The predictability of excess returns in the emerging bond markets

Abstract

This study examines the relationships that exist between excess bond returns and global and country-specific factors, focusing on a sample of 12 developing countries. Our results show a significantly negative autocorrelation with regard to the excess returns of bonds in the emerging markets; with growth in the size of the local bond market, there is a corresponding increase in the excess bond returns. For most of the developing economies, with an increase in emerging market bond returns, there are discernible reductions in the level of domestic interest rate and increases in the volatility of bond returns. A higher sovereign bond spread predicts higher excess returns for emerging market bonds. Overall, we find that world factors have relatively less predictive power in the emerging market bonds.

Keywords:

• emerging market bonds
• foreign capital flow
• predictability

JEL Classification::

• F21
• F30
• G15

Notes

¹ The JP Morgan EMBI Global is a total returns index which tracks trading in the market in US dollar-denominated Brady bonds, loans and Eurobonds with an outstanding face value of at least US$500 million. As documented by Jostova (2006), Brady bonds are the primary financing vehicle in the markets of the emerging economies, accounting for 80% of their total external government debt. The JP Morgan index calculates individual bond returns based on daily changes in the bid prices and the interest earned according to coupon accruals and payment conventions. The EMBI Global for each of the countries examined is derived by weighting the daily returns of all individual bonds in proportion to their market capitalization.

² Examples include, amongst others, Bekaert and Hodrick (1992), Ferson and Harvey (1993), Ilmanen (1995) and Barr and Priestley (2004).

³ Examples include, amongst others, Harvey (1995), Bekaert and Harvey (1995, 2003).

⁴ On the other hand, however, any increase in the market interest rate will lead to capital losses for bond investments.

⁵ Erb et al. (1996) recommend the use of risk ratings in emerging market countries when explaining the expected excess bond returns. Min et al. (2003) find that macroeconomic fundamentals play a significant role in the determination of the bond spreads of 11 emerging market countries, and that liquidity and solvency variables explain most of the spread variations. Jostova (2006) finds that the emerging bond spread can be predicted primarily by credit spread deviations from fundamentals. Jüttner et al. (2006) conclude that the price of oil significantly affects the bond returns of oil importing countries, whereas country risk affects the bond returns of oil exporting countries. Lin et al. (2007) indicate that macroeconomic fundamental factors make greater contributions to emerging bond returns than world stock and bond excess returns. Furthermore, Dailami et al. (2008) point out that US interest rates affect emerging market bond spreads in different ways, noting that this is largely dependent upon the debt levels of the different countries, and that moderate debtors suffer little impact from an increase in US interest rates, and vice versa.

⁶ In both Ilmanen (1995) and Barr and Priestley (2004), emphasis is placed on the importance of global factors for predicting developed market bond returns. Ilmanen (1995) finds that the excess returns of government bonds in the US, Canada, Japan, Germany, France and the UK are linked to world excess bond returns, but not world excess stock returns. Barr and Priestley (2004) conclude that government bond markets in the US, the UK, Japan, Germany and Canada are integrated with the world markets.

⁷ Both the openness and further development of national financial markets are likely to contribute to economic growth through the removal of friction and barriers to trade, as well as the more efficient allocation of capital. As noted by La Porta et al. (1997), Levine (1999) and Rajan and Zingales (2003), the continuing development of the national financial markets particularly reflects the economic, political and sociological situation of a country, as well as its legal and regulatory environment.

⁸ Since the available data frequency on GDP is quarterly, we use cubic spline interpolation to convert the data to monthly frequency. This method assigns each value in the low frequency series to the last high frequency observation associated with the low frequency period, and then replaces all intermediate points on a natural cubic spline connecting all the points.

⁹ We derive the size of each country's bond market from unpublished data on international bonds from the Bank for International Settlements (BIS). Detailed calculations of the size of local bond market are provided in BIS (2002) and Burger and Warnock (2007).

¹⁰ We use International Finance Corporation (IFC) indices to calculate the monthly returns for the emerging equity markets. The IFC provides value-weighted indices of a representative sample of equities in each country covering at least 60% of the market capitalization. Connolly et al. (2005) find that stock market volatility has a negative correlation with that between bond and stock returns, whilst Jostova (2006) reports a negative relationship between local equity index returns and EMBI spreads.

¹¹ Jostova (2006) points out that high interest rates may indicate significant local debt as the government increases its demand for loanable funds, leading to further underinvestment. Therefore, a higher interest rate will hamper the future growth of the economy and reduce the government's capacity to service its debt.

¹² See, amongst others, Bekaert and Harvey (2000), Min et al. (2003), Jostova (2006) and Jüttner et al. (2006).

¹³ A brief summary of the ICRG risk ratings is provided in the Appendix (Table A1).

¹⁴ For example, Arora and Cerisola (2001) use sovereign bond spreads as a proxy for country risk to quantify the impact of monetary policy on country risk.

¹⁵ We would like to thank an anonymous referee for raising this issue about local crises.

¹⁶ For most developing countries, debt or sovereign default is usually accompanied with currency devaluations.

¹⁷ According to Kaminsky (2006, Table 2), dates of crises are February 2002 for Argentina; January 1999 for Brazil; September 1997, September 1998, and August 1999 for Colombia; December 1997 for the Philippines; July 1997, January 1998, September 1999, and July 2000 for Thailand; and February 2001 for Turkey. We also consider the Russian crisis starting in August 1998, following Dailami et al. (2008). As addressed by Dailami et al. (2008), the period of crisis with very high excess returns has been limited to a few months, when a resolution has been in sight, therefore we use 6 months following the crisis date as the local crisis period. Argentina's 2002 default is an exception, and we use 24 months as the duration crisis for Argentina. For more detail, Del Nergo and Kay (2002) also provide a chronicle of the Argentina's crisis.

¹⁸ As noted above, our choice of a small crisis window (6 months) limits the number of crisis periods. Moreover, our reliance on Kaminsky (2006) for the crisis dates may miss some crises, since not all the countries in our sample are on her list. Due to the limitation of observation number in the crisis periods, we consider a jump in the intercept instead of specifying possibly different coefficients on country-specific and global factors for crisis and noncrisis periods.

¹⁹ See for example, Barr and Priestley (2004), Jostova (2006), Jüttner et al. (2006), Lin et al. (2007) and Dailami et al. (2008).

²⁰ The Chicago Board Options Exchange Volatility Index (CBOE VIX) can gauge the future volatility of the stock market. The VIX is commonly regarded as a market ‘fear index’ that reflects investor sentiment.

²¹ As noted by Barr and Priestley (2004), Jüttner et al. (2006) and Lin et al. (2007).

²² Examples include Eichengreen and Mody (1998), McGuire and Schrijvers (2003) and Ferrucci and Taylor (2004).

²³ See for example, Arora and Cerisola (2001), McGuire and Schrijvers (2003), Min et al. (2003) and Jüttner et al. (2006).

²⁴ We obtain similar results for these four dummy variables using various period length settings.

²⁵ All of the variables used in this study are defined in the Appendix (Table A2).

²⁶ In order to avoid the problem of multicollinearity, we use only one variable from each set of high correlation variables in our subsequent analysis.

²⁷ The statistics are available from the US Treasury International Capital (TIC) system. See Tesar and Werner (1994, 1995) for more comprehensive details.

²⁸ During the period from January 1987 to June 2007, the SD in the foreign bond flow (foreign equity flow) is found to be 1241.34 (643.70).

²⁹ As a result of the limitations on data availability, we follow Bekaert and Harvey (2000) to set initial ownership at zero. Although, as noted in Bekaert and Harvey (2000), such treatment may cause certain problems, there is no other way available to us to calculate the capital accumulation on the emerging bond markets.

³⁰ We calculate the test statistics by adjusting the serial correlation and heterogeneity in the error term. We first carry out the UD max and WD max tests to determine whether at least one break is present, and then determine the number of breaks based on the sup-F statistics and estimate the break dates through a global minimization of the sum of the squared residuals.