Multivariate Forecasting Using Hybrid VARIMA Neural Network in JCI Case

Hybrid methods is one of relatively new modern methods for multivariate forecasting. This paper focuses on the comparison between VARIMA (Vector Autoregressive Integrated Moving Average), neural network, and hybrid methods for multivariate forecasting of stock prices. RMSE (Root mean of squares errors) are used for evaluating the forecast accuracy. Four variables are used in this paper, i.e. open, high, low, and close of JCI (Jakarta Composite Index). The result show that hybrid methods with nine input neurons yield more accurate than VARIMA and neural network in training datasets. Whereas in testing datasets show that hybrid methods with eight input neurons yield more accurate than individual VARIMA and neural network. The three methods can't be use for long term forecasting.