H:cobidder-profile-distinct — FL cobidders are economically distinct from other frequent losers

If the frequent-loser ranking captures more than raw participation intensity, then cobidders inside the FL stratum should look different from other firms in the same stratum along economically meaningful dimensions: buyer-portfolio breadth, product-portfolio concentration, proximity to legal cartel anchors, and bid-level behavior consistent with credible losing roles.

Intuition (plain-language)

Within the frequent-loser group, are cobidders qualitatively different from other firms in the same group, or are they just the highest-volume members? The robust answer is structural: cobidders specialize in narrower product portfolios, spread across more distinct winners, sit closer to legal cartel anchors, and bid closer to the winning price. Cobidders bid in ~1.8× more tenders, so the obvious worry is volume — but the volume-matched audit (AN-041) strongly supports the distinction: matching on participation count (SMD 0.49 → 0.00), the structural distinctness holds or strengthens. That is the strongly-supported core. The bid-conduct side is narrower and honestly bounded: one channel survives (bidding closer to the winner), while the bid-dispersion sub-signal (AN-041) and the bid-timing channel (AN-042) are documented nulls. The remaining negatives — the across-cell price/mechanism heterogeneity (AN-032) and the first-time-FL channel (AN-021) — are about a causal-mechanism reading the hypothesis does not assert, not about the structural distinctness. Promotion beyond strongly-supported needs non-BEC replication.

Evidence strength: Partial (strongly supported). The structural/profile distinctness is robust across many dimensions and survives volume matching (the strongly-supported core). The bid-conduct side is one surviving channel with two documented nulls; the across-cell price/mechanism heterogeneity and the first-time-FL channel are scope limits on a causal-mechanism reading the hypothesis does not assert. Eight lines of evidence: (i) Participation profile (AN-008): Cohen's d +0.67 tenders, +1.00 unique winners, +0.46 CADE-facing, −0.32 n item groups, vs non-cobidder FLs. (ii) Network / HHI (AN-009): cobidder portfolio HHI 0.380 vs 0.288 (d = +0.39); FL-bidder-present markets have lower winner-side HHI (0.178 vs 0.303) — cobidders sit in contestable winner environments. (iii) Standardized-diff battery (AN-028): 7 dimensions, d 0.19–1.00, all Wilcoxon p < 10⁻⁵ given N = 191. (iv) Bid-level behavior (AN-031): cobidder median gap-to-winner 0.582 vs 0.809 for non-cobidder FLs (d = −0.28, p < 10⁻⁶); SD of gap +0.15. Distinctness at the bid level, not just participation. (v) Unified mechanism quadrants (AN-024): descriptive cell heterogeneity, with the Low-Low cell carrying the mass. (vi) Matched-heterogeneity audit (AN-032): the quadrant heterogeneity LARGELY DOES NOT SURVIVE PS matching — Low-HHI × Low-pairs cell drops from +0.090 (p = 0.048) unmatched to +0.066 (p = 0.227) matched. Reported as the negative finding that defines the causal-mechanistic limit. First-time-FL channel (AN-021) demoted to appendix (+0.10 unconditional → +0.062 PS-matched, p = 0.312). (vii) Volume-matched profile audit (AN-041): matching cobidders to FL non-cobidders on tenders_count (SMD 0.49 → 0.00), product-portfolio HHI (d +0.47), winner-pair spread (−0.56), and median gap-to-winner (−0.25) hold or strengthen — the structural distinctness is not a volume artifact. The AN-031 bid-dispersion elevation collapses (sd_gap +0.05, p = 0.87) and is dropped. (viii) Volume-matched timing audit (AN-042): a candidate second bid-conduct channel — pregão bid timing (revision intensity, inter-bid interval, last-bid position, engagement span) — does NOT survive matching (all Wilcoxon p ≥ 0.23). Documented null: the bid-conduct distinctness stays single-channel (gap-to-winner). The strongly-supported core is the structural distinctness; promotion to 🟢 (Confirmed) requires non-BEC replication (the binding constraint, as for every other hypothesis), since the claim is descriptive and single-source.

Theory

Cartel theory predicts that cover bidders have specific operational profiles: deployment patterns that put them in the relevant tenders, and bid-level submission patterns consistent with credible non-winning roles \citep{marshall2012economics,kawai2022detecting,clark2021collusion}. The hypothesis is that these distinguishing characteristics survive within the FL stratum — i.e., conditional on persistent zero-win participation.

Prediction

Within the always-loser FL stratum, cobidders should:

1. be deployed across more buyers and tenders (broader operational footprint) than non-cobidder FLs;
2. operate in more concentrated product portfolios;
3. appear closer to legal cartel anchors in network terms (e.g., shared tenders, shared buyers);
4. exhibit bid-level patterns consistent with — but not diagnostic of — credible losing roles.

Competing prediction

Volume confound within stratum. Cobidders could just be the highest- volume FLs, producing the same operational profile mechanically. The test must compare within the stratum, controlling for tenders-count.

Case evidence

CADE adjudication records describe operational patterns that match the predicted profile in specific cases (e.g., coordinated bidding in hospital-supply procurement). The systematic test asks whether these patterns hold in the cross-section of FL cobidders.

Empirical test

• Sample: always-loser FL14 stratum.
• Outcomes: buyer breadth, product concentration (HHI), network proximity, bid-level moments (mean bid, dispersion).
• Specification: cobidder indicator regressed on each outcome with tenders-count and exposure-stratum controls.

Data requirements and limitations

Requires the firm-level panel with buyer / product / network features. Limitation: bid-level moments are consistent with credible losing roles but are not diagnostic — the manuscript explicitly distinguishes consistent patterns from forensic proof.

Evidence

Analysis	Bearing	Status	Key takeaway
AN-008 (FL characterization)	Direct	done	Cohen's d +0.67 tenders, +1.00 unique winners
AN-009 (network proximity + HHI)	Direct	done	Cobidder HHI 0.380 vs 0.288 (d = +0.39); FL-winner HHI 0.178 vs 0.303
AN-024 (unified mechanism)	Supports	done	Profile descriptive — drop "assinatura de cartel" framing
AN-028 (within-FL standardized diffs)	Direct	done	7 dimensions, Cohen's d 0.19–1.00 vs FL_non_cobidder; all Wilcoxon p < 10⁻⁵
AN-031 (bid-level behavior)	Direct	done	Cobidder median gap-to-winner 0.582 vs FL 0.809 (d = −0.28, p < 10⁻⁶); bid-level signal beyond volume
AN-021 (first-time FL — demoted)	Scope (mechanism)	done	+0.062 (p = 0.312) under PS matching; a causal-channel extension that does not survive — not the structural distinctness claim
AN-032 (matched-heterogeneity audit)	Scope (mechanism)	done	Across-cell price/mechanism heterogeneity attenuates under matching: Low-Low loses significance (p 0.048 → 0.227). Bounds the mechanism reading, not the profile
AN-041 (volume-matched profile audit)	Direct	done	Structural distinctness survives PS matching on tenders_count (SMD 0.49 → 0.00): HHI d +0.47, repeat-spread −0.56, median gap −0.25 all hold or strengthen. Casualty: bid-dispersion (sd_gap) collapses to +0.05 (n.s.) — a volume artifact
AN-042 (volume-matched timing audit)	Null (bid-conduct)	done	Candidate 2nd bid-conduct channel: pregão bid timing does NOT survive matching (all Wilcoxon p ≥ 0.23). Bid-conduct distinctness stays single-channel (gap-to-winner)

Open tests

• Robustness to alternative cobidder definitions.
• Heterogeneity within FL stratum by sub-period.
• Bid-timing / inter-bid-interval moments under volume matching — done (AN-042): a candidate second bid-conduct channel that returns a null, so the bid-conduct distinctness stays single-channel (gap-to-winner, AN-041). The structural distinctness remains the strongly-supported core.
• Non-BEC replication (ComprasNet) of the structural distinctness — the binding path to 🟢 Confirmed.

Why not 🟢 Confirmed?

H5 is at Partial (strongly supported) on the structural distinctness: that core survives the strongest within-data audit (volume matching) across multiple dimensions. Confirmed is blocked by the same constraint as every other hypothesis — non-BEC replication — and the bid-conduct side is honestly narrower than the structural side.

1. The negatives are about mechanism, not the profile. The structural distinctness survives volume matching (AN-041): product specialization, winner-pair spread, and bid proximity to the winner hold or strengthen after matching on tenders_count (SMD 0.49 → 0.00). What does NOT survive is (a) the across-cell price/mechanism heterogeneity — the quadrant interaction (AN-032) attenuates 26–35% and Low-Low loses significance — and (b) the first-time-FL channel (AN-021), which loses significance entirely. These bound a causal-mechanism reading the hypothesis does not assert; the structural firm-type distinctness stands.

2. Bid-conduct distinctness is single-channel. The only bid-conduct signal that survives matching is the median gap-to-winner (AN-031: d = −0.28 raw, −0.25 volume-matched). Two candidate companions are documented nulls: the within-firm bid dispersion collapses to d = +0.05 (p = 0.87, AN-041), and the bid-timing battery (revision intensity, inter-bid interval, last-bid position, engagement span) does not survive matching (AN-042: all Wilcoxon p ≥ 0.23). The distinctness is therefore structural, not a multi-channel behavioral signature.

3. No non-BEC replication — the binding Confirmed constraint. Same boundary as H1, H3, H4: the claim is descriptive and single-source (BEC × CADE), so the structural signature could reflect a feature of the BEC mechanism rather than loser-role behavior in general. ComprasNet (or another panel) is the path to 🟢.

H5 stays at Partial (strongly supported): the structural distinctness is robust and survives volume matching; the bid-conduct side is one channel with two documented nulls; and Confirmed awaits non-BEC replication.