AN-003: DiDiR distance effect¶
Intuition (plain-language)
Larger non-SME firms can ship farther, so opening the auction pulls the winning supplier about 12 km farther from the buyer — but only on high-value items, where a large order covers the extra freight. On small orders distance is flat. Geography is the visible signature of larger firms re-entering the pool.
Reduced-form descriptive (not in the v8 manuscript)
Distance is a v1–v4 reduced-form outcome (scripts/02_analysis.R →
tab_distance.tex). The v8 manuscript does not carry distance: it is
a site-level descriptive that documents the geographic signature of
larger non-SME firms re-entering the pool under open competition — the
visible counterpart to the first-stage participation facts
(AN-002, Table 1). The canonical policy
result is the structural decomposition (AN-010)
and welfare arithmetic (AN-011). See the
finding Distant suppliers enter.
Question¶
Does open competition draw geographically distant non-SME suppliers into the BEC procurement pool, widening the average buyer-winner distance? And does this geographic-catchment channel vary with item value as transport-cost amortization predicts?
Design¶
- Sample: completed items with non-missing buyer and winner CEP geocodes; high-value/low-value split at the within-item value median.
- Variation: DiDiR as in AN-001.
- Specification: outcome is buyer-winner distance in km; item-FE, PBU-FE alternating, item-clustered SE.
- Outcomes: distance (km), by value class.
- Identification threats: freight/logistics shock specific to group 65; ruled out by placebo (AN-004).
Results¶
| Sample | Distance β (km) | SE | p |
|---|---|---|---|
| High-value items (above median) | +12.0849 | (2.9018) | <0.01 |
| Low-value items (below median) | +2.8312 | (2.2944) | >0.10 |
| Pooled (18m, base specification) | +10.78 | (2.232) | <0.01 |
| Pooled (18m, PBU FE) | +10.80 | (2.233) | <0.01 |
Output: output/tables/tab_distance.tex;
heterogeneity in
output/tables/tab_heterog_value.tex.
Interpretation¶
The pooled DiDiR finds a meaningful distance widening (~11 km, p<0.01) that decomposes cleanly along the value margin: +12.1 km on high-value items, null on low-value items. The asymmetry tracks the transport-cost amortization story directly — non-SMEs span a wider geographic radius and can profitably bid distant only when the contract value is large enough to absorb logistics. The null on low-value items is consistent with the prediction but is not diagnostic of an alternative reading.
Confidence: yellow. Distance is a coarse geographic proxy (single CEP). The RAIS validation in AN-009 provides a richer firm-footprint measure but does not enter the distance specification. The pooled estimate is stable across windows; the value heterogeneity is the most-informative cut.
Follow-ups¶
- Firm-size cross-cut: split the distance distribution by RAIS-employment quartile. Larger firms should drive the high-value compression; small firms should be unaffected.
- Item-class heterogeneity within group 65: distance compression by PADRAOdesc subcategory (medical equipment vs disposables vs reagents) would isolate where geographic competition matters most.
- Full operational footprint: replace single-CEP distance with the firm's RAIS establishment locations to bound the CEP-mismeasurement concern.