Integrating Multi-Source Data to Model Consumer Confidence: A 3D-SDIM Application in Urban Contexts

Traditional consumer confidence indices (CCIs) rely primarily on low-frequency surveys, limiting their timeliness and resilience during periods of rapid change. This study addresses this shortcoming by proposing a structured multi-source framework (3D-SDIM). This framework views confidence as a latent construct composed of four dimensions—expectations, behavior, context, and sentiment—and estimates it based on quarterly data from a major metropolitan area (2000–2024). This study fuses heterogeneous indicators using a hybrid weighting scheme and estimates the quarterly CCI using a variational autoencoder supplemented by a dynamic feedback step; its performance is benchmarked against existing machine learning benchmarks. Key findings: (i) The estimated CCI closely tracks the official data series and captures inflection points around major events, such as the global financial crisis, the COVID-19 pandemic, and the 2022 turmoil. (ii) Out-of-sample errors are consistently lower than the baseline, with smaller overshoots at the onset of shocks and faster post-shock adjustments. (iii) Dimensional attribution remains interpretable, linking forward-looking expectations, actual behavior, macroeconomic fundamentals, and social sentiment. The framework advances measurement by connecting macro-level statistics and micro-level digital traces in a single latent model, providing actionable signals for policy—early warnings, targeted stimulus timing, and recovery monitoring. While demonstrated in a single city, the modular design supports portability through local indicator substitution and weight recalibration.