Prediction of pricing and hedging errors for equity linked warrants with Gaussian process models

Gaussian process (GP) model is a Bayesian kernel-based learning machine. In this paper, we propose a GP model with a various mixed kernel for pricing and hedging ELWs (equity linked warrants) traded at KRX with predictive distribution. We experiment with daily market data relevant to KOSPI200 call ELWs from March 2006 to July 2006, comparing the performance of the GP model with those of various neural network (NN) models to show its effectiveness. The applied NN models contain early stopping, regularized NN, and bagging. The proposed GP model shows that its forecast capability outperforms those of the three NN models in terms of both pricing and hedging errors, thereby generating consistent results.