Concordance and complementarity in IP instruments

ABSTRACT

This work investigates the relationship between proxies of innovation activities, such as patents and trademarks, and firm performance in terms of revenues, growth, and profitability. By resorting to the virtual universe of Italian manufacturing and service firms, this work provides a rather complete picture of the Intellectual Property (IP) strategies pursued by Italian firms, in terms of patents and trademarks, and studies whether the two instruments for protecting IP exhibit complementarity or substitutability. In addition, and to our knowledge novel, we propose a measure of concordance (or proximity) between the patents and trademarks owned by the same firm and we then investigate whether such concordance exerts any effect on performance. The results suggest that while patents and trademarks independently exert a relevant impact on firm performance, there is no convincing evidence in favour of a complementary role of IP.

KEYWORDS:

• Innovation
• intellectual property
• complementarity
• concordance
• firm performance

Acknowledgments

We thank the participants to the Department of Economics Unibo internal seminar (June 2018), the European Commission Joint Research Center seminar (JRC-Seville, June 2018), the European Association for Evolutionary Political Economy (EAEPE) conference (September 2018), the IP Statistics for Decision Makers (IPSDM) conference in Alicante (October 2018) and the Sant’Anna Institute of Economics seminar (October 2018). We are grateful to Roberto Susanna (Infocamere press office) and to Bureau van Dijk for their technical assistance at various stages. We are also indebted to Caterina Giannetti, Hanna Hottenrott, Laura Magazzini, Arianna Martinelli, Daniele Moschella, Gabriele Pellegrino, Emanuele Pugliese, Enrico Santarelli, Antonio Vezzani, and Nikolas J. Zolas for insightful comments. All remaining errors are our own. This work received fundings by the Fondazione Cassa di Risparmio of Forlì, project ORGANIMPRE. This work has been partly supported by the European Commission under the H2020, GROWINPRO, Grant Agreement 822781.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

¹ Specifically, a trademark is a sign that identifies a firm’s product or service, so that it allows consumers to distinguish between different goods (see, e.g. Mendonça, Pereira, and Godinho 2004; Ramello 2006).

² We use two AIDA ‘historic’ disks in the release version of December 2015 and December 2016, respectively. For a detailed description of the procedures followed in building the dataset, refer to Grazzi, Piccardo, and Vergari (2018).

³ As suggested by the previous literature, see OHIM (2015); Schautschick and Greenhalgh (2016); Graham et al. (2018) among others, patenting activity is mainly observed within manufacturing sectors, while trademarks are of more widespread use independently of the sector of activity. Consistent evidence emerges using the AIDA-AMADEUS dataset in this work. However, in Section 5.2, we replicate our main analysis for firms in service sector, in order to check for potential differences.

⁴ We have opted to focus on applied, rather than granted patents, to enlarge as much as possible the sample of observations, given the very small percentage of Italian firms with patents. Results on the smaller sample (granted patents) are also available upon request.

⁵ Patents could be owned by more than one firm; in these cases, we associated the patents to each owner, as suggested by the existing literature.

⁶ Czarnitzki and Kraft (2004), Xu and Chiang (2005), Belderbos et al. (2014) and de Rassenfosse and Jaffe (2018), among others, have highlighted the importance to account for the decline in patents’ value during the life of patented inventions. Our choice to censor the stock of patents for each firm, in each year, at 20 years is in line with the application of a constant 15% depreciation rate (that is commonly used in the literature in order to discount R&D, patent, and trademark values).

⁷ An alternative attempt is provided, still on Italian firms, by Lotti and Marin (2013) on EPO applications which the authors match to a restricted sample of AIDA firms (the so-called, BvD AIDA TOP). In the end, their effort results in 5485 patenting firms over the period 2000–2007. It is however not possible to directly compare the two final datasets. It is true that our displays a higher number of firms with patents (8616 in 2006, see Table 1), but this is most likely due to employing the ‘full’ AIDA version and, although to a lesser extent, to counting USPTO and IPTO patent applications.

⁸ As for patents, also trademarks can be jointly owned by more than one firm; in these cases, we associated the trademarks to each owner, as suggested by the existing literature. Note however that the phenomenon of joint trademark ownership is much more limited with respect to patents.

⁹ The lack of applications for trademarks at the national level is of course expected to reduce both the overall stock of trademarks in the country as well as the share of firms holding trademarks. It is however very difficult to find a reference to assess how large is the actual impact on our dataset. To the best of our knowledge, OHIM (2015) provides some guidance in terms of aggregate statistics even if it is at the EU level with no possibility to distinguish among countries. According to OHIM (2015), Table 8 (page 40), ‘38.1 per cent of all large companies and 8.6 per cent of all SMEs own trademarks’, either at national or EU. In our dataset, which includes USPTO and EUIPO trademarks, the corresponding percentages are, respectively (see Table 2) 57.9 $\approx$ (679/1172) and 12.4 $\approx$ (3805 + 2653)/(44148+8016). Hence, it would not appear that the lack of national trademarks greatly compromises the sample that we employ.

¹⁰ The IPC is a classification for patents and utility models according to the different areas of technology to which they pertain. We use it at the 3-digit level. NICE is a 2-digit international classification of goods (codes from 1 to 34) and services (codes from 35 to 45) applied for the registration of trademarks that has been adopted with the Nice Agreement concluded in 1957 (a summary of the Nice Agreement is available at: https://www.wipo.int/treaties/en/classification/nice/summary_nice.html).

¹¹ Clearly, we do not account for patents and trademarks without information on the IPC and NICE codes.

¹² For the methodology and weights refer to works of Lybbert and Zolas (2014), Zolas, Lybbert, and Bhattacharyya (2017) and Goldschlag, Lybbert, and Zolas (2016).

¹³ The overlapping coefficient is a measure of agreement (or similarity) which refers to the area under two probability density functions simultaneously. Notice that the overlapping coefficient between an IPC and a NICE code is not firm or time specific.

¹⁴ Note that this overlapping measure is quite intuitive. Indeed, for each pair of IPC and NICE codes we end up with a number of common ISIC codes, and for each common ISIC code we have two weights: one representing the probability of concordance between that particular IPC code and that particular ISIC code, the other representing the probability of concordance between that particular NICE code and that same particular ISIC code. The probability of concordance can be calculated as the minimum of the pairwise likelihood of the match between those IPC and NICE codes. This is in line with the proximity measure proposed by Hidalgo et al. (2007) to analyse the role of the existing production structure on the regions pattern of specialisation as well as with the technology proximity measure developed by Colombelli, Krafft, and Quatraro (2014) that study the path-dependent emergence of new technological fields focusing on the nanotechnology sector, in the EU 15 countries in the period 1986–2006.

¹⁵ Alternative procedures such as mapping IPC to ISIC codes and then ISIC to NICE codes of the firm (or, vice versa going from NICE to ISIC codes and then from ISIC to IPC codes) would of course be possible, but they would complicate the analysis and would underestimate the concordance measure as they would imply going through double weighting.

¹⁶ Note that, as common to empirical investigations with firm-level data, our measure is potentially sensitive to relevant events affecting the life-cycle of a firm such as mergers and acquisitions. Consider two firms characterised by the maximal degree of concordance: each of them has only patents in technological fields perfectly related to its products (trademarks). Clearly, if these two firms merge, and their sets of IPC and NICE codes do not coincide, the degree of concordance of the new firm decreases. Of course, this issue is related to the impossibility to appropriately deal with such events in standard firm-level data.

¹⁷ Such similar percentages of firms with patents and firms with trademarks might also be related to the absence of IPTO trademarks.

¹⁸ A similar trend in the number of patents has been reported also by the Italian Observatory for patents, Osservatorio Italiano Brevetti, Osservatorio Italiano Brevetti (2014). Considering patents applied at the Italian Patent and Trademark Office (IPTO) from both Italian and foreign firms, the Observatory highlights a reduction in the number of applied patents starting from 2011.

¹⁹ In the interest of space, we do not show descriptive statistics considering co-ownership of patents. They are available upon request.

²⁰ The classification is available at: http://ec.europa.eu/eurostat/statistics-explained/index.php/Glossary:Enterprise_size

²¹ Eurostat classification refers to the number of workers, thus including also the entrepreneur(s). AIDA dataset originally reports the number of employees and, as obvious, many SMEs report no employees. Hence, to make the two figures comparable, we approximate the number of workers as the number of employees plus one.

²² Dinlersoz et al. (forthcoming) perform a similar exercise on the relative timing for R&D, patents and trademark filings and report mild evidence in the opposite direction for US firms. Notice however that they do not employ any concordance measure.

²³ As a robustness check, we alternatively compute firms’ growth rates following the methodology suggested by Davis, Haltiwanger, and Schuh (1998), that allows us to account for firms’ entry and exit. The use of this alternative specification for the growth rate does not change our conclusions on the relationship between firms’ growth and their use of IP instruments. Results are available upon request.

²⁴ In order to mitigate problems related to measurement errors, we do not account for values of the profitability larger than 100 and lower than −300, the latter corresponding roughly to the bottom percentile of the distribution. In this way, over a total number of 1,180,933 observations for ebitda_sales, we remove around 15,000 observations.

²⁵ In the interest of space, we do not show graphical comparisons for growth and profitability, but they are available upon request.

²⁶ The graphical comparison between the two groups of firms in terms of growth and profitability is available upon request.

²⁷ Note that the non-parametric analysis only provides descriptive evidence. In order to take into consideration the size affect, we control for size in all the specifications included in the parametric analysis below.

²⁸ In the interest of space, we do not show the FP tests, but they are available upon request.

²⁹ The median number of patents and trademarks (excluding firms without such instruments) is, respectively, 3 and 2. Patents vary in the range 1–7174 while trademarks take values between 1 and 100.

³⁰ We build the four dummy variables based on the yearly number of patents and registered trademarks owned by firms. For more details on the construction of patents and trademarks stocks refer to Section 2.1.

³¹ Labour productivity is computed as the ratio between added value and number of workers.

³² In order to account for the potential endogeneity between labour productivity and the indicators for innovation activities, we have also estimated a structural model following the methodology originally proposed in Crepon, Duguet, and Mairesse (1998). The results, which are quite reassuring, are not shown here in the interest of space, but are available upon request.

³³ Focusing on firms’ growth as a measure of performance, our regression model is in line with specifications suggested by the applied literature which empirically tests the validity of Gibrat’s law (see Santarelli, Klomp, and Thurik (2006) for an exhaustive survey of empirical studies testing Gibrat’s law). In this literature, firms’ growth is modelled as a function of the initial size. Moreover, this baseline specification has been extended in order to account for several major control variables in level, such as productivity, export, external finance, and R&D activity as well as others. We refer to, within the very vast literature that investigates the growth of firms, Gibrat (1931); Penrose (1959); Evans (1987a, 1987b); Doms, Dunne, and Roberts (1995); Del Monte and Papagni (2003); Garca-Manjón and Romero-Merino (2012).

³⁴ Correlation tables for the variables included in our specifications are available upon request.

³⁵ Results accounting for firms fixed effects are available upon request.

³⁶ The effect of having both IP on firms’ performance is computed as follows: (${\beta }_4$-${\beta }_1$)*100 when we consider total revenues and growth as a measure of performance and (${\beta }_4$-${\beta }_1$) when use profitability as a proxy for performance. Similarly, it is possible to identify the effect of having either only trademarks or patents on firms’ performances, using the estimated coefficients ${\beta }_2$ and ${\beta }_3$, respectively. Our results suggest that having only trademarks results in around a 31% increase in total revenues, a 5% increase in firms’ growth and a 0.75% increase in profitability compared with not having both IP instruments. Moreover, having only patents results in around a 17% increase in total revenues, a 3% increase in firms’ growth and a 0.74% increase in profitability compared with not having IP instruments.

³⁷ In Table 3 (columns 1,4 and 7) we reject the following null hypotheses: H0_a: ${\beta }_2\le {\beta }_1$, H0_b: ${\beta }_3\le {\beta }_1$ and H0_c: ${\beta }_4\le {\beta }_1$.

³⁸ The null hypotheses are H0_d: ${\beta }_4\le {\beta }_2$ and H0_e: ${\beta }_4\le {\beta }_3$.

³⁹ The same condition can be expressed as follows: ${\beta }_4-{\beta }_2>{\beta }_3-{\beta }_1$.:

⁴⁰ More precisely, the one-sided Wald-tests comparing the magnitude of coefficients ${\beta }_2$ and ${\beta }_3$ of Table 3 (not shown in the interest of space), support this evidence: in columns 1 and 4, coefficients ${\beta }_2$ are bigger than coefficients ${\beta }_3$ (they are 3.078 and 2.941 in column 1 and 0.338 and 0.321 in column 4, respectively); whereas, coefficients ${\beta }_2$ and ${\beta }_3$ in column 7 do not statistically differ (they are 1.654 and 1.638, respectively). These tests are available upon request.

⁴¹ Also, note that the lack of significance of the concordance coefficients might also be related to the nature of the innovation itself and specifically product versus process innovations. Unfortunately, for the time being, we cannot distinguish patents (or trademark) as related to product or process innovation.

⁴² The negative impact of initial firms’ size on their growth is a common result in the literature, also known as a violation of Gibrat’s (see, e.g. Mazzucato and Parris 2015; Castaldi and Dosso 2018; Grazzi and Moschella 2018).

⁴³ We refer to Section 2.1 for more details on the construction of patent and trademark stocks.

⁴⁴ This is pointed out in, among others, Sandner and Block (2011). Further evidence (see, e.g. Levin et al. 1987; Cohen, Nelson, and Walsh 2000) shows that in many cases firms rely on patents for strategic reasons rather than for appropriating the returns from their innovations; and Blind, Cremers, and Mueller (2009) suggests that patents filed for traditional motives (i.e. protect their inventions) are associated with higher quality.

⁴⁵ Results in Table 4 are robust to the use, for patent and trademark counts, of inverse hyperbolic sine transformation rather than logarithmic transformation.

⁴⁶ Estimates of the baseline specification on more disaggregated sub-samples of firms according to their sector of activity (2-digit ATECO sectors for manufacturing firms, and ATECO Sections for firms operating in the service sector), show that the relationship between firms’ IP strategies and their performances are rather homogeneous across sectors, with the noteworthy exception of the regression on profitability, which exhibits clear sectoral patterns. Conversely, we do not find relevant differences on the relationship between IP instruments and firms’ performances across small, medium, and large firms, operating in both the manufacturing and the service sectors. Results are available upon request.

⁴⁷ A common 2-digit ISIC code is identified only for the pairs of IPC and NICE codes A61-2 and C09-2. In particular, both the IPC code A61 and the NICE code 2 are linked to the ISIC code 20; while, both the IPC code C09 and the NICE code 2 are linked to the ISIC code 20. Thus, the overlapping coefficients are different from zero only for these pairs on IPC and NICE codes (0. 0505005 and 0.1203539 for pairs A61-2 and C09-2, respectively).

Additional information

Funding

This work was supported by the European Commission H2020, project name: GROWINPRO [822781]; Fondazione Cassa di Risparmio of Forlı̀, project ORGANIMPRE [ORGANIMPRE];