AN-027: Within-Group-65 split — CMED-regulated medications vs other supplies¶
Intuition (plain-language)
Could the price rise just be pharmaceutical inflation, since CMED-regulated drugs sit inside Group 65? Split Group 65 into regulated medications and other medical supplies: the price effect appears in both (about −17% and −10%). It is competition, not a drug-price-inflation confound.
Question¶
Group 65 (medical/hospital supplies) is heterogeneous. The CMED-regulated medication sub-class (class 6531) is subject to pharmaceutical price regulation, which could generate its own differential inflation pattern unrelated to the SME-only policy. If the headline DiDiR price effect were driven by CMED-side inflation rather than by competition, the effect should be concentrated in Panel A (medications) and small or absent in Panel B (other supplies). The split tests this directly.
Design¶
- Sample: BEC completed items, 18-month window. Same DiDiR as AN-001 but with the treated group split:
- Panel A: \(g65_{med}\) = group 65 restricted to class 6531 (medications, subject to CMED pharmaceutical price regulation).
- Panel B: \(g65_{other}\) = group 65 restricted to non-6531 medical supplies (hospital equipment, dental supplies, etc.).
- Comparison: all non-group-65 items.
- Variation: same DiDiR; the sub-treatment indicator picks up only one panel at a time.
- Specification: item FE; PBU FE in alternating columns; item-clustered SE.
- Outcomes: log prices, log firms (per panel).
Results¶
Within-Group-65 panels (table tab_within_g65.tex):
| Panel | Outcome | Base coef | PBU-FE coef | N |
|---|---|---|---|---|
| A: Medications (6531) | Log prices | −0.166*** (0.017) | −0.174*** (0.017) | 573,080 |
| A: Medications (6531) | Log firms | +0.152*** (0.010) | +0.161*** (0.010) | 678,547 |
| B: Other supplies (non-6531) | Log prices | −0.099*** (0.009) | −0.097*** (0.009) | 593,920 |
| B: Other supplies (non-6531) | Log firms | +0.044*** (0.006) | +0.050*** (0.006) | 705,616 |
Output: output/tables/tab_within_g65.tex.
Interpretation¶
The price effect appears in BOTH panels. Medications (CMED-regulated) show β = −0.174 (PBU FE); other medical supplies show β = −0.097 (PBU FE). Both are highly significant. The pharmaceutical-inflation alternative is rejected: if differential CMED inflation drove the headline, the non-6531 panel should show no effect; instead it shows a ~10% price effect, half the magnitude but the same direction and significance.
The medication panel is larger. β = −0.174 in 6531 vs −0.097 in non-6531. This is consistent with the competition channel being stronger in the medication cell where bidder pools are likely deeper (many pharma suppliers nationally vs more-specialized hospital equipment suppliers). It is not consistent with CMED inflation: CMED regulation tends to cap pharmaceutical prices, not let them rise — if anything, CMED would attenuate the medication panel's effect, not amplify it.
Firm-count direction. Both panels show positive log-firm effects: +0.16 in medications, +0.05 in non-6531. The medication market is more amenable to non-SME entry under open competition (3× the firm-count gain of non-6531).
Interpretation for the headline. The DiDiR price effect is not mechanically driven by pharmaceutical inflation. The headline operates on both medication and non-medication subcells. The medication panel weighs more heavily in the pooled headline because of its larger sample, but the effect is present and significant in the non-medication cell that pharmaceutical inflation cannot explain.
Confidence: yellow. The within-group split is a clean falsification test against CMED-inflation. The yellow caveat is that other cell-specific shocks could in principle differ between medication and non-medication panels (e.g., regulatory changes specific to hospital equipment); the test does not rule out all class-specific confounds.
Follow-ups¶
- Item-class heterogeneity beyond the 6531 split: further split by 4-digit CADMAT class to expose where the price effect is concentrated within Panel B.
- CMED price-cap alignment: a direct test would compare the medication-cell price effect to the gap between CMED ceiling prices and observed BEC prices. If the BEC-vs-CMED gap shrinks post-policy in Panel A, that confirms the competition channel is operating against the CMED ceiling.