Do deep regional trade agreements facilitate international research collaboration?

ABSTRACT

We examine whether regional trade agreements (RTAs) facilitate international research collaboration. First, using a two-country model of a continuum of oligopolistic industries with process research and development (R&D) investment and spillovers, we analyze whether trade liberalization through a trade agreement with shallow or deep economic integration increases the number of firms that collaborate. We then empirically investigate the effects of shallow and deep RTAs by employing data on patents with multiple inventors from different countries at the United States Patent and Trademark Office for 114 countries/regions over 1990–2015. We interpret co-inventions by inventors from different countries as evidence of international research collaboration. We find that both shallow and deep RTAs are positively associated with research collaboration among member countries, and that the effect of deep RTAs on international research collaboration is stronger than that of shallow RTAs under certain conditions.

KEYWORDS:

• Research collaboration
• international co-invention
• regional trade agreement
• deep integration

JEL CLASSIFICATION:

• L13
• O31
• O33
• F15

Acknowledgments

We thank Matilde Bombardini, Banri Ito, Hiro Kasahara, Ryo Nakajima, Jie Sun, anonymous reviewers, and participants of the 77th Annual Meeting of the Japan Society of International Economics, the 2nd China-Japan Youth Conference on Trade, Exchange Rate and Labor in Kyoto, the 15th International Symposium on Econometric Theory and Applications (SETA) in Osaka, 2019 Spring Meeting of the Japanese Economic Association, the Workshop on International Trade and Intellectual Property Rights at the University of Bologna (Rimini), and seminars at Kwansei Gakuin University and University of British Columbia for their helpful comments and suggestions on earlier versions of the paper. The authors are solely responsible for any remaining errors.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 A number of previous studies have adopted the same approach (Cappelli and Montobbio 2016; Hoekman, Frenken, and van Oort 2009; Montobbio and Sterzi 2013; Tsukada and Nagaoka 2015).

2 Incidentally, Antonelli and Scellato (2013) empirically proved that innovation and technological improvements activities of each firm are significantly affected by interactions among three factors, i.e. reginal external knowledge, industrial external knowledge, and localized external knowledge, by using data on Italian manufacturing companies from 1996 to 2005 and firm level total factor productivity (TFP).

3 Other related recent studies include Giuliani, Martinelli, and Rabellotti (2016) and Iino et al. (2021).

4 See, for example, Goel and Haruna (2011) for the analysis of the relationship between R&D with spillovers and trade.

5 This model setting is similar to that of the Ricardian model with a continuum of goods developed by Dornbusch, Fischer, and Samuelson (1977).

6 Although $x_i\left(\omega \right)$ represents the outcome of R&D in terms of a reduction in marginal production cost, we also refer to $x_i\left(\omega \right)$ as ‘R&D investment’ for simplicity.

7 To simplify the analysis, we assume that there are no cross-industry spillovers.

8 This case corresponds to ‘R&D cartelization’, not ‘research joint venture (RJV) cartelization’, in Kamien, Muller, and Zang (1992). The spillover rate in RJV cartelization increases to its maximum rate, that is, $\text{β}=1$. Falvey and Teerasuwannajak (2016) analyze how the governments’ R&D policies (such as R&D subsidies and taxes) and the coordination of the R&D policies affect firms' choice of the R&D alliance form from several possible forms, including R&D cartelization and RJV cartelization.

9 In our model setting the presence of the R&D spillover is the primary reason for firms to collaborate. However, there could be another reason for this. For example, research subjects solved by a single firm are obviously limited in terms of research funds and professional human resources. One solution to these limitations is to organize research collaborations with other firms.

10 We do not consider the possibility of domestic research collaboration. As our main interest is the effect of deep RTAs on international research collaboration, we focus only on international research collaboration.

11 Based on their empirical findings, Giuliani, Martinelli, and Rabellotti (2016) argued the importance of coordination costs when firms collaborate internationally.

12 Proofs of propositions are presented in the Online Appendix.

13 In addition to the initial degree of the spillover, our theoretical analysis demonstrates that the parameters included in Equation (15(15) $\begin{array}{rl}{\pi }_i^C\left(\omega \right)-{\pi }_i^N\left(\omega \right)& =\frac{\gamma (2\text{β}-1)\left[2\right\{\alpha -c\left(\omega \right)\}-t]\left\{x_i^C\right(\omega )-x_i^N(\omega \left)\right\}}{A^N}-f\\ & =\frac{18{\gamma }^2(2\text{β}-1{)}^2[2\{\alpha -c(\omega \left)\right\}-t{]}^2}{A^C(A^N{)}^2}-f.\end{array}$(15) ) affect firms' incentive to collaborate and countries' incentive to sign a shallow or deep RTA. As we discuss in Section 6.2.1, this may cause the endogeneity issue that we need to address in our empirical analysis.

14 We calculate the average tariff rate, or the tariff burden ratio, for each country using the value of customs and other import duties divided by total imports of goods and services. All data are taken from the World Bank.

15 Note that $q_{m,z}^{\mathrm{AC}}=1$ and $q_{m,z}^{\mathrm{LE}}=0$ may hold at the same time. However, for $q_{m,z}^{\mathrm{LE}}$ to take either the value of one or two, it must hold that $q_{m,z}^{\mathrm{AC}}=1$.

16 Although Mattoo, Mulabdic, and Ruta (2022) use similar indexes to measure the depth of RTAs, their indexes reflect only the count of provisions.

17 Note that none of those indexes reflects the information on how much the signatories of an RTA actually implement collaborative activities among them. Thus, we do not need to worry about the issue of double counting by dependent and explanatory variables in our estimations.

18 In one specification we include the year and country fixed effects, whereas we include time-varying country fixed effects and country-pair fixed effects in another specification. These fixed effects address the issue of the omitted variable bias. In addition, we use a number of variables related to the country-pair-specific characteristics between the two countries as our instrumental variables, as explained in Section 6.2.1.

19 Note that a formation of an RTA by itself does not necessarily facilitate patent applications among member countries in our data because we use data from the USPTO rather than from patent offices in individual countries. Thus, we do not need to worry about the possibility of such a ‘facilitation effect’ being included in the WTO+ and WTO–X dummies and indexes.

20 Out of the 167,466 observations that we use in our empirical analysis, there are 149,396 (89.2%) for which the dependent variable (i.e. $C{A}_{\mathrm{ijt}}$) takes the value zero. With regard to the estimation technique, although the OLS estimates may be biased and inefficient, we believe that it is worth presenting the results of the OLS because it is the basic estimation method in econometrics. In addition, we can directly compare the results of the OLS with those of the two-stage least squares (2SLS), which we employ in Section 6.2.1 to address the issue of endogeneity.

21 The data on patents at the USPTO are more suitable for our study than data from other patent offices such as the EPO. The main reason is that the share of patent applications from foreign inventors is much higher at the USPTO than any other patent office.

22 There are some limitations of the use of the data on patents with multiple inventors from different countries as an indicator of international research collaboration. As discussed in Section 8, our measure of international research collaboration may include co-application of a patent by, for example, two different subsidiaries of the same multinational firm that are located in different countries. However, the decision of engaging in research collaboration in such a case is unlikely to be affected by the formation of RTAs. Thus, we do not think that such a concern is a serious issue in this study.

23 http://rtais.wto.org/UI/PublicMaintainRTAHome.aspx.

24 The sample included free trade areas, customs unions, and economic integration agreements, but excluded partial scope agreements.

25 https://datacatalog.worldbank.org/search/dataset/0039575.

26 When we include country-pair fixed effects, Comcol and Smctry cannot be used as IVs because they are perfectly collinear with the fixed effects. $\mathrm{ContagRT}{A}_{\mathrm{ijt}}$ is still a valid IV, but we find that all estimation results with country-year and country-pair fixed effects using $\mathrm{ContagRT}{A}_{\mathrm{ijt}}$ as the IV show that the endogeneity test of the endogenous regressor cannot reject the null hypothesis that the specified endogenous regressors are treated as exogenous. These results suggest that the endogeneity issue can be properly addressed by including country-pair fixed effects, as argued by Baier and Bergstrand (2007) and Yotov et al. (2016).

27 Although we only report the results of the PPML estimations, those of the OLS remain the same when using two-year lagged $\mathrm{RTA}\mathrm{\_Variabl}{e}_{\mathrm{ij}}$.

28 We thank an anonymous reviewer for pointing out this important issue.

29 Our sample in this subsection consists of 23 advanced countries.

Additional information

Funding

Financial support from the Japan Society for the Promotion of Science under the Grant-in-Aid for Scientific Research (B) Nos. 16H03619, 20H01507, and 22H00846 is gratefully acknowledged.