Analyses — The Price of Exclusion¶

This page is the directory of analyses for The Price of Exclusion: SME Set-Asides in Public Procurement. Each AN-NNN page is anchored by a YAML frontmatter block that drives the generated index below.

This page is the directory of designs and results. For curated claims about the world — sentences that could go in the paper — see docs/findings/. For the testable predictions and their cross-references to AN pages, see docs/hypotheses/.

How to read an AN page¶

Each an-NNN-<slug>.md follows the same skeleton:

Frontmatter: identifier, status, hypothesis it bears on, script and output path, design block (sample, specification, FEs, etc.).
## Question — the precise research question.
## Design — sample, variation, specification, identification threats.
## Results — headline number + headline table or figure, with a link to the underlying script.
## Interpretation — what the result means; confidence color with justification.
## Follow-ups — forward-looking only; analyses not yet run.

Status. pending (scaffolded, not yet run or not yet interpreted) → done (script run, headline + interpretation written, confidence set) → stale (superseded by a newer AN or by upstream-data change).

Confidence. 🟢 green (clean identification, large sample, robust), 🟡 yellow (informative with caveats), 🔴 red (unreliable, kept for the record only). Set only when status: done.

Generated index¶

The table below mirrors the frontmatter of every an-*.md in this folder. The full machine-readable index is regenerated by python3 scripts/gen_analysis_index.py and lives at docs/reference/analysis-index.yaml.

ID	Status	Type	Hypothesis	Question
AN-001	done · 🟡	causal	H:price-discipline-loss	Does open competition lower negotiated procurement prices in switched group 65, identified by DiDiR around the March 2018 PGE-SP reinterpretation?
AN-002	done · 🟡	causal	H:entry-thickens-pool	Does open competition raise the number of bidder firms and valid bids in switched group 65?
AN-003	done · 🟡	causal	H:distance-widens-under-open	Does open competition widen the average distance between the public buyer and the winning supplier?
AN-004	done · 🟡	placebo	H:parallel-trends-hold	Do placebo treatment dates applied to pre-treatment data yield null price effects, as required for parallel trends?
AN-005	done · 🟡	robustness	H:parallel-trends-hold	How tight are Lee (2009) sample-selection bounds on the DiDiR price coefficient?
AN-006	done · 🟡	robustness	H:parallel-trends-hold	Does the price effect survive HonestDiD (Rambachan-Roth 2023) robust confidence intervals under controlled violations of parallel trends?
AN-007	done · 🟡	causal	H:price-discipline-loss	How does the price treatment effect vary across the conditional price distribution?
AN-008	done · 🟡	descriptive	H:sme-winner-share-falls	What share of the reduced-form price effect is attributable to the competition (more bidders) vs composition (winner-type) channels?
AN-009	done · 🟡	descriptive	H:sme-winner-share-falls	Does the probability of an SME winner fall under open competition, and is the composition shift symmetric across PBU types?
AN-010	done · 🟡	descriptive	H:exclusion-dominates	What is the structural decomposition of the set-aside price effect into the lost-discipline channel (\(S_2-S_1\)) and the protected-pool offset (\(S_3-S_2\))?
AN-011	done · 🟡	descriptive	H:static-welfare-loss-large	What is the per-auction static welfare loss of the full SME-only set-aside, decomposed into allocative DWL and MCPF distortion at λ=0.30?
AN-012	done · 🟡	descriptive	H:preference-near-zero-cost	What is the simulated price and SME win-rate effect of a 10% SME price preference relative to the open benchmark \(S_1\)?
AN-013	done · 🟡	descriptive	H:ipv-clock-admissible	How are losing-bidder drop-out prices extracted from BEC Pregão event-log data, and how robust are downstream simulations to the cost-distribution recovery regime?
AN-014	done · 🟡	robustness	H:ipv-clock-admissible	Does the Krasnokutskaya-style auction-level heterogeneity correction (a) remove common scale shocks without mechanically generating the exclusion-dominant decomposition, and (b) does relaxing within-auction independence to a Gaussian copula leave the decomposition intact?
AN-015	done · 🟡	robustness	H:no-collusion-confound	Do Conley-Decarolis close-pair and Bajari-Ye bid-coordination screens detect a post-policy intensification of bidder clustering that would invalidate the IPV-clock reading of Pregão drop-outs?
AN-016	done · 🟡	robustness	H:static-welfare-loss-large	How does the structural decomposition and welfare arithmetic behave in pharmaceuticals, and what specifically makes pharmaceuticals a boundary case rather than a second headline?
AN-017	done · 🟡	robustness	H:exclusion-dominates	Does the exclusion-dominant decomposition survive a strict-invariance specification that holds the SME pool composition fixed at the pre-policy distribution while keeping observed post-policy SME entry counts?
AN-018	done · 🟡	robustness	H:parallel-trends-hold	Do Callaway-Sant'Anna (2021) and Sun-Abraham (2021) staggered-DiD estimators reproduce the DiDiR price effect once heterogeneous treatment timing is accounted for?
AN-019	done · 🟡	robustness	H:ipv-clock-admissible	Does a GPV (Guerre-Perrigne-Vuong 2000) recovery of the bidder cost distribution from Convite first-price bids converge with the Pregão drop-out recovery, providing cross-modality discipline on the IPV-clock interpretation?
AN-020	done · 🟢	robustness	H:parallel-trends-hold	Does the Goodman-Bacon (2021) decomposition of the TWFE DiD coefficient show that all weight falls on the clean treated-vs-never-treated 2×2 comparison, with no contamination from forbidden treated-vs-already-treated comparisons?
AN-021	done · 🟡	robustness	H:parallel-trends-hold	Does a synthetic-control match of group 65 to a weighted donor pool of never-treated product groups validate parallel trends and reproduce the DiDiR price effect?
AN-022	done · 🟡	robustness	H:exclusion-dominates	What is the 95% bootstrap confidence interval on the absolute exclusion share, and does the lower endpoint exclude 50% (the dominance threshold)?
AN-023	done · 🟡	robustness	H:parallel-trends-hold	Does the DiD price coefficient survive dropping the 8-month BEC enablement window (Jul 2017–Feb 2018) during which the SME-only OC functionality was operationalized?
AN-024	done · 🟡	robustness	H:static-welfare-loss-large	How does the static welfare loss vary across the MCPF λ grid, and what is the 95% bootstrap CI on the loss as a percent of the open-regime price?
AN-025	done · 🟡	descriptive	H:static-welfare-loss-large	How does the annualized fiscal welfare cost on Group 65 vary across plausible SME-only adherence rates, and what is the baseline empirical adherence?
AN-026	done · 🟡	robustness	H:no-collusion-confound	Does a persistent-pair coordination screen conditional on CADMAT product class detect within-class bidder coordination, and does the persistent-pairing signature survive class conditioning?
AN-027	done · 🟡	robustness	H:price-discipline-loss	Does the price effect appear both inside CMED-regulated medications (class 6531) and outside it (other medical supplies), as required if competition — not pharmaceutical-specific inflation — drives the headline?
AN-028	done · 🟡	causal	H:sme-winner-share-falls	Does the price effect survive restricting the sample to RAIS-validated SME winners, and does the distance effect vanish in that subsample as the firm-size composition channel would predict?
AN-029	done · 🟡	causal	H:entry-thickens-pool	Is the open-competition price effect driven purely by more bidders (headcount channel), or does it persist when the bidder count is held fixed (the composition / order-statistic channel)? And does open competition raise the extensive margin (tender completion rate)?
AN-030	done · 🟡	robustness	H:protected-pool-responds	What is the Pre→Post entry-rate shift in each modality × class × type cell, what is the calibrated entry-cost asymmetry that explains why SMEs cannot fully replace non-SMEs, and does the dominance ordering survive across the 2×2 method grid?

Index by hypothesis¶

This view groups AN pages by the hypothesis they bear on. Mirrors the scorecard in hypotheses/index.md — that view groups by prediction.

Cluster A — Reduced-form policy effects¶

H:price-discipline-loss: AN-001, AN-007, AN-010, AN-018, AN-020, AN-027
H:entry-thickens-pool: AN-002, AN-029
H:distance-widens-under-open: AN-003, AN-028
H:sme-winner-share-falls: AN-008, AN-009, AN-028

Cluster B — Price formation and decomposition¶

H:exclusion-dominates: AN-010, AN-014, AN-017, AN-022
H:protected-pool-responds: AN-010, AN-030
H:ipv-clock-admissible: AN-013, AN-014, AN-015, AN-019

Cluster C — Static welfare and policy design¶

Cluster D — Identification robustness¶

H:parallel-trends-hold: AN-004, AN-005, AN-006, AN-018, AN-020, AN-021, AN-023
H:no-collusion-confound: AN-015, AN-026