AN-022: Pregão-only falsification¶
Intuition (plain-language)
Does the FL-price relationship hold across modalities, and is it the same size? Same direction, very different magnitude: binary FL raises prices +9.6% in Pregão vs +3.9% in Convite (2.45× larger), and the joint specification flips the binary sign altogether. The magnitude asymmetry is again scope — it tells you where the FL-price association concentrates — not a clean modality-specific damages estimate you could bank.
Question¶
Do FL-margin price effects differ by procurement modality, and does the Pregão-only subsample replicate the full-sample direction? The test isolates the modality contribution from the cross-modal scope variation in AN-016.
Design¶
- Sample: BEC items 2009–2019:
- Pregão: N = 543,752;
- Convite: N = 1,105,852;
- Full: N = 1,653,658.
- Specifications:
- Binary FL presence indicator;
- Continuous log(1 + tenders_count);
- Joint binary + continuous.
- Fixed effects: item, year, PBU.
- SE: item-clustered.
Results¶
Single-score specifications (binary):
| Sample | Coefficient | SE | p |
|---|---|---|---|
| Pregão only | +0.0959*** | 0.0256 | < 10⁻³ |
| Convite only | +0.0392** | 0.0188 | 0.037 |
| Full sample | +0.0653*** | 0.0216 | 0.003 |
Pregão/Convite ratio: 2.45×.
Single-score specifications (continuous):
| Sample | Coefficient | SE | p |
|---|---|---|---|
| Pregão only | +0.0262*** | 0.0063 | < 10⁻⁴ |
| Convite only | +0.0124** | 0.0049 | 0.011 |
Pregão/Convite ratio: 2.11×.
Joint specifications (binary + continuous):
| Sample | Binary coef | SE | Continuous coef | SE |
|---|---|---|---|---|
| Pregão only | −0.0652 | 0.0449 | +0.0399*** | 0.0118 |
| Convite only | −0.0828** | 0.0350 | +0.0305*** | 0.0101 |
Macros: \valFalBinPregCoef, \valFalBinConvCoef, \valFalBinFullCoef,
\valFalBinRatio, \valFalContPregCoef, \valFalContConvCoef,
\valFalContRatio, plus the \valFalJoint* series.
Figure: FL-margin coefficient on log price across four modality × score cells, with 95% CIs. Pregão binary +0.096 vs Convite binary +0.039; Pregão continuous +0.026 vs Convite continuous +0.012. Pregão coefficient ~2× larger than Convite in both binary and continuous specs.
Interpretation¶
The modal asymmetry from D2 (AN-016) replicates in the price equation: Pregão coefficients are ~2× larger than Convite across both binary and continuous specs. The asymmetry is robust to modality-restriction.
In the joint specifications, the binary coefficient flips negative in both modalities — the continuous score absorbs the persistent-loss signal, leaving the binary picking up a residual that flips sign. This sign reversal under joint specification is the same pattern as in AN-015 and is part of the scope-vs-damages evidence (H:price-scope-sign-reversal).
The cross-modal direction (Pregão > Convite) is reported as scope information — telling us where the loser-side ranking applies more strongly — not as a positive test of an institutional minimum-bidder- rule theory (per the locked rule from D2).
Follow-ups¶
- Sub-period sensitivity within each modality.
- Convite-only counterpart with small-N power diagnostic.
- Cross-sector decomposition (commodities vs services from script 52).
