Gaussian Weighted Loss Functions for Neural Network Option Pricing
Designed a gaussian weighted objective function to price options using Neural Network, reducing pricing error by 66% to baseline in high-gamma at-the-money region while maintaining millisecond-inference (~1.1 million contracts/sec) by using an interpretable and compact 4.5k param feed-forward neural network.
Keywords: Derivative pricing, Neural Network, Custom Objective Function, Market Microstructure, Machine Learning in Finance, Option Pricing, Objective Function Design
View Full Paper on SSRN →