DEVELOPMENT OF STOCK MARKET PREDICTION MODEL USING BAYESIAN NETWORK (CASE STUDY OF JAKARTA COMPOSITE INDEX)

Shanty Yuliana, Prof. Ir. Nur Aini Masruroh, S.T., M.Sc., Ph.D., IPU., ASEAN Eng

2012 | Skripsi | S1 TEKNIK INDUSTRI

A share of stock is the smallest unit of ownership in a company. People see stocks as a kind of interestng investment because stocks are highly liquid investment with no risk of losing the owner’s asset when the corresponding company is in crisis. But, stock price is highly volatile and hard to predict, while the money involved in it usually big. Therefore, research about predicting stock movement has always been fascinating. Bayesian network is proposed in this project to predict the movement of Jakarta Composite Index (JCI) because of its ability to update the model with current available information, which is very critical to accommodate the volatility of stock price. Other than the macroeconomic parameters of Indonesia (i.e. inflation rate, Bank of Indonesia rate, exchange rate, world oil price, and national subsidized fuel price), the Bayesian network constructed in this final project also includes getting information from breaking business and economic news articles. News is the easiest and most accessible way to see most recent significant changes around the world, thus will give the best picture of condition of the nation and the world at the moment. The first step of building the model is to find causal relationship among variables to build the Bayesian network structure. Next step is to process the news articles, in which text mining procedure is conducted to classify documents into two clusters, positive and negative, based on their effects on stock movement. The final step is to build the Bayesian network and determine conditional probabilities. The proposed model is capable of predicting correctly 80% of total data tested. There are two other conventional forecasting techniques performed to predict the same data: Single Exponential Smoothing which results in 20% accuracy and ARIMA which could not model the JCI movement. Thus the developed Bayesian network has a better model performance.

Kata Kunci : operational research, stock prediction, Bayesian network, computer science, data mining, text mining