Changes in market liquidity are of significant reference value to investors. This paper uses matrix processing of historical financial data to achieve a high-dimensional representation of time series. Combining the preprocessed data, a time convolution-Bayesian neural network (TCN-BNN) model is constructed to process financial time series data. Furthermore, through ensemble empirical mode decomposition (EEMD) and an improved multidimensional k-nearest neighbor (MKNN) algorithm, the financial market situation under changes in economic policy is predicted. The study shows that in the constructed financial time series model, with a time unit of 1 year, the autocorrelation range between financial markets and economic policies is -0.31 to 0.17, and the partial autocorrelation range is -0.23 to 0.18, indicating a high degree of correlation. The uncertainty of economic policies leads to significant fluctuations in financial markets during the 5th to 10th lag periods and time-varying periods.