AN-030: Entry rates, entry costs, and decomposition robustness grid¶
Intuition (plain-language)
Why can the protected SME pool not simply replace the excluded non-SMEs? Calibrate entry costs: SME entry nearly doubles while non-SME entry falls more, for a small net drop in bidders — and non-SMEs face about 4.7× higher entry costs yet win far more often. The exclusion-dominant decomposition holds across every method-choice combination tested.
Question¶
The protected-pool offset \(S_3 - S_2\) in AN-010 shows that SME participation roughly doubles after the cutoff but does not fully replace non-SMEs. Three structural questions:
- Entry-rate symmetry: Does the SME entry response exactly replace the non-SME exit, or is there a net loss of bidders? Does the pattern differ across Pregão vs Convite modalities?
- Why doesn't the protected pool fully respond? What is the calibrated entry cost for SMEs vs non-SMEs, and does the asymmetry explain the partial offset?
- Methodological robustness: Does the exclusion-dominant decomposition survive across the 2×2 grid of UH-clean-vs-raw and endogenous-vs-fixed-pool entry choices?
Design¶
- (a) Entry rates: per-cell average bidder count by class × type × period; Pre→Post Δ.
- (b) Entry cost calibration: zero-profit condition at the entry margin, \(\kappa^k = P(\text{win} | k) \cdot \bar\pi^k \mid \text{win}\), estimated as the marginal mean profit per entry in 5,000 Monte Carlo auctions under the Pre-policy entry profile.
- (c) Decomposition grid: BNE simulation under 4 method × 2 class combinations:
- Method 1: clean UH (Krasnokutskaya correction applied)
- Method 2: raw (no UH correction)
- Method 3: endogenous entry (observed Pre→Post pool change)
- Method 4: fixed-pool entry (counterfactual no entry response)
- Cross-product: 4 method combinations × 2 classes = 8 cells.
Results¶
(a) Pre/Post entry rates (avg bidders per auction)
(tab_v3_entry_rates.tex):
| Modality | Class | SME Pre | non-SME Pre | SME Post | non-SME Post | Δ SME | Δ non-SME |
|---|---|---|---|---|---|---|---|
| Convite | NP | 1.13 | 1.72 | 1.60 | 1.24 | +0.47 | −0.48 |
| Convite | PH | 0.92 | 1.81 | 0.87 | 1.95 | −0.05 | +0.14 |
| Pregão | NP | 0.94 | 2.68 | 1.87 | 1.50 | +0.93 | −1.18 |
| Pregão | PH | 0.55 | 2.61 | 1.22 | 1.66 | +0.67 | −0.95 |
In Pregão non-pharma the SME pool nearly doubles (+99%) and non-SME falls by 44%, with net total bidders falling by 0.25 per auction. Pharma shows the same direction but smaller absolute SME gain.
(b) Entry-cost calibration (Pre-policy) (tab_v3_entry_cost.tex):
| Class | \(\tilde p^{\mathrm{ref}}\) (R$) | \(P(\mathrm{win} \mid \mathrm{SME})\) | \(P(\mathrm{win} \mid \mathrm{non}\text{-}\mathrm{SME})\) | \(\bar\pi \mid \mathrm{SME}\) | \(\bar\pi \mid \mathrm{non}\text{-}\mathrm{SME}\) | \(\kappa^{\mathrm{SME}}\) (R$) | \(\kappa^{\mathrm{non}\text{-}\mathrm{SME}}\) (R$) |
|---|---|---|---|---|---|---|---|
| Non-pharma | 13.98 | 0.176 | 0.824 | 0.222 | 0.223 | 0.54 | 2.57 |
| Pharma | 2.96 | 0.160 | 0.840 | 0.267 | 0.196 | 0.13 | 0.49 |
The entry cost is ~4.7× higher for non-SMEs than for SMEs in non-pharma (and 3.8× in pharma). Non-SMEs face higher entry costs but win 4–5× more often per attempt, with similar per-win profits in non-pharma — they enter despite the cost because the win probability is high enough.
(c) Decomposition grid (tab_v3_decomp_grid.tex):
| Class | Method | \(S_1\) | \(S_2\) | \(S_3\) | Δ total | % intensive | % entry |
|---|---|---|---|---|---|---|---|
| NP | clean + endogenous (canonical) | 0.770 | 1.110 | 1.002 | +0.233 | 75.9 | 24.1 |
| NP | clean + fixed-pool | 0.761 | 1.132 | 1.097 | +0.337 | 91.4 | 8.6 |
| NP | raw + endogenous | 0.806 | 1.230 | 1.064 | +0.259 | 71.9 | 28.1 |
| NP | raw + fixed-pool | 0.805 | 1.267 | 1.189 | +0.384 | 85.5 | 14.4 |
| PH | clean + endogenous (canonical) | 0.645 | 1.161 | 0.964 | +0.318 | 72.3 | 27.7 |
| PH | clean + fixed-pool | 0.650 | 1.265 | 0.974 | +0.324 | 67.9 | 32.1 |
| PH | raw + endogenous | 0.726 | 1.285 | 1.095 | +0.369 | 74.6 | 25.4 |
| PH | raw + fixed-pool | 0.711 | 1.213 | 1.214 | +0.504 | 99.7 | 0.3 |
The intensive (exclusion) share ranges from 67.9% to 99.7% across the 8 cells; the lowest value is well above 50%.
Interpretation¶
The protected pool responds, but asymmetrically. SME entry rises sharply (+99% in Pregão NP), but non-SME entry falls more (−44%), producing a net loss of ~0.25 bidders per auction. The protected pool cannot replace the lost discipline because (i) the SME pool starts much smaller (0.94 SMEs vs 2.68 non-SMEs per Pre auction) and (ii) doubling a smaller base does not match a 44% cut to the larger base.
Entry cost asymmetry explains the partial offset. The calibrated entry cost for non-SMEs is 4.7× SME's in non-pharma. Non-SMEs face high entry costs but high win probability — they enter when the expected value of entry justifies the cost. SMEs face low entry costs but low win probability — they enter when they have a local or specialized advantage. Removing non-SMEs from the auction does not lower the entry cost barrier for SMEs; it just removes the most productive competitors. The protected pool's response is bounded by the pre-existing SME participation — it can roughly double a smaller base, but the base itself was small because of the underlying type-cost distribution, not because of the policy.
Modality contrast. Convite shows a much smaller SME response (+0.47 in NP vs Pregão's +0.93). Convite is invitation-based first-price; the structural mechanics differ from Pregão's English-clock auction. The asymmetric SME response across modalities is itself an artifact of differing auction-format incentives, not a defect.
Decomposition grid: dominance ordering robust. Across all 8 cells of the 2×2 method × class grid, the intensive (exclusion) share stays above 50%. The lowest value, 67.9% (pharma clean + fixed-pool), is still 17 pp above the dominance threshold. The canonical specifications (clean + endogenous) for both classes align with the AN-010 headline at 75.9% NP and 72.3% PH. The exclusion-dominant ordering does not depend on the methodological choices that distinguish v3 from earlier paper versions.
Confidence: yellow. This is the cleanest within-project test of the protected-pool-response prediction. Three independent sub-analyses (entry rates, entry costs, method grid) converge on the same conclusion: the protected pool responds asymmetrically and this is not a defect of measurement or method.
Follow-ups¶
- Newcomer-vs-returning SME decomposition: which of the additional SMEs post-policy are new to BEC, vs already-active SMEs that scaled up participation? The pure-entry channel is the newcomer share; the rest is participation intensification by existing SMEs.
- Drug-class cross-cut in pharma: refines the pharma boundary statement of AN-016. The aggregate pharma response masks therapeutic-class heterogeneity.
- Endogenous entry vs structural entry model: the current
treatment uses observed Pre→Post pool change. A structural entry
model (with \(\kappa^k\) as parameter) would let the simulation
generate counterfactual entry rates under hypothetical
preference-vs-set-aside regimes; partly done in
v7-jpube-tight/scripts/61_optimal_preference.R.