Skip to content

Changelog

v10 — JPubE short paper publication package (2026-05-25)

The public site now points to the v10 publication package in v10-causal-mechanism/. Main PDF, Online Appendix, analysis-page script paths, results pages, and site figures were refreshed from the v10 outputs.

  • Current manuscript. v10-causal-mechanism/manuscript/paper/main.pdf.
  • Current Online Appendix. v10-causal-mechanism/manuscript/paper/OnlineAppendix.pdf.
  • Site figures. Rebuilt from v10 macros/outputs: urgent-vs-ordinary, Lee bounds, within firm-buyer-item forest, pricing-versus-sourcing decomposition, winner churn, and event-study diagnostics.
  • Core result. The paper separates same-firm pricing from supplier-set sourcing under judicial urgency. In deep repeated urgent pharmaceutical markets, the procurement-cost margin is not mainly a broad same-firm markup; it operates through fragmented sourcing, lost scale, and supplier-set reallocation, with a residual within-firm gap able to reappear in thinner or earlier markets.

Refinement pass — 2026-05-24 (disambiguation, referee-proofing, figures)

A skeptical-referee pass that disambiguated the heterogeneity, ported the highest-value robustness diagnostics into the Online Appendix, and enriched the documentation site with figures. Headline estimates and the identified objects are unchanged; the main-paper exhibit set stays at four (≤5).

Heterogeneity reframed (disambiguation). Two referee tests separate the two axes of the within firm-buyer-item heterogeneity:

  • The quantity axis is the scale channel, not same-firm pricing: holding firm, buyer, and item fixed, the within-triple log-quantity coefficient is −0.259 (SE 0.074) and the order-size gradient collapses under a quantity control (61_h4_quantity_quartiles.R). This coefficient is now cited in Results §5.3.
  • The earlier-period gap survives a within-triple quantity control (+0.117 → +0.168, p = 0.007) but is administrative-dearer and time-declining (62_h4_period_axis.R).
  • The phrase "supplier leverage reappears in thinner and earlier markets" is replaced throughout paper and site by a bounded statement — the deep-market null is not universal, a residual within-firm gap persists in the earlier period — without asserting litigated-buyer leverage. The H4 hypothesis and finding pages are retitled "The deep-market null is not universal."

Online Appendix — four referee-defense diagnostics added (kept to five pages):

  • Bounding the null (H3, Appendix C): a TOST equivalence test on the within-triple coefficient rules out broad same-firm markups above ~10.8% (TOST p = 0.070 at a 10% margin, 0.364 at 5%; MDE 12.2%; power 0.64 against a 10% markup).
  • Multiple testing (H4, Appendix C): Holm and Romano-Wolf step-down adjusted p-values for the below-median-quantity (0.041 / 0.053) and earlier-period (0.010 / 0.025) heterogeneity splits.
  • Lee-bound monotonicity robustness (H2, Appendix B): the bounded litigated-over-administrative gap stays positive under extra trimming for monotonicity slack until the additional trim exceeds 0.60.
  • Supplier-turnover benchmark (H6, Appendix C): a within-regime placebo Jaccard of 0.378 vs the cross-regime 0.268 (gap 0.109) — reallocation exceeds normal churn.

Documentation site — figures added. Four new color-blind-safe figures (63_site_figures.R, numbers read from values.tex, no hardcoding): the within firm-buyer-item coefficient forest (AN-003/AN-004), the urgent-vs-ordinary coefplot (AN-001), the Lee-bound interval (AN-002), and the winner-set churn benchmark (AN-006). Each hypothesis page (H1–H7) now embeds one hero figure after its evidence-strength callout.

Build fix. Referee-test macros carrying hypothesis IDs (h2/h3/h4/h6) are illegal LaTeX control-sequence names. Site-only diagnostics now emit to analysis/referee_macros.tex, while paper-grade numbers emit to values.tex with valid names. The paper had not recompiled since those scripts first ran (deploys only copy the existing PDF), so the break was latent; both documents now build cleanly (main 17 pp, Online Appendix 5 pp).

Final cleanup. The abstract's heterogeneity sentence was simplified to "the within-firm gap reappears only in the earlier period, while order-size differences operate through scale." The appendix now discloses the Romano-Wolf bootstrap replications (999) and the supplier-turnover benchmark's seeded 200-split averaging. A reproducible word counter (analysis/64_wordcount.py) puts the main text at 4,402 words (4,445 with captions), well under the JPubE 6,000-word cap, with four exhibits and a five-page appendix.

Reproducibility run. Re-ran the entire numbered analysis pipeline (analysis/run_pipeline.sh, 23 scripts) from the prepared cache: all exit 0, and values.tex plus the 31 output tables regenerate byte-identically (figure PDFs differ only by an embedded timestamp; rendered content is identical). The v10 results are fully reproducible from the cache forward.

v9 — JPubE short paper (2026-05-24)

Current submission version: Sourcing under Sanctions: Judicial Urgency and Pharmaceutical Procurement Costs.

  • Contribution. Separates same-firm pricing from supplier-set sourcing under court-mandated urgent procurement. In deep repeated urgent markets the sanction-related cost margin does not appear as a broad same-firm markup; it appears through fragmented sourcing — lost scale and supplier-set reallocation. The deep-market null is not universal: a residual within-firm gap persists in the earlier period (the quantity dimension reflects scale, not same-firm pricing).
  • Central estimates. Urgent-vs-ordinary margin (+5.4% negotiated price, ~5.4% fewer bidders, +2.1pp tender success); under-the-gun gap selection-bounded in [15.9%, 21.1%] (Lee bounds; naive 29.5%); within firm-buyer-item coefficient 0.035 (SE 0.041); modal winner differs in 70.2% of item-buyer pairs; administrative orders 3.3× larger.
  • Identification. Lee trimming bounds on administrative-channel selection; within firm-buyer-item pricing test; direct winner-switching evidence; Rademacher wild-cluster bootstrap (p=0.0080, preferred); never-litigated placebo (null, −0.020); BJS event study + Honest-DiD as diagnostics.
  • Fiscal procurement-cost implication. ~**\(27.8M/yr** (\)23.9–31.7M); a fiscal procurement-cost calculation, not a full welfare estimate.
  • Short-paper compliance. Main paper ~5,000 words and four exhibits; Online Appendix five pages.
  • Site. Restructured to mirror the project's documentation conventions: Hypotheses (with scorecard), Analyses (AN pages), Findings (claim-altitude), and Results / Robustness / Extensions / Advanced Methods.

Earlier versions

  • v8 (sourcing-reframe). Earlier reframing around pricing versus sourcing, with the under-the-gun gap and the within firm-buyer-item pricing test; the v10 short paper supersedes it with selection-bounded language, combined exhibits, and a compact appendix.
  • v7 and earlier. Reduced-form and three-channel framings; superseded.