Technical Report

Executive technical summary

This report documents a reproducible branch-level analytics pipeline for a Bolivian microfinance portfolio observed from 2012-03-01 to 2014-07-01. The pipeline transforms operational Excel workbooks into a tidy analytical panel, validates forecast models, monitors mora risk, and reframes portfolio expansion as a financial-inclusion and territorial-access problem.

At the global level, active portfolio increased from 109.8 to 69,850.2 thousand BOB, while clients increased from 18 to 3,827. The maximum observed global mora rate was 0.25%, so the core analytical question is not only growth, but whether growth remained responsible and territorially balanced.

Data model

• Observed panel: branch-month records with portfolio, disbursements, clients and mora.
• Projection panel: workbook business assumptions kept separate from observed history.
• Branch dimension: 16 de Julio, Ceja, and a constructed Global portfolio view.
• Development layer: outreach, credit depth, territorial balance and risk-adjusted inclusion metrics.
• Privacy rule: officer-level names remain only in raw workbooks; public analytical outputs are branch-level.

Analytical modules

1. Branch maturity profile: launch, scale-up and consolidation stages.
2. Risk-growth positioning: monthly portfolio growth against mora.
3. Responsible inclusion score: client outreach, portfolio depth and risk penalty.
4. Forecast validation: Naive, ETS and ARIMA holdout backtesting.
5. Forecast-versus-plan gap: workbook projections compared with statistical forecasts.
6. Stress testing: responsible inclusion, tightening, high-growth and mora-shock scenarios.
7. Policy decision matrix: branch-level strategic priorities and development interpretation.

Forecasting design

The pipeline compares Naive, ETS and ARIMA models on a holdout window and ranks models by forecast error. For the Global portfolio, the best model by holdout error is ARIMA. The purpose is governance-grade model comparison rather than a black-box forecast.

Risk and stress-testing design

Stress scenarios are expressed as analytical governance cases, not predictions. They test how portfolio size, client outreach and mora risk would behave under responsible inclusion, credit tightening, high-growth risk and mora-shock assumptions.

Development economics interpretation

The technical contribution is to connect credit-risk analytics with poverty, inequality and development without overclaiming causality. Client outreach is treated as a formal financial-inclusion proxy. Territorial balance is treated as an inequality-of-access proxy. Mora control is treated as a responsible-finance safeguard.

Key statistical caution

The monthly sample is small and branch-level. Correlations and model diagnostics are useful for governance and hypothesis generation, but they are not causal estimates of poverty reduction or welfare impact.

Quality and privacy controls

• Raw operational files are preserved under data/raw/.
• Processed files are regenerated by scripts/01_run_analysis.R.
• Public outputs exclude officer-level personal names.
• Observed history and workbook projections are kept separate.
• Forecasts are validated before being used in the dashboard narrative.

Main outputs

• docs/index.html: dashboard ready for GitHub Pages.
• docs/research-note.html: public research-note page served by GitHub Pages.
• docs/technical-report.html: public technical-report page served by GitHub Pages.
• outputs/tables/risk_return_matrix.csv: branch risk-growth diagnostics.
• outputs/tables/forecast_vs_business_plan.csv: model-vs-plan gap analysis.
• outputs/tables/stress_test_scenarios.csv: 12-month stress testing.
• outputs/tables/policy_decision_matrix.csv: strategic interpretation table.
• reports/model_results.csv: compact model results for senior analyst review.
• reports/references.bib: citation file for the development-finance framing.