Robust Local Polynomial Regression for Irregular Time Series

Nonparametric curve fitting and prediction of unequally spaced nonstationary time series with additive outliers and jumps in the level.

Nonparametric curve-fitting and predictions of unequally spaced (irregularly sampled) nonstationary time series with additive outliers and structural jumps in level. The estimation techniques is based on Gaussian kernels and local polynomial regression (LPR), with robust censoring of prediction errors. The filter is estimated iteratively with a pseudolinear algorithm; it is resistant to outliers and is jump-tracking. Cross validation selections of the smoothing coefficients is also performed.

The statistical methods are developed in the paper by: Carlo Grillenzoni (2009), "Robust Non-parametric Smoothing of Non-Stationary Time-Series". Journal of Statistical Computation & Simulation, 79(4), 379-393 https://www.tandfonline.com/doi/abs/10.1080/00949650701786390