Low correlations between asset returns increase the portfolio diversification benefits and for US investors emerging market equities are one such class of assets. Several studies indicate that the correlations between asset returns are time varying and using unconditional estimates of correlation in a portfolio optimization model can result in misallocation of assets. To overcome this problem we use multivariate Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models to estimate the time-varying correlations and use the same in portfolio optimization models. Ex-post return calculations show that unrestricted portfolios created with emerging stock indices and S&P 500 index outperform the S&P 500 index by itself. Since investors exhibit strong home bias in their portfolio choice, restricted optimization models are tested. Results indicate that if the total investment in emerging markets is restricted, a minimum investment of 20% in emerging markets is required to obtain significant diversification benefit. With investments in each of the emerging market restricted to less than 3%, there was no significant diversification benefit. Copyright © 2007 John Wiley & Sons, Ltd.