Local polynomial modeling and bandwidth selection for time-varying linear models

This paper proposes a local polynomial modeling approach and bandwidth selection algorithm for estimating time-varying linear models (TVLM). The time-varying coefficients of a TVLM are modeled locally by polynomials and estimated using least-squares estimation with a kernel having a certain bandwidth or support. Asymptotic behavior of the proposed estimator is established and it shows that there exists an optimal local bandwidth which minimizes the weighted mean squared error (MSE). A data-driven variable bandwidth selection method is also proposed to estimate this optimal bandwidth. Simulation results show that the proposed LPM method with adaptive bandwidth selection outperforms conventional TVLM identification methods in a large variety of testing conditions.