Regulating Capital Flows at Both Ends
Does it Work?†
Author:
Mr. Atish R. Ghosh
Search for other papers by Mr. Atish R. Ghosh in
Current site
Google Scholar Close
,
Miss Mahvash S Qureshi
Search for other papers by Miss Mahvash S Qureshi in
Current site
Google Scholar Close
, and
Naotaka Sugawara
Search for other papers by Naotaka Sugawara in
Current site
Google Scholar Close

Contributor Notes

Author’s E-mail address: aghosh@imf.org; mqureshi@imf.org; nsugawara@imf.org
Publication Date:
17 Oct 2014
eISBN:
9781498358743
Language:
English
Keywords:
WP; recipient country; source country; capital control; bank flow; outflow control; real interest rate; inflow control; asset flow; FX-denominated assets; control measure
Download PDF (1,014.6 KB)

This paper examines whether cross-border capital flows can be regulated by imposing capital account restrictions (CARs) in both source and recipient countries, as was originally advocated by John Maynard Keynes and Harry Dexter White. To this end, we use data on bilateral cross-border bank flows from 31 source to 76 recipient (advanced and emerging market) countries over 1995–2012, and combine this information with a new and comprehensive dataset on various outflow and inflow related capital controls and prudential measures in these countries. Our findings suggest that CARs at either end can significantly influence the volume of cross-border bank flows, with restrictions at both ends associated with a larger reduction in flows. We also find evidence of cross-border spillovers whereby inflow restrictions imposed by countries are associated with larger flows to other countries. These findings suggest a useful scope for policy coordination between source and recipient countries, as well as among recipient countries, to better manage potentially disruptive flows.

Abstract

This paper examines whether cross-border capital flows can be regulated by imposing capital account restrictions (CARs) in both source and recipient countries, as was originally advocated by John Maynard Keynes and Harry Dexter White. To this end, we use data on bilateral cross-border bank flows from 31 source to 76 recipient (advanced and emerging market) countries over 1995–2012, and combine this information with a new and comprehensive dataset on various outflow and inflow related capital controls and prudential measures in these countries. Our findings suggest that CARs at either end can significantly influence the volume of cross-border bank flows, with restrictions at both ends associated with a larger reduction in flows. We also find evidence of cross-border spillovers whereby inflow restrictions imposed by countries are associated with larger flows to other countries. These findings suggest a useful scope for policy coordination between source and recipient countries, as well as among recipient countries, to better manage potentially disruptive flows.

I. I ntroduction

“Control will be more difficult to work by unilateral action on the part of those countries which cannot afford to dispense with it …if movements of capital cannot be controlled at both ends.”

- J. M. Keynes1

“Almost every country, at one time or another, exercises control over the inflow or outflow of investments, but without the co-operation of other countries such control is difficult, expensive and subject to considerable evasion.”

- H. D. White1

Capital flows to emerging markets (EMs) have been extraordinarily volatile since the global financial crisis (GFC), prompting debates on whether—and how—to better manage cross-border flows. While most of the literature has focused on the policy options of countries at the receiving end (e.g., Ostry et al., 2010, 2011; IMF, 2011a), some recent studies and policy papers call for a more coordinated approach to regulating these flows by acting at both the source and recipient country ends (e.g., Ostry et al., 2012a; IMF, 2012; Brunnermeier et al., 2012). But can such restrictions help tame destabilizing cross-border flows? And is there any merit to a cooperative approach of operating at both ends? These are the questions we take up in this paper.

The idea of regulating short-term speculative flows “at both ends” to reduce excessive capital flow volatility is not new. In fact, this is one issue on which the main architects of the Bretton Woods system—John Maynard Keynes and Harry Dexter White—were in complete agreement when debating the post-war international monetary system.2 But what would be the incentives for countries to coordinate their capital account policies? While such incentives are clear for recipient countries seeking to limit temporarily large or risky inflows (e.g., to prevent economic overheating, currency overvaluation, and financial instability), coordination may benefit source countries as well—for instance, by lowering the likelihood of an eventual financial crisis in recipient countries that could inflict losses on source country financial institutions (and hence on their taxpayers), or even threaten the stability of the international monetary system. Limiting outflows may also make accommodative monetary policy more effective in stimulating the source country economy by preventing “leakage” of liquidity abroad.3

Historically, there have been a few instances when coordinated regulation of capital flows was contemplated, but political constraints and the precedence of domestic interests prevailed, leaving countries to contend with large capital inflows (or outflows) on their own.4 A prerequisite for successful coordination, of course, is that measures adopted by source and recipient countries actually affect capital flows. Evidence on that is decidedly mixed. At the recipient end, studies focusing on the impact of capital controls on inflows generally find no effect of such controls on the total volume of inflows or on exchange rate appreciation, but a statistically significant impact on the composition of liabilities, and on financial-stability risks.5 The evidence on capital outflow controls points to their effectiveness in some cases (e.g., in Malaysia), but not more generally.6 More recently, studies find that some prudential measures (especially those related to foreign currency transactions) could also influence capital flows indirectly by altering the composition of external liabilities (Ostry et al., 2012b; Forbes et al., 2013).

In this paper, we combine these strands of research and investigate whether the adoption of either capital controls or prudential measures (that may act like capital controls)—hereafter collectively referred to as capital account restrictions, CARs—on outflows by source countries, and on inflows by recipient countries, could present an effective strategy to moderate large (and possibly) disruptive capital movements, as was originally proposed by Keynes and White. This is a hitherto unexplored issue in the literature, perhaps because, as mentioned above, there have been no significant instances of coordinated policy action on CARs, and also because comprehensive data on inflow and outflow restrictions—including both residency-based capital controls and prudential measures that may act like them—have been lacking.7 To get around these issues, we use bilateral data on cross-border bank flows from 31 major source to 76 major recipient countries over 1995–2012, and combine this information with different types of CARs in these countries—constructed using detailed information from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER) and the OECD Code of Liberalization of Capital Movements.

The use of bilateral data on bank flows in this context provides several distinct advantages. First, the bilateral nature of the data allows us to determine simultaneously the association between capital flows and CARs from both the outflow and inflow sides—while controlling for a range of source and recipient country characteristics—and examine the interaction between these policy measures. Second, cross-border bank flows, which have increased exponentially over the years, form a sizable proportion of total cross-border lending and tend to be highly procyclical—with the potential to create serious economic and financial instabilities (Brunnermeier et al., 2012).8 Analyzing their association with both general capital controls, as well as with more targeted prudential measures that can act like capital controls (e.g., foreign currency related measures) could provide useful insights into how to better regulate these flows. Third, to the extent that bilateral flows across country pairs are not perfectly correlated, using bilateral data helps to mitigate potential endogeneity concerns in econometric estimations, and to better identify the impact of CARs on cross-border flows (since such measures are generally adopted in response to the aggregate—and not bilateral— volume of flows).9 Finally, the bilateral cross-border bank flows data comprises information on flows from major source countries to both advanced and EM recipient countries, enabling us to analyze the pre- and post-GFC capital inflow surge to different regions, and exploit the cross-country and intertemporal variation in CARs to assess their effectiveness.10

We begin our analysis by examining the association between cross-border bank flows and CARs in source and recipient countries individually. The results show that such restrictions are associated with a significantly lower volume of cross-border flows: on the source side, moving from the lower to the upper quartile on measures of overall, bond, equity, direct investment or financial credit outflow controls is associated with about 50–100 percent lower flows, while prudential measures such as restrictions on lending to nonresidents also imply a similar reduction in flows. On the recipient side, moving from the lower to the upper quartile on overall and bond inflow controls, and on the existence of foreign currency (FX) related prudential measures (such as restrictions on lending in FX, and open FX position limits) is associated with some 50–80 percent lower inflows. Among other factors, we find a strong effect of global risk aversion and monetary policy (proxied by the interest rate) in source countries on bank flows—highlighting the procyclical nature of such flows—as well as of the domestic interest rate and exchange rate regime of the recipient countries.

Controlling simultaneously for both source and recipient country CARs, the estimated effects remain largely unchanged. Individual measures are, however, associated with a larger reduction in flows when the other side is financially more open—though not necessarily fully open. Inasmuch as capital controls are effective, this result makes intuitive sense: when the source country is already restricting outflows, the incremental effect of inflow restrictions on the volume of flows will be smaller than if the source side is completely free. Likewise, the incremental effect of source country restrictions on outflows will be smaller when the recipient is already restricting inflows. The estimated effects of source and recipient country restrictions are thus partially (but not fully) additive, making it possible to operate CARs at both ends, achieving either a larger reduction in flows, or the same reduction with less intensive (and therefore perhaps less distortive) measures at either end.

Our estimates suggest that pre-GFC flows to emerging Europe and the Eurozone peripheral countries would have been substantially lower in the presence of CARs (including prudential measures) at either end. In addition, we find evidence of potential spillover effects from inflow controls—the volume of bank flows received by the recipient country is significantly larger when its neighbors (defined in terms of regional or economic similarity) are financially relatively closed. Overall, our results survive a battery of robustness tests, including addressing potential endogeneity concerns through the use of instrumental variable approach (where we take (lack of) monetary/central bank freedom and the presence of a democratic left-wing government as instruments for the existence of CARs); using alternate samples (e.g., restricting the source (recipient) country sample to advanced (EM) countries, excluding the post-GFC years, or removing offshore financial centers from the sample); or defining the dependent variable in alternate ways (such as in stock, rather than in flow, terms).

Our findings have important policy implications. First, given the close connection between cross-border bank flows and risks to global financial stability, our analysis suggests that adoption of relevant CARs in boom times could help to dampen the procyclicality of these flows, thereby lowering the risk of systemic financial crises. Second, the traction of both source and recipient country CARs in regulating flows implies that coordination could be useful to achieve a given reduction in cross-border flows by adopting relatively lower levels of restrictions at both ends—instead of more intensive controls at one end—which is a globally more efficient outcome when the cost function of capital controls is convex, as seems plausible (Ostry et al., 2012a).11 Such coordination may be especially beneficial when the scope to act at one end is limited—for instance, because of weak institutional or administrative capacity to enact measures, or because international legal obligations constrain the availability of certain restrictions. Third, considering the role of source country monetary policies in pushing bank flows to recipient countries, our results allude to the importance of coordination on the monetary policy dimension as well. To the extent that countercyclical monetary policy is required in source countries, its effectiveness on the domestic economy could be enhanced, and adverse cross-border spillovers could be reduced, by the adoption of suitable capital account related policies.12 Finally, considering the spillover effects of inflow related CARs, our results suggest potential gains from coordination amongst recipient countries as well, to avoid costly “capital control wars.”

We make several contributions to the existing literature. First, unlike previous studies, which focus on the impact of either capital inflow or outflow restrictions, we examine the joint effect of restrictions on outflows adopted by the source country, and on inflows by the recipient country, while controlling for a range of source and recipient country factors. In doing so, we not only establish the effect of outflow controls on the country implementing the control, but also that on the recipient countries. Second, while existing literature has analyzed extensively the effect of capital controls and prudential measures on aggregate and portfolio flows, we analyze their impact on banking flows—an increasingly important and volatile component of total cross-border capital flows—which remains largely unexplored.13 Third, for our purposes, we construct a comprehensive dataset of capital controls (disaggregated by asset type) and prudential measures that may act as capital controls for both source and recipient countries over 1995–2012. Exploiting this rich bilateral dataset, and the cross-country and time variation in measures, we are able to establish a significant association between CARs and the volume of cross-border bank flows from source to recipient countries, as well as the potential spillover effects of CARs in recipient countries. This allows us to gauge the viability of regulating capital flows internationally.

The remainder of the paper is organized as follows. Section 2 describes the data used in the empirical analysis, and presents some stylized facts. Section 3 examines whether CARs imposed by source and recipient countries affect the volume of capital flows between them. Section 4 analyzes the implications of implementing CARs simultaneously in both source and recipient countries, examines the spillover effects of imposing controls, and presents an extensive sensitivity analysis to establish the robustness of our results. Section 5 concludes.

II. Cross-Border Bank Flows and Capital Account Related Measures

To examine whether source and recipient country CARs influence cross-border bank flows, we obtain annual data on bilateral bank asset flows from the Bank of International Settlement’s (BIS) Locational Banking Statistics by Residence for 31 (reporting) source countries to 76 recipient countries over 1995–2012.14 Our source and recipient countries include both advanced countries and EMs, though data availability varies across countries (Table A1). Flows are estimated by the BIS as the exchange rate-adjusted changes in the gross international financial claims of resident banks in the reporting country on the bank and nonbank sectors of recipient countries (i.e., as changes in total stock, amounts outstanding, of reporting country banks’ foreign assets, accounting for repayments and exchange rate effects). The data—originally complied by reporting country central banks, and then provided to the BIS—cover over 90 percent of the international assets of the domestic banking institutions, and comprise cross-border bank loans, bank credit lines (used portions), trade-related credit, as well as debt securities, equity holdings and participations of banks.15

To obtain information on prudential measures and capital controls, we draw on the IMF’s AREAER database, supplemented with information from the OECD Code of Liberalization of Capital Movements.16 By definition, prudential measures are provisions specific to the regulated domestic financial sector (notably banks, but sometimes also other financial institutions). These could discriminate by the currency denomination of the capital transaction (e.g., restrictions on local FX lending), or could be general (nondiscriminatory) measures to preserve financial system stability (such as cyclical capital requirements, maximum loan-to-value ratios, etc.). Capital controls, by contrast, may be economy-wide and apply to all residents (but could also be sector-specific), and restrict capital transactions by virtue of the residency of the parties to the transaction. In this respect, measures specific to the financial sector that discriminate based on residency (such as restrictions on lending to, or borrowing from, nonresidents) could be considered as financial sector-specific capital controls. It is also important to recognize that certain prudential measures—especially those that discriminate based on currency denomination of the capital transaction—can also influence flows, effectively acting as capital controls (Ostry et al., 2012b; IMF, 2012).17

For our empirical analysis, we consider several capital outflow and inflow related CARs from the source and recipient sides, respectively, that could potentially affect cross-border bank flows. These include economy-wide capital controls disaggregated by asset class (i.e., on bond, equity, direct investment and financial credit flows), and measures specific to the financial sector. Specifically, for source countries, we consider capital controls on bonds, equity, direct investment, and financial credit outflows (as well as a measure of the overall restrictiveness on capital outflows), and the following prudential measures: (i) restrictions on lending to nonresidents; (ii) restrictions on maintenance of accounts abroad; and (iii) open FX position limits. For recipient countries, we include capital controls on bond, equity, direct investment and financial credit inflows (along with a measure of the overall restrictiveness on capital inflows), and the following prudential measures: (i) restrictions on lending locally in FX; (ii) restrictions on purchase of locally issued securities denominated in FX; and (iii) open FX position limits.18

These various measures are expected to reduce the volume of cross-border flows, with the exception of open FX position limits in source countries, where the effect is potentially ambiguous. If banks in source countries have few FX deposits, then open position limits are likely to reduce cross-border bank lending as the bank would be limited in the foreign currency (and hence foreign) assets it could acquire. Conversely, if banks in source countries have large FX deposits, then position limits would force them to acquire FX assets, which may well take the form of cross-border bank loans. In recipient countries, however, open FX position limits are likely to reduce cross-border bank borrowing as the bank would have to acquire FX-denominated assets, which could entail greater credit risk if domestic borrowers lack a natural hedge. (Note that banks in countries with reserve currencies face fewer constraints from such limits than banks in other countries.)

We capture prudential measures through binary variables with a value of one indicating the presence of a restriction (and zero otherwise). To construct the capital control measures, we follow the approach of Schindler (2009), which—importantly for our purposes, and unlike other capital account openness indices—allows us to differentiate between controls on inflows and on outflows, while also allowing differentiation between restrictions by asset type. These measures are constructed by taking averages of binary variables reflecting the existence of controls at the level of individual (resident/nonresident) transactions, and range between 0 and 1. For example, controls on bond and equity inflows are an average of binary variables reflecting the presence of restrictions on the sale of securities abroad by residents, and on the purchase of locally issued securities by nonresidents. By contrast, controls on bond and equity outflows are an average of binary variables indicating restrictions on the purchase of securities abroad by residents, and on the sale of locally issued securities by nonresidents. These measures can thus assume three values: 0 (no restriction), 0.5 (restriction on one type of transaction), or 1 (restrictions on both types of transactions). For controls on direct investment and financial credit flows, where the AREAER provides less disaggregated information, the inflow/outflow controls can take on only two values, 0 or 1.19

The overall restrictiveness on capital outflows and inflows is captured by taking the average of measures over different types of asset classes (e.g., bond, equity, direct investment, financial credit flows, as well as over money market instruments and collective investment flows), and hence can assume a range of values between 0 and 1. Ideally, both prudential and capital control measures would capture the intensity, rather than just the existence, of various restrictions. In practice, however, this is almost impossible to do, especially for administrative measures, without making arbitrary choices. For this reason, like existing literature, we rely on indicators of de jure restrictiveness that capture the presence of measures. Nevertheless, since some measures—such as simple notification requirements—may reflect mere formalities that are unlikely to have a substantial impact on cross-border flows, or there may be restrictions that apply to specific countries (that are not part of the sample), we treat these measures as no restrictions. Moreover, we create alternative outflow/inflow control measures where provisions unlikely to be directly related to cross-border banking flows (such as restrictions on investment in a limited number of sectors for national security purposes), or milder regulations such as registration requirements, are not treated as restrictions, and use them to check the robustness of our results below.

A. Stylized Facts

By any estimate, cross-border bank flows have ballooned over the past couple of decades. Total bank asset flows from reporting advanced source countries to advanced recipient countries increased from about USD 435 billion in 1995 to almost USD 4 trillion in 2007, before dropping sharply during the GFC in 2008–09 (Figure 1[a]). Recovery has been gradual and volatile, with flows totaling about USD 280 billion in 2010, but falling again in 2011–12. Flows from advanced countries to EMs, though smaller in absolute terms, present a similar picture—but with a somewhat sharper post-crisis recovery. The post-crisis bounce back in flows is, however, not uniform across regions—Latin America and Asia have been the major recipients, with total flows received in 2010 close to the pre-crisis peak, while recovery in emerging Europe has lagged. Flows from (reporting) EMs to advanced and to other EMs also increased sharply before the GFC, and have picked up since the sharp fall in 2009, but the recovery is largely driven by EM-to-EM flows (Figure 1[b]). Expressed in percent of (either source or recipient country) GDP, the pre-crisis rise in flows remains striking, and the recovery appears modest but volatile (Figure 1[c], [d]).

Figure 1.
Figure 1.

Cross-Border Bank Asset Flows, 1995–2012

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Source: BIS Locational Statistics.Note: Statistics reflect exchange rate adjusted changes in the total stock (amounts outstanding) of assets (all instruments). Advanced and EME source countries in the Figure include those for which data is available from 1995 and 2004 onward, respectively (see Table A1).

In terms of the importance of bank inflows in the total inflows received by countries, Figure 2 shows that their share gradually increased in the runup to the GFC. On the eve of the crisis in 2007, they constituted about 50 and 40 percent of gross inflows to advanced countries and EMs, respectively.20 Among EMs, perhaps the starkest increase has been for the Baltics, where their share increased from 11 percent in 2000 to about 70 percent just before the crisis. It is also pertinent to note the highly procyclical nature of these flows, as documented in earlier studies (e.g., Milesi-Ferretti and Tille, 2011; Brunnermeier et al., 2012), with surges in bank inflows often followed by sharp declines (such as in Asia and Latin America in the late 1990s, and more generally across all countries in the GFC).

Figure 2.
Figure 2.

Cross-Border Bank and Nonbank Liability Flows, 1995–2012

(in USD tril.)

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Source: BIS Locational Statistics and IFS database (based on BPM5 presentation).Notes: Bank liability flows for recipient regions computed as the sum of gross bank asset flows from all source countries to that region. Nonbank liability flows computed as the difference between total (gross) liability flows to the region (obtained from IFS) and bank liability flows.

Along with the rise in banking flows has been an increasing trend of using prudential measures likely to affect capital inflows—especially in EMs. For example, Figure 3[a] shows that the proportion of EMs in the sample with some form of restriction on lending locally in FX increased from about 50 percent in 2001 to 57 percent in 2007, and further to 70 percent in 2012. Similarly, the proportion of EMs with restrictions on purchase of locally issued securities denominated in FX increased from 42 percent in 2001 to 46 percent in 2007, and to 49 percent in 2012. These statistics, however, mask important variations across regions. FX-related regulations are much more common in Asia, and the least prevalent in emerging Europe, though the trend seems to have reversed somewhat in the latter after the GFC (Figure A1). By contrast, capital controls on inflows generally declined in EMs for the most part of 2000s, but have become relatively more prevalent post-GFC.

Figure 3.
Figure 3.

Capital Controls and Prudential Measures, 1995–2012

(in percent of observations)

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Source: Based on IMF’s AREAER and the OECD Code of Liberalization of Capital Movements (various issues).Note: Statistics on prudential measures reflect the proportion of countries in the sample with the specific measure in place. Controls indices reflect the average of the overall (outflow and inflow) restrictiveness indices. The jump in the outflow controls index for advanced countries in 2003 in Figure 3[b] is mainly because of measures introduced by some EU countries on the purchase of securities abroad by insurance companies and pension funds, as reported by the OECD and AREAER.

Both FX-related prudential measures and capital inflow controls are, however, much less common in advanced countries—though there seems to have been a slight increase in the former in recent years, possibly because of the fallout from the GFC. (For example, France, Italy and Portugal adopted open FX position limits in 2009–10.) Similarly, CARs on outflows seem to be much less prevalent in advanced source countries as compared to EM source countries, but there exists considerable cross-country and time variation across different measures even among the former (Figure 3[b]). For instance, of the restrictions considered here, the United Kingdom has in place only the open FX position limits on banks, while since the year 2008, Iceland has imposed an extensive set of outflow controls. Some other countries have mild capital controls (on outflows) in the form of registration requirements for public offerings of securities by foreign issuers, or restrictions targeting specific sectors such as insurance companies or pension funds. Overall, the existence of prudential measures and capital controls (pertaining to both inflows and outflows) tends to be positively correlated, though the correlation is less strong for advanced countries (Table 1).

Table 1.

Correlation between Capital Controls and Prudential Measures, 1995–2012

article image
Note: ***, **, and * indicate statistical significance of the correlation coefficients at the 1, 5, and 10 percent levels, respectively.

DI=Direct investment

FC=Financial credit

Have these restrictions been successful in curtailing flows? Figure 4[a] presents a snapshot of countries with below and above median capital controls, and bilateral cross-border bank asset flows (in logarithmic terms).21 On average, bilateral flows are significantly lower if restrictions on capital outflows and inflows are in place in source and recipient countries, respectively. The combined presence of these measures further dampens flows. Similarly, prudential restrictions on maintenance of accounts abroad and on lending to nonresidents in source countries have a significant association with bank outflows, while FX-related measures in recipient countries also seem to discourage inflows. These statistics are however unconditional averages; in what follows, we explore the link between source and recipient country CARs and bank flows more formally through regression analysis.

Figure 4.
Figure 4.

Cross-Border Bank Flows, Capital Controls, and Prudential Measures, 1995–2012

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Note: Banking flows measured as log of exchange rate adjusted changes in the total stock (amounts outstanding) of assets (all instruments). *** indicates statistically significant different means between the two groups at the 1 percent level.

III. Do Capital Account Restrictions Matter?

We begin by examining the association between bank asset flows and CARs on capital outflows in source countries, and on capital inflows in recipient countries, individually. To do so, we draw on existing literature and estimate the following gravity-type models:

F i j t = X i t ′ β i + X j t ′ β j + γ R i t + μ i j + λ t + ɛ i j t ( 1 )
F i j t = X i t ′ δ i + X j t ′ δ j + χ R i t + μ i j + λ t + η i j t ( 2 )

where Fijt is (the log of) gross bank asset flows from source country i to recipient country j in year t; Xi and Xj include control variables for source and recipient countries, respectively; Ri and Rj are source and recipient country’s outflow and inflow related CARs, respectively, that are likely to affect bilateral bank flows between them; μij are the source-recipient country specific effects to capture time-invariant factors (such as geographical distance, political and cultural ties, etc.) that may affect bilateral flows, but could also be correlated with the regressors; λt are time effects to capture common shocks across country pairs; and εijt and ηijt are random error terms. We estimate (1) and (2)—which constitute our benchmark models— by including the relevant CARs individually to avoid potential multicollinearity issues.

Following existing literature on cross-border bank flows (e.g., Papaioannou, 2009; Blank and Buch, 2010; Herrmann and Mihaljek, 2010; Bruno and Shin, 2013), our control variables include several proxies for source country “push” and recipient country “pull” factors. These include (log) real GDP and real GDP per capita (to proxy for the economic size and level of economic development, respectively), real interest rate (to reflect return on investment), and real GDP growth rates of both source and recipient countries. In addition, we include the current account balance, expressed in percent of GDP (to reflect the external financing need), and the exchange rate regime (equal to one if the country has a pegged regime, and zero otherwise) of the recipient country. While the aggregate nature of our control and CAR variables (which tend to respond to the total, rather than the bilateral, volume of flows) helps to identify their effect on cross-border flows, we nevertheless lag (by one period) all source and recipient country-specific variables when estimating (1) and (2) to mitigate potential reverse causality concerns (endogeneity of CARs is also addressed through the instrumental variable approach below).22 Further, considering the long time span of our data and the possible correlation in the error term, we cluster standard errors at the country-pair level. For comparative purposes, we first estimate (1) and (2) by the Ordinary Least Squares (OLS) without controlling for country-pair or year effects, but include several time-invariant country-pair specific variables, as well as global factors (such as global market uncertainty— proxied by the VIX index—and commodity prices) that could potentially affect cross-border capital flows. We then estimate the benchmark model as specified above with country-pair and year effects (CPFE/TE).23

A. Source Country Restrictions

The OLS results for (1), presented in Table 2 (col. [1]), show that the estimated coefficients for most control variables are of the expected sign and are statistically significant.24 Among the specific variables, global market uncertainty has a significantly negative effect on bank flows—with a one percent increase in the VIX index decreasing flows by about 8 percent. Geographical distance between the country pair also reduces flows, likely capturing the impact of informational asymmetries between countries (Ghosh and Wolf, 2000; Buch, 2003; Portes and Rey, 2005). An increase in source and recipient country sizes (proxied by real GDP and land area) and their real growth rates leads to significantly larger flows between them. Recipient countries with higher real per capita GDP attract significantly larger flows (suggesting that better developed financial institutions matter; Dell’Ariccia et al., 2008; Papaioannou, 2009; Binici et al., 2010), while source countries with higher real GDP per capita appear to remit relatively smaller flows. Higher real interest rate and larger external financing need of the recipient country also imply larger inflows: a 100 basis points rise in the real interest rate and a 1 percentage point increase in the external financing need, for instance, are associated with about 10 and 6 percent larger inflows, respectively; while the estimated coefficient of source country real interest rate is statistically insignificant. The estimated coefficient of recipient country exchange rate regime is positive (implying larger flows to countries with less flexible regimes), but statistically insignificant.

Table 2.

Cross-Border Bank Flows and Source Country CARs, 1995–2012

article image
Note: Dependent variable is (log of) bank asset flows from country i to j. All variables except for VIX and commodity price index are lagged one period. Constant is included in all specifications. R2 reported for CPFE estimations is the within-R2. Clustered standard errors (by country-pair) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

The estimated coefficient on the variable of interest—the overall restrictiveness on capital outflows in the source country—in the OLS regression is negative and statistically significant (at the 1 percent level), suggesting that moving from the 25th to the 75th percentile of the index is associated with about 70 percent lower outflows. The result remains similar if we estimate the benchmark model with country-pair and year effects (CPFE/TE)—the estimated coefficient for the overall restrictiveness on capital outflows in the source country now implies about 77 percent lower outflows if we move from the bottom to the top quartile of the index (col. [2]). With the inclusion of CPFE/TE, however, the time invariant (country-pair specific) variables drop from the model, and some variables (e.g., source country’s real GDP and real GDP per capita, and recipient country’s external financing need and real GDP per capita) lose statistical significance. By contrast, the estimated coefficients of source country real interest rate and recipient country exchange rate regime turn statistically significant, implying that a 100 basis points increase in the former reduces bank outflows by about 20 percent, while the existence of a pegged exchange rate in the recipient country implies a six-fold increase in flows relative to a floating regime.

Going beyond the overall capital outflow controls index, the results for more disaggregated measures show that controls on bond, equity, direct investment and financial credit outflows are associated with significantly lower bank outflows (cols. [3]-[6]). Moving from the bottom to the top quartile of these indices reduces flows by some 50–100 percent. These results are in line with Binici et al. (2010), who use aggregate (instead of bilateral) data on capital flows, and find that controls on debt and equity outflows reduce these flows by 57–63 percent. That different types of outflow controls are associated with lower outflows suggests that these measures are able to target the different components of cross-border bank asset flows (loans, debt, equity and direct investment).25 Among the prudential measures considered here, the estimated coefficient of restrictions on lending to nonresidents—which could equally be classified as a financial sector capital control—is negative and statistically significant implying a reduction in bank asset flows by about 100 percent, while those of maintenance of accounts abroad and open FX position limits are statistically insignificant (cols. [7]-[9]).

B. Recipient Country Restrictions

What about the impact of CARs on inflows imposed by recipient countries and the volume of bank flows received? The estimation results for (2) suggest a varying impact of different measures on inflows (Table 3). The estimated coefficient for the overall capital inflow controls index is significantly negative (at the 5 percent level) in the CPFE/TE estimation, implying that moving from the lower to the top quartile of the index would be associated with a reduction in bank asset flows by about 50 percent (col. [2]). Bond inflow controls also have a statistically significant association with bank inflows—moving from the 25th to 75th percentile on the bond inflow controls index lowers bank inflows by about 64 percent (col. [3]). The estimated coefficients for equity and direct investment inflow control indices are, however, wholly statistically insignificant, but that of financial credit inflow controls is marginally insignificant (with a p-value of 0.11; cols. [3]-[5]). This suggests a somewhat asymmetric effect of such controls—their adoption by the source country tends to significantly reduce banking outflows (as noted above), while their imposition by the recipient country does not appear to have a statistically strong impact on inflows, perhaps partly reflecting differences in institutional/administrative capacity across the source and recipient countries mostly imposing controls.26

Table 3.

Cross-Border Bank Flows and Recipient Country CARs, 1995–2012

article image
Note: Dependent variable is (log of) bank asset flows from country i to country j. All variables are lagged one period. Constant is included in all specifications. R2 reported in the CPFE estimations is the within-R2. Clustered standard errors (at the country-pair level) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

Importantly, FX-related prudential measures are strongly related with lower cross-border bank flows: inflows are about 70–80 percent lower in the presence of restrictions on lending locally in FX, and open FX position limits in the recipient countries. Inasmuch as domestic lending in foreign currency largely relies on foreign (especially, bank) financing—as in emerging Europe before the GFC—these findings are intuitive, and support those of earlier studies which report a significant effect of such restrictions on local FX-denominated lending by banks, as well as on the composition of external liabilities (e.g., Ostry et al., 2012b).

IV. Restrictions at Both Ends

The analysis above establishes a strong and significant association between outflow and inflow restrictions and flows from source to recipient countries. To determine the extent to which these restrictions can jointly influence cross-border flows, we modify the benchmark specification to simultaneously include both outflow and inflow related CARs, as follows:

F i j t = X i t ′ φ i + X j t ′ φ j + ρ i R i t + ρ j R j t + μ i j + λ t + ν i j t ( 3 )

where the definition of all variables remains the same as above.

Doing so, the estimated coefficients of outflow and inflow restrictions remain very similar— both in terms of magnitude and statistical significance—to those reported above (Table 4). Thus, for instance, overall, bond, equity, direct investment and financial credit controls, and restrictions on lending to nonresidents on the source side are statistically significant and imply a reduction in outflows of about 50–100 percent across different specifications. On the recipient side, inflow controls (overall and bond) are again associated with significantly lower cross-border flows by about 50–60 percent (cols. [2]-[3]). Among measures specific to the financial sector, the impact of restrictions on lending to nonresidents in source countries, and on local FX lending in recipient countries, is statistically significant even when considered jointly with CARs from the other side (cols. [6]-[9]).

Table 4.

Cross-Border Bank Flows, Source and Recipient Country CARs, 1995–2012

article image
Note: Dependent variable is (log of) bank asset flows from country i to country j. All variables are lagged one period. Constant is included in all specifications. R2 reported in the CPFE estimations is the within R2. Clustered standard errors (at the country-pair level) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

While the joint significance of inflow and outflow measures suggests that it is possible to work at either—or both—ends of the flows, it is important to recognize that the estimated effects are not fully additive. By construction, the log specification implies that the combined impact of simultaneous inflow and outflow restrictions will be less than the sum of the individual implied effects. For example, in Table 4 (col. [1]), moving from the 25th to the 75th percentile on the source country overall outflow controls index (while holding the recipient country inflow index constant) is associated with a reduction in flows by about 80 percent, while a similar increase for the recipient country bond inflow controls index (while holding the source country outflow index constant) is associated with a reduction in flows by about 50 percent. The estimated impact of both measures together, however, is a 63 percent reduction in flows. That the joint impact of the measures may not be fully additive is plausible since in practice there may be some overlap in the flows that the source and recipient country restrictions attempt to target.

To explore this issue further, Table 5 re-estimates the impact of source and recipient country measures by segmenting the data according to whether the other end is relatively more or less open (overall controls index is below or above the 75th percentile). The top panel of Table 5 shows that outflow related measures generally have a quantitatively larger and statistically significant impact when the recipient country is more open to inflows (cols. [1]-[8]) than when it is already mostly closed (cols. [9]-[16]). Likewise, the bottom panel of Table 5 shows that inflow related CARs have a quantitatively larger and statistically significant impact when the source country is more open to outflows, than when it is mostly closed. The impact of inflow or outflow restrictions is thus mostly statistically insignificant once the other side is already relatively closed. Inasmuch as controls are effective in reducing outflows/inflows, this finding makes intuitive sense since a closed capital account at one end implies that the incremental impact of restrictions at the other end will be smaller (also implying a lesser need for such restrictions).

Table 5.

Cross-Border Bank Flows and CARs by Openness, 1995–2012

article image
Note: Dependent variable is (log) bank asset flows from country i to j. More (less) open recipient countries are those with below (above) 75th percentile overall inflow controls index. More (less) open source countries are those with below (above) 75th percentile overall outflow controls index. All specifications include control variables as in Tables 2-4, country-pair and year effects, and constant. Clustered standard errors (by country-pair) are reported in parentheses. ***, ** and * indicate significance at the 1, 5, and 10 percent levels, respectively.

Importantly, however, the results do not imply that inflow and outflow restrictions are mutually redundant (so that they should not operate on both ends). Rather, the results remain similar if we exclude the fully open recipient countries (i.e., those with no inflow restrictions) when estimating the effect of outflow measures, and exclude fully open source countries (i.e., those with no outflow restrictions) when estimating the effect of inflow measures (Table 6). While the number of observations declines in doing so, the estimated impact and statistical significance of the different measures are barely affected. Effectiveness of inflow or outflow restrictions is thus not dependant on the other side of the transaction being fully open—only on the other side not being fully closed. 27

Table 6.

Cross-Border Bank Flows and CARs by Openness—Excluding Fully Open, 1995–2012

article image
Note: Dependent variable is (log) bank asset flows from country i to j. More (less) open recipient countries are those with below (above) 75th percentile overall inflow controls index. More (less) open source countries are those with below (above) 75th percentile overall outflow controls index. From both more open recipient and source countries, those with full openness (i.e., no inflow and outflow controls, respectively) are excluded. All specifications include control variables as in Tables 2-4, country-pair and year effects, and constant. Clustered standard errors (by country-pair) are reported in parentheses.***,** and * indicate significance at the 1, 5, and 10 percent levels, respectively.

The upshot of these findings is that it should be possible to influence the volume of cross-border bank flows through outflow or inflow restrictions, or by some combination of both. Imposing measures on both sides allows a further reduction in flows, or possibly the same reduction but achieved by using the measures less intensively. If, as could be the case, the costs associated with the imposition of CARs are convex in the “tax” rate, then it may be globally more efficient to use a combination of low outflow and inflow restrictions than to put the full burden at either end (Ostry et al., 2012). Conversely, when it is not possible to operate at one end—either because the administrative capacity is lacking or because international treaty obligations prohibit the use of such restrictions—then it would nevertheless still be possible to reduce flows by imposing relevant CARs at the other end.

A. Counterfactual analysis

To illustrate more clearly the implications of the results obtained above, we simulate some counterfactual scenarios of the level of pre-GFC flows to various regions under greater use of CARs in both source and recipient countries. We do so by using the estimates reported in Table 4 (col. [6]) of financial credit outflow controls in source countries, and of restrictions on lending locally in FX in recipient countries, and considering three different scenarios. Specifically, we assess the change in (predicted) flows if (i) all source countries had financial credit outflow controls in place, while holding everything else, including recipient country inflow restrictions, at the actual 2007 level; (ii) all recipient countries had restrictions on lending locally in FX in place, while holding everything else, including source country outflow restrictions, at the actual 2007 level; and (iii) all source and recipient countries had imposed financial credit outflow controls and FX lending restrictions, respectively, holding other variables at the 2007 level.

The simulations suggest that flows to emerging Europe in 2007 would have been about 20 percent lower if all source countries (largely other European countries) had financial credit outflow controls in place; about 80 percent lower if all recipient countries had imposed restrictions on lending locally in FX; and about 85 percent lower if all source and recipient countries had imposed the financial credit control and FX lending restriction, respectively (Figure 5[a]). Similar estimates are obtained for the Eurozone peripheral countries (Greece, Ireland, Italy, Portugal and Spain), where it is estimated that flows might have been almost 95 percent lower if all source and recipient countries had imposed financial credit outflow controls and FX lending restrictions in 2007, respectively.

Figure 5.
Figure 5.

Potential Impact of CARs on Cross-Border Bank Flows

(in percent)

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Note: Figure obtained using estimates reported in Table 4 (col. [6]) and summing predicted values for all countries in the identified region. Left and right hand panels show the change in predicted flows (in percent) if all source countries imposed a financial credit outflow control; if all recipient countries imposed restriction on FX lending; and if all source and recipient countries imposed these measures together in 2007 and 2010, respectively.

The effect of all source countries imposing the financial credit outflow control (compared to the actual level of restrictions in 2007) comes out to be smaller for Latin America, as a few relevant source countries already had some financial credit outflow restriction in place. (Of course, more intensive restrictions in source countries would imply a larger effect, but that cannot be captured here because of the binary nature of our CARs.) The impact of imposing FX lending restrictions in the recipient country is however substantial for the region, and lowers inflows by some 80 percent. By contrast, for Asia, the effect of action by all recipient countries comes out to be smaller since most of these countries already had some type of restriction on local FX lending in 2007—but that by source countries comes out to be relatively larger (reducing flows by over 30 percent). Looking at the effectiveness of measures in controlling the post-GFC surge in flows to Asian and Latin American EMs, a similar picture emerges, and we find that action by all source and recipient countries would have lowered inflows to these regions by about 10 and 90 percent, respectively (Figure 5[b]).

B. Spillover Effects

If recipient country restrictions are effective in reducing the volume of cross-border bank inflows—as the results above suggest—then they could (inadvertently) divert flows to other more open countries in the region. While such diverted flows may finance productive investments (and could thus be welcome), they could also exacerbate existing distortions in countries already contending with an inflow surge. If the latter respond by imposing or intensifying their own inflow restrictions, then a costly “capital control war” may ensue.28

To examine the possibility of spillovers from inflow related CARs, we extend (3) to include a measure of inflow controls in other relatively similar countries, which could be likely substitutes from the perspective of foreign investors. We define similarity in two ways: based largely on regional characteristics such that EMs in the same geographic region are grouped together (while advanced countries are considered as one group); and based on economic characteristics such that countries with similar per capita income, real GDP growth rate and institutional quality are grouped together. The first grouping is time invariant; whereas in the second case, we capture the dynamic nature of economic characteristics by grouping countries over three sub-periods (1995–99, 2000–04, and 2005–12). Specifixcally, to construct the second grouping, we first take the average of the relevant variables for each country over the sub-periods, and then apply the k-means clustering method to form three (relatively similar) groups for each sub-period.29

Once the groupings (reported in Table A4) are obtained, we compute the distance-weighted average of overall capital inflow controls in the regional and economic “neighbors” of each recipient country in the sample, and use these composite measures (lagged one period) to examine the possibility of capital flow deflection within groups. The estimation results reported in Table 7 show the existence of strong spillover effects from the imposition of inflow controls in recipient countries on their neighbors, irrespective of which method is used to define neighbors. Cols. [1]-[7], for instance, show a significantly positive estimated coefficient of the regional inflow control measure—implying that raising average inflow controls in regional neighbors from the median to the 75th percentile increases bank flows to the recipient country by, on average, about 80 percent. Similar results are obtained in cols. [8]-[14], which show that an increase in inflow controls in countries with similar economic characteristics leads to a significant increase in capital flows to the recipient country.30

Table 7.

Spillover Effects of CARs, 1995–2012

article image
Note: Dependent variable is (log of) bank asset flows from country i to j. Source and recipient country control variables (lagged one period) as in Tables 2-4 and constant are included in all specifications. Clustered standard errors (at the country-pair level) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. a/ In cols. (1)-(7), inflow controls in neighbors is defined as a distance-weighted average of the overall capital inflow controls index of countries in the region of the recipient country. See Table A4 for the country groupings. b/ In cols. (8)-(14), inflow controls in neighbors is defined as a distance-weighted average of the overall capital inflow controls index of countries with similar economic (per capita real income and real GDP growth) and institutional characteristics as of the recipient country. See Table Ax for the country groupings.

Including inflow controls in neighbors also does not affect much the estimated coefficient of the recipient country CARs, suggesting that the latter’s effect is largely independent of restrictions in other countries. These results remain similar if instead of distance-weighted measures, we use GDP-weighted (or GDP- and distance-weighted) measures of inflow control in neighbors; if we extend the economic characteristics to account for commodity exporters; or if we define economic neighbors by considering attributes annually (instead of period averages). Given the magnitude and volatility of cross-border bank flows, and the potential spillovers from CARs in other recipient countries implied by these results, the possibility of countries imposing beggar-thy-neighbor CARs cannot be ruled out—which strengthens the case for multilateral coordination among recipient countries as well.31

C. Sensitivity Analysis

To examine the robustness of the results obtained above, we conduct a battery of sensitivity checks. These include estimating alternate specifications of the benchmark models with additional control variables and different samples, using alternate formulations of the dependent and CAR variables, and addressing potential endogeneity concerns through the two-stage least squares estimation.

Alternate specifications

Table 8 shows that the benchmark results are reassuringly robust to the inclusion of other variables in the model that could potentially affect the volume of cross-border bank flows. For example, controlling for source and recipient country institutional quality; financial development (proxied alternately by stock market capitalization, private credit, and deposit money bank assets to GDP ratios); financial soundness (proxied by bank return on assets and on equity); bank concentration and stability (proxied by the fraction of assets held by the three largest commercial banks in the country and bank z-score, respectively); and contagion effects through a common lender (i.e., exposure of source country to other countries experiencing a financial crisis; Van Rijckeghem and Weder, 2003; Hermann and Mihaljek, 2010), we find that the estimated coefficients of CARs remain mostly similar to those in Tables 2 and 3 in both magnitude and statistical significance (cols. [2]-[6]). Taking into account the volume of bilateral trade between the country-pair, and their direct exchange rate relationship against each other (as in Ghosh et al., 2014b) does not affect the results much either.32 The results also carry through if we control more generally for recipient country time-varying characteristics when estimating the impact of outflow related measures by including recipient country-year fixed effects (and vice versa for inflow measures; col. [8]).33

Table 8.

Cross-Border Bank Flows and CARs: Robustness Analysis

article image
Note: Table presents robustness for benchmark models (1) and (2). Col. [2], for example, presents the results of CARs when (source and recipient country) institutional quality is added to the benchmark specifications. Dependent variable is (log of) bank asset flows from country i to j unless otherwise stated. All specifications include control variables listed in Tables 2-4, and country-pair and year effects. Sample size varies across specifications based on data availability for the variables. Standard errors are clustered at country-pair level in all specifications. *, **, and *** indicate statistical significance at 10, 5, and 1percent levels, respectively.

Institutional quality for source and recipient countries is added to the benchmark specification (reported in Tables 2 and 3).

Bilateral exchange rate regime and bilateral (log) real trade between country pair is added to the benchmark specification.

Stock market capitalization for source/recipient countries is added to the benchmark specification. Results remain similar if private credit, or deposit money bank assets (to GDP) are used as financial development indicators.

Return on assets and equity variables for source and recipient countries are added to the benchmark specification.

Measures for bank stability (z-score) and bank concentration for source and recipient countries are added to the benchmark specification.

Common lender variable is added to the benchmark specification.

Source (recipient) country-year effects are added in the estimations with inflow (outflow) related CARs.

Sample excludes EM E source countries.

Sample excludes advanced recipient countries.

Sample excludes those source and recipient countries which are off-shore financial centers.

Estimated sample is restricted up to 2007.

Dependent variable is (log) real flows from source country i to recipient country j.

Dependent variable is (log) total stock of bank assets (amount outstanding) of source country in recipient country.

Dependent variable is (log) flows to the nonbank sector in recipient country.

Alternate capital controls indices with mild restrictions treated as zero are used.

In addition, while the inclusion of EMs as source countries permits greater variation in our CAR variables, the results remain similar if we restrict the source countries to advanced countries only (col. [9]). They are also largely unaffected if we restrict the recipient country sample to EMs only, and exclude offshore financial centers (both advanced and EM) or post-GFC years from the sample when international bank deleveraging occurred (cols. [10]-[12]).

Alternate dependent variables

Defining the dependent variable in (log) real terms (deflated by US CPI) does not have much impact on the results (col. [13]), nor does using data on total cross-border stock of bank assets instead of flows (col. [14]). In fact, the latter strengthens the results in most cases: the estimated coefficients of almost all inflow related measures (including equity and financial credit controls, as well as restrictions on purchasing locally issued securities in FX, which were statistically insignificant above) become significantly negative. Moving from the 25th to the 75th percentile on outflow related measures reduces the bilateral stock of bank assets by some 12–15 percent. Similarly, moving from the 25th to the 75th percentile on inflow related measures reduces the bilateral stock of bank assets by about 8–23 percent.

The benchmark results reported in Tables 2 and 3 pertain to total bank asset flows from source countries, comprising flows to the banking as well as nonbanking sectors in the recipient countries. While the former typically dominate the latter, CARs (and other variables) could potentially have a differential effect on the two types of flows. Thus, Blank and Buch (2010) find that banks’ cross-border assets against banks respond more to financial variables (like interest rate differential), and less to real variables, than those against nonbanks. Similarly, prudential measures (that primarily target banks) in recipient countries may have less influence on inflows to the nonbanking sector than to the banking sector. To examine whether this is the case, we re-estimate (1) and (2) taking the (log) flows to the nonbanking sector as the dependent variable. The sample size drops by about one-fifth in these estimations due to lack of data availability; the estimation results however do suggest some differences. As might be expected, financial sector specific measures are not significantly related with flows to the nonbanking sector—on the source side, equity and financial credit outflow controls appear to somewhat restrain lending to the nonbanking sector (col. [15]), while on the recipient side, controls on bond inflows are associated with lower inflows.

Alternate CAR measures

As noted above, our CAR measures are based on the existence of a restriction, with no differentiation by their intensity. While this is unavoidable considering the nature of available information, some restrictions are less likely to be material or binding, with a correspondingly lower impact on flows. As a robustness test, here we construct some alternative CAR indices, where relatively mild regulations such as registration requirements or restrictions on investments in only a few selected sectors for national security purposes are considered as no restrictions. The results are similar to those reported above. For example, Table 8 (col. [16]) shows that with the alternate measures, the estimated coefficients of all outflow related CAR measures are almost the same as those reported in Table 2. On the inflow side, the results for overall and bond inflow controls remain similar to those reported in Table 3, while the estimated coefficient for direct investment inflow controls now also becomes statistically significant (at the 10 percent level).

Endogeneity

An important concern when estimating the effect of CARs on capital flows is that of reverse causality—i.e., countries may strengthen CARs in response to a surge in capital inflows (or outflows).34 Since we focus on bilateral components of total capital flows (whereas the imposition of CARs tend to be in response to the aggregate volume of flows)—unless bilateral flows are perfectly correlated across country pairs—reverse causality is less likely to be a concern in our case than it is in other studies which consider the total volume of flows. Moreover, to the extent that there is any such endogeneity, it is likely to reduce the estimated coefficients of capital controls and prudential measures. The strong findings above on both source and recipient country restrictions are therefore despite, rather than because of, any potential endogeneity (which would tend to bias the results toward finding no effect).

Nevertheless, following earlier studies, we use the first lag of CARs in all of the estimations above to mitigate potential endogeneity concerns. Here we also consider two alternate approaches for robustness purposes. First, instead of a one year lag, we use 5-year lagged average of the measures; second, we apply an instrumental variable two-stage least squares (IV-2SLS) approach. In the first case, we find that the results remain very similar to those reported above (these are therefore not reported here). For the second case, we require at least one valid instrument that is correlated with CARs in (1) and (2), but is not expected to affect the dependent variable directly. We consider two such variables as our potential instruments: monetary (or central bank) independence, and the presence of a democratic left-wing government in country i (or j) in year t. Existing studies (e.g., Grilli and Milesi-Ferretti, 1995) find these variables to be important determinants of capital controls: countries with lower monetary independence and a left-wing government are more likely to implement capital controls. There is, however, no a priori reason to believe that both these variables would be directly related to the amount of cross-border bank flows (especially, since we control for per capita income as a proxy for institutional quality/polity in all specifications).

We obtain information on monetary freedom from the Heritage Foundation’s Economic Freedom Index, which ranks countries on a scale of 0 to 100—with larger values depicting greater monetary independence. A particular advantage of this data is its comprehensive and up to date availability, and cross-country and temporal comparability. Information on the presence of a left-wing government is obtained from Beck et al. (2001), and is summarized as a binary variable (with one indicating a left-wing government, and zero otherwise).

We begin by using only monetary freedom as our instrument since data availability on whether the government is left-wing or center/right-wing is relatively limited. The validity of this instrument is supported by the results from the first stage of the IV-2SLS estimation: the estimated coefficient of monetary freedom is negative and significant (at the 1 percent level) in almost all specifications, indicating the lower prevalence of CARs in countries with greater monetary freedom (Table 9, panel A). The other control variables included in the first stage regression for outflow (inflow) restrictions are those relevant for the source (recipient) countries—such as economic size, real GDP growth rate, real per capita income, real interest rate, exchange rate regime, and current account balance—and are also generally statistically significant (results not shown).35 The F-test of the hypothesis that the estimates in the first stage regression are jointly equal to zero is thus strongly rejected, and the R-squared across specifications is about 0.9, offering evidence on the appropriateness of our instrument and the overall fit of the first stage regression.

Table 9.

Cross-Border Bank Flows and CARs: IV-2SLS Estimates, 1995–2012

article image

Panel A reports the first stage estimation results for outflow and inflow related CARs in cols. (1)-(8) and (9)-(16), respectively, where monetary freedom in source/recipient countries is used as an instrument. Log of real GDP and real GDP per capita, real GDP growth, real interest rate, current account balance, exchange rate regime, lagged CAR, and regional dummies for the source side are included in cols. (1)-(8), while those for the recipients are included in cols. (9)-(16). All regressors are lagged one period. The sample size drops slightly in cols. [9]-[16] because of data unavailability for a few countries for some years. Constant and year effects are included in all specifications. F-test (p-value) reports the joint significance of all regressors. Robust standard errors are reported in parentheses. *** indicates statistical significance at the 1percent level.

Panel B reports the two-stage least squares estimates with (log of) bank asset flows from country i to country j as the dependent variable. CARs are predicted values obtained from the corresponding first stage regression in Panel A. Control variables as specified in Table 4 (lagged real GDP (log), real per capita income (log), real GDP growth rate, and real interest rate of both source and recipient countries; current account balance and exchange rate regime of recipient country; country pair and year effects), as well as constant are included in all specifications. Standard errors computed with jackknife (1,943 replications) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

The results obtained from the second stage of the estimation—using the predicted values of CARs from the first stage regression—are mostly in line with the benchmark results reported in Tables 2 and 3. On the source side, the overall restrictiveness on capital outflows remains significantly important, as do restrictions on bond, equity, direct investment, and financial credit outflows (Table 9, panel B). Among prudential measures, the estimated coefficient of restrictions on lending to nonresidents remains significantly negative. The magnitude of the estimated coefficients implies that moving from the lower to the top quartile on the (predicted) outflow control measures reduces flows by about 50–100 percent. On the recipient side, the second stage results imply a significantly negative effect of overall capital inflow controls, as well as of bond inflow controls, on bank inflows. The estimated coefficients for restrictions on lending locally in FX and open FX position limits also remain significantly negative. Together these estimates imply a reduction in inflows by some 50–85 percent if we move from the lower to the top quartile of the (predicted) measures.

The second stage estimation results hold if we control for institutional quality/polity or financial development of source and recipient countries, in addition to including income per capita. The results also remain similar when we include the existence of left-wing government as an instrument in the first stage estimations, where, as expected, the variable itself is generally positively and significantly associated with the existence of capital controls (Table A5, panel A). In fact, we use the availability of the second instrument to further establish the validity of our main monetary freedom instrument by following the “easy-to-interpret” overidentification test proposed by Acemoglu et al. (2001). Specifically, while instrumenting CARs with the left-wing government variable in the first stage, we add the monetary freedom variable as an exogenous regressor in the second stage. If monetary freedom has a direct effect on cross-border bank flows, we would expect it to be statistically significant in the second stage. By contrast, in all cases, we find the estimated coefficient of the monetary freedom variable to be wholly statistically insignificant (Table A5, panel B), which supports its excludability from the second stage estimation, and confirms that its impact on cross-border bank flows likely works through capital controls.

Overall, these findings support the robustness of our results to potential endogeneity bias, and suggest that CARs—both capital controls and prudential measures—in source and recipient countries can play an important role in moderating large cross-border bank flows.

V. Conclusion

In this paper, we examine whether cross-border capital flows can be regulated by imposing capital account restrictions at both (source and recipient country) ends, as originally advocated by the main architects of the Bretton Woods system, John Maynard Keynes and Harry Dexter White. To this end, we use data on bilateral cross-border bank flows from 31 major source to 76 major recipient countries—both advanced and EMs—over 1995–2012, and combine this information with a comprehensive dataset on different types of capital controls and prudential measures (that can act like capital controls) in both source and recipient countries.

Our findings suggest that CARs at both ends can significantly influence the volume of cross-border bank flows. On the source side, restrictions on capital outflows—specifically, bond, equity, direct investment and financial credit outflow controls, and restrictions on the financial sector to lend to nonresidents—are associated with significantly smaller flows. On the recipient side, controls on bond inflows, as well as FX-related prudential measures such as restrictions on local FX lending and open FX position limits, are associated with significantly smaller inflows. Controlling simultaneously for both source and recipient country CARs, their estimated impact remain largely unchanged. While the effects of simultaneous inflow and outflow restrictions may not be fully additive, it is possible to operate at both ends of the flow and achieve either a larger reduction in the volume of flows, or the same reduction with less intensive measures at either end. The results also point to cross-border spillovers from inflow controls such that a country is likely to receive significantly larger inflows when its (regional or economic) neighbors are financially relatively closed.

These findings suggest that there may be scope for greater international cooperation in managing large and volatile cross-border flows. Where administrative capacity and treaty obligations permit, tackling flows at both the source and receiving ends can result in globally more efficient outcomes if the cost of imposing restrictions is convex (as seems plausible). Likewise, coordination among recipient countries could help prevent costly “capital control wars” in the presence of cross-border spillovers from measures in recipient countries. While the results presented here imply that capital account related measures could be effective in better regulating cross-border bank flows, further research is warranted to fine tune these measures to better differentiate their intensity across countries, and analyze more precisely how they may interact and impact other types of flows; the extent of spillover effects across types of measures; and how international cooperation may be put into effect.

References

  • Acemoglu, D., S. Johnson, and J. Robinson, 2001, “The Colonial Origins of Comparative Development: An Empirical Investigation,” American Economic Review, 91(5): 1369–1401.

    • Search Google Scholar
    • Export Citation

  • Beck, T., G. Clarke, A. Groff, P. Keefer, and P. Walsh, 2001, “New Tools in Comparative Political Economy: The Database of Political Institutions,” World Bank Economic Review, 15(1), pp. 165–176.

    • Search Google Scholar
    • Export Citation

  • Brunnermeier, M., J. Gregorio, B. Eichengreen, M. El-Erian, A. Fraga, T. Ito, P. Lane, J. Pisani-Ferry, E. Prasad, R. Rajan, M. Ramos, H. Rey, D. Rodrik, K. Rogoff, H. Shin, A. Velasco, B. Mauro, and Y. Yu, 2012, Banks and Cross-Border Capital Flows: Policy Challenges and Regulatory Responses (Washington DC: Brookings Institution).

    • Search Google Scholar
    • Export Citation

  • Blank, S., and C. Buch, 2007, “The Euro and Cross-Border Banking: Evidence from Bilateral Data,” Comparative Economic Studies, 49(3): 389–410.

    • Search Google Scholar
    • Export Citation

  • Blank, S., and C. Buch, 2010, “International Bank Portfolios: Short- and Long-Run Responses to Macroeconomic Conditions,” Review of International Economics, 18(2): 286–306.

    • Search Google Scholar
    • Export Citation

  • Bruno, V., and H. Shin, 2013, “Capital Flows, Cross-Border Banking and Global Liquidity,” NBER Working Paper 19038 (Cambridge, MA: NBER).

    • Search Google Scholar
    • Export Citation

  • Buch, M. 2003, “Information or Regulation: What Drives the International Activities of Commercial Banks?” Journal of Money Credit and Banking, 35(6): 851–869.

    • Search Google Scholar
    • Export Citation

  • BIS, 2009, Guide to the International Financial Statistics (Basel: Bank for International Settlements).

  • Binici, M., M. Hutchison, and M. Schindler, 2010, “Controlling Capital? Legal Restrictions and the Asset Composition of International Financial Flows,” Journal of International Money and Finance, 29(4), pp. 666–684.

    • Search Google Scholar
    • Export Citation

  • Dell’Ariccia, G., J. Di Giovanni, A. Faria, A. Kose, P. Mauro, J. Ostry, M. Schindler, M. Terrones, 2008, “Reaping the Benefits of Financial Globalization,” IMF Occasional Paper 264 (Washington DC: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Forbes, K., M. Fratzscher, T. Kostka, and R. Straub, 2012, “Bubble Thy Neighbor: Portfolio Effects and Externalities from Capital Controls,” NBER Working Paper No. 18052 (Cambridge, MA: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

  • Forbes, K., M. Fratzscher, R. Straub, 2013, “Capital Controls and Macroprudential Measures: What Are They Good For?” DIW Discussion Paper 1343 (Berlin: DIW).

    • Search Google Scholar
    • Export Citation

  • Ghosh, S., and H. Wolf, 2000, “Is There a Curse of Location? Spatial Determinants of Capital Flows to Emerging Markets,” in S. Edwards (ed.), Capital Flows and the Emerging Economies (Chicago: University of Chicago Press), pp 137–158.

    • Search Google Scholar
    • Export Citation

  • Ghosh, S., N. Sugawara, and J. Zalduendo, 2012, “Banking Flows and Financial Crisis: Financial Interconnectedness and Basel III Effects,” Journal of International Commerce, Economics, and Policy, 3(1).

    • Search Google Scholar
    • Export Citation

  • Ghosh, A., M. Qureshi, J. Kim, and J. Zalduendo, 2014a, “Surges,” Journal of International Economics, 92(2): 266–285.

  • Ghosh, A., M. Qureshi, C. Tsangarides, 2014b, “Friedman Redux: External Adjustment and Exchange Rate Flexibility,” IMF Working Paper WP/14/146 (Washington DC: IMF).

    • Search Google Scholar
    • Export Citation

  • Giordani, P., M. Ruta, H. Weisfeld, and L. Zhu, 2014, “Capital Flow Deflection,” mimeo, International Monetary Fund.

  • Grilli, V., and G.M. Milesi-Ferretti, 1995, “Economic Effects and Structural Determinants of Capital Controls,” IMF Staff Papers, 42(3): 517–551.

    • Search Google Scholar
    • Export Citation

  • Helleiner, E., 1994, States and the Reemergence of Global Finance (Ithaca, NY: Cornell University Press).

  • Herrmann, S., and D. Mihaljek, 2010, “The Determinants of Cross-Border Bank Flows to Emerging Markets: New Empirical Evidence on the Spread of Financial Crises,” BIS Working Paper No. 315 (Basel: Bank for International Settlements).

    • Search Google Scholar
    • Export Citation

  • Horsefield, J., 1969, The International Monetary Fund 1945–65: Twenty Years of International Monetary Cooperation—Volume 1 (Washington DC: IMF).

    • Search Google Scholar
    • Export Citation

  • IMF, 2011a, Recent Experiences in Managing Capital Inflows—Cross-Cutting Themes and Possible Policy Framework (Washington DC: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • IMF, 2011b, The Multilateral Aspects of Policies Affecting Capital Flows (Washington DC: International Monetary Fund).

  • IMF, 2012, The Liberalization and Management of Capital Flows: An Institutional View (Washington DC: International Monetary Fund).

  • Jeanne, O., and A. Korinek, 2010, “Managing Credit Booms and Busts: A Pigouvian Taxation Approach,” NBER Working Paper 16377 (Cambridge MA: NBER).

    • Search Google Scholar
    • Export Citation

  • Korinek, A., 2010, “Regulating Capital Flows to Emerging Markets: An Externality View,” mimeo, University of Maryland.

  • Lambert, F., J. Ramos-Tallada, and C. Rebillard, 2011, “Capital Controls and Spillover Effects: Evidence from Latin American Countries,” Banque de France Working Paper No. 357 (Paris: Banque de France)

    • Search Google Scholar
    • Export Citation

  • Magud, N., C. Reinhart, and K. Rogoff, 2011, “Capital Controls: Myth and Reality—A Portfolio Balance Approach,” NBER Working Paper w16805 (Cambridge, MA: NBER).

    • Search Google Scholar
    • Export Citation

  • Milesi-Ferretti, G., and C. Tille, 2011, “The Great Retrenchment: International Capital Flows during the Global Financial Crisis,” Economic Policy, 26(66): 289–346.

    • Search Google Scholar
    • Export Citation

  • Portes, R., and H. Rey, 2005, “The Determinants of Cross-Border Equity Flows,” Journal of International Economics, 65(2): 269–296.

    • Search Google Scholar
    • Export Citation

  • Ostry, J., A. Ghosh, K. Habermeier, M. Chamon, M. Qureshi, D. Reinhardt, 2010, “Capital Inflows: The Role of Controls,” IMF Staff Position Note SPN/10/04 (Washington DC: IMF).

    • Search Google Scholar
    • Export Citation

  • Ostry, J., A. Ghosh, K. Habermeier, L. Laeven, M. Chamon, M. Qureshi, A. Kokenyne, 2011, “Managing Capital Flows: What Tools to Use?,” IMF Staff Discussion Note SDN/11/06 (Washington DC: IMF).

    • Search Google Scholar
    • Export Citation

  • Ostry, J., A. Ghosh, and A. Korinek, 2012a, “Multilateral Aspects of Managing the Capital Account,” IMF Staff Discussion Note SDN/12/10 (Washington DC: IMF).

    • Search Google Scholar
    • Export Citation

  • Ostry, J., A. Ghosh, M. Chamon, and M. Qureshi, 2012b, “Tools for Managing Financial-Stability Risks from Capital Inflows,” Journal of International Economics, 88(2): 407–421.

    • Search Google Scholar
    • Export Citation

  • Schindler, M., 2009, “Measuring Financial Integration: A New Data Set,” IMF Staff Papers, 56(1): 222–238.

  • Van Rijckeghem, C., and B. Weder, 2003, “Spillovers through Banking Centers: A Panel Data Analysis of Bank Flows,” Journal of International Money and Finance, 22(4): 483–509.

    • Search Google Scholar
    • Export Citation

  • Wooldridge, J., 2010, Econometric Analysis of Cross Section and Panel Data (Cambridge, MA: MIT Press).

Appendix: Data and Summary Statistics

Table A1.

List of Countries in the Sample

article image
Note: Years in parentheses reflect years of data availability for reporting countries in our sample. The sample of emerging markets is based on those included in the IMF’s Vulnerability Exercise for Emerging Markets.
Table A2.

Data Description and Sources

article image
Table A3.

Summary Statistics of Selected Variables

article image
Notes: Based on the estimated sample of the benchmark specification. DI=Direct investment; FC=Financial credit.
Table A4.

Recipient Country Groupings by Region and Economic Characteristics

article image
Note: Advanced countries and regional groupings are consistent with the IMF’s WEO. Groupings based on economic characteristics (real income per capita, real GDP growth, and institutional quality) are obtained by taking the average of the characteristics over the sub-period, and then applying the k-means clustering approach. Countries not included (Georgia, Macedonia, and Malta (1995–99)) are those for which data on institutional quality (proxied by the ICRG index) is unavailable.
Table A5.

IV-2SLS Estimation: Left-Wing Government as Instrument

article image

Panel A reports the first stage estimation results for outflow and inflow related CARs in cols. (1)-(8) and (9)-(16), respectively, where the presence of left-wing government in source/recipient countries is used as an instrument. Log of real GDP and real GDP per capita, real GDP growth, real interest rate, current account balance, exchange rate regime, lagged CAR, and regional dummies for the source side are included in cols. (1)-(8), while those for the recipients are included in cols. (9)-(16). All regressors are lagged one period. Constant and year effects are included in all specifications. F-test (p-value) reports the joint significance of all regressors. Robust standard errors are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5 and 10 percent levels, respectively.

Panel B reports the two-stage least squares estimates with (log of) bank asset flows from country i to country j as the dependent variable. CARs are predicted values obtained from the corresponding first stage regression in Panel A. Control variables as specified in Table 4, country-pair/year effects, and constant are included in all specifications. Standard errors computed with jackknife (1,943 replications) are reported in parentheses. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

Figure A1.
Figure A1.

Inflow-Related Capital Controls and Prudential Measures in EMs, 1995–2012

(in percent of observations)

Citation: IMF Working Papers 2014, 188; 10.5089/9781498358743.001.A001

Source: Based on IMF’s AREAER and the OECD Code of Liberalization of Capital Movements (various issues).
†

We are grateful to Olivier Blanchard, Eugenio Cerutti, Stijn Claessens, Kristin Forbes, Sebnem Kalemli-Ozcan, Annamaria Kokenyne, Camelia Minoiu, Jonathan Ostry, IMF colleagues, and participants at the Graduate Institute of Geneva BCC 2nd Annual Research Conference and Bank of England International Finance Seminar for very helpful comments and suggestions, and to the Bank of International Settlements for providing us with locational banking statistics on a bilateral basis. Any errors are our responsibility.

2

The Bretton Woods Agreement enshrined capital controls in the IMF’s Articles of Agreement, giving member countries generally the right to exercise such controls as necessary to regulate international capital flows (as Keynes famously remarked in 1944, “What used to be a heresy is now endorsed as orthodox”; Helleiner, 1994). The right of members is, however, qualified by their obligations subject to IMF surveillance under Article IV.

3

To the extent that source countries are large international creditors, they may also be able to manipulate the terms of trade (i.e., the world interest rate) in their favor by restricting outflows (Ostry et al., 2012a).

4

For instance, in the face of increasing speculative capital movement in the early 1970s, both Japan and Western Europe suggested introducing cooperative controls to preserve the stable system of exchange rates. The proposal, however, did not win US backing (Helleiner, 1994). Even in cases where capital controls have been introduced congruently by countries to deal with large capital flows—e.g., the US’ interest equalization tax to stem outflows in 1963–74, and Germany’s capital controls program over 1968–73 to reduce inflows, mostly originating from the US (in search of high yield)—the actions have not been explicitly coordinated.

5

Ostry et al. (2010) and Magud et al. (2011) survey the literature on the effectiveness of capital controls. Several papers (e.g., Korinek, 2010; Jeanne and Korinek, 2010) theoretically make a case for using capital controls for prudential purposes to increase social welfare.

6

The literature on capital outflow controls has mostly focused on their effectiveness in stemming capital flight during crises (Magud et al., 2011). Binici et al. (2010) examine their effectiveness more generally, and find that they are associated with significantly lower outflows in advanced countries.

7

An early example of cooperation on capital controls is that between Western Europe and Britain, when to tackle massive capital flight from the former in 1945–47, the two signed bilateral economic agreements with provisions for cooperative control of speculative flows. The cooperation was however unlikely to be successful without US support, which refused to consider inflow controls on West European capital (Helleiner, 1994).

8

To the extent that cross-border bank flows comprise bank intermediated credit flows that rely on short-term wholesale funding, as is typically the case when global liquidity is abundant and risk aversion is low, they pose a higher risk than those funded by stable deposits (Brunnermeier et al., 2012). Milesi-Ferretti and Tille (2011) find that across different types of flows, the retrenchment in banking flows was the largest during the GFC.

9

Ostry et al. (2010) argue that the lack of convincing evidence on the effectiveness of controls in reducing the total volume of capital inflows may partly reflect econometric identification problems. (This is because if countries facing large inflows are also those that impose controls, the resulting bias in econometric estimations will likely imply reduced or no effectiveness of CARs on the magnitude of inflows.)

10

Bilateral data on other types of flows, e.g., portfolio flows is mostly available for advanced countries (Portes and Rey, 2005). More recently, the IMF has initiated coordinated investment surveys that document the total stock of direct and portfolio investment assets/liabilities of reporting countries against major partner countries, but their cross-country and time coverage is limited.

11

Several (administrative and efficiency) costs may be associated with the imposition of controls, which are likely to increase at an increasing rate. For example, the imposition of an inflow tax may generate incentives for circumvention, leading to an increase in its breadth of application. But that, in turn, may restrict desirable inflows that otherwise may not have been restricted.

12

Indeed, much of the Keynes-White impetus for controlling capital flows was to enhance the effectiveness of domestic macroeconomic policies.

13

Several studies focus on the determinants of cross-border bank asset flows (or stocks), and include proxies for financial openness in their analysis (e.g., Blank and Buch, 2010; Ghosh et al., 2012). They however rely on aggregate measures of financial openness, and do not systematically explore the impact of different types of CARs in both source and recipient countries.

14

We use the BIS Locational Statistics by Residence (instead of the consolidated statistics) as these report bank claims based on the residence of the reporting banks and of the counterparties, and capital controls and most prudential measures considered here also apply to resident versus non-resident transactions.

15

The reporting institutions are mostly deposit-taking banks and similar financial institutions. In some countries, specialized non-deposit-taking, trade-related financial entities also report (BIS, 2009). Since the creditor data is reported on residence (rather than nationality) basis, the measurement of flows is consistent with the Balance of Payments Statistics.

16

The detailed data on different types of measures that we use in our analysis is available in the AREAER from 1995 onward, which precludes us from including the pre-1995 years in the sample.

17

Recently, the IMF (2012) has adopted the terminology of capital flow management measures (CFMs) for capital controls and prudential measures that are designed to limit capital flows. The classification of a particular measure as a CFM, however, requires an assessment of the country-specific circumstances and a “judgment as to whether the measure is, in fact, designed to limit capital flows.” To ensure consistency in treatment of measures across countries, and to avoid exercising subjective judgment, we therefore refrain from using the CFM terminology here. Instead, we collectively refer to measures (both capital controls and prudential) that are likely to affect cross-border capital transactions as capital account restrictions (CARs).

18

Our choice of prudential measures is driven by data availability. In some countries with pegged exchange rates, exposure in the anchor currency is excluded from the calculation of the open position. We code such cases as not having limits on open FX positions. The appendix provides a description of variables and data sources.

19

For direct investment, the AREAER provides information on inward and outward restrictions, as well as on the liquidation of direct investment. Following Schindler (2009), we consider controls on direct investment inflows as the maximum of restrictions on inward direct investment and the liquidation of direct investment, which recognizes that liquidation restrictions indirectly impose costs on direct investment inflows.

20

In fact, considering large net capital inflow (or “surge”) observations in EMs over 1995–2011 as identified in Ghosh et al. (2014a), we find that, on average, about 40 percent have been driven predominantly by (net) bank flows as opposed to (net) nonbank flows. Further, the share of bank-flow driven surges has increased over the years—from some 38 percent in 1995–99 to about 43 percent in 2005–11.

21

Following existing literature (e.g., Papaioannou, 2009; Herrmann and Mihaljek, 2010), when taking the log, we transform the negative asset flow observations by taking the log of the absolute value and then changing the sign. This transformation preserves the original sign on the flow observations, and retains symmetry in the data. Unlike bilateral trade flows data, the share of zero observations here is quite small (about 9 percent), so their exclusion from the estimation does not pose any significant issues. The results presented below are however robust to including the zero observations by adding a small constant to all flow values and then taking the log.

22

Formal panel data tests of serial correlation (e.g., Wooldridge, 2010) do not indicate the presence of serial correlation in the errors, lending support to the use of lagged variables in (1) and (2).

23

In the CPFE estimations, inference about the association between cross-border flows and CARs is derived from the time series variation in the latter, since all cross-country variation is absorbed by the CPFE.

24

The R-squared statistic—indicating the goodness of fit of the model—is consistent with those obtained in other studies examining the determinants of cross-border bank flows (e.g., Papaioannou, 2009; Herrmann and Mihaljek, 2010; Ghosh et al., 2012). Studies generally obtain a higher R-squared when estimating the bilateral cross-border stock (rather than flow) of bank assets as a function of similar explanatory variables (e.g., Buch, 2003; Blank and Buch, 2007, 2010). This is also true when we estimate the model using the stock of bank assets (amount outstanding) as the dependent variable in the robustness analysis below (the obtained R-squared is then about 0.7).

25

While loans (targeted directly by financial credit controls) tend to be the dominant component of cross-border bank flows, restrictions on direct investment flows, by limiting the establishment of branches/subsidiaries abroad, could indirectly affect loans by restricting intrabank transactions.

26

Binici et al. (2010) find a similar result that outflow controls on direct investment/equity (and in their case also debt) flows have a statistically much stronger impact than inflow controls. Using aggregate measures of capital account openness, Hermann and Mihaljek (2010) and Ghosh et al. (2012) find that lower financial openness in EMs significantly reduces bank inflows.

27

We draw similar implications if instead of splitting the sample by openness, we include an interaction term between the measures and a dummy variable indicating if the other side is more open/closed (where the dummy variable is equal to one if the other side is more closed, and zero otherwise).

28

Existing empirical evidence on such spillover effects is limited and mixed. Lambert et al. (2011), for example, find that the increase in Brazil’s tax on portfolio bond inflows during 2009–11 raised portfolio equity and bond flows to other Latin American countries, but the effect was short-lived. Looking across EMs, Forbes at al. (2012) find that the average spillover effect of Brazilian controls was small—with some countries experiencing a rise in inflows, while the others experiencing a decline. IMF (2011b) reports similar findings that CARs have occasionally increased or decreased flows to neighboring countries. By contrast, Giordani et al. (2014) find that inflow restrictions have significant spillover effects as they deflect capital flows to countries with similar economic characteristics.

29

We rely on k-means clustering to avoid imposing ad hoc thresholds to define similarity across characteristics. The k-means clustering method groups observations such that the within-cluster sum of squared differences from the mean is minimized (and the between-cluster difference in means is maximized). As a result, each observation belongs to the cluster with the nearest mean, and clusters comprise statistically similar observations.

30

The sample size drops in cols. [8]-[14] as data on institutional quality is unavailable for two countries. Retaining the full benchmark sample by grouping countries just based on income per capita and growth rate, however, yields similar results.

31

Ostry et al. (2012a) argue that coordination among recipient countries implies that in the presence of generalized flows to EMs, the response should involve less intensive controls than would be the case if flows were going to one or a few recipient countries only.

32

The results remain similar if other variables such as source/recipient country (log) population; fiscal balance, external debt and foreign reserves (to GDP) ratios; polity index; and a measure of real exchange rate overvaluation are included in the model. The estimated coefficients for the additional control variables are generally in line with those reported in earlier studies: e.g., recipient countries with better institutional quality, greater financial development and soundness, higher reserves and fiscal balance, and lower external debt attract more flows. The estimated coefficients for bilateral trade flows, bank stability and concentration, and common lender effects are, however, statistically insignificant (detailed results are available upon request).

33

We cannot control for time-varying characteristics of both source and recipient countries simultaneously by including source and recipient country-year fixed effects together since then the effect of our CAR measures—which vary by country-year—would not be identified. To consider potential nonlinearities in the effect of CARs, we interact these measures with economic size and financial development of source/recipient countries but find the interaction terms to be statistically insignificant.

34

Another potential source of endogeneity arises if capital controls and cross-border bank flows are determined by some omitted third factor (e.g., institutional quality). We, however, control for such endogeneity bias by including country-pair fixed effects in the benchmark estimations (which capture the effect of time-invariant and slow moving factors), as well as by augmenting the benchmark specification with a range of variables in the sensitivity analysis above (Table 8, cols. [2]-[8]).

35

We do not include country fixed effects in the first stage because our instruments are slow moving variables, but include region-specific and year effects, as well as the first lag of CARs to capture their persistence.

Same Series

Other IMF Content

Other Publishers

  • Collapse
  • Expand
Cover IMF Working Papers
Regulating Capital Flows at Both Ends: Does it Work?
Author:
Mr. Atish R. Ghosh
,
Miss Mahvash S Qureshi
, and
Naotaka Sugawara
Volume/Issue:
Volume 2014: Issue 188
Publisher:
International Monetary Fund
ISBN:
9781498358743
ISSN:
1018-5941
Pages:
46
DOI:
https://doi.org/10.5089/9781498358743.001
  • Figure 1.
    View in gallery
    Figure 1.

    Cross-Border Bank Asset Flows, 1995–2012

  • Figure 2.
    View in gallery
    Figure 2.

    Cross-Border Bank and Nonbank Liability Flows, 1995–2012

    (in USD tril.)

  • Figure 3.
    View in gallery
    Figure 3.

    Capital Controls and Prudential Measures, 1995–2012

    (in percent of observations)

  • Figure 4.
    View in gallery
    Figure 4.

    Cross-Border Bank Flows, Capital Controls, and Prudential Measures, 1995–2012

  • Figure 5.
    View in gallery
    Figure 5.

    Potential Impact of CARs on Cross-Border Bank Flows

    (in percent)

  • Figure A1.
    View in gallery
    Figure A1.

    Inflow-Related Capital Controls and Prudential Measures in EMs, 1995–2012

    (in percent of observations)

Figure 1.

Cross-Border Bank Asset Flows, 1995–2012
Figure 2.

Cross-Border Bank and Nonbank Liability Flows, 1995–2012

(in USD tril.)
Figure 3.

Capital Controls and Prudential Measures, 1995–2012

(in percent of observations)
Figure 4.

Cross-Border Bank Flows, Capital Controls, and Prudential Measures, 1995–2012
Figure 5.

Potential Impact of CARs on Cross-Border Bank Flows

(in percent)
Figure A1.

Inflow-Related Capital Controls and Prudential Measures in EMs, 1995–2012

(in percent of observations)