Skip to content

Manuscript

Download

JPubE submission, May 2026.

Title

The Price of Exclusion: SME Set-Asides in Public Procurement

Contribution

The paper makes three contributions:

  1. A price-formation decomposition. I decompose the price effect of an SME set-aside into a lost-discipline component (non-SMEs removed from the price-forming pool, SME pool held fixed) and a protected-pool offset (post-policy SME pool replacing the pre-policy one). The decomposition turns the set-aside price effect into a design question: how much is mechanical exclusion of rival bidders, and how much is the protected pool's response?

  2. Implementation via a reverse-auction setting. In São Paulo's Pregão (electronic English-reverse auction), losing bidders' drop-out prices reveal type-specific willingness to supply under the maintained independent-private-values clock interpretation (Vickrey 1961; Milgrom and Weber 1982; Haile and Tamer 2003; Athey and Haile 2002). I use the recovered primitives, an auction-level heterogeneity correction in the spirit of Krasnokutskaya (2011), and observed equilibrium entry to simulate counterfactual price-formation objects under three pools.

  3. Recast of policy design. The relevant policy frontier is not SME support versus no support, but exclusionary redistribution versus support that preserves the price-forming bidder pool. A 10 percent SME price preference enters as a static design benchmark: it preserves the non-SME bidders that discipline the price-forming order statistic but delivers less redistribution than full exclusion.

In standardized non-pharmaceutical procurement, the protected pool responds (SME participation roughly doubles) but does not replace the excluded discipline: the full set-aside generates a static welfare loss of 28.9% of the open-regime price at λ=0.30, with the exclusion component accounting for ~72% of the decomposition. The implied SME welfare weight required for a planner to prefer full exclusion is 2.42. Pharmaceutical procurement is reported as a boundary case, not as a second headline: there the protected pool is thinner, composition changes more under the policy, and the welfare ranking becomes sensitive to how the post-policy SME pool is modeled.

Institutional Background

Since 2005, all public buyer units (PBUs) in São Paulo state procure common goods and services through Bolsa Eletrônica de Compras (BEC), the centralized electronic platform. Brazil's federal SME statute (LC 123/2006) requires exclusive SME tenders for items valued at R$80,000 or less, and LC 147/2014 made the SME-only rule mandatory. For most goods on BEC, that rule already operated before the period studied here.

Group 65 (medical, dental, and hospital equipment and supplies) was different for a legal reason, not a market one. The prevailing interpretation applied the R$80,000 threshold to the total value of the purchase notice, not to each item. Because medical-supply notices bundle many individually small items inside large purchase orders, the total-notice reading kept many individually eligible items outside the SME-only default. A sequence of administrative acts moved them in: BEC enabled SME-only functionality (COMUNICADO BEC 02/2017, July 2017), and PGE-SP Parecer 151/2017 (December 2017) held that the R$80,000 threshold applies item-by-item regardless of the total notice value. The operational cutoff is March 2018, when Group 65 buyer units began mass take-up of the functionality; TCE-SP ratified the new reading in May 2018, after the cutoff. The trigger was legal and administrative — not a medical-supply policy, a price intervention, or a technology shock.

Group Before March 2018 After March 2018
Group 65 (switched) Open competition (threshold applied to total notice) SME-only default (threshold applied item-by-item)
Other groups (always treated) SME-only default SME-only default

Data and Sample

Feature Detail
Source BEC administrative records (SEFAZ/SP)
Coverage All standardized goods procurement in São Paulo state
Period 18-month structural window Sep 2016 – Aug 2019 (each side of the March 2018 cutoff); broader reduced-form panel spans 2016–2019
Raw extract 3.7M bid-level observations; 1,344 PBUs; 82,569 items; 832,984 purchase orders
Treatment group Group 65 (medical/hospital supplies, ~27% of platform transactions)
Controls 76 never-treated product groups (reduced-form benchmark)
Structural sample 297,967 firm-auction obs across 97,993 Group 65 Pregão auctions (48,740 pre / 49,253 post)
Treatment date March 2018 (operational mass take-up)

Empirical Strategy

The argument rests on three pieces of evidence that carry separate identifying weight. A weakness in one does not mechanically overturn the others; it changes which interpretation the evidence can support.

1. Reduced-form benchmark (timing, sign, scale)

A difference-in-differences compares Group 65 to 76 never-treated product groups around March 2018, written in a DiD-in-reverse convention (a negative pre-period coefficient is the open-regime discount that disappears under the SME-only extension):

\[y_{pigt} = \eta_i + \gamma\,\text{Pre}_t + \beta\,(g65_{pgt} \times \text{Pre}_t) + x\,\delta + \varepsilon_{pigt}\]

where \(\eta_i\) are item fixed effects, \(\text{Pre}_t = 1\) before March 2018, \(g65\) marks Group 65, \(x\) includes log quantity and tender format, and errors are item-clustered. The 18-month coefficient is −0.113 (a 10–11% price movement), stable across 6-, 12-, and 18-month windows. This benchmark is not asked to carry the structural magnitude — it verifies that the institutional episode left the predicted footprint in timing and sign.

2. Structural price-formation decomposition (the mechanism)

The core of the paper. In the Pregão (English-reverse auction), losing bidders' drop-out prices reveal type-specific willingness to supply under the maintained IPV-clock interpretation. Recovering type-specific cost distributions (with a Krasnokutskaya-style auction-level heterogeneity correction) and using observed equilibrium entry, the set-aside price effect is simulated under three counterfactual bidder pools and decomposed:

\[\underbrace{p_{S_3}-p_{S_1}}_{\text{net set-aside effect}} = \underbrace{p_{S_2}-p_{S_1}}_{\Delta^{\text{excl}}\ \text{lost discipline}} + \underbrace{p_{S_3}-p_{S_2}}_{\Delta^{\text{pool}}\ \text{protected-pool offset}}\]

This conditions on within-period type-specific primitives and equilibrium bidder counts, so it requires neither cross-group balance nor the DiD magnitude. See Results for the decomposition.

3. Static welfare comparison (the policy)

Mapping the recovered primitives into allocative deadweight loss, the MCPF distortion (\(\lambda = 0.30\)), and SME producer surplus (Saez–Stantcheva weights), the full set-aside \(V_0\) is compared to a static 10% price preference \(V_3\) that preserves the non-SME price-forming pool.

Time windows: 6 months (Sep 2017–Aug 2018), 12 months (Mar 2017–Feb 2019), and 18 months (Sep 2016–Aug 2019).

Software and Estimation

Component Detail
Language R 4.5
Fixed effects fixest::feols() with lean estimation
Data handling data.table + arrow (parquet cache)
Clustering Item-level (main), group-level (event study)
Figures ggplot2 with cairo_pdf, grayscale theme
Threads 16 (fixest and data.table)