Replication data and code for "Inter-class co-movement of retail electricity tariffs and the residential adjustment gradient: evidence from a new Latin American panel"

Abstract

This repository contains the data and code required to reproduce every result, table and figure in the article "Inter-class co-movement of retail electricity tariffs and the residential adjustment gradient: evidence from a new Latin American panel" (Utilities Policy). It comprises: (i) a reconstructed quarterly panel of published sectoral retail electricity tariffs—residential at five consumption blocks, commercial and industrial—for fourteen Latin American countries (2013Q1–2025Q4), reconciled cell-by-cell against the original OSINERGMIN Tarifas Internacionales source workbooks; (ii) an annual policy–price panel for the same fourteen countries (2011–2024), together with a quarterly consumer price index for all fourteen from CEPALSTAT; (iii) a three-layer replication suite comprising a Stata implementation, an independent Python verification engine, and a Python layer that produces the second-round estimates and generates each figure directly from the estimation output, the φ column of the central table being reproduced independently by all three; (iv) the build-and-validation script that reconstructs the quarterly panel from source, documents each divergence and reconciles the result under an --audit flag; and (v) a README detailing software requirements, run order, the specification and which layer produces each result, and the documented data corrections, with a MANIFEST listing the SHA-256 digest of every file. Source tariff reports are publicly available from the Peruvian energy regulator (GRT-OSINERGMIN) and are not redistributed here; the specific workbook underlying each country–quarter is identified in the file column of the quarterly panel. This version supersedes the previous deposit and accompanies the second round of revision. The admissibility rule of the previous version, which admitted a country to a mean-group average on the basis of its estimated error-correction coefficient, has been withdrawn in response to referee comments; a single criterion remains, a property of the data. The long-run coefficient is now estimated directly from the country cointegrating regression rather than as the ratio −ψ/φ. The correction rates are unchanged in two of the three inter-class relations and move from −0.288 to −0.265 in the third; the long-run coefficients move by up to 0.04, and the inter-class gradient they document is now established by a paired test across countries. The first-round specification remains reproducible for audit ($SPEC = "r1" in the Stata master, --spec r1 in the verification engine), and 10_canonical_ecm.py --legacy reproduces the estimates of the originally submitted manuscript.