سرمایه گذاری مستقیم خارجی در فضا: روابط خودبازگشت (اتورگرسیو) فضایی در سرمایه گذاری مستقیم خارجی

There are a number of theoretical reasons why foreign direct investment (FDI) into a host country may depend on the FDI in proximate countries. Such spatial interdependence has been largely ignored by the empirical FDI literature, with only a couple recent papers accounting for such issues in their estimation. This paper conducts a general examination of spatial interactions in empirical FDI models using data on US outbound FDI activity. We find that estimated relationships of traditional determinants of FDI are surprisingly robust to inclusion of terms to capture spatial interdependence, even though such interdependence is estimated to be significant. However, we find that both the traditional determinants of FDI and the estimated spatial interdependence are quite sensitive to the sample of countries one examines.

Since 1980, worldwide foreign direct investment (FDI) has grown at a remarkable rate. According to Markusen (2002), in the latter half of the 1990s FDI flows grew annually by nearly 32%. When compared to the 1.5% annual growth in exports and the 0.6% annual increase in world gross domestic product (GDP), it comes as no surprise that this same period has seen the development of formal economic models of multinational enterprises (MNEs) and increased empirical investigation of factors driving FDI patterns. Development of formal MNE theory stems from Markusen (1984) and Helpman (1984). Markusen (1984) provides a general-equilibrium model where MNEs arise due to a market-access motive to substitute for export flows, or what is termed “horizontal” FDI. In contrast, Helpman (1984) develops a general-equilibrium model where MNEs arise due to the desire to access cheaper factor inputs abroad, or what is termed “vertical” FDI. Both are developed in a two-country framework and have spawned significant theoretical work on MNEs. Empirical work on the determinants of FDI over recent decades has mainly relied on a gravity-type framework, where market size and distance provide explanatory power, and have primarily used data on bilateral country-level FDI activity.2 A potential weakness of the standard theoretical and empirical work on MNEs and FDI is this reliance on the two-country (or bilateral) framework. Recent theoretical work has begun to relax the two-country assumption, leading to the development of alternative motivations for FDI. For example, recent work by Ekholm et al. (2003), Yeaple (2003), and Bergstrand and Egger (2004) develop models of export-platform FDI, where a parent country invests in a particular host country with the intention of serving “third” markets with exports of final goods from the affiliate in the host country.3 and 4 Alternatively, an MNE may set up its vertical chain of production across multiple countries to exploit the comparative advantages of various locales. This motivation has been developed in a model by Baltagi et al. (forthcoming) and termed “complex vertical.” While both of these forms of FDI would involve exports to third markets, the difference is that complex-vertical MNE activity would be associated with exports of intermediate inputs from affiliates to third market for further (or final) processing, before being shipped to its final destination. However, both export-platform and complex-vertical motivations imply that FDI decisions are multilateral in nature and, therefore, cannot be captured by a two-country framework. Other factors may also create interdependent FDI decisions across host destinations, including agglomeration externalities and imperfect capital markets that limit the funds an MNE has to invest abroad.5 The existence of multilateral decision-making has significant implications for empirical work on FDI, as multilateral decision-making means that FDI decisions across various host countries are not independent. Yet, estimating models of FDI where each observation measures activity between a separate bilateral country-pair does not allow for the potential interdependence between FDI decisions across host destinations. Empirical work allowing for the impact of third-country (or third-region) effects—much less, general interdependence across multiple host markets—is sparse. Head et al. (1995) look for evidence of agglomeration externalities in determination of Japanese FDI location in the US by examining patterns of related producers in states adjacent to the US state chosen by a Japanese affiliate. Their conditional-logit specification explicitly models an interdependence of the location decisions across all possible locales and their estimates provide evidence of agglomeration effects between bordering states for the Japanese automobile industry's FDI into the US. Head and Mayer's (2004) analysis of Japanese FDI patterns into developed Europe examines the effect of market potential using measures that include not only the host region's GDP, but also GDP of adjacent regions weighted by distance and other trade frictions. They find that regions with higher market potential attract more FDI and that this effect is robust to a variety of alternative measures of market potential and inclusion of agglomeration measures as in Head et al. (1995). While the discrete choice models used in Head et al. (1995) and Head and Mayer (2004) allow for potential interdependence of FDI decisions, such models impose significant restrictions on the data, (e.g., the assumption of the independence of irrelevant alternatives). Furthermore, they limit one to examining a discrete measure of FDI choice, not the magnitude of the FDI activity.6 A more flexible alternative is offered by standard spatial econometric techniques, which directly model spatial interdependence in a linear regression framework. The first paper to use spatial econometric techniques to examine FDI behavior is Coughlin and Segev (2000), which considers US FDI across Chinese provinces. The paper finds that a region's FDI is positively correlated with FDI into neighboring regions (a positive spatial lag), which is attributed to agglomeration economies. The only other paper to use spatial econometric techniques to examine FDI patterns is Baltagi et al. (forthcoming) whose approach is more closely related to ours.7 The paper first develops a model of MNE activity that allows for a variety of MNE motivations and then maps these into the implied spatial interactions that should be associated with each type of MNE motivation. The resulting econometric specification is then estimated using US outbound FDI for seven manufacturing industries across both developed and less-developed destinations. Their results find substantial evidence of spatial interactions, though they cannot definitively conclude whether export-platform or complex vertical FDI is more prevalent. In this paper we take a more general look at empirically modeling spatial interactions in FDI and ask some fundamental questions not yet addressed by the previous literature. First, to what extent does omission of spatial interactions bias the coefficients on the traditional regressor matrix in empirical FDI studies? Significant bias would call into question much of the existing empirical work and inference. Second, how robust are estimated spatial relationships in FDI patterns across specifications and samples? Given the existing literature, an obvious issue to examine in this regard is differences across samples of developed and less-developed countries. In addition, because of the nature of space and how this influences the interpretation of estimated coefficients, it is necessary to examine differences across geographic sub-samples. Finally, we ask, to what extent can we uncover evidence of various theories of FDI using these techniques and available data? To explore these issues, we use various samples of US outbound FDI from 1983 through 1998. We find that the estimated relationships of traditional determinants of FDI are surprisingly robust to the inclusion of terms to capture spatial interdependence, even though empirical patterns in the data suggest that such interdependence can itself be significant. Furthermore, after controlling for country-specific dummy variables, estimated effects of spatial terms are often insignificant. This result is analogous to Feenstra's (2002) finding that fixed effects can adequately control for third-country effects in gravity trade models, which Anderson and Van Wincoop (2003) show are crucial in modeling trade in the gravity framework. However, our analysis also reveals that both the traditional determinants of FDI and the estimated spatial interdependence are sensitive to the sample of countries examined. The fragility of estimated spatial interdependence in the country-level data suggests, generally, that tying such results back to motivations of FDI is a difficult task and depends crucially on the sample chosen. Nevertheless, our estimates are broadly suggestive of export-platform FDI in the developed European countries. The remainder of the paper proceeds as follows. In the next section we discuss hypotheses concerning the implications of various models of multinational firm behavior for spatial relationships between FDI into various regions. Section 3 provides a brief overview of spatial econometric methods and discusses our data. Section 4 reports our estimates and highlights the importance of including both market potential and a spatially-weighted dependent variable. Section 5 concludes.

There are a number of theoretical reasons why FDI into a host country may depend on the FDI in proximate countries. Such spatial interdependence has been largely ignored by the empirical FDI literature with only a couple recent papers accounting for such issues in their estimation. This paper conducts a more general examination of spatial interactions in empirical FDI models using data on US outbound FDI activity. We find that estimated relationships of traditional determinants of FDI are surprisingly robust to inclusion of terms to capture spatial interdependence, even though such interdependence is estimated to be significant in the data. However, we find that both the traditional determinants of FDI and the estimated spatial interdependence are quite sensitive to the sample of countries one examines. In particular, the geographic scope of the sample and level of disaggregation are quite important in trying to separate evidence supporting different motivations for FDI. These general results are quite important for the extensive previous work on FDI. Omitted variable bias from not modeling spatial interdependence is apparently quite small in these cross-country FDI estimations across the variety of samples we explore. This is good news for the statistical inference drawn by previous empirical studies regarding determinants of FDI. On the other hand, it is worth noting that we find significant omitted variable bias for the market potential measure or spatial lag when not including both in the specification. This point is particularly applicable to the few previous studies of spatial effects in empirical FDI patterns, as ours is the first to include both spatial effects. Furthermore, our results highlight that estimates of cross-country determinants of FDI are not very robust to changing the sample of countries. In a related vein, the fragility of estimated spatial interdependence in the country-level data suggests that tying such results back to motivations of FDI is a difficult task and depends crucially on the sample chosen. This is a potential explanation for why the Baltagi et al. (forthcoming) study that pools data across a wide variety of countries and industries does not reach unambiguous conclusions. However, once we pursue estimation of sub-samples and disaggregate our data we find evidence suggestive of export-platform FDI for most industries within the developed European countries.