H:distance-widens-under-open — Open auctions pull in geographically distant non-SME suppliers¶
Non-SME firms tend to be larger and span a wider geographic radius than SME firms in São Paulo's BEC ecosystem. If open auctions bring non-SMEs back into the pool, the average buyer-winner distance should widen. Conversely, SME-only rules should compress the geographic catchment of the public buyer. The DiDiR distance effect tests this directly.
Intuition (plain-language)
Non-SMEs are typically larger firms that can serve a wider radius; SMEs are local. So opening the auction pulls the winning supplier farther from the buyer — about 12 km — but only on high-value items, where the larger order covers the extra freight. On small orders distance barely moves. The widening distance is the visible footprint of larger, more efficient firms re-entering the pool.
Evidence strength: Partial (strongly supported). AN-003 reports +12.1 km high-value, null low-value. The cleanest test is now in AN-028: the full-sample distance β = +14.25 km vanishes in the SME-validated subsample (β = −0.28, p>0.10) and stays null in the Micro (≤9) subsample (β = +3.06). The geographic-catchment widening is entirely* a non-SME composition effect — RAIS-validated SMEs win locally; non-SMEs win at a wider radius. This is the strongest possible within-project decomposition of the distance channel.
Theory¶
Procurement competition is shaped by transport and search costs (Carril 2021; Best, Hjort and Szakonyi 2023). Larger firms amortize logistics across more contracts and can profitably bid in distant markets; smaller SMEs face binding transport costs that confine them to a local catchment. Removing larger firms from the pool therefore mechanically shrinks the geographic supply radius — an effect that is bigger when transport is a small share of contract value (i.e., on high-value items).
Prediction¶
Average distance (km) from the public buyer to the winning firm in switched group 65 should be larger in the pre-period (open) than in the post-period (SME-only), relative to always-treated controls. Operationalized as DiDiR on distance: the coefficient on \(g65 \times \text{Pre}\) should be positive and significant. The effect should be larger for high-value items where transport-cost amortization favors larger firms.
Competing prediction¶
Logistics shock. A change in interstate freight cost, fuel price, or toll structure specific to group 65 would shift the distance distribution without the SME-composition interpretation. The placebo test on distance (AN-004) comes back null (coefficients ~3.8 and −1.2 km, both p>0.10), ruling out the freight-shock confound.
Setting evidence¶
BEC records the CEP (postal code) of every registered firm and PBU.
Geocoding from these CEPs to lat/lon gives a usable buyer-winner
distance for every completed item. The administrative grain is
item-level, so the distance variable is well-defined wherever
oc_item_status == 1.
Empirical test¶
- Outcome variable: distance (km) from PBU to winner firm CEP.
- Variation: DiDiR.
- Specification: same DiDiR equation; item-clustered SEs.
- Fixed effects: item; PBU FE in second specification.
- Sample: completed items only (winner CEP defined).
Data requirements and limitations¶
CEP-to-lat/lon geocoding inherits any errors in the original CEP database. The distance variable is undefined for items without a winner. Distance is a coarse proxy for the geographic-catchment channel — a richer specification would use the firm's full operational footprint (e.g., RAIS-validated establishment locations) rather than a single CEP. The RAIS validation is partially done in AN-009 but does not enter the distance specification yet.
Evidence¶
| Analysis | Bearing | Key takeaway |
|---|---|---|
| AN-003 | Mixed | High-value items: +12.1 km (p<0.01). Low-value items: +2.8 km (n.s.). Heterogeneous along the value margin as predicted by transport-cost amortization. |
| AN-004 | Supports (via rule-out) | Pre-treatment placebo on distance null — rules out freight/logistics shock. |
| AN-028 | Supports | Distance β = +14.25 km full → −0.28 (null) in SME (≤49); stays null in Micro (≤9). Geographic-catchment effect is entirely a non-SME composition channel. |
Open tests¶
Decompose by firm-RAIS-employment size¶
A clean test of the SME-composition reading would split the distance distribution by firm employment quartile (from the AN-009 RAIS validation). Larger firms should drive the post-2018 compression on high-value items; small firms should be unaffected. Not yet run.
Item-class heterogeneity¶
The high-value/low-value split is a coarse value cut. A finer cut by PADRAOdesc (item subcategory) would isolate where geographic competition matters most — e.g., medical equipment vs disposables vs reagents within group 65.