KOSPI200 Prediction through Low-Pass Filtered Long Short-Term Memory Algorithm

Abstract

Diversification of the modern financial market has led to an increase in the importance of the stock market index. Trend of the stock market at large can be identified through the analysis of the stock market index. Movements of stock indices serve as a key measure for investors while trading individual stocks and play an important role in establishing basic asset allocation strategy. Currently, in the field of computer science, research on data prediction through machine and deep learning is being actively conducted. Using these algorithms, research on financial time-series data is being conducted in the financial domain. Similar to other time series data, stock market indices have typical characteristics such as regularity, wavelength, and noise. In this study, we focus on the time-series characteristics of stock indices by adopting the low-pass filter as a method of denoising data, rather than simply analyzing the index using basic deep learning. Through this research, we aim to increase the predictivity of stock price index using the ensemble model of the low-pass filter and Long Short- Term Memory algorithm (LSTM) and empirically analyze the result through KOSPI200 stock index data. The result of the studies shows that proposed model had surpassed other denoising LSTM models and simple LSTM model in every test period. In conclusion, further studies in denoising data can be resulted in improvement of prediction in financial area.