AN-002: DiDiR firm and bid count effects¶
Intuition (plain-language)
Reopening an auction to non-SMEs mechanically adds bidders; the question is how many. Switched Group 65 gains roughly 18% more bidding firms right after the change, fading to about 10% as the control groups themselves finish rolling into SME-only. The fade reflects the controls moving, not the treatment effect vanishing.
Reduced-form motivation layer
The numbers below are from the v1–v4 reduced-form DiDiR pipeline
(scripts/02_analysis.R + companions), which the v8 manuscript
carries as motivation in §1 but does not headline. The canonical
v8 result is the structural counterfactual decomposition — see
AN-010 (decomposition) and
AN-011 (welfare arithmetic).
Question¶
Does opening the auction to non-SMEs raise the number of participating firms and the number of valid bids in switched group 65? The reduced-form DiDiR identifies the open-vs-SME-only difference in bidder participation.
Design¶
- Sample: all items (not restricted to completed) in the BEC parquet cache for the 6/12/18-month windows around March 2018.
- Variation: same DiDiR as AN-001.
- Specification: same DiDiR equation; outcomes are log firms and log valid bids.
- Outcomes: log number of bidder firms; log number of valid bids.
- Identification threats: secular trend in control groups' rollout of SME restrictions across other product groups — see Interpretation below.
Results¶
| Window | Base log firms | PBU-FE log firms | N |
|---|---|---|---|
| 6m | +0.1776*** (0.0079) | +0.1821*** (0.0081) | 261,450 |
| 12m | +0.1495*** (0.0062) | +0.1540*** (0.0063) | 524,745 |
| 18m | +0.0926*** (0.0059) | +0.1004*** (0.0060) | 773,263 |
Item FE; PBU FE in alternating columns; item-clustered SE. *** p<0.01.
Bid-count effects (table tab_validbids.tex) move in the same direction
with similar magnitude attenuation across windows.
Output: output/tables/tab_participants.tex,
output/tables/tab_validbids.tex.
Maps to the v8 first stage (Table 1)
This aggregate count is the net of two by-type moves that the v8 structural first stage (Table 1) reports separately. In the non-pharmaceutical structural sample, SME bidders rise from 0.94 to 1.87 per auction while non-SMEs fall from 2.68 to 1.50 (pharma: SME 0.55 → 1.22, non-SME 2.61 → 1.66). Open competition therefore fields more total bidders than the SME-only counterfactual — consistent with the positive reduced-form coefficient here. The reduced-form DiD cannot separate SME entry from non-SME removal; the structural decomposition does (AN-010).
Interpretation¶
The 6-month estimate implies ~22% more firms in switched group 65 under open competition; the 18-month estimate ~10%. The attenuation does not mean the policy effect weakened over time — it reflects the control groups' gradual SME restriction rollout. As the controls converged to SME-only-everywhere, the cross-group DiDiR contrast shrank mechanically.
The mechanical removal of non-SMEs from the eligible pool sets a floor on the firm-count effect: the post-period bidder count cannot fall below the pre-period count by more than the count of non-SMEs that participated in the pre-period. The 6-month estimate is probably the cleanest reading of the policy effect itself; the 18-month is the conservative cross-window benchmark.
Confidence: yellow. The direction is stable; the magnitudes are window-dependent in a known and interpretable way. The pre-treatment placebo on firm counts is significant (AN-004) but reflects the same secular trend rather than a group-65-specific confound.
Follow-ups¶
- Type-decomposition of the firm-count response: how much is non-SME removal versus SME additional entry? The structural decomposition (AN-010) separates them; the DiDiR cannot.
- Intensive vs extensive bid margin: are bid-count gains driven by more firms registering or by within-firm re-bidding? Pregão allows multiple bids per firm; a within-firm bid-count decomposition would expose strategic re-bidding.
- Item-class heterogeneity: the firm-count response should be larger on items where non-SMEs were over-represented pre-policy.