摘要
基于Zheng等人[1]的研究框架,即将先验金融知识纳入神经网络的设计和训练,提出了一种预测隐含波动率曲面的集成GRU神经网络模型。该模型使用了一种包含波动率微笑的激活函数,并将无套利、左右边界和渐进斜率等金融条件纳入神经网络的训练过程中。利用上证50ETF期权2015年2月9日至2023年3月31日期间的交易数据进行了实证分析。实证结果显示:与SSVI模型和基准神经网络模型相比,集成GRU模型在训练集上的平均绝对百分比误差为8.56,在测试集上的平均绝对百分比误差为11.17,是所有模型中预测精度最高的,同时满足了嵌入的金融条件。
Based on the research framework of Zheng et al,which incorporates prior financial knowledge the design and training of neural networks,an integrated GRU neural network model for predicting implicit tility surfaces is proposed.The model uses an activation function that includes a volatility smile,and incorpo-rates financial conditions such as arbitrage,left and right boundaries,and asymptotic slopes into the training process of the neural network.Empirical analysis was conducted using trading data from the Shanghai 50 ETF options fromFebruary 9,2015 to March 31,2023.The empirical results show that compared to the Ssvl model and the benchmark neural network model,the integrated GRU model has the highest prediction accuracy among all models with an average absolute percentage error of 8.56 on the training set and 11.17 on the test set,while satisfying embedded financial conditions.
作者
白祥
张金良
靳慧娜
BAI Xiang;ZHANG Jinliang;JIN Huina(School of Mathematics and Statistics,Henan University of Science and Technology)
出处
《上海节能》
2024年第2期296-304,共9页
Shanghai Energy Saving
基金
国家自然科学基金(51675161)。