Hypotheses

Testable predictions for The Price of Exclusion: SME Set-Asides in Public Procurement. Each hypothesis lives in its own file and is referenced project-internally by H:<slug>.

Hypotheses are organized into four thematic clusters aligned with the canonical manuscript v8-jpube/manuscript/paper_v8.tex.

How to read a hypothesis page

Each page follows the same structure — the design → what's known → what's left:

• Title + lede: the H:<slug> title, then a plain-words statement of the claim.
• Evidence-strength callout: an at-a-glance verdict (Very strong / Strong / Moderate / Weak / Not yet tested) plus a one-line status.
• Theory: the theoretical motivation, with references.
• Prediction: the precise directional claim.
• Competing prediction: what a non-exclusion explanation predicts instead — the alternative the test has to rule out.
• Setting evidence: institutional and descriptive grounding from the BEC/PGE-SP record and paper.md setting.
• Empirical test: the concrete specification — outcome, variation, fixed effects.
• Data requirements and limitations: datasets needed and threats to identification.
• Evidence: a table of the project's analyses bearing on the hypothesis — Analysis (AN-ID, linked) · Bearing (Supports / Against / Mixed / Pending) · Key takeaway. Click the analysis ID for the full design and results.
• Open tests: forward-looking only — analyses not yet run, or not yet stubbed.

The scorecard below rolls up the current status of every hypothesis against the analyses (AN-NNN) that bear on it. Individual hypothesis pages carry the detailed Evidence table and the Evidence-strength callout; those may lag the scorecard until refreshed.

---

Cluster A: Reduced-form policy effects (H1–H4)

Maps to paper §1 (Introduction reduced-form motivation) and §6 (Robustness), and to docs/results.md. The reduced-form DiDiR results that motivated the paper.

• H:price-discipline-loss — Open auctions yield lower negotiated prices than SME-only auctions in standardized procurement.
• H:entry-thickens-pool — Open auctions attract more firms and more valid bids than SME-only auctions.
• H:distance-widens-under-open — Open auctions pull in geographically distant non-SME suppliers, widening the average buyer-winner distance.
• H:sme-winner-share-falls — Open auctions reduce the probability that the winner is an SME (composition shift).

Cluster B: Price formation and decomposition (H5–H7)

Maps to paper §4 (Exclusion and the Price-Forming Pool). The structural decomposition built on Pregão drop-outs.

• H:exclusion-dominates — The lost-discipline component \(|S_2-S_1|\) dominates the protected-pool offset \(|S_3-S_2|\) in the absolute price decomposition.
• H:protected-pool-responds — Post-policy SME participation rises but does not recreate the competitive discipline supplied by the excluded non-SMEs.
• H:ipv-clock-admissible — Pregão drop-out prices admit the maintained IPV-clock interpretation as willingness-to-supply observations.

Cluster C: Static welfare and policy design (H8–H10)

Maps to paper §5 (Static Welfare and Policy Design). The welfare arithmetic and the static price-preference benchmark.

• H:static-welfare-loss-large — The full SME-only set-aside generates a static welfare loss of ~28.9% of the open-regime price in standardized non-pharmaceutical procurement at λ=0.30.
• H:preference-near-zero-cost — A 10% SME price preference preserves the non-SME bidders that discipline the price-forming pool at near-zero static welfare cost.
• H:implied-welfare-weight-large — A utilitarian planner needs an SME welfare weight ~2.42 to prefer the full set-aside over open auctions in non-pharma.

Cluster D: Identification robustness (H11–H12)

Maps to paper §6 (Robustness) and to docs/robustness.md + docs/advanced.md. The identification-robustness gauntlet.

• H:parallel-trends-hold — Pre-treatment placebos are null; HonestDiD confidence intervals survive substantial M̄ violations; Lee bounds are tight around the main estimate.
• H:no-collusion-confound — The price drop is not produced by post-policy changes in bidder coordination — Bajari-Ye, Schurter, and pair-classification screens do not flag a collusion shift.

---

Scorecard

The theory-driven prediction, the intuition behind it, the analyses that bear on it, and the current verdict. Click a hypothesis for the full design and detailed Evidence table; click an AN-NNN for the backing analysis. Status runs Not yet tested → Not confirmed / Mixed / Partial → Confirmed.

#	Prediction	Intuition	Evidence	Status
H1	Open auctions → lower negotiated prices than SME-only	Adding the price-forming non-SMEs back to the pool tightens the second-order statistic	AN-001 v8 benchmark DiDiR β=−0.113 (18m, p<0.01); AN-007 quantile DiD strongest at low quantiles; AN-010 structural confirms; AN-018 Sun-Abraham ATT 0.108 matches DDR 0.113 within 0.005; AN-020 GB clean 2×2 🟢; AN-027 survives medications/non-medications split (rules out CMED inflation)	Partial (strongly supported)
H2	Open auctions → more firms and more valid bids; price effect is composition-driven, not headcount-only	Eligibility expansion brings non-SMEs back in; at a fixed bidder count, their lower costs shift the price-forming order statistic	AN-002 +10–22% firms; AN-029 completion +10.7 pp at 18m; price β survives fixing N=2/3/≥5; price-forming gap shrinks −2.7 pp. Pure-headcount explanation rejected (composition/order-statistic, not behavioral aggressiveness)	Partial (strongly supported)
H3	Open auctions → larger buyer-winner distance	Non-SME firms span a wider geographic radius than SMEs	AN-003 +12.1km on high-value items; null on low-value; AN-028 distance β = +14.25 km full → null in RAIS-SME (≤49). Distance widening is entirely a non-SME composition channel	Partial (strongly supported)
H4	Open auctions → lower probability winner is SME	Non-SMEs outbid SMEs on price when allowed in	AN-008 composition channel +185% of Gelbach gap; AN-009 symmetric across PBU types; AN-028 RAIS-validated SME winner indicator falls −21 pp at 18m; price β survives SME-only subsample (−0.121***)	Partial (strongly supported)
H5	Lost-discipline \(\|S_2-S_1\|\) dominates protected-pool offset \(\|S_3-S_2\|\)	The excluded non-SMEs were the price-forming bidders	AN-010 absolute exclusion share 72.0% NP / 68.8% PH; AN-017 strict invariance: share rises to 85/79%; AN-022 95% CI [64.5, 86.8] NP / [61.6, 85.2] PH — lower endpoint exceeds 50% in every cell × regime	Partial (strongly supported)
H6	SME participation rises post-cutoff but does not close the gap	Protected firms enter; entry alone cannot replace discipline	AN-010 SME bidders roughly double; AN-030 NP Pregão SME +99% / non-SME −44%, net −0.25 bidders/auction; entry cost 4.7× higher for non-SMEs (R\(2.57 vs R\)0.54); intensive share ∈ [67.9%, 99.7%] across 2×2 method grid	Partial (strongly supported)
H7	Pregão drop-outs admit IPV-clock interpretation	English clock + private values + truthful exit pin down willingness-to-supply	AN-013 Turnbull NPMLE exclusion share 74%/82% (NP/PH); AN-014 UH-corrected ICCs 0.36–0.59, Gaussian-copula ρ_c=0.3 drifts share <5pp; AN-015 Conley + BY screens flat or falling; AN-019 Convite GPV aligns with Pregão in load-bearing pharma non-SME pre cell (0.712 vs 0.704), other 7 cells show ~0.2 first-price wedge	Partial (strongly supported)
H8	Full SME-only set-aside generates large static welfare loss	DWL\(_{\mathrm{alloc}}\) + λ · MCPF arithmetic at λ=0.30	AN-011 28.9% NP / 44.8% PH; AN-024 bootstrap CI [20.5, 34.8] NP / [34.9, 55.9] PH at λ=0.30; stable 24–33% NP across λ ∈ [0.15, 0.45]; AN-025 annualized R\(38–89M/yr across 30–70% adherence (~R\)55M at the empirical 43%)	Partial (strongly supported)
H9	10% SME price preference preserves the price-forming pool at near-zero static cost	Eligible non-SMEs continue to discipline the price	AN-012 Δp = −0.004 NP / +0.002 PH; SME win-rate gain +4.3pp NP / +1.4pp PH; AN-024 ranking \(V_3 \succ V_0\) stable across λ ∈ [0.15, 0.45] in NP	Partial (strongly supported)
H10	Planner needs SME welfare weight ~2.42 to prefer full set-aside	Saez-Stantcheva inversion of the welfare-loss arithmetic	AN-011 implied weight 2.42 NP / 2.61 PH; AN-024 ω > 1.5 even at λ=0.15; λ choice does not rescue the set-aside	Partial (strongly supported)
H11	Placebo nulls + HonestDiD survives M̄ + Lee bounds tight + staggered-DiD reproductions converge + Goodman-Bacon clean + synth pre-fit exact	DiDiR-in-reverse exploits the 2018 PGE-SP legal reversal	AN-004 price placebos small (−0.013/−0.030/−0.034) vs real cutoff −0.108 to −0.142; AN-005 Lee bounds tight; AN-006 survives substantial M̄; AN-018 SA 0.108 vs DDR 0.113 (Δ<0.005); AN-020 GB weight=1.000 on clean 2×2 (🟢); AN-021 synth pre-gap 0.0000, post-ATT 0.171, placebo p=0.103	Partial (strongly supported)
H12	The price drop is not a collusion shift	If bidder coordination loosened post-2018, drop-outs lose the IPV interpretation	AN-015 Conley flat NP / falls PH; Bajari-Ye T1 falls in both; AN-026 class-conditional persistent-pair \(T_3\) at/below null in all 4 cells (p ∈ [0.730, 1.000]) — persistent meeting is segmentation, not coordination	Partial (strongly supported)