Beschreibung:
This study employs machine learning models to forecast and comprehend the implied volatility of China ETF50. We develop a hybrid model named LSTM-ML, leveraging historical implied volatility, moneyness, and time-to-maturity as input features. The LSTM component captures dynamic hidden information from time series of volatility to generate temporal features. The empirical results reveal enhanced and stable predictive performance, particularly when the features capture long-memory patterns. Furthermore, the analysis of feature importance highlights implied volatility as the most influential factor affecting the forecast outcome. Lastly, we construct a trading strategy to showcase the profitability of the forecast results from our model