Option Pricing Model Combining Ensemble Learning Methods and Network Learning Structure

Option pricing based on data-driven methods is a challenging task that has attracted much attention recently. There are mainly two types of have been widely used, respectively, the neural network method and ensemble learning method. The option model high complexity, large number hyper-parameters will be generated during training, resulting in difficult adjustment. Furthermore, lot training data needed. not ideal for feature extraction, because each calculation to reduce final residual. Therefore, this paper adopts framework embeds modular into structure, an deep proposed. composed parts: features reorganization random forest, used calculate importance features, combined with original as input; multilayer structure modules, it also designs stop algorithm automatically determine layers. This enables retain effect extraction adapt small medium sets without generating many hyper-parameters. Moreover, order make fully absorb advantages methods, we adopt cross-training data. From experimental results, can concluded compared current optimal method, prediction performance proposed improved by 36% root mean square error (RMSE), which proves superiority from quantitative direction.