Variable selection in sparse GLARMA models

In this paper, we propose a novel and efficient two-stage variable selection approach for sparse GLARMA models, which are pervasive modelling discrete-valued time series. Our consists in iteratively combining the estimation of autoregressive moving average (ARMA) coefficients models with regularized methods designed performing regression Generalized Linear Models (GLM). We first establish consistency ARMA part coefficient estimators specific case. Then, explain how to efficiently implement our approach. Finally, assess performance methodology using synthetic data, compare it alternative illustrate on an example real-world application. approach, is implemented GlarmaVarSel R package, very attractive since benefits from low computational load able outperform other terms recovering non-null coefficients.