Knowledge-based integrated financial forecasting system

This paper seeks to implement and test a financial forecasting agent which employs time series, derived time series data, and news that are retrieved and extracted from the Web. This research focuses on the time series data of some individual stocks from the Indonesian Stock Exchange as well as the index data. The financial forecasting agent implemented is based on a Multilayer Neural Network trained with the Genetic Algorithm. An incrementally trained Naïve Bayesian classifier to classify the news stories automatically is also implemented. The agent is measured for its performance on the specified historical stock price data. The first testing surveys its performance against the benchmark agents in a range of stocks. Although the agent does not consistently outperform the benchmarks, it does have its own advantages in terms of stability. The second test tries to find out whether news improves the agent´s prediction accuracy. We find that in certain stocks and the index, the inclusion of news does improve the prediction accuracy.