FEASIBLE GENERALIZED LEAST SQUARE (FGLS) METHOD TO ESTIMATE THE PARAMETERS IN THE MULTIPLE LINEAR REGRESSION MODEL WITH AUTOCORRELATED ERRORS

• Main Author: Rezky Friesta Payu (NIM: 20113047) , Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/20328
• Institution: Institut Teknologi Bandung

An important assumption in the multiple linear regression model is uncorrelated errors. Offences againts the assumption; called auto-correlated errors, makes the estimator of Ordinary Least Square (OLS) is not Best Linear Unbiased Estimators (BLUE) because it does not have minimum variance among the other estimators. The main purposed of this research is to investigate the Feasible Generalized Least Square (FGLS) method as the remedial measures with considering two procesures of estimating correlation coefficient; those are the Cochrane-Orcutt (CO) iterative procedure and Theil-Nagar procedure. In using FGLS, one has to be careful in omitting the first observation. Therefore, it is advisable to transform the first observation according to the Prais-Winsten transformation. This thesis combines each procedures with using the Prais-Winsten transformation and compared with OLS. The MSE of the resulting intercept and slope parameter estimates of those methods are compared with OLS estimates via simulation data. In addition, that is also comparing those methods based on the resulting of forecasting using the real data set. It is obtained that FGLS method is better for large sample, while for small sample, OLS method tend to be better. Meanwhile, transformation of Prais-Winsten gives significance to small sample size.