Financial time series data exhibit characteristics such as complexity and high noise levels, rendering traditional clustering methods prone to limitations when processing such data. Consequently, this paper introduces rough set clustering, leveraging its advantages in financial time series analysis and forecasting to construct the macroeconomic prediction model required for research. Principal component analysis is employed to transform high-dimensional data into low-dimensional representations, thereby reducing the complexity of the prediction model. A comprehensive digital economy forecasting model is constructed, revealing its data processing workflow. This workflow emphasizes calculating the fitness of each solution and designing targeted fitness functions. Quantitative analysis across multiple datasets is conducted on the digital economy forecasting model, comparing traditional K-means clustering with the proposed model using economic indicators from multiple cities. Empirical applications demonstrate that the clustering results from the proposed macroeconomic forecasting model better reflect the developmental tiers of 36 cities, highlighting the significant value of rough clustering in financial time series analysis and forecasting.