ARMA Model Based On Bayesian Analysis For The Prediction Of Net Value

In the process of purchasing and redeeming trust fund, we need to predict the variation trend of the net value so as to achieve maximum benefits. However, because the trust industry starts late and the sample size-the data of net value is still relatively less, it just can be analyzed as a short-term sequence. In this paper we use autoregressive moving average (ARMA) model appropriate for short series to analyze and forecast the net value of trust for the first time.Based on Bayes theory, firstly, we apply Bayesian method to estimate the coefficients of ARMA models, and then make numerical argument about the autocorrelation and partial correlation coefficient of Stationary time series to test trailing or censored, finally obtain the order of the model.Among them, we prove that the posterior distribution with prior information is more efficient than that without prior information, which requires to use existing information as much as possible in the choice of prior distribution, such as sample information, past experiences and so on. Then by using Gibbs sampling algorithm to sample the posterior distribution of parameters, random sampling error and its ultra-parameters, we obtain the parameter estimation of the model and its final form. Finally, the known sample values can be substituted into the ARMA model to forecast the next value of the net value.This paper also compare ARMA model based on Bayes theory with the one based on traditional least squares method, the results show that the net value of the former is closer to the real value with smaller relative error. Hence we conclude that the ARMA model based on Bayes theory is more effective.