5

Talent, Geography, and Offshore R&D

Paper Author and Presenter: Jingting Fan (The Pennsylvania State University)

Moderator: Eduardo Morales (Princeton University)

Jingting Fan, assistant professor of economics at The Pennsylvania State University, introduced his presentation by providing the motivation behind his new model of trade and innovation. Global economies have become more integrated because of increasing trade volumes and the activities of multinational enterprises (MNEs). There is a growing body of literature on the costs and benefits of globalization. Past literature has focused on offshore production of MNEs; in contrast, this paper focuses on offshore research and development (R&D), defined as the fraction of R&D that occurs in a host country by an affiliate firm of a foreign corporation. Fan’s workshop paper extends prior models of trade and R&D to enable exploration and understanding of the patterns of the global diffusion of R&D.

Fan explained that offshore R&D, which is R&D performed by affiliates of foreign corporations within a country relative to total domestic R&D activity, is the majority or substantial share of total R&D activity. Smaller economies are more reliant on R&D expenditures by these affiliates of foreign firms than are larger economies and, in general, R&D performed by affiliates of foreign firms is growing over the sample period (1985–2012). For example, in the United States, by the end of the sample period, offshore R&D increased from about 7 percent to 15 percent of total U.S. R&D (see Figure 5-1).

Fan described two components required for innovation and commercialization of new products: a talented workforce and know-how of firms. But he said there is a spatial mismatch between these two components. Countries like China and India have some of the largest talent pools, while the vast majority of well-run firms are in highly industrialized countries. Thus, offshoring R&D can have a direct impact by helping with this spatial mismatch, as well as an indirect effect on countries through trade and offshore production. There are several challenges to analyzing the impact of offshoring R&D:

• First, R&D and production are jointly determined and depend on market access.
• Second, calibrating the model to fit realistic geographic features requires firm-level data on the activities of MNEs in different regions, and such data are relatively scarce. Additionally, most firm-level data focus on one home country at a time.

Fan stated that his paper contributes to the literature by assembling a panel dataset that merges production and ownership data from Bureau van Dijk’s Orbis database with patent data from the European Patent Office’s PATSTAT database. Using this panel dataset, the paper documents several empirical relationships, including the role of human capital in affiliate R&D, colocation of affiliate R&D and production, and headquarter (HQ) effects for both affiliate R&D and production. These patterns are then interpreted and explored using counterfactual experiments in a structural model. There are two proposed mechanisms within the model: talent acquisition and market access. Used first as a measurement tool and second as a counterfactual experiment, the model produces the following results:

• First, around 70 percent of R&D in overseas affiliates is for domestic production, and offshore R&D is an important source of profit for firms headquartered in advanced economies. This

• indicates that the market-access motive is a strong incentive for firms to innovate abroad.
• Second, the movement from zero offshore R&D to positive offshore R&D generates 3.3 percent welfare gains on average and amplifies total gains from globalization by 33 percent. These estimates have significant bias from advanced economies, as the advanced economies benefit more from offshore R&D. There are also important interactions between offshore R&D and trade and offshore production.

DATA

Fan continued by stating that he merged data from the Orbis database with patent data from PATSTAT covering 37 countries with four periods over 1996–2016. Parent firms in the Orbis database are defined as an entity with at least 50 percent control of a firm. Affiliate behavior in a foreign country is the aggregate of all firms controlled by the parent in the given country. Patent behavior was matched to firms before being aggregated to three dimensions: parent firm, inventor country, and time. The paper accounts for a large amount of this variation by controlling for firm, industry, and host fixed effects and their interactions, in addition to other factors. The estimates of the model align well with expenditure data from the Organisation for Economic Co-operation and Development (OECD).

Fan moved on to describe the structure of the data. Here he explained an inference problem due to the difficulty in assigning invention location for given production. For example, the innovating location may be the home country for production in a host country, or vice versa. There are more dimensions to this problem, and the picture can quickly become quite complicated. To work around this inference concern, he uses the data to document a systematic distribution of innovation activity. Then, knowing the size and location of production, one knows the thickness, or size, of the flows of innovation going to each node in the network. Fan then parametrized the value of each link in the network, under some assumptions.

STYLIZED FACTS

Fan presented four stylized facts:

• First, firm heterogeneity plays an important role, as firms with more HQ innovation also have higher affiliate innovation in the extensive and intensive margins, and the affiliate firms have higher sales per invention.

• Second, there is a positive correlation between host human capital and affiliate innovation intensity, or the ratio of patents and sales, in cross-section and panel dimensions.
• Third, there are synergies to the colocation of innovation and production. For example, firms that conduct innovation in one host country are also more likely to produce in that country. This relationship is true when focusing on the intensive margin and changes over time.
• Lastly, there are benefits to being geographically close to the firm HQ, as both affiliate innovation and production decrease with distance to HQ.

MODEL

Fan moved next to discussing the model environment. The model is nested in the work of Arkolakis et al. (2018), with the inclusion of offshore R&D as the major change. There are N countries endowed with L workers who have ability (alpha), drawn from a distribution A. Workers either work in manufacturing and earn a common wage w, or work in a high-skill job and earn a premium wage (w * alpha). Firms are heterogeneous and are differentiated in their manufacturing productivity and their efficiency at innovation. A representative consumer maximizes a constant elasticity of substitution (CES) demand function, subject to a budget constraint.

The firm’s decision begins at the firm HQ in the home country (see Figure 5-2). The parent firm chooses to invest in R&D in two host countries. The affiliate firms can then choose to produce and sell in either the host country or another destination market. Fan provided the example of DuPont, a U.S. MNE with R&D labs in the United States, Brazil, China, Switzerland, South Korea, Germany, and Japan. DuPont produces in 19 countries around the world and sells final products in close to 90 countries.

Fan explained that firms that wish to invest and innovate in a host country pay a fixed entry cost. These firms retain some of the innovation efficiency gained from the parent firm, which is located in the home country, depending on the distance between the home and host countries, and gain a production efficiency drawn from a distribution. Firms can enter multiple host countries at once, incurring multiple fixed costs and drawing a unique productivity efficiency in each. Firms face a unit delivery cost of production determined by the retained productivity efficiency, the production wage, the manufacturing productivity, and the distance between the host country and the parent firm, as well as the distance between the host firm and the final market.

Fan stated that firms also incur a marketing cost in the final market and a shipping cost from the production location to the final market. For technical

reasons, there is an idiosyncratic component that varies across regions. The revenue generated by the final sale of products will be split across the value chain, leaving the net profit with the HQ.

The model includes horizontal innovation, which Fan described as new product blueprints/inventions that are differentiated from each other. Inventions are produced at a cost that takes into account both the distance between the production and R&D (via wage premiums and production efficiency), as well as between production and the final market (via shipping and marketing costs).

To calibrate the model to the real world, Fan explained, the paper uses country size as a proxy for manufacturing employment. For the distribution of talent, the paper follows Hanushek and Woessmann (2012), which is based on a country’s PISA score (from the OECD’s Programme for International Student Assessment) and the distribution of cognitive skill. For the firm know-how distribution, the World Management Survey from Bloom et al. (2012) is utilized. The production efficiency is the average of target, operations, and monitor scores, and the innovation efficiency is based on the talent score.

Fan modeled offshore production as a weighted average of the distance between the final market and the firm’s HQ, and the distance between the final market and the production site with host-specific parameters. This is similar to the approach used in Arkolakis et al. (2018), but with the inclusion of the offshore R&D centers. Offshore R&D costs are modeled as a function of host-specific parameters and a distance measure between the home and host countries. The host-specific parameters measure the overall openness to foreign production and R&D for both the intensive and extensive margins. The paper does this by taking

the share of foreign-owned firms that either produce or innovate in a host country. The distance measure is weighted by a coefficient from a regression measuring the HQ effect and benefits for colocation. A prediction of this model is that the proximity to the R&D center is more important for decreasing production costs than proximity to the HQ. Fan noted that extensive literature is available on the importance of HQ effects, but these results suggest that the proximity to the site of innovation is much more important.

RESULTS

Fan then transitioned to discussing the main results of his study. About 65 percent of domestic R&D is completed by domestic firms, and about 35 percent is conducted by foreign-owned firms, on average. The model predicts that the share of R&D activity for the purpose of production in the host country is higher among domestic than among foreign-owned firms. As shown in Figure 5-3, emerging economies have a higher share of R&D dedicated to domestic production than developed economies, both for domestic and foreign R&D. Foreign R&D contributes about 1.5 percent of total profits; however, there is large heterogeneity within foreign R&D contributions. For example, foreign R&D contributes 7.6 percent of total profits in the United States, whereas in Brazil, foreign R&D contributes 0.01 percent of total profits.

Fan also presented several counterfactual experiments. Offshore R&D generates about 3.3 percent welfare gains on average, which are larger for

developed countries. The introduction of offshore R&D amplifies the total gains from liberalization by 1.3, or a 30 percent increase. Offshore R&D acts as a substitute for trade and offshore production for developing countries, and as a complement for developing countries. The combination of general equilibrium effects and firm linkages leads to the conclusion that incorporating offshore R&D is crucial when evaluating trade and offshore production policies.

SUMMARY

In conclusion, Fan stated that his work calibrates a model of trade, offshore production, and offshore R&D, studying the determinants and welfare implications of offshore R&D. According to Fan, the inclusion of offshore R&D is a novel contribution, building on recent literature and noting several new stylized facts from the merging of the Orbis and PATSTAT databases. The theory-based measurements show that the empirical patterns are rich and informative, and the counterfactual experiments highlight the importance of offshore R&D for understanding the implications of globalization for welfare and income distribution. Finally, he concluded with a caveat, that his work includes no industry-level analysis and overlooks the role of outsourcing in both production and R&D.

DISCUSSION

Discussant: Gary Lyn (Iowa State University)

Gary Lyn, assistant professor of economics at Iowa State University, began his discussion by summarizing Fan’s work as a theoretical model consistent with three empirical facts: (1) the innovation intensity of an affiliate firm increases in a host country with a higher-quality talent pool; (2) innovation and production tend to be colocated; and (3) geography matters for the location of affiliate production and innovation. The model contains two motives for the observed trends in offshore R&D: talent acquisition and market access. Fan also conducted several counterfactual experiments to quantify the gains from R&D offshoring.

Lyn offered three broad comments on Fan’s work:

• First, the introduction in Fan’s paper would be enhanced by a discussion of the relationship of this work to Arkolakis et al. (2018), as the latter describes innovation as occurring only in the HQ country, and finds that countries that specialize in production may be hurt by offshoring R&D. This is related to the counterfactual in which no innovation efficiency is transferred. Additionally, in Fan’s paper, the variety-level labor productivity is drawn from a multivariate distribution. Arkolakis et al. (2018) allows for labor productivity to be correlated across draws. Allowing such correlation may offer interesting predictions.

• Second, related to work on R&D spillovers, Lyn cited Bilir and Morales (2020), which describes HQs as experiencing stronger spillovers to overseas affiliates than spillovers from one affiliate to another. In other words, there appear to be HQ effects for knowledge transfer. Lyn asked if there is an easy way to implement this behavior in Fan’s model. He suggested associating scale effects with employment in R&D in the home country, and then interacting these scale effects with the HQ innovation efficiency. Lyn’s discussion prompted an audience member to ask if there are empirical R&D spillovers from foreign affiliate firms to foreign nonaffiliate firms.
• Third, Lyn raised a question about the broader story being presented: Is there a systematic difference in the type of R&D activities of HQs and affiliates? In other words, are affiliates innovating on different parts of the value chain to make a single product? This relates to the paper presented by Anna Gumpert (Gumpert et al., Chapter 2) that breaks down R&D activities by basic, applied product, and applied process innovation, and presents some stylized facts about these relationships. In addition, Lyn wondered why vertical foreign direct investment is not a first-order segment of the narrative.

Fan began his response with a comment about the relation of his work to Arkolakis et al. (2018), saying that labor productivity correlation is already captured in the firm-level productivity draws. This means that all varieties of R&D will be produced with a common level of efficiency and generate firm-level correlation.

Second, regarding the difference in innovation types, Fan returned to the work of Gumpert and her coauthors as a path forward.

Third, regarding the spillovers between affiliate firms in different hosts, Fan pointed out a positive correlation between production and distance to affiliate R&D centers. As an example, an affiliate R&D center in China will have some influence on an affiliate production center in India. This implies a scope for cross-country within-firm spillovers. Fan said he has not thought about the within-country spillovers across different affiliates, but thinks it is an important angle for future work.

During the presentation, one audience member asked about how the model handles the relationship between sales and R&D expenditures in the context of HQ and affiliate firms. Fan responded that this would appear in the model when innovation occurs in the host country and production occurs in surrounding countries. Eduardo Morales of Princeton University asked a question about the impact of human-capital accumulation and the firm’s decision to offshore innovation, suggesting that the PISA score may not be the best measure. Fan responded that, while the PISA score is not perfect, it is the best measure available.