Abstract: The scientific and reasonable exchange traded fund (ETF) options price contributes to implementing risk hedging function. This complex modeling process needs to consider the economic significance and accurately grasp the market rules. The issue of pricing ETF options is studied and a hybrid ETF options model is proposed. It combines the Nested-LSTM neural network model and the Heston model for the modeling, and dynamically corrects the option pricing deviation. The high-frequency data of ChinaAMC China 50 ETF, Harvest SZSE SME-CHINEXT 300 ETF and Huatai-PB CSI 300 ETF are taken as examples to verify the effectiveness of the proposed model. The experiment results show that the volatility characteristics of different types of ETF options prices are significantly different. Therefore, neither the Black-Scholes model nor the Heston model can be adapted to handle complex variation rules of ETF option prices accurately. By introducing Nested-LSTM neural network model into the Heston model, the proposed model can effectively capture the dynamic change rules of different types of ETF options, thus improving the estimation accuracy of ETF option prices effectively.

Key words: option pricing, ETF options, deep neural network, financial risk