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 non-pharma¶

The set-aside raises government payments and reassigns the contract to a higher-cost supplier. The static welfare cost combines real allocative waste (DWL$_{\mathrm{alloc}} = c_{(1)}^{V_0} - c_{(1)}^{S_1}$) with the fiscal distortion from raising the additional public funds at the marginal cost of public funds λ. The headline magnitude in standardized non-pharma is large — ~28.9% of the open-regime price at λ=0.30. The implied annualized welfare cost on Group 65 alone spans R$38–89M per year (US$11–25M).

Intuition (plain-language)

The set-aside costs society twice: it hands the contract to a higher-cost supplier (real resource waste) and it forces the government to raise more tax money to pay the higher price (costly because public funds carry a deadweight markup). Adding both at λ=0.30 gives a static welfare loss of about 29% of the open-regime price in ordinary goods — on Group 65 alone, roughly R$38–89M per year.

Evidence strength: Partial (strongly supported). AN-011 reports 28.9% non-pharma / 44.8% pharma at λ=0.30. AN-024 extends this with: (i) λ sensitivity — loss runs 24.1% → 33.8% across λ ∈ [0.15, 0.45] in non-pharma, 38.1% → 51.6% in pharma; (ii) bootstrap CI at λ=0.30 = [20.5, 34.8] non-pharma / [34.9, 55.9] pharma; (iii) welfare ranking $V_3 \succ V_0$ stable across the entire λ range in non-pharma (under both main and strict-invariance specs); pharma ranking determined by main vs strict-invariance, not by λ. AN-025 translates to R$38–89M/yr (US$11–25M) annualized on Group 65 alone across adherence rates 30%–70%; empirical baseline 43% → R$55M/yr (US$16M).

Theory¶

The static per-auction welfare loss combines the allocative wedge $\mathrm{DWL}_{\mathrm{alloc}} = c_{(1)}^{V_0} - c_{(1)}^{S_1}$ with an MCPF distortion $\lambda \cdot (p^{V_0} - p^{S_1})$ (Ballard, Shoven, and Whalley 1985; Hendren and Sprung-Keyser 2020; Finkelstein and Hendren 2020). The first term is real allocative waste; the second is the fiscal distortion. The remainder of the payment increase is a transfer to the protected winner and enters welfare only weighted by the planner's SME welfare weight (see H:implied-welfare-weight-large).

Prediction¶

In standardized non-pharma, the welfare loss as a share of the open-regime price $p^{S_1}$ should be large — paper headline ~28.9% at λ=0.30. The bootstrap lower endpoint should remain economically meaningful (~20%+). Pharma should show a larger point estimate but wider model sensitivity.

Competing prediction¶

The transfer-to-SMEs is welfare-positive at conventional weights. A utilitarian planner with no SME-specific weight (weight = 1) would count the additional SME producer surplus dollar-for-dollar against the fiscal+allocative cost. The welfare arithmetic in the paper deliberately excludes this transfer from the social-cost accounting — the loss number is the non-transfer component. A planner who counts the transfer at full weight would still face the allocative + MCPF cost calculated here, but the implied welfare weight required for a preference reversal would be different (see H:implied-welfare-weight-large).

Setting evidence¶

Group 65 (medical/hospital supplies) accounts for ~27% of BEC volume. São Paulo's BEC platform handles ~R$13B/year in standardized goods procurement. The R$38–89M/year welfare cost is therefore on a single product group of a single state's procurement platform.

Empirical test¶

Outcome variables: $\Delta_{\mathrm{gov}}$ (payment increase), DWL$_{\mathrm{alloc}}$ (allocative waste), MCPF distortion, total Loss / $p^{S_1}$. All in reference-price units.
Variation: counterfactual policy regime $V_0$ (full SME-only) vs $S_1$ (open).
Specification: BNE simulation with recovered type-specific cost distributions, $\lambda = 0.30$ baseline (Ballard-Shoven-Whalley benchmark); bootstrap CIs from v7-jpube-tight/scripts/56_welfare_bootstrap.R.
Sample: structural cells (non-pharma standardized; pharma boundary case).

Data requirements and limitations¶

The welfare arithmetic depends on (i) the IPV-clock interpretation holding (H:ipv-clock-admissible); (ii) the λ benchmark — λ=0.30 is the Ballard-Shoven-Whalley midpoint; sensitivity to λ is reported in §5 of the paper; (iii) the SME-pool composition under the policy (H:protected-pool-responds).

Evidence¶

Analysis	Bearing	Key takeaway
AN-011	Supports	Non-pharma welfare loss = 28.9% of open-regime price at λ=0.30; pharma 44.8% but boundary case.
AN-016	Mixed	Pharma magnitudes larger but more model-sensitive; reported as boundary case, not headline.
AN-017	Supports	Strict invariance preserves non-pharma dominance ordering; exclusion share rises to 85% (NP) / 79% (PH). Welfare ranking $V_3 \succ V_0$ in NP under both specs.
AN-024	Supports	Bootstrap 95% CI at λ=0.30: [20.5, 34.8] NP / [34.9, 55.9] PH. Loss > 18% even at λ=0.20 lower-CI endpoint. Ranking stable across λ ∈ [0.15, 0.45].
AN-025	Supports	Annualized fiscal cost R$38–89M/yr across adherence 30%–70%; baseline 43% → R$55M/yr (US$16M) on Group 65 alone.

Open tests¶

Adherence-rate sensitivity¶

v7-jpube-tight/scripts/57_welfare_adherence_sensitivity.R varies the SME-eligible adherence rate to bound the realized fiscal cost. The R$38–89M annual range comes from this; documenting it as an AN page would expose the calibration explicitly.

Optimal preference comparison¶

AN-012 compares the full set-aside ($V_0$) against a 10% SME price preference ($V_3$). The welfare-cost gap between $V_0$ and $V_3$ is the magnitude of the "exclusion premium" relative to a non-exclusionary alternative; the headline reading is in H:preference-near-zero-cost.