latin-america-financial-development-lab

Methodology

Research Design

The project studies how credit composition and productive-sector financing evolve across Latin America. It combines descriptive panel analytics, sectoral concentration metrics, country rankings, Bolivia-centered comparison, clustering, and panel econometric specifications when the data support estimation.

Unit of Analysis

The intended unit is country-period, with period inferred from the data. The legacy scripts use date variables and monthly updates. Annual indicators are constructed from monthly data using annual means unless a source-specific stock definition is documented.

Core Data Products

Indicator Construction

The pipeline attempts to construct:

When the required source variables are unavailable, the script records the indicator as unavailable.

Descriptive Evidence

The descriptive workflow produces panel coverage, variable dictionary, missing value audit, descriptive statistics, country rankings, and editorial figures.

Econometric Strategy

The preferred dependent variable is:

productive_credit_share_it

Candidate explanatory variables:

Minimum planned specifications:

If an adequate dependent variable or panel structure is missing, the model script creates status outputs rather than numerical claims.

Robustness Strategy

The robustness script is prepared for outlier exclusion, winsorization at the 1st and 99th percentiles, crisis-year exclusion, pre/post COVID comparisons, Bolivia versus rest-of-region comparisons, fixed-effect alternatives, lagged variables, and alternative standard errors.

Clustering Strategy

Country typologies are planned using productive credit share, sectoral concentration, diversification, credit growth, and volatility. Methods include k-means clustering, hierarchical clustering, and PCA when sufficient complete observations exist.

The central comparative question is: Which countries share Bolivia’s financial development profile?

Limitations

The current repository lacks versioned raw or cleaned data. As a result, no substantive empirical finding should be read from placeholder outputs. All results must be regenerated after the source data are restored.