Piecewise model selection for non-stationary long memory data

Li Song (1) , Pascal Bondon (1)

1 L2S - Laboratoire des signaux et systèmes

We study the model selection problem for a locally stationary long memory signal by dividing the signal into stationary blocks. In this piecewise model, the number and the locations of the break points are unknown as well as the parameters of each regime. We propose a model selection criterion based on the minimum description length (MDL) principle. Monte Carlo simulations show that our criterion performs better than BIC and the criterion proposed by Davis et al. (2006). The application of our method to the Nile river data for the years 622-1284 AD confirms previous studies which conclude that a structural break exists around the year 722 AD

Domaines: Statistiques [math.ST]; Théorie [stat.TH]; Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Li Song, Pascal Bondon. Piecewise model selection for non-stationary long memory data. GRETSI 2011, Sep 2011, Bordeaux, France. pp.1-4. ⟨hal-00662562⟩