Models trained under assumptions in the complete market usually don't take effect in the incomplete market. This paper solves the hedging problem in incomplete market with three sources of incompleteness: risk factor, illiquidity, and discrete transaction dates. A new jump-diffusion model is proposed to describe stochastic asset prices. Three neutral networks, including RNN, LSTM, Mogrifier-LSTM are used to attain hedging strategies with MSE Loss and Huber Loss implemented and compared.As a result, Mogrifier-LSTM is the fastest model with the best results under MSE and Huber Loss.

翻译：基于完备市场假设训练的模型通常在不完备市场中失效。本文解决了具有三种不完备性来源（风险因子、非流动性及离散交易日期）的不完备市场中的对冲问题。提出了一种新的跳跃扩散模型用于描述随机资产价格。采用三种循环神经网络（RNN、LSTM、Mogrifier-LSTM）获取对冲策略，并实现了均方误差损失与Huber损失的对比。结果表明，Mogrifier-LSTM在均方误差损失与Huber损失下均表现出最快的收敛速度与最优效果。