Innovation and Patent Protection: A Multicountry Study on the Determinants of R&D Offshoring

Abstract

This paper looks at the role that intellectual property rights (IPR) protection plays in the decision of multinational corporations (MNCs) to locate their R&D activities abroad, a phenomenon which has been labelled in the literature as innovation offshoring. Do countries with stronger IPRs attract more offshored innovation? Do different types of innovation offshoring respond equally to IPR variations? Using a novel multicountry and multisector database gathering information on the innovation activity of more than 15,000 MNCs from all around the world, I am able to distinguish among two types of innovation offshoring: innovation carried out in nations different from the home country, where the firm undertakes production activities directly or indirectly through a subsidiary (commercial innovation), and research done in countries, where the MNC only collaborates with local firms or inventors, with no on-site production involved (external innovation). In order to better isolate the impact of property rights protection on R&D, my identification strategy takes into consideration IPR’s variation across industries. I find that firms tend to locate commercial innovation in countries with strong IPR protection. This is true especially for long life-cycle industries which rely longer on patents. In contrast, short life-cycle technologies with faster obsolescence rate (e.g. high-tech products) are less responsive to IPR protection. External innovation, on the other hand, is less affected by patent protection, suggesting other motives behind its location decisions.

An earlier version of this paper was presented in the 11th Annual Conference of Knowledge Forum in December 2016.

Appendices

Appendix 1: Theoretical Model

In this section, I examine in more detail the resolution of the model in Sect. 7.3, deriving the main formula of the paper. Firm i wants to maximize, with respect to the obsolescence time t, its expected profits:

$$\mathop {\hbox{max} }\limits_{{t_{s} }} E\left( {\Pi _{i,s,c} } \right)$$

(7.8)

just taking the expectation of Eq. 7.1:

$$E\left( {{\Pi}_{i,s,c} } \right) = \left( {\pi_{i,s,c} + k_{i,s,c} } \right)E\left( {{\text{MIN}}\left[ {t_{s} ,m_{c} } \right]} \right) + \mathop \sum \limits_{z \ne i} k_{z,s,c} \left( {T_{\hbox{max} } - E\left( {{\text{MIN}}\left[ {t_{s} ,m_{c} } \right]} \right)} \right) + E\left( {\varepsilon_{i,s,c} } \right)$$

(7.9)

where $\varepsilon$ is a white noise process and m has a uniform distribution accordingly to Assumption 1. It follows that the probability density function and the cumulative distribution function for m are, respectively, as follows:

$$f\left( {m_{c} } \right) = \frac{1}{{m_{c} }}$$

(7.10)

$$F\left( {m_{c} } \right) = P\left( {m_{c} \le x} \right) = \frac{x}{{\overline{{m_{c} }} }}$$

(7.11)

To ease the calculation, I assume just one competitor firm in the market: company j.

Notice that the expectation in (7.9) can be rewritten as follows:

$$E\left( {{\text{MIN}}\left[ {t_{s} ,m_{c} } \right]} \right) = t_{s} \cdot P\left( {t_{s} < m_{c} } \right) + E\left( {m_{c} \cdot P\left( {t_{s} \ge m_{c} } \right)} \right)$$

(7.12)

with

$$P\left( {t_{s} < m_{c} } \right) = 1 - P\left( {t_{s} \ge m_{c} } \right) = 1 - \frac{{t_{s} }}{{\overline{{m_{c} }} }}$$

(7.13)

and

$$E\left( {m_{c} \cdot P\left( {t_{s} \ge m_{c} } \right)} \right) = \mathop \int \limits_{0}^{{t_{s} }} m_{c} \cdot f\left( {m_{c} } \right)dm_{c} = \frac{{t_{s}^{2} }}{{2\overline{{m_{c} }} }}$$

(7.14)

I can therefore simplify (7.9) into:

$$E\left( {\varPi_{i,s,c} } \right) = \left( {\pi_{i,s,c} + k_{i,s,c} } \right)t_{s} - \frac{{t_{s}^{2} }}{{2\overline{{m_{c} }} }}\left( {\pi_{i,s,c} + k_{i,s,c} } \right) + k_{j,s,c} T_{\hbox{max} } - k_{j,s,c} t_{s} + \frac{{k_{j,s,c} t_{s}^{2} }}{{2\overline{{m_{c} }} }}$$

(7.15)

A comparison between expected profits in North and in South leads to (7.2). Finally, extracting the FOC for Eq. (7.9), I arrive at the expression in Eq. (7.3).

For the external innovation case, I just followed all precedent steps considering $\pi_{i,s,c} = 0.$

Appendix 2: Database Creation

2.1 MNC Group Identification

Orbis database, compiled by Bureau Van Dijk, is a commercial dataset containing financial and administrative data on over 150 million firms across the planet. While coverage of firms is not exhaustive, it has been proved that it offers a fair representation of economic activity in each state, arriving to cover almost 75–80% of firms in developed countries such as European ones (Kalemli-Ozcan et al. 2015). National censuses are, by far, more complete including a large number of small companies, but they typically lack of annual representation of the firms as surveys are not conducted every year. For the purposes of my study, given the focus on multinational activity and innovation, I am not concerned about the exclusion of smaller firms, which are rarely conducting R&D activities, and I prefer more systematic data on bigger companies offered by Orbis. The Bureau Van Dijk’s platform presents two sections: “Companies” which contain financial data on each firm present in the database, and “Patents” which include all information on patents hold by represented firms and accessed through PATSTAT database. Orbis advantage is to connect these two parts through a unique BvD ID number which exclusively identify each enterprise.

I start my analysis downloading all granted patents owned by a firm with a publication date between 1 January 2005 and 1 July 2015 (initial date of my research). Orbis does not allow you to select patents based on their priority date; therefore, even if my analysis is limited to the interval of time 2005–2013 (years in which I have data for IPR at country level), I extended the time of selection in order not to lose any observation, knowing that typically patents are published after 18 months from the priority date except for certain patents at the USPTO which are published only if/when granted. For each patent, I download IPC codes, BvD ID of the firm which currently owns it, priority date, application number, inventors’ names and countries of residence.

Once obtained all the innovating firms, I need to build, for each of them, the corporate group in order to understand if they are the head of a corporation or just subsidiaries held by other companies. Additionally, since my paper focuses on multinationals, I want to rule out national enterprises which only have subsidiaries within their national territory. Building precisely the ownership structure of the MNCs is crucial to attribute the correct patents to each multinational. In the Companies section, I download the Global Ultimate Owners (GUO) associated with the previously extracted BvD ID; for all firms which lack this information, I assume that they are themselves GUOs. Subsequently, I download all their subsidiaries owned at more than 25% by all the GUOs in my sample: the participation level threshold, fixed at 25%, is intended to include only effectively controlled subsidiaries. Also, I make sure to unfold up to the 10th, and last, subsidiary level. Subsidiaries can be controlled at different levels. As Fig. 7.5 shows if firm A holds 100% of firm B, and firm B holds 100% of firm C, then indirectly firm A holds 100% of firm C: firm C is a second-level subsidiary, while firm B is a first-level subsidiary for A.

Fig. 7.5

Different levels of ownership

Here, a limitation of the platform arises: Orbis, according to his settings, only gives a maximum of 1.000 subsidiaries at a time. Since some MNCs have many more, I isolate them in a group of “big” GUOs, and I download manually all their subsidiaries from their reports one by one. This task is very time consuming, but it is necessary since the bigger multinationals in my sample are more likely the more active ones in terms of R&D, and excluding them would inevitably bias my findings. Also, there is a limit of 40.000 subsidiaries that can be downloaded in Excel from Orbis, but none of my GUO exceeds this threshold.

2.2 Commercial and External Innovation Identification

In order to distinguish the two categories of commercial and external R&D, I have to identify the countries where the MNCs undertake some manufacturing activity. I use the distinction that Orbis provides about different types of entities. I select as “commercial” subsidiaries only these registered as industrial companies. “This category includes all companies that are not banks or financial companies nor insurance companies. They can be involved in manufacturing activities but also in trading activities (wholesalers, retailers, brokers, etc.)”.²⁵ They also figure as a separate category from research institute; therefore, making use of this classification, I am sure not to include in my dataset any isolated research laboratory which does not operate in combination with a manufacturing or resale activity.

Appendix 3: Additional Robustness Checks

I perform additional robustness checks to validate my findings. In Table 7.7, all MNCs are included in the sample, not only those undertaking both commercial and external innovation. The results remain valid.

Table 7.7 Robustness checks—all MNCs

As the distinction between commercial and external R&D can be tackled, I run an estimation restricting commercial innovation only to these cases where at least two industrial subsidiaries belonging to the MNC are present in the country. In this way, I can be sure that there is a real presence of the MNC in the country in terms of production and sales. Results are presented in Table 7.8.

Table 7.8 Robustness checks—multiple subsidiaries