Author name:
علي سيف الدين عبد الحافظ
Supervisor name:
ظافر حسين رشيد النجار
Abstract:
ان نماذج المعاملات المتغيرة زمنيا (Time - varying coefficient models) هي ادوات مهمة جدا لشرح الديناميكية في كثير من العلوم مثل الاقتصادية,المالية,السياسية,علم الاوبئة...الخ ,اي كيفية تغير معاملات المتغيرات زمنيا , وكذلك استعمالها هيكل خفي (Hidden Stru | Time - varying coefficient models are very significant instruments to explain the dynamics in many sciences such as economics, financial, politics, epidemiology, etc. That means, how the variables coefficients vary chronologically, and how they use Hidden Structure to avoid the Curse of Dimensionality for the nonparametric models. In the last ten years, varying coefficient models encountered deep and exciting developments on the theoretical and practical aspects particularly in the longitudinal data applications.In this research, some of the nonparametric technologies have been presented to estimate the coefficients functions, which are varying chronologically for the nonparametric marginal model of the balanced longitudinal data that characterized by observations obtained through (n) from the independent subjects, each one of them is measured repeatedly by group of specific time points (m). Although the measurements are independent among the different subjects; they are mostly connected within each subject, and the applied technologies are Local Linear Polynomial kernel (LLPK) and the Cubic Smoothing Splines (CSS).To avoid the problems of dimensionality, thick computation and the programming effort, the two - steps method has been used to estimate the coefficients functions by using the two former technologies. Since, the two - steps method depends, in estimation, on (OLS) method, which is sensitive for the existence of abnormality in data or contamination of error; it has been proposed using some robust methods such as LAD & M to strengthen the two - steps method towards the abnormality and contamination of error as well as using the robust method in any step or both steps to give high flexibility in terms of applying the robust methods. In this research, simulation experiments have been performed to imitate the used models, with verifying the performance of the traditional and robust methods for both of CSS & LLPK technologies by using two criteria, for different sample sizes and disparity levels. One of the important aims of this research is to create a dynamics for the data, i.e., how variables coefficients vary over time for a group of balanced longitudinal data for nine economic sectors in Iraq (agriculture and forestry sector, mining and quarrying sector, industrial sector, electricity and water sector, construction sector, transport and communication sector, trade of retail and wholesale sector, finance and insurance sector, and social development services sector for the period (1990 - 2009)); formulated according to time - varying coefficients, for the nonparametric marginal model of Cobb - Douglas Production Function. The most important conclusions of this research are : using Cubic Smoothing Splines is better than using Local Linear kernel, as well as the progress of the proposed robust methods over the traditional estimation methods in all contamination cases. Concerning the practical side, it has shown weakness of capital investments comparing to the volume of revenues, and the Gross Domestic Production (GDP) of the economic sectors depends mainly on the volume of employment, as well as the volume of revenues since(1990 - 2008) are considered decreasing revenues. i.e., any double of the capital and employment volume would not lead to double of the Gross Domestic Production (GDP), except in (2009) which witnesses increasing revenues, i.e., any double of the capital and employment volume would lead to double of the Gross Domestic Production (GDP).
