Evaluating Deep Learning Models On Monte Carlo-Simulated Basket Option Pricing Under Uniform Setting
In the fast-paced domain of computational finance, the valuation of derivatives demands models that are both precise and computationally efficient. This report presents an in-depth comparative analysis of several deep learning architectures applied to the pricing of basket options under uniform setting. This report uses Monte Carlo simulations for basket options dataset creation. We systematically investigate the accuracy, execution speed, and cost-efficiency of Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Transformer models, Long Short-Term Memory networks (LSTM), and combined CNN-LSTM models, to determine their efficacy in predicting derivatives pricing. Our findings demonstrate a nuanced landscape where the CNN-LSTM architecture excels in accuracy due to its dual ability to process spatial and temporal information. CNNs follow closely, offering a harmonious balance of accuracy and simplicity, while LSTMs provide a reliable option with consistent, albeit varied, performance. In contrast, DNNs and Transformer models, while powerful, face challenges in slower convergence and higher susceptibility to overfitting as complexity increases. The report underscores the nuanced trade-offs between model complexity, resource allocation, and performance, guiding the selection of optimal computational strategies in the fast-paced and intricate world of financial trading.