International bond diversification strategies: the impact of currency, country, and credit risk

Abstract

We investigate the incremental role of emerging market debt and corporate bonds in internationally diversified government bond portfolios. Contrary to earlier results, we find that international diversification among government bonds does not yield significant diversification benefits. This result is obtained using mean–variance spanning and intersection tests, with restrictions for short sales, both for currency unhedged and hedged internationally developed market government bonds. Currency hedged international corporate bonds in turn do offer some diversification benefits, and emerging market debt, in particular, significantly shifts the mean–variance frontier for a developed market investor. Since especially unconstrained mean–variance spanning and intersection tests can indicate significant diversification benefits, but lead to frontier portfolios with extreme weights, we also consider some ex-ante global government bond portfolio strategies. We find that passive global benchmarks such as GDP-weighed government bond portfolios perform quite well within developed countries.

Keywords:

• international bond diversification
• mean–variance spanning and intersection
• emerging market debt
• corporate bond

JEL Classification :

• G11
• G15
• F34

Acknowledgements

Highly valuable comments by two anonymous referees are gratefully acknowledged. We are grateful for comments received at the Joint Finance Research Seminar in Helsinki, the conference on Asset Management and International Capital Markets in Frankfurt, the European Financial Management Association's Annual Meeting in Athens, and the Southern Finance Association's meeting in Florida.

Notes

Among the pioneers in this line are, e.g. Grauer and Hakansson (1987), Jorion (1989), and Odier and Solnik (1993). Although some analysis of bond portfolios is also included in such papers, the number of fixed income assets is typically low.

See, e.g. Dewachter and Maes (2001) studying US government bonds in UK portfolios, or Varotto (2003) studying various forms of diversification solely within the asset class of corporate bonds.

For equities, Chiou (2008) has shown that, as expected, developing countries benefit more from both regional and global diversification.

Interestingly, Hunter and Simon (2005) find that the mean and volatility spillovers are much weaker among major international bond markets than those between equity markets, and that the bond diversification benefits do not decrease during periods of high volatility or after extremely negative bond returns. These results are also supported by those of Polwitoon and Tawatnuntachai (2006) for international bond funds.

Later evidence for the euro-area by Baele et al (2004) indicates fully integrated money markets, a reasonably high integration in the government and corporate bond markets, whereas the credit market is the least integrated.

Fresh indirect international evidence on the existence of significant diversification benefits from foreign debt is offered by studies on home bias. Sørensen et al (2007) show that the home bias is actually smaller in the debt portfolios of many countries than in the corresponding equity portfolios. If the markets behave rationally, this empirical evidence would indicate higher actual diversification benefits of foreign debt, and/or lower transaction costs or informational asymmetries for foreign debt securities. Evidence from fund managers’ views by Lütje and Menkhoff (2007) also indicate lower informational advantages and less relative optimism for international bond managers, again supporting less home bias for bond investments. Polwitoon and Tawatnuntachai (2006) report international diversification benefits from global bond funds for a US investor.

Bekaert and Urias (1996) investigate small sample properties of mean-variance spanning test methods and find that regression-based methods appear to have better power than GMM-based alternatives.

The weights in the GDP index are annually rebalanced such that a particular year's weight is based on previous year's GDP converted to USD as reported by the International Monetary Fund (IMF) for each country in Group 1.

When analyzing diversification benefits from another investor perspective (UK, German, or Japanese investors), we correspondingly use the MSCI index for that country.

For example, the correlation between our choice for the US government bonds (the MSCI US 7–10 Year Sovereign Debt Index) and a potential alternative, the Merrill Lynch US Treasury 7–10 Year Index, is as high as 0.9958 during our study period.

Even if Merrill Lynch had the best geographical coverage and longest time-series of corporate indices, it was not possible to find exactly the same type of index for all countries. The indices for the USA and EMU are industrial/non-financial, while the indices for UK, Japan, and Canada are broad corporate indices. The ratings (Standard & Poor's) of the issuers included are BBB for the USA, A for Japan, and investment grade for the EMU, the UK, and Canada.

The return for the Lebanese index in November and December 1997 would else have been +151% and −148%, respectively. As we found no financial news to explain this, and as the price index data series in contrast to the total return index data used in our study did not exhibit the same extreme fluctuations, the November index value was smoothed. All other large returns were accompanied with either a simultaneous equity market return in the same direction, and/or a substantial weakening of the local currency.

Germany is the only country in our sample entering the Eurozone from the beginning of year 1999. For the German Government bond index in Group 1, we, therefore, use the DEM to USD currency rate up to the end of 1998, and the EUR to USD rate from 1999 onwards.

If Japan would be excluded, the average intra-group correlation for developed markets (Group 1) would rise to 0.71. For the group containing corporate bonds (Group 3), the exclusion of Japan would also increase the correlation to 0.71.

Regressing a time trend on the correlation for each group results in positive regression coefficients between 0.0040 and 0.0045. The coefficient estimates are statistically significant on at least a 5% level using heteroskedasticity and autocorrelation consistent standard errors (Andrews 1991).

The Japanese results can be available from the authors. Of these analyzed four perspectives, only one represents the Euro area. However, the integration of the money and bond markets in the Euro area has been fast, as reported in, e.g. Baele et al (2004). When we calculated the correlations between total return bond indices that could be found for our study period for both France, Italy, and Germany (Citigroup Government Bond Index, maturity 7–10 years), the correlation was 0.99 between the returns on German and French bonds during our study period, and 0.94 between each of these two against Italian bonds. Since also the currency risk for these countries is the same from 1999 onwards, and the money market benchmark (EURIBOR) is the same, we believe that our results for German investors can be rather representative also for the other Euro area investors.

In line with related literature we assume a perfect hedge. These calculations are detailed in the Appendix.

The main reason for reduced diversification benefits seems to lie in the more recent data used in our study. During our time period (from 1997 to May 2006), the average correlation between US and foreign government bonds was 0.67 (ranging from 0.62 to 0.82 for the different country pairs, see column 1 in Table 4, Panel A), while the correlations between the USA and three foreign bonds analyzed in the Hunter and Simon (2004) study (for the years from 1992 to September 2002) vary between 0.15 (USA–Japan) and 0.56 (USA–Germany) on a comparable basis. Given higher correlations in our study, there are less diversification benefits in the first place. Although the removal of exchange rate does lead to improvements in Sharpe ratios, these are not large enough to give significant diversification benefits.

While insignificant in the total sample period our sub-sample analysis shows that emerging market bonds do offer significant diversification benefits (intersection rejected) against all four benchmarks in the 2002–2006 sub-period. This is in line with the spanning test rejections and is due to the relatively higher Sharpe ratios in 2002–2006 compared with 1997–2001.

For example, considering the ML benchmark investor: He should abandon his benchmark offering a Sharpe ratio of 0.053 totally for a maximum Sharpe ratio (, column 4 of Table 9) portfolio consisting of 74.6% India, 13.4% Mexico, 3.5% Russia, 5.6% Turkey, and 2.8% Venezuela. Note that currency risk and variability in local short risk free rates used in computing excess returns can significantly alter return and risk levels compared with local raw returns in Table 3.

We also compared these benefits with measures for transaction costs in the bond and currencies markets. For example, Chordia, Sarkar, and Subrahmanyam (2005) report an average quoted spread of $0.022 per $100 par value for US government bonds around the start of our study period, indicating pretty low transaction costs. For corporate bonds, the bid-ask spreads are wider, e.g. Chen, Lesmond, and Wei (2007) report spreads of about 31 and 44 basis points (bp) for BBB bonds in maturity classes of 1–7 and 7–15 years. Developing market bonds exhibit a larger variation in spreads, ranging from low spreads of 5 bp for certain countries (e.g. Mexico, Chile, and Brazil) to 50–100 bp for Venezuela (Jeanneau and Tovar 2006). The hedging costs, in turn, can be approached by looking at bid-ask spreads on the currency markets. Hau, Killeen, and Moore (2002) report median post-euro spreads between 5 and 8 bp on the spot market for currencies such as USD, JPY, GBP, and EUR. The forward spreads are typically wider than the spot spreads. However, our analysis of these transaction costs as compared with the diversification benefits from our study indicate that there still would be sizable bond diversification benefits net of transaction costs.

During our study period, however, relatively few international bond market ETFs existed. According to Gebler and Tucker (2003), the first bond ETFs (both Canadian) were launched in 2000. In 2003, there were only four bond ETFs in the USA.

These normality-related robustness results are available from the authors upon request.