We generalize the Cooper and Kaplanis (1994) methodology for estimating the costs that could reconcile international portfolio holdings with CAPM predictions. First, we simultaneously estimate inward and outward investment costs and even interactions between home and host country. Second, the risk aversion parameter is estimated rather than postulated. Third, we detect costs for domestic investments. We find that the home bias in equity portfolios is related to a mixture of market frictions,

more »

... h as information asymmetries, institutional factors and explicit costs. Over the period 2001-2004, the average implicit investment costs range from 0.26 (US) to 16 (Turkey) percent per annum. JEL classification: G11, G15, F36 Keywords: international portfolio diversification, information asymmetries, international CAPM In this paper we build on Cooper and Kaplanis' (CK, 1994) idea of estimating a set of deadweight costs that can reconcile actual international portfolio weights with the predictions of the International CAPM (InCAPM). The CK approach provides point estimates of each country's cost of either inward or outward investments, conditional on a postulated value of relative risk aversion. In contrast, we adopt a regression approach: home bias depends on deadweight costs which, in turn, depend on regressors related to international transaction and information costs. The key advantage of this route is that we can measure far more. First, we can estimate simultaneously a home-country cost vector, a host-country one, and even interactions, thus ending up with a complete matrix of costs for all combinations of home and host countries. Second, we are able to estimate relative risk aversion rather than having to assume one. Third, we get more than just point estimates: we can in fact distinguish between coincidences or transient factors and more substantial ones, and we can therefore obtain confidence intervals and significance tests as to both the level of the implied overall deadweight costs and the contribution of the various variables to those overall costs. Fourth, we allow for costs of domestic investment too; and while we can still only estimate the differential cost of investing abroad versus at home, we are able to demonstrate that the domestic-investment cost does vary over countries and years and, therefore, is non-zero. The existence of such a cost may explain part of the equity premium puzzle or the divergence between risk-aversion estimates from mean returns versus from intertemporal studies or inflation-hedging asset demand. Fifth, we do not have to assume that the capitalization of domestic equity equals the wealth of a country. Lastly, in our computations we also tie up some loose ends in the original CK methodology, like the role of fixed-interest securities denominated in the various currencies. None of this would have been possible without the better data that have become available since CK's work. We were also inspired by recent work on home bias-Berkel (2004), Coval and Moskowitz (1999), Faruqee, Lee and Yan (2004) , or Portes and Rey (1999)-that tries to directly explain capital flows or deviations between actual portfolio holdings and InCAPM predictions. In a way, we even merge both approaches. Conducting this type of research firmly within portfolio theory instead of via stand-alone regression offers a neat and rigorous way of controlling for expectations, to which mean-variance portfolio weights are very sensitive, and for the correlations between each and every country's index. Also, our two-layer approach,