Skip to content

Hypotheses

Testable predictions for Cheap Signals, Costly Proof: Award-Layer Triage for Cartel Enforcement (Genicolo-Martins & Furquim de Azevedo). Each hypothesis lives in its own file and is referenced project-internally by H:<slug>.

Hypotheses are organized into thematic clusters aligned with the paper's sections in work/v18-editor/submission_clean/.

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-strategic explanation predicts instead — the alternative the test has to rule out.
  • Case evidence: anecdotal grounding from CADE decisions and case documents.
  • Empirical test: the concrete specification — outcome, variation, specification, 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.


Cluster A: Award-layer triage scope

Maps to paper §3 (award-layer triage) and §4.1, §4.3 (baseline + scope check).

  • H:cobidder-concentration — The frequent-loser ranking concentrates CADE-adjacent cobidders in the always-loser stratum.
  • H:direct-defendants-null — The same ranking does not discriminate direct CADE defendants (loser-side scope by design).

Cluster B: Audit discipline

Maps to paper §4.2 (exposure and timing discipline) and Appendix Validation Audits.

  • H:exposure-discipline — Cobidder concentration survives participation-volume placebos and opportunity-set audits.
  • H:timing-discipline — Cobidder concentration survives relaxed and strict timing rules.

Cluster C: Cobidder economic profile

Maps to paper §5 (the economic profile of cartel-adjacent losers).

  • H:cobidder-profile-distinct — Inside the FL stratum, cobidders differ from other frequent losers along buyer breadth, product concentration, network proximity, and bid-level patterns.

Cluster D: Sequential architecture

Maps to paper §6 (award → bid layer sequencing).

  • H:award-bid-complementarity — Award-layer and bid-distribution screens enter at different stages; the joint signal is the full-observability upper bound.
  • H:gatekeeping-cost-of-evidence — Award-layer gatekeeping cuts the bid-microdata pool ≥80% while recovering the majority of adjudicated cobidders.

Cluster E: Price scope

Maps to paper §7 (price evidence as scope, not damages).

  • H:price-scope-sign-reversal — The price coefficient sign reverses across validation targets, supporting a scope interpretation rather than a damages reading.

Scorecard

Status runs Not yet testedNot confirmed / Mixed / PartialConfirmed. Single-source own-project estimates anchored in BEC 2009–2019 and CADE adjudications; promotion to Confirmed would require independent replication on a non-BEC procurement panel.

# Prediction Headline evidence Evidence Status
H1 FL score concentrates CADE cobidders. AUC 0.924 (FL14) / 0.939 (continuous); 131/193 cobidders recovered; survives 8 audits. AN-004, AN-005, AN-006, AN-011, AN-014, AN-017, AN-025, AN-026 Partial (strongly supported)
H2 FL score is null on direct CADE defendants. AUC 0.491 [0.461, 0.520]; 14.9% of direct defendants are always-losers (D4). AN-007, AN-018 Partial (strongly supported)
H3 Concentration survives exposure-disciplined placebos. Sham permutation p < 1/2,000 (observed 32 σ above sham mean); leakage audit retains AUC 0.864; scope matrix row 4 = 0.383 (score repels winner-heavy defendants); within-FL Cohen's d 0.19–1.00. AN-005, AN-014, AN-027, AN-028 Partial (strongly supported)
H4 Concentration survives strict timing discipline. Three-classifier battery preserves AUC 0.79–0.90 across clf_2015/clf_2017 × cobid_all/cobid_post2019; firm persistence between windows = 8.7% (genuinely fresh population); strict ex ante firm AUC 0.767; precision retention 53%. AN-006, AN-013, AN-014, AN-029, AN-030 Partial (strongly supported)
H5 FL cobidders are economically distinct from other FLs (structural). Structural distinctness survives volume matching (SMD 0.49→0.00): HHI d +0.47, repeat-spread −0.56, median gap −0.25 hold or strengthen — not a volume artifact. Bid-conduct is single-channel: dispersion (n.s.) and timing (AN-042, all p≥0.23) are documented nulls. Scope limits (not the profile claim): quadrant price-heterogeneity attenuates 26–35%; first-time-FL demoted (p=0.31). AN-008, AN-009, AN-024, AN-028, AN-031, AN-032, AN-021, AN-041, AN-042 Partial (strongly supported)
H6 Award and bid layers are complementary. Imhof+FL Δ vs Imhof = +0.096 (DeLong p = 1.2e-26); FL alone vs Imhof Δ = +0.035 (p = 0.014); sequential K=2,000 captures 74% joint recall at 17% bid-microdata footprint. AN-010, AN-011, AN-015, AN-033, AN-034 Partial (strongly supported)
H7 Gatekeeping cuts the bid pool ≥80% with majority cobidder recovery. 83% pool cut + 131/193 cobidders (in-sample); sequential K=2,000 captures 103% of joint TP at 24% microdata cost (temporal holdout); CV precision SD ≤ 0.011. AN-012, AN-013, AN-014, AN-034, AN-035, AN-036 Partial (strongly supported)
H8 Price coefficient sign reverses across targets — scope, not damages. Broad +0.064 → overlap ATT −0.097 (p = 1.7e-10); negative in 14 of 15 subgroup cells; item-group decomp predictably structured; modal asymmetry 2.45×. Decomposition (descriptive, not mechanism identification): selection into high-price cells (Δ 5.58 log-pts) plus a within-cell bidder-count association (+0.507 log-bidders). AN-019, AN-020, AN-016, AN-022, AN-037, AN-038, AN-039, AN-040 Partial (strongly supported)

Generated index

All 8 hypothesis pages, auto-generated from the YAML frontmatter of each docs/hypotheses/<slug>.md via scripts/render_indexes.py. Maintained in lockstep with the scorecard above.

H# Cluster Paper section Status Slug + title
H1 A §3 + §4 Partial (strongly supported) cobidder-concentration — The frequent-loser ranking concentrates CADE-adjacent cobidders
H2 A §4.3 Partial (strongly supported) direct-defendants-null — The frequent-loser ranking does not discriminate direct CADE defendants
H3 B §4.2 Partial (strongly supported) exposure-discipline — Cobidder concentration survives exposure-disciplined placebos
H4 B §4.2 Partial (strongly supported) timing-discipline — Cobidder concentration survives timing discipline
H5 C §5 Partial (strongly supported) cobidder-profile-distinct — FL cobidders are economically distinct from other frequent losers
H6 D §6.1 + §6.2 Partial (strongly supported) award-bid-complementarity — Award-layer and bid-layer information are complementary
H7 D §6.3 + §6.4 Partial (strongly supported) gatekeeping-cost-of-evidence — Award-layer gatekeeping cuts the bid-microdata pool cost-effectively
H8 E §7.1 + §7.2 Partial (strongly supported) price-scope-sign-reversal — Price evidence carries scope information, not damages

Status legend. Not yet testedNot confirmed / Mixed / PartialPartial (strongly supported)Confirmed. Promotion to Confirmed requires non-BEC replication; see COMPRASNET_PATH_TO_CONFIRMED.md.