One-Size-Fits-All? Cooperative vs. Non-Cooperative Financial Institutions Under Full Basel-Style Regulation in Brazil

Author: Arthur Gomes Nery Affiliation: Sistema OCB Date: 2025–2026

Overview

This repository contains the replication materials for "One-Size-Fits-All? Cooperative vs. Non-Cooperative Financial Institutions Under Full Basel-Style Regulation in Brazil." The study uses institution-level quarterly data from Brazil's Central Bank (BACEN) IF.data system to compare the financial behavior of credit cooperatives and conventional financial institutions operating under the same full Basel-style prudential methodology.

Key Findings

• Cooperatives hold 7–8 percentage points less regulatory capital than non-cooperatives of comparable size and geography (CEM specification: −7.16pp, 95% CI [−7.89, −6.42], p < 0.001)
• Cooperatives are 11–15 percentage points more leveraged, reflecting structural dependence on deposit funding rather than equity markets
• Cooperatives operate with lower net interest margins (−3.48pp) but higher operational efficiency (−1.51 cost-to-income) and marginally higher ROA (+0.28pp)
• Cooperatives exhibit 4 percentage points lower NPL ratios — lower credit risk despite lower capital buffers
• Credit portfolio allocation and deposit reliance results do not survive common-support restriction, suggesting they reflect size composition effects rather than genuine behavioral differences
• Six of eight outcomes are stable across all four specifications, including unconditional differences, size-controlled OLS, coarsened exact matching, and common-support restriction

Repository Structure

paper1_v4.ipynb              # Main analysis notebook (current, consolidated)
paper1_comprehensive.ipynb   # Comprehensive version integrating all IF.data sources
expanded_abstract.md         # Extended abstract with full results narrative
requirements.txt             # Python dependencies
figures/                     # All output figures (PNG)
tables/                      # All output tables (CSV)
  table1_balance.csv         # Balance test: matched vs. unmatched
  table2_four_specs.csv      # All outcomes × four specifications
  table3_main.csv            # Main CEM results
  tableA1_descriptive.csv    # Descriptive statistics
data/
  raw/
    if.data/                 # BACEN IF.data bulk CSVs (semicolon-delimited)
      prudential_conglomerates/
        summary/             # Balance sheet overview (primary source)
        segmentation/        # S1–S5 tier + simplified methodology flag
        assets/              # Asset decomposition
        liabilities/         # Funding structure
        income_statement/    # P&L
        capital_information/ # RWA decomposition, CET1/Tier1/Basel ratios
      financial_conglomerates/
        portfolio_risk_level/          # AA–H loan quality buckets
        portfolio_legal_person_*       # SME vs. large lending
        portfolio_geographic_region/   # Regional breakdown
        portfolio_number_clients_operations/
    bcb/
      agencias/              # Monthly branch office locations (2007–2024)
      postos/                # Monthly ATM/service points (2007–2024)
  processed/
    if.data/                 # Processed panels and results

Data Sources

All data are publicly available from BACEN's IF.data bulk download system:

• Prudential conglomerates panel: 61,380 institution-quarter observations, 1,959 institutions (2014–2024)
• Analysis sample: 15,004 observations, 650 institutions under full Basel methodology (168 cooperatives, 483 non-cooperatives)
• Monetary values: All figures in R$ thousands (R$ mil)
• Frequency: Quarterly (March, June, September, December)

Data portal: https://www.bcb.gov.br/estatisticas/ifdata

Methodology

Identification Strategy

Comparison is restricted to institutions already operating under full Basel-style prudential methodology (i.e., Instituição Utiliza Metodologia Simplificada == Não). This holds constant the regulatory rules — capital requirements, risk-weighting methodology, reporting standards — and isolates the association between institutional form and financial outcomes.

Four Specifications

Spec	Description	N institutions
1	Unconditional mean difference	650
2	OLS: log(assets) + calendar-quarter FE	650
3	Coarsened Exact Matching on size quintile × macro-region + OLS	599
4	Common support restriction (log assets ∈ [11.47, 15.96]) + OLS	varies

Heteroskedasticity-robust (HC3) standard errors throughout.

Outcomes (8)

Dimension	Outcome
Capital structure	Basel capital adequacy ratio (%), leverage ratio
Profitability	Return on assets, net interest margin, cost-to-income ratio
Credit quality	Credit portfolio / assets, NPL ratio (provisions / gross credit)
Funding	Deposit ratio (captações / assets)

Main Results (CEM Specification)

Outcome	Coop differential	95% CI	p-value
Basel Capital Ratio (%)	−7.16pp	[−7.89, −6.42]	<0.001
Leverage (Passivo / Assets)	+14.89pp	[+14.37, +15.42]	<0.001
Return on Assets	+0.28pp	[+0.21, +0.35]	<0.001
Net Interest Margin	−3.48pp	[−3.67, −3.29]	<0.001
Cost-to-Income Ratio	−1.51	[−1.63, −1.39]	<0.001
Credit Portfolio / Assets	+4.90pp	[+3.76, +6.04]	<0.001
NPL Ratio	−4.08pp	[−4.37, −3.79]	<0.001
Deposit Ratio	+6.48pp	[+5.33, +7.63]	<0.001

Credit portfolio and deposit ratio results do not survive common-support restriction — see Section 3.2.

Robustness

All six stable results are robust to:

• Unconditional comparison, size-controlled OLS, coarsened exact matching, and common-support restriction (Specifications 1–4)
• Sign and significance maintained across all four specifications (p < 0.001 in each)
• Coefficient stability plots across the four specifications confirm monotonic attenuation or strengthening patterns

Setup

# Python 3.12+
python -m venv .venv
source .venv/Scripts/activate   # Windows/bash

pip install pandas numpy matplotlib seaborn scipy statsmodels linearmodels

Running the Analysis

Open paper1_v4.ipynb in Jupyter and run all cells sequentially. The notebook expects data at data/raw/if.data/ as downloaded from BACEN IF.data.

Raw CSVs use semicolon delimiters (;), Portuguese column names, and Brazilian number format (. = thousands separator, , = decimal). The parse_br() helper handles conversion automatically.

Citation

@unpublished{nery2026onesizefitsall,
  title={One-Size-Fits-All? Cooperative vs. Non-Cooperative Financial Institutions
         Under Full Basel-Style Regulation in Brazil},
  author={Nery, Arthur Gomes},
  institution={Sistema OCB},
  year={2026}
}

---

Keywords: credit cooperatives, prudential regulation, Basel, coarsened exact matching, institutional form, Brazil, BACEN JEL codes: G21, G28, P13