External Balance in Low Income Countries
Author: Lone Engbo Christiansen, Mr. Alessandro Prati, Mr. Luca A Ricci, and Mr. Thierry Tressel

Contributor Notes

Author’s E-Mail Address: lchristiansen@imf.org; lricci@imf.org; ttressel@imf.org
Publication Date:
01 Oct 2009
eISBN:
9781451873689
Language:
English
Keywords:
WP; current account balance; fiscal policy; terms of trade goods; emphasizing saving; income effect; trade balance
Download PDF (430.3 KB)

This paper offers a coherent empirical analysis of the determinants of the real exchange rate, the current account, and the net foreign assets position in low income countries. The paper focuses on indicators specific to low income countries, such as the quality of policies and institutions, the special access to official external financing, and the role of shocks. In addition to more standard factors, we find that domestic financial liberalization is associated with higher current account balances and net foreign asset positions, while capital account liberalization is associated with lower current account balances and net foreign asset positions and with more appreciated real exchange rates. Negative exogenous shocks tend to raise (reduce) the current account in countries with closed (opened) capital accounts. Finally, foreign aid is progressively absorbed over time through net imports, and is associated with a more depreciated real exchange rate in the long-run.

Abstract

This paper offers a coherent empirical analysis of the determinants of the real exchange rate, the current account, and the net foreign assets position in low income countries. The paper focuses on indicators specific to low income countries, such as the quality of policies and institutions, the special access to official external financing, and the role of shocks. In addition to more standard factors, we find that domestic financial liberalization is associated with higher current account balances and net foreign asset positions, while capital account liberalization is associated with lower current account balances and net foreign asset positions and with more appreciated real exchange rates. Negative exogenous shocks tend to raise (reduce) the current account in countries with closed (opened) capital accounts. Finally, foreign aid is progressively absorbed over time through net imports, and is associated with a more depreciated real exchange rate in the long-run.

I. Introduction

This paper investigates empirically the external balance of low income countries, by offering a coherent analysis of determinants of medium-to-long-term real exchange rates, current accounts, and net foreign assets, and highlighting factors that are more likely to be specific to these countries. The rise and persistence of large external imbalances in recent years have renewed interest in this area from an empirical and theoretical perspective, and have also highlighted the need for a multi-pronged approach to the analysis of external balances based on multiple indicators. In this paper, the simultaneous analysis of the three aforementioned indicators of external balance allows us to check the consistency of the results across indicators, an effort generally absent in the literature. The focus on low income countries aims at filling another gap. Although the literature on the determinants of the real exchange rate and of the current account is very vast, very few contributions focus specifically on low income countries, or account for features that are quite specific to—or more important for—this set of countries. Our analysis emphasizes factors such as structural policy and institutional distortions, access to special external financing, and a larger macroeconomic sensitivity to exogenous shocks. For the purpose of the empirical analysis, extensive efforts were required to create a wide database, which is unique in terms of the set of indicators and countries covered.

A large literature has based the analysis of medium term determinants of current accounts (CA) on the standard intertemporal approach to the current account emphasizing saving and investment decisions (see for instance Chinn and Prasad (2003) and Lee et al. (2008)).2 A more recent empirical literature has aimed at explaining the patterns of global imbalances that have widened over the past decade as function of financial crisis, financial development and distortions, and institutional variables (Gruber and Kamin (2007, 2008); Chinn and Ito (2007); and, from a theoretical perspective, Mendoza et al. (2008); Caballero et al. (2008) and Gourinchas and Jeanne (2009)). Others have illustrated the role of labor market policies and exchange rate regimes in influencing the persistence and dynamics of the current account (Ju and Wei (2007) and Chinn and Wei (2008)) and the relationship between labor market, financial frictions, and fiscal policies in shaping the optimal current account responses to shocks (Blanchard (2006)).

The literature on real exchange rates (RER) is vast and we cannot do justice to all contributions. Broad surveys are offered by Froot and Rogoff (1995), Rogoff (1996) and, for developing countries, by Edwards (1989), Hinkle and Montiel (1999), and Edwards and Savastano (2000). 3 The traditional findings of Meese and Rogoff (1983) on the unpredictability of exchange rates at short horizons are still undisputed, and the literature has converged towards explaining the behavior of real exchange rates at medium to long-term horizons as a function of fundamentals (see for example Engel and West (2005) and Engel et al. (2007)). Empirical analysis of long-run real exchange rates are typically guided by steady-state relationships in models involving the intertemporal and intratemporal allocation of resources between tradable and nontradable sectors (Obstfeld and Rogoff (1996), Vegh (2009), Montiel (2003), and Ricci et al. (2008)).

A growing literature has uncovered the medium-term determinants of gross and net foreign assets (NFA), after the creation of the Lane and Milesi-Ferretti database of external positions (for the latest version, see Lane and Milesi-Ferretti (2006)). Lane and Milesi-Ferretti (2001) offer a theoretical and empirical discussion of long-term determinants of the net foreign asset position. Faria et al. (2007) show that more open economies with better institutions have a greater equity share in external liabilities.

Few studies have focused on low-income countries (LICs) with the notable exceptions of Edwards (1989) and Hinkle and Montiel (1999).4 In this paper, we argue that LICs differ from other countries mainly along three broad dimensions, which simultaneously affect the current account, the real exchange rate, and the net external asset position: (i) structural policies or distortions, in particular those related to the capital account and the domestic financial system; (ii) exogenous shocks, in particular natural disasters (whose effect may depend on the degree of capital account openness) and terms of trade shocks; and (iii) official external financing (grants and concessional loans).

We believe that these factors are particularly important for our sample of countries. First, low income countries face greater distortions—some of which are policy-induced—than other countries. For example, capital account controls—which were prevalent for a large number of countries over the sample analyzed—may reduce the ability of LICs to borrow in order to bring consumption and investment forward, as required by a lower level of development or the occurrence of negative shocks. They may therefore affect domestic demand, the current account, the net foreign assets, and the real exchange rate.5 Domestic financial liberalization, which occurred during the 1980s and the 1990s in many developing countries, may reduce borrowing constraints and boost investment, which would tend to lower the current account and the net foreign assets position, and appreciate the real exchange rate. But financial liberalization may also raise private savings, which, everything else equal, would improve the current account and the net foreign assets position, and depreciate the real exchange rate.

Second, low-income countries are in general more exposed to shocks than other countries, and may—as a result of the lack of diversification of their production structure—experience larger macroeconomic consequences associated with these shocks. 6 For example, LICs are exposed to frequent terms of trade fluctuations associated both with their exports (for example, main crop or natural resources), and with their imports (for example, oil). Such terms of trade fluctuations affect the real exchange rate and the current account through income and intra- and inter-temporal substitution effects. Moreover LICs frequently experience natural shocks, such as droughts, floods, windstorms, and earthquakes, which have larger macroeconomic consequences than in high and middle income countries—including on the external position. Finally, wars and violent political transitions between regimes have often occurred in the historical sample. Such events, by disrupting investment, consumption, and capital flows, can have a bearing on the current account and the real exchange rate at a relatively short horizon.

Lastly, capital flows are typically of a different nature in low income countries than in other countries. A large part of their foreign borrowing is in the form of official development assistance (grants or concessional loans). Such capital flows do not respond to market incentives, and often do not need to be repaid, thus contributing to financing larger trade deficits over the medium-term. Finally, aid flows have often been associated with the risk of Dutch disease as resource transfers, and are expected to lead to more appreciated real exchange rates in the short-run by increasing aggregate demand (Van Wijnbergen (1984)). In the long run, however, the effect on the real exchange rate is uncertain, depending on the relative impact on the productivity of tradables versus the one of nontradables (Torvik (2001)).

To anticipate some of results for the indicators that are more important for low income countries, we find that domestic financial liberalization is associated with higher current account balances and net foreign asset positions, suggesting a positive effect on domestic savings. Capital account liberalization tends to be associated with lower current account and net foreign asset positions, and more appreciated real exchange rates, as predicted by standard theories. Negative exogenous shocks tend to raise (respectively reduce) the current account in countries with closed (respectively opened) capital accounts pointing at the importance of capital account frictions in shaping intertemporal consumption-smoothing decisions. Finally, foreign aid is progressively absorbed over time through net imports, and is associated with a more depreciated real exchange rate in the long-run, a result that may reflect larger productivity gains in the non-tradable sector relative to the tradable one.

The paper is organized as follows. Section II reviews the theoretical literature on the determinants of the external balances. Section III presents the results of the empirical analysis. Section IV concludes.

II. Determinants of External Balance

This section mainly reviews the determinants of the real effective exchange rate and of the current account, with particular emphasis on factors that are important fundamentals for LICs. Towards the end of the section we will discuss the more limited literature on the determinants of net foreign assets (which generally follows similar intuitions as the current account). The main emphasis will be on the theoretical arguments that can guide our empirical analysis, but we will also highlight the empirical contributions related to each conceptual argument, in order to ease comparison with our results. Potential determinants are grouped into 4 main groups: (i) main determinants already identified in the literature (macroeconomic policies, pre-determined characteristics, and stage of economic development); (ii) structural policies, distortions, and institutions; (iii) shocks, and (iv) external financing.

Economic theory underpins the relationship between the real exchange rate, the current account, and a number of macroeconomic variables. In principle, factors that affect the real exchange rate should also affect the current account: for example factors influencing aggregate demand will generally impact both the current account and the real effective exchange rate. However, theoretical foundations for the empirical analysis of the real exchange rate have usually been derived from long-run steady state analysis of models with tradable and nontradable goods in the presence of balanced trade. 7 At the same time, empirical analysis of the current account has been underpinned by the intertemporal approach to the current account, often in single goods models, hence without a meaningful exchange rate (Edwards (1989), Obstfeld and Rogoff (1996), Hinkle and Montiel (1999), and Vegh (2009)). In discussing the various determinants, we will highlight the possible joint effects.

Macroeconomic policies, pre-determined characteristics, and economic development

Fiscal policy

In absence of Ricardian equivalence, fiscal policy affects aggregate demand, national savings and therefore the current account balance and the real exchange rate. 8 Empirically, Chinn and Prasad (2003) and Lee et al. (2008) find that the fiscal balance is significantly and positively associated with the current account in pooled OLS regressions.9 Fiscal policy affects also the real exchange rate through a composition effect in a multigood economy even in the presence of Ricardian equivalence (Obstfeld and Rogoff, 1999). If government spending falls relatively more on non-traded goods than on private consumption (which is often the case for government consumption), it will lead to an appreciation of the real exchange rate, as the relative price of nontraded goods must increase in order to maintain internal and external balance (Vegh (2009) and Hinkle and Montiel (1999)). Consistent with this prediction, the empirical literature tends to find a positive coefficient (see for example Ricci et al. (2008) and De Gregorio et al. (1994)).

Net foreign assets.

Countries with initially higher net foreign assets can afford higher spending (above income flow)—and therefore a lower current account—while remaining solvent.10 However, in economies with uncertain horizon, non-zero steady state current accounts are positively associated with steady-state net foreign assets (see Lane and Milesi-Ferretti (2002), Chinn and Prasad (2003), and Lee et al. (2008), for consistent empirical evidence in pooled OLS regressions).11 Moreover, in steady-state, higher net foreign assets allow higher consumption of both tradable and non-tradable goods while remaining solvent, implying a more appreciated real exchange rate (Lane and Milesi-Ferretti (2002) and (2004) and Ricci et al. (2008)). This relationship may not hold in low-income countries experiencing debt relief: if an increase in debt is expected to benefit from debt relief in the future, lower net foreign assets resulting from the increase in debt may not be associated neither with lower consumption needed to service external liabilities through trade surplus, nor with changes in the real exchange rate. The effect of such expectations cannot be disentangled in the data and would also be reflected in the coefficient of net foreign assets in current account and real exchange rate regressions.

Demographics.

Under the life-cycle hypothesis, a higher share of inactive dependent population reduces national savings and the current account balance, and therefore results in a more appreciated real exchange rate. In an overlapping generation model, a higher share of working-age agents raises national savings, thus increaseing the current account. Population growth and fertility have a negative effect on the current account and a positive one on the real exchange rate if they are correlated with the share of young inactive in the population. These predictions are confirmed empirically in the analysis of the current account (see for example, Lee et al. (2008)), of the real exchange rate (see Rose et al. (2008)) and of net foreign assets (see Lane and Milesi-Ferretti (2001)).

Stage of development and economic growth.

Neoclassical theory predicts that countries at an early stage of development should import capital and borrow against their future income to finance their investment needs and smooth out consumption, given high marginal utility of consumption (Obstfeld and Rogoff (1999)). Similarly, fast growing countries with higher expected productivity gains should invest more, implying a deterioration of the current account.12 Finally, high productivity growth in the tradable sector relative to the non-tradable should be associated with a more appreciated real exchange rate (Balassa-Samuelson effect): an increase in productivity in the tradable relative to the nontradable sector, with respect to trading partners, will lead to higher wages in the tradable sector (whose price is given in world markets if the country is small) and subsequently put upward pressures on wages and prices in the nontraded sector. Ricci et al. (2008) and Choudhri and Khan (2005) find that a 10 percent increase in the productivity of tradables relative to non-tradable tends to appreciate the real exchange rate by about 1 to 2 percent on average. Moreover, higher income will result in an upward pressure on prices of nontraded goods relative to traded goods, as traded goods are priced on the international market, leading to a real exchange rate appreciation.13 However, a good measure of relative productivity is not easily available in low-income countries. Therefore, this paper uses real GDP per capita as a proxy variable, as in most of the literature. As this variable may not accurately capture the relative productivity effects—on the contrary, it averages out productivity in tradables and nontradables—the expected sign on this proxy is not clear.

Policy distortions and institutions

Domestic financial reforms.

A more developed financial system facilitates investment and helps attract foreign capital, thereby lowering the current account balance and appreciating the real effective exchange rate.14 A more developed financial system may also improve the current account balance and depreciate the real exchange rate if it stimulates domestic savings (Mac Kinnon (1973) and Edwards (1995)).15 Gourinchas and Jeanne (2009) model an open economy in which both investment and saving decisions are distorted by “wedges” affecting the return to capital. Their model predicts that financial liberalization can have ambiguous effects on the external position of a developing country: a reduction of the saving distortion tends to reduce capital inflows by increasing domestic savings, but a reduction of the investment distortion tends to increase capital inflows by raising capital scarcity.16

Empirical analysis has usually relied on measures of financial development as a proxy for the degree of financial liberalization, and has found at best weak effects on the current account (Gurber and Kamin (2007) and Chinn and Ito (2007)).

Capital account openness.

Neoclassical theory predicts that, over the development process, capital account liberalization should be associated with a deterioration of the current account (capital inflows) and a real exchange rate appreciation in developing countries, and with an improvement of the current account (capital outflows) and a real exchange rate depreciation in advanced countries (Lucas (1988) and Edwards (1989)).17 Moreover, a more open capital account allows countries to borrow against future income and therefore to run a lower current account balance when hit by a temporary negative income shock (Vegh (2008)). However, Kraay and Ventura (2000) show that, if the marginal unit of wealth is invested in the same way as the average unit of wealth, transitory positive income shocks will lead to a current account deficit (surplus) in countries with negative (positive) net foreign assets.

Institutions.

Broad institutional characteristics such as the quality of property rights and contract enforcement can have first order effects on the current account balance and capital flows. Countries with better institutions may be more able to attract a steady flow of foreign capital as a result of lower expropriation risks, and therefore can sustain lower current account balances and net foreign asset position (Alfaro et al. (2007) and Gruber and Kamin (2007)). However, in countries with better institutions, the political process may produce exchange rate policies less likely to favor overvalued real exchange rates, and therefore result in higher current account and net foreign asset positions. The same outcome may arise from better institutions generating an environment more conducive to saving.

Trade reforms.

The effect of trade reforms on the current account and the real exchange rate is theoretically ambiguous (Edwards (1990)). Trade reforms that are temporary (or perceived as temporary) may worsen the current account by reducing the price of imported goods relative to domestically produced goods (intra-temporal substitution effect). The inter-temporal effect is ambiguous: the current account should improve as a result of the income effect, but a lower price for today’s consumption relative to future consumption negatively impacts the current account balance (Ostry, 1990).18 The effect of trade liberalization on the real exchange rate depends on whether income or substitution effects dominate. As trade is liberalized, the increase in real income resulting from lower import prices tends to appreciate the real exchange rate.19 However, trade liberalization may also shift demand away from nontraded to traded goods, resulting in a depreciation. A similar result would arise from the direct effect of liberalization (say tariff reduction) on the reduction of the domestic price of tradables. The evidence is generally in favor of the latter effect (see Ricci et al. (2008) and Goldfajn and Valdes (1999)).

Price controls and black market premium.

Administered prices keep prices below the market level and are therefore associated with a more depreciated real exchange rate (see evidence in Ricci et al. (2008)). However, price controls may also take the form of a marketing board pushing domestic prices up, which would therefore be associated with a more appreciated real exchange rate. Finally, a black market rate that is more depreciated than the official exchange rate (i.e. a positive black market premium indicating expectations of a devaluation of the official rate) is likely to be associated with a more appreciated real exchange rate (based on the official rate, which is the prevalent basis for measuring real exchange rates) for given fundamentals.

Shocks

Terms of trade.

The effect of an improvement in the terms of trade is uncertain and mainly depends on whether the substitution or the income effect dominate. An improvement in the terms of trade arising from an increase in the price of exports raises the current account, as part of the positive income shock is saved to smooth out consumption over time, and appreciates the CPI-based real exchange rate as the increase in domestic demand (associated with the income effect) bids nontraded goods prices up (Ostry (1988) and Edwards and Ostry (1992)). However, an improvement in the terms of trade arising from a decrease in the price of imports may also result in a worsening of the current account and a real depreciation if agents substitute imported goods for domestically produced goods or for future imported goods (Obstfeld and Rogoff (1999) and Vegh (2009)). If imports are used as intermediate inputs in production, the last effect on the real exchange rate would be compounded, as domestic goods produced with those imports would tend to experience a decline in price. Overall, evidence shows a positive effect on the real exchange rate from a terms of trade improvement (see Ricci et al. (2008), Cashin, Cespedes, and Sahay (2004), Chen and Rogoff (2003), and Ostry and Reinhart (1992); for an analysis of the separate effect of import and export prices, see Christiansen and Tokarick (2009)).

Natural disasters.

A negative income shock positively affects the current account balance if national savings increase, or if investment falls relative to savings, as a consequence of the shock. However, the current account could worsen if the country can smooth consumption out by borrowing on international financial markets (Obstfeld and Rogoff, 1999). Thus, we expect the effect of natural disasters to depend on the degree of capital account openness.20

When considering the long-run relationship between the real exchange rate and its fundamentals, we may expect shocks not to play any role, as it is likely that they have only temporary effects.

External financing

Official Aid

In the short-run, a surge in aid could push up domestic prices and induce a real exchange rate appreciation as supply has a limited ability to respond to an increase in aggregate demand financed by aid. In the long-run, however, the effect of aid on the real exchange rate is theoretically ambiguous, as aid flows may cause an increase (decrease) in the productivity of the nontraded goods sector relative to the productivity of the traded sector, hence leading to a real exchange rate depreciation (appreciation).21

To empirically estimate the impact of aid on the current account, aid must be broken down into its two main components (grants and concessional loans), given that they are accounted for in different parts of the balance of payment (the former enters the current account and the latter the financial account). Also, to the extent that aid flows are usually redistributed to private agents through government expenditures and are not intermediated by the domestic financial system, their effect on the external position is independent of the impact of domestic financial liberalization. Conceptually, and as a first order approximation, aid flows can be modeled as exogenous transfers.

Grants.

Countries receiving a steady flow of grants are able to sustain a lower trade balance in the medium-term. Given that grants are accounted for in the current account section of the balance of payment, the current account should remain unchanged if a grant fully finances a deterioration of the trade balance. If, on contrary, part of the grant is saved in the form of international reserves, the current account will improve.22

Concessional loans.

Concessional loans allow to finance a lower current account in the medium-term. Moreover, debt on concessional terms poses a measurement issue as it creates a gap between the nominal and the present market value of net foreign assets.23 This paper examines the effect of net foreign assets when accounting for the present value of public and publicly guaranteed debt. 24 This valuable correction has of course limitations as, given data availability, the additional effect from expected future debt relief cannot be not accounted for.

The literature on the net foreign assets is more limited, in part because the empirical analysis was impaired by the lack of good data until recently (see Lane and Milesi-Ferretti (2006) for a second edition of their dataset). Lane and Milesi-Ferretti (2001) offer a theoretical and empirical discussion of the main determinants for advanced economies and developing countries. First, public debt tends to reduce net foreign assets, similarly to the effect of budget deficits on the current account. Second, a higher share of dependent population implies the need to run down savings (and thus reduce net foreign asset), in order to consume more. Third, the relation between income and net foreign assets is more uncertain. A positive relation is suggested by the standard development model where poor countries borrow and rich countries lend, but can also be derived in models with habit formation, or nonlinearities in the utility function. However, a negative relationship with income may arise as a result of limited access to international markets or a high desire for precautionary saving in developing countries. Finally, richer countries tend to invest more in equity (see Faria et al. (2007)), which is more likely to offer a higher long term return and result in a higher net foreign assets.

III. Empirical Results

A. Data

The exercise required an extensive data gathering and cleanup exercise. We constructed a dataset containing 134 countries over the period 1980-2006 for various indicators. Countries used in the main analysis were classified based on their income group. Our LICs sample (see Appendix Table A1) comprises “low income” or “lower middle income” according to the World Bank classification, and excludes emerging markets to make the sample as homogenous as possible (China, Colombia, India, Indonesia, Pakistan and Thailand). “High income” and “higher middle income” countries (in the World Bank classification), including the 6 countries above, were mainly used as a comparator group of high income countries (HICs). A summary statistic of the main data is provided in Appendix Table A2. A description of all variables is provided in the Appendix. The number of low income countries entering the regressions varies across specifications based on data availability for the specific indicators, but the largest low income country set (used in the trading partner calculations and in regressions with standard fundamentals) includes 59 low- and lower middle-income countries.

B. Analysis of Medium-Term Current Accounts

This section analyzes the medium-term relationship between the current account and a set of fundamentals. Following the existing literature, the estimations consist of an unbalanced panel of non-overlapping four-year averages over the period 1981-2005 with 6 observations for most countries.

Benchmark CA regressions for low-income countries25

Our preferred current account regressions for our sample of LICs are reported in Table 1. In addition to country fixed effects, the regressions include traditional variables (see for example Chinn and Prasad (2003), Chinn and Ito (2007), and Lee et al. (2008)) such as the fiscal balance, demographic variables (the old age dependency ratio, population growth), the initial net foreign asset position, the oil trade balance, and variables related to the stage of development (GDP per capita relative to the US, and real per capita GDP growth). Most notably, the ratio of the fiscal balance to GDP in relation to trading partners, and population growth remain strongly significant in our sample of LICs with the expected sign.

Table 1.

Medium-term determinants of the current account: main results.

article image
*** p<0.01, ** p<0.05, * p<0.1 (Robust p-values in parentheses)Robust panel regressions of the current account to GDP ratio on country fixed effects, a set of control variables (the fiscal balance to GDP ratio, demographic variables, the initial net foreign asset to GDP, the oil trade balance to GDP ratio, income per capita relative to the US, per capita real GDP growth), and a set of medium-term deteminants of the current account including: (i) aid flows; (ii) financial and capital account liberalization; and (iii) shocks (natural disasters, terms of trade fluctuations). The dependent variable and explanatory variables are averaged over 4 year periods (except when stated otherwise) and covering 1981-2005.

Our set of control variables also include a measure of the terms of trade which has traditionally not been included in current account regressions. Our findings are consistent with the interpretation that income effects resulting from terms of trade fluctuations have a strong impact on the current account: a temporary improvement in the terms of trade tends to improve the current account as part of the increase in income is saved to smooth consumption over time.

Columns 1 to 7 report regressions in which we introduce—one by one or in groups—our new medium-term determinants of the current account, in addition to the previous set of medium-term determinants of the current account. Column 8 and 9 summarize our preferred results, either when maintaining standard determinants, or when using a parsimonious specification with only significant variables.

Regarding official financing, an aggregate measure of foreign aid is significantly and negatively correlated with the current account.26 The estimated coefficient implies that, for any 1$ of aid, the current account deteriorates by about 0.1$ during the same period. Consistent with theoretical predictions of normative models (see for instance Torvik and Matsen (2004)), on average, a large share of aid is not immediately spent on net imports, and is indirectly saved as foreign exchange reserves (Berg et al. (2007) and Prati and Tressel (2006) provide consistent evidence that macroeconomic responses to aid surges may limit the short-run absorption of aid in LICs).

As the two components of foreign aid (grants and concessional loans) are accounted for in separate parts of the balance of payments (the current account and the financial account, respectively), we present regressions separating these components in column 2, 6, 8, and 9. The estimated coefficients imply that, on average, about 80 percent of grants are fully absorbed through an increase in net imports over a 4 years period, while only 30 to 50 percent of concessional loans received are absorbed through net imports over the same horizon (see Prati and Tressel (2006) for a model of macroeconomic policy responses to aid flows).27

Domestic financial reforms are associated with higher current accounts, suggesting that these reforms boost domestic private savings more than investment. This interpretation is consistent with the Mc Kinnon hypothesis (initially formulated by Mc Kinnon (1973) and also conceptualized by Fry (1995)) according to which a liberalization of domestic banking systems in developing countries results in higher saving rates by raising the return on financial savings.28 The estimated effect implies that shifting from complete financial repression to the sample average level of liberalization would lift the current account by 1.5 percent of GDP. Evidence consistent with this interpretation is shown in Appendix Table A3, which suggests that real deposit rates increase when countries liberalize their domestic financial systems.29

The opposite effect on the current account of concessional loans and of domestic financial liberalization may be explained by the fact that aid flows are typically not intermediated by the domestic financial system, as discussed in the theory section (aid flows typically finance government current expenditures and public investment). In addition, part of aid flows, by being redistributed through transfers, temporarily increase private agents disposable income, thus raising private consumption and savings.30

Exogenous negative income shocks (as proxied by natural disasters) have a negative impact on the current account, as predicted by standard theory, but only in countries with a fully open capital account. 31 In these countries, a temporary negative income shock is associated with a deterioration of the current account and with capital inflows, which is consistent with smoothing of aggregate consumption through international borrowing.32 Since we are already controlling for official capital flows, our estimate must be capturing the role of private capital flows.

On the contrary, natural disasters are associated with current account improvements in countries with closed capital accounts. A possible explanation is that these negative shocks trigger a procyclical increase in saving (relative to investment) when there is no access to international capital markets. An alternative explanation is suggested by the work of Kraay and Ventura (2000) who show that, under certain conditions, the marginal unit of savings following temporary income shocks is invested as the average one, suggesting that a positive change in savings should lead to current account surpluses in countries for which the real return on foreign assets is higher than the return on domestic assets, but would lead to deficits in countries with lower return. This implies that negative income shocks should lead to current account surpluses (deficits) in debtor (creditor) countries because domestic investment falls more (less) than savings following such shocks.33 When interacting the natural disaster variable with the initial net foreign asset position, we obtain a negative and significant coefficient on the interaction term, which tends indeed to support the prediction of the Kraay and Ventura model, and a non-significant coefficient on the shock variable itself (column 5). Controlling for aid, however, weakens this estimated effect of the portfolio composition (column 6).

Another interpretation of this evidence is that the impact of capital account liberalization on the current account depends on income fluctuations: as discussed, when countries face negative income shocks, a more open capital account allows to borrow on international capital market, which is consistent with consumption smoothing. However, in good times, capital account liberalization seems to be associated with capital outflows in low income countries, which is consistent with the presence of financial repression during the sample (Appendix Table A3 shows that 6 out of 11 developing countries covered exhibited real deposit rates below US deposit rates in the 1980s).34

Turning to institutional factors, we find that violent political events tend to improve the current account, suggesting that political unrest may trigger significant capital flight.

Robustness

Our main baseline results are generally robust to the inclusion of various additional explanatory variables as shown in Table 2. Column 1 allows for a dynamic effect of the current account by controlling for the lagged current account.35 In our sample of LICs, we find limited persistence of the current account beyond a 4 year horizon. Column 2 shows that trade reforms (as measured by an index of average tariffs) have no additional explanatory power. In column 3, we split the terms of trade index in his two components, and show that export prices are strongly positively correlated with the current account, while import prices are weakly and negatively correlated with the current account: a weaker correlation for the latter is consistent with income effects and substitution effects of import price movements having opposite effects on the current account. Next, as shown in column 4, the results are robust to the inclusion of emerging markets classified as lower middle income countries to the sample (these countries are China, Colombia, India, Indonesia, Pakistan and Thailand). Furthermore, our results are robust to the use of a broader measure of capital account liberalization (updated version of Quinn (1997)), as reported in column 5. In columns 6, 7, and 8, we consider alternative interaction terms with the index of capital account openness. In column 6, we control for an interaction between the fiscal balance and the capital account index, to test whether the effect of fiscal policy may depend on the capacity of private agents to offset fiscal policies by borrowing or lending in international markets. We find that, in our sample of LICs, the effect of fiscal policy on aggregate demand and on the current account is not significantly affected by the degree of capital account openness. However, the negative sign is consistent with standard theories, as it would imply a lower effect of the fiscal balance on the current account in countries with more open capital account. In column 7, we consider an interaction with population growth, but again do not find any significant effect. Still, the negative sign on the interaction term is consistent with neoclassical theories according to which countries with a younger non-active population are more able to borrow against future income when the capital account is open. The last column considers an interaction term with real per capita GDP growth; the positive sign of the interaction term, although not significant, would suggest that countries with more open capital accounts tend to run higher current account balances when growth is temporarily above its trend.36

Table 2:

Medium-term determinants of the current account: robustness.

article image
*** p<0.01, ** p<0.05, * p<0.1 (Robust p-values in parentheses)Robust panel regressions of the current account to GDP ratio on country fixed effects, a set of control variables (the fiscal balance to GDP ratio, demographic variables), a set of medium-term deteminants of the current account identified in Table 1. The dependent variable and explanatory variables are averaged over 4 year periods (except when stated otherwise) and covering 1981-2005. Additional control variables are added one at a time: a 4 year lag of the medium-term current account, an index of average import tariff, price indices for exports and imports, an alternative broader measure of capital account liberalization (based on an extension of Quinn, 1997), and interaction terms between the capital account liberalization index and (i) the fiscal balance; (ii) population growth; and (iii) per capital real GDP growth. Column (4) considers a broader country sample including emerging markets that are classified as lower middle income countries (China, Colombia, India, Indonesia, Pakistan and Thailand)

Are low income countries different?

We expect low income countries to differ from high income countries because of their higher exposure to shocks, greater distortions, and different sources of external financing. Tables 3a and 3b explores the extent to which various medium-term determinants of the current account have significantly different effects on the current account of low income countries on the one hand, and of high income countries on the other hand.

Table 3a.

Current account regressions with different slopes for LICs and HICs.

article image
*** p<0.01, ** p<0.05, * p<0.1 (Robust p-values in parentheses)

Robust panel regression of the current account to GDP on a set of medium-term determinants of Table1, column (8). Independent variables are interacted with dummy variables for our sample of LICs and HICs. Sample and period averages are as in Tables 1 and 2.

Table 3b.

F-tests of equality (p-values) of coefficients across income groups for regressions reported in Table 3a.

article image
Note: Ho: coeff HIC = coef LIC (p value of F test is reported).Regressions are those reported in Table 3a

Among standard determinants of the current account, we find that (i) the fiscal balance; (ii) the old age dependency ratio, and (iii) initial net foreign assets have significantly different impacts on the current account in low-income countries relative to other countries. In particular, we find no significant association between the current account and the fiscal balance in high income countries, a finding consistent with Chinn and Prasad (2003) and Isard and Faruqee (1999) who did not find any significant association in industrialized countries. This result is consistent with the hypothesis that Ricardian equivalence is more likely to be a reasonable first order approximation of consumption and saving decisions in advanced countries than in other countries. The negative sign of the coefficient for the initial net foreign asset position in advanced countries is consistent with the standard intertemporal approach to the current account that predicts that initially richer countries can sustain lower current account balances in the medium-term.

Surprisingly, we seem to find a negative association between the current account and capital account openness for more developed countries when they do not experience natural disasters (column 1)—a finding seemingly inconsistent with the standard prediction of the neoclassical theory, as richer countries should experience capital outflows when opening up their capital account. Moreover, for this group of countries, we obtain a positive coefficient on the interaction term between the capital account openness variable and the natural disaster variable. It is likely that this result is due to the fact that the specification in column 1 was ideal for LICs, but not for high income countries; two possible explanations relate to the way income and capital account liberalization interact, and the role of shocks. In column 2, we introduce an interaction term between the capital account openness variable and GDP per capita as a way to test more directly the prediction of neoclassical theory that capital should flow from more developed to less developed countries. We find a positive coefficient of the interaction term, which is consistent with the prediction of this theory, and the total effect of capital account liberalization on the current account is indeed positive for the richer countries.37 It is also possible is that the natural disasters are not relevant from a macroeconomic perspective for high income countries (see the non-significant coefficient on this variable for the higher income group). In regression 3, we drop the natural disaster variable and the interaction term, and show that the interaction of the capital account liberalization index with GDP per capita is consistent with the prediction of the standard development theory.

C. Empirical Analysis of the Real Exchange Rate

This section investigates the long-run relationship between the real effective exchange rate and a set of fundamentals. The estimation relies on an unbalanced panel of annual data covering 1980-2006. Panel unit root tests show the unit root nature of the variables involved in the estimation, apart from the natural shocks (see Table 4). Panel cointegration test have been performed for the benchmark regressions of interest (columns 3 in Tables 5 and 6) and reject the null of no cointegration.38 Under the assumption of I(1) cointegrated variables, dynamic ordinary least squares (DOLS) with fixed effect provide—from the coefficients of the variables in levels—an estimate of a long-run cointegrating relationship between the real exchange rate and a set of fundamentals. As part of the DOLS specification, in addition to the variables in levels, we enter changes in right hand side variables and—given the short length of the sample—one lead and one lag of these changes.

Table 4.

Panel Unit Root test statistics.

article image

Test based on Pesaran (2007).

Samples A and B refer to restricting the sample to having at least 14 or 20 observations per country, respectively.

means reject null HO of unit root at 5% one-sided

indicates that the variable is constructed relative to the weighted average of the trading partners

Table 5.

Real effective exchange rate (IMF definition) regressions.

Panel dynamic OLS with fixed effects (only long-run coefficients reported).

article image
*** p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the real effective exchange rate (INS source) on net foreign asset (with NPV value of external debt) to trade ratio, log of relative productivity, terms of trade if goods and services, government consumption to GDP, aid to GDP (or its components: concessional loans and grants), capital account liberalization, trade restrictions, agricultural price reforms, fertility, natural disaster, and black market premium. Unbalanced panel with annual data 1980-2006.
Table 6.

Real effective exchange rate (Penn World Table) regressions.

Panel dynamic OLS with fixed effects (only long-run coefficients reported).

article image
*** p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the real effective exchange rate (Penn World Table source) on net foreign asset (with NPV value of external debt) to trade ratio, log of relative productivity, terms of trade if goods and services, government consumption to GDP, aid to GDP (or its components: concessional loans and grants), capital account liberalization, trade restrictions, agricultural price reforms, fertility, natural disaster, and black market premium. Unbalanced panel with annual data 1980-2006.

Benchmark REER regressions for low income countries39

Tables 5 and 6 report the preferred specifications for LICs with two different measures of real effective exchange rate. The first is the CPI-based one offered by the IMF-INS source. The second one is constructed from the price relative with respect to the U.S. as reported in the Penn World Table, then turned into a real effective exchange rate by applying the same weights employed in the calculation of the IMF-INS real effective exchange rate. Results are virtually identical, with few exceptions discussed below. The regression specifications include traditional variables such as the net foreign assets, productivity, government consumption, terms of trade, and trade restriction, but also LIC-specific variables such as aid flows and capital account liberalization. Moreover, demographics and price controls are likely to be very relevant for low income countries, even though these factors have been shown to matter also in other samples (see Rose et al. (2008) and Ricci et al. (2008), respectively). The first column of both tables includes a dummy for natural shocks defined as before, given that low income countries are widely affected by such natural occurrences. The results would hint to a negative effect of these shocks on the real exchange rate. However, given that the econometric nature of these [0,1] indicators is uncertain (a panel unit root test rejects the unit root hypothesis), in column 2 (and subsequent regressions) we exclude this indicator.

Columns 3 drops insignificant variables and derives a benchmark regression. Consistently with the previous literature, government consumption is associated with a real exchange rate appreciation, which is usually the case under the presumption that government spending is spent on non-tradables in a higher proportion than private spending. An improvement in the terms of trade appreciates the real exchange rate with effects similar to those found in other sample of countries (see Ricci et al. (2008)). Fertility is associated with an appreciation of the real exchange rate, as in Rose et al. (2008). The net foreign assets position is not significant, as possible expectations of debt relief may blur the intertemporal role of this variable in our sample of LICs. Overall productivity is not significant with the IMF-INS real exchange rate and is negative with the Penn World Table one (note that for dataset consistency, the productivity indicator is based on GDP per worker from the Penn World Table dataset in the second case). One possible explanation is that, in low income countries, measures of overall productivity equally reflect both tradable and non tradable sectors’ productivities, or even more the nontradable sector productivity than the productivity of the tradable one (the opposite assumption stands behind the standard presumption for using aggregate productivity as a proxy for the Balassa-Samuelson effect). Unfortunately, it was not possible to construct a better proxy for the Balassa-Samuelson effect despite extensive efforts.

Turning to variables specific to low income countries, aid inflows (from Roodman (2006)) are in the long run associated with a more depreciated exchange rate, potentially indicating a positive effect on productivity in the nontradable sector relative to the tradable sector. Aid is generally considered to push up domestic prices (especially of nontradables), thus leading to a real exchange rate appreciation (Dutch disease) in the short-run, i.e. when the supply side of the economy has not had a chance to adjust. In the long run, however, an increase in aid would be consistent with a real exchange rate depreciation if it would raise productivity of nontradables relative to the productivity of tradables (Torvik, 2001).40

Capital account liberalization is associated with an appreciation of the real exchange rate, suggesting that, in the long run, such liberalization promotes persistent net capital inflows. Price distortions are also somewhat significant. In particular, the presence of marketing boards (as captured by the indicator “max agric. price intervention”) is likely to keep prices high and thus appreciate the real exchange rate.

The last column of Tables 5 and 6 includes the black market premium, which unfortunately halves the sample size. As real exchange rates are normally measured at official rates, the positive and significant coefficient is consistent with the standard interpretation that the presence of a black market premium usually signals an overvalued official exchange rate. Generally, in these circumstances, most public transactions are done at the official rate, while private transactions tend to occur at the black market rate, so the actual average exchange rate is likely to be between the official and the black market rates. This would correspond to a coefficient between zero and one, which is indeed what we find. However, the sample size decreases substantially, which limits the usefulness of the regressions. Measuring correctly the exchange rate is an important issue which should deserve wide attention in real exchange rate analysis—especially when focusing on low income countries which have traditionally been more prone to dual exchange rate systems and problems of measurements of price levels—and would require additional efforts in data collection.

Are low income countries different?

Low income countries differ from high income ones mainly because of the specific factors (distortions, financing, and shocks), but traditional factors do not show great difference when specific factors are included in the regression. However, neglecting the presence of the specific factors would lead to misspecifications, and even coefficients on traditional factors would appear different. Columns 1 and 2 of Table 7a present a specification typically used for high income and higher middle income countries (see for example Ricci et al., 2008), but estimated with separate coefficients for low- and lower middle-income countries (LIC) and for high- and upper middle-income countries (HICs). All coefficient appears to be significantly different (see Table 7b column 1). The next column in Table 7a show a regression equivalent to the one in Table 5 column 2—i.e. the benchmark before dropping variables that have been found relevant for a broader sample of advanced economies—but again with different slopes for LICs and HICs.

Table 7a.

Real effective exchange rate (IMF definition) regressions with different slopes for LICs and HICs.

Panel dynamic OLS with Fixed Effects (only long-run coefficients reported).

article image
*** p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the real effective exchange rate (INS source) on same determinants as previous two tables. Independent variables are interacted with dummy variables for LICs and HICs. First two columns present standard determinants only. Second two columns encompass determinants in columns 2 of previous two tables. Last two columns present determinants in columns 3 of previous two tables. Unbalanced panel with annual data 1980-2006.
Table 7b.

F-tests of equality (p values) of coefficients across income groups for regressions reported in in Table 7a.

article image
note: p-values for the Ho: coeff HIC = coeff LIC

As this regressions encompasses indicators that are relevant for both set of countries, it may be the best one to assess the different roles of these indicators across sample of countries. The difference in the coefficients of the traditional variables is now insignificant in most indicators (Table 7b column 2). Net foreign assets appear to play a different role in the two set of countries, which is not surprising given that a possible expectation of debt relief may reduce the relevance of this variable in LICs. The key LIC factors (apart from fertility) seem to play different a different role in the two samples, which is again not surprising given that these indicators are likely to be less relevant in high income countries. The last column of Table 7a shows the income split for a regression equivalent to the baseline in Table 5 column 3, and Table 7c present the corresponding test of equality of coefficient; results are now somewhat different, but it may be due to the fact that such regression is tailored to low income and is mispecified for the other group.

Robustness

The benchmark model for LICs is generally robust to alternative specifications. Tables 8 and 9 repeat in column 1 the benchmark derived in column 3 of Tables 5 and 6 and then explore the robustness of alternative indicators. In particular, columns 2 and 3 allow for the terms of trade (respectively in goods only or goods and services) to be split in the two components (price of exports and imports), and show that the effect is mainly due to the price of exports. This is something we would expect, and is consistent with the results for the current account regressions (Table 2): an improvement in the terms of trade from a decline in import prices may generate not only a positive income effect (increasing demand for domestic goods), but also an additional substitution effect away from domestic goods, thus with offsetting effects on the real exchange rate (see Christiansen and Tokarick (2008) for a broad theoretical and empirical analysis of the effect of the components of the terms of trade). Results for the other variables are quite consistent with the previous regressions.

Table 8.

Real effective exchange rate regressions (IMF def.) regressions, robustness.

Panel dynamic OLS with Fixed Effects (only long-run coefficients reported).

article image
*** p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the real effective exchange rate (INS source) on same determinants as previous two tables. Additional control encompass: splitting terms of trade in price of exports and of imports (either of goods only or of goods&services), and alternative measures of capital account liberalization, of demographics, and of productivity. Unbalanced panel with annual data 1980-2006.
TABLE 9:

Real effective exchange rate (Penn World Tables) regressions robustness.

Panel dynamic OLS with Fixed Effects (only long-run coefficients reported).

article image

*** p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the real effective exchange rate (Penn World Table source) on same determinants as previous two tables. Additional control encompass: splitting terms of trade in price of exports and of imports (either of goods only or of goods&services), and alternative measures of capital account liberalization, of demographics, and of productivity. Unbalanced panel with annual data 1980-2006.

Speed of adjustment

In order to assess the speed at which the real exchange rate adjusts towards its long run cointegrating relationship, we impose the estimated cointegrating relationship in an error-correction specification. We construct the error correction term via the difference of the real exchange rate from the sum of the products of the fundamentals entering the baseline regression in Table 5 column 3 times the corresponding level coefficients. We then run changes of the real exchange rate on the lag of the error correction term as well as on lagged changes of the real exchange rate and of the other right hand side variables entering the baseline. In the four specifications derived from entering progressively from 1 up to 4 lags, the robust OLS coefficient of the lagged error correction term hovered around 0.2, suggesting that a shock to the gap would have a half life of about three years. We replicated the exercise with the alternative measure of real exchange rate (hence using Table 6 column 3) and obtained a somewhat higher speed of adjustment, in the order of 0.3, indicative of a half life of about 2 years. These results are in line with the previous literature (see Rogoff (1996) for example).

D. Empirical Analysis of the Net Foreign Asset Position

This section investigates empirically the net foreign asset position of LICs. The estimation relies on an unbalanced panel of annual data covering 1980-2006. Again, panel unit root tests confirm the unit root nature of the variables involved in the estimation (see Table 4); and panel cointegration tests performed for the benchmark regressions (Table 10 column 1) reject the null hypothesis of no cointegration (see discussion of these tests in the previous section). Similarly to the RER regressions, the panel cointegration estimation is based on dynamic ordinary least squares with fixed effects. In addition to determinants identified in the literature (public debt, demographics, and income per capita), we analyze the role of policy distortions (capital account and domestic financial liberalization) and of the quality of institutions.

Table 10.

Net foreign asset regressions.

Panel dynamic OLS with Fixed Effects (only long-run coefficients reported).

article image
***p<0.01, ** p<0.05, * p<0.1 (robust p-values in parentheses)

indicates that the variable is constructed relative to the weighted average of the trading partners

Robust dynamic OLS panel regressions with fixed effects of the Net Foreign Asset to trade ratio (or NFA to GDP in column 2, or NFA adjusted for the net present value of external debt in column 3) on public debt to trade ratio, fertility, log of relative productivity, quality of institutions (constraint on executive), financial liberalization, and capital account liberalization. Additional control encompass: alternative measures of debt, demographics, capital account liberalization, relative income or productivity, terms of trade, trade restrictions, and agricultural price reforms. Unbalanced panel with annual data 1980-2006.