This paper uses financial and macroeconomic variables to predict currency returns, by using a two-step procedure. The first step consists of a cointegration equation that explains the exchange rate level as a function of global and domestic financial factors. The second step estimates an error-correction equation, for modeling the expected returns. This approach is a factor model analysis, where a Lasso derived technique is used for variable selection. This paper will focus on the five most frequently traded Latin American currencies, Brazilian Real (BRL), Chilean Peso (CLP), Colombian Peso (COL), Mexican Peso (MXN) and Peruvian Sol (PEN), during the time horizon from December 2001 until February 2016. The first finding is that the Global Exchange Rate Factor offers information about the exchange rate movements. In addition, this paper shows that commodity, equity prices and domestic risk premium are important variables for explaining exchange rates. Moreover, it confirms the existing results for the carry and slope variables.