Official Financial Flows, Capital Mobility, and Global Imbalances
Author:
Mr. Tamim Bayoumi
Search for other papers by Mr. Tamim Bayoumi in
Current site
Google Scholar
Close
,
Mr. Joseph E. Gagnon https://isni.org/isni/0000000404811396 International Monetary Fund

Search for other papers by Mr. Joseph E. Gagnon in
Current site
Google Scholar
Close
, and
Christian Saborowski https://isni.org/isni/0000000404811396 International Monetary Fund

Search for other papers by Christian Saborowski in
Current site
Google Scholar
Close

Contributor Notes

We use a cross-country panel framework to analyze the effect of net official flows (chiefly foreign exchange intervention) on current accounts. We find that net official flows have a large but plausible effect on current account balances. The estimated effects are larger with instrumental variables (42 cents to the dollar on average compared to 24 without instruments), reflecting a possible downward bias in regressions without instruments owing to an endogenous response of net official flows to private financial flows. We consistently find larger impacts of net official flows when international capital flows are restricted and smaller impacts when capital is highly mobile. A further result is that there is an important positive effect of lagged net official flows on current accounts that we believe operates through the portfolio balance channel.

Abstract

We use a cross-country panel framework to analyze the effect of net official flows (chiefly foreign exchange intervention) on current accounts. We find that net official flows have a large but plausible effect on current account balances. The estimated effects are larger with instrumental variables (42 cents to the dollar on average compared to 24 without instruments), reflecting a possible downward bias in regressions without instruments owing to an endogenous response of net official flows to private financial flows. We consistently find larger impacts of net official flows when international capital flows are restricted and smaller impacts when capital is highly mobile. A further result is that there is an important positive effect of lagged net official flows on current accounts that we believe operates through the portfolio balance channel.

I. Introduction1

Government-directed, or official, financial flows (dominated by purchases of foreign exchange reserves) have exploded over the past 15 years and are now running at more than $1 trillion per year. Current account imbalances also reached record levels in recent years and they remain a major source of tension in international economic policy, despite a partial retrenchment since 2007. Advanced economies see emerging ones as frustrating needed current account adjustment via reserve accumulation aimed at holding down the values of their currencies. Emerging economies see their advanced brethren as trying to export their way out of recession via loose monetary policies that tend to weaken their currencies. Hence the much publicized talk of currency wars.

This paper explores the first of these two arguments: Are official flows frustrating current account adjustment? Of particular interest is the extent to which official flows have a greater impact on current accounts in the presence of capital controls or other barriers to capital mobility. In addition, we explore whether there is a longer lasting impact of official flows on current accounts through the portfolio balance channel.

The paper follows in the footsteps of Chinn and Prasad (2003), Chinn and Ito (2008), Lee et al. (2008), and others who use a cross-country time-series approach to estimate the underlying determinants of current accounts. Bayoumi and Saborowski (2014), Gagnon (2012, 2013), and IMF (2012, 2013) augment the Chinn and Prasad framework to include reserve accumulation or net official financial flows as a measure of a government’s exchange rate policy.2 All five studies find a significant effect of net official flows on the current account balance. However, Bayoumi and Saborowski and IMF find that the effect of official flows is significant only in countries with capital controls, whereas Gagnon finds a large effect of official flows that is not sensitive to measures of capital controls.3 Understanding the differences in data and specification that give rise to these conflicting results is a key objective of this paper.

We use a cross-country panel approach to analyze the effect of net official flows on current accounts. The sample runs up to 26 years, from 1986 through 2011 although the baseline sample is de-facto restricted to the period of 1995–2010. A key empirical issue is the potential endogeneity of official flows to shocks to current account balances and net private flows. Endogenous movements are most likely to arise from attempts to stabilize the exchange rate in the face of trade or financial market shocks. The model is thus estimated via two-stage least squares. We propose a set of instruments for net official flows chosen to reflect possible exogenous reasons why policymakers may choose to accumulate official assets including for precautionary reasons, to save resource revenues for future generations, to borrow for economic development, or to achieve economic growth through higher net exports.

We find that net official flows have a large but plausible effect on current account balances. This result is robust to an array of samples, specifications, and estimation techniques. The estimated effects are larger with instrumental variables (42 cents to the dollar on average compared to 24 without instruments), reflecting possible downward bias in regressions without instruments owing to an endogenous response of net official flows to private financial flows.

We also find that the impact of net official flows is importantly affected by the extent of international capital mobility. We explore various measures of capital mobility. Nearly all of these measures show that the effect of net official flows on the current account declines as mobility increases. In our baseline specification, across all our measures of capital mobility, current accounts increase on average by 18 cents for each dollar of intervention in countries with capital mobility above the median compared to 66 cents to the dollar in countries with low mobility.

These results are robust to varying samples and specifications. However, coefficient magnitudes for lower capital mobility are sensitive to the countries and years included in the analysis. While our preferred specification predicts an average effect of 42 cents to the dollar, our robustness checks illustrate that the confidence interval around this number remains rather wide.

A further result is that there is an important effect of lagged net official flows, captured by the coefficient on the lagged stock of net official assets. We believe this effect operates through the portfolio balance channel. Persistent changes in the relative supplies of assets in different currencies have persistent effects on exchange rates and current account balances. This effect often, but not always, appears to increase with capital mobility, probably indicating that the portfolio channel is less important when private flows are tightly restricted. There is also some tradeoff across samples and specifications in the estimates of the net official flow and net official stock effects. When flow effects are estimated to be larger, stock effects typically are estimated to be smaller.

The paper is organized as follows: Section 2 presents the theoretical arguments underlying the hypotheses tested in this paper and discusses various measures of capital mobility. Section 3 illustrates our empirical specification and motivates the use of instrumental variables while Section 4 analyzes the regression results under the baseline specification. Section 5 presents robustness checks to the baseline regressions. Section 6 compares our results to previous research, and Section 7 illustrates the fitted model for individual countries. Section 8 concludes.

II. Background and Motivation

A. Current Accounts and Official Flows

We define official financial flows as the acquisition and disposition of assets and liabilities denominated in foreign currencies by public-sector institutions in the reporting country.4 The dominant form of official flows is purchases of foreign exchange reserves. However, public-sector borrowing in foreign currency counts as a negative official flow. Foreign asset purchases by sovereign wealth funds (SWFs) also count as official financial flows.5 We exclude countries with significant SWFs for which data do not allow the construction of comprehensive official flows.

Note that purchases and sales of a country’s assets by official institutions in other countries are not classified as official flows of the country in question. For example, purchases of US assets by Norway’s SWF count as official outflows for Norway but private inflows for the United States. Gagnon (2013) and Bayoumi and Saborowski (2014) used IMF data on the currency composition of official foreign exchange reserves (COFER) to estimate data on foreign official flows but found that these data had little effect on the coefficient of interest.6 To some extent, the use of time effects in our regressions controls for the spillover of net official flows onto other countries’ current account balances, although this would implicitly assume the spillover is equal across all countries.

According to the balance of payments (BOP) accounts, in the absence of statistical errors and omissions, a country’s current account must equal its financial account.7 A current account surplus implies net lending abroad (positive financial flows) whereas a current account deficit implies net borrowing from abroad (negative financial flows). The financial account, in turn, is the sum of net official financial flows and net private financial flows. These relationships are defined in equation 1.

C u r r e n t A c c o u n t = N e t O f f i c i a l F l o w s + N e t P r i v a t e F l o w s ( 1 )

As shown in Figure 1, net official flows grew rapidly in the years before the global financial crisis and have fluctuated around $1 trillion per year since then. The solid line in Figure 2 displays the sum of all the positive current account balances in each year (in percent of world GDP), which is a measure of global current account imbalances.8 The figure shows that these imbalances reached record levels late last decade.

Figure 1.
Figure 1.

Net Official Financial Flows

(USD billions)

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Note: Excludes countries with large unreported sovereign wealth funds.Sources: IMF Balance of Payments, Norges Bank, and authors’ calculations.
Figure 2.
Figure 2.

Net External Accounts of Countries with Current Account Surpluses

(percent of world GDP)

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Note: Excludes countries with large unreported sovereign wealth funds.Sources: IMF Balance of Payments, Norges Bank, and authors’ calculations.

The dashed line in Figure 2 is net official flows, and the dotted line is net private flows, for the same countries whose combined current account surplus is displayed in the solid line. Thus, the dashed and dotted lines sum up to the solid line, except for a relatively small statistical error. The rise in current account imbalances since 2000 is clearly associated with an increase in net official flows of a strikingly similar magnitude, whereas net private flows declined slightly and appear unrelated to the combined current account surplus.

The focus of this paper is on establishing whether this close correlation reflects a causal relationship running from official flows to current accounts, although causality need not run in only one direction. There are two ways in which purchases of official reserves could drive the current account. The first is through their impact on monetary policy and interest rates and hence domestic demand and activity; this is unsterilized intervention. The second is the impact of reserve accumulation on the exchange rate and the current account even when intervention is sterilized. In this paper, we focus on the latter effect.

B. The Case of No Private Flows

In the absence of private financial flows, Equation 1 implies that a country’s current account balance must equal its net official financial flows. In this case, an increase in net official flows increases the current account via depreciation of the exchange rate regardless of whether the official flow is fully sterilized or not. A net official financial outflow implies a transfer of capital from the home country to the rest of the world. The reduction in the domestic capital stock raises the marginal product of capital at home and the increase in the foreign capital stock lowers the marginal product abroad. This bids up domestic rates of return and pushes down foreign rates of return.9 In a world without private financial flows, interest rates and other rates of return on financial assets and the underlying capital stock can remain different across countries for extended periods of time.

C. Private Flows and Arbitrage

When private investors are allowed to send capital across borders, they will tend to arbitrage these different rates of return. Starting from a position of equal rates of return across countries, a net outflow of official capital that is fully sterilized creates an arbitrage opportunity through incipient differences in rates of return.10 Private investors will take advantage of this opportunity and send capital in the opposite direction, from the rest of the world to the home country. Thus, positive net official flows will give rise to negative net private financial flows. The standard benchmark with fully open private financial markets is uncovered interest rate parity (UIRP), according to which private financial flows keep expected exchange-rate-adjusted rates of return equal across countries. Under UIRP, sterilized official flows have no effect on the current account because they are fully offset by private financial flows.

So far, we have shown that there is a one-to-one relationship between net official flows and the current account when private financial markets are closed. And, in the opposite extreme of efficient financial markets with perfect capital mobility, sterilized official flows have no effect on the current account because they are fully offset by private flows.11 Next, we consider intermediate cases in which capital mobility is imperfect, implying that the UIRP relationship breaks down, allowing official flows to have an effect on current accounts.

D. Capital Controls

Capital controls are one potential source of imperfect capital mobility. However, the implications for the link between official flows and the current account depend on the nature of the capital control. We consider two broad types of controls: taxes and quantity controls.

  • An across-the-board withholding tax on interest, dividends, and profits earned by foreigners creates a fixed wedge between domestic and foreign rates of return. If the withholding tax rate stays constant, and UIRP would otherwise hold, then official flows have no effect on the current account because private flows adjust to maintain the fixed differential in the rates of return.

  • Quantity controls place limits on the volume of private financial flows.12 Binding quotas on inward and outward private financial flows imply that, ceteris paribus, a change in net official flows must be exactly matched by a change in the current account. In the extreme, as quotas approach zero, private cross-border financial flows are eliminated.

If financial markets are segmented, so that arbitrage is limited between foreign direct investment, portfolio equity, portfolio debt, bank debt, and other forms of capital, then it is possible that net official flows can have an effect between zero and one when quotas bind on some but not all financial instruments. As quotas become binding on more financial instruments, the effect of official flows on the current account should increase.

Menzie Chinn and Hiro Ito (2006), Dennis Quinn (1997), and Martin Schindler (2009) created indexes of the number of legal constraints on capital flows across different forms of capital for many countries and years. The Schindler index only spans the period 1995-2005. In order to allow using more recent data, this paper employs a variation of the Schindler index that employs some limited judgment, namely the Fund staff’s narrow de jure restrictiveness index (Fund index).13 Figure 3 plots median values of these measures in each year, where measures are normalized to be bounded between zero and one with higher values denoting fewer controls. Two of these measures show a trend increase in financial openness, which should imply a declining effect of net official flows on the current account for the median country. Notably, the Quinn measure finds that more than half of all countries had removed all quantitative controls on financial flows by 2009, yielding the highest possible median value of 1. The Chinn-Ito measure displays substantial liberalization over time, but significant controls remained as of 2011. The Fund measure starts in 1995 and shows little trend between 1995 and 2011 for the median country.

Figure 3.
Figure 3.

Median Values of Capital Control (Inverted)

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Source: International Monetary Fund. Excludes low-income countries.

E. Institutional Quality

There are strong reasons to believe that legal controls on financial flows are not the only factor influencing the mobility of capital. Private investors may not send capital freely into countries with few or no controls if they have reason to doubt the safety of their investments. Potential concerns include the quality of financial supervision and regulation, the ability to obtain redress of fraud and negligence in the court system, the stability of the economic environment, and the risk of expropriation or discriminatory treatment by host governments.

The World Bank’s Worldwide Governance Indicators (WGI) are a widely used source of indicators of institutional quality.14 The PRS Group’s International Country Risk Guide (ICRG) provides measures of political, economic and financial risk by country.15 We experimented with the full set of measures and many of them are highly correlated. The paper focuses on three: the WGI rule of law and regulatory quality indexes and the ICRG composite risk index.

Figure 4 displays the median values of these three indicators. The median financial risk index (the dotted line) increased sharply after 1990, representing a marked decline in perceived financial risks at that time. There is little trend in this measure since the early 1990s. The solid line shows that the median value of the regulatory quality index has increased somewhat since its inception in 1995, but this increase is small relative to the overall scale of the index (0-100). The rule of law index (the dashed line) has declined somewhat over time, but this change is also small relative to the scale of the index.

Figure 4.
Figure 4.

Median Values of Institutional Quality Measures

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Source: International Monetary Fund. Excludes low-income countries.

F. Financial Market Measures

Financial market outcomes provide alternative proxies for capital mobility. We use size indicators, both of cross-border financial transactions and of the domestic financial system. Intuitively, the magnitude of cross-border financial flows may be seen as the direct outcome of capital mobility. Alternatively, a country with a large domestic financial system may be viewed by investors as closely integrated into the global financial system. We consider three financial market measures: (1) the ratio of gross private financial transactions to the sum of gross current and gross private financial transactions in the BOP accounts; (2) the ratio of gross private financial transactions in the BOP accounts to nominal GDP; and (3) the ratio of total bank assets to GDP.16 17 Box 1 presents a simple model of cross-border investment driven solely by diversification which implies that the effect of net official flows on the current account should be inversely related to the first of these measures.

The solid line in Figure 5 is the median value of the share of private financial transactions in total BOP transactions (excluding reserve accumulation). This measure has trended up over time, but it has given back some of its gains since 2007. The dashed line is the median value of private financial BOP transactions relative to GDP. The gains over time are even more pronounced for this measure, reflecting the growing size of cross-border transactions in the world economy. The dotted line is the median value of bank assets to GDP, which has also grown over time and has retrenched by less since its peak than the BOP-based measures.

Figure 5.
Figure 5.

Median Values of Financial Market Measures

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Source: International Monetary Fund. Excludes low-income countries.

Investment as Pure Diversification

The model is based on the idea that uncertainty about expected rates of return across countries is a potential impediment that could constrain private flows from offsetting the impact of official flows. In an extreme case, market participants may have no views on differences in expected returns across countries. Nevertheless, investors may wish to reduce the overall variance of their portfolio returns by diversifying across countries. Private financial flows will then occur purely to reap the benefits of diversification. Private investors at home (US) and in the rest of the world (ROW) send financial outflows that are fixed in terms of their respective domestic currencies. In addition, for simplicity, we assume that trade flows in the current account have unitary price elasticities of demand. This implies that imports into each country are constant in terms of local currency.

Let M be US imports, fixed in $ terms; X is US exports, fixed in € terms; PFO is US outward private financial flow, fixed in $ terms; PFI is US inward private financial flow, fixed in € terms; NOF is US net official flow in $ (ROW assumed to be 0); E is the exchange rate, in $/€. The BOP identity is E * X − M = PFO − E * PFI + NOF

The effect of NOF on the current account is: Δ(E*XM)ΔNOF=E*XE*(X+PFI)

The effect of NOF on net private flows, in turn, is given by: Δ(PFOE*PFI)ΔNOF=E*PFIE*(X+PFI)

The exchange rate is determined statically based on trade flows and financial flows according to the BOP identity. An increase in net official flows pushes down the value of the domestic currency against the foreign currency, implying a rise in E. The effect on the current account is proportional to the ratio of exports to the sum of exports and private financial inflows. The effect on net private flows is -1 times the ratio of private financial inflows to the sum of exports and private financial inflows. An analogous ratio expresses the effect on ROW net private flows as the ratio of PFO to M+PFO. In order to generalize this effect from the point of view of all countries, we compute our measure of capital mobility as: Mobility=PFO+E*PFIE*X+M+PFO+E*PFI

G. Private Flows and Portfolio Balance

In a world of risk-averse investors, UIRP need not hold even in the absence of legal controls and even with high-quality regulatory regimes. Volatile exchange rates are a particularly important source of risk. The portfolio balance theory holds that relative supplies of assets in different currencies will influence the exchange rates between these currencies through investors’ desire to maintain a specific balance of portfolio holdings (Branson and Henderson 1985). An increase in domestic-currency assets will depreciate the domestic exchange rate, setting up expectations of higher future returns relative to returns on foreign currency and thus inducing investors to hold the additional supply. The link between capital mobility and the portfolio balance channel is ambiguous. One the one hand, a lack of mobility may prohibit investors from balancing their portfolios, on the other hand, portfolio rebalancing may be inherently more important when investors face tight exposure limits in countries that are less closely integrated into the global financial system.

III. Empirical Specification

The regressions are run on a sample period of up to 26 years, from 1986 through 2011. In principle, the maximum number of observations is 2054. However, owing to limitations on data availability, the baseline regressions use only 794 observations for 72 countries (see list of countries in the Appendix) while robustness checks on larger samples use up to 1213 observations.18 The coefficient standard errors in all regressions, including those using instruments, are robust to heteroskedastic and first-order autoregressive errors. Further information on the data is contained in the appendix. Equations 2 and 3 present the two basic specifications used in the analysis.

CAX it GDP it = α 1 ( NOF it / GDP it ) + α 2 ( NOF it HIGHMOB it-1 / GDP it ) + β 1 ( NOA it-1 / GDP it 1 ) + β 2 ( NOA it HIGHMOB it-1 / GDP it 1 ) + γ ( MOBILITY it 1 ) + AUX it + year t ( 2 )
NPFX it GDP it = ( α 1 1 ) ( NOF it / GDP it ) + α 2 ( NOF it HIGHMOB it-1 / GDP it ) + β 1 ( NOA it-1 / GDP it 1 ) + β 2 ( NOA it HIGHMOB it-1 / GDP it 1 ) + γ ( MOBILITY it 1 ) + AUX it + year t ( 3 )

Where CAX is the current account excluding net investment income, NPFX is net private flows excluding net investment income, NOF is net official flows, NOA is net official assets (stock) and MOBILITY is a measure of capital mobility, ranging from 0 to1. NOF_HIGHMOB is an interaction term between net official flows and a dummy that takes the value 0 when the relevant measure of capital mobility is below its median and 1 otherwise. NOA_HIGHMOB is an interaction term between net official assets and the same dummy. Auxiliary variables (AUX) include lagged PPP GDP per capita relative to the United States, the 10-year forward change in old-age dependency ratio, the lagged real GDP growth rate over the previous 5 years, net energy exports relative to GDP, and the cyclically adjusted fiscal balance relative to GDP.

Equation 2 presents the current account as a function of net official flows and other control variables. The coefficient α1 represents the effect of net official flows on the current account and the coefficient α2 allows for a differential effect when capital mobility is above its median value. The coefficient β1 represents the effect of lagged net official asset stocks on the current account and the coefficient β2 allows for a differential effect with higher capital mobility. The coefficient γ represents the direct effect of capital mobility on the current account, if any. The regressions include a standard set of controls for other potential determinants of the current account similar to those used in Gagnon (2013, 2014) and Bayoumi and Saborowski (2014).19

Equation 3 is a restatement of the link between official flows and the current account in equation 2 that takes advantage of the BOP identity: any effect of net official flows on the current account that is less than 1 must show up as a negative effect on net private flows. When net official flows have no effect on the current account (α1=0) then they must cause a one-for-one reduction of net private flows. Because of errors and omissions in the BOP data, these regressions are not identical. The bias from measurement error in net official flows in the estimate of α1 is downward in Equation 2 and upward in Equation 3, helping to put a range on its true value.

The dependent variable in Equation 2 excludes investment income from the current account in order to remove the influence of steady-state differences in stocks of net foreign assets.20 Countries with higher net foreign assets tend to have higher current accounts because of the associated net investment income. Previous research has shown that the stock of net foreign assets is a robust and important regressor when the dependent variable is the total current account and the coefficient on net foreign assets is typically in the range of 0.02 to 0.06, which is close to real rates of return on portfolio assets.21 By excluding net investment income from the dependent variable, we eliminate the need to include the stock of net foreign assets as a regressor.22 This allows us to use the stock of net official assets, a subset of net foreign assets, to estimate the lagged effect of net official flows on exchange rates and current accounts, working through the portfolio balance channel. We confirm in variants of the basic regressions (not shown) that the results are robust to including the stock of net private foreign assets (constructed as the difference between net foreign assets and net official foreign assets); the stock of net private foreign assets is no longer an important regressor after net investment income is excluded from the dependent variable and is often insignificant in the regressions.

In order to minimize the effect of outliers, we weight the observations in most of our regressions by each country’s share of world GDP, but we also present some unweighted regressions to show robustness. Weighting by GDP is appropriate if a country’s ratio of current account to GDP is interpreted as an average of the current account ratios of individual economic agents. Larger countries have less noise and idiosyncratic movements in their data than smaller countries and thus deserve greater weight. However, just two countries—the United States and the euro area23—account for nearly half of global GDP. As these two countries are unusual in having close to zero net official flows and are the dominant issuers of reserve assets, we exclude them from most of our regressions, but we run some robustness regressions, with and without GDP weights, in which we include both. Our results are qualitatively robust to including these two countries.

A key empirical issue is the potential endogeneity of official flows to shocks to current account balances and net private flows. Endogenous movements are most likely to arise from attempts to stabilize the exchange rate in the face of trade or financial market shocks. On the other hand, examples of exogenous movements in official flows include increasing holdings of foreign assets for precautionary reasons, saving resource revenues for future generations, borrowing for economic development, and achieving economic growth through higher net exports. Gagnon (2012, 2013) shows that endogeneity through stabilization of the exchange rate leads to a positive bias of the coefficient on net official flows if current account shocks dominate and a negative bias if private financial shocks dominate. Conventional wisdom suggests that financial shocks are important; witness the complaints from central banks in emerging markets about private capital flows driven by monetary policy in advanced economies. Aizenman (2006) finds that current account balances are more stable in countries with larger stocks of official reserves, corroborating the view that official flows move to offset shocks to private flows and stabilize the current account. Thus, it is likely that the coefficient on net official flows is biased downward. However, Ghosh, Ostry, and Tsangarides (2012) find that motivations for official flows shift over time and across countries, suggesting that the bias may not be constant over time and across countries.

We use instrumental variables to address the potential endogeneity of net official flows to shocks to current account balances and net private flows. The challenge is to isolate the variation in net official flows that is not driven by shocks that simultaneously drive the current account and/or private financial flows. Valid instruments must reflect exogenous motives for reserve accumulation.

Gagnon (2013, 2014) and Bayoumi and Saborowski (2014) used sets of instruments that include country dummies interacted with the lagged ratio of gross official assets to imports of goods and services as instruments. Months of import cover is a common metric for adequacy of foreign exchange reserves. The intuition is thus to capture the variation in net official flows that is related to country-specific precautionary motives governing the desired stock of reserves. One major drawback with this set of instruments is the resulting very large instrument count and the attendant risk of over-fitting. What is more, the import cover may not be a good proxy for some country’s desired stocks of foreign reserves. Finally, in countries with very persistent current account shocks, country dummies may not succeed in fully dealing with a potential endogeneity bias. We nevertheless use this set of instruments as part of our robustness checks.

Our preferred set of instruments aims to more carefully reflect country-specific motives to accumulate reserves unrelated to exchange rate or current account developments without the profligate use of country-specific intercepts and slope coefficients. The instruments are defined in Table 1. The IMF dummy and its interaction with import cover follow the intuition that countries under IMF programs, and especially those with low import cover, may be more prone to accumulate reserves for precautionary reasons. The emerging markets dummy and its interaction with relative GDP per capita, in turn, proxy for mercantilist motives: emerging markets may be more prone to accumulate reserves than advanced markets especially when they are at an early stage of development.24 25 Similarly, countries with SWFs in place are likely to have taken a forward looking decision to save; this decision would have stronger implications for reserve accumulation the higher are a country’s resource revenues.26

Table 1.

Definition of Instruments

article image

IV. Baseline Regression Results

This section describes the results of estimating Equations 2 and 3 using two-stage least squares using our preferred set of instruments described in Table 1. Tables 2a and 2b display estimates using each of the nine measures of capital mobility shown in Figures 3 through 5. In order to compare goodness of fit across different measures of capital mobility, all regressions are run on the same set of observations that are common to all measures. The baseline sample is thereby de-facto restricted to the period 1995–2010. The baseline sample excludes the United States, the euro area, and all low income economies. As discussed in the previous section, we also drop countries with SWFs whose asset purchases cannot be accounted for and we use weighted least squares with GDP weights.27

Table 2a.

Baseline Specification with Various Capital Mobility Measures and Preferred Set of Instruments

article image
Sources: Authors’ calculations.
Table 2b.

Baseline Specification with Various Capital Mobility Measures and Preferred Set of Instruments

article image
Sources: Authors’ calculations.

Note that we add one to the estimated coefficients on the net official flows term in the tables for all regressions in which net private financial flows (NPFX) is the dependent variable. As illustrated in Equation 3, adding one to the coefficient gives us an estimate of α1, the coefficient on the net official flows term in a regression in which the current account is the dependent variable.

The auxiliary variables included in the regressions are relatively standard in the literature and are similar to the ones used in Bayoumi and Saborowski (2014) and Gagnon (2012, 2013). Except for relative PPP GDP per capita, the coefficients on the auxiliary variables all have the expected signs and are in the range found by previous studies. The unexpected negative effect of relative PPP GDP per capita is small and is not significant in some cases. An increase in relative GDP of 10 percentage points is estimated to reduce the current account by 0.2 to 0.5 percent of GDP.

The first stage results are encouraging in that our instruments appear relevant. The instruments generally show the expected signs—illustrated in Table 1—in the first stage regressions when they are individually significant. In the case of the net official flows term, the F-test statistic takes values between 7.9 and 13.1 in our baseline regressions in Tables 2a and 2b; the null hypothesis that our instruments are irrelevant is comfortably rejected in all of the regressions. Similarly, the Angrist-Pischke first stage chi-squared statistic rejects the null that the net official flows term is unidentified almost throughout. We find similar evidence in the case of the interaction between the net official flows term and the various measures of capital mobility for which the F-test statistic takes values between 5.6 and 11.7.

Moving to the second stage estimation results, columns 1 and 2 of Table 2a display results using the Chinn-Ito measure of capital mobility. Column 1 is based on a regression of the current account excluding investment income (Equation 2) and column 2 is based on a regression of net private flows (Equation 3). The estimated effect of net official flows on the current account when capital mobility is below the median (α1) is 0.66 in column 1 and 0.71 in column 2. The coefficient on the interaction term between net official flows and the capital mobility dummy (α2) captures the difference in the effect of net official flows between low-mobility and high-mobility situations. The overall effect of net official flows when mobility is above the median is the sum of these two coefficients, or 0.11 in column 1 and 0.03 in column 2.

The coefficients α1 and α2 reflect the immediate effect of net official flows on the current account. However, because official flows have permanent effects on the relative supplies of assets in different currencies, they are likely to have long-lasting effects on exchange rates and current accounts through the portfolio balance channel. These effects are captured in the coefficients on the lagged stocks of net official assets. An interesting result is that the total effect of lagged official assets is positive and significant in most regressions only when capital mobility is above the median.

Taking these results together, it appears that when capital mobility is high, net official flows have a smaller immediate impact on the current account but a larger lagged impact through the reserves stock. This may occur because international capital mobility allows better smoothing short-run fluctuations in net official flows, but the greater volume of private flows increases the importance of portfolio effects. When private flows are tightly restricted, agents have less ability to maintain diversified portfolios and so accumulated stocks of official assets have less effect. A competing hypothesis with the opposite implication could have been that investors have tighter exposure limits in countries that are less integrated into the global financial system.

The finding that interventions have smaller contemporaneous but larger lagged effects in countries with high capital mobility may at first appear surprising but is line with intuition. Low mobility implies that the flow effect is strong because investors cannot take advantage of arbitrage opportunities to offset official flows. Under low mobility, portfolio rebalancing effects – which require capital to be mobile – exist but matter less as the flow effect dominates. With high mobility, the immediate effect of interventions is greatly attenuated, but the implied changes in relative asset supplies have a small but persistent effect through the portfolio balance channel. It is important to note that the flow coefficient, especially under low mobility, is much larger than the lagged stock coefficient under either low or high mobility.

In both columns 1 and 2, for countries with high capital mobility, a one percentage point higher lagged stock of net official assets increases the current account by 0.07 percent. Because stocks of official assets typically are larger than flows, this is an important effect. For example, a relatively open economy with net official assets equal to 10 percent of GDP would have a current account that is higher by 0.7 percent of GDP than a comparable country with 0 net official assets, even if net official flows were the same in both countries. Because we have excluded net investment income from the dependent variable, this effect must reflect a lasting effect through the exchange rate rather than simply the earnings on official assets.

The coefficient on the Chinn-Ito measure itself is positive but very small. This is not necessarily surprising, as there is no a priori presumption that capital controls should on average either increase or decrease the current account balance independently of official financial flows.28

Turning to columns 3 to 6, we see that the variables of interest remain significant in the regressions using the Fund index and the Quinn measure, and the coefficients retain the expected signs. The effect of net official flows on current accounts is somewhat higher on average with these measures than with the Chinn-Ito measure. The total effect of net official flows under high mobility increases to between 0.15 and 0.21, while the coefficient under low mobility is roughly unchanged, between 0.59 and 0.77. The coefficients on the lagged stock variable are very similar to those found when using the Chinn-Ito measure. Once again, the direct effect of mobility on the current account is essentially zero.

Turning to the institutional measures of capital mobility, columns 7 and 8 display results using the ICRG composite risk index. The interaction terms are not significant in the two regressions but the coefficients remain correctly signed. The effect of net official flows is again around 0.7 in the low mobility situation but when mobility is above the median, the total effect, at about 0.4, is somewhat higher than in previous regressions. There is once again no significant effect of lagged official stocks in the low mobility case and a moderately large effect with high mobility. The ICRG composite risk index itself has no effect on the current account.

The institutional measure with the best regression fit (R2) is the regulatory quality index (columns 9 and 10). The net official flows terms along with their interaction with capital mobility are all highly significant. Their coefficients have the expected signs, and the coefficient magnitudes are almost identical to those in the Chinn-Ito regressions. The lagged stock effect under high mobility is slightly higher than in previous regressions and not significantly different from zero in the low mobility case. Regulatory quality has a significant negative direct effect on the current account, but this effect is rather small economically given the relatively small range of this variable. The alternative institutional indicator of the rule of law (columns 11 and 12) is the indicator with the worst regression fit in Table 2. While the net official flows term and its interaction are significant and correctly signed in both regressions, three of the coefficients become larger than one. The average effect of net official flows, at 0.6, takes a value that is higher than in the case of all other mobility measures. What is more, somewhat surprisingly, the effect of the lagged stock of foreign assets falls to negative territory with low mobility.

Columns 1 through 6 of Table 2b focus on financial market measures of capital mobility. The results are remarkably similar to previous regressions in terms of the average effect of net official flows, although the interaction term is not significant in the case of the financial share measure discussed in Box 1. The ratio of BOP financial flows to GDP (columns 3 and 4) turns out to be the best fitting of all nine individual measures of capital mobility. Here we again find evidence in favor of an important role for capital mobility as the effect of net official flows falls from (0.47 to 0.57) in the low mobility case to around 0.2 under high mobility. The official stock effect is once again zero in the low mobility case and around 0.05 with high mobility. The direct effect of the BOP financial ratio to GDP is not significant. The final mobility measure is the ratio of bank assets to GDP (columns 5 and 6). The results here are fairly similar to those for the BOP financial flows to GDP measure.

The remaining columns of Table 2b attempt to construct a better overall measure of capital mobility by extracting information common to the various indicators. Columns 7 and 8 display results based on the first principal component of all nine mobility measures. Columns 9 and 10 are based on the first principal component of the best measures within each sub-group.29 Columns 11 and 12 display a variant of the measure used in columns 9 and 10, substituting the financial share of BOP transactions for the ratio of BOP financial flows to GDP and the ICRG financial risk index for the regulatory quality index. This final measure is the overall best fitting measure.

The results across all three principal component measures are broadly similar in terms of the average effect of net official flows on the current account. Focusing on the best-fitting “alternate” measure, the effect of net official flows with low capital mobility is 0.46 to 0.51. This effect declines significantly to around 0.21 to 0.25 with high mobility. The effect of net official asset stocks is insignificant with low mobility and 0.03 to 0.04 higher with high mobility. The direct effect of mobility on the current account is negative and significant but relatively small. An increase in capital mobility equal to half of the total range between the lowest value and the highest value in the sample would lower the current account by 0.4 percent of GDP.

The green bars in Figure 6 display the average effects of net official flows on the current account across all the regressions of Tables 2a and 2b. The first bar, labeled High, is the sum of α1 and α2, which is the effect under high capital mobility. The second bar, labeled Low, is α1, which is the effect under low capital mobility. The bars in the other colors refer to alternative sets of regressions discussed below. On average across the regressions in Tables 2a and 2b, each dollar of net official flows raises the current account 18 cents with high capital mobility and 66 cents with low capital mobility, for an average effect of 42 cents.

Figure 6.
Figure 6.

Coefficients on Net Official Flows Term under Low and High Mobility

Citation: IMF Working Papers 2014, 199; 10.5089/9781484380239.001.A001

Source: Authors’ calculations.

V. Robustness to Sample and Specification

Tables 3a and 3b show the results from the same specification as in Tables 2a and 2b except that we do not use instruments to identify the effect of net official flows on current accounts. The tables show that both the net official flows term and its interaction generally are significant with correctly signed coefficient. What is more, the coefficients are more stable across regressions than when using two-stage least squares. Interestingly, the effect of net official flows on the current account is similar to the results found in Tables 2a and 2b in the high mobility case but is significantly smaller in the low mobility case. The blue bars in Figure 6 illustrate this finding by averaging the relevant effects across all regressions in the tables. While the average impact under low mobility is 0.66 in the case of the instrumented regressions (the green bars), it falls to half of that in the regressions without instruments (the blue bars). The average effect of net official flows on current accounts is thus 24 cents to the dollar in the regressions without instruments compared to 42 cents to the dollar in the two-stage least squares regressions in Tables 2a and 2b. This result is consistent with the idea that the bias in a regression without instruments when net official flows move to stabilize the exchange rate tends to be negative because financial shocks to the exchange rate are more important than trade shocks. Finally, the lagged stock of official assets is now often significant as a determinant of the current account independently of capital mobility. However, the magnitude of the effect remains broadly the same.

Table 3a.

Regressions Including Various Mobility Measures without Instruments

article image
Sources: Authors’ calculations.
Table 3b.

Regressions Including Various Mobility Measures without Instruments

article image
Sources: Authors’ calculations.

Tables 4a and 4b again show the same set of regressions as in the previous tables but use two-stage least squares with the alternative set of instruments described in Table 1. The instruments include country dummies as well as their interaction with the lagged ratio of gross reserves to imports of goods and services. The net official flows term and its interaction with capital mobility are once again significant in almost all regressions. The coefficients are correctly signed and are more stable than in our preferred set of regressions. As illustrated by the red bars in Figure 6, the effect of net official flows under high mobility is on average only slightly lower than that found in Tables 2 and 3; the average effect in the low mobility case, in turn, is 0.50 and thus lies between that found in the case of the regressions without instruments and those with the preferred instrument set. In the high mobility case, on the other hand, the coefficient shrinks to only slightly over 0.10.

Table 4a.

Regressions Including Various Mobility Measures with Alternative Set of Instruments

article image
Sources: Authors’ calculations.
Table 4b.

Regressions Including Various Mobility Measures with Alternative Set of Instruments

article image
Sources: Authors’ calculations.