Stock market volatility is a widely and deeply studied subject across academic and industrial settings. The increase in internet data usage, collection, and availability has also led to studies examining the links between internet search history and the stock market. In this paper we attempt to predict future stock market volatility, otherwise known as realized volatility, in Korean equity markets using Naver search trend data in a Long Short-Term Memory (LSTM) machine learning model. We find that LSTM networks showed an improvement in predicting realized volatility over the traditional GARCH model when compared using the mean average percentage error (MAPE) metric in both the KOSPI and KOSDAQ indices. We also find that we can improve on our original LSTM network with feature selection and expansion of the observation interval and normalization window.
주식식장의 변동성은 학계와 산업계 전반에 걸쳐 광범위하고 심도있게 연구된 주제다. 인너텟 데이터 사용및 수집의 증가와 인터넷 접근성의 개선은 이터넷 검색 기록과 주식시장의 연관성에 대한 연구의 증가로 이어졌다. 본연구에서는 Long Short-Term Memory (LSTM) 머신러닝 모델에서 네이버 검색 트렌드 데이터를 활용하여 한국 주식 시장에서 실현된 변동성 으로도 알려진 미래 주식시장의 변동성을 예측하려 한다.
