Abstract : This work focuses on the study of financial time series using multifractal processes including processes MRW (Multifractal Random Walk), introduced by Bacry, and Muzy DeLoura. In this context, it addresses the problem of extreme events, the approximate limits of small intermittent and statistical estimation of model parameters log-normal MRW. The results obtained allow the use of the MRW model for risk prediction (prediction of conditional volatility and Value-at-Risk conditional). A final section offers a more exploratory modeling of intraday financial time series modeling, consistent with the multifractal approach and to improve risk prediction. Results The Digital! ues obtained on real data show that the log-normal modμele MRW provides predictions of risk of much better quality than those obtained using more traditional econometric models (GARCH and tGARCH).