Exchange Rate Regimes and Foreign Direct Investment Flows to Developing Countries

Exchange Rate Levels and FDI

Early work (e.g. Mundell, 1968) denied the possibility of any link between exchange rates and FDI, arguing that changes in the exchange rate level cannot provide systematic cost-of-capital advantages to foreign or domestic firms since, under perfect capital mobility, risk-adjusted expected returns on all international assets will be equalized. These early conclusions did not go unchallenged. Alexander and Murphy (1975) investigated the impact of exchange rate changes on capital flows using the internal rate-of-return method and their findings supported the contention that the US dollar devaluations of the 1970s induced FDI flows into the USA. Cushman (1985) emphasized the substitution effect between exports and FDI in serving the host-country market. An appreciation of host-country currency would make FDI less attractive as exports rose. His empirical results were consistent with this interpretation. To rationalize the apparent contradiction between traditional theory and subsequent evidence, additional models (Froot and Stein, 1991; Campa, 1993; Blonigen, 1997) emerged in the 1990s to shed light on the relationship between the level of the exchange rate and FDI. While the models by Froot and Stein (1991) and Blonigen (1997) predict, albeit through different channels, a negative relationship between the exchange rate level and inward FDI, Campa (1993) argued that a higher value of the host country's currency would increase investment into the host country since the expectation of future profits would be higher.

Effects of Exchange Rate Volatility on FDI

Studies that have examined the relationship between exchange rate volatility and FDI flows can be categorized into two broad strands. One strand uses a straightforward “risk-aversion framework,” where exchange rate risk may reduce the certainty-equivalence value of expected revenues from foreign investments. For example, Cushman (1985) considered the relationship between exchange rate volatility and FDI, and concluded that volatility could either discourage or encourage FDI, depending on the differences in the locations of production and sales. For example, if higher volatility discouraged exports, serving the market via FDI could actually become more attractive, leading to a positive relationship between FDI and volatility. The second strand is rooted in the work of Dixit and Pindyck's (1994) “option-value framework,” where volatility increases the value of holding on to the option thus leading the firm to postpone the investment (see Campa, 1993). Against such theoretical blueprints some studies also juxtapose a “production flexibility” approach, whereby MNCs have the flexibility to shift production to the country where it is cheapest in response to exchange rate movements (see Goldberg and Kolstad, 1995).

The empirical results have been mixed, with findings of positive effects on FDI (Cushman, 1985, 1988; Goldberg and Kolstad, 1995; Görg and Wakelin, 2002), negative effects (Campa, 1993; Chakrabarti and Scholnick, 2002; De Vita and Abbott, 2007; Udomkerdmongkol et al., 2009), and insignificant effects (De Menil, 1999). To some extent the varying empirical results may reflect the varying theoretical predictions that depend on the assumptions concerning the risk preferences of foreign investors, the marginal utility of capital, where finance is raised and output sold, cost reversibilities, host-country demand and input cost conditions, and the timing of entry and production decisions. Regression models adopted by large scale applied studies of FDI flows among countries are evidently incapable of accounting for the theoretical assumptions germane to the properties of individual investment transactions let alone the attitude to risk of individual investors. Lack of data simply precludes this. Conflicting empirical evidence may also stem from specification problems, most notably, endogeneity issues. Russ (2007) showed that results of partial equilibrium analyses based on regressing FDI flows on any measure of exchange rate volatility may be subject to endogeneity bias.

Effects of Exchange Rate Regimes on FDI

We now turn to the explanatory variable of interest, the exchange rate regime. In many presentations, the issue is viewed as being directly related to exchange rate volatility. For example, when analyzing fixed vs flexible exchange rates, textbook discussions (e.g. Krugman and Obstfeld, 2009, p. 539) often include the argument that a flexible rate regime will discourage international trade and investment in general because of higher risk relative to a fixed rate regime.

It has, of course, been recognized that exchange regime effects on FDI can involve more than just the reduction of exchange rate volatility. For example, Schiavo's (2007) empirical model did allow for a currency union to have an effect in addition to that from reduced exchange rate volatility. To motivate the idea, Schiavo (2007) suggested that a currency union could reduce “transactional and informational barriers” (Schiavo, 2007, p. 545). He went on to search for a regime effect from the introduction of the European Monetary Union (EMU) in a sample of Organisation for Economic Cooperation and Development (OECD) countries for the period 1980–2001. Applying ordinary least squares (OLS) and Tobit estimation to a gravity model, he found that the EMU increased FDI both between EMU members and also between EMU members and non-members, thus supporting the notion that the exchange rate regime has an effect distinct from that of reducing volatility.

The general equilibrium model in Aizenman (1992) clearly distinguishes between exchange rate volatility and exchange rate regimes. In the model, two exchange rate regimes are available, fixed and flexible. The economy is subject to exogenous monetary (nominal) and productivity (real) shocks and the investors are assumed to be risk neutral. The result is that, for a given level of monetary or real volatility, investment (including FDI) is always higher under fixed exchange rates than under flexible. Monetary volatility under fixed rates has no effect on employment and real wages, whereas under flexible rates employment and real wages fluctuate. Expected profits and thus investment are higher under fixed rates. Meanwhile, productivity shocks lead to employment volatility under both regimes. The improvement in the marginal product of labor under a positive productivity shock is higher than its reduction under a negative productivity shock of equal magnitude. Thus, volatility encourages investment because the expected value of profits is higher with than without it. However, under flexible exchange rates, the beneficial effects on investment are reduced by partially offsetting exchange rate changes. Thus, again, higher investment occurs under fixed rates than under flexible, for a given level of productivity volatility.

These results imply possible correlation between exchange rate volatility and FDI under a flexible exchange rate regime because both respond to changes in the volatility of the exogenous shocks. Suppose first that monetary volatility increases. The extent to which volatility discourages FDI under flexible rates will grow along with exchange rate volatility. Thus, FDI and exchange rate volatility will be negatively correlated. Now suppose that productivity volatility grows. The extent to which this volatility encourages FDI under flexible rates will grow along with exchange rate volatility. FDI and exchange rate volatility will be positively correlated. In general, the volatility of both kinds of shocks will be fluctuating and so the overall correlation between exchange rate volatility and FDI will be uncertain. Consequently, from the point of view of the Aizenman (1992) model, the ambiguous empirical results are no surprise. In contrast, the relationship between the exchange rate regime and FDI is unambiguous.

Model and Econometric Approach

Our regression model is specified as follows:

(1)

where y_i,t is net FDI inflows³ expressed as a percentage of gross domestic product (GDP) of country i at time t, χ_i,t is a vector of explanatory variables, RD_i,t is a vector of exchange rate regime dummies, θ(L) and φ(L) are vectors of associated polynomials in the lag operator, q is the maximum lag length, γ_t are time specific effects, α_i are unobserved country specific effects, ε_i,t are error terms, and the β's, θ's and φ's are parameters to be estimated. Model (1) accounts for dynamic effects, as the set of right-hand side variables includes the lagged dependent variable. However, in both fixed and random effects settings, substantial complications arise from estimation of (1) because the lagged dependent variable will be correlated with the disturbance, even if it is assumed that ε_i,t is not itself autocorrelated. To address this problem, we employ instrumental variables within the SYS-GMM estimation procedure (Blundell and Bond, 1998).⁴ This econometric procedure suits our analysis well. FDI flows, factors influencing regime choice and variables typically entering the FDI equation may be simultaneously determined. With SYS-GMM every regressor is instrumented and including both level and first difference equations in a stacked system allows us to overcome potential issues of endogenity bias.

Exchange Rate Regimes Data

We draw from three different classifications to construct the regime dummies. Our first classification is that published by the International Monetary Fund (IMF) in its Annual Report on Exchange Rate Agreements and Restrictions (various issues). Since 1999 the IMF moved from a purely de jure classification scheme (one that records the exchange rate policy countries say they are pursuing) to a hybrid one that combines data on the actual behavior of the exchange rate since many countries deviate from their declared exchange rate policy framework (for recent evidence, see Gumus, 2011). However, concerns over the IMF classification have prompted researchers to develop alternative schemes that attempt to characterize more accurately countries' de facto regimes. Genberg and Swoboda (2005) examined the divergences between de jure and de facto classification schemes and concluded that empirical investigations can benefit from the simultaneous use of both types of classification. We, therefore, also use two such alternative de facto classification schemes. The first is the one developed by Levy-Yeyati and Sturzenegger (2005). They use cluster analysis techniques to group countries' regimes on the basis of the volatility of the exchange rate relative to the relevant anchor currency, the volatility of exchange rate changes, and the volatility of reserves. This classification scheme is now commonly employed in empirical work. The second de facto regime from which we draw on, is the one developed by Reinhart and Rogoff (2004) which incorporates data on market determined exchange rates, employs a rolling 5-year horizon to capture the “true” flexibility of the regime, and strips off the floating category of observations characterized by high inflation (which are floats by necessity) through the creation of a separate “freely falling” regime category.

Following the standard approach, we use a tripartite categorization to classify observations according to fixes, intermediates, and floats. According to the theories on FDI and exchange rate regimes in our literature review, we expect a positive effect of the fixed-regime dummy on FDI relative to floating. The models do not, however, explicitly indicate what to expect for the intermediate-regime dummy. Nevertheless, intuitively extending the Aizenman (1992) model suggests that dummy, too, will have a positive effect compared with floating.

Control Variables

Although here interest centers on establishing whether exchange rate regimes have any impact upon FDI flows to developing countries (over and above the effect of exchange rate volatility), we control for the most important variables that can be expected to exert a systematic influence. It is a standard hypothesis that economic growth promotes FDI inflows, with increases in income in the host-economy expected to generate an expanding market for the producer's goods and services. Consistent with the most popularly adopted measure (e.g. De Vita and Abbott, 2007) we use the rate of growth of GDP. As noted by Noorbakhsh et al. (2001), because labor typically costs less for a given level of productivity in developing than in developed countries, a measure of relative labor costs weighted by the productivity of labor vis-à-vis that of competitor host-developing countries might also prove an important determinant. Accordingly, the measure we employ is the relative efficiency wage (average wage divided by labor productivity) for country i formed from the difference between the efficiency wage of the country and the average efficiency wage of all developing countries in our sample. The availability of natural resources is another factor cited to explain MNCs' location decision, with a non-trivial amount of cross-border investment to developing countries still being of this type. Following Asiedu and Esfahani (2001), the measure we employ is the share of fuel and oil in total exports. FDI usually entails costs in coordinating activities with suppliers and distributors hence a well developed informational infrastructure may stimulate FDI flows (Mody et al., 2003). Following Asiedu (2002), the proxy we use is the number of telephone lines per 1000 population. We also control for macroeconomic stability by including the rate of inflation in our specification. Higher inflation implies increased macroeconomic uncertainty and therefore less FDI. Our selection of explanatory variables is further enriched by two country risk indices. The first is an index of government stability that is expected to better capture the host-country political risk connotations. This index measures the ability of a government to carry out its declared policy programme(s) as well as its ability to stay in office. The second country risk variable is a composite index of investment profile, which measures the host country's attitude to international investment drawing from several indicators, including risk to operations, taxation and repatriation of profits. Capital controls can be broadly classified as direct or indirect (Asiedu and Lien, 2004). Direct controls refer to administrative barriers in the form of limitations on foreign ownership, the stringency of screening requirements, and various restrictions on the use of funds. Indirect controls relate to various market-based and post-entry operational restrictions in the form of multiple exchange rate arrangements, taxation of cross-border flows, and other indirect regulatory controls. Since our exchange rate regime dummies and investment profile variable already account for the influence of the latter, our chosen measure of capital controls (from Chinn and Ito, 2008) focuses on the institutional restrictions that a country places on current account and capital account transactions. The contention that a well educated pool of labor can increase the attractiveness of a country's locational advantage has strong theoretical backing (Noorbakhsh et al., 2001). To capture also high technical and managerial skills, our human capital variable includes tertiary education. We also include the ratio of trade (imports plus exports) to GDP as an additional regressor. This is a standard measure of openness or trade restrictions in the literature, whose impact mostly depends on the type of investment. Given the likely presence of many types of FDI within our broad sample, the prevailing net effect remains an empirical question.

To isolate the impact of exchange rate regimes we additionally control for the independent effects of both the level and volatility of the exchange rate. To avoid arbitrariness in the selection of the volatility measure, we tested alternative specifications (e.g. GARCH) and found that the standard deviation provided the best fit to the data. The latter is specified as the standard deviation of the monthly percentage changes in the US dollar exchange rate.⁵

Results

Table 1 provides a first pass at the data, showing annual averages of net FDI flows as a percentage of GDP (over 1985–2004) and, in parentheses, the number of observations within each exchange rate regime. Observations are grouped according to the three classification schemes adopted. Inspection of Table 1 foretells the main results of the estimations that follow. De facto fixed and intermediate exchange rate arrangements considerably outperform the de facto floating regime in inducing FDI flows, though an altogether different picture emerges using the IMF classification.

Table 2 reports the results of estimating equation (1) for the various classification schemes. The regime dummies compare the average FDI inflows (as a percentage of GDP) for fixes and intermediates (whose proportions vary with the classification) with those for the regime category of floats. Although we lose many observations (owing to lack of FDI data and data relating to other variables for some countries and years) the proportion of usable data remains fairly stable across exchange rate regimes, with no particular bias of our unbalanced panel towards a specific regime. The model works well on several diagnostic dimensions. The Sargan test for instrument validity, the exclusion restriction test, and the second order serial correlation test do not reject the chosen econometric specification.

The main finding is that under both de facto classification schemes fixed and intermediate regimes are associated with significantly higher FDI inflows than the floating policy option. Specifically, the Reinhart and Rogoff (Levy-Yeyati and Sturzenegger) coefficients of the fixed regime dummies indicate that fixers attract on average 125.1% per year (68.2% per year) FDI flows more than floaters, from 1.66% (1.98%) of GDP under floating to 3.737% (3.331%) of GDP under fixed regimes, respectively. Similarly, the Reinhart and Rogoff (Levy-Yeyati and Sturzenegger) coefficients of the intermediates reveal that developing countries opting for intermediate regimes attract on average 143.9% per year (71% per year) FDI flows more than floaters.

These results also suggest that exchange rate regimes' impact on FDI flows does not merely reflect the degree of exchange rate volatility inherent in them, because we have controlled separately for exchange rate volatility. Accordingly, we might expect the fixed-regime dummy to be bigger than the intermediate-regime dummy by intuitive extension of the Aizenman (1992) model. Of course, the volatility measure is unlikely to be perfect, and so the regime dummies may still proxy for some of their effects, which are ambiguous. The regime effects may additionally reflect factors not in the Aizenman (1992) model, such as the role of underlying determinants of regime choice or simply the ability of developing countries to defend a fixed exchange rate. Indeed, countries with weaker fundamentals, such as insufficient foreign exchange reserves, may be unable to defend a fixed rate and be forced to float their currencies or adopt the intermediate option. Thus, a fixed rate may lack credibility. Although the sustainability of intermediate regimes may also be questioned, Williamson (2000) suggests that, especially for developing countries, intermediate regimes will continue to be a viable option to reap the benefits of both fixed and flexible rates without having to incur some of their costs.

The insignificance of fixed and intermediate regimes under the IMF classification may be rationalized simply in terms of the statistical distribution of the de jure floaters that de facto have a stable or fixed exchange rate. As can be seen in Table 1, particularly noticeable is the difference in the proportion of floating observations under the Reinhart and Rogoff classification vs the IMF scheme. Another explanation might be that, in addition to signaling the likely degree of exchange rate volatility, exchange rate regimes capture differences in institutional quality that, in turn, may influence the FDI location decision. Previous evidence supports this supposition. Alesina and Wagner (2006) show that countries that announce a fixed exchange rate but end up in the de facto floating category have relatively “bad” legal and policy institutions, whereas countries that fix de facto but float de jure have “good” institutions. The finding that developing countries' de facto flexible exchange rate systems are detrimental to FDI relative to other regimes is also consistent with previous evidence on the wider macroeconomic performance of the floating option in developing vs advanced economies. Husain et al. (2005) find that, for developing countries with little exposure to international markets, pegs are notable for their durability and relatively low inflation. In contrast, for advanced economies, floats are more durable and tend to be associated with higher growth.

Although our focus in this discussion is on the interpretation of the regime dummies, it is worth highlighting that some of the estimated coefficients of the control variables are consistently statistically significant across all three specifications, with sensible economic interpretations. These are trade openness, economic growth, and investment profile. Government stability and educational attainment are significant in one out of three specifications. The remaining control variables are not statistically significant. Within the latter category, particularly notable is the insignificance of relative wage costs, though this may be rationalized in terms of the type of investment, with possibly only a limited proportion of efficiency-seeking FDI within our broad panel. Of special interest is also the insignificance of the exchange rate volatility coefficient across all specifications. This may reflect its ambiguous sign in many theoretical discussions.

Extensions and Robustness Checks

We also investigate whether there may be a de facto“skew” that is inherently unfavorable to floating. Bleaney and Francisco (2007) argued that de facto classifications may be biased in their floating categories towards countries with inflationary or slow growth (e.g. Sub-Saharan Africa) and away from fast-growing ones (East and South Asia).

Accordingly, we re-estimated our model while disaggregating our sample by region. The results of this exercise (see Table 3) are generally consistent with those of Table 2. The regime coefficients using the IMF classifications are insignificant as before. The coefficients using the de facto classifications mostly show the same positive effects as in Table 2, although in many cases they are not significant. The lack of significance is not surprising given the loss of power from the smaller sample sizes of the disaggregated approach. Finally, a few negative coefficients appear, seeming to contradict the aggregated results. However, they are not statistically significant.

Although there is no a priori reason to expect exchange rate regimes to operate differently purely on the basis of an economy's income level, to test for sensitivity to sample selection, we also re-estimated the model by excluding low-income and lower-middle income countries from our sample. As can be seen from Table 4, the results are broadly consistent with those reported in Table 2. Some of the control variables lose while others gain statistical significance but the estimated coefficients of the regime dummies prove robust to this exercise.

Finally, to test further for robustness, we subjected the model to a number of perturbations. This process involved the inclusion of the log of real GDP, to check for omitted variable bias, of the money supply (M3), as a substitute for inflation in proxying macroeconomic stability, and of the squared volatility measure, to pick up any possible bias stemming from our model's approximation of the “true” model's non-linearities (estimations omitted for conciseness). None of these additional regressors proved significant, with no significant variations shown in the estimates of the regime dummies.