Financial Forecasting by Modified Kalman Filters and Kernel Machines

This study combines a modified Kalman filter (MKF) and support vector machines (SVMs, a type of kernel machines) to implement a fast online predictor for option prices. The latent variables in Black-Scholes formula are estimated by the MKF. The residuals in MKF predictions are handled by an SVM. Using option data of Taiwan Futures Exchange, the proposed model is compared with traditional predictors. Empirical results confirmed that the new model is superior to traditional neural network models, which remarkably reduce the root-mean-squared forecasting errors.