Learning through foreign market participation: the relative benefits of exporting, importing, and foreign direct investment

Abstract

A growing literature explores the degree to which firms learn from exporting. Although this literature finds that firms that export subsequently enjoy enhanced innovative performance, there has been little research that compares the effect of exporting to that of alternative internationalization activities. In this paper, we extend the literature to explore theoretically the differential effects of a firm’s exporting, foreign direct investment, and importing activity on its innovative outcomes. We test the resulting hypotheses using a sample of Spanish manufacturing firms from 2000 to 2008. We find that (1) learning associated with exporting is more pronounced than that associated with a firm’s FDI activities, (2) exporting and FDI operate as substitutes in their effect on a firm’s learning, and (3) although importing is positively associated with learning as manifested in new product introductions, it is not associated with learning as manifested in patenting activity.

Appendix: Testing for substitutability between export activity and FDI

Building on the supermodularity literature (Milgrom and Roberts 1990), Athey and Stern (1998) demonstrate that a comprehensive test of complementarity (substitutability) of discrete practices entails a “production function” approach that evaluates organizational performance differences across combinations of practices. As described by Lokshin et al. (2004), when there exist only two relevant practices, one may directly interpret the interaction term between these practices—a positive (negative) coefficient on the interaction term indicates that the practices are complements (substitutes). However, when there exist more than two practices, the interaction term is not easily interpreted. Put simply, in our study exporting and FDI will be substitutes only if their joint adoption reduces learning (compared to the sum of independent adoptions of the practices) both for the set of firms that do not import and for the set of firms that do import. As Lokshin et al. (2004) demonstrate, this will be true if both of the following conditions regarding the practices’ coefficients hold:

$$ \upbeta _{{{\text{Export}}\& {\text{FDI}}}} + \upbeta _{{{\text{None}}}} - \upbeta _{{{\text{ExportOnly}}}} - \upbeta _{{{\text{FDIOnly}}}} < 0 $$

(3)

$$ \upbeta _{{{\text{All}}}} + \upbeta _{{{\text{ImportOnly}}}} - \upbeta _{{{\text{Export}}\& {\text{Import}}}} - \upbeta _{{{\text{FDI}}\& {\text{Import}}}} < 0 $$

(4)

Recent studies have introduced and employed appropriate techniques to assess complementarity among types of cooperative R&D (Belderbos et al. 2006) and between IT and human resources practices (Aral et al. 2012). The basic test involves three steps. First, create a categorical variable for each possible combination of practices. As noted in the Data section above, in this study these encompass None, ExportOnly, FDIOnly, ImportOnly, Export&FDI, Export&Import, FDI&Import, and All. Thus, for this study we will be able to estimate:

$$ \begin{aligned} \lambda_{it} & = \exp (\beta_{1} None_{i,t - p} + \beta_{2} ExportOnly_{i,t - p} + \beta_{3} FDIOnly_{i,t - p} + \beta_{4} ImportOnly_{i,t - p} \\ & \quad + \beta_{5} Export\& FDI_{i,t - p} + \beta_{6} Export\& Import_{i,t - p} + \beta_{7} FDI\& Import_{i,t - p} \\ & \quad + \beta_{8} All_{i,t - p} + \beta_{9} W_{i,t - p} + \Sigma \rho_{j} y_{i,t - j} + u_{it} ) \\ \end{aligned} $$

where λ_it, W _i,t−p , Σρ _j y _i,t−j and u _it are defined as they were in the Data section above.

Next, estimate the model three ways: unconstrained, subject to constraint (3), and subject to constraint (4). Each estimation generates a log-likelihood that denotes the explanatory power of the model. Finally, use the likelihood-ratio test to compare the explanatory power of the unconstrained model to each of the constrained models. If the unconstrained model offers statistically significantly better explanatory power than each of the constrained models, then we can reject the null hypothesis that exporting and FDI are not substitutes.

Table 9 presents the estimation for the effect of each combination of exporting, FDI, and importing on new product introduction, using 1-year lags of internationalization activity. Model 1 presents the unconstrained estimation, while models 2 and 3 present the constrained estimations. All of the control variables from the paper’s previous estimations are included, with the exception of NewProductsLag3 (whose inclusion prevents convergence of the estimation). Since the variables span every combination of practices, the model does not include a constant term. As an aside, although the coefficients on the practices variables are negative, these do not indicate that each combination of practices depresses a firm’s new product introductions. Proper interpretation requires comparing the coefficients on various practices to that on NoneLag1, which represents the baseline condition of no internationalization activities at all. Compared to NoneLag1, six of the other seven variables have a positive effect on new product introduction (e.g., ExportOnlyLag1 has a coefficient of −2.147; −2.147 − (−2.949) = 0.802 > 0.) Thus, internationalization appears to have a positive effect on a firm’s new product introductions.

Table 9 Test of substitutability of a firm’s exporting and FDI on its new product introductions

As for the precise test of substitutability, a likelihood ratio test indicates that model 2 offers significantly less explanatory power than model 1 (χ² = 22.4, p < .01). Model 3 also offers significantly less explanatory power than model 1 (χ² = 11.6, p < .01). Thus, both constraint (3) and constraint (4) are satisfied in these estimations, indicating that exporting and FDI act as substitutes with respect to enhancing a firm’s rate of new product introduction. In unreported results we replicate these estimations using 2-year and 3-year lags of exporting, FDI, and importing activities. The results are qualitatively identical. Although we made no predictions about complementarity or substitutability between FDI and importing and between exporting and importing, in unreported results we also tested these relationships. Likelihood-ratio tests indicated no evidence of either complementarity or substitutability between these sets of activities.

Table 10 replicates these estimations using Patents instead of New Products as the dependent variable. Likelihood-ratio tests again indicate that the unconstrained model offers significantly better explanatory power than either constrained model. Again, results using 2-year and 3-year lags of exporting, FDI, and importing yield qualitatively identical results, and tests of complementarity/substitutability between exporting and importing and between FDI and importing offer no evidence of such relationships. We interpret the body of evidence as support for Hypothesis 2: exporting and FDI are substitutes in terms of their effect on a firm’s innovative learning.

Table 10 Test of substitutability of a firm’s exporting and FDI on its patenting activity