The Closed-Loop Paradox: Sequential Bayesian Recalibration and Competitive Decoupling in Two-Sided Payment Networks

Title: The Closed-Loop Paradox: Replication Package — Sequential Bayesian Recalibration and Competitive Decoupling in Two-Sided Payment Networks

Description:

This deposit is the full replication package for the manuscript "The Closed-Loop Paradox: Sequential Bayesian Recalibration and Competitive Decoupling in Two-Sided Payment Networks" by Restrepo Morales, Rodríguez Flores, Zea Restrepo, Giraldo Betancurt, Garcés Giraldo, and Camacho Vargas (2026), submitted to the International Review of Financial Analysis (Elsevier, Q1).

The study integrates two-sided market theory with sequential Bayesian inference and Vector Autoregression (VAR) econometrics to re-examine the long-standing claim that closed-loop payment networks are competitively inferior to open-loop networks. Using American Express, Visa, and Mastercard financial data over FY2021–FY2025, the analysis (i) recalibrates prior beliefs across four pieces of empirical evidence — revenue growth, net income growth, customer retention, and valuation multiples — yielding a base-case posterior of 64.9% for open-loop superiority, 30.1% for equivalence under segmentation, and only 4.9% for closed-loop superiority; (ii) documents — using Johansen cointegration and Granger causality on quarterly data (Q1 2021–Q1 2026) — that American Express is statistically decoupled from the Visa–Mastercard cointegrated dyad; and (iii) shows via impulse-response functions that a single macroeconomic shock reduces Amex net income growth roughly 2.9× more severely than Visa's. The deposit lets any reader reproduce every number, table, and figure in the paper.

Contents:

data/ — annual and quarterly financial data for AmEx, Visa, and Mastercard (FY2021–FY2025), the Bayesian likelihood worksheets (CSV and XLSX).

code/ — Python source: financial_analysis.py (CAGR, margins, ROE, P/E) and generate_figures.py (reproduces all eight publication-quality figures).

figures/ — eight figures × three editable formats: PDF (vector), SVG (vector, text preserved as text), and PNG @ 300 dpi.

Reproducibility: the code runs on plain Python 3 with numpy, pandas, and matplotlib — no proprietary software is required. See README.md for the exact commands.

Licensing: manuscript, figures, and data are released under CC-BY-4.0; the Python code is released under the MIT License.

JEL classification: G14, G21, L13, L86, C11, C32.

Suggested citation: Restrepo Morales, J. A., Rodríguez Flores, E. A., Zea Restrepo, F., Giraldo Betancurt, E. A., Garcés Giraldo, L. F., & Camacho Vargas, A. (2026). The Closed-Loop Paradox: Replication Package (Version 1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.XXXXXXX