Home Bias of Korean Resident Bond Investors: The Role of FX Hedging

ABSTRACT

This paper analyzes the home bias of Korean resident bond investors, as their overseas bond investments have rapidly increased after 2000. For this purpose, we introduce the Won/invested currency swap basis (interest rate differential adjusted FX swap return) and the Won/invested currency uncovered basis (interest rate differential adjusted expected FX spot return) along with the interest rate gap based on CIP and UCIP respectively in order to address the return factors in detail. The model coefficients are estimated by the static pooled OLS and the dynamic system GMM over the period of 2002–2018 and 2009–2018 respectively. With these estimation results, we have obtained the following implications. First, it is expected that the level of home bias will decline as Korean resident overseas bond investments increase, as the factors that have traditionally caused home bias continue to mitigate. Second, when Korean residents invest in overseas bond markets, the FX risk has partially hedged at the country level. This suggests that the rapid in/out flows for overseas bond transactions make a significant impact on the FX swap rates as well as spot rates. Third, both groups of investors have taken into heavy consideration both the FX swap basis (or the FX uncovered basis) related to the invested currency/USD as well as to Won/USD. Our findings suggest that the policy makers need to reconsider their conventional monitoring methods which are focused on the Won/USD transactions.

KEYWORDS:

• Home bias
• bond investor
• FX swap basis
• FX uncovered basis
• FX hedging

JEL:

• G11
• G15
• G20
• F30

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. For more conceptual and theoretical background information related to this section, refer to Baba, Packer, and Nagano (2008), Borio et al. (2016).

2. Readers may refer to Ian, Piet, and Rosanne (2018) for additional information on the calculation methods, and pros and cons of the HB measures.

3. The oversea bond investment of Korean residents increased rapidly from 4.5 billion dollars at the end of 2000, to 194.2 billion dollars at the end of 2018. As of the end of 2018, the major investment countries were USA (39.8%), France (8.5%), UK (8.3%), Brazil (4.7%), Australia (4.1%). Japan (3.4%), Cayman Islands (2.8%), Luxembourg (2.6%), Netherlands (2.6%) among others.

4. As recommended by a reviewer, if the inflation hedge-related variable is explicitly added into the estimation model in future studies, the economic meaningfulness of the estimated result would be improved greatly. Please refer to the Supplemental Material section for short definitions and sources of variables Table S3.

5. Buying USD spot <selling Won spot> & selling USD forward <buying Won forward>

6. Selling USD spot <buying invested currency spot> & buying USD forward <selling invested currency forward>

7. By the insightful advice by the anonymous Reviewer, we were also able to confirm that there was a statistically significant structural change by adding the dummy variables based on before and after GFC into the models of 2002–2018.

8. Bond, Hoeffler, and Temple (2001) showed that $\hat{\phi }$ tends to exhibit upward bias in OLS estimation, and exhibit downward bias in the difference GMM estimation, and recommended the system GMM when the $\hat{\phi }$ of difference GMM is close to or below the $\hat{\phi }$ of the fixed effect estimation.

9. Han & Kim (2014)showed that the estimated coefficients have considerable bias and inefficiency if the constant term is omitted when estimating the dynamic panel GMM model.

10. As Blundell and Bond (1998)recommend for the accounting of the possibility that there are too many instrumental variables when using the system GMM estimation, we conduct the difference in Hansen tests for the instrumental variables which were added when estimating with the system GMM. These test statistics did not reject the null hypothesis, implying that the choice of instrumental variables was also adequate.

11. According to the persuasive comments of an anonymous Reviewer, we checked up the likelihood that the difference in the estimation coefficients in Table 3 and Table 6 may be caused by multicollinearity between explanatory variables.