Estimation of Missing Values for BL (p, 0, p, p) Time Series Models with Student-t Innovations

摘要: In this study optimal linear estimators of missing values for bilinear time series models BL (p, 0, p, p) whose innovations have a student-t distribution are derived by minimizing the h-steps-ahead dispersion error. Data used in was simulated using R Statistical Software where 100 samples size 500 each were generated model (1, 1, 1). The data numbered from 1 to 500. sample, three positions 48, 293 and 496 selected at random value these points removed create artificial values. For comparison purposes, two commonly non-parametric techniques neural network (ANN) exponential smoothing (EXP) estimates also computed. performance criteria ascertain efficiency mean squared error (MSE) Mean Absolute Deviation (MAD). found that ANN most efficient estimating with innovations. recommends use student errors.