Introduction

Although the effectiveness of foreign aid in promoting economic growth in the recipient countries is controversially debated,Footnote 1 there appears to be a broad consensus that foreign direct investment (FDI) inflows bring many benefits to host countries. According to the so-called Monterrey Consensus achieved at the UN summit on Financing for Development in 2002, ‘foreign direct investment … is especially important for its potential to transfer knowledge and technology, create jobs, boost overall productivity, enhance competitiveness and entrepreneurship, and ultimately eradicate poverty through economic growth and development. A central challenge, therefore, is to create the necessary domestic and international conditions to facilitate direct investment flows’ (United Nations, 2003, p. 9, Paragraph 20).

Consequently, policymakers around the world have liberalized regulations and offered incentives to attract FDI inflows.Footnote 2 Yet FDI continues to be highly concentrated in a few host countries, whereas various developing countries hardly participated in the FDI boom. The distribution of FDI is skewed even within relatively advanced regions such as Latin America. Some countries, notably Chile and Panama, hosted outstandingly high FDI stocks, relative to GDP, in the mid-1980s already and still belonged to the top group 25 years later (Figure 1). In sharp contrast, countries such as Cuba and Venezuela ranked at the bottom at both points in time. Other Latin American countries changed their position considerably during this period: Whereas Guatemala suffered a steep decline, neighbouring Honduras and Nicaragua jumped from poor rankings in the mid-1980s to close the top in recent years.

Figure 1
figure 1

FDI stocks in percentage of GDP, 21 Latin American sample countries, 1984 and 2008. Source: UNCTAD online database.

Full size image

This raises the question of whether the international community could support the diffusion of FDI-related benefits by using aid as a means to ease access to FDI. Theoretically, foreign aid has ambiguous effects on FDI (Harms and Lutz, 2006; Kimura and Todo, 2010). On the one hand, aid may increase the productivity of private capital by improving the supply of complementary factors of production (Selaya and Sunesen, 2012). On the other hand, aid could have adverse effects on FDI by giving rise to rent-seeking (for example, Economides et al, 2008) and by crowding out foreign investment in the tradable goods sector (Beladi and Oladi, 2007). Yet a widely cited OECD report called on donors to improve ‘the synergies between FDI flows and ODA’ (OECD, 2002, p. 30). Beerfeltz (2011, p. 417), the under-secretary in the German Ministry for Economic Cooperation and Development, declared that German development aid shall ‘motivate companies to make more direct investments in our [development cooperation] partner countries’.

This could be achieved if well-targeted foreign aid removed critical impediments to higher FDI inflows, for instance by improving the endowment of host countries with sufficiently skilled labour on which foreign direct investors draw. Previous studies on the aid-FDI nexus have often employed aggregate aid data. This approach does not account for the heterogeneous nature of aid. Some recent studies disaggregate aid, as we do in the following. Nevertheless, an important gap remains that we attempt to fill. Apart from disaggregating aid, we consider the specific needs of host countries by identifying major bottlenecks to higher FDI inflows that foreign aid may help overcome. Our analysis addresses this issue by focusing on host countries in one particular region, Latin America, and aid in one particular category, education.

Our central hypothesis is that aid for education is an effective means to increase FDI flows to host countries where schooling and qualification can reasonably be considered inadequate from the viewpoint of foreign direct investors. This appears to be the case in large parts of Latin America. At the same time, the recent literature suggests that analysing disaggregated aid and its impact on narrowly defined outcome variables offers a more promising way to assess the effectiveness of aid, compared with earlier studies on the economic growth effects of aggregate aid. In particular, it has been found that aid for education improved educational outcome variables such as enrolment rates (Dreher et al, 2008), as well as completion rates, repetition rates and gender parity (D’Aiglepierre and Wagner, 2010). This provides an important channel through which aid for education could have promoted FDI in Latin America.

We employ panel data techniques covering 21 Latin American countries over the period from 1984 to 2008. We find that aid for education has a statistically significant and positive effect on FDI. This effect is robust to potential outliers, the selection of Latin American sample countries, alternative specifications and different estimation methods. Before presenting our results in detail, we discuss the relevant literature, which offers several building blocks on which our central hypothesis on the effects of aid for education on FDI flows to Latin American countries rests.

Analytical Background and Related Empirical Evidence

Relevance of Education for FDI

North–South models of FDI suggest that a sufficiently qualified workforce is an important pull factor as foreign direct investors rely on relatively skilled labour in developing host countries. Similarly, empirical studies consider the endowment of host countries with human skills to be an important determinant of FDI (for example, Noorbakhsh et al, 2001). In the Latin American context, Mexico has received particular attention in this regard (for example, Aitken et al, 1996; Feenstra and Hanson, 1997). Hanson (2003) concludes from a survey of the earlier literature that FDI has increased the relative demand for skilled labour in Mexico. The observation that FDI draws on relatively skilled labour in developing host countries supports the theoretical predictions of North–South models of FDI. In particular, Feenstra and Hanson (1997) argue that FDI may increase the skill premia not only in the advanced source countries of FDI (by offshoring the relatively unskilled labour intensive lines of production), but also in the less advanced host countries. FDI-related activities tend to be relatively skilled labour intensive in the host country, even though they are relatively unskilled labour intensive by the standards of the source country.

Insufficient education and worker qualification could discourage FDI inflows, particularly in middle-income countries where local governance structures and essential physical infrastructure are no longer binding constraints. Comparative evaluations and surveys on schooling, education and qualification indicate that various Latin American host countries may fall into this category. The Global Competitiveness Report of the World Economic Forum, which ranks a large set of countries across the whole spectrum of GDP per capita with respect to several educational indicators, reveals that our Latin American sample countries cluster at the bottom of the ranking with respect to quality aspects of education. Furthermore, almost all Latin American sample countries fall below the ‘normal pattern’ when the quality of education is related to the countries’ average GDP per capita. In other words, Latin American countries typically fall behind the quality of education to be expected at their level of economic development.Footnote 3

Aid and Educational Outcomes

According to data from UNESCO, public expenditure on education (in percentage of GDP) varies considerably within Latin America.Footnote 4 However, Michaelowa and Weber (2007) find no compelling evidence that higher domestic expenditure on education improves outcomes in terms of school enrolment and completion rates. Similarly, domestic expenditure on education has insignificant effects on educational outcome variables considered by Dreher et al (2008), as well as D’Aiglepierre and Wagner (2010).

In striking contrast to domestic expenditure, foreign aid for education appears to be effective in improving educational outcome variables. Michaelowa and Weber (2007) find that aid for education increases primary education, even though the impact of aid is rather small and conditional on local governance. Similarly, Dreher et al (2008) show that higher per-capita aid for education significantly increases primary school enrolment. This result proves to be robust to the method of estimation, the use of instruments to control for the endogeneity of aid and the set of control variables included in the estimations. D’Aiglepierre and Wagner (2010) focus on aid for primary education and consider a broader spectrum of educational outcome variables, including completion and repetition rates, as well as gender parity. Aid in this particular category proves to be strongly effective. Christensen et al (2011) compare aid for primary education from bilateral and multilateral donors. They find that bilateral donors condition their aid on better control for corruption in the recipient countries, which renders bilateral aid more effective in boosting school enrolment. As shown by Birchler and Michaelowa (2013), however, the quality of education may suffer when enrolment rates increase rapidly. Even though appropriate data on the quality of education in developing countries are scarce, it appears that ‘donors may have focused on quantity to the detriment of quality’ (Birchler and Michaelowa, 2013, p. 16).Footnote 5

Taken together, these strands of the literature invite the hypothesis that aid for education helps attract FDI inflows, notably to where schooling and education appear to be deficient, as in large parts of Latin America. This is despite the fact that previous research offers little insights on whether the link between aid for education and educational outcomes would also hold in the specific Latin American context.Footnote 6 It cannot be taken for granted that this is the case.Footnote 7 However, we collect some stylized facts to render our hypothesis more plausible. Figure 2 portrays the development of completion rates at different levels of schooling for our sample of Latin American countries, drawing on the well-known Barro-Lee database.Footnote 8 As can be seen, the increase in aid for education since the mid-1980s for the Latin American sample is correlated positively with completion rates at the secondary and tertiary levels of schooling.

Figure 2
figure 2

Aid for education and highest schooling level attained (percentage of population age 15 and more that completed schooling levels): Average of 21 Latin American sample countries, 1985–2010.Solid line gives 5-year averages of aid for education, starting with 1984–1988 (see text for details); right-hand scale.

Full size image

Furthermore, we draw on country-specific information on average years of schooling available for all 21 Latin American sample countries from the Barro-Lee database. More precisely, we calculate the change in average years of schooling, covering all levels of schooling, during the whole period of observation (2010 compared with 1985) and, alternatively, during 5-year intervals since 1985. The changes in average years of schooling are then correlated with aid for education in the corresponding periods of time (that is, average aid, relative to the recipient country’s GDP, throughout the 1984–2008 period or, alternatively, during 5-year sub-periods starting with 1984–1988). This simple exercise results in correlations that are positive, though not particularly strong, providing further indications that aid for education may have contributed to improved educational outcomes in Latin America. Specifically, the correlation coefficient is 0.26 when considering aid for education and changes in years of schooling for the cross-section of the 21 sample countries over the whole period of observation.Footnote 9 Not surprisingly, the correlation coefficient is lower (0.13) when pooling the five sub-periods available for each country in our sample.

Obviously, these simple correlations do not necessarily imply that aid for education is causal for improved educational outcomes in Latin America. Providing a deeper and comprehensive analysis of the effects of aid for education on outcomes in Latin America would also require data on the quality of education, which are hardly available over time. The more modest objective of presenting Latin-American-specific stylized facts is to indicate that the general link between aid for education and educational outcomes may also hold for this region.

Previous Literature on Aid and FDI

Our focus on aid with the explicit purpose of removing educational bottlenecks in the recipient countries deviates from previous studies on the links between aid and FDI. Almost all of these studies apply aggregate aid data, starting with Papanek (1973) who observed a statistically insignificant correlation between aid and FDI across countries in the 1950s and 1960s. Whereas Papanek (1973, p. 123) rejected the view that ‘aid is biased in favour of the countries which are hospitable to (and often exploited by) the private investors of aid donor countries’, Berthélemy and Tichit (2004) find some evidence that donors grant more aid to host countries of FDI.

The study by Harms and Lutz (2006) is the most prominent on whether aid stimulates private foreign investment. Using data for 92 developing host countries during the 1988–1999 period, Harms and Lutz find that aggregate aid per se has no significant impact on foreign investment flows.Footnote 10 Surprisingly, however, the effect of aid proves to be strictly positive ‘where firms have to cope with substantial restrictions on their activities’ (Harms and Lutz, 2006, p. 780). Karakaplan et al (2005) concur that aid per se has no positive effect on FDI. In contrast to Harms and Lutz, however, Karakaplan and colleagues show that aid is more likely to induce FDI in host countries with better governance and more developed financial markets. Asiedu et al (2009) find that aid per se is negatively associated with FDI in low-income host countries, but aid tends to mitigate the adverse effects of country risk on FDI. Unconditionally positive effects of aid from bilateral sources, though not from multilateral sources, are reported by Yasin (2005) whose panel analysis covering the 1990–2003 period is restricted to 11 Sub-Saharan Africa countries.Footnote 11 Blaise (2005, 2009) considers Japanese aid to be a determinant of Japanese FDI in China and, respectively, in four South-East Asian countries. Employing conditional logit analyses based on firm-specific data, both case studies reveal that aggregate Japanese aid had a significantly positive impact on the location choices of Japanese direct investors.Footnote 12

All these studies ignore the sector-wise composition of aid. This may help explain the highly ambiguous results. As a first step towards disaggregating aid, Kimura and Todo (2010) distinguish between five major donors and separate ‘aid for infrastructure’ from other aid (mainly budget support, debt relief and humanitarian aid). Both types of aid prove to be insignificant as determinants of FDI, except for FDI from Japan. Kimura and Todo focus on the so-called vanguard effect of Japanese aid promoting Japanese FDI, although not FDI from other sources, but they hardly consider truly sector-specific aid. Their definition of ‘aid for infrastructure’ is extremely broad and includes aid for projects related to social and economic infrastructure, as well as aid for production activities and so-called multi-sector aid.

Selaya and Sunesen (2012) refine major aid categories to address the theoretical ambiguity mentioned in our introduction above. Specifically, projects related to social and economic infrastructure are supposed to attract FDI by improving the supply of complementary factors of production. By contrast, aid is supposed to crowd out FDI when granted as ‘pure physical capital transfers’ (Selaya and Sunesen, 2012, p. 2155). Indeed, the empirical analysis reveals the expected opposing effects of both types of aid on FDI, even though the categorization of aid is still fairly broad and not related to specific ‘needs’ or bottlenecks to FDI in the recipient countries.Footnote 13 Kapfer et al (2007) focus on aid for economic infrastructure (communication, transportation, energy), which they find to have a significant effect on FDI. Mayer’s (2006, p. 34) analysis of dyadic aid and FDI patterns suggests, however, that the ‘very strong effect [of aid for infrastructure] seems entirely caused by the cross-sectional variation in the data’ and largely disappears once country-pair fixed effects are included.

Mayer (2006) represents the only study that also considers aid for social infrastructure as a distinct determinant of FDI, with similarly sensitive results as in the case of aid for economic infrastructure. We suspect that social infrastructure is still too broad a concept to capture specific bottlenecks to FDI that sector-specific aid may help overcome. The OECD’s aid statistics subsume not only aid for education, health, and water and sanitation under ‘social infrastructure’, but also aid projects related to government administration and civil society. As a matter of fact, education accounted for just about 20 per cent of total aid commitments listed under ‘social infrastructure’ by all donors in recent years (2005–2010).Footnote 14

Empirical Analysis

The analysis in this section examines the relationship between aid for education and FDI in Latin America. We first describe the empirical model and the data. Subsequently, we present fixed-effects estimates of the impact of aid for education on FDI.

Empirical Model and Data

Our model is of the general form:

where i=1, 2, …, N is the country index; t=1, 2, …, T is the time index; FDI represents net FDI inflows relative to GDP; and Aid stands for net aid flows relative to GDP.Footnote 15 In line with our reasoning above, we decompose Aid into aid for education, Aidedu, and all other (non-education) aid, Aidother. X is the usual vector of m time-varying control variables. Following Harms and Lutz (2006), we control for GDP per capita (GDPpc), the trade-to-GDP ratio (Trade), governance (Governance) and investment risk (Investment risk). We include fixed effects, αi, to control for any country-specific omitted factors that are stable over time. We also include period dummies, αt, to account for common time effects such as shocks affecting all countries at the same time, as is standard in the literature.

The empirical analysis covers the period from 1984 to 2008.Footnote 16 As is common practice in panel studies, we use time-averaged data to eliminate business cycle effects. Specifically, we construct 5-year averages as in Selaya and Sunesen (2012). This gives us five periods for our panel (1984–1988 until 2004–2008). We include all Latin American countries with available data, with the exception of countries with a population below 1 million.Footnote 17 This yields a sample of 21 Latin American countries: Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Cuba, Dominican Republic, Ecuador, El Salvador, Guatemala, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay and Venezuela.

We now describe the data used in the empirical analysis. For our FDI variable, we use net FDI inflows from the United Nations Conference on Trade and Development (UNCTAD) FDI database (available at unctadstat.unctad.org). As noted before, we distinguish two types of aid: aid for education and aid for other purposes. Both categories of aid are based on aid commitments reported in the Creditor Reporting System (CRS) database of the Development Assistance Committee (DAC) of the OECD (available at www.oecd.org/dac/stats/idsonline). Aid for education includes aid for basic education, secondary education, post-secondary education and unspecified levels of education according to CRS purpose code 110. Aidother comprises all other CRS purpose codes except Code 110.

Principally, it would be more appropriate to use aid disbursements, instead of aid commitments, as the effects of aid should depend on actual flows rather than donor promises that are often not fully met. However, data on sector-specific aid disbursements suffered from serious underreporting until recently. The reporting by donors of sector-specific disbursements is almost complete only since 2002, whereas it covered only a small fraction of actual disbursements throughout the 1980s and 1990s.Footnote 18 The evidence for our sample corroborates serious underreporting of disbursed aid for education until recently.Footnote 19

Coverage of commitments is generally better than for disbursements, though far from perfect in the more distant past. The reporting of sector-specific commitments is widely perceived to be satisfactory since the mid-1990s. We extend the period of observation to 1984–2008 as careful inspection of the data points to relatively small data gaps between the sum of reported sector-specific commitments in the CRS and the DAC statistics on overall commitments (where underreporting is no major issue) for our sample of 21 Latin American countries.Footnote 20

Therefore, we follow previous studies and use aid data on a commitment basis in our baseline estimations (see, for example, Mayer, 2006; Dreher et al, 2008; D’Aiglepierre and Wagner, 2010; Kimura and Todo, 2010; Christensen et al, 2011). In addition, we perform estimations with estimated disbursements of aid for education along the lines suggested in some previous studies, notably Michaelowa and Weber (2007) and Selaya and Sunesen (2012). Specifically, we combine the CRS data on sector-specific commitments with the aggregate DAC statistics on disbursements, as well as commitments to arrive at estimated sector-specific disbursements. The CRS data on commitments of aid for education are adjusted to mitigate two biases: (i) we use the ratio of total aid disbursements over total aid commitments as available from DAC statistics to account for a potential upward bias as commitments tend to exceed actual disbursements to the extent that donors renege on earlier pledges; (ii) we multiply with the ratio of total aid commitments from DAC statistics over the accumulated project-based commitments in all sectors as given in the CRS to account for the downward bias due to underreporting of project-based aid in the CRS.

Our control variables are drawn from the previous empirical literature on the aid–FDI relationship. Data for GDP per capita (in $US1000) and trade (exports plus imports relative to GDP) are from the World Development Indicators (WDI) online database. Following Harms and Lutz (2006) and Kapfer et al (2007), we lag both variables one period to alleviate potential endogeneity problems.

As far as the measure of governance is concerned, the often used World Governance Indicators from the World Bank are available only from 1997 onwards. Therefore, we follow Kapfer et al (2007) and include the democracy index from the POLITY IV database (available at www.systemicpeace.org/polity/polity4.htm). This index ranges from –10 (strongly autocratic) to +10 (strongly democratic). The choice of this variable is based on the following considerations: there is evidence to suggest that the level of democracy is both a determinant of aid (for example, Alesina and Dollar, 2000) and a determinant of FDI (for example, Jensen, 2003). Thus, one should control for democracy to avoid omitted variable bias. Moreover, the level of democracy appears to be a good proxy for the quality of governance. Li and Resnick (2003, p. 187), for example, point out that ‘democratic institutions […] collectively serve to secure private property rights and lower the risks of expropriation, contract repudiation, ineffective rule of law, and government corruption […]’. Nevertheless, the relation between democracy and FDI is theoretically ambiguous. For instance, Li and Resnick (2003) argue that democratic governments are more likely to impose restrictions on multinational enterprises to prevent them from taking advantage of monopolistic positions. FDI could be discouraged by such restrictions. Therefore, we consider two alternative, more specific measures of governance in the robustness tests: the Freedom House Civil Liberties index and the Freedom House Political Rights index (available at www.freedomhouse.org).

Finally, in line with Harms and Lutz (2006) and Selaya and Sunesen (2012), we include a measure of investment risk: the investment profile index from the International Country Risk Guide, published by the Political Risk Services Group (see www.prsgroup.com/prsgroup_shoppingcart/pc-75-7-icrg-historical-data.aspx). This measure assesses several factors affecting the risk to investment, including contract viability and payment delays, and ranges from 0 (very high risk) to 12 (very low risk).Footnote 21

Baseline Results

In Column (1) of Table 1, we present the baseline results with our aid variables defined in terms of commitments as reported in the CRS. The effects of most of the control variables on FDI are in line with previous studies. The coefficients on the volume of trade and GDP per capita are positive, although only GDP per capita seems to have a statistically significant influence on FDI flows. As in previous studies (for example, Harms and Lutz, 2006), we also find lower investment risk and FDI to be significantly and positively correlated (recall that higher values imply a less risky business environment). More surprisingly, the negative coefficient on our governance variable suggests that more democratic regimes attract less FDI. Although Kapfer et al (2007) report the same finding, it is clearly at odds with the view that democratization induces more FDI through better governance in the broadest sense. It rather appears that FDI is discouraged by restrictions on the activities of multinational enterprises that democratic governments are more likely to impose (Li and Resnick, 2003).

Table 1 Baseline results for aid commitments and estimated disbursements
Full size table

Turning to the variables of major interest, we do not find any statistically significant influence of aid to other sectors (Aidother) on FDI inflows. This is in line with the ambiguous results of previous studies using aggregate aid data. By contrast, aid for education is positively associated with FDI in Latin American countries. The t-value of Aidedu is highly significant, and the point estimate implies that an increase of Aidedu by one standard deviation increases the FDI-to-GDP ratio by more than 1 percentage point – an economically large effect.

As discussed above, sector-specific aid data are underreported in the CRS. The degree of underreporting becomes more serious the further one moves back in time. Therefore, we replicate the baseline estimation by excluding the first 5-year interval (1984–1988) in Column (2) of Table 1. Indeed, Aidedu loses its statistical significance when omitting the most distant past in this way.Footnote 22 The finding that aid for education is no longer effective in stimulating FDI could have different explanations. On the one hand, the poor quality of CRS data in the more distant past may bias our results and lead us to overstate the impact of aid for education in Column (1), even though underreporting appears to be relatively modest for our Latin American sample. On the other hand, the variation over time is limited once the fixed-effects estimations are based on a reduced number of time intervals.Footnote 23

A definite conclusion on the validity of these two explanations is almost impossible unless longer time series of fully reliable aid data become available. Yet, the following estimations suggest that the impact of aid for education is understated in Column (2), rather than being overstated in Column (1). In Columns (3) and (4), we adjust the sector-specific CRS commitments in two steps to arrive at estimated disbursements. In the first step, we multiply sector-specific CRS commitments with the ratio of overall aid disbursements over overall aid commitments from the DAC’s aggregate aid statistics, in order to correct for a possible upward bias of sector-specific commitments. In the second step, we also account for possible underreporting in the CRS. The estimations with these two variants of estimated disbursements of aid for education closely resemble the baseline findings on the effect of Aidedu on FDI inflows.

Finally, we need to ensure that our results are not subject to spurious regression problems because of the potential non-stationarity of the data. As is well known from the growing literature on non-stationary panel data, even panel regressions may be spurious when the regression residuals are non-stationary. Following Eberhardt et al. (2013), we therefore apply the ADF Fisher panel unit root suggested by Madalla and Wu (1999) to the residuals of our models. The corresponding p-values are reported in the bottom part of Table 1. As can be seen, the null hypothesis of non-stationarity is rejected at the 1 per cent level for the residuals from all models, suggesting our results are not spurious.

Delayed Effects

The baseline estimations reported in Table 1 assess the effects of aid granted during a 5-year interval on FDI inflows in the same 5-year interval. This approach may capture delayed effects of aid granted at the beginning of the interval, whereas aid granted at the end of the interval could stimulate FDI inflows only if foreign investors anticipated its effectiveness in improving the host country’s endowment of sufficiently qualified labour. This approach probably fails to capture delayed effects on FDI fully, in particular when using commitments of aid. It takes time until committed aid is disbursed, and still more time until disbursed aid eventually improves the education of the workforce.Footnote 24

In the following, we modify the estimation approach to assess delayed effects more fully and, correspondingly, reduce the reliance on anticipation effects. More precisely, we estimate the effects of our aid variables in t–t+4 on FDI inflows in (i) t+1–t+5, (ii) t+2–t+6 and (iii) t+3–t+7. The results are presented in Columns (2)–(4) of Table 2, together with the benchmark estimation from Column (1) of Table 1 for ease of comparison. The results are strikingly similar across the different lag structures. In particular, Aidedu enters significantly positive in all estimations shown in Table 2, even though the level of significance weakens to the 5 per cent level when delayed effects on FDI are taken into account.Footnote 25

Table 2 Delayed effects
Full size table

At least implicitly, the results in Table 2 suggest that foreign investors anticipate the effects of aid for education on the country’s endowment of sufficiently qualified labour. To the best of our knowledge, the relevance of anticipation effects on FDI has rarely been addressed in the earlier aid literature. However, other strands of the literature on the determinants of FDI provide ample evidence on anticipation effects. In particular, several studies find that FDI inflows increase well in advance of the widening and deepening of regional integration. Egger and Paffermayr (2004) conclude that anticipation effects are common to various steps of EU integration. Indeed, these authors find that the effects on FDI are ‘mainly anticipatory’ and ‘seem to be exhausted with the formal completion’ (p. 108). According to Alguacil et al (2008), new accession countries experienced a boom in FDI inflows before their actual adhesion to the EU. In the Latin American context, it has been shown that Mexico attracted rising FDI flows before the ratification of the North American Free Trade Agreement (NAFTA). Kose et al (2004, p. 11) argue that ‘there was an anticipation effect after the member countries agreed to pursue negotiations for a free trade agreement in 1991’, that is, 3 years before NAFTA became operative (see also Salvatore, 2007).

The evidence that foreign direct investors anticipate changes in relevant FDI parameters is not restricted to the effects of regional integration. The model of Belderbos et al (2004) predicts FDI in anticipation of the imposition of anti-dumping duties. Azrak and Wynne (1995) support this proposition empirically for Japanese FDI in the United States. Graham and Wada (2001, p. 6) suspect that the rise of FDI approvals by Chinese authorities at the beginning of this century was ‘in anticipation of reforms that are likely to accompany Chinese entry into the World Trade Organization’. Kwok and Tadesse (2006, p. 776) even ponder the possibility that ‘the prediction of low corruption in the future attracts more FDI today’. In the context of foreign aid, Mayer (2006, p. 45) argues that ‘commitments can have a large signalling role for foreign investors, who can be affected by them even if not all commitments actually end up in disbursements’. Against this backdrop, it appears not far fetched to attribute the positive correlation between aid commitments and concurrent FDI decisions at least partly to the expectation of foreign investors that aid for education may prove effective in improving educational outcomes in host developing countries where education traditionally constituted a major bottleneck to FDI.

Robustness Tests

We perform several sensitivity tests in order to examine the robustness of the significantly positive effect of aid for education on FDI.Footnote 26 First, we re-estimate Equation (1) by excluding one country at a time from the sample. In this way, we test whether the positive effect of aid for education on FDI is the result of an individual outlier. This is not the case. The estimated coefficients are relatively stable and always significant at the 5 per cent level (not shown).

Second, we address sample selection issues by excluding various sub-groups of Latin American countries – for example, countries in a particular part of the region or countries with particular characteristics (Table 3). The resulting coefficients on Aidedu are all significant at least at the 5 per cent level, suggesting that the positive effect of aid for education on FDI is not due to sample selection bias. Third, we re-run the estimation for two samples outside Latin America. The bottom part of Table 3 reveals that the robust and positive coefficient on Aidedu does not carry over to Sub-Sahara Africa and 12 Asian countries. This tends to support our reasoning that aid for education is effective in stimulating FDI only where education appears to be the major bottleneck to higher FDI inflows.

Table 3 Estimates of the effect of aid for education on FDI using different samples
Full size table

Fourth, our main result on Aidedu in Latin America proves to be robust to using alternative definitions of aid for education and FDI, as well as alternative indicators of governance. Fifth, the same applies when extending the baseline specification by a measure on government ideology (not shown).

Alternative Estimation Methods

Finally, we employ alternative estimation methods to address endogeneity concerns with regard to Aidedu. In Column (1) of Table 4, we re-estimate Equation (1) by two-stage least squares (2SLS) to account for the potential endogeneity of aid, using as instruments lagged variables of aid for education, the log of population, the lagged log of population and the log of the infant mortality rate. These instruments are widely used in the aid-growth literature and prove to be relevant and exogenous. The coefficient on Aidedu is again statistically significant, corroborating the positive effect of aid for education in the baseline specification. Surprisingly, re-estimating Equation (1) by 2SLS suggests that the fixed-effects estimate tends to be biased downwards. However, a Durbin–Wu–Hausman test reveals that the 2SLS results are not necessarily superior.

Table 4 Alternative estimation methods
Full size table

Indeed, there is no compelling evidence for a downward bias of aid effects on FDI when employing a Generalized Methods of Moments (GMM) IV estimator in Column (2) of Table 4, our alternative approach to account for the possible endogeneity of aid. We use the well-known Blundell and Bond system GMM estimator. As our sample includes only 21 countries, we reduced the number of instruments from a maximum of 31 unrestricted instruments to 19 restricted instruments by using only two-period lags of aid for education (and the lagged dependent variable) as instruments. The Hansen J-test and a serial correlation test (AR) suggest that the instruments are valid and that the errors in the first-difference regression exhibit no second-order serial correlation.

It should be noted that the lagged FDI variable is not significant in Column (2) of Table 4. This implies that aid for education would have only short-run effects on FDI. Importantly, however, the coefficient on Aidedu is positive and highly significant once again. According to the GMM results, the short-run effect of Aidedu on FDI is of similar magnitude as in the baseline estimation in Table 1. All in all, we tend to prefer the estimates in Table 1 over those in Table 4. This is also because the GMM estimator is designed for large N (and small T) and thus may be biased in small country samples (as here).

Conclusion

Well-targeted sector-specific foreign aid could possibly remove critical impediments to higher FDI inflows and thereby help diffuse FDI-related benefits across a wider spectrum of developing host countries. Specifically, we raise the hypothesis that aid for education is an effective means to increase FDI flows to host countries where schooling and qualification appear to be inadequate from the viewpoint of foreign direct investors. This is the case in large parts of Latin America.

Our results provide strong empirical evidence that aid for education is indeed associated with higher net FDI inflows to developing countries in Latin America. Employing data for 21 countries over the period from 1984 to 2008, we find no statistically significant effect of other (non-education) aid. By contrast, aid for education proves not only statistically significant, but also has a quantitatively important positive impact on FDI flows to Latin American economies. In our baseline estimation, an increase in the ratio of aid for education over GDP by one standard deviation raises the FDI-to-GDP ratio by more than 1 percentage point. This finding is robust to potential outliers, sample selection and different variable definitions. The impact might even be stronger when using different estimation techniques to correct for potential endogeneity issues.

Our findings suggest that aid can be effective even though the relation between aggregate aid and economic growth appears to be elusive. This underscores the need to disaggregate aid and assess its effects on more specific outcome variables. To further explore possible synergies between aid and FDI flows (OECD, 2002), the case of aid for education in Latin America invites future research into the alignment of sector-specific aid with FDI-related needs in particular host countries or regions. For instance, aid targeted at fighting HIV/AIDS could improve access to FDI for countries with particularly high infection rates. Similarly, aid may help upgrade physical infrastructure and thereby remove critical impediments to higher FDI flows to where physical infrastructure is particularly deficient. At the same time, deeper insights into the relationship between aid and FDI could be gained if inward FDI was differentiated by sectors and industries. It clearly deserves more attention whether sector-specific aid such as aid for education attracts FDI to certain sectors and particular types of FDI, though not necessarily other types.