We propose a new method for the study of the correlations of fluctuations in time-series, using the multifractal analysis. The investigation is performed on the partial data sets obtained from the original series, for positive and negative fluctuations.
The method offers the possibility of identifying correlations either in each sub-series or between them. The application of the method to financial timeseries allows estimation on the predictability of evolution. As case study, the exchange rates Leu-USD and Leu-EUR are considered.