Skip to content

Changelog

v5 — April 2026 (Submission-ready, JPubE format)

Structural decomposition of the SME set-aside replaces the v1--v3 reduced-form headline:

  • Asymmetric IPV reading of the Pregao auction; type-specific cost distributions point-identified from drop-out bids (Haile and Tamer 2003)
  • Krasnokutskaya (2011) auction-level UH correction via method-of-moments + best-linear-predictor shrinkage
  • Athey--Seira (2011) treatment of equilibrium SME entry as observed Poisson arrival rates
  • BNE Monte Carlo (B = 2,000) over scenarios S1, S2, S3, V0--V3
  • Cluster bootstrap at the auction level (Bbs = 500)
  • Saez--Stantcheva (2016) generalized social-marginal-welfare-weight identity

Headline magnitudes. Within-auction share of the simulated total effect: 74.5% (non-pharma), 73.3% (pharma), stable in the 73--85% range across cost-distribution estimators and across the strict-invariance benchmark. Welfare cost at λ = 0.30: 28.7% of pS1 (non-pharma), 47.0% (pharma). 10% price preference welfare-dominates the set-aside in non-pharma under both specifications; ranking bifurcates in pharma.

Robustness. Strict primitive invariance, Fc regime (losers-only / all-bidders / Turnbull NPMLE), kernel bandwidth, sample filter, temporal window (6m / 12m / 18m), alternative policy variants (V0--V3).

Online appendix. 15 sections including IPV-relaxation Gaussian copula sensitivity, Maskin--Riley FPSB calibration of V3, full λ-grid welfare CIs, Convite cross-modality consistency, Conley--Decarolis collusion screen, primitive-invariance KS tests.

Title. Renamed from The Price of Protecting Small Firms to The Cost of Inclusion: Decomposing Bidder Exclusion in Public Procurement, reflecting the structural focus on within-auction (sheltered bidding) vs. participation-margin decomposition.

March 2026 (v4)

  • Fiscal Cost Quantification subsection added to reduced-form manuscript (R$84.5--85.8 million)
  • 5 advanced econometric methods bolted onto DiDiR: HonestDiD, Lee bounds, causal forest, quantile DiD, Gelbach decomposition
  • 18 tables / 15 figures total

March 2025 (v2--v3)

  • Full R/fixest pipeline with 14 tables and 10 figures
  • Robustness: placebo tests, alternative clustering, winsorization, randomization inference
  • Extensions: real prices, extensive margin, efficiency, winner composition, heterogeneity by item value and PBU type
  • MkDocs documentation site

January 2021 (v1, first version)

  • Initial manuscript with core DiDiR results
  • 4 main outcome variables: prices, firms, bids, distance
  • 3 time windows with baseline and PBU-controlled specifications