Content area
Full text
I nvestors everywhere exhibit ahome bias that leads them to heavily overweight domestic stocks relative to foreign stocks.1 The home bias is not confined to "naïve" individual investors, but also affects the choices of investment managers (Coval and Moskowitz [1999]). Familiarity with and relative competence in analyzing local firms have been cited as possible reasons for this bias. Other researchers argue that failure to diversify internationally may not be suboptimal considering the tendency of markets to fall at the same time in the most volatile periods.
This article does not attempt to find the most plausible explanations for home bias. Our goal is to provide a simulation of international diversification for a long-term investor whose investment horizon is measured in decades. Our interest in this topic is motivated by some aspects of portfolio theory that may have been overlooked in the literature. In particular, a long-term benefit from holding assets is weakly correlated with the risk reduction that comes from moving away from the most volatile assets. The diversification benefit raises a portfolio's geometric mean return, which also reduces the likelihood of underperforming a given target return as the holding period is lengthened. In other words, given the one-to-one relationship between geometric mean and terminal wealth, long-term investors may have much to gain by curbing the home bias and diversifying internationally. Our empirical results confirm this conjecture.
Our study is based on monthly returns of 15 stock markets--Australia, Belgium, France, Germany, Hong Kong, Italy, Japan, Netherlands, Norway, Singapore, Spain, Sweden, Switzerland, the United Kingdom, and the United States--over the period from January 1975 to December 2014. Historical data are of limited use for the purpose of studying long-horizon returns because, even with a century of data, there are only 10 independent decade-long holding period returns, and most markets have far less than a century of data available. To address this limitation, we simulated holding period returns of up to 20 years using the block bootstrap method pioneered by Knusch [1989]. The method is nonparametric, except for the choice of block size, to preserve unspecified forms of time series dependencies within the blocks. Following previous studies that use the block bootstrap (e.g., Hansson and Persson [2000], Fong and Koh [2015]), we set the...