We analyze the Nikkei daily stock index and verify how wavelets can help in identifying, estimating and predicting its volatility features. While we study the conditional mean and variance dynamics, by utilizing statistical parametric inference techniques, we also decompose the observed signal with a data de-noising procedure. We thus investigate how wavelets discriminate among information at different resolution levels and we attempt to understand whether the de-noised data lead to a better identification of the underlying volatility process. We find that the wavelet data pre-processing strategy, by reducing the measurement error of the observed data, is useful for improving the volatility prediction power.