Multiscale stochastic dynamics in finance

Semimartingale probabilistic setups lead to very useful volatility estimation. The integrated volatility can be consistently estimated by the realized one according to the quadratic variation principle, even if the convergence speed can result relatively slow, depending on noise and market microstructure effects. We show, experimentally, that scale transforms based on wavelets and the corresponding cumulative periodogram estimators may offer comparable numerical performance in measuring the quadratic variation limit, thus minimizing the discrepancy between realized and integrated volatility.