This paper denoises the collected stock market return data during the audit period based on wavelet transform and constructs a GARCH-MIDAS model to capture the volatility characteristics of the return data. The autoregressive conditional heteroskedasticity (ARCH) test, Q-test and ADF test are used to demonstrate the reasonableness of the GARCH-MIDAS model construction in this paper. The parameter estimation of the GARCHMIDAS model is carried out using the great likelihood estimation method to further illustrate the validity of the model in this paper. The results show that the denoised yield data using wavelet transform is smoother than the original data, and can retain the main volatility characteristics in the original data, providing good data support for the subsequent capture of volatility characteristics. The macroeconomic and economic policy uncertainty variables are basically significant in the parameter estimation of one-factor and two-factor GARCH-MIDAS models, which can effectively reflect the overall and long-term volatility characteristics of yield data.
