PEMODELAN REGRESI NONPARAMETRIK DATA LONGITUDINAL MENGGUNAKAN POLINOMIAL LOKAL (Studi Kasus: Harga Penutupan Saham pada Kelompok Harga Saham Periode Januari 2012 – April 2015)

Abstract

Stocks are securities that can be bought or sold by individuals or institutions as a sign of participating or possessing a company in the amount of its proportions. From the lens of market capitalization values, stocks are divided into 3 groups: large capitalization (Big-Cap), medium capitalization (Mid-Cap) and small capitalization (Small-Cap). Longitudinal data is observation which is conducted as n subjects that are independent to each subject observed repeatedly in different periods dependently. Smoothing technique used to estimate the nonparametric regression model in longitudinal data is local polynomial estimator. Local polynomial estimator can be obtained by WLS (Weighted Least Square) methods. Local polynomial estimator is very dependent on optimal bandwidth. Determination of the optimal bandwidth can be obtained by using GCV (Generalized Cross Validation) method. Among the Gaussian kernel, Triangle kernel, Epanechnikov kernel and Biweight kernel, it is obtained the best model using Gaussian kernel. Based on the application of the model simultaneously, it is obtained coefficient of determination of 97,80174% and MSE values of 0,03053464. Using Gaussian kernel, MAPE out sample of data is obtained as 11,74493%. Keywords: Longitudinal Data, Local Polynomial, Stocks