The Model of Artificial Neural Network and Nonparametric MARS Regression for Indonesian Composite Index

Abstract

Abstract The Indonesian composite stock price index is an indicator of changes in stock prices as a guide for investors to invest in reducing risk. The regression model for Indonesian Composite Index (ICI) has the response variable as stock prices with fluctuation behavior and several financial predictor variables, the model tends to violate the assumptions of normality, homoscedasticity, autocorrelation and multicollinearity. This problem can be overcome by modeling the composite stock price index by using the Artificial Neural Network (ANN) and nonparametric regression of Multivariate Adaptive Regression Spline (MARS). In this study, the time series data from the composite stock price index starting from April 2003 to March 2018 with its predictor variables are crude-oil prices, interest rates, inflation, exchange rates, gold prices, Dow Jones price, and Nikkei 225 Index. The both methods give better goodness of fit, where the coefficient of determination ANN is 0.98925 and the MARS determination coefficient is 0.99427. While based on the Mean Absolute Percentage Error (MAPE) of ANN was obtained 6.16383 and the MAPE value of MARS is 4.51372. This means that the ANN method and nonparametric MARS regression method have good performance to forecast the value of the Indonesian composite stock price index in the future, but in this case of data the nonparametric MARS regression method shows the accuracy of the model is slightly better than ANN.