# UPGRADE_REPORT — Scholarly upgrade of the paper (2026-08-05)

Scope executed: Phase 1 (PAPER_REVIEW.md), Phase 2 (literature expansion, all references verified via OpenAlex), Phase 3 (section rewrites), Phase 4 (this report). No results, data, figures, or tables were changed. Paper version bumped 1.0 → 1.1. Compiled: paper/main.pdf, 61 pages (was 52), 0 errors, 0 undefined references, 69/69 bibliography entries cited.

Parameters chosen for the bracketed placeholders: field = housing / real estate economics (journal-article standard, e.g., Journal of Housing Economics / Real Estate Economics); reference target ≈ 60–65 (final: 69, from 37); length target ≈ +25–35 % prose (achieved: intro ×2.2, literature ×4, discussion ×2.7, conclusion ×2.5 in source lines; total document +9 pages ≈ +17 % because tables and figures are unchanged).

# 1. New references added (32 net: 33 added, 1 removed) — with justification

All entries verified one-by-one against OpenAlex (exact title, authors, venue, volume/pages, DOI). None were added from memory.

# Hedonic foundations & identification (7)

# Spatial econometrics (4)

# Quantile regression (3)

# Listing prices & seller behavior (3)

# Capitalization of local public goods (3)

# ML in economics & valuation (6)

# Explainability (2)

# Spatial validation (4)

# Climate risk (2)

# 2. Corrections to existing entries (verified against OpenAlex)

• bourassa2019machine → bourassa2010predicting : year was wrong (2019 → 2010), pages 139–159 → 139–160, DOI added.
• meyer2019importance → meyer2021predicting : year was wrong (2019 → 2021), DOI added.
• mak2010quantile : confirmed; author initials completed; DOI added.

# 3. ⚠ Fabricated reference found and removed

chen2020housing — “Chen, Hu & Lin (2020), Housing price prediction using machine learning: A systematic review, Expert Systems with Applications, 145:113142” — does not exist. No such work in OpenAlex; neither candidate DOI resolves; no ESWA article with that article number matches. This is precisely the citation-hallucination pattern the verification pass was designed to catch. Replaced in the text by real literature (park2015using for ML housing prediction; bishop2020best/rosen1974hedonic for the SHAP-is-not-WTP claim).

# 4. Sections expanded and how

# 5. Claims flagged for your verification

1. Black (1999) magnitude — ✅ verified post-hoc against the QJE abstract: “parents pay 2.5% more for a 5% increase in test scores, about half the naive hedonic estimate.” Text corrected from “approximately 2%” to 2.5%.
2. Campbell, Giglio & Pathak (2011) magnitude — ✅ verified against the AER abstract: average foreclosure discount of 27% on Massachusetts transactions. Text unchanged (accurate).
3. “Neighborhood scores carry market-specific meaning” conjecture — ✅ now partially grounded: Duncan et al. (2011, IJERPH 8(11), 4160–4179, DOI 10.3390/ijerph8114160; verified via PubMed/MDPI) validate Walk Score across four US metropolitan areas and find its correspondence with walkability constructs varies by metro. Cited in the discussion; the mechanism remains flagged as “plausible rather than established” since we do not test it.
4. Positioning sentence — ✅ validated by the author (2026-08-05): “To our knowledge, this framing has not previously been brought to bear on national-scale hedonic housing models” is kept as written.
5. Waltl citation year — ✅ decided (author, 2026-08-05): cite as 2019, the print issue of Real Estate Economics 47(3), 723–756. (OpenAlex records the online-first year 2016.)

# 6. Literature potentially in tension with the paper's findings

• Wadoux et al. (2021) argue spatial cross-validation is pessimistically biased when the estimand is map accuracy over a sampled region. Rather than ignoring it, the paper now engages it in three places (literature, methodology, limitations) and confines its own claims to the extrapolation setting. This is the most important "contradicting" reference; the engagement strengthens the argument but review it.
• Murfin & Spiegel (2020) find limited sea-level-rise capitalization, in tension with Bernstein et al. (2019); both are cited to avoid one-sided support for the "climate variables matter" limitation.
• Bayer, Ferreira & McMillan (2007) imply naive cross-sectional school coefficients confound neighbor characteristics — this supports the paper's caution but contradicts any residual temptation to interpret the school coefficient; the text now explicitly disclaims that interpretation.

# 7. Build status & typographic fixes

latexmk clean build: 60 pages, 0 errors, 0 undefined references/citations, 0 overfull vboxes, 70/70 entries cited, hyperlinks resolving. Version 1.1; title-page date frozen to August 5, 2026 for submission.

Two pre-existing typographic defects were found and fixed during the final QA pass: the main OLS table (Table 5) and the appendix full-OLS landscape table were taller than the page and were being clipped at the bottom (Panel F, fit statistics, and table notes were cut off in the rendered PDF). Table 5 is now split across two pages via \ContinuedFloat; the appendix table was compacted (scriptsize, tighter row spacing). No values were changed.