Patent rights, product market reforms, and innovation

Abstract

In this paper, we provide empirical evidence to the effect that strong patent rights may complement competition-increasing product market reforms in fostering innovation. First, we find that the product market reform induced by the large-scale internal market reform of the European Union in 1992 enhanced, on average, innovative investments in manufacturing industries of countries with strong patent rights since the pre-sample period, but not so in industries of countries with weaker patent rights. Second, the positive response to the product market reform is more pronounced in industries where, in general, innovators tend to value patent protection higher than in other industries, except for the manufacture of electrical and optical equipment. The observed complementarity between competition and patent protection can be rationalized using a Schumpeterian growth model with step-by-step innovation. In such a model, better patent protection prolongs the period over which a firm that successfully escapes competition by innovating, actually enjoys higher monopoly rents from its technological upgrade.

Appendices

Appendix 1: Additional tables

See Tables 8, 9, 10, and 11.

Table 8 Definitions of main variables and descriptive statistics

Table 9 Variants to model specifications in Table 5

Table 10 Identification using alternative data variation: part 1

Table 11 Identification using alternative data variation: part 2

Appendix 2: Data sources and variables

1.1 Appendix 2.1: Research and development expenditures

Our main measure of innovation is R&D intensity, that is nominal R&D expenditures as a percentage of nominal value added. For calculating that measure we use country-industry level panel data on nominal research and development expenditures from the OECD Analytical Business Enterprise Research and Development (ANBERD) database (edition 2011).⁵⁰

In addition, we use data on nominal value added from the EU KLEMS database (March 2008 release), adapting the set of national currencies in the EU KLEMS 2008 database to the one of the ANBERD 2011 database.⁵¹ We also use ANBERD 2011 data on real R&D expenditures in US dollar purchasing power parities at constant prices of the year 2005 (in billion).

1.2 Appendix 2.2: Patenting

For several purposes, we use country-industry-year-specific US patent counts constructed by the EU KLEMS consortium.⁵² The patent counts are based on the NBER patent data (edition 2002, http://elsa.berkeley.edu/~bhhall/patents.html), specifically on the patents granted by the US Patent and Trademark Office after 1962 and applied for before 2000 (O’Mahony et al. 2008; Hall et al. 2001). Each US patent was assigned to a year according to the application date recorded in the patent document and to a country according to the country of the first inventor. For our main empirical analysis we use the fractional patent counts where each patent is counted in all n OTAF classes it was assigned to with a weight of 1 / n. The patents were assigned to up to 7 OTAF classes per patent and all classes (41 OTAF classes, plus one “other industries” class) were mapped into EU KLEMS 2008 industry classes.⁵³

To construct the knowledge stock built up during the pre-sample period per country-industry we use the US patent data for the pre-sample period, 1980–1986. Applying the perpetual inventory method, we calculate the knowledge stock as the sum of all pre-sample fractional patent counts (in 1000 patents) that are depreciated to the last year of the pre-sample period with an annual knowledge depreciation rate of 20 percent.

1.3 Appendix 2.3: Patent rights

To measure the strength of patent protection, that is intellectual property rights (IPRs) as laid down in patent laws,⁵⁴ we distinguish between different country groups. To do so, we use information on patent law reforms, as well as related regulation, and data on a time period with high variation of patent protection across European Countries, that is a time period before international harmonization of patent systems started to dominate.⁵⁵ The following countries had strong patent protection regimes already in the pre-sample period, 1980–1986, and maintained strong regimes throughout the whole sample period, 1987 to 2003: Belgium, Denmark, France, Germany, Italy, Netherlands, Sweden, and the United Kingdom (plus the United States). The countries with weaker patent protection regimes completed the major patent law reform preparing the ground for a strong patent protection regime in 1992, or later: Czech Republic, Finland, Greece, Hungary, Ireland, Poland, Portugal, Spain, and the Slovak Republic.

All European countries that we classified as having strong patent rights, except for Denmark and Italy, were among the initial contracting states of the European Patent Organisation (EPOrg) in October 1977.⁵⁶ All the countries classified as having weaker patent rights joined the EPOrg between October 1986 and March 2004 (European Patent Organization (EPOrg) 2010) and none of these countries completed the reforms preparing the ground for a strong patent protection regime before 1992 (Branstetter et al. 2006; Qian 2007; World Intellectual Property Organization 2012). Our classification is consistent with the groupings in Branstetter et al. (2006) or Qian (2007). It also fits with the patent right index of Maskus and Penubarti (1995) and Rapp and Rozek (1990): the patent laws of all the countries that we classify as countries with strong patent rights were fully conforming to the minimum standards of the US Chamber of Commerce Intellectual Property Task Force in 1984; all other countries lacked fully conforming patent laws in 1984.

For robustness checks, we also use the index of patent protection that was developed by Ginarte and Park (1997), and updated by Park (2008a, (2008b). Walter Park provides the index for more than 100 countries, updating it quinquennially for the years from 1960 onwards.⁵⁷ The index takes values between zero and five and higher values indicate patent laws with stronger IPRs. The index coding scheme aggregates information on 1) membership in international treaties (Paris Convention, International Convention for the Protection of New Varieties of Plants, Patent Cooperation Treaty, Budapest Treaty, Agreement on Trade Related Aspects of Intellectual Property Rights), 2) enforcement mechanisms (preliminary injunctions, contributory infringement pleadings, burden of proof reversal), 3) restrictions on patent rights (working requirements, compulsory licensing, revocation of patents), 4) duration of protection and 5) extent of coverage (pharmaceuticals, chemicals, food, surgical products, microorganisms, utility models, software, plant and animal varieties). Relevant in the context of our study is the updated coding scheme as described by Park (2008a); Ginarte and Park (1997) give details on the original coding scheme.

We classify industries according to the extent to which innovators consider the strength of patent rights as relevant, and rely on patenting in appropriating returns to invention. To do so, we form industry groups with different levels of patent relevance. For our main measure of patent relevance, we first calculate the nominal patent intensity for US industries in the pre-sample years, 1980–1986, dividing the fractional EU KLEMS patent counts by nominal value added in million US dollars, determine the ranking of US industries based on the intensity variable in each year, and average across the pre-sample years. Then, we generate different sets of industry groups. First, we define the group with high patent relevance as covering the industries at or above the 75th percentile of the average pre-sample US patent intensity ranking. These are the four sampled industries that constitute in all the relevant pre-sample years the industries with the highest patent intensities, reflecting the fact that the ranking is very persistent across these years. The group with low patent relevance covers three industries, two rank below the 25th percentile in all the relevant pre-sample years, and one ranks below in 5 of 7 years. The intermediate group covers all remaining industries. Second, we distinguish between the group of industries with average pre-sample US patent intensity rankings above the median (these 6 industries constitute in all relevant pre-sample years the industries with the highest patent intensities), and the complementing group with rankings below or at the median.

Our alternative measure of patent relevance builds on Cohen et al. (2000) who present survey-based evidence on the importance of patenting in appropriating returns to invention from the Carnegie Mellon Survey (CMS) on Industrial R&D in the US manufacturing sector. In the survey, which was conducted in 1994, about 1100 R&D unit or laboratory managers reported, among others, the percentage of their innovations for which patenting had been effective in protecting the firm’s “competitive advantage from those innovations” during the prior three years, 1991 to 1993.⁵⁸ For 34 manufacturing industries, Tables 1 and 2 in Cohen et al. (2000) provide the mean shares of product and process innovations for which the survey respondents judged patenting to be effective. We aggregate these shares to the level of the 13 industries in our main data set (1993 NACE, revision 1), weighting by the respective numbers of respondents. Then, we generate different sets of industry groups. First, we define the group with high patent relevance as covering the industries at or above the 75th percentile of the share of innovations with effective patent protection, the group with low patent relevance as covering the industries below the 25th percentile, and the intermediate group covering all remaining industries. Second, we distinguish between the group of industries above the median of the share variable, and the complementing group with share values below or at the median.

1.4 Appendix 2.4: Product market reform

The product market reform that we consider is part of the large-scale internal market reform of the European Union (EU) in 1992, named the Single Market Program (SMP). The SMP was designed by the European Commission and, thus, a supra-national institutional body. It was meant to bring down internal barriers to the free movement of products and production factors within the EU in order to foster competition, innovation and economic growth. Recent empirical evidence supports the view that the product market reform increased product market competition in manufacturing industries.⁵⁹ The SMP was officially implemented in 1992 in all EU member countries in 1992: Belgium, Denmark, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain, United Kingdom, and Luxembourg. Eleven of these countries are included in our main sample, and all these had joined the EU much earlier, at the latest in 1986.⁶⁰

For all initial SMP countries, the European Commission report by Buigues et al. (1990) provides a common list of 40 three-digit manufacturing industries that researchers expected ex ante to be affected by the product market reform. Country-specific additions to and removals from the common list are also reported. These additions and removals reflect recommendations of experts, who were asked whether they expected the reform to change the product market conditions in an individual industry in a specific SMP country differently than in the corresponding average industry. In country-industries that were ex ante expected to be affected the initial level of competition was typically low. The information in Buigues et al. (1990) allows for constructing reform measures that vary across industries within countries, across countries and across time. We exploit that fact for identifying the reform impact from confounding influences.

To generate our main product market reform measure, $R_{cit}$, we first link the 40 three-digit industry codes (1970 NACE) that are on the common list of Buigues et al. (1990) to the corresponding 109 four-digit industry codes (1993 NACE, revision 1). Then, we link the 39 three-digit codes (1970 NACE) of the additions to the corresponding 119 four-digit codes (1993 NACE, revision 1). Finally, we aggregate the four-digit industry codes on the common list, as well as the data on country-specific removals and additions of industry codes, to the country-industry level of our data-set:

$$\begin{aligned} R^{m}_{cit}= \left\{ \begin{array}{ll} \frac{1}{n_i} \sum _{k \in i} A_{kct} &{} \text{ if } \; t \ge 1992\\ 0 &{} \text{ if } \; t<1992 \; \text{ or } \; \sum _{k \in i} A_{kct}=0. \end{array} \right. \end{aligned}$$

For each of the 13 manufacturing industries i in each of the 11 SMP countries c in our main sample (see Tables 1 and 2), the main reform measure $R_{cit}$ is set equal to zero in all years t before the implementation of the SMP. From 1992 onwards, it is equal to the share of the four-digit industry classes k per country-industry that were ex ante expected to be affected by the SMP.⁶¹ The dummy variable $A_{kct}$ is coded one from 1992 onwards if a four-digit industry k in country c was ex ante expected to be affected, and zero otherwise. The number of four-digit industry codes per industry i is denoted by $n_{i}$. For non-SMP-countries, as well as for service industries, the main reform measure is always equal to zero.

We also apply an alternative measure of the product market reform. To construct it, we follow Griffith et al. (2010) in using employment shares, including those that are reported in Buigues et al. (1990), for weighting purposes. First, we link each of the three-digit industry codes (1970 NACE) that are on the common list to the main corresponding industry among the 13 industries (1993 NACE, revision 1) in our main sample and, then, aggregate as follows:⁶²

$$\begin{aligned} R^{a}_{cit}= \left\{ \begin{array}{ll} \sum _{j \in i} w_{jic} A_{jt} &{} \text{ if } \; t \ge 1992\\ 0 &{} \text{ if } \; t<1992 \; \text{ or } \; \sum _{j \in i} A_{jt}=0. \end{array} \right. \end{aligned}$$

For each three-digit industry j in country c, the weight $w_{jic}$ indicates the share that industry j contributes to the employment in industry i in country c, averaged over the years 1985 to 1987. For constructing weight $w_{jic}$, we divide the employment share $w_{jc}$, as directly reported by Buigues et al. (1990), by the appropriate employment share $w_{ic}$. The weight $w_{jc}$ of Buigues et al. (1990) indicates the share of the tree-digit industry j in total manufacturing employment in country c, averaged over the years 1985 to 1987. The weight $w_{ic}$ is constructed from EU KLEMS data and indicates the share of industry i in total manufacturing employment in country c, averaged over the years 1985 to 1987.

In all years t before the implementation of the SMP, the alternative reform measure is set equal to zero. From 1992 onwards, it is equal to the share of the employment-weighted three-digit industry classes per industry i that were ex ante expected to be affected by the SMP. The dummy variable $A_{jt}$ is coded one from 1992 onwards if a three-digit industry j was ex ante expected to be affected according to the common list of Buigues et al. (1990), and zero otherwise.⁶³

1.5 Appendix 2.5: Financial conditions

The financial variables which we use in Sect. 5.3 are based on data from the EU KLEMS 2008 database and from the November 2010 version of the Financial Development and Structure Database (FDSD) by Beck et al. (2000, (2009, (2010a, (2010b).⁶⁴

First, we measure financial sector development at the country-level using data from the Financial Development and Structure Database on stock market capitalization and on the channeling of savings to investors via financial intermediaries, relative to the size of the economy. Our proxy of stock market capitalization is the value of listed shares, relative to gross domestic product (GDP). As values of listed shares are hardly available before 1989, we determine the country ranking based on stock market capitalization for the years 1989 and 1990, and then we average across both years.⁶⁵ The channeling of savings to investors via financial intermediaries is captured by the following private credit ratio: claims on the private sector by deposit money banks and other financial institutions, relative to GDP. We use the ratio values between 1980 and 1990 to determine a country ranking per year, and then we average across all years.⁶⁶ Finally, we generate our financial development measure by averaging across both the average stock market and private credit ranking and classify countries with averages above the median rank as having a highly developed financial sector in the time period 1980–1990. These countries are France, Germany, the Netherlands, Spain, Sweden, and the UK. All other countries in our main sample are classified as having a weakly developed financial sector up to 1990.⁶⁷

Second, we classify industries according to their capital needs, proxied by the capital intensity of production in the corresponding US industries in the pre-sample period, 1980–1986. To that aim we use the relevant EU-KLEMS 2008 industry-level data on the ratio of capital relative to labor compensation in the pre-sample years, determine the respective industry ranking in each year, average across the pre-sample period, and generate different sets of industry groups. First, we define the group with high US pre-sample capital intensity as covering the industries at or above the 75th percentile of the average pre-sample US capital intensity ranking. The group with low capital intensity covers the industries below the 25th percentile and the intermediate group covers all remaining industries. Second, we distinguish between the group of industries above the median of the average pre-sample US capital intensity ranking, and the complementing group.

1.6 Appendix 2.6: Import penetration

For constructing the measures of import penetration which we use in Sect. 5.3, we start by calculating the EU 15 import penetration for industry i in country c in year t,

$$\begin{aligned} P^{EU \; 15}_{cit}=\frac{M^{EU \; 15}_{cit}}{Q_{cit} + M_{cit} - X_{cit}}, \end{aligned}$$

where $M^{EU \; 15}_{cit}$ denotes the nominal value of imports from EU 15 member countries, $M_{cit}$ represents the nominal value of imports from all countries worldwide, and $X_{cit}$ represents the nominal value of exports to all countries worldwide. Domestic production is denoted by $Q_{cit}$. We take data on $M^{EU \; 15}_{cit}$, $M_{cit}$, and $X_{cit}$, as well as on all other trade measures mentioned below, from the OECD STAN Bilateral Trade Database (BTD, edition 2010).⁶⁸ From the EU KLEMS 2008 database, we take data on $Q_{cit}$, that is nominal gross output. When calculating the import penetration ratios, we harmonize the set of national currencies in the BTD 2010 database and the EU KLEMS 2008 database.

Then, we average the country-industry-year specific ratios across the years 1988 to 1990 to generate a measure of the initial EU 15 import penetration. The years 1988 to 1990 are the earliest years for which BTD 2010 provides the relevant trade data, although not for all country-industries in our main sample. The group of the EU 15 member states covers the eleven SMP countries in Table 1, Finland, Sweden, and two non-sampled EU 15 member states (Luxembourg and Austria).

Finally, we set the indicator EU 15 import penetration $^{high}_{ci,1988-90}$ equal to one for all industries i in countries c where the initial EU 15 import penetration is at or above the relevant 75th percentile, and zero otherwise.⁶⁹

To generate an indicator for high exposure to imports from all countries worldwide, except the EU 15 countries, we proceed analogously, using $M^{World \; \setminus \; EU \; 15}_{cit}$ instead of $M^{EU \; 15}_{cit}$. In case of the indicator for high exposure to imports from China, we average across a longer time period, 1988 to 2003, taking into account that imports from China were typically low until the mid-1990s (Bloom et al. 2015).

Appendix 3: Construction of main estimation sample

Our main estimation sample is an unbalanced panel of 2736 observations on 13 manufacturing industries in 17 countries between 1987 and 2003.

We apply the following standard data cleaning routines. We drop country-industry-year observations with missing values of variables that are relevant to our main regression analysis (see Tables 3, 4, 5 and 6). In addition, we eliminate all observations with absolute growth of more than 200 percent in R&D intensity, or in real R&D expenditures. Finally, we eliminate all country-industries where we observe less than five consecutive years during the time period from 1987 to 2003.⁷⁰