Multi-scale Digital Twin Modeling for Macroeconomic Forecasting: Integrating Econometric Models with Deep Generative Networks
Year: 2026
Type: conference-paper
Abstract: This paper proposes a multi-scale digital twin (DT) modeling framework that integrates econometric models with deep generative networks (DGN) to achieve dynamic simulation and forward-looking prediction of macroeconomic indicators. The method includes: (1) constructing a multi-scale data fusion module to unify national, industry and micro-enterprise level data; (2) establishing structural constraints through a generalized vector autoregression (GVAR) model to achieve short-term causal characterization between economic variables; (3) using a variational autoencoding generative adversarial network (VAE-GAN) to generate multi-scale dynamic features; and (4) synchronously calibrating parameters in the digital twin engine and updating the generative network weights through Bayesian optimization to achieve dynamic consistency between prediction and real data. The experiments were validated using Chinese macroeconomic data from 2015 to 2023. The average relative error of the GDP quarterly growth rate forecast was concentrated between 1.2% and 2.7%, and the accuracy of capturing CPI fluctuations converged to 1.32% in 2023. The model showed an average response time of approximately 1.1 quarters in policy shock simulations. The results indicate that this method has high accuracy and interpretability in multi-timescale economic forecasting.
Authors: Ping Zhang, Jindao Zheng
Institution: Hubei Industrial Polytechnic
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7208744933