Volatility Estimation Algorithm Based On Particle Filter

Volatility in financial markets plays an important role in economic performance and financial stability.This article uses standard particle filtering,auxiliary particle filtering and improved regularized particle filtering to estimate volatility:1.For the standard particle filter in the volatility estimation of return on assets,because the transfer density is used as the important sample density,which leads to the sample particles lose a large number of low weight particles after updating and the problem that the estimation effect is not accurate.A new method based on auxiliary particle filter for the estimation of volatility is proposed in this thesis,that is APFVE algorithm.In the predicted stage of the APFVE algorithm,the two-factor asymmetric realized stochastic volatility model is used to estimate the current volatility.In the updated stage of the APFVE algorithm,the measurement information of the current time is fully utilized to adjust the importance sample density in real time and introduce the sample particle into the high likelihood region,which make the weight of the updated particle more stable.Simulation results show that the APFVE algorithm has better effect than the standard particle filter algorithm in terms of volatility variation feature estimation and volatility estimation accuracy.2.In the volatility estimation of return on assets by standard particle filter,with the iteration of the algorithm,because the resampling algorithm is carried out in the discrete distribution,which leads to a large proportion of the particle with large weight in the particle set while the particles with small weight are hardly used,resulting in the particle being relatively single and lack of diversity,so it is difficult for the algorithm to estimate the volatility of return on assets correctly when the return on assets change rapidly to a new level.In order to solve this problem,a algorithm based on improved regularized particle filter volatility estimation(RPFVE)is proposed in this dissertation.In the prediction stage,the improved RPFVE algorithm uses GARCH diffusion model to predict the volatility;in the update stage,it resamples from the continuous density of posterior distribution and selects the particles by Markov Chain Monte Carlo(MCMC),the weight distribution of the selected particles is more uniform,so that particles are no longer single and no longer lack diversity.According to the simulation results,we can see that compared with the standard particle filter volatility estimation algorithm and RPFVE algorithm,improved RPFVE algorithm has better results in both volatility variation featureestimation and volatility estimation accuracy.