Skip to content

The protected SME pool responds but does not replace lost discipline

Intuition (plain-language)

This is not a story where SMEs simply fail to show up. After the set-aside returns, the number of SME bidders per auction roughly doubles — the protected pool does respond. The problem is the quality of competition, not the headcount: even with twice as many SMEs bidding, the winning price stays above the open-auction benchmark. More of the protected type cannot replicate the discipline of the type you excluded.

🟡 After the March 2018 SME-only cutoff, the number of SME bidders per auction roughly doubles in switched group 65 — the protected pool does respond. But the realized post-policy price \(S_3\) remains above the open benchmark \(S_1\) (AN-010). The set-aside cannot be rationalized as a mechanical non-entry story: SMEs show up; the issue is that the protected pool's cost distribution cannot recreate the competitive discipline supplied by the excluded non-SMEs.

The structural decomposition splits the protected-pool offset \(S_3 - S_2\) into a participation channel (more SMEs) and a composition-of-active-SMEs channel (the active SME cost distribution may itself shift under the policy — selection, sourcing or technology changes, product-mix changes). The offset is sizable but does not flip the net price effect from positive to negative — the net \(S_3 - S_1\) remains positive in both non-pharma and pharma cells.

Caveat. The interpretation that the SME pool cannot recreate the discipline rests on the maintained IPV-clock reading (H:ipv-clock-admissible): the post-policy SME drop-outs are read as cost observations. The threat that drop-outs are strategically suppressed (more cover-bidding among the smaller eligible pool) is now bounded by the collusion screens (AN-015): Conley close-pair shares are flat or falling post-cutoff and Bajari-Ye T1 ratios fall in both classes — the differential-coordination shock that would explain the protected-pool offset away is rejected. Residual baseline clustering (real > null in every cell) remains a limitation, hence 🟡.

Sources.

  • Own analysis: AN-010 (decomposition of \(S_3 - S_2\)); AN-017 (strict-invariance isolating the participation channel from the composition channel).
  • Reports: none direct.
  • News anchors: none direct.
  • Cross-refs: H:protected-pool-responds; H:exclusion-dominates; paper §4 (The protected pool responds).
  • Validation: v7-jpube-tight/scripts/44_entry_pool.R for the bidder-count tabulation; 45_bne_simulation.R for the \(S_3\) simulation.