This project is ongoing. Descriptions, methods and results may be updated after further validation.
Research question
Do global forecasting models, coherently reconciled across the municipality–state–region–Brazil hierarchy, forecast the geography of Pix better than local models — and in which territories is the forecast least reliable?
Motivation
Pix changed the economics of payments in Brazil within a few years, but adoption was not uniform across the territory. Forecasting its volume by municipality matters for banks, retailers and financial-inclusion policy, and coherence across geographic levels prevents local forecasts that do not add up to the national total.
Data
- Monthly Pix statistics by municipality (Banco Central do Brasil open data), from Nov 2020 onwards
- IBGE municipal mesh and the municipality–state–region–Brazil hierarchy
Methodology
- Temporally stable geography and municipality-month panel with snapshot hashes and metadata
- Global forecasting models evaluated by rolling origin against baselines
- Coherent hierarchical reconciliation and scaled error metrics (MASE/RMSSE)
- Structural-break tests and heterogeneity analysis of forecast reliability
Limitations
- Early stage — extraction, descriptives and structural breaks run; global models not yet estimated.
- The availability and contract of the Central Bank API are external to the project; the snapshot used in the paper will be frozen with a hash.
Implications
- Territorially coherent forecasts for financial-services planning and inclusion policy.
Planned applied products (possible outreach, subject to registration)
- Municipal Pix-diffusion dashboard and forecast-reliability maps