Claim Missing Document

Found 1 Documents

Price Difference Embedded Multivariate Long Sort-Term Memory for Stock Movement Prediction Wawan Yunanto
International ABEC 2021: Proceeding International Applied Business and Engineering Conference 2021
Publisher : International ABEC

Show Abstract | Download Original | Original Source | Check in Google Scholar | Full PDF (335.556 KB)


The course of the capital market is complicated, unpredictable, and volatile for investors to formulate. Fundamental and technical analysis are the main common approaches to predict stock prices used by the economic experts nowadays. The fundamental aspect is determined by the internal factor of the companies, but the technical one is clearly represented on a daily basis in the stock market. Stock price is not the only important item for investors to make investment policy. The fluctuation in the trading floor has become the most important issue to be considered. In this research, we propose a prediction framework, namely Price Difference Embedded Multivariate Long Sort-Term Memory (PDEM-LSTM), to combine stock price and movement prediction into a single pipeline. We employ recurrent deep learning modeling technics into stock market forecasting since there are sequential properties in the technical components. Our work solely based on these sequences or timeseries features to simplify the experimental setting and more focus on the improvement compared to most previous studies. Our benchmark compares results from univariate scheme on the same sequences with 3 difference features which are current day and next day price along with the price difference between those days. We use 5 stock issuers from 5 different stock indices and the market data taken from January 1, 2000, to December 31, 2020. The results showed that price difference feature embedded into LSTM in multivariate setting greatly improve stock movement prediction without degrading stock price forecasting too much. It is simple and robust; it can be attached on most stock prediction techniques in the feature engineering phase.