Macro-Prudential Policies & Capital Controls, Financial Development and the Interaction Effects on External Debt Liabilities

Macro-Prudential Policies & Capital Controls, Financial Development and the Interaction Effects on External Debt Liabilities 1, a, b Wenwen Sheng Abstract This paper investigates the role of domestic financial system in determining the effectiveness of macro-prudential policies and capital controls (CFMs) in reducing external debt liabilities. Using a panel data for 51 emerging countries over 1995-2008 periods, we find the effectiveness of CFMs, especially of FX-related prudential policies, enhances significantly with banking sector development, and to a lesser extent, with stock market development. Moreover, we claim that FX-related prudential measures can be used as part of capital flow management toolkit only when substantial level of banking sector development has been achieved. The results of this paper have important policy implications: in designing and implementing capital flow management framework to manage risks associated with debt inflows, policy-makers should consider the function of domestic financial sector carefully and tailor toolkit to country circumstances. Keywords: Macro-Prudential Policies; Capital controls; Financial development; Debt liabilities 1 Post-doctoral researcher, a State Administration of Foreign Exchange, b Research Institute, People s Bank of China, Chengfang Street No. 32, Xicheng District, Beijing, China. Email: valen86429@hotmail.com Cellphone: (86) 134 0115 3699

Table of Contents Page Abstract Table of Contents i ii 1. Introduction 1 2. Data and empirical specification 5 2.1 Empirical specification 5 2.2 Data descriptions 6 2.3 Some stylized facts 8 3. Estimation Results 11 3.1 Benchmark results 11 3.2 Robustness: alternative measures of CFMs 15 3.3 Robustness: alternative measures of financial development 16 3.4 Robustness: sub-samples re-estimations 17 3.5 Robustness: cross-section estimations 18 3.6 Further robustness: external debt liabilities as dependent variables 20 3.7 Endogeneity issues 23 4. Conclusion 24 References 25 Appendix 28

Macro-Prudential Policies & Capital Controls, Financial Development and the Interaction Effects on External Debt Liabilities 1. Introduction In past decades, most of emerging market economies (henceforth referred to as EMEs), which historically have restrictive capital account have been going through capital flow liberalization reform. Capital flow liberalization is fundamentally beneficial as it eases financing constraint for domestic projects, fosters the diversification of investment and promotes technology progress. However, liberalization also carries certain financial-stability risks, and one major concern is on the surging debt liabilities as a response to capital account liberalization. Overreliance on foreign debt is detrimental to a country since compared with equity-like flows (FDI and portfolio equity) that are commonly viewed to be stable, and productive to growth, debt flows (consisting of bank loans and bonds) have historically proven more volatile and risky (Sachs, 2006;Tong and Wei, 2010; Alfaro et.al, 2007). When economic environment changes, debt inflows could fly out of the country abruptly, which might further trigger macroeconomic and financial turmoil (Ostry et al., 2011a,b; IMF, 2012a,b). Therefore, excessive reliance on debt liabilities is widely deemed as a more crisis-prone external liability structure. Despite that, as shown in Figure 1, one can see that debt instruments still form the primary source of external liabilities for EMEs. Figure 1. Composition of external liabilities in emerging economies In view of the risks, both researchers and policy-makers have been seeking appropriate policies for solution. In recent decade, International Monetary Fund 1

(henceforth referred to as IMF) introduces the concept of prudential regulations into capital flow toolkits, and refers policies aiming at affecting capital flows collectively as capital flow management measures (henceforth referred to as CFMs). And to the extent that capital flows are the source of systemic financial sector risks, CFMs used to address those risks can be also seen as part of Macro-prudential measures (IMF, 2012b, 2013). CFMs toolkit includes: (1) restrictions on cross-border financial activity that discriminate based on the residency; and (2) macro-prudential measures which are not discriminate on the basis of residency, but relate to cross-border transactions. Prudential policies can be further grouped into foreign currency- (FX-) prudential measures and other domestic macro-prudential measures (IMF, 2011a,b, 2012a,b, 2013). Generally speaking, compared with capital controls that are mainly consisted of administrative restrictions, prudential measures tend to be price-based (or market-based), thereby less discriminatory and more flexible in most cases. Therefore, for EMEs that aiming at liberalizing capital account transactions and meanwhile managing risks, prudential policies should be given primacy from a long-term perspective. Capital flow management measures (CFMs) Residence-based: Capital controls Currency-based: FX-related prudential measures Figure 2. CFMs Toolkit Other domestic macroprudential measures However, till now, the academic and practical understanding of CFMs, especially for the macro-prudential measures remains incomplete. Among the few existing literature, Ostry et al. (2012) has done one of the most outstanding work. By constructing new de jure indices of CFMs (which are based on the IMF s AREAER), Ostry et al. (2012) quantitatively investigate the effectiveness of capital controls and prudential policies in mitigating risks of capital inflows for 51 emerging economies over 1995-2008. The results suggest both capital controls and FX-related prudential measures are able to mitigate the risk of over-reliance on debt liabilities. However, the effects of FX-prudential measures turn out to be rather weak due to the fact that its effect lost significance when considering the effects of capital controls simultaneously. Therefore, the evidence from their work is inadequate to support the effectiveness of macro-prudential measures in reducing foreign debt. In a broader sense, although the systematic studies on CFMs is relatively scant, there is a body of works focusing on the traditional capital controls, from individual country level (e.g. De Gregorio et al., 2000; Edwards and Rigobon, 2009), or multicountry level (e.g. Montiel and Reinhart, 1999; Binici et al., 2010; Kokenyne and Baba, 2011). Magud et al. (2011), Ostry et al. (2011) did great jobs in synthesizing and reviewing these results. According to Magud et al. (2011), Ostry et al. (2011), it is difficult to get an unambiguous conclusion on the merits of capital flow restriction since the effectiveness of capital controls varies across countries, yet, they all mention that there exist some country-specific fundamentals that determine the effects of capital flow restrictions. These characteristics rationalizes why the same policy tools are effective in some countries while are not in other countries, and the effectiveness 2

of capital flow restrictions could be better understood in light of specific country characteristics. To this end, we attempt to identify the domestic circumstances, under which CFMs should be seen as an effective way to tackle the risk of over-reliance on debt. Among various potential factors that might affect the effects of policies, we conjecture domestic financial system is a crucial mechanism linking CFMs and the structure of external liabilities, and three hypothesizes are illustrated to explain why this should be the case. Hypothesis 1. Banking sector expansion is associated with a less reliance on FDI financing, and larger reliance on foreign debt. It is arguably that the banking development may influence the share of debt in total liabilities through two channels. The first channel is the direct impact, which is the development of banking sector increases the cross-border debt inflows. From the standpoint of firms, local firms find it more easily to acquire banking loans abroad in forms of cross-border lending with more active banking sector. And from the standpoint of banks, when bank assets are growing rapidly, especially if the funding required outstrips the growth of the domestic deposit base, the growing credit is often met by capital flows from the international banks and is reflected in the growth of short-term foreign currency-denominated liabilities of the domestic banking system (Shin and Shin, 2011). As a result, debt liabilities tend to increase with the deepening of banking system. For the empirical evidence, Balakrishnan et al. (2012) find that emerging Asia economies that experienced higher credit growth, such as Mainland China, Hong Kong and Korea have been followed by surge in debt flows. The second channel of banking development affecting debt ratio is the indirect impact, which is banking development shifts the financing preference of local firms toward less FDI. Abundant of literature argue that financial market imperfections, as reflected by information asymmetries or poor enforceability of contracts, are important determinant of foreign investment, and FDI is a preferred instrument when these obstacles are present. Based on the corporate financing theories of Meyer (1984) and Myers and Majlauf (1984), Razin et al. (1998), Neumann (2003) extend the pecking order theory to capital flows under an important hypothesis that FDI evolves direct control by investors, and even the ownership in the form portfolio equity is related to some degree of control and thereby information on the investment. Assuming that monitoring costs are decreasing in ownership, it implies that FDI and equity are less costly ways of financing instruments than loans or debt that do not convey some degree of ownership and thereby information. From another standpoint, Albuquerque (2003) argues that the lack of enforceability of claims (as a reflection of inefficiency financial system) in many emerging markets makes portfolio equity and debt less desirable than FDI for foreign investors. Based on these arguments, in a lessdeveloped financial system where information frictions are severe or enforceability is low, local firms may have no other choice but to rely on FDI to carry out an investment project. The theory may also be interpreted to predict that - as local banking sector develops, the role of FDI as a substitute for a well-functioning financial mechanism weakens, and the proportion of debt in total external liabilities is expected to rise. To support this in the real world, a good case in point is Mainland China, where the banking sector has been expanded rapidly in recent decades, has experienced a shifting inflow structure from FDI to banking flows (Balakrishnan et al., 2012). 3

Hypothesis 2. The impact of capital markets on the share of debt in total liabilities is less clear-cut. A series of literature have documented that the role of banking sector and equity markets does not substitute to each other. Demirguc-Kunt and Maksimovic (1998), Levine and Zervos (1998), Beck and Levine (2004) provide evidence that both the operation of banks and the functioning of equity markets exert robust and independent influence on economic activity. Focusing on the external liability structure, the impact of equity market on the share of debt liabilities is more complicated and less clear-cut, compared with banking development. For one hand, the development of financial markets, in terms of both active stock and bond market, provide local firms with more diversified financial instruments to finance their investment in the way of reducing the issuance cost or simplifying issuance procedures. Therefore, portfolio equity and portfolio debt that are intermediated mainly outside the banking system might comprise a much larger share of aggregate flows (Pradhan et al., 2011). Since debt instruments as well as equity-like liabilities tend to rise at the same time, the net effect of financial market development on the share of debt liabilities might be ambiguous. Hypothesis 3. The impact in reducing the debt liabilities for CFMs, especially prudential measures, enhances with financial development, and to a lesser extent with equity market development. It is natural to expect the impact of CFMs in reducing debt liabilities is enhancing with financial development. First, by definition, macro-prudential regulations are a set of prudential measures that are designed to deal with systemic risks in the financial sector. Therefore, as it is put in Ostry et al. (2011a), when capital inflows are largely intermediated through the regulated financial system, prudential tools can be the main instruments, and when inflows bypass regulated markets and institutions (e.g., because domestic entities borrow directly abroad), then prudential regulations will have little traction and capital controls may be the only option. As a result, the impact of CFMs, especially prudential measures in reducing the debt liabilities is critically dependent on local financial system. Furthermore, the enhancement of effectiveness should be greater along with banking development compared with stock market development for two reasons. First, the expanding of banking sector has an inherent tendency to increase debt via crossborder banking transactions, whereas equity market development is not necessarily associated with increasing reliance on debt, thereby the significance of CFMs is supposed to be greater for countries with rapid banking development. Second, and as illustrated above, most of CFMs, especially prudential measures, are directly aiming at regulating activities of financial intermediaries, namely, banks. This is may be another reason why the effects of CFMs enhance more with banking development. However, despite abundant work have identified CFMs and financial development are recognized as two crucial factors in determining external capital structure of countries, pioneering literature tends to study their role separately. To the best of our knowledge, there is hardly any empirical study investigating the interaction effects of these two factors in external liabilities, which, we believe is an important hint to understand why the effectiveness of CFMs varies substantially across countries. Our study aims to fill this gap. Using data from 51 emerging market countries over period 1995 2008, this paper econometrically investigates the nexus between financial development, capital controls and prudential policies and the debt liabilities. Our findings can be summarized as follows: first, we find domestic banking sector development has a tendency to raise the share of debt liabilities, whereas the deepening of capital market development does not necessarily lead to an increase in 4

debt. Second, there is interaction effect between financial development and CFMs in reducing debt liabilities, namely, whether a specific policy tool is effective or not crucially hinges on the function of local financial system. Furthermore, the effectiveness of CFMs in curbing foreign debt enhances more significantly with banking sector development, and to a lesser extent with stock market development. Third, among CFMs toolkit, FX-related prudential measures is the most sensitive to financial development, while economy-wide capital controls seems to be least affected by the status of domestic financial sector. These results are supported by a serious of robustness checks. In particular, our results are not affected by using alternative proxies for capital controls, prudential measures and financial development; by sub-sample regressions according to financial structure; by using cross-sectional estimation; by using debt liabilities as alternative dimension of dependent variables; and by employing IV-2SLS approach to tackle the endogenous problem. The results of our paper provide several important implications. First, the findings highlight that there is certain threshold level of financial development that a country needs to achieve before it can use CFMs to manage debt liabilities. This means when a country design and deploy CFMs to improve its composition of external liabilities, the appropriate framework of CFMs would depend on the specific circumstances of local financial system. Specifically, if a country is lack of wellfunctioning financial markets, strengthening CFMs, especially prudential measures would have very limited effect in reducing foreign debt. Second, when deploying CFMs is inevitable in face of surging debt inflows, our finding provides a reference in choosing between policy tools. Considering that FXrelated prudential measures is more sensitive to financial development, while economy-wide capital controls are least affected by the status of domestic financial sector, capital controls are more appropriate for countries with lower level of financial development. However, for a country with highly developed financial system, FXrelated prudential measures can obtain better results. This statement can be further put in another way, that is, when a country aims to liberalize capital account, it must have a well-functioning financial system first, otherwise a switch from capital controls to macro-prudential measures will leave the country losing control on debt liabilities. This also supports the widely hold opinion that domestic financial development should be seen as pre-requisites for successful liberalization. Third, our paper also shed light on the choice between different modes of financial systems, namely, bank-based financial system and market-based financial system. To certain extent, there is trade-offs between financial structure and capital flow management. That is, for countries with bank-oriented financial system that has an inherent tendency to accumulate more foreign debt, the effectiveness of CFMs in reducing foreign debt is greater. In contrast, debt liabilities tend to accumulate in a slower way in countries with market-oriented financial systems, however, CFMs is playing smaller role there compared with bank-based financial system. From this point of view, market-based financial system might be a better solution for large and systemically important EMEs such as China in the long run, because after all, her financial systems and capital movement regimes will need to evolve to support her increasingly important roles in the global economy, and a sophisticated capital market helps to tackle excessive debt through its own function rather than relying on capital flow restrictions. The remainder of the paper is organized as follows. Section 2 describes the data set and variables. Section 3 presents the estimation results. Section 4 concludes. 5

2. Data and empirical specification 2.1 Empirical specification As indicated above, the main focus in this paper is the interaction effect of financial system and CFMs on composition of external liabilities. In particular, we postulate that capital controls and prudential measures are more effective in countries with highly developed financial market than countries with less-developed financial market. To test for this relation, we interact capital controls and prudential measures with financial variables and regard this as the regressors of key interests. In order to maker sure that the interaction term does not proxy for capital flow management or the level of development of financial markets, both of the latter variables were included in the regressions independently. Since we are interested in the pattern of external liability structure changing over time, and at the same time, to best take advantage of extensive time-series data, this study primarily employs panel data analysis, and a cross-sectional estimation is provided in the later part of paper to further inform the analysis. Accordingly, the basic specification for the empirical analysis can be written as "#$ " = "# " + " " + "# " " " + " + + + " Here, i stands for a country, t presents a time period. is time dummies to account for time-specific effects, controls for unobserved region-specific effects, and is an error term. CFM is explanatory variables of key interest in this paper, which are policy tools restricting capital inflows, and FD represents the other key explanatory variable-financial development. In order to detect the complementary effect between CFMs and financial development, we introduce the interaction term of CFMs and financial development into regressions, which is indicated by "# ". To tackle the potential endogeneity problem, we take one period lag of CFMs and financial development into regressions. Furthermore, " is a set of the other potential factors affecting the share of debt in total external liabilities. Finally, to control for the unobservable effect of country characters and time trend, we also include region and year dummies into the regressions. 2.2 Data Descriptions The data used in our econometric analysis are drawn from various databases. First, we use the share of debt in total external liabilities as our dependent variables, and the data for foreign debt and total liabilities come from External Wealth of Nations Mark II database assembled by Lane and Milesi-Ferretti (2007). Compared with flow data, stock data is less likely to subject to short-term volatility, thus ensuring us to focus on the long-term correlation between key variables. This is in line with most of existing literature investigating external capital structure, for example, Faria et al. (2007, 2009), Ostry et al.(2012). What is worth to point out, the classification of Lane and Milesi-Ferretti (2007) is based on broad financial instruments rather than the functional category. In particular, total liabilities include direct investment, portfolio equity assets, debt, financial derivatives, official reserves and measurement error. Debt is in turn consisted by portfolio debt securities, bank loans and deposits and other debt instruments. Policy tools at policymaker s dispose to restrict capital flow are our key explanatory variable of interest. Here we consider three categories of capital flow management measures: 6

(i) Economy-wide capital controls, which by definition, are measures that restrict capital transactions (or transfers and payments necessary to effect them) by virtue of the residency of the parties to the transactions. Depending on the type of the flow, economy-wide capital controls include controls on debt, equity or direct investment. (ii) Financial sector-specific capital controls, which is provisions that apply only to the financial sector, and that discriminate based on the residency of the parties to the transaction. Toolkits in this category include: impose limits on financial sector borrowing from abroad; (ii) restrict the maintenance of accounts abroad; or (iii) impose differential treatment of accounts held by nonresidents (e.g. setting different reserve/liquid asset requirements, interest rate, or credit controls). (iii) Foreign currency (FX)-related prudential policies, that is, restrictions by virtue of the currency denomination of the capital transaction. It should be pointed out that Ostry et al. (2012) also considers the effect of other prudential measures that are intended reduce systemic risk generally, such as LTV ratio, reserve requirements and limits on credit concentration. However, these prudential policies turn out to have insignificant impact on foreign debt due to its indirect effect on capital flows, as claimed by other related studies, e.g., Balakrishnan et al. (2012). Therefore our study restricts prudential measures to FX-related prudential measures. For financial-sector specific capital controls and FX-related prudential measures, we rely on the data set provided by Ostry et al. (2012), which are assembled based on information contained in the IMF s Annual Report on Exchange Arrangement and Exchange Restrictions (AREAER). For economy-wide capital controls, we employ the de jure measure of capital controls, which is constructed by Fernández et al. (2015). This dataset is building on data first presented in Schindler (2009), and other information from AREAER by IMF. As a robustness check, we also use Chinn and Ito (2008) KAOPEN index for an alternative proxy for economy-wide capital controls. Regarding the other key variable of interest financial development, we closely follow the classic literature on financial markets and growth (e.g. King and Levine, 1993a,b; Beck et al, 2000a, b; Levine et al., 2000). In order to comprehensively reflect the condition of financial markets, four different indicators are used to measure financial development two for the development of banking system, and two for stock markets respectively. For the development of banking system, we use (i) deposit money bank assets to GDP (%) and (ii) private credit by deposit money banks to GDP (%) as two proxies. And for stock markets, we consider (iii) stock market capitalization (%) and (iv) stock market total value traded (%). These four financial variables all come from World Bank Global Financial Development (GFD) database (Beck, 2010), which is originally compiled by Beck, Demirguc-Kunt, and Levine (1999). For other control variables, the choice is closely following Ostry et al. (2012), and they include institutional quality, real GDP per capita, M2 to GDP, and the index of vulnerability. Regarding institutional quality, we rely on political risk rating created by PRS Group, and use the simple average of each subcomponents. To control for the different level of economic development, we add real GDP per capital and M2 to GDP that comes from World Bank WDI database. In relation to indicator of external vulnerability, ideally, we would follow Ostry et al. (2012) by using a composite measure of vulnerability to capital account crisis, obtained from IMF-FSB early warning exercise database. Unfortunately, this dataset is unavailable to the public. As an alternative, we construct another indicator to measure external vulnerability of countries in a way proposed by Ramakrishnan and Juan Zalduendo (2006). At last, in 7

order to control other uncovered country characters and time trend, dummy variables for region and year are also included in the specifications. As a whole, the empirical analysis is built on a sample consisted of 51 countries over the period 1995-2008. A list of country name and descriptive statistics can be found in Appendix. 2.3 Some stylized facts Before the econometric analysis, we use a few simple graphs to illustrate some stylized facts that constitute the point of departure of our analysis. Figure 3. depicts the index for capital controls and prudential measures, as well as the composite index of three types of CFMs imposed by EMEs. Apparently, the attitude of EMEs toward capital flows shows substantial distinctions across regions. Figure 3 Evolution of capital controls and prudential measures by regions Historically, most of EMEs have made significant progress in moving toward greater capital mobility in 1980s. However, Figure 3. clearly captures the fact that after the 1997 financial crisis, capital flow management is somewhat reinforced in Middle East & Africa, as well as East & South Asia. However, for European & Central Asia, there has been a gradual trend toward liberalization of capital flows in past two decades, in terms of both lower degree of capital controls and prudential measures. And for emerging Latin America & Caribbean, economy wide and financial sector-specific capital controls are less frequently used than before, for an alternative, FX-related prudential measures have been on the rise in recent years. In general, all three types of CFMs are commonly used in EMEs. 8

Figure 4. Evolution of financial markets in EMEs, 1995-2008 Figure 4. presents a glance of the evolution of financial system over the last two decades. As shown in Figure 4., most of EMEs experience development in domestic financial markets, as reflected by more active banking sector and stock market. However, there is also substantial variation in financial system among EMEs. For instance, emerging Europe and Central Asia, which historically have under-developed financial system have experienced rapid expansion in private credit over the past decade. In contrast, the financial development of Middle East and Africa is characterized mainly by its rapidly developing stock markets. For Latin America & Caribbean, both banking sector and stock market have been relative less active than other regions and the performance of stock market is to some extent more remarkable than banking sector. And as to East and South Asia, both financial intermediaries and financial markets are better functioning than EMEs from other regions, with banking sector playing a more important role and at the same time stock market is developing substantially. 9

Figure 5. The relation between CFMs and the share of debt liabilities across countries As illustrated in the introduction, we are interested whether the association between CFMs and foreign debt is affected by the function of local financial system. To get a crude guide, we divide the samples into two groups according to the function of financial system and turn to some stylized facts on the relation between CFMs and 10

debt liablities. In particular, we classify samples with the private credit by deposit money bank to GDP above (below) the mean level as banking sector-developed (less developed) countries, and classify samples with market capitalization above (below) the mean level as the stock market-developed (less developed) countries. Figure 5. plots the ratio of debt to total external liabilities on the Y axis against the degree of capital flow management measures (average index of CFMs) on the X axis. As depicted in Figure 5., the effect of CFMs in reducing foreign debt liabilities displays striking difference depending on whether domestic financial system (financial intermediaries and financial markets) is highly developed or not. Specifically, the scatter plots show a negative relationship between the share of debt liabilities and the degree of CFMs for countries with better performed financial system. By contrast, for countries with less developed financial sector, there is no strong evident to support CFMs is associated with a lower debt ratio, and the correlation even appear to be somewhat positive in the case of countries with less developed banking sector. In a whole, the stylized facts from Figure 3-5 illustrate two facts. First, CFMs and financial development varies substantially among EMEs. Second, a reinforcement of capital flow management is not necessarily associated with lower share of foreign debt. In the following section, we investigate the nexus between CFMs, financial development and the composition of external liabilities by using econometric analysis. 3. Estimation Results 3.1 Benchmark results Table 1 Estimated effects of financial development and CFMs on the share of debt liabilities Private credit by deposit money banks to GDP Market capitalization to GDP VARIABLES Model 1.1 Model 1.2 Model 1.3 Model 1.4 Model 1.5 Model 1.6 Model 1.7 Model 1.8 Model 1.9 Model 1.10 Financial dev. (FD) 0.18** 0.16** 0.15* 0.21*** 0.21*** 0.02 0.01 0.02 0.27 0.03 [0.07] [0.07] [0.07] [0.07] [0.06] [0.06] [0.06] [0.05] [0.07] [0.06] Kai -15.70*** -14.06** -18.51*** -14.54*** -11.07** -11.93** [4.36] [5.65] [5.66] [4.02] [5.34] [5.46] Fincont -10.21* -3.35-11.29* -5.49 [5.158] [6.02] [5.67] [7.09] Fxreg -6.99 0.83-9.09-4.91 [5.55] [6.43] [5.57] [6.64] Vulnerability 2.53** 1.91 2.40* 2.35* 2.76* 3.13** 2.29* 2.70** 3.06** 3.46*** [1.21] [1.31] [1.32] [1.32] [1.38] [1.20] [1.21] [1.21] [1.26] [1.252] Institutional quality -53.29*** -32.16-29.71-41.31* -35.06-27.32-18.84-19.36-19.21-18.16 [18.82] [22.84] [24.66] [21.20] [22.04] [19.34] [24.33] [25.03] [23.92] [24.39] M2/GDP -0.31*** -0.28*** -0.25** -0.39*** -0.36*** -0.22*** -0.17* -0.18** -0.23*** -0.24*** [0.09] [0.10] [0.10] [0.09] [0.09] [0.08] [0.08] [0.08] [0.07] [0.075] Real GDP per capita 0.47-0.12 0.16-0.92-0.78 1.52 1.53 1.36 1.06 0.83 [2.34] [2.23] [2.10] [2.40] [2.15] [2.64] [2.53] [2.47] [2.57] [2.49] Observations 482 469 480 405 416 431 428 444 368 384 R-squared 0.53 0.48 0.45 0.54 0.54 0.50 0.44 0.44 0.48 0.49 Notes: Financial development (FD) is measured by private credit by deposit money banks to GDP (in percent), and stock mareket capitalization to GDP (in percent). Kai is a measure of capital controls suggested by Schindler (2009); Fincont is defined as an average of three binary variables measuring restrictions in the financial sector: (1) borrowing abroad; (2) maintenance of accounts abroad; and (3) differential treatment of deposit accounts held by nonresidents. Fxreg is the average of binary variables measuring foreign exchange restrictions in the financial sector: (1) lending locally in foreign exchange; (2) purchase of locally issued securities denominated in foreign exchange; (3) differential treatment of deposit accounts in foreign exchange; and (4) open foreign exchange position limits. Data on Fincont and Fxreg are obtained from Ostry et al (2012). Kai, Fincont, Fxreg, Financial development, vulnerability and M2/GDP are lagged one period. Constant, region dummies and time effects are included in all regressions. Clustered standar errors (on country level) reported in brackets; *, **, and *** indicate significance at 10, 5 and 1% levels, respectively. 11

We begin our analysis by focusing on the separate role of financial development and CFMs in determining the share of debt liabilities. The setting of regressions in Table 1 is identical to Ostry et al. (2012), except private credit to GDP as banking sector development and capitalization as stock market development are added into regressions. As shown in Table 1, economy-wide capital controls have a robust effect in reducing the share of debt in total liabilities, given the significant signs in each column. In contrast, the impact of financial sector-specific capital controls is relatively weak, considering that the coefficients of Fincont lose its significance when introducing into regressions with Kai simultaneously. With respect to FX-prudential measures, there is no clear relationship between prudential measures and debt liabilities since the estimates for Fxreg are insignificant in each of specifications. For other control variables, higher external vulnerability index is associated with increasing debt share, and better institutional quality and higher M2/GDP are correlated with lower debt share, while real GDP per capita does not help to explain the variation in debt ratio. These preliminary results are well consistent with Ostry et al. (2012). We then focus on the variables of our key interest - financial development. As presented in Model 1.1-1.5 of Table 1, the coefficients for private credit to GDP display positive and significant sign in each of columns, implying that the deepening of bank system alone is associated with larger share of debt in external legalities. In fact, this finding is in line with a strand of pioneering studies that also claim a crucial role of financial markets in determining the external capital structure. For instance, McQuade (2014) have demonstrated that domestic credit growth is associated with increase in debt inflows but not to net equity inflows. Theoretical work such as Mendoza et.al (2009), Ju and Wei (2010), Hagen and Zhang (2013) have explained why this may be the case through credit market imperfections theory. However, when we look at market capitalization as stock market development, as presented in Model 1.6-1.10 of Table 1, there is no clear evidence to support stock market development has a significant impact on the share of debt liabilities. In summary, these findings well confirm Hypothesis 1 and 2 claimed in introduction part, and one may ask: does this mean the deepening of banking sector would inevitably lead to a detrimental external liability structure in the end? And are capital controls the only choice to manage foreign debt? The answer is no. Table 2 presents the results based on Eq. (1) with interaction terms between capital flow restrictions and financial variables added into regressions. It is interesting to see when introducing the interaction term into the regressions, all of these three types of CFMs show a statistically significant interaction effect with banking sector development, namely, private credit by deposit money banks to GDP. It is also worth noting that when introduced jointly with economy-wide capital controls, as presented in Model 2.4 and 2.5, not only the estimates for Kai remain significant, but also those interaction terms for Fincont and Fxreg show robust signs. This result suggests a strong interacted relation between the effectiveness of CFMs and financial development, irrespective of whether the country sets economy-wide capital controls or not. 12

Table 2 Estimated effects of CFMs on the share of debt liabilities conditional on the level of financial development Private credit by deposit money banks to GDP Stock market captitalization to GDP VARIABLES Model 2.1 Model 2.2 Model 2.3 Model 2.4 Model 2.5 Model 2.6 Model 2.7 Model 2.8 Model 2.9 Model 2.10 Financial Dev. (FD) 0.27*** 0.24*** 0.29*** 0.25*** 0.29*** 0.13 0.09 0.17 0.18** 0.18 [0.07] [0.04] [0.07] [0.05] [0.06] [0.14] [0.02] [0.12] [0.08] [0.13] Kai -2.22-10.21* -16.24*** -9.30-12.66** -12.36** [6.68] [5.81] [5.77] [6.85] [5.08] [5.51] Kai FD -0.31*** -0.19 [0.11] [0.18] Fincont 9.48 10.97-3.6 5.40 [7.79] [8.34] [6.74] [7.415] Fincont FD -0.51*** -0.42*** -0.23** -0.33*** [0.10] [0.11] [0.10] [0.10] Fxreg 8.87 10.45-2.25 2.03 [8.28] [8.30] [7.01] [7.80] Fxreg FD -0.41*** -0.27** -0.29-0.28 [0.12] [0.12] [0.189] [0.19] Vulnerability 2.28* 2.42** 2.46** 2.61** 2.75** 2.84** 2.36* 2.67** 3.19** 3.42*** [1.16] [1.172] [1.16] [1.21] [1.27] [1.26] [1.23] [1.20] [1.29] [1.22] Institutional quality -47.86** -35.23* -25.89-39.91* -30.96-24.25-21.44-18.41-21.38-17.15 [19.21] [19.37] [22.19] [20.01] [21.37] [19.70] [22.48] [24.53] [21.68] [23.68] M2/GDP -0.23** -0.20*** -0.16* -0.29*** -0.28*** -0.21** -0.17* -0.18** -0.23*** -0.24*** [0.08] [0.07] [0.08] [0.08] [0.08] [0.07] [0.08] [0.085] [0.07] [0.07] Real GDP per capita -0.39-0.47-0.9-1.04-1.31 0.75 1.46 0.5 0.15 0.003 [2.46] [2.13] [2.13] [2.51] [2.24] [2.38] [2.35] [2.24] [2.39] [2.33] Observations 482 469 480 405 416 431 428 444 368 384 R-squared 0.55 0.54 0.49 0.57 0.55 0.51 0.47 0.45 0.53 0.51 Notes: Financial development (FD) is measured by private credit by deposit money banks to GDP (in percent), and stock market capitalization to GDP (in percent). Kai is a measure of capital controls suggested by Schindler (2009); Fincont is defined as an average of three binary variables measuring restrictions in the financial sector: (1) borrowing abroad; (2) maintenance of accounts abroad; and (3) differential treatment of deposit accounts held by nonresidents. Fxreg is the average of binary variables measuring foreign exchange restrictions in the financial sector: (1) lending locally in foreign exchange; (2) purchase of locally issued securities denominated in foreign exchange; (3) differential treatment of deposit accounts in foreign exchange; and (4) open foreign exchange position limits. Data for Fincont and Fxreg are obtained from Ostry et al (2012). Kai, Fincont, Fxreg, Financial development, vulnerability and M2/GDP are lagged one period. Constant, region dummies and time effects are included in all regressions. Clustered standar errors (on country level) reported in brackets; *, **, and *** indicate significance at 10, 5 and 1% levels, respectively. Generally speaking, Model 2.1-2.5 in Table 2 implies the effectiveness of CFMs in lowering debt liabilities is enhancing with the deepening of local banking system. Moreover, in term of the magnitude of interaction effects, the interaction term between private credit and financial-sector-specific capital controls displays the largest magnitude, followed by FX-related prudential measures, and economy-wide capital controls appears to be least affected by the level of domestic private credit. This interactional effect confirms our previous Hypothesis 3 that financial market plays a crucial role in determining the effectiveness of CFMs in reducing the debt liabilities, especially for regulations related to financial sector. When financial variable is measured by stock-market capitalization over GDP, as shown in Model 2.6-2.10 in Table 2, the estimation results are slightly different. In particular, one can see that the interaction effect between CFMs and financial development still display negative signs in each column, yet the effect is less pronounced here since only the estimates for interaction between financial-sectorspecific capital controls and market capitalization sustain significant. Furthermore, one also sees that the interaction effect between CFMs and stock market capitalization is substantially smaller in magnitude than the interaction with banking development, indicating that the effectiveness of CFMs enhance more with banking sector development than stock market development. 13

Table 3 Total effect of CFMs and financial development on the share of debt liabilities Model 2.1 Model 2.2 Model 2.3 Model 2.6 Model 2.7 Model 2.8 Private credit by deposit money banks to GDP Market capitalization to GDP Kai Fincont Fxreg Kai Fincont Fxreg Panel (a) Total effect of CFMs: how much a one standard deviation increase in the CFMSs would change the debt share for different level of financial development? FD at the mean level -4.82-3.71-2.33 0.75-4.66-4.02 FD at the minimum level -0.83 3.05 2.62 3.07-1.22-0.69 FD at the maximum level -28.66-44.11-31.94-9.53-16.94-18.76 Panel (b) Total effect financial development: how much a one standard deviation increase in the FD would change the debt share for different level of CFMs? CFMs at the mean level 0.04 0.01 0.02 0.02 0.001 0.02 CFMs at the minimum level 0.09 0.08 0.09 0.05 0.03 0.07 CFMs at the maximum level -0.01-0.09-0.04-0.02-0.06-0.05 Threshold level of CFMs 0.87 0.47 0.7 0.68 0.39 0.58 Number of countries that FD had a net negative effect on debt liabilities in 2008 14 19 29 19 19 29 To get an estimate of how effective CFMs has been in reducing the external indebtedness, one can ask the question of how much a one standard deviation increase in the CFMs index would reduce debt share of a country with particular level of financial development in the sample. The total effects of CFMs can be calculated by ( " " "# + "# ) at the mean, minimum and maximum values of financial development. As shown in Panel (a) of Table 3, the total effect of Kai / Fincont / Fxreg at the mean level of private credit is 4.82 / -3.71 / -2.33 respectively. The same effects are -0.83 / 3.05 / 2.62 when evaluated at the minimum level of private credit, and -28.66 / -44.11 / -31.94 at the maximum level of private credit. In the same way, the total effects of CFMs for the different levels of market capitalization are also calculated. Not surprisingly, the effects decline when we look at market capitalization as financial development. This is consistent with our previous statement that the effectiveness of CFMs enhances significantly with banking development, and to a lesser extent with stock market development. Based on the calculation results, it is also interesting to notice that economy-wide capital controls is more effective on reducing debt liabilities than the other two instruments only when banking sector (as reflected by private credit) is relatively inactive. In other cases, the effectiveness of FX-related prudential measures and financial sector-specific capital controls appear to be greater. In a quite similar way, we also calculate the total effects of financial development at the mean, minimum and maximum values of CFMs, where the total effects are measured by ( "# " " + " ). As presented in Panel (b) of Table 3, our estimation shows that one standard deviation increase in the private credit would increase the debt share of liabilities by 0.04 / 0.01 / 0.02 points if a country sets mean level of Kai / Fincont / Fxreg. Likewise, having better stock market would increase the debt share by 0.02 / 0.001 / 0.02 if mean level of Kai / Fincont / Fxreg is imposed. Of course, the total effects financial development on debt ratio turn from positive to negative as CFMs strengthen, and the effects on reduction in debt liabilities maximize when CFMs are set to their maximum levels. It is worth noting that both the coefficients for interaction terms and financial variables alone display statistically significant yet opposite signs, therefore it would be meaningful to calculate the threshold level of CFMs that the total effects of financial development turn positive to negative. Based on our estimation, when Kai / Fincont / Fxreg take values larger than 0.97 / 0.47 / 0.7, private credit would exert a reducing effect on external indebtedness, and there are 14 / 19 / 29 out of 51 EMEs in 14

2008 had set Kai / Fincont / Fxreg above the threshold level and are benefit from banking development. When financial development is measured by market capitalization, the threshold values drop to 0.68/ 0.39 / 0.58. Still, in total, there are relatively same numbers of countries with CFMs above the threshold levels in 2008 and experiencing a reducing share of debt liabilities stemming from equity market development. In summary, the results in Table 1-3 imply that the effectiveness of CFMs critically hinges on the development of domestic financial system, specially banking sector development. And among toolkit, the effects of FX-related prudential measures more likely to be affected by local financial system, whereas the effectiveness of economy-wide capital controls is less sensitive to banking activities. This finding fits well with existing literature that also argues that capital controls might be the only option when the domestic financial system are inactive, whereas prudential rule should be the main instruments when financial markets work well. 3.2 Robustness: alternative measures of CFMs Table 4 The estimation results of alternative measures of CFMs Private credit by deposit money banks to GDP Stock market capitalization to GDP VARIABLES Model 4.1 Model 4.2 Model 4.3 Model45.4 Model 4.5 Model 4.6 Financial Dev. (FD) 0.20** 0.27*** 0.22*** 0.03 0.09 0.13 [0.77] [0.04] [0.05] [0.06] [0.07] [0.09] Kaopen 3.46** -0.004 [1.38] [1.1] Kaopen x FD -0.17*** -0.67 [0.51] [0.05] Fincont2 16.36*** 2.6 [5.78] [5.87] Fincont2 FD -0.53*** -0.22** [0.08] [0.09] Fxreg2 8.37 3.12 [5.46] [5.61] Fxreg2 FD -0.32*** -0.22* [0.10] [0.12] Other control variables Yes Yes Yes Yes Yes Yes Observations 579 472 512 517 431 463 R-squared 0.5 0.53 0.5 0.45 0.44 0.45 Notes: Financial development is measured based on: (1) private credit by deposit money banks to GDP ; and (2) stock market capitalzation to GDP. Kaopen is the reciprocal of Chinn-Ito index. Fincont2 is defined as an average of two binary variables measuring restrictions in the financial sector: (1) borrowing abroad, and (2) differential treatment of deposit accounts held by nonresidents. Fxreg2 is the average of binary variables measuring foreign exchange restrictions in the financial sector: (1) lending locally in foreign exchange, and (2) differential treatment of deposit accounts in foreign exchange; and (4) open foreign exchange position limits. Kaopen, Fincont2, Fxreg2, Financial development, vulnerability and M2/GDP are lagged one period. Constant, region dummies and time effects are included in all regressions. Clustered standar errors (on country level) reported in brackets; *, **, and *** indicate significance at 10, 5 and 1% levels, respectively. In addition to the benchmark estimation, we conduct a series of sensitivity tests to ensure the robustness of the results. In this section, we start with robustness checks by using alternative ways to measure CFMs. Here, for economy-wide capital controls, we replace Schindler index with financial openness index constructed by Chinn and Ito (2008). Chinn-Ito index is based on three categorical indicators of financial current account restrictions (current account restrictions, export proceeds surrender requirements, and presence of multiple exchange rates), thus may not reflect the stringency of restrictions on cross-boarder transactions as directly as Schindler index does. To ensure higher value indicates more restrictive capital controls, we take the reverse of Chinn and Ito index into regression. And for financial-sector-specific 15

capital controls and prudential measures, a set of narrower definitions that come from Ostry et al. (2012) are being used here. Specifically, financial-sector-specific capital controls now only include (i) impose limits on financial sector borrowing from abroad; and (ii) restrict the maintenance of accounts abroad, and FX-related prudential regulations only include (i) limit local lending in foreign currency; and (ii) impose differential treatment of deposit accounts in foreign exchange. Table 4 presents the estimation results using alternative measures of CFMs, and note that some variables have been omitted for brevity. According to Table 4, one sees that the benchmark results survive by using alternative indicators of capital controls and prudential policies. The share of debt liabilities still increases with higher private credit to GDP, and market capitalization turns out to have no robust effects on the share of debt. More importantly, the interaction terms between CFMs and financial variables remain significant and negative. Still, it is worth noting that the magnitude of estimates for interaction terms is somewhat smaller than they are in Table 2. It is not surprising, though, giving that now we are using a relatively indirect measure to proxy economy-wide capital controls, and using narrower definition to measure financial-sector specific capital controls and prudential measures. 3.3 Robustness: alternative measures of financial development Table 5 Robustness: alternative indicators of financial development Deposit money bank assets to GDP Stock market total value traded to GDP VARIABLES Model 5.1 Model 5.2 Model 5.3 Model 5.4 Model 5.5 Model 5.6 Financial Dev. (FD) 0.23** 0.22*** 0.27*** 0.05 0.07 0.1 [0.08] [0.05] [0.07] [0.12] [0.07] [0.16] Kai -6.56-12.14** [8.29] [4.93] Kai x FD -0.17-0.23 [0.12] [0.16] Fincont 13.38-6.72 [8.23] [5.93] Fincont FD -0.48*** -0.36*** [0.11] [0.13] Fxreg 9.73-6.45 [9.11] [6.05] Fxreg FD -0.37*** -0.26 [0.13] [0.25] Other control variables Yes Yes Yes Yes Yes Yes Observations 482 469 480 424 426 442 R-squared 0.54 0.52 0.48 0.51 0.47 0.44 Notes: Financial development is measured based on: (1) deposit money bank assets to GDP (%) ; and (2) an stock market total value traded to GDP (%). Kai is a measure of capital controls suggested by Schindler (2009); Fincont is defined as an average of three binary variables measuring restrictions in the financial sector: (1) borrowing abroad; (2) maintenance of accounts abroad; and (3) differential treatment of deposit accounts held by nonresidents. Fxreg is the average of binary variables measuring foreign exchange restrictions in the financial sector: (1) lending locally in foreign exchange; (2) purchase of locally issued securities denominated in foreign exchange; (3) differential treatment of deposit accounts in foreign exchange; and (4) open foreign exchange position limits. Kai, Fincont, Fxreg, Financial development, vulnerability and M2/GDP are lagged one period. Constant, region dummies and time effects are included in all regressions. Clustered standar errors (on country level) reported in brackets; *, **, and *** indicate significance at 10, 5 and 1% levels, respectively. In addition to using alternative proxies for CFMs, we also consider other two choices in the literature to measure financial development. For the development of banking system, we replace private credit by deposit money banks to GDP with 16