Journal of Macroeconomics 39 (2014) 215225

Contents lists available at ScienceDirect

Journal of Macroeconomics
journal homepage: www.elsevier.com/locate/jmacro

What determines external debt tipping points?

Alice Y. Ouyang a,, Ramkishen S. Rajan b
a b

China Academy of Public Finance and Public Policy, Central University of Finance and Economics, #39, S. College Rd., Haidian Dist., Beijing 100081, China School of Public Policy, George Mason University, VA, USA

a r t i c l e

i n f o

a b s t r a c t
This paper is interested in the nexus between external debt and export competiveness. Specically, while we nd that once external debt exceeds a certain threshold it is negatively associated with export growth, we are interested in determining whether the tipping points vary based on country characteristics. We test various hypotheses, including extent of exchange rate exibility, size of foreign exchange reserve holdings, bond market development, degree of banking sector concentration and history of nancial crises. 2013 Elsevier Inc. All rights reserved.

Article history: Received 12 November 2012 Accepted 7 November 2013 Available online 22 November 2013 JEL classication: F21 F34 F43 Keywords: Exchange rate regimes Debt threshold External debt Exports Financial development Reserves

1. Introduction Over the last two decades, the scal positions of many economies have deteriorated rather precipitously, leading to evergrowing levels of public debt (in absolute terms and as share of GDP) and mounting concerns about debt sustainability. However, on a global scale, external indebtedness (public plus private) has been rising quite sharply especially since the mid1990s (Fig. 1). There is a growing body of literature that has attempted to estimate public debt tolerance in the sense that if public debt rises above a certain point it may start to impair economic growth. In a much-cited study, Reinhart and Rogoff (2009, 2010, 2011) highlighted a possible inverted-U relationship between growth and debt. They examined histograms of data from 44 countries over two centuries and found that there appears to be a tipping point at debt-to-GDP ratio of about 90% growth rates decline signicantly beyond the threshold. Many other studies since have revisited this issue using more rigorous threshold econometric techniques with broadly consistent results (see Reinhart et al., 2012 for a succinct summary).1 While most of the literature has focused on public debt, the literature on external debt is rather sparse. While some amount of external leverage may be helpful to growth (i.e. overcoming liquidity constraints), once it exceeds a certain
Corresponding author. Tel.: +86 10 6228 8769; fax: +86 10 6228 8501.
E-mail addresses: Alice.Ouyang@gmail.com (A.Y. Ouyang), rrajan1@gmu.edu (R.S. Rajan). A central result of Reinhart and Rogoff (2010) was that the average growth rates of countries with debt-to-GDP ratios above the 90% threshold are negative. Herndon et al. (2013), in trying to replicate Reinhart and Rogoffs work have now found that the original ReinhartRogoff results were actually driven by coding errors. They nd that once the coding errors are rectied, the estimated mean growth rates of these countries are 2% points more than what Reinhart and Rogoff found but still lower than growth rate below the tipping point.
1

0164-0704/$ - see front matter 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jmacro.2013.11.001

216

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

160 140 120 100 80 60 40 20 0

External Debt Public Debt

1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Fig. 1. Average public and external debt (over GDP) sample countries.

threshold it could have deleterious effects by increasing uncertainty (of future tax increases, possible credit ratings concerns), appreciating the real exchange rates, raising risk premia and thus pushing up overall cost of borrowing. There have been a handful of studies that have estimated a threshold for external debt. For instance, Pattillo et al. (2004) use data for a panel of 93 developing countries for the period 1969 and 1998 and found that external debt is associated with negative per capita growth at above 3540% of GDP and around 160170% of exports.2 This paper builds on the foregoing literature but is more narrowly focused on estimating the possible tipping point for external debt on export growth. More importantly, the paper takes the discussion forward by trying to understand factors that might affect these debt thresholds? We test various hypotheses, including the extent of exchange rate exibility, size of foreign exchange reserve holdings, bond market development, degree of banking sector concentration, and history of nancial crises. The paper is organized as follows: Section 2 discusses the empirical methodology linking export growth with debt thresholds. Section 3 discusses the data, sets out the hypotheses and summarizes main ndings. The nal section concludes the paper. 2. Methodology and data In exploring the impact of debt, much of the literature focuses on overall growth. The problem here is that there is no denite consensus on the exact determinants of growth more generally (Sala-i-Martin et al., 2004). On the other hand, there is a long-established literature highlighting factors that impact export growth more specically, which is our area of interest. In order to investigate the nonlinear relation between export growth and external debt we can set up a quadratic specication as follows:

Exportit ait bX it c1 Debt it c2 Debt i;t eit

where Exportit is the log difference in exports as a share of GDP. Xit is the set of fairly standard control variables in an export function, including global economic growth, real exchange rate appreciation, terms of trade growth, and proxies for supply capacity. Debti,t is the logarithm of external debt variable (share of GDP). We also include a lagged exports-to-GDP growth term to account for inertial effects. To avoid cyclical effects, we use three-year moving averages of all variables. We employ system-GMM to correct for endogeneity of debt and other control variables.3 The empirics are based on a data set of 59 countries (including 27 developed countries and 32 developing countries), covering the sample period from 1980 to 2010 (depending on data availability for each economy) (Table 1). As a robustness check, we re-estimate Eq. (1) using a spline function (IMF, 2011 and Pattillo et al., 2004). The rst step is to estimate the model to ascertain the debt threshold (Debt) and the second step is to estimate the regressions by OLS with xed effects for different thresholds and evaluate which regression produces the highest R-squared.4 The spline function is as below:

2 They nd that the main reason for the negative impact on growth is a decline in efciency of investment (i.e. distortion of resource allocation) as opposed to decline in investment per se (also see Pattillo et al., 2004). Reinhart and Rogoff (2009, 2011) also found a lower threshold of 60% for external debt (government plus private) to GDP for developing market economies. There is a tangential literature on debt Laffer curve which posits that larger debt stocks are associated with lower probabilities of debt repayment (Krugman, 1988; Sachs, 1989). 3 The system-GMM has an advantage over differenced-GMM since it provides a more efcient estimation than the latter, and does not entirely eliminate the cross-country dimension of the data by rst-differencing (such as differenced-GMM) or taking differences with respect to country means (such as xed effects). 4 We try all possible combinations and choose the pair that delivered the best t in terms of the R-squared in order to ensure that the threshold levels are not derived in a non-arbitrary way.

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225 Table 1 List of sample economies.

217

Developed Countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong, Hungary, Iceland, Ireland, Italy, Japan, Korea, Netherland, New Zealand, Norway, Poland, Portugal, Singapore, Spain, Sweden, Switzerland, UK, USA Developing Countries: Algeria, Argentina, Bolivia, Brazil, Bulgaria, Central African Republic, Chile, China, Columbia, Costa Rica, CoteDIvoire, Dominican Republic, Ecuador, Ghana, India, Indonesia, Malaysia, Mexico, Morocco, Nicaragua, Nigeria, Paraguay, Peru, Philippines, Romania. South Africa, Thailand, Tunisia, Turkey, Uruguay, Venezuela, Zambia

Table 2 Variable denition and data sources. Variables REER TOT Export Debt RGDP World GDP Electric Equator Law and order Variable denition REER (2005 = 100) Term of Trade Exports of goods and services (% of GDP) growth Total gross external Debt/GDP (public plus private) (%) Real GDP, US Dollars at constant prices (2005) and constant exchange rates (2005) World Real GDP, US Dollars at constant prices (2005) and constant exchange rates (2005) Electric power consumption (kW h per capita) Distance to the equator (km), measured as the absolute latitude of the capital city (Maximum 6 points) Two measures comprising one risk component. Each subcomponent equals half of the total. The law sub-component assesses the strength and impartiality of the legal system, and the order sub-component assesses popular observance of the law Three exchange regimes, i.e. de facto peg, intermediate regimes and free oating, reclassied from IMF classied exchange rate regimes (Coarse) Amount of public international debt securities (amounts outstanding), as a share of GDP. It covers long-term bonds and notes and money market instruments placed on international markets Assets of ve largest banks as a share of total commercial banking assets. Percentage of the total banking assets that are held by foreign banks. A foreign bank is a bank where 50% or more of its shares are owned by foreigners Data source WDI and IFS UNCDADstat WDI Reinhart and Rogoff (2009) UNCDADstat UNCDADstat WDI European Association for Astronomy Educations Webpage International Country Risk Guide (ICRG)

Exchange rate regimes International public debt securities/GDP Bank asset concentration Foreign bank assets to total bank assets

Ilzetzki et al. (2008) GFDD

GFDD GFDD

Note: WDI stands for World Development Indicator; IFS stands for International Financial Statistics; UNCDADstat stands for United National Conference on Trade and Development (UNCTAD). GFDD stands for Global Financial Development Database.

Exportit ait bX it c1 Debt it c2 Debtit Debt Z eit

where Debt is the debt threshold and Z is a dummy variable for debt ratio that is above the threshold. Both methods can help us identify the turning point where the marginal impact of debt on export growth becomes negative.5 We proxy world demand by the annual growth rate of world real GDP (US dollars at constant prices and constant exchange rates, 2005). We also include the logarithm of potential output to proxy a countrys supply capacity, measured by a HP-trend of real domestic GDP (lagged one-year). As a robustness check, we replace potential output with a set of other proxies for output: electricity, distance to the equator, and law and order index to proxy the infrastructure, geography and institutional environment for a country, respectively.6 We also include two price variables, viz. terms of trade which is dened as the ratio of price of exportable goods over price of importable goods as well as real effective exchange rate (REER). We use gross external debt data (both public and private) from Reinhart and Rogoff (2009). Debt variables are scaled by nominal GDP. Table 2 summarizes the data sources. 3. Empirics 3.1. Are there debt thresholds? The results in Table 3 (Model 1) show that there is a denite inverse U-shaped relationship between export growth and external debt ratio. The tipping point of external debt-to-GDP ratio is around 75% for the whole sample. All controls are statistically signicant and have the expected sign. Higher the world GDP growth, the faster is the domestic export growth (income effect). Both real exchange rate depreciation and terms of trade improvement appear to stimulate domestic export
5 The spline model is not without its problems. In particular it does not overcome the criticism of identication failure under the null of the linear model, i.e. under the null hypothesis of the linear model, c2 = 0, the parameter Debt cannot be not observed. See Hansen (1996) who analyzed this problem and suggested bootstrapped inference. We thank an anonymous referee for pointing this out. 6 We have tried various other proxies for infrastructure and geography (for instance, telephone trafc for the former and proximity to coastal area for the latter). Results remain largely unchanged and are available on request. Choice of institutional proxies is also limited by data availability.

218 Table 3 Impact of external debt on export growth. Dep Var.: export

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

Quadratic function Model 1 Model 2 0.138*** (0.00703) 0.168*** (0.00445) 1.612*** (0.0125) 0.00544*** (0.000434) 0.00253*** (0.000226) 0.00342*** (0.000538) 0.171*** (0.00269) 0.119*** (0.00174) 0.0556*** (0.00229) 0.00654*** (0.000276) 70.16 1078 58 0.391 1.29 [0.197] 0.83 [0.406] 55.01 [0.988] Model 3 0.142*** (0.00968) 0.167*** (0.00732) 1.613*** (0.0117) 0.00175*** (0.000420) 0.00652*** (0.000530) 0.00210*** (0.000207) 0.00263*** (0.000642) 0.171*** (0.00313) 0.118*** (0.00181) 0.0593*** (0.00382) 0.00685*** (0.000426) 75.87 1078 58 0.391 1.28 [0.199] 0.82 [0.410] 54.92 [0.988]

Spline function Model 4 0.0381 (0.0373) 0.488*** (0.0232) 0.942*** (0.193) 0.00587 (0.00585) 0.101*** (0.0285) 0.0962*** (0.00875) 0.0129** (0.00545) 0.0319*** (0.0105) 72.97 1108 59 0.476

Constant Export (1) World GDP growth ln(RGDP_hp) (1) ln(Electric) ln(Equator) Law and Order TOT growth REER growth Debt Debt2 (Debt Debt) Z Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify

0.146*** (0.00367) 0.177*** (0.00158) 1.567*** (0.0103) 0.00119*** (0.000252) 0.163*** (0.00451) 0.121*** (0.00137) 0.0524*** (0.00214) 0.00606*** (0.000310) 75.41 1108 59 0.384 1.49 [0.136] 0.93 [0.353] 56.79 [0.981]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. Signicance at 10%. ** Signicance at 5%. *** Signicant at 1%.

growth (price effects). Supply capacity proxied by lagged potential output is positive and statistically signicant. By way of a robustness check, we replace potential output with infrastructure, geography and institutional determinants of growth (Model 2). Thresholds remain largely unchanged (70%) as do the economic and statistical signicance of the other explanatory variables. Consistent with our prior, the law and order index has a signicant and positive effect on export/GDP growth. Similarly, equator enters with a positive sign, implying that temperate zones tend to export more than tropics.7 However, the proxy for infrastructure (electricity) enters with a negative sign, possibly suggesting that improvements in infrastructure have stronger impact on GDP growth than on exports.8 As a further check, we combine Models 1 and 2, but again results are broadly similar (Model 3). This suggests that, while there may be other supply-side variables that have a marginal impact on export growth, their exclusion ought not to have any signicant impact on the basic relationship of interest, i.e. between export growth and external debt. Given the robustness in results, for the sake of parsimony, we use Model 1 in the next set of results. We also re-run the regressions using spline methodology noted previously. The results are also listed in Table 3 for comparison (Model 4). We identify the thresholds using the regression with the highest R-squared. The turning points are estimated at around 7072%, which are very consistent with the previous ndings using quadratic functions.9
7 Rodrik et al. (2004) suggest that geography variables tend to have weak direct effects on income once institutional variables are included. Lorenz et al. (2005) summarizes the debate on the importance of institutions versus geography. 8 The same result holds if we replace the electricity variable with telephone trafc. We did not have comparable data on some trade-specic infrastructure such as ports. 9 However, the spline model does not allow us to assess the statistical signicance of the thresholds by providing condence intervals. Following Hansen (1996, 2000) the literature has made use of threshold estimation techniques to account for a nonlinear debtgrowth relationship. Some recent papers that estimate debt thresholds using threshold regression models include Caner et al. (2010), Cecchetti et al. (2011), and Kourtellos et al. (2013). While it is not clear if threshold regression is necessarily a superior methodology (see Cordella et al. (2005) for a discussion), an important merit of threshold regression models is that one can compute condence intervals for the threshold parameter. We thank an anonymous referee for pointing this out.

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

219

300 Developed Countries 250 Developing Countries

200

150

100

50

0 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Fig. 2. Average external debt (over GDP) developed and developing countries.

Table 4 Level of economic development (World Bank classication). Dep Var.: Export Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Developed Economies 0.119 (0.0157) 0.250*** (0.0211) 1.688*** (0.0896) 0.000295 (0.000570) 0.0861*** (0.0192) 0.475*** (0.0156) 0.0362*** (0.00583) 0.00381*** (0.000654) 115.78 342 27 0.553 2.66 [0.008] 0.65 [0.515] 24.93 [1.000]
***

Developing Economies 0.227 (0.0271) 0.186*** (0.0243) 1.474*** (0.132) 0.00329*** (0.000796) 0.148*** (0.0225) 0.113*** (0.0146) 0.0896*** (0.0131) 0.0104*** (0.00145) 73.30 766 32 0.396 1.54 [0.123] 0.95 [0.341] 29.67 [1.000]
***

All Samples with developed country dummy 0.188*** (0.00317) 0.172*** (0.00509) 1.582*** (0.00904) 0.00215*** (0.000337) 0.160*** (0.00520) 0.126*** (0.00132) 0.0750*** (0.00150) 0.00919*** (0.000208) 0.0169*** (0.00117) 0.00334*** (0.000267) 0.05802*** [0.000] 0.00584*** [0.000] Developed: 142.86 Developing: 59.15 1108 59 0.385 1.47 [0.142] 0.92 [0.355] 56.33 [0.983]

c1: Debt c2: Debt2 c3: Debt Ddeveloped c4: Debt2 Ddeveloped
Test: c1 + c3 = 0 Test: c2 + c4 = 0 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify

Note: Ddeveloped is the dummy variable for developed economies. The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. Signicance at 10%. Signicance at 5%. *** Signicant at 1%.

220

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

140 120 100 80 60 40 20 0

Fig. 3. Average external debt over GDP in different groups. Note: The levels of reserve/GDP, public debt/GDP, bank asset concentration and FBA ratio are based on their own median values. The pattern looks similar if we classify the groups based on the mean values.

Table 5 Exchange rate regimes (IMF classications). Dep Var.: export Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify De facto Peg 0.0817*** (0.0159) 0.273*** (0.00623) 1.132*** (0.0924) 0.00265*** (0.000714) 0.104*** (0.0111) 0.0874*** (0.00282) 0.0335*** (0.00547) 0.00500*** (0.000603) 28.49 339 39 0.432 1.94 [0.052] 1.39 [0.163] 35.39 [1.000] Intermediate regimes 0.159*** (0.0150) 0.148*** (0.0117) 1.375*** (0.0647) 0.00516*** (0.000652) 0.248*** (0.0117) 0.540*** (0.0153) 0.0502*** (0.00908) 0.00554*** (0.00127) 92.54 410 41 0.520 1.29 [0.197] 0.03 [0.972] 35.76 [1.000] Free oating 0.205*** (0.0154) 0.177*** (0.0134) 1.169*** (0.126) 0.000715* (0.000391) 0.228*** (0.0120) 0.345*** (0.0158) 0.0885*** (0.00814) 0.00998*** (0.00102) 84.48 330 40 0.453 0.07 [0.945] 0.28 [0.780] 36.21 [1.000]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c Turning points of quadratic function can be calculated as exp 2c 1 . 2 # 1-RSS/TSS reported for system GMM. * Signicance at 10%. Signicance at 5%. *** Signicant at 1%.

Fig. 2 divides external debt between developed and developing countries. Interestingly, while both sets of countries had external debt ratios of around 75% of GDP in the mid-1990s, the trajectories of the two since then have been quite different. By 2010, the external debt ratio had declined to just over 30% for developing countries but surged to almost 240% for developed countries. Dividing the sub-samples into developed and developing countries and re-running the regression (Table 4),

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225 Table 6 Foreign reserves/GDP levels. Dep Var.: export Median = 6.74% Reserves/GDP 6 6.74% Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify 0.187*** (0.00629) 0.140*** (0.00387) 1.444*** (0.0364) 0.000320 (0.000318) 0.138*** (0.00491) 0.103*** (0.00201) 0.0788*** (0.00275) 0.00950*** (0.000358) 63.35 462 44 0.375 0.62 [0.536] 0.70 [0.483] 41.32 [1.000] Reserves/GDP > 6.74% 0.134*** (0.0129) 0.168*** (0.00703) 1.625*** (0.0290) 0.00257*** (0.000285) 0.171*** (0.00958) 0.143*** (0.00571) 0.0396*** (0.00625) 0.00403*** (0.000793) 136.78 646 49 0.436 1.94 [0.052] 0.71 [0.476] 40.87 [1.000] Mean = 10.32% Reserves/GDP 6 10.32% 0.185*** (0.00343) 0.191*** (0.00237) 1.168*** (0.0319) 0.00111*** (0.000320) 0.107*** (0.00264) 0.116*** (0.000811) 0.0793*** (0.00194) 0.00956*** (0.000254) 63.47 658 51 0.400 1.20 [0.230] 0.44 [0.661] 47.53 [0.999]

221

Reserves/GDP > 10.32% 0.153*** (0.0114) 0.228*** (0.0153) 1.877*** (0.118) 0.00207*** (0.000344) 0.194*** (0.00843) 0.0868*** (0.00313) 0.0461*** (0.00492) 0.00477*** (0.000515) 125.01 450 41 0.422 2.36 [0.018] 1.01 [0.314] 36.47 [1.000]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. Signicance at 10%. Signicance at 5%. *** Signicant at 1%.

we note that the former has a larger threshold (115%) compared to their developing country counterparts.10 We re-estimated the regression using dummies for developed versus developing. Results remain intact though now the thresholds are about 140% and 60%, respectively. There may be good reasons to expect this outcome. For one, we use gross debt which tends to overstate indebtedness in developed countries generally because of the greater amount of nancial assets owned by developed country governments. In addition, with some exceptions, most of the developed countries also operate reserve currencies (US, UK, Euro, Swiss Franc, Japanese yen) and thus enjoy exorbitant privilege. Nonetheless, the choice of countries based on income levels is somewhat arbitrary and not very helpful from a policy perspective. Can we be more specic in understanding country characteristics that account for differences in debt thresholds? 3.2. What determinants debt thresholds? We outline a set of hypotheses below: (i) Countries with more exible exchange rate regimes are likely to have a smaller portion of their debt that is unhedged, thus making them more able to withstand larger external debt thresholds.11 (ii) Countries with larger buffer of foreign exchange reserves (to withstand sudden stops) are likely to be able to hold larger external debt ratios without detrimental impact on growth.12 (iii) Countries with deeper, more liquid and well-developed nancial markets are more likely to be able to manage larger debt shares without negative real sector repercussions. (iv) Countries in which a few large banks dominate may be able to force domestic banks to hold debt via nancial repression and are less likely to run larger public debt. The captive source of domestic funds may concomitantly also raise their external debt thresholds. In addition, a banking scene dominated by a few large domestic banks may be perceived as carrying an implicit guarantee from the government, thus allowing them to borrow more externally.

This is consistent with the larger literature on public debt which nds a higher threshold for developed countries compared to developing ones. Of course, the size of the debt itself may be endogenous to the type of exchange rate regime, i.e. do xed or exible regimes impose more policy discipline? (Chiu et al., 2007). 12 We also used alternative benchmark for reserves, such as imports. Overall conclusion remains unchanged.
11

10

222

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

Table 7 Bond market development (International Public Debt Securities/GDP). Dep Var.: export Median = 5% Public Debt/GDP 6 5% Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify 0.0864*** (0.0238) 0.0314*** (0.00741) 2.294*** (0.0462) 0.00280*** (0.000813) 0.0335** (0.0144) 0.765*** (0.0118) 0.0134 (0.00920) 0.00190* (0.00102) 33.48 323 39 0.521 2.45 [0.014] 0.21 [0.830] 35.02 [0.947] Public Debt/GDP > 5% 0.145*** (0.0120) 0.196*** (0.00269) 1.285*** (0.0226) 0.000510 (0.000324) 0.183*** (0.00707) 0.115*** (0.00170) 0.0568*** (0.00647) 0.00652*** (0.000872) 77.97 785 50 0.403 1.42 [0.156] 1.06 [0.288] 47.46 [0.999] Mean = 8% Public Debt/GDP 6 8% 0.111*** (0.0156) 0.0110* (0.00655) 2.183*** (0.0413) 0.00265*** (0.000467) 0.0737*** (0.00750) 0.609*** (0.00951) 0.0279*** (0.00657) 0.00365*** (0.000754) 45.53 430 46 0.405 1.59 [0.113] 0.77 [0.438] 43.84 [0.718] Public Debt/GDP > 8% 0.132*** (0.00774) 0.211*** (0.00511) 1.183*** (0.0417) 0.00111*** (0.000275) 0.188*** (0.00599) 0.111*** (0.00195) 0.0513*** (0.00368) 0.00584*** (0.000459) 81.17 678 46 0.415 1.34 [0.180] 0.99 [0.321] 42.02 [1.000]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. * Signicance at 10%. ** Signicance at 5%. *** Signicant at 1%.

(v) Countries with a history of one or more banking or debt crises are likely to have lower debt tolerance than those which have not had any history of such crises. The proxies and sources for each of the hypotheses are listed in Appendix 1. Fig. 3 reveals external debt ratios of countries grouped on the basis of the ve categories noted above. Eyeballing the data reveals that intermediate regimes have lower external debt ratios than peggers or oaters. Countries with low reserves seem to have higher external debt burdens than those with high reserves.13 External debt burdens do not seem to vary too much between low and high public debt countries. Similarly the external debt burdens of low and high bank asset concentration does not appear very different. However, countries with low share of foreign bank asset in the domestic banking system have much higher external debt burdens than those with high extent of foreign bank entry. While countries with no history of banking crisis do not seem to have signicantly higher external debt ratios than those with a history of at least one banking crisis, default risk seems to matter much more. More specically, countries with no history of debt default have much higher external debt ratios than countries with at least one past episode of default. Table 5 separates the sample based on whether a country has a de facto pegged, intermediate or exible exchange rate regime. It reveals clearly that debt tolerance levels for pegged regimes (30% of GDP) are much lower than their managed and freely exible regime counterparts (about 8090%), consistent with our hypothesis above.14 Table 6 divides the sample into those with below mean/median reserves (as a share of GDP) ratio and those above. Countries with above-average reserves have higher external debt thresholds than those below average, consistent with our hypothesis. Table 7 uses the size of international public debt outstanding as an indication of the development of bond market. Countries with relatively more developed bond markets (i.e. above sample average) have higher debt thresholds than those with less developed bond markets. Table 8 reveals that the greater the bank asset concentration (assets of ve largest banks) the higher the debt tolerance level. We also used an alternative proxy, viz. the share of foreign banks in total assets (Table 9). Once again, the greater the share of foreign banks, the lower is the debt tolerance. These results are consistent with the nancial repression hypothesis noted previously. Table 10 divides the sample into countries with no history of nancial crises

13 While this may seem counter-intuitive, most developed countries tend to have low reserves and high external debt, while developing countries are opposite. 14 The intermediate regime consists of managed oaters and crawling bands. The same result holds if we use two broader categories of pegs (combining de facto pegs and crawling bands) and exible regimes (combining oaters and managed oaters). Results are available on request.

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225 Table 8 Extent of bank asset concentration (ve largest banks). Dep Var.: export Median = 70% Bank asset concen. 6 70% Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify 0.112*** (0.0119) 0.00679*** (0.00170) 1.809*** (0.0648) 0.00275*** (0.000710) 0.175*** (0.00419) 0.745*** (0.00904) 0.0336*** (0.00437) 0.00435*** (0.000447) 47.72 340 43 0.480 1.66 [0.097] 2.64 [0.008] 36.01 [0.692] Bank asset concen. > 70% 0.120*** (0.00605) 0.221*** (0.00511) 1.551*** (0.0159) 0.00102*** (0.000232) 0.164*** (0.00384) 0.103*** (0.000612) 0.0421*** (0.00338) 0.00501*** (0.000464) 67.35 768 53 0.432 1.20 [0.229] 0.67 [0.505] 52.19 [0.995] Mean = 70% Bank asset concen. 6 70% 0.112*** (0.0119) 0.00679*** (0.00170) 1.809*** (0.0648) 0.00275*** (0.000710) 0.175*** (0.00419) 0.745*** (0.00904) 0.0336*** (0.00437) 0.00435*** (0.000447) 47.72 340 43 0.480 1.66 [0.097] 2.64 [0.008] 36.01 [0.692]

223

Bank asset concen. > 70% 0.120*** (0.00605) 0.221*** (0.00511) 1.551*** (0.0159) 0.00102*** (0.000232) 0.164*** (0.00384) 0.103*** (0.000612) 0.0421*** (0.00338) 0.00501*** (0.000464) 67.35 768 53 0.432 1.20 [0.229] 0.67 [0.505] 52.19 [0.995]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. Signicance at 10%. Signicance at 5%. *** Signicant at 1%. Table 9 Share of Foreign Bank Assets (FBA) to total bank assets. Dep Var.: export Median = 19% FBA ratio 6 19% Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify 0.134*** (0.0118) 0.0932*** (0.0174) 2.164*** (0.143) 0.00346** (0.00163) 0.150*** (0.0292) 0.798*** (0.0268) 0.0357*** (0.00401) 0.00420*** (0.000562) 70.47 193 35 0.369 0.96 [0.336] 0.89 [0.372] 28.21 [0.105] FBA ratio > 19% 0.155*** (0.00399) 0.195*** (0.00219) 1.478*** (0.00703) 0.00126*** (0.000216) 0.173*** (0.00159) 0.119*** (0.000369) 0.0606*** (0.00170) 0.00735*** (0.000219) 62.00 915 59 0.394 1.49 [0.135] 1.05 [0.296] 55.32 [0.987] Mean = 29% FBA ratio 6 29% 0.114*** (0.0127) 0.0993*** (0.0155) 2.248*** (0.149) 0.00158 (0.00143) 0.141*** (0.0268) 0.815*** (0.0164) 0.0277*** (0.00502) 0.00314*** (0.000678) 82.83 245 42 0.462 1.19 [0.236] 1.68 [0.093] 30.15 [0.068] FBA ratio > 29% 0.132*** (0.00245) 0.217*** (0.00180) 1.326*** (0.00563) 0.00110*** (0.000193) 0.178*** (0.00116) 0.116*** (0.000174) 0.0535*** (0.000752) 0.00671*** (8.73e05) 54.13 863 59 0.407 1.51 [0.130] 0.98 [0.328] 55.63 [0.986]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. Signicance at 10%. ** Signicance at 5%. *** Signicant at 1%.

224 Table 10 History of banking crisis or debt crisis. Dep Var.: export Constant Export (1) World GDP growth ln(RGDP_hp) (1) TOT growth REER growth Debt Debt2 Turning Point: Debt (%) Observations Number of id R-squared# Arellano-Bond Test for AR(1) Arellano-Bond Test for AR(2) Hansen J-Test of over-identify

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

No banking crisis 1.799** (0.867) 0.422 (0.320) 1.889*** (0.532) 0.0135** (0.00679) 0.341*** (0.0744) 0.830*** (0.102) 0.721** (0.362) 0.0746* (0.0384) 125.86 152 8 0.082 0.93 [0.352] 0.49 [0.622] 0.00 [1.00]

Number of banking crisis P 1 0.171*** (0.00596) 0.178*** (0.00395) 1.608*** (0.00991) 0.00195*** (0.000164) 0.157*** (0.00536) 0.119*** (0.00143) 0.0657*** (0.00306) 0.00791*** (0.000364) 63.66 956 51 0.388 1.53 [0.125] 0.95 [0.344] 50.27 [0.997]

No debt crisis 0.101*** (0.00748) 0.168*** (0.0325) 1.826*** (0.0956) 0.00008 (0.000589) 0.0815*** (0.0193) 0.508*** (0.0172) 0.0257*** (0.00257) 0.00270*** (0.000335) 116.89 432 30 0.538 1.42 [0.156] 0.70 [0.486] 26.49 [1.00]

Number of debt crisis P 1 0.258*** (0.0188) 0.191*** (0.0335) 1.253*** (0.188) 0.00294*** (0.00105) 0.177*** (0.0385) 0.127*** (0.0139) 0.109*** (0.00816) 0.0130*** (0.00110) 68.03 638 27 0.396 1.44 [0.150] 0.94 [0.348] 24.28 [1.000]

The p-values of Arellano-Bond test for serial correlation and Hansen J-Test of over-identifying restrictions are reported in baskets. All the difference-in-Hansen tests cannot reject the null hypothesis, indicating that the instruments are exogenous. c1 Turning points of quadratic function can be calculated as exp 2c2 . # 1-RSS/TSS reported for system GMM. * Signicance at 10%. ** Signicance at 5%. *** Signicant at 1%.

versus others. Credit history clearly matters. Countries with no history of banking crises or debt crises clearly have higher debt tolerance levels than those with such a history. 4. Concluding remarks This paper has investigated the nexus between external debt and export growth. While the existence of debt thresholds is conrmed, this paper has gone further by exploring their determinants, something that seems missing in the existing literature. We nd that countries with exible currencies, greater reserve holdings, solid credit history (no previous nancial crisis), well-developed bond market but highly concentrated banking system (limited extent of deregulation) are more likely to have accumulated larger levels of external debt (as a share of GDP) without negative impact on export growth. Acknowledgements The authors appreciate the detailed comments by an anonymous referee. Valuable research assistance by Sasidaran Gopalan is gratefully acknowledged. The usual disclaimer applies. Financial support from Social Science Foundation of Ministry of Education of China (10YJC790200) and CUFE Youth Innovation Fund are greatly acknowledged. Appendix A Based on IMF classied exchange rate regimes (Coarse) (Ilzetzki et al., 2008), we re-classify exchange rate regimes into three categories: de facto peg, intermediate regimes and free oating. The countries are dened as de facto exchange rate regimes if the IMF classication code equals 1. The countries are dened as intermediate regimes if the code equals 2 or 3. The countries are dened as free oating if the code values are above 3. With regard to other hypotheses, total outstanding international public debt securities to GDP is used to proxy the development of bond market, while both of assets of ve largest banks as a share of total commercial banking assets and the share of foreign bank assets to total bank assets are used to proxy the development of nancial markets. The ratio of foreign reserves over GDP and the proxies for bond market and nancial market development are taken from Global Financial Development Database. Banking crisis identication is based Laeven and Valencia (2010), while debt crisis identication is based on Reinhart and Rogoff (2009). A banking crisis is dened as systemic if two conditions are met: (1) signicant signs of nancial distress in the banking system (as indicated

A.Y. Ouyang, R.S. Rajan / Journal of Macroeconomics 39 (2014) 215225

225

by signicant bank runs, losses in the banking system, and/or bank liquidations) and (2) signicant banking policy intervention measures in response to signicant losses in the banking system. The external debt crisis is dened as the failure of a government to meet a principal or interest payment on the due date (or within the specied grace period). References
Caner, M., Grennes, T., Koehler-Geib, F., 2010. Finding the Tipping Point When Sovereign Debt Turns Bad. Policy Research Working Paper Series, 5391. The World Bank. Cecchetti, S., Mohanty, M., Zampolli, F., 2011. The future of public debt. In: Gokarn, S. (Ed.), Challenges to Central Banking in the Context of the Financial Crisis. Academic Foundation, New Delhi, pp. 183217. Chiu, E.M.P., Dechsakulthorn, S.B.J., Walter, S., Walton, J.C., Willett, T.D., 2007. The Discipline Effects of Fixed Exchange Rates: The Distinction between Hard and Soft Pegs. Mimeo, Claremont Graduate University. Cordella, T., Ricci, L.A., Ruiz-Arranz, M., 2005. Debt Overhang or Debt Irrelevance? Revisiting the Debt-Growth Link. IMF Working Paper, WP/05/223, December. Hansen, B., 1996. Inference when a nuisance parameter is not identied under the null hypothesis. Econometrica 64, 413430. Hansen, B., 2000. Sample splitting and threshold estimation. Econometrica 68, 575603. Herndon, T., Ash, M., Polin, R., 2013. Does High Public Debt Consistently Stie Economic Growth? A Critique of Reinhart and Rogoff. Political Economy Research Institute, University of Amherst, MA. Ilzetzki, E., Reinhart, C.M., Rogoff, K.S., 2008. Exchange Rate Arrangements Entering the 21st Century: Which Anchor Will Hold? University of Maryland and Harvard University (unpublished). IMF, 2011. Modernizing the Framework for Fiscal Policy and Public Debt Sustainability Analysis. IMF Policy Paper. August 5, 2011. <http://www.imf.org/ external/np/pp/eng/2011/080511.pdf>. Kourtellos, A., Stengos, T., Tan, C.M., 2013. The Effect of Public Debt on Growth in Multiple Regimes, Journal of Macroeconomics. Part A 38, 3543. Krugman, P., 1988. Financing versus forgiving a debt overhang. Journal of Development Economics 29, 253268. Laeven, L., Valencia, F., 2010. Resolution of Banking Crises: The Good, the Bad, and the Ugly. IMF Working Paper, WP/10/146. Lorenz, A., Hemmer, H.R., Ahlfeld, S., 2005. The Economic Growth Debate Geography versus Institutions: Is There Anything Really New? Institute for Development Economics at Justus-Liebig-University, Giessen. Pattillo, C., Poirson, H., Ricci, L., 2004. What Are the Channels Through which External Debt Affects Growth. IMF Working Paper 04/15, IMF. Reinhart, C.M., Rogoff, K., 2009. This Time is Different: Eight Hundred Centuries of Financial Folly. Princeton University Press, Princeton. Reinhart, C.M., Rogoff, K., 2010. Growth in a Time of Debt, NBER Working Paper No.15639. Reinhart, C.M., Rogoff, K., 2011. A Decade of Debt. Peterson Institute of International Economics, September. Reinhart, C., Reinhart, V., Rogoff, K., 2012. Debt Overhangs: Past and Present. Working Paper No. 18015, NBER. Rodrik, D., Subramaniam, A., Trebbi, F., 2004. Institutions rule: the primacy of institutions over geography and integration in economic development. Journal of Economic Growth 9, 131165. Sachs, J., 1989. The debt overhang of developing countries. In: Calvo, A.G., Findlay, R., Kouri, P., Braga de Macedo, J. (Eds.), Debt Stabilization and Development: Essays in Memory of Carlos Diaz Alejandro. Basil Blackwell, Oxford, pp. 80102. Sala-i-Martin, X., Doppelhofer, G., Miller, R., 2004. Determinants of long-term growth: a Bayesian Averaging of Classical Estimates (BACE) approach. American Economic Review 94, 813835.