應用LSTM探討選擇權流動性對市場波動方向之預測

Applying Long Short-Term Memory Networks to Predict Market Trend Based on Options Market Liquidity

本研究旨探討市場臺指選擇權各履約價之買賣方掛單行為，聚焦於其掛單現象進一步分析其市場流動性與其影響期現貨價格波動之現象，並建構以選擇權委買賣價差衡量之流動性指標，並利用長短期記憶網路（Long Short-Term Memory Networks, LSTM）模型結合卷積神經網路（Convolutional Neural Network, CNN），建立一混合深度學習模型，由市場買賣方之掛單變化預測對未來期現貨價格波動。實證結果顯示，本研究所提之LSTM模型可用於預測短線波動，而LSTM-CNN模型整體預測效果更優於原LSTM模型。整體而言，所提出之模型可提升對市場動態的辨識能力與波動預測準確性，除了有效辨識期貨日內波動方向，並有助於即時掌握市場變化，對交易策略設計與風險管理展現應用潛力。

This study examines the order submission behaviors of buyers and sellers across different strike prices in the Taiwan Index Options (TXO) market, with a particular focus on order placement phenomena. It further analyzes how these behaviors impact market liquidity and influence price volatility in both futures and spot markets. A liquidity indicator based on option bid-ask spreads is constructed for the purpose of this research. Additionally, a hybrid deep learning model is developed, combining Long Short-Term Memory Networks (LSTM) and Convolutional Neural Networks (CNN), to predict future futures and the spot price volatility based on the changes of buyer and seller orders. Empirical results indicate that the proposed LSTM model is effective in forecasting short-term volatility, whereas the CNN-LSTM hybrid model achieves the superior predictive performance compared to the standalone LSTM model. Overall, the proposed models enhance the identification of market dynamics and improve the volatility prediction. Besides identifying intraday directional movements of futures prices effectively, these two models also demonstrate significant potential ability for the real-time market monitoring, the strategic trading decision-making, and the risk management applications.