Implementation of differential evolution algorithm for mars modeling optimization on Indonesia composite index data

Abstract

Nonparametric regression is applied as an alternative to parametric regression if the data to be studied does not meet all the assumptions of parametric regression. This study aims to model using the Multivariate Adaptive Regression Splines model integrated with the Differential Evolution algorithm on the Indonesia Composite Index data. Indonesia Composite Index reflects the overall performance of stocks on the Indonesia Stock Exchange, which is influenced by various macro and microeconomic factors that lead to erratic patterns. Thus, the MARS method was chosen because of its ability to capture nonlinear relationships and interactions among the independent variables. At the same time, the Differential Evolution algorithm was implemented to optimize the selection of model parameters. The MARS model is found by combining Basis Function (BF), Maximum Interaction (MI), and Minimum Observation (MO), with concern to the minimum value of Generalized Cross-Validation (GCV). The study results using MARS-DE indicate that the optimal combination of models is BF = 35, MI = 1, and MO = 2 with a GCV value of 0.0068. In this study, 6 independent variables were used. The study results showed that the monthly exchange rate (USD-IDR), inflation, interest rates in Indonesia, the Dow Jones stock index, and the Shanghai Stock Exchange Composite index are independent variables that affect the Indonesia Composite Index (ICI). However, the Nikkei 225 stock index does not influence the Indonesia Composite Index. The resulting model is based on the smallest Mean Squared Error value of 0.0034.