Under fairly general assumptions, expected stock returns are a linear combination of two accounting-based characteristics—book to market and ROE. Empirical estimates based on this relation predict the cross section of out-of-sample returns in 26 of 29 international equity markets, with a highly significant average slope coefficient of 1.05. In sharp contrast, standard factor-model-based proxies fail to exhibit predictive power internationally. We show analytically and empirically that the importance of ROE in forecasting returns depends on the quality of accounting information. Overall, a tractable accounting-based valuation model provides a unifying framework for obtaining reliable proxies of expected returns worldwide.