Warrants price forecasting using kernel machine & EKF-ANN: A comparative study

Due to the six unreasonable assumptions companioned with the Black-Scholes options pricing model (BSM), which often make the miss-pricing result because of the difference of market convention in practical. This study try to combine the BSM and extended Kalman filters-based artificial neural networks (EKF-ANN) to deal with the limitation of consideration of the influences from many unexpected real world phenomena. If we were to soundly take these phenomena into account, the pricing error could be reduced. In this paper, we try to make a comparative study with examined the forecasting accuracy between the BSM-based kernel machines (KM-BSM) and the BSM-based EKF-ANN (EKF-ANN-BSM). From the evidence of Taiwan Warrants market, we found that the performance indicates the KM is superior to the others, and the hybrid EKF-ANN-BSM framework is also better than the pure EKF-ANN. The results show that the KM-BSM and hybrid model could significantly reduce the normalized root-mean-square-errors (NRMSE) of forecasting, it helps to provide an alternative way to refine the options valuation.