The aim of this research is to evaluate to what extent the actual level and dynamics of financial dollarization prevalent in Latin America can be explained by the portfolio approach proposed by Ize and Levi Yeyati (2003). Inasmuch as the result of a portfolio optimization is sensible to the estimation of the variance-covariance matrix, this paper considers several alternatives to estimate the expected volatility by means of historical data so as to resemble two polar cases: short-memory and long-memory adaptive expectations. The main finding is that, for the period 1995-2005, financial dollarization in Latin America can be partially explained by the MVP ratio only if long-memory of past volatility events is assumed. Additional results are reported for highly dollarized countries.