In order to improve the performance of financial time series data analysis, evaluation and prediction, a sequential Bayesian learning method is designed to estimate the asymmetric generalized autoregressive moving average (GARCH) model based on the study of data mining, maximum likelihood estimation and sequential parameter learning. The leverage effect is considered to describe the negative correlation between return and volatility, thus solving the complex numerical problems in the estimation of stock simulation models. Through the simulation analysis, the results show that the model can better simulate the stock volatility and price trends, and is effective.