Daily Forecasting Trend Jakarta Composite Index (JCI) Using Multivariate Long Short Term Memory

Abstract

The need to be able to predict stock price movements is one of the problems that is difficult for investors to solve.Forecasting price movements serves as a signal to buy and sell. Forecasting stock price movements can provide us with a reference in making investment decisions. This study aims to predict the daily trend JCI. The datasets used in this research are JCI, NASDAQ, and NYSE with a range of 21 years (01/01/2000 to 31/12/2021). The features used as input are the opening prices of the JCI, NASDAQ, and NYSE. The amount of data used is 5211 lines. The deep learning method used is multivariate LSTM. optimization of the model used using Adam. There are 4 LSTM models made using loss metrics MSE and MAE, using 2 epochs of 100, and 300 epochs. The results showed that the 4 LSTM models could predict the daily trend (d+1) of the JCI. The most optimum model is the model made using MSE with 300 epochs.