Study on the VaR model based on the simulation of Support Vector Machine

Three computational methods are applied to traditional VaR model at present, including delta positive, Monte Carlo simulation and history simulation, however, some defects exist in the traditional methods such as fat tail, nonlinearity, big estimated error, complexity of the calculations, etc. In this paper, SVM theory is applied to VaR model by choosing Gaussian normal distribution function as kernel function. The new VaR model overcomes the defects, and is effective in approximating and generalizing compared with traditional ones; therefore, it is a significant complement to VaR system.