Journal Issue
Share
Article

What Slice of the Pie? The Corporate Bond Market Boom in Emerging Economies1

Author(s):
Diana Ayala Pena, Milan Nedeljkovic, and Christian Saborowski
Published Date:
July 2015
Share
  • ShareShare
Show Summary Details

I. Introduction

Over the past two decades, emerging market economies (EM) have become increasingly integrated into global capital markets. While the development of equity markets picked up pace in the 1990s, the growth of private bond markets was initially slower and limited to a subset of industries in a smaller number of EMs. The period immediately following the global financial crisis (GFC) saw private bond market issuance catching up. The annual value of EM non-financial corporate (NFC) issuance increased more than threefold between 2009 and 2014, grossly outpacing equity and syndicated loan issuance. The boom contributed to growing debt stocks and sizable exposures to both foreign exchange risk and asset managers with portfolios highly concentrated in EM assets (IMF, 2014). On the bright side, it allowed a more diversified set of borrowers to diversify their funding sources. A key question is whether the borrowing spree can be seen, at least in part, as a structural rather than a cyclical shift in bond market development.

Policymakers in EMs have long pursued initiatives to promote capital market development more generally, and bond market development in particular.2 Intuitively, the diversification of funding sources should lead to more efficient capital allocation and better risk sharing, with a positive impact on long-term economic growth.3 What is more, evidence from advanced economies (Kashyap et al, 1993, Adrian et al, 2012, Becker and Ivashina, 2014) suggests that local bond issuance does not share the strongly pro-cyclical behavior of bank lending. It is in this spirit that the Asian financial crisis led observers to proclaim bond market development as an effort to develop “spare tires” that borrowers can rely on when bank balance sheets are strained (Greenspan, 1999).45

This paper studies the determinants of shifts in debt composition among EM corporates. Our primary aim is to identify both global and domestic factors - other than those related to the demand for borrowing more generally - that explain why financial systems shift away from bank lending and towards bond market finance. Our focus is on the recent bond market boom and the question why it was stronger in some countries than in others. In particular, we aim to understand whether EMs that experienced the largest booms relative to bank lending were those with strong fundamentals and institutions or whether it was cyclical factors that drove flows into the largest and most liquid markets. In this context, we also explore the role of cross-border bank linkages.

To facilitate the analysis, we propose a measure of corporate debt that can be decomposed both into bank loans and bonds, and into local and foreign currency instruments. The dependent variable throughout the analysis is the share of bond finance in total outstanding corporate debt. This choice has some important advantages, including that the dependent variable can be directly interpreted in relation to the size of the NFC sector’s outstanding debt.6 What is more, it implicitly controls for potentially endogenous factors that drive the overall demand for borrowing (from both bond markets and banks). The main focus of the empirical analysis is thus on factors that drive bond issuance beyond what can be explained based on shifts in the demand for funding. Potential drivers include (a) local fundamentals that provide an enabling environment for bond market development and foster investor interest such as institutions, macro fundamentals or market development initiatives, (b) domestic bank supply related variables that constrain or facilitate bank borrowing and (c) supply factors that drive the relative availability and cost of bond market finance.

We tackle our question of interest in two ways. First, we estimate censored panel regressions with fixed effects (Honore, 1992).7 While these enable us to identify a wide range of global and local drivers of bond market shares, they do not allow testing reliably whether a prominent finding of our descriptive analysis continues to hold, namely that market size is an important conditioning variable for the influence of global factors on increasing bond market access during the post-crisis period.8 In order to test this hypothesis, we cast the model in a panel quantile regression setup and employ the recently proposed censored quantile regression estimator for panel data with fixed effects (Galvao et al, 2013). The quantile regression offers a parsimonious framework to trace the varying importance of determinants at different levels of bond market development. In this way we can analyze whether the search for yield in global markets during the post-crisis period affected countries differently depending on whether their bond markets were more or less developed.

Our main hypothesis is that the recent boom was driven primarily by the global financial cycle (Rey, 2013). In particular, we conjecture that the search for yield accounted for most of the variation of bond shares in total corporate debt, with investor interest in specific EMs mostly driven by market size and the associated easy entry and exit.9 The analysis indeed confirms that the role of bond markets in NFC finance during the post-crisis period increased considerably more in EMs with more market based financial systems. While macro fundamentals and strong institutions are shown to be important determinants of bond market development throughout the sample period, their relative role declined substantially during the post-crisis period as global factors took center stage, paired with a growing investor focus on market size.10 We also find evidence for a role for global bank leverage in driving cross-border banking, building on the findings of Bruno and Shin (2015), among others.11

Our paper is related to empirical literature on the determinants of corporate bond issuance at the firm and country level. Earlier studies predominantly for the developed countries have shown that both firm-specific characteristics and the macroeconomic environment matter for firms’ decisions to issue bonds (Houston and James, 1996, Johnson, 1997; Datta et al, 2000; Dennis and Mihov, 2003; Hale and Santos, 2008; Mizen and Tsoukas, 2013, Didier et al, 2014, Gozzi et al, 2014). Important firm characteristics include firm size, growth and financial conditions while various other factors such as market depth, information asymmetries and market timing also play a key role. In addition, the literature emphasizes the role of reputation as past issuers are more likely to issue again than firms that have never issued before. Relatedly, the probability that a firm will issue a bond in domestic markets (relative to either not issuing at all or issuing in foreign markets) grows with the level of local bond market development. However, the fact that individual firms are more likely to issue when markets are more developed does not necessarily imply that initially well-developed bond markets continue to grow faster at the macro level. For instance, Burger et al (2012) find in a cursory analysis of the changes in US investors’ portfolio weights from 2006 to 2008 that investors tended to move towards markets in which they had smaller initial positions. Conversely, Hale et al (2014) show that countries better suited to issuing in home currency prior to the crisis also gained more in terms of home currency issuance following the Crisis.

At the same time, economic fundamentals are important drivers of bond investor interest (Laeven, 2014). Goldstein and Turner (2004) argue that economic policies and institutions are key determinants of bond market development in EMs. Eichengreen and Luengnaruemitchai (2006) indeed find that institutional impediments - and to some extent macro policies - can help explain the smaller size of Asian and Latin American bond markets relative to advanced economies. Hale (2007) suggests that country risk is the key macroeconomic fundamental that explains a large share of the variation in corporate financing choices between bonds and syndicated loans in EMs.

The choice between bond and bank financing can also be time-varying and related to cyclical drivers or the incidence of financial crises. Becker and Ivashina (2014) find evidence of a cyclical substitution between bank credit and bond financing at the firm level in the US, confirming earlier findings by Ramey (1992) and Kashyap et al (1993) at the macro level. Adrian et al (2012) provide additional empirical evidence on loan-bond substitutability in the US during the GFC and relate this pattern to the cyclicality of bank leverage. However, empirical evidence on the substitution channel is weaker in the case of developing economies. Indeed, Eichengreen (2007) notes that there is no guarantee that bond markets will continue to function as banking sectors collapse. Arteta and Hale (2008) find that both bank loan and bond financing to NFCs decrease following sovereign crises. Allen et al (2012), similarly, show that banking sectors and bond markets behave as complements rather than substitutes in the aftermath of banking crises.

Finally, our paper is related to the literature on capital flow surges during the post-crisis period. The importance of global conditions for fixed income flows to EMs has long been recognized in the literature. Early studies (Calvo et al, 1993, Chuhan et al, 1998, Antzoulatos, 2000) find that factors related to global liquidity and interest rates are more important than local fundamentals in explaining bond and equity issuance in Asian and Latin American economies in the 1990s. Rey (2013) establishes the existence of a global financial cycle—driving capital flows, asset prices and credit - which is not aligned with country-specific macroeconomic conditions and co-moves with uncertainty and risk aversion in global markets. Similarly, Forbes and Warnock (2012) show that global risk proxies such as the VIX consistently predict waves of capital flows. Bruno and Shin (2015) highlight the key role of the global bank leverage cycle in explaining cross-border banking flows and its close relationship with the role of the VIX. On the other hand, Fratzscher (2012) emphasizes the growing role of macro fundamentals during the post-crisis period, showing that countries with stronger macro fundamentals suffered lower capital outflows during the crisis and were able to attract more flows after the initial shock. Ghosh et al (2014) confirm the role of fundamentals in other episodes of capital flows surges. The cross-country variation and the relative role of local and global conditions in the recent EM NFC bond market boom, however, are still largely unexplored in the literature.12

This paper contributes to the existing literature in three ways: first, we propose a measure of NFC debt stocks in EMs that allows for a breakdown both by currency and by instrument. This allows studying the time and cross-country variation in the relative importance of bond versus bank financing for a large set of EMs. Second, we analyze the drivers of bond market shares in NFC debt at the macro level, allowing their impact to vary across different levels of bond market development, while controlling for the impact of demand side factors and time-invariant drivers of financial development. Finally, we show that the determinants of bond market access in EM vary importantly with global cyclical conditions. In particular, we confirm earlier findings in the literature on the importance of local fundamentals and global bank leverage for the EM corporate debt structure. However, we show that the relative role of local fundamentals declined substantially during the post-crisis period as global factors took center stage, paired with a growing investor focus on market size.

The finding that global cyclical factors explain most of the variation in EM bond market development during the post-crisis period is important from a policy perspective. To the extent that bond markets in EMs boomed largely because their large and liquid markets attracted investor flows during a cyclical upswing in the global financial cycle, these countries may be hit severely by capital outflows as the cycle turns. As such, our findings highlight the importance of strong institutions and macro fundamentals in facilitating a gradual diversification of funding sources.

The remainder of this paper is organized as follows: Section 2 proposes a measure of non-financial corporate debt stocks for emerging markets. Section 3 discusses trends in emerging market debt stocks and their composition with a particular focus on the role of bond market finance. Section 4 presents the empirical specification used in the regression analysis covered in Sections 5 (simple panel model) and 6 (quantile regression setup). Section 7 concludes.

II. A Measure of Non-financial Corporate Debt Stocks

This section proposes a measure of aggregate NFC debt stocks in EMs. While the Bank of International Settlements (BIS) provides NFC debt data for a number of countries, a comprehensive and comparable measure does not exist for a larger sample of EMs. Besides broadening the sample of countries, our focus is on putting together a measure of aggregate corporate debt that allows breaking the stock down into bank loans and bonds as well as into local and foreign currency debt. Moreover, we aim to gather sufficient information on the currency composition of foreign currency debt to allow identifying the influence of valuation effects. In particular, our measure has three components: domestic bank loans, foreign bank loans and bonds (Table 1).

Table 1.A Measure of NFC Debt
DefinitionDefinitionSourceAvailability
Outstanding stock of bondsNFC bonds outstanding by currency on an ultimate risk basisDealogicFull country sample
IFS – Other Depository Corporations (ODC) survey- Loans Other Non-financial Corporations and Loans Public Non-financial CorporationsAlgeria, Armenia, Belarus, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Egypt, Georgia, Guatemala, Indonesia, Jamaica, Kazakhstan, Macedonia, Mexico, Morocco, Pakistan, Panama, Philippines, South Africa, Thailand, Turkey and Uruguay
Country authoritiesAlbania*, Argentina, Bosnia and
Domestic bank loansDomestic bank loans to non-financial corporationHerzegovina*, Bulgaria*, China, Ecuador, El Salvador, India, Jordan, Latvia*, Lebanon, Lithuania, Malaysia, Peru, Poland*, Romania*, Russia, Serbia*, Tunisia, Ukraine* and Venezuela
ECB data Statistical Data Warehouse – MFIs loans deposits and security holdings by sectorCroatia* and Hungary*
Foreign bank loansExternal loans from BIS reporting banks to domestic non-bank sectorBIS -External loans of reporting banks vis-à-vis non-banking sectors (Table 6)Full country sample

Indicates countries whose data allows for a breakdown of bank loans into EUR and other currencies

Indicates countries whose data allows for a breakdown of bank loans into EUR and other currencies

The data source for the stock of outstanding bond market debt is the Dealogic Debt Capital Markets database (DCM).13 Dealogic DCM incorporates global primary market bond data since 1980, with details on almost half a million international and domestic deals. For each transaction, the database includes detailed information, including on the dates of issuance and maturity, the currency of issuance, the residence and sectoral classification of the issuer and the nationality of its parent company.14 We calculate the stock of bonds outstanding in country c at time t as the sum of bonds issued since 1980 in country c minus the sum of all those bonds that have matured by time t.15 We calculate the aggregate stocks in US dollars but take account of valuation effects in constructing the dollar value of the outstanding stock of bonds at a given point in time.16 We distinguish local and foreign currency bond stocks based on the currency at time of issuance. In countries in which the NFC sector never issued a bond, the stock of bonds outstanding is zero. Our country classification is based on the nationality of the parent company unless the issuer does not have a parent. This allows associating offshore issuance by foreign incorporated subsidiaries of parent companies located in country c with country c. In other words, debt stocks are calculated based on an ultimate risk basis (Avdjiev et al, 2014).

The second component of our measure is domestic loans, broken down into local and foreign currency loans. For the majority of countries in our sample, this information is taken from the IMF’s International Financial Statistics (IFS). For those countries for which the data is not available in IFS, the data is directly sourced from the relevant country authorities (Table 1).17

The third component of our measure is cross-border loans from BIS reporting banks to country c’s non-bank sector. We take two crucial assumptions: first, we assume that all cross-border loans are in foreign currency; second, we assume that cross-border loans to non-bank financial corporations are zero.

There are a few additional caveats to be considered. Most importantly, our measure does not include intercompany loans which constitute a large component of NFC debt in some EMs. The reason is, first, that data on intercompany debt is only available for a limited set of EMs; second, intercompany loans arguably have a different risk profile than other forms of debt; third, including intercompany debt would double count offshore issuance by foreign incorporated subsidiaries (reflected in bond stocks) to the extent that the proceeds are channeled back to the country of nationality of the parent company. Another caveat is that we do not separately include syndicated loans. In principle, syndicated loans are available from Dealogic and stocks can be calculated in the same way as bond stocks. However, including the stock of syndicated loans separately would lead to double counting to the extent that these are already included in domestic and foreign bank loans. This would be the case for all syndicated loans but a small minority that is tradable in secondary markets (Gadanecz, 2004). Finally, our measure does not comprise non-bank, non-bank lending.

Table 1 illustrates data availability by country. As can be seen, our complete measure is available for 47 EMs, spanning the period of 2000–13. While data on corporate debt is otherwise not available for a wider set of countries, there are at least two available sources that provide a comprehensive measure for some EMs. These are, first, the BIS measure of total NFC credit and, second, a measure of NFC debt employed in various issues of the IMF’s Global Financial Stability Report (GFSR). 18 However, neither measure would suffice for the purpose of this paper, as both cover a significantly smaller set of countries and permit neither a breakdown into foreign and local currency debt—including valuation adjustment—nor a breakdown into bank and bond market debt.19

Nevertheless, a comparison of our measure to the two alternatives is useful to ensure that the aggregates are of broadly similar magnitudes. In order to compare our measure on equal grounds, we add intercompany loans to our measure and choose countries for which all three measures are available. Figure 1 illustrates how NFC debt stocks in 2013 compared between our measure and the two alternatives. As illustrated in the chart, the overall magnitudes are mostly very similar.

Figure 1.Comparing Different Measures of Aggregate NFC Debt (2013)

Finally, our analysis also employs valuation adjusted measures of corporate debt. The motivation is our interest in the determinants of shifts in the composition of outstanding debt. Since corporate debt stocks comprise debts in both local and foreign currencies in many EMs, not accounting for valuation effects would omit an important variable driving movements in outstanding stocks. Our approach is to attempt to calculate all components of the total debt stocks at a constant exchange rate, namely that of December 2013.

In the case of the bonds data, the valuation adjustment can be performed in a straightforward fashion as Dealogic data allows calculating outstanding stocks by individual currencies. The challenge is greater in the case of domestic loans. In most EMs, a case can be made that the vast majority of domestic FX loans is denominated in US dollars. European EMs are an exception to this rule. In all European EMs other than Turkey and Russia (in which USD denominated loans constitute the vast majority of domestic bank loans) we therefore distinguish euro denominated loans.

Our strategy is thus as follows: for European EMs with the exception of Russia and Turkey, we break domestic bank loans down into EUR and USD denominated loans where loans denominated in currencies other than EUR and USD are assumed to be denominated in USD as well. For all other EMs, we assume that domestic bank loans in FX are fully denominated in USD. While this assumption may be a strong one in some cases, to our knowledge USD denominated loans constitute the majority of domestic bank loans in FX in all non-European EMs in our sample. Moreover, to the extent that the true currencies of denomination correlate more closely with the USD than with the local currency in each EM, it is still a preferable assumption to not controlling for valuation effects at all.

In the case of cross-border loans, a currency breakdown is not publically available from BIS. Our assumption is therefore that cross-border loans follow the same composition as domestic FX loans. While this may not be exactly true, there is likely to be a strong correlation in most cases. Moreover, cross-border loans constitute the smallest component of total corporate debt across EMs such that possible inaccuracies should have a relatively small impact on the results.

III. Trends in Non-financial Corporate Debt Stocks and Composition

This section discusses recent trends in EM corporate indebtedness, with a focus on developments since the global financial crisis. In the context of unconventional monetary policies in advanced economies, and the search for yield in global financial markets, EM corporate bond markets have boomed in recent years (Figure 2, left panel). Both foreign and local currency issuances contributed as bond finance increased in relative importance compared to other forms of financing. Equity finance by NFCs dropped from 1.7 percent of EM GDP in 2008 to about 1.1 percent in 2010 and 0.5 percent in 2014. At the same time, bond issuances increased from about 0.8 percent of EM GDP in 2008 to 3.3 percent in 2014. The right panel in Figure 2 illustrates that, since about 2010, bond markets have increasingly replaced syndicated loans as conduits of channeling liquidity to EMs.

Figure 2.EM NFC Bond vs. Equity and Syndicated Loan Issuance

Importantly, despite the recent boom, bond finance to EM NFCs is still small relatively to loans from domestic and foreign banks. The right panel in Figure 3 illustrates this based on a breakdown of our measure of total NFC debt by country: the mean outstanding stock of NFC bonds in our sample amounted to 5.3 percent of GDP in 2013 while domestic and foreign bank loans together amounted to an average of 40.5 percent. At the same time, however, the importance of bonds as a share of total corporate debt has grown substantially in recent years. The stock of outstanding bonds more or less doubled since 2009 in GDP terms while the outstanding stock of bank loans remained broadly constant. In other words, on average, the bond market boom has driven most of the increase in overall debt stocks over this period.

Figure 3.Change in EM Bond Market Debt 2009–13

The left panel in Figure 3 shows that the increase in debt ratios has indeed been dramatic in most EMs, with FX debt contributing notably.20 The handful of European EMs in which NFC debt stocks dropped are the exception. When we control for valuation effects, evaluating FX debt at end-2013 exchange rates, we see that the valuation adjusted debt changes are smaller for some EMs, for example in the cases of Hungary, Kazakhstan, Romania, Argentina and Turkey (countries with relatively high FX shares whose currencies depreciated significantly against the euro/dollar between 2008 and 2013). In other words, debts in these economies would have increased less/decreased more had it not been for the depreciation. The opposite is the case for Peru, the only country in the chart with sizeable FX debt whose currency appreciated notably over the period.

While the bond market boom boosted outstanding debt ratios, it also allowed NFCs to diverify their funding sources that were previously not able to issue bonds in domestic and international markets. As shown in Figure 4 (left panel), the total number of issuers, and particularly the number of new issuers, edged up quite a bit since 2008 and especially since 2010. In addition, the concentration of issuance declined. What is more, the share of issuers with ratings below investment grade also increased, mostly since 2010, signaling that foreign investors became less selective. Interestingly, both the number of issuers and the share of issuers with below investment grade ratings edged down in 2014.

Figure 4.Diversification of EM Bond Issuance

Taking a cross-country perspective, the key question this paper asks is what determined the extent to which the global bond market boom boosted access to bond finance – relative to bank loans—in some EMs more so than in others.

There are a number of factors that may have driven the allocation of global liquidity across EM bond markets. We aim to disentangle these factors in the econometric analysis presented in subsequent sections. It is useful, however, to illustrate some interesting descriptive findings beforehand. Figure 5 illustrates that the importance of foreign bank loans in total EM corporate debt has declined since the global financial crisis, in line with weaker balance sheets and tighter regulatory regimes in global banks. With regards bond finance, we see that it is largely access to international bond markets that increased in recent years relative to total NFC debt. We observe that the FX bonds share increased from 5.6 percent in 2008 to 8.0 percent since 2008, while in previous years it remained almost unchanged. The share of domestic bond finance, in turn, grew rapidly from 2003 to 2007, but has all but leveled off since 2009.

Figure 5.EM NFC Debt Composition Over Time

If we look at the same chart by region, we see that there is one exception that stands out, namely Asia (Figure 6). In Asia, it was clearly local bond markets that have grown in recent years while access to foreign bond markets at best stagnated. A possible explanation might be the strong policy push towards local bond market development since the launch of the Asian Bond Market Intitiative (ABMI) in 2003 and the Asian Bond Fund 2 ABF2 in 2005 (Chan, 2011). In Asia, we also see another interesting development that stands out: While foreign bank loans declined across the other regions in recent years, it was the share of domestic bank loans in Asia whose share in total debt has fallen.

Figure 6.EM NFC debt Composition Over Time by Region

With view to the econometric analysis, it is interesting to establish whether it was EMs with larger, more diversified access to domestic and international bond markets that grew strongest in recent years or rather those that were initially still more constrained in terms of bond finance.

The top right panel of Figure 7 illustrates that it is indeed EMs with the largest access to international bond markets in which access grew the most since 2009. The larger a country’s access in 2009, the more its access grew over the subsequent years. The top left panel however, shows that this is not business as usual: between 2003 and 2009, this relationship did not exist. If anything, countries with the largest initial access grew the least while countries with the smallest initial access grew the most. Moving to the lower panel, we see that declining pattern also holds for local currency bond markets. In fact, over the entire sample period, it is the countries with the largest local bond market access that grew the most.

Figure 7.Change in the Stock of NFC Bonds by Initial Quantile

Overall, this finding suggests that market size and easy entry and exit for investors are important in explaining why bond market access grew more in some EMs than in others during the post-crisis period. In subsequent sections, our aim is to assess whether this finding continues to hold in a regression setup. In particular, we aim to understand the relative roles of domestic structural factors—such as institutions and macro fundamentals—versus global cyclical factors in explaining bond market development across EMs.

IV. Empirical Specification

In this section, we move to the econometric analysis. In particular, we estimate different variants of the following model:

Throughout our analysis, the dependent variable yit is the share of bond finance (total, local or foreign currency) in total outstanding corporate debt. The advantage of our dependent variable of choice - compared to more commonly used measures of bond market development such as bond market debt over GDP - is that it implicitly controls for factors driving the overall (both bond and bank) demand for borrowing. In other words, it alleviates the need to control for variables such as economic activity on the right-hand side and thus does not require dealing with the related reverse causality issues.

In order to ensure parsimony, we group potential determinants into subsets and include only a limited number of variables from each subset in our baseline regression. The first group of regressors, EE, includes domestic factors that create an enabling environment for bond market development such as the quality of institutions or policy initiatives specifically aimed at market development. The second group of covariates, MF, comprises macro fundamentals. The third group of regressors, BC, includes proxies for local banking system characteristics. The fourth group of explanatory variables, included in G, comprises global factors driving capital flows to EMs such as proxies for the search for yield. Finally, our particular interest in the recent bond market boom episode leads us to interact all regressors in our model with a dummy that takes the value one for all observations during the period 2010 to 2013 and zero otherwise. The interaction terms are included in the vector Z. The definition of the dummy variable follows the literature (Cetorelli and Goldberg, 2011; Shin, 2013; Bremus and Fratzscher, 2014) who classify 2010–13 as the post-crises episode. However, we show that our results are generally robust to defining the post-crisis period as 2009–13 as well.

We make use of the time series dimension in y by using a panel regression setup for the entire sample period to explain developments in bond market shares. We tackle our question of interest in two ways. We begin with censored panel fixed effects regressions (Honore, 1992) of y on our control variables. The need to account for censoring arises because the dependent variable, y, is censored at zero (a modest share of the observations in our sample do take the value y=0); the need to control for unobserved cross-sectional heterogeneity arises from, inter alia, time-invariant drivers of financial development. While these regressions enable us to identify a wide range of global and local drivers of bond market development, they do not allow us to test reliably whether the key result of our descriptive analysis continues to hold, namely that the market size is an important conditioning variable for the effect of global factors on bond market access during the post-crisis period. While we could include (lagged) market size among the regressors, the arising simultaneity problem would be difficult to deal with.

In order to allow testing the proposition that market size matters for bond market development, we therefore, in the second step, cast the model in a panel quantile regression setup. This framework offers two main advantages for our analysis. First, the quantile regression estimator is robust to outliers in the dependent variable and imposes fewer restrictions on the distribution of the error term relative to conditional mean estimators. It thus provides a useful robustness check of the conditional mean results. Second, it provides a parsimonious way of tracing the varying importance of determinants at different levels of bond market development.21 In other words, it allows assessing how global factors and domestic conditions affect countries based on their position in the conditional distribution of bond market shares in total debt. Throughout the analysis we will be using the term “market development” and “market size” rather synonymously with “bond market shares in total NFC debt”. While a more typical definition would be bond market debt over GDP, the advantage of our measure is that it allows relating changes in the dependent variable directly to the size of the NFC sector’s total debt stock. It is further important to note that the correlation between NFC bond market debt over GDP and over total NFC debt is very high, amounting to more than 70 percent.

In order to control for both fixed effects and the censoring character of the dependent variable in a quantile regression setup, we use the recently proposed censored quantile regression estimator for panel data (CPQR) with fixed effects (Galvao et al, 2013). The CPQR estimator is an extension of Chernozhukov and Hong’s (2002) three-step censored quantile regression estimator. The general idea behind the CPQR estimator is to estimate a standard panel fixed effects quantile regression on a suitably defined subset of observations. The subset of observations for a particular quantile (τ) is selected by estimating a probability model for the non-zero bond share of NFC financing and selecting the observations for which the estimated propensity score is higher than 1 – τ. This ensures that only the data for which the conditional quantile line is above the censoring point is used in the estimation of the quantile regression parameters. The estimation procedure is done in three steps which are briefly summarized in Appendix 1.

V. Estimation Result Using Panel Model

We begin by discussing the results of the censored panel regressions with fixed effects. The dependent variable in our baseline regressions is the percent share of bond market debt in total NFC debt. All regressors we employ are defined in Table 2 in the Appendix. Tables 3 to 7 each show our benchmark specification in the first column as well as, in the remainder of the columns, robustness checks in which we deviate from the benchmark by adding/replacing one indicator at a time from a given subset of regressors (G, EE, MF and BC). All tables report estimated average marginal effects (Honore, 2008, Alan et al, 2014) with bootstrapped standard errors. We use standard clustered bootstrap (with 200 repetitions) and calculate significance levels as percentiles of the bootstrap distribution (Abrevaya and Shen, 2014).22

Table 2.Definitions and Sources of Variables
NameDefinitionSource
Dependent variables
Adjusted Bond sharePercent share of bonds in total NFC debt, adjusted for valuation effects (see section II)See Table 1
Unadjusted Bond SharePercent share of bonds in total NFC debt (see section II)See Table 1
Adjusted LC Bond SharePercent share of LC bonds in total NFC debt, adjusted for valuation effects (see section II)See Table 1
Adjusted FX Bond SharePercent share of FX bonds in total NFC debt, adjusted for valuation effects (see section II)See Table 1
Regressors
Enforcement proceduresMeasures the average number of procedures to enforce a contractWorld Bank Doing Business
Bond market concentration (lagged)Share of largest issuance in total issuances in given yearDealogic; author’s calculations
Asian Bond Fund dummyTakes the value 1 during year in which a country was a member of the ABFAuthor’s calculations
Current account ratio, 3-year average (lagged)Lagged 3-year average of current account ratio to GDP, in percentWorld Economic Outlook; authors’ calculations
Local bank capital to assets (lagged)The ratio of local bank capital and reserves to total assets, in percentWorld Development Indicators and IMF GSFR
US high yield spreadMoody’s Baa-Aaa Corporate Bond Yield, in percentFED St. Louis
US BD leverage growthUS Broker-dealer leverage growthAuthor’s calculations based on Adrian and Shin (2011)
Dummy for 2010-13Takes the value 1 during years 2010 to 2013Author’s calculations
Creditor rightsMeasures the degree to which collateral and bancrupcy laws protect borrowers and lendersWorld Bank Doing Business
Creditor informationMeasures rules and practices afecting the coverage, scope and accessibility of credit informationWorld Bank Doing Business
Rule of lawMeasures whether confidence in and adherence to rules of society, e.g. contracts and property rightsWorldwide Governance Indicators
Number of bond market issuers (lagged)Simple count of the numberyDealogic; author’s calculations
Financial development (lagged)Financial development indexIMF (2015b)
PPP GDP per capita, (logged and lagged)Gross domestic product per capita converted to international dollars using purchasing power parity ratesWorld Development Indicators
Reserves in percent of ST external debtGross international reserves, percent of short term external debtWorld Economic Outlook; authors’ calculations
External debt in percent of exports of G&STotal external debt, percent of exports of goods and servicesWorld Economic Outlook; authors’ calculations
Growth, 3-year average (lagged)Lagged 3-year average of growth rate, in percentWorld Economic Outlook; authors’ calculations
Inflation, 3-year average (lagged)Lagged 3-year average of inflation rate, in percentWorld Economic Outlook; authors’ calculations
ICRG composite risk indicatorComposite risk indicatorICRG
ICRG financial risk indicatorFinancial risk indicatorICRG
Local bank NPL ratio (lagged)The value of nonperforming loans divided by the total value of the local bank’s loan portfolio, in percentWorld Development Indicators and IMF GSFR
Cross-border claims (bank-to-bank), percent GDPExternal position of BIS reporting banks vis-à-vis domestic banks, percent GDP (BIS Table 6)BIS
US term spreadUS Term spread (3M T-bill vs 10yr bond)Bloomberg, Author’s calculation
VIXImplied volatility of S&P 500 indexFED St. Louis
Money market spreadSpread between US federal funds rate and domestic interbank rate, in percentIFS; central bank websites
Chinn-Ito IndexMeasuring a country’s degree of capital account opennesshttp://web.pdx.edu/~ito/Chinn-Ito_website.htm
Commodity exports, percent GDPTotal commodity exports as a percentage of GDPGruss (2014)
Table 3.Baseline Regression and Enabling Environment (EE)
Reg 1Reg 2Reg 3Reg 4Reg 5Reg 6
Enforcement procedures−1.13* (.58)−1.21 (1.07)
Interaction with 2010-13 dummy.02 (.1)−.13 (.16)
Bond market concentration (lagged)−2.29*** (.7)−2.58*** (.74)−3.06*** (.78)−3.08*** (.74)−2.25** (.99)
Interaction with 2010-13 dummy−.2 (1.66)−.34 (1.48)−.56 (1.52)−.32 (1.61).01 (1.65)
Asian Bond Fund dummy4.28** (3.11)4.17** (3.66)3.84** (2.25)4.18* (3.31)6.21** (3.62)3.31* (2.21)
Interaction with 2010-13 dummy1.12 (1.52).65 (1.57).5 (1.78).51 (1.75)−.75 (1.9)−.85 (1.6)
Current account ratio, 3-year average (lagged).16** (.09).19* (.1).21** (.1).21** (.1).26** (.14).26*** (.1)
Interaction with 2010-13 dummy0 (.1)0 (.13)−.01 (.13)−.01 (.14).09 (.22).08 (.12)
Local bank capital to assets (lagged)−.22 (.21)−.2 (.21)−.22 (.2)−.21 (.21)−.26 (.34)−.2 (.19)
Interaction with 2010-13 dummy.35* (.2).36 (.23).38 (.23).36 (.27).43** (.22).43 (.21)
US high yield spread−.13 (.28)−.19 (.31)−.2 (.3)−.06 (.33)−.91** (.49)−.59*** (.23)
Interaction with 2010-13 dummy−1.81** (.73)−1.77** (.84)−1.57* (.86)−1.87* (.87)−3.12** (1.41)−1.43* (.81)
US BD leverage growth−1.11** (.54)−1.02** (.5)−.83 (.51)−1.05** (.5)−2.23** (.83)−1.05** (.51)
Interaction with 2010-13 dummy.04 (1.07)−.67 (1.23)−1.31 (1.48)−.91 (1.44)2.15 (2.29).22 (1.36)
Dummy for 2010-13−.19 (5.95)1.48 (3.61).8 (4.38).94 (3.24)5.93 (7.09)13.16 (12.02)
Creditor rights.6 (.24)*
Interaction with 2010-13 dummy−.14 (.27)
Credit information.18 (.16)
Interaction with 2010-13 dummy−.08 (.6)
Rule of law.3 (2.57)
Interaction with 2010-13 dummy−.84 (1.2)
Number of bond issuers (logged and lagged)1.15*** (.38)
Interaction with 2010-13 dummy.33 (.57)
PPP GDP per capita, (logged and lagged)7.32*** (2.99)
Interaction with 2010-13 dummy−1.52 (1.34)
Number of observations/units476/11476/11476/11476/11230/07439/10
ChiSq333.0134.8111.2139.8294.7180.7
Prob > ChiSq0.0000.0000.0000.0000.0000.000
Fraction uncensored0.680.680.680.680.680.69
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
Table 4.Baseline Regression and Macro Fundamentals (MF)
Reg 1Reg 2Reg 3Reg 4Reg 5Reg 6
Enforcement procedures−1.13* (.58)−1.22 (.73)−1.09 (.61)−1.24** (.61)−.73 (.58)−.7 (.63)
Interaction with 2010-13 dummy.02 (.1)−.01 (.11)−.03 (.11)−.02 (.11)−.01 (.1)−.03 (.11)
Bond market concentration (lagged)−2.29*** (.7)−2.26*** (.68)−1.76*** (.64)−2.23*** (.66)−2.22*** (.64)−1.96*** (.62)
Interaction with 2010-13 dummy−.2 (1.66).14 (1.41).15 (1.52).23 (1.32)−.66 (1.33)−.72 (1.51)
Asian Bond Fund dummy4.28** (3.11)3.82* (2.77)3.44* (2.58)3.81** (1.8)3.86* (1.77)3.46* (1.67)
Interaction with 2010-13 dummy1.12 (1.52).81 (1.56)1.3 (1.61)1.02 (1.48).92 (1.13).8 (1.14)
Current account ratio, 3-year average (lagged).16** (.09)
Interaction with 2010-13 dummy0 (.1)
Local bank capital to assets (lagged)−.22 (.21)−.14 (.23)−.22 (.2)−.16 (.19)−.01 (.19).01 (.21)
Interaction with 2010-13 dummy.35* (.2).31 (.22).35** (.19).38* (.19).14 (.15).11 (.15)
US high yield spread−.13 (.28)−.31 (.3)−.28 (.29)−.23 (.29)−.11 (.3).25 (.28)
Interaction with 2010-13 dummy−1.81** (.73)−1.71** (.68)−2.03** (.74)−1.93** (.77)−2** (.77)−2.63*** (.84)
US BD leverage growth−1.11** (.54)−.87 (.58)−.74 (.57)−.88 (.57)−1.23** (.55)−.98* (.59)
Interaction with 2010-13 dummy.04 (1.07)−.8 (1.04).27 (1.04).75 (1).32 (1.19).54 (1.47)
Dummy for 2010-13−.19 (5.95).22 (5.89)1.36 (6.1)−.28 (5.86)4.61 (9.7)7.13 (6.65)
Reserves in percent of ST external debt (lagged)0.00*** (0.00)
Interaction with 2010-13 dummy0.00 (0.00)
External debt in percent of exports of G&S (lagged)−.01** (.01)
Interaction with 2010-13 dummy0 (.01)
Growth, 3-year average (lagged).04 (.18)
Interaction with 2010-13 dummy.19 (.23)
ICRG composite risk indicator.09 (.08)
Interaction with 2010-13 dummy−.02 (.11)
ICRG financial risk indicator.26** (.09)
Interaction with 2010-13 dummy−.06 (.11)
Number of observations/units476/11449/11467/11473/11445/11442/11
ChiSq333.0166.8214.5156.1132.5166.1
Prob > ChiSq0.0000.0000.0000.0000.0000.000
Fraction uncensored0.680.680.680.680.710.72
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
Table 5.Baseline Regression and Domestic Bank Characteristics (BC)
Reg 1Reg 2Reg 3
Enforcement procedures−1.13* (.58)−1.03* (.64)−1.15** (.59)
Interaction with 2010-13 dummy.02 (.1)−.06 (.1).01 (.09)
Bond market concentration (lagged)−2.29*** (.7)−2.21*** (.65)−2.43*** (.68)
Interaction with 2010-13 dummy−.2 (1.66)−1.05 (1.24).23 (1.59)
Asian Bond Fund dummy4.28** (3.11)4.57** (2.27)3.93** (2.64)
Interaction with 2010-13 dummy1.12 (1.52)1.83 (1.53)1.07 (1.54)
Current account ratio, 3-year average (lagged).16** (.09).12 (.09).16* (.09)
Interaction with 2010-13 dummy0 (.1)−.11 (.12)−.02 (.1)
Local bank capital to assets (lagged)−.22 (.21)−.17 (.21)
Interaction with 2010-13 dummy.35* (.2).34* (.19)
US high yield spread−.13 (.28)−.07 (.25).02 (.28)
Interaction with 2010-13 dummy−1.81** (.73)−1.46** (.67)−1.63* (.8)
US BD leverage growth−1.11** (.54)−1.26** (.58)−1.18** (.53)
Interaction with 2010-13 dummy.04 (1.07)−.49 (.87).09 (1.13)
Dummy for 2010-13−.19 (5.95)5.57 (4.54).35 (5.33)
Local bank NPL ratio (lagged).07 (.07)
Interaction with 2010-13 dummy.06 (.1)
Cross-border claims (bank-to-bank), percent GDP−.1** (.06)
Interaction with 2010-13 dummy−.03 (.03)
Number of observations/units476/11492/11486/12
ChiSq333.0637.7355.3
Prob > ChiSq0.0000.0000.000
Fraction uncensored0.680.660.65
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
Table 6.Baseline and Global Variables (G)
Reg 1Reg 2Reg 3Reg 4
Enforcement procedures−1.13* (.58)−1.08 (.59)−1.11 (.59)−1.15* (.56)
Interaction with 2010-13 dummy.02 (.1).02 (.1).02 (.1)−.03 (.11)
Bond market concentration (lagged)−2.29*** (.7)−2.29*** (.73)−2.3*** (.72)−2.37** (.81)
Interaction with 2010-13 dummy−.2 (1.66)−.12 (1.65)−.11 (1.64)−.22 (1.76)
Asian Bond Fund dummy4.28** (3.11)4.37** (2.63)4.3** (2.63)4.45** (2.9)
Interaction with 2010-13 dummy1.12 (1.52)1.13 (1.53)1.14 (1.53)1.61 (1.79)
Current account ratio, 3-year average (lagged).16** (.09).16** (.09).16** (.09).2** (.09)
Interaction with 2010-13 dummy0 (.1)0 (.1)0 (.1)−.03 (.11)
Local bank capital to assets (lagged)−.22 (.21)−.23 (.21)−.23 (.21)−.2 (.24)
Interaction with 2010-13 dummy.35* (.2).35 (.21).35 (.21).39 (.26)
US high yield spread−.13 (.28)−.1 (.29).18 (.47)−.14 (.33)
Interaction with 2010-13 dummy−1.81** (.73)−5.03* (2.55)−1 (1.02)−2.56*** (.76)
US BD leverage growth−1.11** (.54)−1.79*** (.6)−1.73** (.68)−1.12** (.57)
Interaction with 2010-13 dummy.04 (1.07)2.33 (1.81)2.32 (2.30).77 (1.33)
Dummy for 2010-13−.19 (5.95)4.28 (8.23).86 (6.55).73 (6.84)
US term spread0 (0)
Interaction with 2010-13 dummy−.01 (.01)
VIX−.04 (.04)
Interaction with 2010-13 dummy−.1 (.1)
Money market spread.03 (.06)
Interaction with 2010-13 dummy.21 (.15)
Number of observations/units476/11476/11476/11459/11
ChiSq333.0388.6352.8501.8
Prob > ChiSq0.0000.0000.0000.000
Fraction uncensored0.680.680.680.69
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
Table 7.Baseline and Specification Checks
Reg 1Reg 2Reg 3Reg 4Reg 5Reg 6Reg 7Reg 8
Enforcement procedures−1.13* (.58)−1.08* (.6)−.99 (.59)−1.03* (.6)−1.05 (.59)−1.08* (.61)−.56 (.54)−1.24** (.56)
Interaction with 2010-13 dummy.02 (.1).03 (.1).02 (.11)0 (.11)−.04 (.13).09 (.13)
Bond market concentration (lagged)−2.29*** (.7)−2.2*** (.55)−2.43*** (.71)−1.62** (.85)−2.59*** (.84)−1.94** (.84)−.7 (.94)−2.18** (1.07)
Interaction with 2010-13 dummy−.2 (1.66)−.01 (1.72).45 (1.74)−.37 (1.49).09 (1.54)1.17 (1.58)
Asian Bond Fund dummy4.28** (3.11)4.29* (6.8)4.09** (3.09)2.79* (2.27)4.48** (3.81).99 (5.03).17 (2.52)4.86** (1.82)
Interaction with 2010-13 dummy1.12 (1.52)1.02 (1.56)1 (1.75)1.37 (1.57)−.82 (1.78)1.24 (1.38)
Current account ratio, 3-year average (lagged).16** (.09).17** (.08).15* (.09).19** (.09).2* (.09).17** (.08).12 (.08).13 (.13)
Interaction with 2010-13 dummy0 (.1)0 (.11)−.02 (.12)−.05 (.11)−.12 (.17).11 (.14)
Local bank capital to assets (lagged)−.22 (.21)−.23 (.21)−.27 (.2)−.2 (.24)−.19 (.22)−.3 (.22).15 (.17)
Interaction with 2010-13 dummy.35* (.2).3 (.17).35* (.2).36 (.27).36 (.24).4 (.24).09 (.2)
US high yield spread−.13 (.28)−.1 (.26).35 (.41)−.24 (.32)−.09 (.3)−.34 (.32).22 (.28)
Interaction with 2010-13 dummy−1.81** (.73)−1.92*** (.69)−2.14** (.79)−1.88** (.83)−1.9** (.71)−1.57** (.75)−.9 (.65)
US BD leverage growth−1.11** (.54)−1.08** (.5)−1.14** (.54)1.57 (1.2)−1.26** (.53)−1.81*** (.58)−.15 (.49)−1.44*** (.47)
Interaction with 2010-13 dummy.04 (1.07).07 (1.13).22 (1.19).68 (1.18)−.68 (1.27)1.47** (.83)
Dummy for 2010-13−.19 (5.95)1.23 (2.07)−.24 (5.76)−.39 (6.95).21 (6.95)0 (6.94)−2.49 (6.39)
Dummy for 2009-13−.11 (6.44)
Interaction with Enforcement procedures.04 (.12)
Interaction with concentration−.26 (1.47)
Interaction with Asian Bond Fund dummy1.51 (1.5)
Interaction with current account.01 (.08)
Interaction with local bank capital ratio.42** (.2)
Interaction with high yield spread−3.19*** (.82)
Interaction with BD leverage growth−1.35 (1.2)
Chinn-Ito Index.47 (.58)
Interaction with 2010-13 dummy−.61 (.41)
Financial development index−1.43 (4.26)
Interaction with 2010-13 dummy1.18 (4.17)
Number of observations/units476/11476/11476/11476/11465/11476/11476/12476/13
ChiSq333.091.8280.8317.3681.4610.1435.3522.3
Prob > ChiSq0.0000.0000.0000.0000.0000.0000.0000.000
Fraction uncensored0.680.680.680.680.690.680.680.68
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.

Table 3, column 1 shows the results from our baseline specification. We estimate the model over the period 2002–13. The number of observations is 476, with 43 cross-sectional units and an average of 11 observations per unit.23 Importantly, note that the dummy for the period 2010–13 is insignificant, illustrating that the specification explains any idiosyncrasies about the post-crisis episode reasonably well.

We begin by examining the findings for the regressors included in EE, namely domestic factors that create an enabling environment for bond market development. The empirical literature has established a strong link between institutions and financial development.24 Given the disadvantages bond market investors face - compared to banks - in information gathering (Holmstrom and Tirole, 1998), seniority (Welch, 1997) and collateral loan immunization (Rajan and Winton, 1995), we would expect stronger institutions to boost investor interest in bond market financing relative to bank lending. Other factors that may create a stronger enabling environment are those that proxy for an established issuer base and financial infrastructure. Both issuers and investors may benefit through limited information gaps and a lower cost of market entry. While we cannot include initial market size as a regressor for reasons discussed in the previous section, we do include proxies such as a measure of bond market diversification. Finally, policies explicitly aimed at bond market development, including through establishing the necessary infrastructure and promoting market access could be important determinants of market access.

Our baseline specification includes three regressors that are designed to proxy for the quality of the enabling environment; first, an indicator of institutional quality, second, a measure of the concentration of bond issuance and, third, a dummy for membership in the Asian Bond Fund initiative, an initiative of 12 major central banks in the Asia-Pacific region to promote local bond market development. The institutional quality indicator of choice is the number of procedures necessary to enforce contracts from the World Bank’s Doing Business indicators, an indicator widely used in the literature. The results shown in column 1 of Table 3 illustrate that the number of enforcement procedures is indeed a significant determinant of bond market development and carries the expected negative sign. The same holds for the concentration indicator (negative sign) and the dummy for membership in the Asian Bond Fund initiative (positive sign). Conversely, the interaction terms between each of the three variables and the dummy for the period of 2010–13 are all insignificant at conventional levels. In other words, a strong enabling environment drives bond market development. However, the importance of these factors has not changed during the post-crisis period and is thus unlikely to explain the strong boost to bond market development in recent years.

This finding is confirmed in our robustness checks in the remaining columns of Table 3. Columns 2 to 5 replace our measure of institutional quality with alternative indicators frequently used in the literature while columns 6 and 7 replace the concentration measure with alternatives. In none of these cases does one of the interaction terms end up being significant. The coefficients on the institutional quality indicators - including creditor rights, credit information and the rule of law - carry the expected signs (stronger institutions are associated with growing bond market shares), but not all of them are significant. Similarly, replacing the concentration measure with alternative proxies for the quality of market infrastructure, we find a positive link between the dependent variable and (a) the number of bond market issuers and (b) GDP per capita. However, the interaction terms are once again not significant, confirming the result that an enabling environment matters but is unlikely to explain much of the cross-country variation in the recent bond market boom. Moreover, note that alternating the enabling environment proxies generally does not change the signs or the statistical significance of the remaining regressors.

The baseline regression further includes one variable from the MF subset, proxying for the quality of macroeconomic fundamentals in relation to the level of country risk (Hale, 2007). In particular, we include the lagged three year average current account balance as a percent of GDP in line with Fratzscher (2012), who illustrates the importance of current account deficits as drivers of global capital flows. The first column of Table 4 illustrates that we find evidence for the expected positive association between the lagged current account and the dependent variable, indicating that increasing bond market access more so than domestic credit growth is predicated on strong fundamentals. The variable is highly significant while the interaction term is insignificant. Table 4, columns 2 to 6 show the results when we replace the current account with alternative measures of macro fundamentals. We find equivalent results when employing other commonly used indicators such as reserves as a percentage of short term debt (positive sign), external debt as a percentage of exports of goods and services (negative sign), and the ICRG country financial risk rating; the lagged three year average growth rate and the ICRG composite risk rating show the correct coefficient sign but are not significant. In all cases, the interaction terms are insignificant. The bottom line, as in the case of the enabling environment, is that strong macro fundamentals increase investor interest in EM bond markets, but their importance did not increase during the post-crisis period.

Table 5 takes a closer look at local banking system characteristics. The variable we include in our baseline is the bank capital to assets ratio, an inverse measure of leverage. In our setup, there are at least two competing hypotheses as to whether the coefficient on the variable should take a negative or a positive sign. Noting that an increase in the capital ratio implies falling bank leverage, a negative sign implies that bond market issuance is a complement rather than a substitute to bank lending (Holstrom and Tirole, 1997). Conversely, a positive sign could arise either if banks and bond markets were substitutes or if they were complements with bond markets less sensitive to cyclical conditions. Intuitively, bank lending and bond issuance may both increase as local bank risk taking takes off since local banks tend to be major holders of corporate bonds in EMs (Eichengreen and Luengnaruemitchai, 2006). In addition, if the bond market investors face information and monitoring deficits compared to banks, uncertainty for bond investors grows - driving down their supply of funds—as the stock of outstanding bank loans falls (Holstrom and Tirole, 1997). On the other hand, evidence from advanced economies (Kashyap et al, 1993, Adrian et al, 2012, Becker and Ivashina, 2014) suggests that local bond issuance does not share the strongly pro-cyclical behavior of bank lending (leverage) and that bonds tend to substitute for cyclical contractions in the supply of bank loans.

In the baseline specification, the local bank capital ratio turns out to be insignificant with a negative sign while its interaction with the 2010–13 dummy is significant and carries a positive sign. In other words, while the evidence of a negative link prior to 2010 is statistically insignificant, decreasing local bank leverage is associated with relatively stronger bond market growth during the post-crisis period. The evidence in favor of this link during the post-crisis period is only weak, however. In particular, we do not find similar results when we replace the capital ratio with the share of non-performing loans in column 2 of Figure 5 (Becker and Ivashina, 2014). Here, the variable and its interaction term are always insignificant. In sum, while there is some evidence that bond market issuance substituted for weak bank lending, the evidence is not very robust.

Finally, we also control directly for the dependence on foreign funding in column 3 of Table 5 using an indicator that captures lagged cross-border exposures of BIS reporting banks to domestic banks as a percentage of GDP. During periods of growing cross-border banking, we may expect the variable to carry a negative coefficient, signaling that EMs highly dependent on cross-border banking would further increase their dependence on foreign funding intermediated through banks. Our results confirm this expectation as the indicator shows a negative coefficient and is significant. However, during the post-crisis period, as global banks reduced cross-border lending, one may expect the opposite, namely that a high initial dependence on foreign funding would put downward pressure on bank credit and thus increase the dependent variable (positive coefficient on the interaction term). However, we find no evidence for such a link. There are at least three possible reasons for this somewhat surprising finding: first, while cross-border exposures of global banks declined in the post-crisis period, domestic subsidiary lending did not, signaling that subsidiaries found alternative sources of financing (IMF, 2015); second, bond market issuance in European EMs—those with comparably high foreign funding dependence—grew only marginally compared to other EMs. In other words, factors that constrained bond market borrowing in European EMs during the post-crisis period – such as initial market size—may explain the overall negative coefficient. Finally, while we lag the variable, it is very persistent, and endogeneity issues are unlikely to be resolved.

The baseline specification further includes two global variables (Table 6). The first is the US high yield spread which we include as a measure of global risk aversion towards high yield fixed income investments.25 Given the EM NFC’s risk profile, we would expect a lower high yield spread in the US market to lead to greater demand for NFC bonds across EMs. The second global factor is the growth rate of US broker-dealer (BD) leverage as a proxy for global bank liquidity and risk taking behavior. Bruno and Shin (2015) highlight the importance of the global bank leverage cycle in explaining cross-border banking flows. Following this reasoning, to the extent that BD leverage falls, bond markets’ role as a conduit of channeling liquidity to EMs could be enhanced.26

We find that the coefficient on the high yield spread and its interaction term are negative, although only the interaction term is statistically significant (Table 6, column 1). Conversely, the coefficient on BD leverage growth is negative and the variable is significant while the interaction term is insignificant with a positive coefficient. This implies that a falling high yield spread is associated with growing investments into bonds issued by EM corporates. Before 2010 this effect is not statistically significant, in line with the still limited integration of EM corporate bond markets into global financial market (Shin, 2013). Indeed, the quantile regression analysis discussed in the next section confirms that the pre-2010 impact of the high yield spread is significant, but only for countries with an already high level of bond development. During the post-crisis period, the effect becomes large and highly significant, indicating that global bond markets largely replaced cross-border banking—plagued by balance sheet weakness and regulatory reform - as conduits of channeling liquidity to EMs. Conversely, BD leverage growth carries a negative sign and is significant while its interaction term is insignificant, suggesting that global bank risk taking behavior significantly reduces bond shares in EM corporate debt independently of the time period under consideration.

The results are robust to including the US term spread and the VIX as additional measures of global risk aversion, the difficulty being that these variables are closely correlated with the high yield spread (columns 2 and 3). Both the term spread and the VIX are not significant when included as additional regressors. When we include the term spread, our results do not change qualitatively but the coefficient magnitudes differ; when we include the VIX, the US HY spread interaction becomes insignificant due to their close correlation over the post-crisis period, but it retains a large negative coefficient.

We also include the differential between local money market rates and the US Federal Funds rate in the regression as a measure of relative funding costs (column 6). The idea is that higher local interbank funding costs should boost demand for bonds in global markets (market timing). The variable has a positive coefficient, as does its interaction. This suggests that interest differentials indeed boosted bond market borrowing although the estimated effect is not statistically significant.

Table 7 includes additional specification checks. In column 2, we drop all the insignificant variables from the regression; in column 3, we include the same dependent variable except that we do not adjust it for valuation effects; in column 4, we replace the dummy variable for the period 2010–13 with one for the period 2009–13; in column 5, we include the Chinn-Ito index of financial openness as an additional regressor; in column 6 we include a measure of broad financial development. Our results are generally robust to these specification checks. The main concern arising from these results is that the BD leverage growth term is no longer significant when we redefine the dummy. Other than that, our main results are qualitatively unchanged. The coefficient for Chinn-Ito indicator of financial openness and its interaction term is statistically insignificant, suggesting a limited role of capital controls in explaining the changes in the NFC debt structure. This result is confirmed when we use the Quinn indicator as an alternative (not shown).

While analyzing the relative determinants of local and foreign bond market development is left to future work, columns 7 and 8 of Table 7 take an initial pass at the issue. We simply run our baseline specification except that we replace the dependent variable with the foreign currency bond share in total NFC debt (column 7) and the local currency bond share in NFC debt (column 8). The results are rather intuitive. While a strong domestic enabling environment is very important for domestic market development, it is not in the case of foreign currency bonds which tend to be issued under foreign law. While the US high yield spread interaction appears to matter more for foreign currency bond shares, BD leverage growth matters mostly for domestic bond shares. This suggests that the search for yields drives investors mostly into EM assets that do not entail currency risk while local currency bond market development benefits less strongly than foreign currency bond market liquidity from the risk taking behavior of global banks.

Finally, we dig a bit deeper into the question whether local fundamentals can explain the bond market boom. We already discussed that that local fundamentals were no more important during the post-crisis period than previously in the sense that their interaction terms with the 2010–13 dummy are insignificant throughout the baseline specification and the robustness checks (Tables 3 and 4). However, even though the interaction terms are insignificant, it would be conceivable that fundamentals themselves improved to a degree that would explain part of the upward shift in bond market shares during the post-crisis period. We investigate this hypothesis in a simple exercise illustrated in Table 8. In particular, we aim to understand how much of the average increase in bond shares during the post-crisis period is explained by each regressor. In particular, we multiply the change in the three-year average of each variable (post-crisis vs. before) with the combined (variable and interaction term) coefficient in the baseline specification to arrive at the predicted change in the dependent variable on account of each regressor. Reassuringly, the aggregate predicted change is very close to the actual change in the dependent variable, indicating a good fit. The key point to note, however, is that—of the total predicted change in the dependent variable of 1.99 percentage points—domestic variables only explain about 0.08 percentage points. In other words, the explanatory power of local fundamentals for the bond market boom is very limited at best.

Table 8.Predicted Effects in the Baseline Specification
MeanMedianCombined coefficientPredicted change in dependent
EnforcementAverage 2007-0937.1737.00
Average 2010-1336.9737.00−1.120.23
Lagged concentrationAverage 2007-090.540.43
Average 2010-130.580.50−2.50−0.10
Lagged 3-year CAAverage 2007-09−1.56−1.34
Average 2010-13−2.07−2.490.16−0.08
Lagged Bank capital ratioAverage 2007-099.869.60
Average 2010-1310.3310.300.140.06
High yield spreadAverage 2007-091.601.15
Average 2010-130.990.99−1.941.18
BD leverageAverage 2007-0951.5259.38
Average 2010-1342.0942.11−1.080.22
Dummy for 2010-13Average 2007-090.000.00
Average 2010-131.001.00−0.19−0.28
Sum of predicted changes in dependent variable1.23
Actual mean change in dependent variable1.99
* The sample used in this table excludes all countries with zero observations on the dependent variable during 2007–13
* The sample used in this table excludes all countries with zero observations on the dependent variable during 2007–13

Having shown that local fundamentals cannot explain the bond market boom by themselves, we move on to testing whether the search for yield may impact countries differently depending on the quality of their institutions or macro fundamentals. For this purpose, in Tables 9a and 9b, we interact each of the domestic fundamentals with the high yield spread.27 Throughout Tables 9a and 9b, we find that these interaction terms are all insignificant. In other words, there is no evidence that the sensitivity to global push factors was higher in countries with strong institutions or macro fundamentals. This result strengthens one of the central findings of this paper, namely that local fundamentals neither explain the bond market boom as a whole nor the extent to which markets boomed in one country relative to the other.

Table 9a.Institutional Quality and the Impact of the High Yield Spread
Reg 1Reg 2Reg 3Reg 4
Bond market concentration (lagged)−2.3*** (.68)−2.72*** (.8)−3.05*** (.75)−3.06*** (.73)
Interaction with 2010-13 dummy−.26 (1.47)−.19 (1.64)−.56 (1.6)−.18 (1.71)
Asian Bond Fund dummy4.26** (3.12)4.11** (3.39)3.8** (2.33)4.1** (4.23)
Interaction with 2010-13 dummy1.08 (1.37).68 (1.48).54 (1.67).67 (1.65)
Current account ratio, 3-year average (lagged).16** (.09).19* (.09).2** (.1).2** (.1)
Interaction with 2010-13 dummy−.01 (.1).01 (.12)−.01 (.13)−.01 (.13)
Local bank capital to assets (lagged)−.21 (.22)−.22 (.21)−.21 (.21)−.22 (.21)
Interaction with 2010-13 dummy.35* (.18).35 (.2).38 (.23).42 (.25)
US high yield spread−.39 (1.57).88 (.92)−.31 (.54)−.24 (.37)
Interaction with 2010-13 dummy−1.79** (.75)−1.63* (.85)−1.52* (.81)−1.75* (.86)
US BD leverage growth−1.11** (.53)−1** (.5)−.84 (.53)−1.06** (.5)
Interaction with 2010-13 dummy.01 (1.06)−.72 (1.23)−1.34 (1.42)−.96 (1.42)
Dummy for 2010-13.59 (2.52).56 (2.87).32 (2.93).35 (3.3)
Enforcement procedures−1.13* (.59)
Interaction with HY spread.01 (.04)
Creditor rights.76* (.42)
Interaction with HY spread−.2 (.17)
Credit information.16 (.2)
Interaction with HY spread.02 (.13)
Rule of law.56 (2.66)
Interaction with HY spread−.61 (.61)
Number of observations/units476476476476
ChiSq321.9152.1167.2187.3
Prob > ChiSq0.0000.0000.0000.000
Fraction uncensored0.680.680.680.68
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
Table 9b.Macro Fundamentals and the Impact of the High Yield Spread
Reg 1Reg 2Reg 3Reg 4
Enforcement procedures−1.13* (.59)−1.24 (.73)−1.09 (.61)−1.31** (.61)
Interaction with 2010-13 dummy.02 (.1).01 (.1)−.03 (.11).01 (.1)
Bond market concentration (lagged)−2.3*** (.68)−2.29*** (.68)−1.73*** (.63)−2.42*** (.73)
Interaction with 2010-13 dummy−.18 (1.36)0 (1.31).1 (1.43).19 (1.32)
Asian Bond Fund dummy4.27** (2.64)3.67* (2.83)3.47** (2.55)3.83** (2.67)
Interaction with 2010-13 dummy1.12 (1.46)1.21 (1.43)1.32 (1.58)1.44 (1.58)
Local bank capital to assets (lagged)−.21 (.21)−.13 (.22)−.23 (.2)−.19 (.2)
Interaction with 2010-13 dummy.35** (.19).28 (.21).36** (.18).35** (.19)
US high yield spread−.12 (.29)−.48 (.51)−.66 (.76)−1.19 (1.14)
Interaction with 2010-13 dummy−1.79** (.73)−1.64** (.66)−2.04** (.71)−1.3 (.77)
US BD leverage growth−1.1** (.53)−.88 (.58)−.72 (.59)−1.11** (.52)
Interaction with 2010-13 dummy.03 (.99)−.78 (1.02).21 (1).34 (1.02)
Dummy for 2010-13−.19 (5.7).01 (5.73)1.27 (5.84)−.81 (5.69)
Current account ratio, 3-year average.15* (.1)
Interaction with HY spread.01 (.05)
Reserves in percent of ST external debt0* (0.00)
Interaction with HY spread0.00 (0.00)
External debt in percent of exports of G&S (lagged)−.02** (.01)
Interaction with HY spread0.00 (0.00)
Growth, 3-year average (lagged)−.05 (.19)
Interaction with HY spread.14 (.16)
Number of observations/units476476476476
ChiSq197.2140.7152.1120.9
Prob > ChiSq0.0000.0000.0000.000
Fraction uncensored0.680.680.680.68
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.
* p<0.1, ** p<0.05, *** p<0.01The Table shows marginal effects with standard errors in parentheses. Significance is based on boostrapped confidence intervals.

To summarize, we find that structural domestic factors such as strong fundamentals and an enabling environment are associated with rising bond market development relative to banks. However, the importance of these factors has not increased during the post crisis period. In other words, structural domestic factors generally cannot explain the large increase in bond market borrowing relative to bank borrowing during the post-crisis period. Conversely, it is global push factors that explain the bulk of the EM bond market boom.

VI. Estimation Results Using Panel Quantile Model

While we have shown that global push factors are crucial in explaining the recent bond market boom, we would also like to better understand which factors (if not domestic macro fundamentals or institutions) determine whether global liquidity flowed into some countries rather than others. In particular, we are interested in confirming the main result in the descriptive section, namely that EMs with well developed bond markets were those that benefited most from the global search for yield during the post-crisis period. In order to investigate this question, we now move to the quantile regression setup discussed in the previous section. In particular, we begin by running our baseline specification one more time, now allowing for varying coefficients along different quantiles of the dependent variable. The key question is whether global factors proxying for the search for yield have larger impacts on bond market development in countries with already developed markets. The main focus is therefore on global push factors and the question whether their coefficients become larger in absolute terms for higher quantiles of the dependent variable.

Figure 8 illustrates the estimation results based on our benchmark specification. Due to modest account of censoring in our sample we report the results starting from the 20th quantile. The solid line in each chart shows the coefficient estimates from the 20th to the 90th quantile of the dependent variable. The shaded area indicates the bootstrapped 10 percent confidence interval around the point estimates.

The average marginal effects across quantiles are broadly in line with those found in our baseline specification in Tables 3 to 7. Similarly, the variables that are insignificant in the panel regressions are also insignificant throughout in the quantile regressions. Interestingly, while the estimated coefficient for enforcement procedures (Row 1 in Figure 8a) is downward sloping, it is only statistically significant for lower quantiles. Bond market concentration and the Asian Bond Fund dummy, in turn, do not show very smooth quantile dynamics (Rows 2 and 3 In Figure 8a). As in the panel regressions, the interaction terms of the three variables are insignificant throughout. Figure 8b confirms the findings from the panel regressions with regard to the current account ratio (Row 1) and the local bank capital ratio (Row 2). The former is significant with a positive coefficient while its interaction term is not, and the latter is insignificant while its interaction term is. The interaction term for the bank capital ratio is significant only for lower quantiles, in line with empirical evidence that local bank deleveraging took place mostly in European EMs with relatively bank based financial systems.

Figure 8a.Quantile Regression Setup: Enabling Environment

Figure 8b.Quantile Regression Setup: Macro Fundamentals and Bank Characteristics

The global factors are the main variables of interest at this point of the analysis. The first row of Figure 8c shows the coefficient estimates for the BD leverage growth variable. As in the panel regression, it is significant with a negative coefficient while its interaction term is insignificant. The second row shows coefficient estimates for the high yield spread variable and its interaction. Once again, the panel regression results are confirmed in that the high yield spread and its interaction show (on average) a negative coefficient. While the coefficient for the pre-2010 period is significant only for the highest quantiles, the interaction term is significant for a wider range of quantiles. Interestingly, both the high yield spread and its interaction term show a steep negative slope in the coefficient estimate across quantiles. This suggests that a given drop in risk aversion would increase bond market access more strongly the larger bond market access is relative to the overall size of the financial system. The coefficient on the high yield spread interaction, for instance, is almost four times larger at the 90th percentile than at the 20th percentile. In other words, bond market access increased significantly more as a ratio to total NFC debt in EMs with bond markets that were already relatively large. We interpret this as an indication that flows into EM bond financing driven by falling global risk aversion tend to go into markets that are liquid and allow for easy entry and exit.

Figure 8c.Quantile Regression Setup: Global Factors

As shown in Figure 8, the confidence intervals in the baseline specification are rather wide. One reason is the large number of insignificant variables included. For this reason, we also ran the model based on a specification that drops all insignificant terms. The results are shown in Figure 9. The solid lines in the charts illustrate that the point estimates are very similar to those under the baseline specification. At the same time, the confidence intervals are considerably tighter—especially on the US high yield spread interaction, thus giving further support to our findings.

Figure 9.Quantile Regression Setup: Dropping Insignificant Regressors

In sum, quantile regressions analysis confirms our earlier findings on the relative importance of individual regressors. More importantly, we find that market size is an important conditioning variable that explains a large share of the cross-country variation in bond market development during the post-crisis period.

VII. Discussion

This paper studies the determinants of shifts in EM corporates’ debt composition. Our primary aim is to identify both global and domestic factors that explain why financial systems shift away from bank lending and towards bond market finance. Our focus is on the recent bond market boom and the question why it was stronger in some countries than in others. In particular, we aim to understand whether EMs that experienced the largest booms relative to bank lending were those with strong fundamentals and institutions or whether it was cyclical factors coupled with easy entry and exit that attracted investors. In this context, we also explore the role of cross-border bank linkages.

Our main hypothesis is that the recent boom was driven primarily by the global financial cycle. In particular, we conjecture that the search for yield accounted for most of the variation of bond shares in total corporate debt, with investor interest in specific EMs mostly driven by market size and the associated easy entry and exit. The analysis confirms that the role of bond markets in NFC finance during the post-crisis period increased considerably more in EMs with initially more market based financial systems. While macro fundamentals and strong institutions are shown to be important determinants of bond market development throughout the sample period, their relative role declined substantially during the post-crisis period due to a growing investor focus on market size. We also find evidence for a role for global bank leverage in driving cross-border banking.

The finding that global cyclical factors explain most of the variation in EM bond market development during the post-crisis period is important from a policy perspective. To the extent that bond markets in EMs boomed largely because their large and liquid markets attracted investor flows during a cyclical upswing in the global financial cycle, these countries may be hit severely by capital outflows as the cycle turns. As such, our findings highlight the importance of strong institutions and macro fundamentals in facilitating a gradual diversification of funding sources.

Appendix 1. Galvao et al’s (2013) Three-step Censored Quantile Panel Regression Estimator.

In the first step, a parametric propensity score model is estimated. We use a panel fixed effect logit model as in Galvao et al (2013). We denote the estimated propensity score from the logit model as π^it. The subsample J0 is selected as

The constant cN takes a value strictly between 0 and τ and serves to control for the potential inconsistency of the propensity score estimator π^it by providing a more conservative criterion for the selection of observations. Following Chernozhukov and Hong (2002) we choose cN as the value that minimizes the equivalent of Powel’s (1986) criterion function. In the minimization process we discard the values of cN for which more than 10% observations from J0 were excluded from J1 as this could signal possible misspecification of the separation (subset selection) model or the conditional quantile model (Chernozhukov and Hong, 2002). Such events, however, appeared only a few times and only in the estimation of lower quantiles.

We denote the vector including all regressors as Xit, with the corresponding coefficients denoted as φ. In the second step, a preliminary estimator φ^0 is obtained by minimizing the quantile criterion function over the subsample J0 which is equivalent to minimizing the quantile objective function:

where ρτ(u) = u(τ – 1{u ≤ 0}). The estimator φ^0 is a consistent estimator of the quantile regression parameters, though not necessarily efficient. To improve the efficiency of the estimator, another round of data selection is performed. Define the subsample J1 as:

where ωNT is a small positive number that converges to zero when N and T go infinity and NTωNT is bounded. We choose the ωNT=(1/3)(NT)1/3 th quantile of the estimated quantile function in (4) as in Galvao et al (2013). In the final step, the quantile objective function is minimized over the subset J1 yielding the final estimate φ^.28 The confidence intervals are computed as the corresponding 5th and 95th percentiles of the bootstrapped distribution. We use the bootstrap procedure for censored quantile regression models in Bilias et al (2001) with 200 bootstrap draws to save computing time.

References

    Abrevaya, Jason and ShuShen (2014) Estimation of Censored Panel-data Models with Slope Heterogeneity. Journal of Applied Econometrics29: pp. 523–548.

    Adrian, Tobias, PaoloColla and HyunSong Shin (2012) Which Financial Frictions? Parsing The Evidence From The Financial Crisis of 2007–9. NBER Macroeconomics Annual 2012, Volume 27 edited by DaronAcemoglu, JonathanParker, and MichaelWoodford,May2013, pp. 159 – 214.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Alan Sule, Bo Honoré, LuojiaHu and SorenLeth-Petersen (2014). Estimation of Panel Data Regression Models with Two-Sided Censoring or Truncation. Journal of Econometric Methods3(1): pp. 1–20.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Allen, Franklin Allen, XianGu and OskarKowalewski (2012) Financial Crisis, Structure and Reform. Journal of Banking and Finance36: pp. 2960–2973.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Antzoulatos, Angelos A. (2000) On the Determinants and Resilience of Bond Flows to LDCs, 1990–1995. Journal of International Money and Finance19: pp 399–418.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Arteta, Carlos and GalinaHale (2008) Sovereign Debt Crises and Credit to the Private Sector. Journal of International Economics74: pp 53 – 69.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Avdjiev, Stefan, Michaeul, Chui and HyunSong Shin (2014) Non-financial Corporations from Emerging Market Economies and Capital Flows. BIS, Quarterly Review, December2014, Bank for International Settlements.

    • Search Google Scholar
    • Export Citation

    Becker, Bo and VictoriaIvashina (2014) Cyclicality of Credit Supply: Firm Level Evidence. Journal of Monetary Economics62: pp. 76–93.

    Bilias, Yannis, SongnianChen and ZhiliangYing (2000) Simple Resampling Methods for Censored Regression Quantiles. Journal of Econometrics99, pp. 373–386.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Borensztein, Eduarco, KevinCowan, BarryEichengreen and UgoPanizza (2008) Bond Markets in Latin America. On the Verge of a Big Bang?. The MIT Press.

    • Search Google Scholar
    • Export Citation

    Bremus, Franziska and MarcelFratzcher (2014) Drivers of Structural Change in Cross-Border Banking Since the Global Financial Crisis. Journal of International Money and Finance52: pp. 32–59.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bruno, Valentina and HyunSong Shin (2015) Cross-Border Banking and Global Liquidity. Review of Economic Studies, 82 (2), pp. 535–564.

    Burger, John D., Francis E.Warnock and VeronicaWarnock (2012) Emerging Local Currency Bond Markets. Financial Analysts Journal68: pp.73–93.

    Calvo, Guillermo, LeonardoLiederman and Carme M.Reinhart (1993) Capital Inflows and Real Exchange Rate Appreciation in Latin America. IMF Staff Papers 40, no.1: pp 108–151.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Chan, Eric, MichaelChui, FrankPacker and EliRemolona (2011) Local Currency Bond Markets and the Asian Bond Fund 2 Initiative. BIS Papers63: pp 35–61, Bank for International Settlements.

    • Search Google Scholar
    • Export Citation

    Chuhan, Punam, StijnClaessens and NlanduMamingi (1998) The Equity and Blond Flows to Asia and Latin America: The Role of Global and Country Factors. Journal of Development Economics, 55(2): pp. 439–463

    • Crossref
    • Search Google Scholar
    • Export Citation

    Cetorelli, Nicola and Linda S.Goldberg (2011) Global Banks and International Shock Transmission: Evidence from the Crisis. IMF Economic Review59, no.1: pp 41–76.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Chernozhukov, Victor and HanHong (2002) Three-Step Censored Quantile Regression and Extramarital Affairs. Journal of the American Statistical Association97, no.459, pp: 872–882.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Cerutti, Eugenio, StijnClaessens and DamienPuy (2015) Push Factors and Capital Flows to Emerging Markets: Why Knowing Your Lender Matters More Thank Fundamentals. Forthcoming as IMF Working Paper.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Datta, Sudip, MaiIskandar-Datta and AjayPatel (2000) Some evidence on the uniqueness of initial public debt offerings. Journal of Finance55: pp 715–743.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Demirguc-Kunt, Asli and ErikFeyen and RossLevine (2013) The Evolving Importance of Banks and Securities Markets. World Bank Economic Review, 27(3): pp. 476–490.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Demirguc-Kunt, Asli and ErikFeyen and RossLevine (2001) Financial Structures and Economic Growth: A Cross-Country Comparison of Banks, Markets, and Development. MIT Press, Cambridge, MA.

    • Search Google Scholar
    • Export Citation

    Demirguc-Kunt, Asli and ErikFeyen and RossLevine and VojislavMaksimovic (2002) Funding Growth in Bank-Based and Market-Based Financial Systems: Evidence from Firm Level Data. Journal of Financial Economics65: pp. 337–363.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Denis, David J. and Vassit T.Mihov (2003) The Choice Among Bank Debt, Non-bank Private Debt, and Public Debt: Evidence from New Corporate Borrowings. Journal of Financial Economics70: pp.3–28.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Djankov, Simeon, CaraleeMcLiesh and AndreiShleifer (2007) Private Credit in 129 Countries. Journal of Financial Economics84: pp. 299–329.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Didier, Tatiana, RossLevine and SergioSchmukler (2014) Capital Market Financing, Firm Growth, and Firm Size Distribution. NBER Working Paper 20336.

    • Search Google Scholar
    • Export Citation

    Eichengreen, Barry (2007) The Mother of All Emerging Market Crises. Mimeo. 22August2007http://eml.berkeley.edu/~eichengr/mother_all_emerging.pdf.

    • Search Google Scholar
    • Export Citation

    Eichengreen, Barry and PipatLuengnaruemitchai (2006) Why Doesn’t Asia Have Bigger Bond Markets?BIS Papers30: pp. 40 – 77. Bank for International Settlements.

    • Search Google Scholar
    • Export Citation

    Forbes, Kristin J. and Francis E.Warnock (2012) Debt and Equity Led Capital Flow Episodes. Journal of International Economics, 88(2), pp. 235–251.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Francis, Johanna L., DilekAykut and EugenTereanu (2014) The cost of private debt over the credit cycle. Journal of International Money and Finance41: pp. 146–181.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Fratzscher, Marcel (2012) Capital Flows, Push versus Pull Factors and the Global Financial Crisis. Journal of International Economics88, no.2: pp 341–356.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Gadanecz, Blaise (2004) The Syndicated Loan Market: Structure, Development and Implications. BIS, Quarterly Review, December2004, Bank for International Settlements.

    • Search Google Scholar
    • Export Citation

    Galvao, Antonio F., CarlosLamarche and LuizRenato Lima (2013) Estimation of Censored Quantile Regression for Panel Data with Fixed Effects. Journal of the American Statistical Association108: pp. 1075–1089.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Ghosh, Atish R., Mahvash S.Qureshi, JunKim and JuanZalduendo (2014) Surges. Journal of International Economics92(2), pp 266–285.

    Goldstein, Morris and PhilipTurner (2004) Controlling Currency Mismatches in Emerging Markets, Institute for International Economics, Washington, DC.

    • Search Google Scholar
    • Export Citation

    Gozzi, Juan Carlos., RossLevine, MariaSoledadMartinezPeria, and SergioSchmukler (2014) How Firms use Corporate Bond Markets under Financial Globalization. Journal of Banking and Finance (forthcoming)

    • Search Google Scholar
    • Export Citation

    Greenspan, Alan (1999) Do efficient financial markets mitigate financial crises?October19, 1999, http://www.federalreserve.gov/boarddocs/speeches/1999/19991019.htm.

    • Search Google Scholar
    • Export Citation

    Hale, Galina (2007) Bonds or Loans? The Effect of Macroeconomic Fundamentals. The Economic Journal117: pp. 196–215.

    Hale, Galina and JoaoSantos (2008) The decision to first enter the public bond market: The role of reputation, funding choices, and bank relationships, Journal of Banking and Finance32: pp. 1928–1940.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Hale, Galina and JoaoSantos, PeterJones, and Mark M.Spiegel (2014) The Rise in Home Currency Issuance. Working Paper 2014–19. Federal Reserve Bank of San Francisco.

    • Search Google Scholar
    • Export Citation

    Holmstrom, Bengt and JeanTirole (1997). Financial Intermediation, Loanable Funds, and the Real Sector, The Quarterly Journal of Economics, 112(3), pp. 663–91.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Honoré, Bo E. (1992) Trimmed LAD and Least Squares Estimation of Truncated and Censored Regression Models with Fixed Effects. Econometrica60: pp. 533–565.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Honoré, Bo E. (2008) On Marginal Effects in Semiparametric Censored Regression Models, Working paper, Department of Economics Princeton University.

    • Search Google Scholar
    • Export Citation

    Houston, Joel and ChristopherJames (1996) Bank Information Monopolies and the Mix of Private and Public Debt Claims, Journal of Finance51: pp. 1863–1889.

    • Crossref
    • Search Google Scholar
    • Export Citation

    International Monetary Fund (2014) Global Financial Stability Report, Chapter 1. Improving the balance between financial and economic risk taking.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (2015) Global Financial Stability Report, Chapter 2. International Banking After the Crisis: Increasingly Local and Safer?

    • Search Google Scholar
    • Export Citation

    Japelli, Tullio and MarcoPagano (2002) Information Sharing, Lending and Defaults: Cross-country Evidence. Journal of Banking and Finance26: pp. 2017–2045.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Johnson, Shane A., (1997) An Empirical Analysis of the Determinants of the Corporate Debt Ownership Structure, Journal of Financial and Quantitative Analysis32: pp. 47–69.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Kashyap, Anil K., Jeremy C.Steinn and David W.Wilcox (1993) Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance. The American Economic Review83, no.1: pp. 78–98.

    • Search Google Scholar
    • Export Citation

    Laeven, Luc (2014) The Development of Local Capital Markets: Rationale and Challenges. Working paper No. 234. International Monetary Fund.

    Levine, Ross (2002) Bank-based or Market-based financial systems: Which is better?. Journal of Financial Intermediation11: pp. 398–428.

    Lo Duca, Marco, GiulioNicoletti and AriadnaVidal Martinez (2014) Global Corporate Bond Issuance. What Role for US Quantitative Easing?Working Paper No. 1649. European Central Bank.

    • Search Google Scholar
    • Export Citation

    Mizen, Paul and SerafeimTsoukas (2014) What promotes greater use of the corporate bond market? A study of the issuance behaviour of firms in Asia?. Oxford Economic Papers, 66(1): pp. 227–253.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Ramey, Valerie A. (1992) The Source of Fluctuations in Money: Evidence From Trade Credit. Journal of Monetary Economics30, no.2: pp. 171–193.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Rey, Helene (2013) Dilemma or Trilemma: The Global Financial Cycle and Monetary Policy Independence. Jackson Hole Conference Proceedings, FED Kansas City (forthcoming).

    • Search Google Scholar
    • Export Citation

    Papaioannou, Elias (2009) What drives international financial flows? Politics, institutions and other determinants. Journal of Development Economics88: pp 269–281.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Rajan, Raghuram and AndrewWinton (1995) Covenants and Collateral as Incentives to Monitor. Journal of Finance50: pp. 1113–1146.

    Sahay, Ratna, Martin.Cihak, PapaN’Diaye, AdolfoBarajas, RanBi, DianaAyala, YuanGao, AnnetteKyobe, LamNguyen, ChristianSaborowski, KatsiarynaSvirydenka and SeyedReza Yousefi (2015) Rethinking Financial Deepening. Stability and Growth in Emerging Markets, Staff Discussion Note 15/08, International Monetary Fund.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Shin, Hyun Song (2013) The Second Phase of Global Liquidity and Its Impact on Emerging Economies. November7, 2013, http://www.princeton.edu/~hsshin/www/FRBSF_2013.pdf.

    • Search Google Scholar
    • Export Citation

    Welch, Ivo (1997) Why is Bank Debt Senior? A Theory of Asymmetry and Claim Priority Based on Influence Costs. Review of Financial Studies10: pp. 1203–1236.

    • Crossref
    • Search Google Scholar
    • Export Citation
1

We would like to thank Marco Umberto Alfano, Vivek Arora, Robert Blotevogel, Julian Chow, Luis Cubeddu, Federico Diaz Kalan, Martin Kaufman, Alexander Klemm, Papa N’Diaye, Marijn Otte, Nathan Porter, Yan Sun, Hui Tong and seminar participants at the IMF and the National Bank of Serbia for helpful comments and suggestions. Any remaining errors are our own.

2

The Asian Bond Fund 1 and 2, an initiative of 12 major central banks in Asia-Pacific region, administrated by the BIS, is one example of such policies. Furthermore, the IMF, World Bank and ECB launched in 2007–08 a joint action plan under the G8 umbrella for developing local bond markets in EMs (“Developing Local Bond Markets in Emerging Market Economies and Developing countries”).

3

A central finding in the literature is that both banks and markets have a role to play in providing access to finance and supporting growth (Demirguc-Kunt and Levine, 2001; Levine, 2002; Demirguc-Kunt and Maksimovic, 2002). In particular, while banks tend to be more adept at lending to smaller companies, bond markets hold a comparative advantage in servicing larger, more established companies. At the same time, financial systems become increasingly market based at higher levels of income (Demirguc-Kunt et al, 2012).

4

However, as discussed in more detail below, the experience has shown that the notion of bond markets as “spare tires” may not hold under sufficiently severe disruptions.

5

Cross-border syndicated lending and international private bond issuances, on the other hand, historically show cyclical variation in volumes and interest rates spreads (Francis et al, 2014). The present EM corporate bond boom thus can be in part driven by the temporary easing of financial conditions in global markets.

6

Note also that the correlation between NFC bond market debt divided by GDP and divided by total NFC debt is more than 70 percent.

7

The need to account for censoring arises because the dependent variable is censored at zero while the need to control for unobserved cross-sectional heterogeneity arises from, inter alia, time-invariant drivers of financial development.

8

While we could include (lagged) market size among the regressors, the arising simultaneity problem would be difficult to deal with.

9

Note that demand for bond market borrowing is also likely to be higher in more developed markets with an established issuer base in which the cost of borrowing is likely to be lower due to better information and lower risk for the investor.

10

Interestingly, prior to the crisis, it was EMs with lower access to bond markets that saw bond market access develop more rapidly.

11

Foreign bank exposures to EM financial systems mostly held up well following the GFC although cross-border exposures declined as foreign banks shifted increasingly from centralized to multinational funding models. While cross-border exposures of global banks to European EMs declined strongly following the crisis, overall exposures did not, and the bond market boom was limited.

12

Lo Duca et al (2014) show a positive effect of US quantitative easing policies on NFC bond issuance in a sample of advanced and emerging economies. Analogously, Bremus and Fratzscher (2014) find a positive effect of expansionary monetary policies in advanced economies on cross-border banking flows over the post GFC period. Cerutti et al (2015) explore the sensitivity of different types of capital flows to EMs to global push factors. They find that macroeconomic fundamentals and the nature of the investor base help explain cross-country variation in the impact of global push factors on public and private bond flows (less so in case of bank flows).

13

More information is available under: http://www.dealogic.com/the-platform/unique-content/#debt. Coverage includes Investment Grade Bonds, High Yield Bonds, Supranational Bonds, Sovereign Bonds, Local Authority Bonds, Agency Bonds, Securitization, Covered Bonds, Medium-Term Notes, Preferred Stock, EMTN programmes and trades, and ECP programmes and trades.

14

We include all issuers not classified by Dealogic as pertaining to either the financial or the government sector.

15

Note that this may imply a flawed stock estimate to the extent issuances were not captured by Dealogic or because the borrower defaulted.

16

The stock of outstanding bonds is calculated as the sum of the stocks of outstanding bonds in all relevant currencies, converted into US dollars using the prevailing bilateral exchange rate at any given point in time. Both stock and flow data may be incomplete to the extent that Dealogic DCM does not fully cover issuances of debt or equity securities in a given sector or country. Coverage is likely to be better in more developed economies and more recent years. There is only very limited coverage of short term debt securities (less than one year).

17

The data includes loans to both private and public non-financial corporations.

18

The BIS measure is available here: http://www.bis.org/statistics/credtopriv.htm. The GFSR measure combines data on non-financial corporate domestic debt securities from Bloomberg with data on domestic bank loans (IFS) and external debt (QEDS).

19

The GFSR measure is, moreover available for a significantly shorter horizon.

20

Please note that, for reasons discussed in Section 2, the measure of corporate debt used here is not necessarily in line with official data in countries in which such data is available.

21

Specifically, the quantile regression allows characterizing the impact of each determinant across the entire conditional distribution of the dependent variable which provides a more complete pattern of influences compared to conditional mean estimates.

22

Estimation is done by adapting the pantob.ado file for our framework. We are grateful to Bo Honore for making it available.

23

Among the list of countries in Table 2, Argentina, Belarus, Jamaica and Venezuela are not included in the baseline regressions due to data availability.

24

Djankov et al (2007), for instance, document a positive association between financial development—measured as total banking sector assets—and both contract enforcement and the protection of creditor rights. They also find that the quality of information sharing is especially important in developing countries relative to advanced economies as discussed in Japelli and Pagano (2002). Papaioannou (2009) shows that institutional development is also a significant correlate of international banking inflows. A further potential driver of bond market development relative to banking system development could be financial deregulation which is, however, difficult to quantify in our cross-country setting.

25

Falling risk aversion towards HY fixed income assets may, in part, be driven by global liquidity conditions.

26

The sample correlation between the two global variables in the benchmark specification is -0.22. We also estimated a specification in which the two series are orthogonalized, with no impact on our results.

27

We also tested including the interaction between local fundamentals and the high yield post crisis interaction. However, since the local fundamentals show very little variation during 2010–13, the high yield interaction and its interaction with local fundamentals are too highly correlated to be included alongside each other in the regression. In none of these regressions, however, do the double interaction terms turn out to be significant.

28

Estimation is done by adapting the authors R file to our setup. We are grateful to the authors for making it available.

Other Resources Citing This Publication