Regression Forward avec fenêtres Tempo-Frequentielles roulantes par ondelettes discretes et continues -Une application à la Droite de Marché -

The Rolling-Regression are currently used to study the parameters stability over time. In finance, we can analyse the time evolutions of systematic risk relaxing the constant-Beta hypothesis. This method can be associated with a wavelet decomposition of the variables in order to the parameters stability of frequency regression. Then, we compare continuous and discrete wavelets methodologies of rolling regression with the standard rolling regression. The discrete methods are based on time-frequency window but we compare if we have to use it on the wavelets filter output or directly on the series and realize the wavelet decomposition at each step of the window. The continuous method is based on wavelets coherence-phase. We use daily data of AXA returns and the CAC 40 index from 2005 to 2015. We show that the differences between discrete methods are more important at Low-Frequencies and we compare the results with the Continuous Time-Frequency Betas.