This paper studies the volatility in financial market returns. We obtain strong evidences in favor of a stochastic volatility model, including an MA(1) term in errors. Also, we estimate companion models build up in the framework of FIGARCH/HYGARCH class of models. Various methods for persistence checks are used. The results suggest that mutual information might be a valid alternative for persistence checking: significant deviations of mutual information from zero can be viewed as an evidence of long-run memory. We illustrate the case of Bucharest Stock Exchange's BET index, which displays a significant persistence in returns.