Neural Networks Boost Multi-Asset Options Pricing Efficiency.

A new application of Neural Networks with Local Converging Inputs (NNLCI) has been developed to enhance the efficiency of numerical methods used for pricing complex multi-asset options. This approach leverages a neural network to refine solutions derived from a coarse computational mesh by locally correcting them with data from a more refined mesh. A key advantage is its ability to achieve substantial improvements with only a small amount of high-fidelity training data. The NNLCI method has been demonstrated across various financial models, including cash-or-nothing options under the Black-Scholes equation in multiple dimensions and down-and-out barrier call options under the Heston stochastic-volatility model. In these tests, NNLCI consistently reduced the root-mean-square error of the refined-mesh solutions by factors ranging from 4 to 12. This innovation promises to significantly cut down computational demands for high-dimensional problems, making real-time options trading and risk management more feasible due to its low training costs and strong generalization capabilities.