009; Qian, 2007). The curve in Figure 1 is based on regression[A6], but compared with the other regressions in Table 2, thereare no qualitative differences in the effect of IPR on innova-tion, since the derivative describes in all regressions very sim-ilar curves. In addition, the quantitative differences are minor;for example, depending on the regression considered, the firstthreshold is located between 1.41 and 1.56 and the secondthreshold is located between 2.86 and 2.97, showing a high de-gree of stability in our findings.

However, the analysis may be criticized for assuming anidentical curve for all countries, by focusing exclusively onthe IPR level to explain the effect of IPR on innovation.To deal with this issue, we present in the next section a uni-fied analysis, which extends the results in this section to thepresence of another major determinant of the link betweeninnovation and IPR, namely the level of economic develop-ment.

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

IPRdINNOVd /

IPR

Figure 1. The influence of the IPR level on the innovation/IPR derivative/elasticity. Note: derived from regression [A6].

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 71

6. FURTHER RESULTS: INNOVATION, IPR, ANDECONOMIC DEVELOPMENT: A UNIFIED

LANDSCAPE

The previous literature (see, e.g., Allred & Park, 2007; Chen& Puttitanun, 2005; Ginarte & Park, 1997; Schneider, 2005)has also focused on the existence of a second variable funda-mentally affecting the impact of IPR on innovation, namelythe level of economic development. The PSTR model (6) up-grades the PSTR model (4) to include the presence of nonlin-earities in the level of economic development, proxied by percapita GDP:

INNOV it ¼ ai þ b1IPRit þ b2IPRitCIPRðIPRitÞ

þPK

k¼1bk3IPRitC

kðGDPCAP it ; :Þ þPJ

j¼1/jX j

it þ eit

CIPRðIPRit ; cIPR; IPRhÞ ¼ 1þ exp �cIPRY2

h¼1

ðIPRit � IPRhÞ !" #�1

CkðGDPCAP it; ck ;GDPCAP khÞ ¼ 1þ exp �ck

QHkh¼1ðGDPCAP it � GDPCAP k

h� �� ��1

8>>>>>>>><>>>>>>>>::

ð6Þ

In addition to a first transition function CIPR, model (6) cap-tures the possible existence of k further transition functionsdepending, this time, on the per capita GDP. To check forthe existence of nonlinearities on the GDP per capita level,we perform in Table 3 an identification procedure similar tothe one for the IPR level, except that the presence of GDPper capita nonlinearities is done assuming the presence of non-linearities in the IPR level already identified.

Regardless of the number of thresholds, the low p-values dis-played by the tests in the first column confirm that the per ca-pita GDP level exerts a nonlinear effect on the innovation/IPRderivative, in addition to the nonlinear effect of the IPR level.Based on the highest rejection criterion (i.e., the lowest p-va-lue), we again specify a first transition function with twothresholds depending on the GDP level, and with no secondfunction in the GDP level. The results are shown in Table 4.

The existence of nonlinear effects is confirmed by a stronglysignificant coefficient of the variable containing the transitionfunction CGDPCAP, which remains robust in significance, sign,and magnitude when introducing different control variablesand period-dummies (regression [B6]). The relatively low valueof the transition parameter cGDPCAP suggests a smooth transi-tion between regimes, confirming that a PSTR model fits bet-

ter the nonlinear properties of the data compared to a PTRmodel. The results of the control variables are similar toTable 2; in particular, higher levels in education and foreigninvestment encourage innovation. A noticeable change, com-pared to Table 2, occurs in the negative effect of the opennessdegree on innovation, which is now significant. Opening totrade allows a larger potential market and is also a possiblesource of knowledge transfer. But it also involves strongercompetition (Eaton & Kortum, 1996), which may reduce theincentive to innovate. Our results suggest the latter effect dom-inates, given the level of GDP per capita which of course open-ness may also influence. Another difference is the existence ofdifferent transitional thresholds for GDP per capita, althoughthere remains still just the common threshold for IPR.

In this analysis, the impact of IPR on innovation is given by:

d INNOVd IPR

¼ b1 þ b2CIPRðIPR; :Þ

þ b2IPRcIPRðCIPRÞ2ððCIPRÞ�1 � 1Þð2IPR� IPR1

� IPR2Þ þ b3CGDPCAP ðGDPCAP ; :Þ: ð7Þ

The resulting curves relating innovation to IPR are shown inFigure 2. Accounting for nonlinearities on the per capitaGDP level in [B6] has two effects on the innovation/IPR elas-ticity. Firstly, an indirect effect, because it changes the coeffi-cients b1 and b2 in (7). This has shifted the curve to theright, and also dampened it, as can be seen in Figure 2. Thus,accounting for the per capita GDP level mitigates the effects ofan IPR strengthening on innovation, suggesting the presenceof important synergies (between the level of economic develop-ment and the level of IPR) that require a unified analysis. 21

Secondly, and more importantly, there exists a direct effect,which corresponds to the last term of Eqn. (7), i.e.,b3C

GDPCAP(GDPCAP; .). The impact is such that the innova-tion/IPR derivative increases (decreases) when the per capitaGDP level is above (below) a threshold of around $4500 (seeAppendix 4). Since 0 6 CGDPCAP(.) 6 1 and b3 > 0, the directeffect is such as 0 6 b3C

GDPCAP(.) 6 b3. Consequently, asemphasized in Figure 2, the values of the innovation/IPRderivative for all the countries in the sample are located abovethe curve derived from (7) without the b3C

GDPCAP(.) term (i.e.,for CGDPCAP(.) = 0). We can see from the figure that countrieswith low IPR tend to be clustered close to this curve. Slightly

Table 4. The impact of IPR on innovation, subjected to the IPR level and the level of economic development

Innovation [B1] [B2] [B3] [B4] [B5] [B6] [B7]

IPR �0.0037 �0.0072** �0.0067** �0.0088** �0.0094** �0.0075* �0.0101**

(0.0027) (0.0035) (0.0034) (0.0038) (0.0039) (0.0044) (0.0049)IPR*C(IPR) 0.0161*** 0.0142*** 0.0130*** 0.0136*** 0.0135*** 0.0106** 0.0060***

(0.0049) (0.0047) (0.0043) (0.0051) (0.0050) (0.0052) (0.0013)IPR* C(GDPCAP) 0.0341*** 0.0339*** 0.0337*** 0.0346*** 0.0352*** 0.0361*** 0.0105***

(0.0037) (0.0038) (0.0038) (0.0038) (0.0039) (0.0038) (0.0017)GDPCAP 0.0066* 0.0031 0.0039 0.0042 0.0030 0.0031 0.0173***

(0.0035) (0.0036) (0.0036) (0.0036) (0.0040) (0.0040) (0.0040)EDUC 0.0370*** 0.0374*** 0.0347*** 0.0317** 0.0434*** 0.0610***

(0.0118) (0.0120) (0.0121) (0.0127) (0.0136) (0.0143)OPEN �0.0059** �0.0066** �0.0068** �0.0067* �0.0043

(0.0028) (0.0028) (0.0028) (0.0028) (0.0030)FDI 0.0019 0.0019 0.0023* 0.0019

(0.0013) (0.0013) (0.0014) (0.0015)INFRA 0.0017 0.0039 �0.0021

(0.0024) (0.0027) (0.0028)

cIPR 20.21 20.05 25.91 20.05 20.56 19.72 1IPR1 (in levels) 3.284 3.263 3.302 3.341 3.310 3.297 2.360IPR2 (in levels) 3.284 3.263 3.302 3.341 3.310 3.297 3.726

cGDPCAP 2.285 2.297 2.242 2.226 2.186 2.228 1GDPCAP 1 (in levels) 1171.6 1163.2 1118.9 1157.6 1170.5 1110.0 1246.4GDPCAP 2 (in levels) 18032.7 17542.9 17764.3 17587.5 17622.7 18287.0 6210.5N/countries 372/62 372/62 372/62 372/62 372/62 372/62 372/62Sum of sq. resid. 0.0395 0.0385 0.0381 0.0379 0.0379 0.0358 0.0415

Note: Regressions [B1]–[B6] were estimated following the algorithm presented above. Regressions [B6] and [B7] include also period-dummies. (.) Denotesstandard errors. N denotes the number of observations. As in Table 2, the two transition points for IPR are identical. [B7] shows an equation with brutaltransition between regimes.* Denotes significance at the 10% levels.** Denotes significance at the 5% levels.*** Denotes significance at the 1% levels.

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

IPR

Original curve from regression [A6], eq. (5)

The curve from regression [B6], without the

( ).3GDPCAPΓβ term

The curves from regression [B6], with the ( ).3

GDPCAPΓβterm, eq. (7)

IPRdINNOVd /

Figure 2. The joint influence of the IPR and economic development level on the innovation/IPR derivative/elasticity.

72 WORLD DEVELOPMENT

less clearly, richer countries, for which b3CGDPCAP(.) is closer

to b3, tend to be located well above this curve 22. If these richercountries also have higher IPR, the difference is substantial. Aswith Figure 1, the curves from the different equations in Ta-ble 4 are relatively similar, particularly with respect to thedifferent threshold points. The curve in Figure 2 is basedon [B6] as more of the control variables are significant

and because it also includes time fixed effects, in additionto country fixed effects. For consistency this is why [A6]formed the basis for Figure 1. Finally, note that the quan-titative results from [B6] can be seen as a lower bound, sincethe magnitude of the responsiveness on innovation to differ-ing levels of IPR is larger for the other PSTR regressions,and particularly for [B2] and [B3].

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 73

7. DISCUSSION: GENERAL CONSIDERATIONS ANDCRITICAL PATTERNS

As suggested by Figure 2, both the IPR and per capita GDPlevels can generate changes in the sign of the innovation/IPRderivative and hence elasticity. The differences between coun-tries can be substantial. For example, Ecuador and Singaporehave had at various times similar levels of IPR (2.042), butsubstantially different levels of GDP per capita ($4863 com-pared to $16,177). For the former, dINNOV/dIPR equals�0.0114 and for the latter it equals +0.0013; hence, increasingIPR for the former will reduce innovation but increase it forthe latter. The figures for Guatemala and Norway, both attimes with IPR of about 3.150, show even greater differences.For the former with GDP per capita of $5527, dINNOV/dIPRequals �0.0094, whereas for the latter with GDP per capita of$24,381, it equals +0.0125. Hence, the optimal policy for Gua-temala, at least in the short-term, would be to reduce IPR if itwanted to stimulate innovation, while for Norway the reverseis the case.

There are also substantial differences in magnitude. Forexample, Nicaragua and Venezuela have both had IPR atabout 0.922, but for the former, with GDP per capita at only$4165, dINNOV/dIPR is 0.004, whereas for Venezuela, withGDP per capita at $13,252, it is 0.0105. There are even differ-ences over time for the same country. Thus, with an un-changed IPR level of 1.408, dINNOV/dIPR for Egypt in thefirst and the third of the five time periods was 0.047 and0.0106, respectively. Equally, there can be substantial differ-ences for countries with similar levels of GDP per capita;hence Jordan in period three and Tunisia in period five hadvery similar GDP per capita levels, but substantially differentIPR levels, 0.742 and 2.317, respectively, leading to values forthe derivative of +0.0049 and �0.0253.

8. CONCLUSIONS

Contrary to previous papers that looked at the impact ofGDP per capita and IPR on innovation in isolation, we con-sider a unified econometric framework, in which the impactof IPR on innovation was subjected to the joint influence ofinitial IPR and per capita GDP levels. We used the PSTRmethodology to search for nonlinear effects with smooth tran-sitions, between innovation, IPR, and economic development,in a panel of developing and developed countries. We showedthat strengthened IPR exert a complex effect on innovation,both in sign and magnitude, which depends on both the initiallevel of IPR and per capita GDP. These results thus help rec-oncile the differing results noted earlier on the different impactof IPR on innovation in developing and developed countries.Previous papers prescribed an ad hoc transition, by way of aquadratic equation, for example, or used a “brutal transition”between regimes by estimating threshold effects. These previ-ous estimates did have the advantage of simplicity, but theywere a simplification and few would have seriously believedthat these approaches were totally satisfactory. In this sense,our approach is an advance on previous work. The curveswe have identified illustrate a complex interplay betweenGDP per capita and the existing level of IPR in determiningthe nature of the response of innovation to a strengtheningof IPR

From a policy perspective, the first conclusion is that there isa range of values within which no country should set its IPR.Figure 2 suggests that such a range lies between values of 1.8and 3.3. The latest data we have show about one third of

countries lie in this range. The idea that it might be appropri-ate for some countries to, at least in the short term, reducetheir IPR to stimulate innovation also runs against the com-mon view that stronger is always better. Based on these results,consider the effects of a global reform that aims at increasingthe IPR level worldwide, rather than setting a target for a min-imum IPR level. As we can see from Figure 2, countries withan initial IPR level roughly below 1.8 or above 3.3 would ben-efit in terms of innovation, irrespective of the per capita GDPlevel. However, the innovation gain would be significantlystronger as per capita GDP increases, as well as the critical lev-els of IPR moving to the left. Countries with close IPR levelsmay experience significantly different effects because of differ-ent economic development levels. However, poorer countrieswith an initial IPR level roughly between 1.8 and 3.3 wouldprobably experience less innovation as a result of an IPRstrengthening. Thus following a “one size fits all approach”can lead to inappropriate policy recommendations.

The presence of the per capita GDP level (in addition to theIPR level) as a second variable influencing the effect of IPR oninnovation complicates the policy lessons that can be drawnfrom a dynamic perspective. In the short-run, when the levelof economic development changes little, the major differencefrom before is that the innovation/IPR curve (for differentIPR levels) is no longer unique, but individual to each countryaccording to its per capita GDP level. On the one hand, wefind again the existence of what is essentially a U-shaped curvebetween the IPR level and the innovation derivative across therange of values for most countries. From the perspective ofinnovation alone, countries should therefore move their IPRup to the point where innovation is maximized. Hence, coun-tries with relatively high IPR levels should support an eco-nomic policy that consists of tightening IPR, as it booststheir innovation; however, the magnitude of this effect de-pends on the per capita GDP level. 23 The strongest impactof tightened IPR on innovation arises for countries with IPRlevels around 4.5 and per capita GDP levels well above thetwo GDP thresholds. Hence the simple view that a strengthen-ing of IPR is always good for innovation is not consistent withour results.

However, global policy tends not to go for a position of agradual strengthening for everyone, but for a single minimumstandard. In considering the “optimal” level of IPR, it needs tobe recognized that IPR policies may have other impacts on theeconomy and social welfare than those on innovation. Thereare also likely to be adjustment costs in moving to a strongerIPR regime, with firms, skill levels, capital, infrastructure of alltypes, all needing to adjust. In the long run, cognizance alsoneeds to be taken of the impact innovation can have onGDP per capita. Hence, we are no longer faced with anunchanging single optimal level of IPR for each country, butone which evolves. An optimal policy should both specifythe long run optimal level of IPR, which could be the samefor all countries given income convergence, and the time pathtoward that optimum, which would not be the same for allcountries.

From this perspective, and focusing just on the impact oninnovation, and thence GDP per capita, what should that sin-gle optimum be and what should be the approach path to it?Given that there are adjustment costs to strengthened IPR,our analysis suggests a multiplicity of standards, rather thana single one, as being optimal. For example, according toour Figure 2, up to an IPR value of 1.5 most countries, regard-less of their level of GDP per capita, are benefiting from stron-ger IPR, in innovation terms. Hence it could reasonably beargued that this should be the minimum IPR standard. Once

74 WORLD DEVELOPMENT

we move beyond this, we fall into a region where stronger IPRare increasingly a deterrent to innovation. It is not until wereach an IPR value of about 4, with it being lower for richercountries, that the innovation is at the level it is with IPRset at 1.5. Hence, a value close to this should form the secondcritical value. But countries at a level of IPR of 1.5 will be ben-efiting from, locally optimal, innovation. This will be stimulat-ing GDP per capita growth and hence closing the gap betweenthe current target IPR and the second one. By waiting beforemoving to a higher level of IPR, the existing one will stimulateinnovation to a degree which, other things being equal, will in-crease GDP per capita. The increase in GDP per capita swivelsthe curve linking innovation to IPR to the left, which will inturn increase the gains, and hence reduce the net adjustmentcosts, of moving to a higher level. The third and final valuefor IPR should equal maximum IPR, i.e., 4.67, but again aperiod of waiting will stimulate GDP which will in this caseenhance the gains from moving to this third and final targetlevel. The trigger for a shift from one standard to the nextcan be reasonably linked to GDP per capita. Higher levelsof GDP per capita both tend to increase the gains fromstrengthening IPR and shift the second target threshold valueto the left. But since this relates to increasing GDP per capita,the optimal waiting period before shifting from one IPR levelto the next would not be a fixed period of years, but dependupon the evolution of GDP per capita. In reality these waitingperiods may be measured in decades rather than years.

This is a long way from the current global standard basedon TRIPS. In the context of patents, this specifies that theterm of protection lasts for at least 20 years from the filingdate. Developing countries had a short period of grace, whichfor lesser developed countries is still operative, to achieve thisminimum, although there are a number of special circum-stances, e.g., linked to health, which allow deviations fromthe guidelines. This 20 year period is much closer to thatadopted by developed countries prior to TRIPS, with develop-ing countries tended to have a much lower level of protection.In addition, as this is a standard bounded from below, devel-oped countries are free to adopt an even stronger standard fortheir own country. Thus, in this sense, TRIPS is not really acompromise at all for them. In terms of our results, there isjust one optimal level of IPR set at a relatively high value, withinitial periods of grace being based on a fixed period of years,rather than a variable one based on the economic conditions inthe particular country. Given that under our suggestion thetrigger for switching to a higher level of IPR would be basedon the latter, i.e., GDP per capita, the country would notswitch until GDP per capita had exceeded a critical level. Thusthe time periods envisaged could be much longer than thoseinvolved with the TRIPS transformation and importantlythose time periods would not be fixed, but vary from countryto country.

NOTES

1. In addition, contrary to models that consider brutal transitionsbetween regimes (see, for example, Falvey, Foster, & Greenaway, 2006,who study the impact of IPR on economic growth), smooth transitionsavoid unjustified sudden jumps in the innovation/IPR derivative andhence elasticity for countries with fairly close IPR or economic develop-ment levels.

2. Previous to these contributions, the impact of stronger IPR oninnovation was unambiguously positive; see, for example, the theoreticalanalysis of Gilbert and Shapiro (1990), Klemperer (1990), and Kamienand Schwartz (1974).

3. At least two important alternative ways of analyzing the IPR-innovation relation exist. First, the international dimension is omitted incountry-level studies; see, for example, Lanjouw (1998) for India,Sakakibara and Branstetter (2001) for Japan, or Hall and Ziedonis(2001) for the United States. Second, a promising strand of literaturecoined by Lerner (2002) investigates multiple IPR reforms instead of asingle patent policy reform; see, for example, the event-based analysis ofLerner (2002, 2009) who focuses on the impact of major patent policyshifts in 60 nations over the past 150 years, Branstetter, Fisman, andFoley (2006) who study the effect of different IPR reforms on FDI in 16countries over the 1982–1999 period, or Qian (2007) who discusses patentprotection on pharmaceutical innovations in 26 countries that establishedpharmaceutical patent laws during the 1978–2002 period. Although resultsfrom this literature are still crude, we will compare them with some of ourfindings (in particular, the existence of an inverted-U curve betweeninnovation and IPR).

4. These papers follow the intuition of the analysis of Nordhaus (1969).

5. With the notable exceptions of Helpman (1993) and McCalman(2001), where the effect of tighter IPR is ambiguous for North countriesand always detrimental for South countries, a result that fuels even morethe pressure from developing countries against a rise in IPR.

6. In addition, several theoretical papers study the impact of tighter IPRin the South on the innovation incentives of the firms in the North; see, inparticular, Yang and Maskus (2001) or Akiyama and Furukawa (2009).

7. The U-curve derived by Allred and Park (2007) for developedcountries concerns the relation between IPR and innovation (throughthe presence of a second order polynomial in IPR), with no explicitreference to the level of economic development.

8. Kanwar and Evenson (2009) also assume exogenous thresholds whenstudying the impact of the innovation-to-GDP ratio on IPR.

9. Chen and Puttitanun (2005) present equally a U-curve based on semi-parametric estimations, which however concerns the relation between IPRand GDP, with no reference to the innovation/IPR elasticity.

10. Consequently, the PTR model may be viewed as a special case of aPSTR model, namely when c ! 1.

11. Namely, b1�2 ¼ �cb2ðQ1 þ Q2Þ=4 and b2�

2 ¼ cb2=4, respectively. Inthis case, the absence of nonlinear effects H0 :c ¼ 0 can be checked basedon a joint nullity test H0 : b1�

2 ¼ b2�2 ¼ 0. The algebra for a model with

three thresholds is easily computable and available upon request.

12. Denoting by S0 and S1 the sum of squared residuals under the nullhypothesis (linear panel) and the alternative hypothesis (PSTR model)respectively, the tests are computed as LM = TN(S0 � S1)/S1 andF = [(S0 � S1)/H]/[S1/(TN � N � H)] (under the null hypothesis, LM

follows a v(H) distribution, while F follows a F(H, TN � N � H)distribution).

13. To eliminate fixed effects, all variables of the model (2) are centeredas usual around their means, except for the term including the transitionfunction G(.) � IPRitC(.), which becomes: eGð:Þ ¼ Gð:Þ � Gð:Þ, with

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 75

Gð:Þ ¼ T�1PT

t¼1Gð:Þ ¼ T�1PT

t¼1ðIPRitCð:ÞÞ. Consequently, the centeredvariable eGð:Þ, used in the PSTR model without fixed effects, depends onthe c and Qh parameters in both G(.) and Gð:Þ, which is why eG must becomputed at each iteration, i.e. each time that c and Qh take differentvalues.

14. The six sub-periods are 1980–84, 1985–89, 1990–94, 1995–99, 2000–04, and 2005–09.

15. However, allowing for country and time fixed effects does takecognizance of scale effects to an extent.

16. Namely, the production of power plants and combined heat, lesstransmission, distribution, and transformation losses and own use by heatand power plants.

17. Recall that, following Gonzalez et al. (2005), we limit the number ofthresholds to three for each function.

18. We were also aware that with limited data the more transitionfunctions we estimate the greater the problems linked to degrees offreedom and multicollinearity become.

19. The next section will allow for possible nonlinear effects of the percapita GDP level on innovation.

20. For example, according to [A7] the effect of IPR on innovationequals +0.058 for relatively low (below the first threshold 2.367) andrelatively high (above the second threshold 3.726) IPR values, while thereis no significant effect for average IPR values between the two thresholds2.367 and 3.726. Considering that the impact of IPR on innovation cantake only two values is of course much more restrictive compared to thecomplex influence we outline based on PSTR estimations.

21. However, despite such synergies, one should avoid concluding thattighter IPR and better economic development go always hand-in-hand;such a conclusion is refuted by the presence of significant slopes for thevariables containing of both IPR and per capita GDP.

22. Note the lowest value for GDP per capita in our data base is $1377.1.This is above the lowest threshold value. Thus for the very poorest ofcountries the relationship between IPR and innovation may be differentfrom that indicated by our regressions.

23. In some cases, an increase in IPR can contract innovation; indeed,much caution is needed for poor countries with IPR levels around 3.1, inwhich an increase in the IPR level can decrease innovation when the GDPlevel increases to values below the GDP threshold.

REFERENCES

Akiyama, T., & Furukawa, Y. (2009). Intellectual property rights andappropriability of innovation. Economics Letters, 103(3), 138–141.

Allred, B., & Park, W. (2007). Patent rights and innovative activity:Evidence from national and firm-level data. Journal of InternationalBusiness Studies, 38(6), 878–900.

Bardhan, P. (1995). The contributions of endogenous growth theory to theanalysis of development problems: An assessment. Handbook ofDevelopment Economics, 3(2), 2983–2998.

Barro, R., & Lee, J.-W. (2010). A new data set of educational attainmentin the world, 1950–2010. NBER Working Paper 15902, NationalBureau of Economic Research, Inc.

Benhabib, J., & Spiegel, M. (2005). Human capital and technologydiffusion. Handbook of Economic Growth, 1(1), 935–966.

Branstetter, L., Fisman, R., & Foley, F. (2006). Do stronger intellectualproperty rights increase international technology transfer? Empiricalevidence from U.S. firm-level data. Quarterly Journal of Economics,121(1), 321–349.

Cadot, O., & Lippman, S. (1995). Barriers to imitation and the incentive toinnovate. University of California Working Paper 434, University ofCalifornia.

Chan, K., & Tong, H. (1986). On estimating thresholds in autoregressivemodels. Journal of Time Series Analysis, 7(3), 179–190.

Chen, Y., & Puttitanun, T. (2005). Intellectual property rights andinnovation in developing countries. Journal of Development Economics,78(2), 474–493.

Chin, J., & Grossman, G. (1991). Intellectual property rights and North–South trade. NBER Working Paper 2769, National Bureau ofEconomic Research.

Deardorff, A. (1992). Welfare effects of global patent protection. Econo-mica, 59(233), 35–51.

Diwan, I., & Rodrik, D. (1991). Patents, appropriate technology andNorth–South trade. Journal of International Economics, 30(1–2),27–47.

Eaton, J., & Kortum, S. (1996). Trade in ideas patenting andproductivity in the OECD. Journal of International Economics,40(3–4), 251–278.

Falvey, R., Foster, N., & Greenaway, D. (2006). Intellectual propertyrights and economic growth. Review of Development Economics, 10(4),700–719.

Furukawa, Y. (2007). The protection of intellectual property rights andendogenous growth: Is stronger always better?. Journal of EconomicDynamics and Control, 31(11), 3644–3670.

Furukawa, Y. (2010). Intellectual property protection and innovation: Aninverted-U relationship. Economics Letters, 109(2), 99–101.

Futagami, K., & Iwaisako, T. (2007). Dynamic analysis of patent policy inan endogenous growth model. Journal of Economic Theory, 132(1),306–334.

Gilbert, R., & Shapiro, C. (1990). Optimal patent length and breadth.Rand Journal of Economics, 21(1), 106–112.

Ginarte, J., & Park, W. (1997). Determinants of patent rights: A cross-national study. Research Policy, 26(3), 283–301.

Glass, J., & Saggi, K. (2002). Licensing versus direct investment:Implications for economic growth. Journal of International Economics,56(1), 131–153.

Gonzalez, A., Terasvirta, T., & van Dijk, D. (2005). Panel smoothtransition regression models. Research Papers Series 165, QuantitativeFinance Research Center, University of Technology.

Grossman, G., & Helpman, E. (1995). Technology and trade. Handbook ofInternational Economics, 3, 1279–1337.

Grossman, G., & Lai, E. (2004). International protection of intellectualproperty. American Economic Review, 94(5), 1635–1653.

Hall, B., & Ziedonis, R. (2001). The patent paradox revisited: Anempirical study of patenting in the U.S. semiconductor industry, 1979–1995. Rand Journal of Economics, 32(1), 101–128.

Hansen, B. (1999). Threshold effects in non-dynamic panels: Estimation,testing, and inference. Journal of Econometrics, 93(2), 345–368.

Helpman, E. (1993). Innovation, imitation, and intellectual propertyrights. Econometrica, 61(6), 1247–1280.

Horii, R., & Iwaisako, T. (2007). Economic growth with imperfectprotection of intellectual property rights. Journal of Economics, 90(1),45–85.

Horowitz, A., & Lai, E. (1996). Patent length and the rate of innovation.International Economic Review, 37(4), 785–801.

Kamien, M., & Schwartz, N. (1974). Patent length and R&D rivalry.American Economic Review, 64(1), 183–187.

Kanwar, S. (2007). Business enterprise R&D, technological change,and intellectual property protection. Economics Letters, 96(1),120–126.

Kanwar, S., & Evenson, R. (2003). Does intellectual property protec-tion spur technological change?. Oxford Economic Papers, 55(2),235–264.

Kanwar, S., & Evenson, R. (2009). On the strength of intellectual propertyprotection that nations provide. Journal of Development Economics,90(1), 50–56.

76 WORLD DEVELOPMENT

Klemperer, P. (1990). How broad should the scope of patent protectionbe?. Rand Journal of Economics, 21(1), 113–130.

Lai, E. (1998). International intellectual property rights protection andrate of product innovation. Journal of Development Economics, 55(1),133–153.

Lai, E., & Qiu, L. (2003). The North’s intellectual property rightsstandard for the South?. Journal of International Economics, 59(1),183–209.

Lanjouw, J. (1998). The introduction of pharmaceutical productpatents in India: ‘Heartless exploitation of the poor and suffering’?NBER Working Paper 6366, National Bureau of EconomicResearch, Inc.

Lerner, J. (2002). 150 Years of patent protection. American EconomicReview, 92(2), 221–225.

Lerner, J. (2009). The empirical impact of intellectual property rights oninnovation: Puzzles and clues. American Economic Review, 99(2),343–348.

Maskus, K., & Penubarti, M. (1995). How trade-related are intellectualproperty rights?. Journal of International Economics, 39(3–4),227–248.

McCalman, P. (2001). Reaping what you sow: An empirical analysis ofinternational patent harmonization. Journal of International Econom-ics, 55(1), 161–186.

Nordhaus, W. (1969). Invention growth, and welfare: A theoreticaltreatment of technological change. Cambridge: MIT Press.

O’Donoghue, T., & Zweimuller, L. (2004). Patents in a model ofendogenous growth. Journal of Economic Growth, 9(1), 81–123.

Park, W. (2008a). Intellectual property rights and international innova-tion. In K. Maskus (Ed.). Frontiers in economic and globalization(intellectual property, growth and trade) (vol. 2). Amsterdam: Elsevier-North Holland, pp. 289–327.

Park, W. (2008b). International patent protection: 1960–2005. ResearchPolicy, 37(4), 761–766.

Park, W., & Ginarte, J. (1997). Intellectual property rights and economicgrowth. Contemporary Economic Policy, 15(3), 51–61.

Porter, M., & Stern, S. (2000). Measuring the ‘ideas’ production function:Evidence from international patent output. NBER Working Paper7891, National Bureau of Economic Research, Inc.

Qian, Y. (2007). Do national patent laws stimulate domestic innovation ina global patenting environment? A cross-country analysis of pharma-ceutical patent protection, 1978–2002. Review of Economics andStatistics, 89(3), 436–453.

Sakakibara, M., & Branstetter, L. (2001). Do stronger patents inducemore innovation? Evidence from the 1988 Japanese patent lawreforms. Rand Journal of Economics, 32(1), 77–100.

Schneider, P. (2005). International trade, economic growth and intellec-tual property rights: A panel data study of developed and developingcountries. Journal of Development Economics, 78(2), 529–547.

Scotchmer, S., & Green, J. (1990). Novelty and disclosure in patent law.RAND Journal of Economics, 21(1), 131–146.

Terasvirta, T. (1994). Specification, estimation, and evaluation of smoothtransition autoregressive models. Journal of the American StatisticalAssociation, 89(245), 208–218.

Varsakelis, N. (2001). The impact of patent protection, economy opennessand national culture on R&D investment: A cross-country empiricalinvestigation. Research Policy, 30(7), 1059–1068.

Yang, G., & Maskus, K. (2001). Intellectual property rights, licensing, andinnovation in an endogenous product-cycle model. Journal of Interna-tional Economics, 53(1), 169–187.

APPENDIX 1. LIST OF COUNTRIES CLASSIFIEDUSING THE WORLD BANK 2002 CLASSIFICATION

Developing countries: Low Income (Bangladesh, Cameroon,Cote d’Ivoire, Haiti, India, Kenya, Nicaragua, Pakistan, Sen-egal, Zimbabwe); Low-Middle Income (Algeria, Bolivia, Bul-garia, Colombia, Ecuador, Egypt, El Salvador, Guatemala,Honduras, Iran, Jamaica, Jordan, Morocco, Paraguay, Peru,Philippines, Sri Lanka, Syria, Thailand, Tunisia); High-Mid-dle Income (Argentina, Brazil, Chile, Costa Rica, KoreaRep., Malaysia, Panama, Poland, South Africa, Turkey, Uru-guay, Venezuela)

Developed countries: Australia, Austria, Canada, Denmark,Finland, France, Greece, Ireland, Israel, Italy, Japan, Nether-lands, New Zealand, Norway, Portugal, Singapore, Spain,Sweden, Switzerland, United Kingdom)

APPENDIX 3. DESCRIPTIVE STATISTICS

Variable Nb. of obs. Mean Std. dev. Min Max

INNOV 372 0.0212948 0.0448169 0 0.2806478IPR 372 2.71846 1.14374 0.58824 4.66667GDPCAP 372 10518.5 8793.6 1377.1 41829.6EDUC 372 8.948925 7.505213 0.2 42.2OPEN 372 68.40846 50.48219 10.74545 446.0643FDI 372 21.40115 23.51062 0 176.4624INFRA 372 3,732,606 4,879,506 33,213.29 29,877,591

APPENDIX 2. VARIABLES’ DEFINITIONS AND SOURCES

Variables Definition Source

Dependent variableInnovation (INNOV) Average of the number of US patents per

capita granted to residents of a givencountry for the years t to t + 4; 5 yearsaverage (in logarithm).

United States Patent and TrademarkOffice (USPTO)

Explanatory variablesInterest variable

IPR Level of IPR protection; the first year ofeach sub-period (in logarithm)

Ginarte and Park (1997), Park (2008b)

Transition variableGDPCAP Average GDP per capita for the years t

to t + 4, constant 2000 USD; 5 years lagwith respect to the first year of theconsidered sub-period (in logarithm).

Penn World Data Tables 6.2

Control variablesEDUC Share of the population over 15 years old

having achieved tertiary education(in logarithm).

Barro and Lee (2010)

FDI Average FDI stock for the years t to t + 4(in logarithm); FDI stock is thevalue of the share of their capital andreserves (including retained profits)attributable to the parent enterprise,plus the net indebtedness ofaffiliates to the parent enterprises.

Penn World Data Tables 6.2

INFRA The production of power plants andcombined heat, less transmission,distribution, and transformation lossesand own use by heat and power plantsfor the years t to t + 4; 5 yearsaverage (in logarithm).

World Bank Development Indicators

OPEN Average of total of exports and importsas % of GDP for the years t tot + 4 (in logarithm)

Penn World Data Tables 6.2

Note: In order to be able to log zero variables, we normalized variables with a lower bound of zero to a lower bound of one. The results howeverwere not sensitive to alternative specifications.

INNOVATION, INTELLECTUAL PROPERTY RIGHTS, AND ECONOMIC DEVELOPMENT 77

78 WORLD DEVELOPMENT

APPENDIX 4. NONLINEAR EFFECTS OF THE LEVELOF PER CAPITA GDP ON THE INNOVATION/IPR

ELASTICITY

In addition to the IPR level, the level of economic develop-ment, measured by the level of per capita GDP, influences theeffect of IPR on innovation, through the term b3C

GDPCAP(.).According to the chart below, this effect is always positive (it

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0 5,000 10,000 15,000 20,000

(GDPCAP)Γ3β

Figure A1. The influence of the GDPCAP

lies between 0 and b3) and depends nonlinearly on the per ca-pita GDP level. The effect is relatively low in the neighbor-hood of a threshold just above 4,500 USD, and relativelyhigh for low and high capita GDP values, with importantmagnitude differences among in the latter case. Finally, forvery high per capita GDP levels, the contribution of the percapita GDP to the innovation/IPR is almost independent ofthe level of economic development (Figure A1).

25,000 30,000 35,000 40,000 45,000

GDPCAP

level on the innovation/IPR derivative.