چه عواملی باعث سوگیری دارایی اصلی می شود و چگونه باید آن را اندازه گیری کرد ؟

Many explanations of home asset bias involve intuitions that should affect the data inputs used by investors in optimizing portfolios: (1) transaction costs affecting expected returns, (2) perceived riskiness of foreign assets affecting standard deviations, and (3) omitted assets affecting correlations. Only the first area has been examined empirically. We examine the empirical feasibility of the second and third explanations, as well as whether combining explanations can fully account for home asset bias. We find that no single set of adjustments can explain home asset bias by itself. Combining adjustments is promising but the implied correlation structure among asset returns is puzzling.

Previous research on international mean-variance portfolio choice consistently predicts that investors should put much more of their wealth into foreign assets than they actually do; i.e., investors appear to engage in a sub-optimal degree ofinternational diversification. This preference for domestic over foreign assets is often referred to as home asset bias, and it has been documented for investors in Ž several countries by many authors including French and Poterba, 1990, 1991; . Tesar and Werner, 1992, 1994, 1995; Cooper and Kaplanis, 1994 . Many researchers wishing to explain home asset bias have focused on the inputs to the investor’s optimization model, examining such issues as the presence of differential transaction costs between countries, additional sources of risk for foreign investing, or the explicit omission of assets from the investor’s opportunity set. We will reference and discuss their work in some detail in following sections, but here we note that the empirical implementation of each of the potential explanations for bias requires directly or indirectly adjusting the inputs to the standard mean-variance model of portfolio choice—the returns, variances, and Ž. covariances correlations of the assets in the investment list. Only the first category of these explanations has been examined empirically and results show that adjusting returns does not explain the home asset bias. 1 In this paper we first examine the empirical feasibility of the remaining explanations by determining how much each of the statistical inputs would have to be individually altered to explain home asset bias. We find that the input values have to be outside reasonable ranges. We then ask whether there are combinations of the explanations that can explain observed bias towards domestic assets without resorting to extreme parameter values. In these cases, the required adjustments to expected returns and standard deviations are reasonable in magnitude, but the implied correlations for the omitted asset do not correspond to any single identifiable asset. In particular, human capital, foreign or domestic bonds, and foreign or domestic real estate do not have the same characteristics as the omitted asset we describe in our exercise. We conclude that completely erasing home asset bias requires a model with multiple omitted assets, as well as transaction costs and risk adjustments. The reader should not view the analysis as merely a tautological exercise. We are not trying to determine whether adjusting model inputs can explain home asset bias; mathematically, many such combinations exist. Rather, we are asking if the necessary adjustments to model inputs are consistent with observable data and r or strong intuitions based on earlier research. The paper is organized as follows. In Section 2, we briefly describe the standard model we use to measure home asset bias. Section 3 addresses the empirical feasibility of the three individual categories of explanations for the home asset bias: adjustments to expected returns, adjustments to standard deviations andadjustments to correlations resulting from the inclusion of an additional asset. In Section 4, we examine the potential for combinations of parameter adjustments to explain home asset bias. We also discuss what the results suggest about asset categories that should be included in a more complete model. We conclude with a summary section.

In this paper, we examine the measurement of and explanations for home asset bias, the documented tendency of investors to A over-invest B in home assets, even in the face of evidence that their risk-return portfolio profile can be improved by less domestic investment. Home asset bias is measured by comparing actual portfolio weights to optimal weights, so a natural question to ask is whether we have the right model and data inputs to generate the optimal weights for the comparison. Previous work has documented that a general model with PPP deviations must be used to allow for all diversification possibilities. However, earlier work also documents that home asset bias persists, even when such a model is used. The remaining explanations of home asset bias can be categorized as leading to Ž. adjustments in the statistical inputs to the portfolio optimization model: 1 adjustments to expected returns to reflect transaction costs and other market-seg- Ž. menting factors, 2 adjustments to standard deviations to reflect the additional Ž. perceived riskiness of foreign assets, and 3 adjustments to correlations by including a previously omitted asset in the investment menu. These explanations of home asset bias are all based on compelling intuitions, but it is an empirical question to determine which, if any, may explain observed behavior. Only the category of adjustments to returns has been examined empirically in prior work, and home asset bias remained. In this paper, we provide the first evidence on the empirical feasibility of explanations based on adjusting standard deviations or adding an omitted asset. We also are the first to empirically examine whether combining these explanationscan fully account for home asset bias. When we consider each explanation individually, we find that the adjustments to statistical inputs are unreasonable in magnitude—either much larger than historical values or inconsistent with the assets involved. However, when we combine adjustments to returns, standard deviations and correlations, we are able to find a combination of parameters whose magnitudes are more reasonable. This conclusion must be qualified because the characteristics of the omitted asset—its return volatility, its weight in the portfolio and its pattern of implied correlations with equity returns—do not correspond to any single asset we can imagine. In particular, we can rule out human capital, domestic or foreign bonds and domestic or foreign real estate as being the omitted asset. Ž. We conclude that the failure to find a single unidentified omitted asset that accounts for the home asset bias means that we should consider multiple omitted assets. However, adding multiple assets means that many more correlations will be involved so that we cannot anticipate what the effects on home asset bias will be. We cannot uniquely solve for implied correlations in our framework when more than one asset is omitted, so we do not explore this possibility here. We believe this is an important avenue for future research.