Authors: Riki Herliansyah and Jamilatuzzahro
Abstract: The Indonesia Stock Exchange (IDX) Composite is an index of all stocks that are traded on the IDX. It is known as a tool used by investors and financial managers to depict market and to compare the return on specific investments. Recently, IDX Composite rate is very fluctuated resulting in difficulty to predict the trend of series data. A lot of time series-based model had been developed to make future predictions of IDX Composite. Neural Networks, one of the most popular approaches in financial worlds due to its parsimonious data requirements and can produce a more accurate model especially nonlinear, was used in this study. The study aims to investigate the suitable model and forecast future of IDX Composite using Feed Forward Neural Networks (FFNN). The IDX Composite series from January 2011 to July 2016 was used to construct the model of FFNN. The study found that FFNN architecture with three inputs and four hidden neurons showed the best forecast accuracy measured by RMSE and MAE. The selected model was then used to predict 12 months ahead of IDX Composite along with their confidence interval. The capital gain of IDX Composite forecasting was calculated to compute the profit with an assumption the investment was 1 million (IDR). The results indicated that if the investors sold the shares in next 12 months, there would be significant profit or capital gain obtained on August and September 2016. While selling the share on remaining periods would yield a noticeable loss or capital loss.
Keywords: Capital gain; confidence interval; IDX Composite; neural networks; predictions;.