The ARCH type of models is a notorious family of models proven to be suitable for predicting financial returns. Their notoriety flourished after Bollerslev (1986) developed the econometric Generalized ARCH model (GARCH). This paper provides a presentation of the main characteristics of the modeling of financial returns with the objective to calibrate an EGARCH (Exponential GARCH) model for the logarithmic returns of the Romanian composite index BET-C on the stocks listed at the Bucharest Stock Exchange. We continue a previous study Lupu (2005) to model the statistical properties of these returns in comparison with the main non-normality properties found in previous research for the US stock index. We found that these properties are generally held on the Romanian market and this provides us reasons to trust the opportunity of an EGARCH model. The article provides the testing of the predictive power of this model for the Romanian index by calibrating the model and then evaluate its performance on an out of sample test.