Priestley Chao Estimator in Nonparametric Multivariable Kernel Regression in Estimating The Value of Indonesia’s Balance Trade

Abstract

Several researchers have speculated that the model for the Indonesian Trade Balance Value uses a parametric model, but this model has not provided accurate results in determining the actual Indonesian Trade Balance Value. The estimation used is a parametric approach which assumes the data follows a certain pattern. This can result in big mistakes. We propose a nonparametric approach using Kernel functions for data that does not follow a particular pattern and has outliers. The Kernel function used for multivariables is the Gaussian Kernel function with the Priestley-Chao estimator. Analysis of Indonesia’s Trade Balance Data for 2019-2020 using the available data on Indonesia’s Trade Balance Rate, shows that this model is able to estimate with a very small Mean Square Error (MSE) of 0.98 at optimal bandwidth value are h1 =8.72 and h2 = 0.39. Optimum bandwidth selection uses minimum Generalized Cross Validation (GCV). With this bandwidth value, it gives very good estimation results. This model can be used to predict Indonesia’s Trace Balance Accurately on data that does not have a specific pattern and there are outlier data.