Multilayer Perceptron and Long Short-Term Memory for Predicting Indonesian Composite Stock Price Index Using Macroeconomic Factors

Capital markets are complex and dynamic, so they have risks and uncertainties. The main activity in the capital market is investment. There are several instruments for investing, namely gold, land, savings, deposits, bonds and stocks. Investors need a variety of information to help determine the right momentum to invest, one of which is the stock price in the future. By making predictions, investors can minimize the risk of loss in investing. The purpose of this study is to compare the Multilayer Perceptron and Long Short-Term Memory algorithm in predicting Indonesian Composite Stock Price Index in the future based on macroeconomic factors. This study uses Dow Jones Industrial Average, Shanghai Stock Exchange, world gold prices, world oil prices, USD to IDR exchange rate, inflation, BI Rate, and money supply as macroeconomic factors to predict the price of the Indonesian Composite Stock Price Index. The time series used for the research data is monthly, from January 1991 to December 2018. Based on the results of the evaluations that have been conducted, this study found that modeling using the Multilayer Perceptron algorithm has a better performance than Long Short-Term Memory. Evaluation using Root Mean Squared Error and R-squared each produces a value 86.86% and 3.19% for Multilayer Perceptron; 74.24% and 3.94% for Long Short- Term Memory.