STATUS — UQO WP36 (Used-car market dynamics, Auto-Ka)
Updated: 2026-11-01 (session 1, end)
Current phase
COMPLETE (v1.0, 2026-11-01): all phases 0–9 executed; final end-to-end rerun (make data models tables figures, E-017) after the E-016 fix; manuscript 44 pp. compiled; pushed to spbgit wp36_uqo.git. Next: author's human read-through; SSRN deposit; extension with a longer panel and registration transfers.
Completed
- Phase 0: audit (
reports/data_access_audit.md), snapshot (SHA-256 verified), governance, environment; no external model API used. - Phase 1:
research/questions.md,research/analysis_plan.md(preregistered 2026-11-01; amendments D-010, D-011, D-014, D-015). - Phase 2: 172 verified references (
literature/), novelty memo, search protocol; external benchmarks inventory (no weekly Quebec used-car benchmark exists). - Phase 3: pipeline 01–03 (ingest → clean → risk set), exclusion log, sample tables, 14 tests, synthetic fixture.
- Phase 4: Models A (hedonic), B/C (hazards), D (competition), E (regions), F (prediction).
- Phase 5: AMLI (
results/amli/), weights, dealer bootstrap, held-out validation. - Phase 7: Figures 1–10, Tables 1–8 (+ tests and region tables).
- Phase 8:
paper/main.pdfv1.0 (43 pages, 0 undefined citations, 1 negligible overfull box). - Phase 9: STATUS, DECISIONS (D-001…D-015), EXPERIMENTS (E-000…E-015), CLAIMS_LEDGER (39 claims), REPRODUCE, Makefile, README.
- Phase 6: robustness suite (
results/robustness/, Table 9, Holm) — three non-robust results reported explicitly (H4 with make-model FE / sampled portals; exit residual under S2L; AMLI validation under rank/two-week). - E-016 bug (first_cut NULL) found by the robustness audit, fixed, full rerun; Model B rewritten.
- Final rerun + recompile + commit + push.
Results summary
H1 supported; H2 supported (listing-age profile; relative price and price vs competitors raise the markdown hazard; competitor density unstable); H3 same-day co-movement only (association), placebo null; H4 small premia (< 4 %), joint p = 0.53 with trim FE but 0.004 with make-model FE; H5 relative-price → exit not supported, exit validated across sources (94 % within 7 d, ≥ 10 % relistings); H6 PCA 56 %, held-out slope 0.97 pp [0.69, 1.24], no external test possible.
Blockers / known limits
- No transaction prices; exit ≠ sale; 81-day window; H3 descriptive; AMLI internal validation only.