SPB Git

spb/wp9_uqo Public

UQO Working Paper No. 9 — A grand hedonic model of the Canadian housing market: decomposing structure and location value.

researcheconomicshousinghedoniccanadauqo
TeX 60.1% Python 39.8%

Scholarly upgrade v1.2: +32 verified references, new Discussion section

Phase 1: critical assessment (PAPER_REVIEW.md). Phase 2: 32 new references
across 11 identified gaps (urban spatial structure, land share, supply
constraints, price indices, submarkets, spatial-dependence empirics,
capitalization theory, list-price microstructure, AVM/mass appraisal,
Canadian evidence, HDFE methodology) — every entry verified against
Crossref DOI records or publisher sources; references.bib 32 → 64.
Phase 3: introduction deepened with numbered contributions; related work
restructured into 7 thematic subsections each ending with an explicit gap
statement; methodology choices grounded in citations; new Discussion
section (magnitudes vs. literature, points of tension, implications,
limitations, future work); conclusion strengthened. Paper 26 → 34 pages.
Phase 4: UPGRADE_REPORT.md with per-reference justifications and flagged
claims. No result, number, figure or table changed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
simon-pierre boucher committed 5 days ago (Aug 5, 2026) parent 6241037

Showing 13 changed files with +983 and −163

added PAPER_REVIEW.md +120 −0
@@ -0,0 +1,120 @@
1 +# PAPER_REVIEW — Critical assessment before the scholarly upgrade (2026-08-05)
2 +
3 +Scope: `paper/` (v1.1, 26 pp., 32 references). Assessment only — no results questioned.
4 +
5 +## 1. Core contribution: is it clearly stated?
6 +
7 +The empirical contribution is clear and genuinely stated in one sentence: *a national-scale
8 +decomposition of Canadian house prices into structure vs. location using 1,153 absorbed FSA
9 +fixed effects on 140,931 listings*. Three supporting claims (elasticity ≈ 0.55, bathrooms
10 +≈ +11%, bedrooms ≈ 0; OOS R² 0.764; premia span ×9) are quantified and reproducible.
11 +
12 +**Weaknesses in the contribution statement:**
13 +- The introduction motivates in three bullets ("first-order question", "granular data",
14 + "transparent benchmark") but never states *what is new relative to specific prior work*.
15 + No named paper is identified as the closest antecedent, so the gap is asserted, not shown.
16 +- "Among the most geographically comprehensive single-equation hedonic exercises for
17 + Canada" is unverifiable as written because the Canadian empirical hedonic record
18 + (Des Rosiers/Thériault's Quebec program, Toronto and Vancouver studies) is never cited.
19 +- The headline result — location ≈ 30 pp of R² — is never connected to the macro
20 + literature quantifying the land/location share of housing value (Davis–Heathcote;
21 + Knoll–Schularick–Steger), which is the natural external benchmark for its magnitude.
22 +
23 +## 2. Literature review: gaps by subfield
24 +
25 +Current coverage (good): Rosen framework and second-stage identification; semi-log
26 +functional form; spatial econometrics vs. fixed effects debate; three amenity
27 +capitalization classics; brief AVM motivation.
28 +
29 +**Missing subfields (each currently has zero citations):**
30 +
31 +1. **Urban spatial structure / monocentric city.** The paper estimates an urban price
32 + gradient (§ Results 5.6) with no reference to Alonso–Muth–Mills, nor to modern
33 + quantitative treatments (Ahlfeldt–Redding–Sturm–Wolf; Duranton–Puga). The gradient
34 + section cites only Combes et al. (2015).
35 +2. **Land value and the location share.** Davis & Heathcote (2007), Knoll, Schularick &
36 + Steger (2017), Albouy's work on urban land values. Directly benchmark the 30-pp claim.
37 +3. **House-price indices.** The hedonic/repeat-sales index literature (Bailey–Muth–Nourse;
38 + Case–Shiller; Hill's survey) — the paper claims relevance for "house-price indices" in
39 + the first paragraph without citing any index paper.
40 +4. **Housing submarkets.** Goodman & Thibodeau; Bourassa et al. on submarket definitions —
41 + the FSA-fixed-effect design is a submarket argument and should engage this strand.
42 +5. **Spatial dependence in hedonics (empirics).** Dubin (1988); Basu & Thibodeau (1998);
43 + and the Canadian spatio-temporal work of Dubé & Legros. Moran's I is used but only
44 + Moran (1950) is cited.
45 +6. **Capitalization theory.** Tiebout (1956), Oates (1969), Roback (1982) — the FSA
46 + premium is interpreted as capitalized amenities with no reference to why amenities
47 + capitalize.
48 +7. **List price vs. transaction price.** The central data caveat (list prices) has no
49 + citation — Genesove & Mayer (2001) on seller behaviour, Han & Strange (2015) on the
50 + listing/search microstructure.
51 +8. **AVM / ML valuation empirics.** One citation (Mullainathan & Spiess 2017). Missing the
52 + applied mass-appraisal accuracy literature and the ML-vs-hedonic horse-race studies the
53 + OOS section implicitly competes with.
54 +9. **Canadian housing.** One CMHC report. Missing the Canadian hedonic tradition
55 + (Des Rosiers and coauthors on Quebec; Toronto/Vancouver-specific studies) and the
56 + supply-side explanations for Vancouver/Toronto price levels (Saiz-style constraints;
57 + superstar-city dynamics of Gyourko–Mayer–Sinai).
58 +10. **High-dimensional FE methodology.** Correia (2017) is cited, but the intellectual
59 + origin (Abowd–Kramarz–Margolis) and the standard computational references
60 + (Guimarães & Portugal) are not.
61 +11. **Sample selection / trimming practice.** The 1% trim and its robustness variants have
62 + no methodological anchor.
63 +
64 +## 3. Weak argumentation / unsupported claims
65 +
66 +- "Competitive with commercial automated valuation models" (§ Robustness, Conclusion) —
67 + no AVM accuracy benchmark is cited; needs at least one accuracy-standard reference
68 + (e.g., IAAO mass-appraisal standards or a published Zestimate-type accuracy figure).
69 +- "Credible national estimates for Canada are scarce" — unsupported; requires citing what
70 + *does* exist to establish scarcity.
71 +- Half-bath negative coefficient interpretation ("older or more compartmentalised floor
72 + plans") is speculative and unflagged — should be presented as conjecture or supported.
73 +- "This is the housing counterpart of the within estimator" — fine, but the AKM/FE lineage
74 + should be cited where the claim is made.
75 +- The bedrooms≈0 result is called "a textbook hedonic finding" without a supporting
76 + citation (Sirmans et al. tabulate mixed signs — an easy, honest anchor).
77 +- Urban gradient "classic urban theory predicts" — no Alonso/Muth/Mills citation.
78 +
79 +## 4. Underdeveloped sections
80 +
81 +- **Introduction** (~1 page): thin on stakes (housing = largest household asset class;
82 + assessment/taxation machinery; index construction) and has no explicit numbered
83 + contribution list.
84 +- **Related work** (~1.5 pages): organized adequately by theme but shallow within themes
85 + (2–4 citations each) and never says *where the gap is* at the end of each strand.
86 +- **Discussion: absent.** Results flow directly into robustness and a short conclusion.
87 + There is no section interpreting magnitudes against prior literature (is 0.55 a typical
88 + elasticity? is an 11% bathroom premium high? is ×9 premia dispersion large vs. the US?),
89 + no engagement with findings that could contradict (e.g., submarket papers arguing
90 + *coefficients*, not just intercepts, vary by location — which the paper's own
91 + heterogeneity section partially confirms), and no consolidated Limitations & Future Work
92 + (currently squeezed into two conclusion paragraphs).
93 +- **Methodology**: several choices uncited (trimming, FSA pooling threshold, kNN Moran
94 + implementation choice, quantile regression estimator details).
95 +
96 +## 5. Positioning: current vs. ideal
97 +
98 +**Current:** a well-executed measurement paper that reads as a technical report — it
99 +presents itself mainly as "we ran a big transparent regression."
100 +
101 +**Ideal:** a contribution at the intersection of (i) the classic structure-vs-location
102 +decomposition, (ii) the land-share macro literature, and (iii) the AVM benchmarking
103 +debate — i.e., "the neighbourhood is the single largest priced component of Canadian
104 +housing; we quantify it with a design that is transparent enough to audit and accurate
105 +enough to rival black-box AVMs." The framing should name its closest antecedents (US
106 +land-share estimates; submarket hedonics; ML-valuation horse races) and state exactly
107 +what is new: national scope for Canada, listing-level granularity, absorbed-FE scale, and
108 +a validated accuracy claim.
109 +
110 +## 6. Consequences for the upgrade (implemented in Phases 2–3)
111 +
112 +1. Add ~28 verified references across the 11 gaps above (target ≈ 60 total).
113 +2. Restructure Related Work into thematic subsections, each ending with the gap.
114 +3. Add a proper **Discussion** section: magnitudes vs. literature, agreements/
115 + disagreements, implications (assessment, taxation, indices, AVM), limitations & future
116 + work.
117 +4. Deepen the introduction (stakes, numbered contributions, roadmap); strengthen the
118 + conclusion without overselling.
119 +5. Ground every methodological choice in a citation; keep every number, figure and table
120 + untouched.
added UPGRADE_REPORT.md +122 −0
@@ -0,0 +1,122 @@
1 +# UPGRADE_REPORT — Scholarly upgrade of WP9 (v1.1 → v1.2, 2026-08-05)
2 +
3 +Scope: literature and framing only. **No result, number, figure or table changed.**
4 +Paper: 26 → **34 pages** (+31%); references: 32 → **64** (63 cited); new **Discussion**
5 +section (§7). Critical pre-assessment in `PAPER_REVIEW.md`. Builds cleanly
6 +(pdflatex + bibtex, zero errors/undefined citations).
7 +
8 +## 1. New references (32), each verified before entry
9 +
10 +Verification method: direct Crossref DOI lookup for 30 entries (metadata in the .bib
11 +copied from the Crossref record); Mills (1967) and the two books confirmed via JSTOR
12 +issue listing / publisher records ([Mills 1967 AER 57(2):197–210](https://www.jstor.org/stable/i331577),
13 +[Alonso 1964](https://www.cambridge.org/core/journals/journal-of-the-history-of-economic-thought/article/abs/william-alonso-richard-muth-resources-for-the-future-and-the-founding-of-urban-economics/A1F19C2F841A0D0AA33D15F46918CFEA),
14 +[Muth 1969](https://www.cambridge.org/core/journals/recherches-economiques-de-louvain-louvain-economic-review/article/abs/rf-muth-cities-and-housing-the-spatial-pattern-of-urban-residential-land-use-chicago-and-london-the-university-of-chicago-press-1969-xxii-p-355-p-90/0AA8AA64AD2123A62CED306A63423327),
15 +[Case–Shiller 1989 AER 79(1):125–137](https://econpapers.repec.org/RePEc:aea:aecrev:v:79:y:1989:i:1:p:125-37)).
16 +The OpenAlex MCP connector was budget-limited, so Crossref's REST API was used instead.
17 +
18 +**Urban spatial structure (motivates the gradient section, previously uncited):**
19 +- `alonso1964location` — founding monocentric land-rent model.
20 +- `muth1969cities` — housing version of the monocentric model.
21 +- `mills1967aggregative` — the third leg of the Alonso–Muth–Mills tradition.
22 +- `ahlfeldt2015density` — modern quantitative estimation of density/amenity gradients (Berlin Wall).
23 +
24 +**Land value & the location share (external benchmark for the 30-pp headline):**
25 +- `davis2007price` — U.S. residential land share ≈ 46%; direct macro counterpart.
26 +- `knoll2017no` — land explains ~80% of global house-price growth since 1950.
27 +- `albouy2016cities` — cross-city dispersion of land rents & amenity values.
28 +- `cheshire1995price` — early land-price/amenity decomposition.
29 +
30 +**Supply constraints / superstar cities (explains WHY premia disperse ×9):**
31 +- `saiz2010geographic` — geographic supply constraints; Vancouver-type geography.
32 +- `gyourko2013superstar` — superstar-city price dynamics; frames Vancouver/Toronto.
33 +
34 +**House-price indices (paper claimed index relevance with zero index citations):**
35 +- `bailey1963regression` — repeat-sales method origin.
36 +- `case1989efficiency` — the index tradition + efficiency questions (also cited in Limitations).
37 +- `hill2013hedonic` — authoritative survey of hedonic price indexes.
38 +
39 +**Submarkets (the FSA design is a submarket argument):**
40 +- `goodman1998housing` — market segmentation tests.
41 +- `bourassa2003submarkets` — submarkets improve prediction; simple geography suffices.
42 +
43 +**Spatial dependence, empirics (Moran's I had one 1950 citation):**
44 +- `dubin1988estimation` — spatially autocorrelated errors distort hedonic inference.
45 +- `basu1998analysis` — canonical measurement of residual spatial autocorrelation.
46 +- `dube2013spatiotemporal` — Canadian spatio-temporal weights (Dubé & Legros).
47 +
48 +**Capitalization theory (why a neighbourhood intercept is 'value' at all):**
49 +- `tiebout1956pure` — sorting on local public goods.
50 +- `oates1969effects` — first capitalization measurement.
51 +- `roback1982wages` — spatial-equilibrium amenity capitalization.
52 +
53 +**Hedonic identification:**
54 +- `epple1987hedonic` — second-stage simultaneity; grounds our first-stage-only stance.
55 +
56 +**List prices & microstructure (the paper's central data caveat, previously uncited):**
57 +- `genesove2001loss` — sellers set list prices strategically (loss aversion).
58 +- `han2015microstructure` — search/bargaining microstructure (Handbook chapter).
59 +- `kiel2008location` — the '3L' structure/location decomposition framework.
60 +
61 +**AVM / mass appraisal (grounds 'competitive with commercial AVMs'):**
62 +- `clapp2003semiparametric` — location-value surface for automated valuation.
63 +- `mccluskey2013prediction` — accuracy comparison across mass-appraisal model classes.
64 +
65 +**Canadian evidence (was: one CMHC report):**
66 +- `desrosiers1996shopping`, `desrosiers2002landscaping` — the Quebec City hedonic program.
67 +- `haider2000effects` — spatial hedonics for Toronto.
68 +
69 +**High-dimensional FE methodology:**
70 +- `abowd1999high` — the AKM origin of the estimator lineage.
71 +- `guimaraes2010simple` — the standard computational reference.
72 +
73 +## 2. Sections expanded and how
74 +
75 +| Section | Before | After |
76 +|---|---|---|
77 +| Introduction | 5 ¶, assertion-style gap | 6 ¶: stakes ¶ (housing wealth, assessment, land share), positioning vs. named literatures, **numbered 3-part contribution**, roadmap incl. §7 |
78 +| Related Literature | 5 loose subsections, 32 refs | **7 thematic subsections**, each ending with an explicit *Gap:* statement; ~55 refs engaged |
79 +| Data | unchanged | one grounding citation for the 1% trim |
80 +| Methodology | unchanged equations | AKM lineage + computational refs added to the absorbing-LS paragraph |
81 +| Results | **numbers/figures untouched** | 3 citation anchors only (bedrooms "textbook finding"; monocentric theory for the gradient; no text changes otherwise) |
82 +| Robustness | **numbers untouched** | 2 citation anchors (spatial-dependence empirics; AVM accuracy claim now sourced) |
83 +| **Discussion (NEW §7)** | — | 6 subsections: magnitudes vs. literature (land share, elasticities, Moran, quantiles, AVM accuracy); **points of tension**; implications (assessment, indices, AVM, policy); Limitations (5, consolidated); Future work |
84 +| Conclusion | 3 ¶ incl. limitations | limitations moved to §7; added an explicit "what the reading is" paragraph; no overselling (contribution restated with the same numbers) |
85 +
86 +## 3. Claims flagged for your verification
87 +
88 +1. **"First listing-level structure-vs-location decomposition spanning the entire
89 + Canadian market" (intro, contribution 1).** I softened it with "(to our knowledge)",
90 + and the Canadian-literature subsection now documents what exists (metropolitan-scale
91 + work). Please confirm you are comfortable with the novelty claim.
92 +2. **Bathroom premium "at the upper end of published estimates" (Discussion §7.1).**
93 + Directionally supported by Sirmans et al.'s tabulations, but I did not run a formal
94 + meta-comparison. Soften to "within the range" if you prefer.
95 +3. **AVM accuracy comparison.** The paper now cites McCluskey et al. (2013) and Clapp
96 + (2003) instead of claiming an unsourced comparison with "commercial" AVMs. If you
97 + have an IAAO-ratio-standard source you prefer, it slots in naturally there.
98 +4. **Half-bath negative coefficient interpretation** remains conjectural in Results
99 + (unchanged text). Consider adding "we conjecture" if a referee pushes.
100 +
101 +## 4. Literature that potentially contradicts the paper
102 +
103 +- **`bourassa2003submarkets` / `goodman1998housing`** — argue implicit prices (slopes),
104 + not just intercepts, vary across submarkets. The paper's own provincial heterogeneity
105 + (elasticity 0.49–0.66) partially *confirms* this, so the grand model's pooled slopes
106 + are an approximation. Now acknowledged explicitly in §7.2 ("Points of tension").
107 +- **`kuminoff2010which`** — hedonic estimates can be specification-fragile; addressed
108 + via the robustness battery but flagged in §7.2 as binding for any welfare use of the premia.
109 +- **`genesove2001loss` / `han2015microstructure`** — list prices embed strategic
110 + behaviour; directly qualifies the paper's dependent variable. Now the first limitation
111 + in §7.4 with citations.
112 +- **LOPO's negative Alberta R²** (paper's own result) — cuts against fully universal
113 + structural prices; now stated plainly in §7.2 rather than left implicit.
114 +- I found **no verified source contradicting the headline location-share result**; the
115 + closest quantitative benchmarks (Davis–Heathcote 46% land share; Knoll et al.) point
116 + the same way.
117 +
118 +## 5. Not changed
119 +
120 +All tables, all figures, all reported numbers, the abstract's quantitative claims, the
121 +data and methodology substance, and `results/reference/`. The uncited legacy bib entry
122 +`gabaix2016power` (orphaned in the original) was left in place.
modified paper/main.pdf +0 −0

Binary file not shown.

modified paper/main.tex +2 −1
@@ -66,7 +66,7 @@
66 66 \newcommand{\WPsubtitle}{Decomposing the Value of Structure and Location across
67 67 140{,}931 MLS Listings with High-Dimensional Neighbourhood Fixed Effects}
68 68 \newcommand{\WPdate}{May 2026}
69 −\newcommand{\WPversion}{1.1}
69 +\newcommand{\WPversion}{1.2}
70 70 \newcommand{\WPkeywords}{Hedonic pricing, Housing markets, Canada, Fixed effects,
71 71 Spatial heterogeneity, Automated valuation}
72 72 \newcommand{\WPjel}{R31, R21, C21, C55}
@@ -109,6 +109,7 @@ between the most and least expensive areas.}
109 109 \input{sections/methodology}
110 110 \input{sections/results}
111 111 \input{sections/robustness}
112 +\input{sections/discussion}
112 113 \input{sections/conclusion}
113 114
114 115 \newpage
modified paper/references.bib +341 −0
@@ -307,3 +307,344 @@
307 307 pages = {15--33},
308 308 year = {1998}
309 309 }
310 +
311 +% ============================================================================
312 +% Added in the v1.2 scholarly upgrade (2026-08-05).
313 +% Every entry verified against Crossref (DOI lookup) or publisher records.
314 +% ============================================================================
315 +
316 +% ---- Urban spatial structure / monocentric tradition ----
317 +@book{alonso1964location,
318 + author = {Alonso, William},
319 + title = {Location and Land Use: Toward a General Theory of Land Rent},
320 + publisher = {Harvard University Press},
321 + address = {Cambridge, MA},
322 + year = {1964}
323 +}
324 +@book{muth1969cities,
325 + author = {Muth, Richard F.},
326 + title = {Cities and Housing: The Spatial Pattern of Urban Residential Land Use},
327 + publisher = {University of Chicago Press},
328 + address = {Chicago},
329 + year = {1969}
330 +}
331 +@article{mills1967aggregative,
332 + author = {Mills, Edwin S.},
333 + title = {An Aggregative Model of Resource Allocation in a Metropolitan Area},
334 + journal = {American Economic Review},
335 + year = {1967},
336 + volume = {57},
337 + number = {2},
338 + pages = {197--210}
339 +}
340 +@article{ahlfeldt2015density,
341 + author = {Ahlfeldt, Gabriel M. and Redding, Stephen J. and Sturm, Daniel M. and Wolf, Nikolaus},
342 + title = {The Economics of Density: Evidence from the {B}erlin {W}all},
343 + journal = {Econometrica},
344 + year = {2015},
345 + volume = {83},
346 + number = {6},
347 + pages = {2127--2189},
348 + doi = {10.3982/ECTA10876}
349 +}
350 +
351 +% ---- Land value and the location share of housing ----
352 +@article{davis2007price,
353 + author = {Davis, Morris A. and Heathcote, Jonathan},
354 + title = {The Price and Quantity of Residential Land in the {U}nited {S}tates},
355 + journal = {Journal of Monetary Economics},
356 + year = {2007},
357 + volume = {54},
358 + number = {8},
359 + pages = {2595--2620},
360 + doi = {10.1016/j.jmoneco.2007.06.023}
361 +}
362 +@article{knoll2017no,
363 + author = {Knoll, Katharina and Schularick, Moritz and Steger, Thomas},
364 + title = {No Price Like Home: Global House Prices, 1870--2012},
365 + journal = {American Economic Review},
366 + year = {2017},
367 + volume = {107},
368 + number = {2},
369 + pages = {331--353},
370 + doi = {10.1257/aer.20150501}
371 +}
372 +@article{albouy2016cities,
373 + author = {Albouy, David},
374 + title = {What Are Cities Worth? {L}and Rents, Local Productivity, and the Total Value of Amenities},
375 + journal = {Review of Economics and Statistics},
376 + year = {2016},
377 + volume = {98},
378 + number = {3},
379 + pages = {477--487},
380 + doi = {10.1162/REST_a_00550}
381 +}
382 +@article{cheshire1995price,
383 + author = {Cheshire, Paul and Sheppard, Stephen},
384 + title = {On the Price of Land and the Value of Amenities},
385 + journal = {Economica},
386 + year = {1995},
387 + volume = {62},
388 + number = {246},
389 + pages = {247--267},
390 + doi = {10.2307/2554906}
391 +}
392 +
393 +% ---- Supply constraints and superstar cities ----
394 +@article{saiz2010geographic,
395 + author = {Saiz, Albert},
396 + title = {The Geographic Determinants of Housing Supply},
397 + journal = {Quarterly Journal of Economics},
398 + year = {2010},
399 + volume = {125},
400 + number = {3},
401 + pages = {1253--1296},
402 + doi = {10.1162/qjec.2010.125.3.1253}
403 +}
404 +@article{gyourko2013superstar,
405 + author = {Gyourko, Joseph and Mayer, Christopher and Sinai, Todd},
406 + title = {Superstar Cities},
407 + journal = {American Economic Journal: Economic Policy},
408 + year = {2013},
409 + volume = {5},
410 + number = {4},
411 + pages = {167--199},
412 + doi = {10.1257/pol.5.4.167}
413 +}
414 +
415 +% ---- House-price indices ----
416 +@article{bailey1963regression,
417 + author = {Bailey, Martin J. and Muth, Richard F. and Nourse, Hugh O.},
418 + title = {A Regression Method for Real Estate Price Index Construction},
419 + journal = {Journal of the American Statistical Association},
420 + year = {1963},
421 + volume = {58},
422 + number = {304},
423 + pages = {933--942},
424 + doi = {10.1080/01621459.1963.10480679}
425 +}
426 +@article{case1989efficiency,
427 + author = {Case, Karl E. and Shiller, Robert J.},
428 + title = {The Efficiency of the Market for Single-Family Homes},
429 + journal = {American Economic Review},
430 + year = {1989},
431 + volume = {79},
432 + number = {1},
433 + pages = {125--137}
434 +}
435 +@article{hill2013hedonic,
436 + author = {Hill, Robert J.},
437 + title = {Hedonic Price Indexes for Residential Housing: A Survey, Evaluation and Taxonomy},
438 + journal = {Journal of Economic Surveys},
439 + year = {2013},
440 + volume = {27},
441 + number = {5},
442 + pages = {879--914},
443 + doi = {10.1111/j.1467-6419.2012.00731.x}
444 +}
445 +
446 +% ---- Housing submarkets ----
447 +@article{goodman1998housing,
448 + author = {Goodman, Allen C. and Thibodeau, Thomas G.},
449 + title = {Housing Market Segmentation},
450 + journal = {Journal of Housing Economics},
451 + year = {1998},
452 + volume = {7},
453 + number = {2},
454 + pages = {121--143},
455 + doi = {10.1006/jhec.1998.0229}
456 +}
457 +@article{bourassa2003submarkets,
458 + author = {Bourassa, Steven C. and Hoesli, Martin and Peng, Vincent S.},
459 + title = {Do Housing Submarkets Really Matter?},
460 + journal = {Journal of Housing Economics},
461 + year = {2003},
462 + volume = {12},
463 + number = {1},
464 + pages = {12--28},
465 + doi = {10.1016/S1051-1377(03)00003-2}
466 +}
467 +
468 +% ---- Spatial dependence in hedonic empirics ----
469 +@article{dubin1988estimation,
470 + author = {Dubin, Robin A.},
471 + title = {Estimation of Regression Coefficients in the Presence of Spatially Autocorrelated Error Terms},
472 + journal = {Review of Economics and Statistics},
473 + year = {1988},
474 + volume = {70},
475 + number = {3},
476 + pages = {466--474},
477 + doi = {10.2307/1926785}
478 +}
479 +@article{basu1998analysis,
480 + author = {Basu, Sabyasachi and Thibodeau, Thomas G.},
481 + title = {Analysis of Spatial Autocorrelation in House Prices},
482 + journal = {Journal of Real Estate Finance and Economics},
483 + year = {1998},
484 + volume = {17},
485 + number = {1},
486 + pages = {61--85},
487 + doi = {10.1023/A:1007703229507}
488 +}
489 +@article{dube2013spatiotemporal,
490 + author = {Dub{\'e}, Jean and Legros, Diego},
491 + title = {A Spatio-temporal Measure of Spatial Dependence: An Example Using Real Estate Data},
492 + journal = {Papers in Regional Science},
493 + year = {2013},
494 + volume = {92},
495 + number = {1},
496 + pages = {19--30},
497 + doi = {10.1111/j.1435-5957.2011.00402.x}
498 +}
499 +
500 +% ---- Capitalization theory ----
501 +@article{tiebout1956pure,
502 + author = {Tiebout, Charles M.},
503 + title = {A Pure Theory of Local Expenditures},
504 + journal = {Journal of Political Economy},
505 + year = {1956},
506 + volume = {64},
507 + number = {5},
508 + pages = {416--424},
509 + doi = {10.1086/257839}
510 +}
511 +@article{oates1969effects,
512 + author = {Oates, Wallace E.},
513 + title = {The Effects of Property Taxes and Local Public Spending on Property Values: An Empirical Study of Tax Capitalization and the {T}iebout Hypothesis},
514 + journal = {Journal of Political Economy},
515 + year = {1969},
516 + volume = {77},
517 + number = {6},
518 + pages = {957--971},
519 + doi = {10.1086/259584}
520 +}
521 +@article{roback1982wages,
522 + author = {Roback, Jennifer},
523 + title = {Wages, Rents, and the Quality of Life},
524 + journal = {Journal of Political Economy},
525 + year = {1982},
526 + volume = {90},
527 + number = {6},
528 + pages = {1257--1278},
529 + doi = {10.1086/261120}
530 +}
531 +
532 +% ---- Hedonic identification (second stage) ----
533 +@article{epple1987hedonic,
534 + author = {Epple, Dennis},
535 + title = {Hedonic Prices and Implicit Markets: Estimating Demand and Supply Functions for Differentiated Products},
536 + journal = {Journal of Political Economy},
537 + year = {1987},
538 + volume = {95},
539 + number = {1},
540 + pages = {59--80},
541 + doi = {10.1086/261441}
542 +}
543 +
544 +% ---- List prices, search and housing microstructure ----
545 +@article{genesove2001loss,
546 + author = {Genesove, David and Mayer, Christopher},
547 + title = {Loss Aversion and Seller Behavior: Evidence from the Housing Market},
548 + journal = {Quarterly Journal of Economics},
549 + year = {2001},
550 + volume = {116},
551 + number = {4},
552 + pages = {1233--1260},
553 + doi = {10.1162/003355301753265561}
554 +}
555 +@incollection{han2015microstructure,
556 + author = {Han, Lu and Strange, William C.},
557 + title = {The Microstructure of Housing Markets: Search, Bargaining, and Brokerage},
558 + booktitle = {Handbook of Regional and Urban Economics},
559 + volume = {5},
560 + publisher = {Elsevier},
561 + year = {2015},
562 + pages = {813--886},
563 + doi = {10.1016/B978-0-444-59531-7.00013-2}
564 +}
565 +@article{kiel2008location,
566 + author = {Kiel, Katherine A. and Zabel, Jeffrey E.},
567 + title = {Location, Location, Location: The 3{L} Approach to House Price Determination},
568 + journal = {Journal of Housing Economics},
569 + year = {2008},
570 + volume = {17},
571 + number = {2},
572 + pages = {175--190},
573 + doi = {10.1016/j.jhe.2007.12.002}
574 +}
575 +
576 +% ---- Automated valuation and mass appraisal ----
577 +@article{clapp2003semiparametric,
578 + author = {Clapp, John M.},
579 + title = {A Semiparametric Method for Valuing Residential Locations: Application to Automated Valuation},
580 + journal = {Journal of Real Estate Finance and Economics},
581 + year = {2003},
582 + volume = {27},
583 + number = {3},
584 + pages = {303--320},
585 + doi = {10.1023/A:1025838007297}
586 +}
587 +@article{mccluskey2013prediction,
588 + author = {McCluskey, William J. and McCord, Michael and Davis, Peadar T. and Haran, Martin and McIlhatton, David},
589 + title = {Prediction Accuracy in Mass Appraisal: A Comparison of Modern Approaches},
590 + journal = {Journal of Property Research},
591 + year = {2013},
592 + volume = {30},
593 + number = {4},
594 + pages = {239--265},
595 + doi = {10.1080/09599916.2013.781204}
596 +}
597 +
598 +% ---- Canadian hedonic evidence ----
599 +@article{desrosiers1996shopping,
600 + author = {Des Rosiers, Fran{\c c}ois and Lagana, Antonio and Th{\'e}riault, Marius and Beaudoin, Marcel},
601 + title = {Shopping Centres and House Values: An Empirical Investigation},
602 + journal = {Journal of Property Valuation and Investment},
603 + year = {1996},
604 + volume = {14},
605 + number = {4},
606 + pages = {41--62},
607 + doi = {10.1108/14635789610153461}
608 +}
609 +@article{desrosiers2002landscaping,
610 + author = {Des Rosiers, Fran{\c c}ois and Th{\'e}riault, Marius and Kestens, Yan and Villeneuve, Paul},
611 + title = {Landscaping and House Values: An Empirical Investigation},
612 + journal = {Journal of Real Estate Research},
613 + year = {2002},
614 + volume = {23},
615 + number = {1--2},
616 + pages = {139--162},
617 + doi = {10.1080/10835547.2002.12091072}
618 +}
619 +@article{haider2000effects,
620 + author = {Haider, Murtaza and Miller, Eric J.},
621 + title = {Effects of Transportation Infrastructure and Location on Residential Real Estate Values: Application of Spatial Autoregressive Techniques},
622 + journal = {Transportation Research Record},
623 + year = {2000},
624 + volume = {1722},
625 + number = {1},
626 + pages = {1--8},
627 + doi = {10.3141/1722-01}
628 +}
629 +
630 +% ---- High-dimensional fixed-effects methodology ----
631 +@article{abowd1999high,
632 + author = {Abowd, John M. and Kramarz, Francis and Margolis, David N.},
633 + title = {High Wage Workers and High Wage Firms},
634 + journal = {Econometrica},
635 + year = {1999},
636 + volume = {67},
637 + number = {2},
638 + pages = {251--333},
639 + doi = {10.1111/1468-0262.00020}
640 +}
641 +@article{guimaraes2010simple,
642 + author = {Guimar{\~a}es, Paulo and Portugal, Pedro},
643 + title = {A Simple Feasible Procedure to Fit Models with High-Dimensional Fixed Effects},
644 + journal = {Stata Journal},
645 + year = {2010},
646 + volume = {10},
647 + number = {4},
648 + pages = {628--649},
649 + doi = {10.1177/1536867X1001000406}
650 +}
modified paper/sections/conclusion.tex +34 −36
@@ -5,45 +5,43 @@
5 5 % ============================================================================
6 6
7 7 Using a nationwide cross-section of 140{,}931 MLS listings and 1{,}153 absorbed
8 −neighbourhood fixed effects, we estimate a grand hedonic model of the Canadian residential
9 −housing market. The exercise yields three robust conclusions. First, location dominates:
10 −resolving geography from the provincial to the neighbourhood scale raises explained price
11 −variation from 57\% to 77\%, and location as a whole accounts for roughly thirty
12 −percentage points of $R^2$---more than every structural attribute combined. Second, the
13 −structural implicit prices behave exactly as hedonic theory predicts and are strikingly
14 −stable across samples: living area carries an elasticity near 0.55, each full bathroom
15 −adds about 11--15\%, and bedroom counts are economically negligible once floor space is
16 −held fixed. Third, the neighbourhood premia are vast---a factor of roughly nine separates
17 −the most and least expensive FSAs---and the model translates into a credible valuation
18 −tool, predicting held-out prices with an out-of-sample $R^2$ of 0.76 and a median error of
19 −16\%. These conclusions survive an extensive battery of checks: the implicit prices are
20 −stable across subsamples and across the price distribution, the neighbourhood effects
21 −absorb 82\% of the spatial autocorrelation in residuals, floor space displays diminishing
22 −returns, value decays with distance to major metros, and the structural prices transfer
23 −across provinces in leave-one-province-out cross-validation.
8 +neighbourhood fixed effects, we estimate a grand hedonic model of the Canadian
9 +residential housing market. The exercise yields three robust conclusions. First,
10 +location dominates: resolving geography from the provincial to the neighbourhood scale
11 +raises explained price variation from 57\% to 77\%, and location as a whole accounts for
12 +roughly thirty percentage points of $R^2$---more than every structural attribute
13 +combined, and a listing-level counterpart to the large land shares documented in
14 +aggregate data \citep{davis2007price,knoll2017no}. Second, the structural implicit
15 +prices behave as hedonic theory predicts and are strikingly stable across samples:
16 +living area carries an elasticity near 0.55, each full bathroom adds about 11--15\%, and
17 +bedroom counts are economically negligible once floor space is held
18 +fixed---magnitudes squarely within the range catalogued by decades of hedonic work
19 +\citep{sirmans2005composition}. Third, the neighbourhood premia are vast---a factor of
20 +roughly nine separates the most and least expensive FSAs---and the model translates into
21 +a credible valuation tool, predicting held-out prices with an out-of-sample $R^2$ of
22 +0.76 and a median error of 16\%. These conclusions survive an extensive battery of
23 +checks: the implicit prices are stable across subsamples and across the price
24 +distribution, the neighbourhood effects absorb 82\% of the spatial autocorrelation in
25 +residuals, floor space displays diminishing returns, value decays with distance to major
26 +metros as the monocentric tradition predicts, and the structural prices transfer across
27 +provinces in leave-one-province-out cross-validation.
24 28
25 −Several caveats temper these conclusions. The prices are \emph{list} prices rather than
26 −closing prices and may embed listing strategy and market tightness. The data lack year of
27 −construction, renovation status and interior quality, so the neighbourhood effects absorb
28 −some dwelling-level quality that is correlated with location; the within-FSA structural
29 −estimates are nonetheless purged of the cross-neighbourhood component of this confound.
30 −Lot information is sparse and noisily reported, limiting the precision of the land-value
31 −component. Finally, the analysis is a cross-section and characterises the spatial and
32 −structural \emph{level} of prices in the current market rather than their dynamics.
33 −
34 −These limitations chart a natural research agenda: linking listings to closing prices and
35 −time-on-market, enriching the attribute set with age and quality, decomposing the
36 −neighbourhood premia into capitalized amenities such as schools, transit and environmental
37 −quality \citep{black1999better,chay2005does}, modelling the modest residual spatial
38 −dependence explicitly \citep{lesage2009introduction}, and extending the cross-section to a
39 −panel to study price dynamics and the incidence of policy shocks. Even in its present
40 −form, the model provides a transparent, reproducible benchmark for automated valuation,
41 −market monitoring and the welfare analysis of local amenities across the Canadian
42 −residential market.
29 +The reading we propose is deliberately modest and, we believe, well supported: in
30 +Canada, the single largest priced component of a home is its neighbourhood. That fact
31 +is simultaneously a measurement result (for assessment, taxation and index
32 +construction), a validation result (a transparent fixed-effects design removes the
33 +spatial dependence that motivates far more elaborate machinery), and a policy result
34 +(the premium dispersion that supply constraints and agglomeration generate is the
35 +dominant term in the decomposition). Section~\ref{sec:discussion} details the
36 +limitations---list prices, unobserved quality, noisy lot data, a single
37 +cross-section---and the research agenda they imply. Even within those bounds, the model
38 +provides a transparent, reproducible benchmark for automated valuation, market
39 +monitoring and the welfare analysis of local amenities across the Canadian residential
40 +market.
43 41
44 42 \vspace{0.6em}
45 43 \noindent\textbf{Reproducibility.} Data engineering was performed in \texttt{DuckDB} and
46 44 \texttt{pandas}; estimation used \texttt{statsmodels} (cluster-robust OLS) and
47 45 \texttt{linearmodels} (absorbing least squares); figures were produced in
48 −\texttt{matplotlib}. All tables and figures are generated programmatically from the source
49 −database by the numbered scripts in the accompanying repository.
46 +\texttt{matplotlib}. All tables and figures are generated programmatically from the
47 +source database by the numbered scripts in the accompanying repository.
modified paper/sections/data.tex +2 −1
@@ -28,7 +28,8 @@ square metres where a numeric value can be recovered, with a missingness indicat
28 28 retained. We restrict the sample to residential dwellings (houses, condominiums, plexes,
29 29 townhouses and apartments) with a strictly positive list price and non-missing core
30 30 structural fields, and we trim the extreme 1\% tails of price and living area to limit the
31 −influence of data-entry errors and ultra-luxury outliers. The resulting estimation sample
31 +influence of data-entry errors and ultra-luxury outliers, a standard precaution in hedonic
32 +work with listing data \citep{malpezzi2003hedonic}. The resulting estimation sample
32 33 contains \textbf{140{,}931 dwellings}, of which 82{,}334 are houses and 57{,}857 are
33 34 condominiums.
34 35
added paper/sections/discussion.tex +136 −0
@@ -0,0 +1,136 @@
1 +% Author: Simon-Pierre Boucher — contact@spboucher.ai
2 +% ============================================================================
3 +\section{Discussion}
4 +\label{sec:discussion}
5 +% ============================================================================
6 +
7 +\subsection{Magnitudes in the light of the literature}
8 +
9 +\paragraph{The location share.} Our headline decomposition---structure explains 46\% of
10 +log-price variance, location a further 30 points---is a cross-sectional, listing-level
11 +statement, but its magnitude aligns closely with what aggregate approaches find from the
12 +opposite direction. \citet{davis2007price} estimate that land accounts for roughly 46\%
13 +of the value of the U.S. housing stock, with much higher shares in coastal metros, and
14 +\citet{knoll2017no} attribute about 80\% of the global house-price run-up since 1950 to
15 +land. Land and ``location'' are not identical objects---our FSA premia capitalize
16 +amenities and access as well as physical land scarcity
17 +\citep{cheshire1995price,albouy2016cities}---but both measurements say the same thing:
18 +the dwelling itself is the minority component of housing value. The ×9 span of our
19 +neighbourhood premia likewise echoes the enormous cross-city dispersion of land values
20 +documented by \citet{albouy2016cities} and is exactly the pattern predicted where supply
21 +constraints bind heterogeneously across locations
22 +\citep{saiz2010geographic,gyourko2013superstar}: Vancouver---sea, mountains and an
23 +agricultural land reserve---anchors the top of our premium distribution, as
24 +superstar-city dynamics would predict.
25 +
26 +\paragraph{Structural implicit prices.} The living-area elasticity of 0.55 sits inside
27 +the range catalogued by \citet{sirmans2005composition} across decades of published
28 +hedonics, and its decline from 0.66 to 0.55 as location controls tighten is the
29 +classic signature of omitted-locational-quality bias that within-neighbourhood designs
30 +are built to remove \citep{kiel2008location}. The near-zero conditional bedroom
31 +coefficient reproduces one of the most robust findings of that meta-literature: floor
32 +space, not room count, carries value. Our bathroom premium (11--15\%) is at the upper
33 +end of published estimates, which we attribute to bathrooms proxying for unobserved
34 +renovation status and finish quality in listing data that lack an age variable. The
35 +diminishing marginal elasticity of floor space (from $\sim$0.65 at 60~m$^2$ to
36 +$\sim$0.45 at 350~m$^2$) gives parametric confirmation of the curvature that
37 +\citet{mcmillen2010issues} detect nonparametrically.
38 +
39 +\paragraph{Spatial dependence.} The 82\% reduction of Moran's~$I$ (0.46 to 0.08) speaks
40 +directly to the oldest empirical worry in housing hedonics
41 +\citep{dubin1988estimation,basu1998analysis}. It quantifies, for a national market, the
42 +claim of \citet{bourassa2007spatial} and \citet{gibbons2012mostly} that geographic
43 +controls---here, three-character postal geography---do most of the work that parametric
44 +spatial-lag structures are designed to do. Our result does not make spatial econometrics
45 +redundant: the residual $I=0.08$ is within-FSA dependence that an explicit
46 +spatio-temporal model \citep{pace1998spatiotemporal,dube2013spatiotemporal} could
47 +exploit, particularly for prediction. It does, however, shift the burden of proof toward
48 +parsimony: a fixed effect per neighbourhood is transparent, imposes no weight-matrix
49 +assumptions, and removes five-sixths of the spatial signal.
50 +
51 +\paragraph{Quantile and provincial heterogeneity.} The rising size elasticity across the
52 +price distribution (0.56 at $\tau=0.1$ to 0.60 at $\tau=0.9$) and the rising lot
53 +elasticity mirror the distributional patterns of \citet{zietz2008determinants}, and the
54 +coastal-versus-Prairies gradient in the size elasticity (0.49 in BC to 0.66 in MB) is
55 +consistent with the submarket literature's core claim that implicit prices---not just
56 +price levels---vary across segments \citep{goodman1998housing,bourassa2003submarkets}.
57 +This heterogeneity qualifies our own grand model: the absorbed intercepts allow every
58 +neighbourhood its own price \emph{level}, but the structural slopes are pooled, and
59 +Section~\ref{subsec:heterogeneity} shows those slopes move within an economically
60 +meaningful band. Fully interacted (submarket-specific) coefficient systems are the
61 +natural next step, at a substantial cost in transparency.
62 +
63 +\paragraph{Valuation accuracy.} A held-out $R^2$ of 0.764 and a median absolute error of
64 +15.8\% place the transparent hedonic model within the accuracy bands reported in the
65 +mass-appraisal comparison literature \citep{mccluskey2013prediction} and close to the
66 +performance that machine-learning methods deliver on comparable tasks
67 +\citep{mullainathan2017machine}. Two caveats temper the comparison: our target is the
68 +\emph{list} price rather than the transaction price, and our evaluation is restricted to
69 +neighbourhoods observed in training. Within those bounds, the result supports the
70 +position of \citet{clapp2003semiparametric}: most of the predictive content of
71 +``black-box'' valuation lies in the location surface, which a fixed-effect design
72 +captures explicitly and auditably rather than implicitly.
73 +
74 +\subsection{Points of tension with prior findings}
75 +
76 +Three tensions deserve explicit statement. First, the submarket literature sometimes
77 +finds that \emph{how} submarkets are defined matters little for prediction
78 +\citep{bourassa2003submarkets}; our single-country evidence cannot adjudicate the
79 +optimal geography, and FSAs---postal artifacts---are surely not it. The 0.08 residual
80 +Moran's~$I$ suggests finer geography would still add value. Second,
81 +\citet{kuminoff2010which} warn that hedonic estimates can be fragile to specification;
82 +our robustness battery (six samples, quantiles, quadratic terms) addresses the
83 +first-stage version of this concern, but any second-stage welfare use of our premia
84 +would inherit the well-known identification problems \citep{epple1987hedonic}. Third,
85 +our LOPO exercise shows structural prices transfer imperfectly (mean held-out $R^2$ of
86 +0.36, negative for Alberta), which cuts against reading the grand model's pooled slopes
87 +as universal constants---and aligns with the heterogeneity that segmented-market models
88 +predict \citep{goodman1998housing}.
89 +
90 +\subsection{Implications}
91 +
92 +For \emph{assessment and property taxation}, the decomposition implies that
93 +neighbourhood-level value---not structure---is where most of the assessable base lives,
94 +so assessment uniformity depends first on getting location surfaces right; the FSA
95 +premia we publish are directly usable as such a surface. For \emph{index construction},
96 +the stability of structural implicit prices across provinces supports pooled hedonic
97 +indices with local intercepts \citep{hill2013hedonic}, while the quantile results
98 +caution that constant-quality adjustment differs across market segments. For
99 +\emph{automated valuation}, the results quantify the price of transparency: a fully
100 +inspectable model concedes little accuracy on within-support predictions, echoing
101 +\citet{mullainathan2017machine}'s point that prediction tasks discipline, rather than
102 +replace, economic structure. For \emph{housing policy}, a ×9 neighbourhood premium span
103 +within one country, concentrated in two metropolitan systems, is the observable imprint
104 +of restricted supply against agglomeration demand
105 +\citep{saiz2010geographic,gyourko2013superstar,combes2015empirics}; policies that expand
106 +supply where premia are highest attack the decomposition's dominant term.
107 +
108 +\subsection{Limitations}
109 +
110 +Five limitations bound the interpretation. (i)~Prices are \emph{list} prices: sellers
111 +set them strategically \citep{genesove2001loss} and they interact with search and
112 +bargaining \citep{han2015microstructure}, so implicit prices measured on listings can
113 +differ from transaction-based ones, especially in tight markets. (ii)~The data lack year
114 +of construction, renovation status and interior quality; the neighbourhood effects
115 +absorb the cross-neighbourhood component of this unobserved quality, but the within-FSA
116 +structural estimates remain exposed to within-neighbourhood quality sorting.
117 +(iii)~Lot information is sparse and noisily parsed from free text, limiting the
118 +precision of the land-versus-structure margin at the listing level. (iv)~The analysis is
119 +a single cross-section: it characterises the spatial and structural \emph{level} of
120 +prices, not their dynamics, and cannot speak to the efficiency questions of the
121 +repeat-sales tradition \citep{case1989efficiency}. (v)~FSA boundaries are postal
122 +conveniences; premia estimated at this resolution average over genuinely finer
123 +neighbourhood variation, as the residual spatial autocorrelation confirms.
124 +
125 +\subsection{Future work}
126 +
127 +The natural agenda follows the limitations: linking listings to closing prices and
128 +time-on-market to estimate the list-to-sale wedge \citep{genesove2001loss,han2015microstructure};
129 +enriching the attribute set with age and quality; decomposing the neighbourhood premia
130 +into capitalized amenities---schools, transit, environmental quality---in the
131 +quasi-experimental tradition \citep{black1999better,chay2005does,linden2008estimates};
132 +modelling the residual within-FSA dependence with spatio-temporal structure
133 +\citep{dube2013spatiotemporal}; extending the cross-section to a panel to study
134 +dynamics and policy incidence; and estimating submarket-specific coefficient systems to
135 +map the heterogeneity that our provincial results only sketch
136 +\citep{goodman1998housing,bourassa2003submarkets}.
modified paper/sections/introduction.tex +74 −52
@@ -4,68 +4,90 @@
4 4 \label{sec:introduction}
5 5 % ============================================================================
6 6
7 −A dwelling is the archetypal heterogeneous good. No two houses are identical: they differ
8 −in floor space, in the number of bathrooms, in tenure, and---above all---in where they
9 −stand. The hedonic approach pioneered by \citet{lancaster1966new} and
10 −\citet{rosen1974hedonic} treats such a good as a bundle of attributes, each commanding an
11 −implicit price determined in equilibrium by the interaction of buyers' marginal
12 −willingness to pay and sellers' marginal cost of supply. Regressing the (log) price of a
13 −dwelling on its measurable characteristics recovers these implicit prices, and the
14 −resulting estimates provide the empirical backbone for house-price indices, property
15 −assessment, mortgage valuation, and the welfare analysis of local amenities.
7 +Housing is the largest asset class on household balance sheets and the collateral behind
8 +most household debt, and its price is the single most consequential relative price most
9 +families ever face. Yet a dwelling is the archetypal heterogeneous good: no two houses
10 +are identical, and the most important attribute---\emph{where the dwelling stands}---is
11 +not a scalar that can be read off a listing sheet. The hedonic approach pioneered by
12 +\citet{lancaster1966new} and \citet{rosen1974hedonic} resolves this heterogeneity by
13 +treating the dwelling as a bundle of attributes, each commanding an implicit price
14 +determined in equilibrium by the interaction of buyers' marginal willingness to pay and
15 +sellers' marginal cost of supply. Regressing the (log) price of a dwelling on its
16 +measurable characteristics recovers these implicit prices, and the resulting estimates
17 +provide the empirical backbone for constant-quality house-price indices
18 +\citep{bailey1963regression,hill2013hedonic}, property assessment and mass appraisal
19 +\citep{mccluskey2013prediction}, mortgage valuation, and the welfare analysis of local
20 +amenities \citep{black1999better,chay2005does}.
16 21
17 −This paper estimates a hedonic price model for the Canadian residential market at national
18 −scale. Using a cross-section of \textbf{140{,}931} active MLS listings spanning the nine
22 +How much of a home's value is the structure, and how much is the location? The question
23 +is old---it is the empirical content of the realtor's adage ``location, location,
24 +location'' \citep{kiel2008location}---but its answer matters far beyond folklore.
25 +Assessment authorities must apportion value between land and improvements; property-tax
26 +reform hinges on the incidence of the land component; the macro literature has shown
27 +that land, not structure, drives both the level and the growth of aggregate housing
28 +wealth \citep{davis2007price,knoll2017no}; and the dispersion of location values across
29 +cities is the fingerprint of supply constraints and agglomeration
30 +\citep{saiz2010geographic,gyourko2013superstar,combes2015empirics}. Credible answers
31 +require holding the structural bundle fixed while letting location vary---precisely what
32 +a hedonic model with fine geographic fixed effects delivers.
33 +
34 +This paper estimates such a model for the Canadian residential market at national scale.
35 +Using a cross-section of \textbf{140{,}931} active MLS listings spanning the nine
19 36 provinces present in the data---from British Columbia to Newfoundland and Labrador---we
20 37 decompose dwelling prices into structural attributes, dwelling type and ownership form,
21 38 and a rich set of \textbf{1{,}153 neighbourhood fixed effects} defined at the Forward
22 39 Sortation Area (FSA) level. To our knowledge this is among the most geographically
23 −comprehensive single-equation hedonic exercises assembled for Canada. It is made feasible
24 −by an absorbing least-squares estimator that sweeps out the high-dimensional location
25 −effects without materialising thousands of dummy variables.
26 −
27 −The motivation is threefold. First, the relative importance of \emph{structure} versus
28 −\emph{location} in house prices is a first-order empirical question with direct
29 −implications for assessment and policy, yet credible national estimates for Canada are
30 −scarce. Second, the granularity of modern listing data---geocoded to the dwelling and
31 −covering the whole country---permits a clean within-neighbourhood identification of
32 −structural implicit prices that earlier, coarser data could not support. Third, a
33 −transparent, reproducible hedonic benchmark is a useful yardstick against which more
34 −complex machine-learning valuation models can be judged \citep{mullainathan2017machine}.
40 +comprehensive single-equation hedonic exercises assembled for Canada, where the hedonic
41 +tradition has remained largely metropolitan in scope
42 +\citep{desrosiers1996shopping,haider2000effects}. It is made feasible by an absorbing
43 +least-squares estimator in the lineage of \citet{abowd1999high} that sweeps out the
44 +high-dimensional location effects without materialising thousands of dummy variables
45 +\citep{guimaraes2010simple,correia2017reghdfe}.
35 46
36 47 Our main findings can be summarised as follows. The purely structural model explains 46\%
37 48 of the variation in log prices; adding province fixed effects raises this to 57\%, and
38 49 replacing them with FSA fixed effects lifts it to \textbf{77\%}. Neighbourhood location
39 50 \emph{alone} therefore accounts for roughly thirty percentage points of explanatory
40 −power---more than all structural attributes combined. Conditional on location, the
41 −living-area elasticity is estimated at 0.51--0.62: a 10\% larger dwelling sells for about
42 −5--6\% more. Each additional full bathroom commands a premium of 11--15\%, whereas the
43 −number of bedrooms is economically negligible once floor space is held fixed---a classic
44 −hedonic result. The implicit prices are remarkably stable across alternative samples,
45 −trimming rules and dwelling types, and the model attains an out-of-sample $R^2$ of 0.76
46 −with a median absolute valuation error of 16\%, competitive with commercial automated
47 −valuation models. We map the estimated neighbourhood premia and show that the
48 −highest-valued FSAs---concentrated in the City of Vancouver and the Greater Toronto
49 −Area---trade at more than triple the national-median level net of structure. A battery of
50 −additional analyses sharpens the picture: the neighbourhood effects absorb 82\% of the
51 −spatial autocorrelation in raw residuals (Moran's~$I$ falls from 0.46 to 0.08); floor
52 −space exhibits clear diminishing returns; the location component of price decays with
53 −distance to the nearest major metropolis; and the implicit prices vary sensibly across the
54 −price distribution and across provinces while transferring well out-of-region.
51 +power---more than all structural attributes combined, and a listing-level counterpart to
52 +the large land shares found in aggregate data \citep{davis2007price,knoll2017no}.
53 +Conditional on location, the living-area elasticity is estimated at 0.51--0.62: a 10\%
54 +larger dwelling sells for about 5--6\% more. Each additional full bathroom commands a
55 +premium of 11--15\%, whereas the number of bedrooms is economically negligible once floor
56 +space is held fixed---a recurrent finding in the hedonic literature
57 +\citep{sirmans2005composition}. The implicit prices are remarkably stable across
58 +alternative samples, trimming rules and dwelling types, and the model attains an
59 +out-of-sample $R^2$ of 0.76 with a median absolute valuation error of 16\%, a level of
60 +accuracy competitive with the mass-appraisal benchmarks reported in the valuation
61 +literature \citep{mccluskey2013prediction,mullainathan2017machine}. We map the estimated
62 +neighbourhood premia and show that the highest-valued FSAs---concentrated in the City of
63 +Vancouver and the Greater Toronto Area---trade at more than triple the national-median
64 +level net of structure. A battery of additional analyses sharpens the picture: the
65 +neighbourhood effects absorb 82\% of the spatial autocorrelation in raw residuals
66 +(Moran's~$I$ falls from 0.46 to 0.08); floor space exhibits clear diminishing returns;
67 +the location component of price decays with distance to the nearest major metropolis, as
68 +the monocentric tradition predicts \citep{alonso1964location,muth1969cities,mills1967aggregative};
69 +and the implicit prices vary sensibly across the price distribution and across provinces
70 +while transferring well out-of-region.
55 71
56 −This paper relates to several strands of the literature. It builds directly on the hedonic
57 −tradition of \citet{rosen1974hedonic}, \citet{griliches1961hedonic} and the applied
58 −syntheses of \citet{sirmans2005composition} and \citet{malpezzi2003hedonic}. It connects
59 −to the spatial-econometric treatment of housing of \citet{can1992specification},
60 −\citet{anselin1988spatial} and \citet{bourassa2007spatial}, and to the literature on the
61 −spatial structure of house prices \citep{case2004modeling,glaeser2005why}.
62 −Methodologically it draws on high-dimensional fixed-effects estimation
63 −\citep{correia2017reghdfe}, cluster-robust inference \citep{cameron2015practitioner}, and
64 −retransformation from logs \citep{duan1983smearing}.
72 +The contribution is threefold. \emph{First}, we provide the first (to our knowledge)
73 +listing-level structure-versus-location decomposition spanning the entire Canadian
74 +market, quantifying the neighbourhood as its single largest priced component.
75 +\emph{Second}, we bring the high-dimensional fixed-effects technology of modern applied
76 +microeconomics \citep{abowd1999high,correia2017reghdfe} to a national housing
77 +cross-section and validate the design directly, showing that more than a thousand
78 +absorbed neighbourhood intercepts eliminate the bulk of spatial dependence that
79 +parametric spatial models are built to address
80 +\citep{dubin1988estimation,basu1998analysis,gibbons2012mostly}. \emph{Third}, we supply a
81 +transparent accuracy benchmark for the automated-valuation debate: every coefficient of
82 +the model is inspectable, yet its held-out performance is competitive with the
83 +mass-appraisal standards reported in the literature
84 +\citep{clapp2003semiparametric,mccluskey2013prediction}.
65 85
66 −The remainder of the paper is organised as follows. Section~\ref{sec:literature} reviews
67 −the related literature. Section~\ref{sec:data} describes the data and variable
68 −construction. Section~\ref{sec:methodology} presents the empirical strategy.
69 −Section~\ref{sec:results} reports the main results. Section~\ref{sec:robustness} provides
70 −robustness checks, heterogeneity analyses, and out-of-sample validation.
86 +The remainder of the paper is organised as follows. Section~\ref{sec:literature}
87 +positions the paper in six related literatures. Section~\ref{sec:data} describes the
88 +data and variable construction. Section~\ref{sec:methodology} presents the empirical
89 +strategy. Section~\ref{sec:results} reports the main results.
90 +Section~\ref{sec:robustness} provides robustness checks, heterogeneity analyses, and
91 +out-of-sample validation. Section~\ref{sec:discussion} interprets the findings against
92 +the literature and discusses implications, limitations and future work.
71 93 Section~\ref{sec:conclusion} concludes.
modified paper/sections/literature.tex +136 −62
@@ -4,7 +4,14 @@
4 4 \label{sec:literature}
5 5 % ============================================================================
6 6
7 −\subsection{The hedonic framework}
7 +This paper sits at the intersection of six literatures: the hedonic framework and its
8 +identification; functional form and distributional heterogeneity in housing hedonics;
9 +the treatment of space---submarkets, spatial econometrics and fixed effects; the theory
10 +of amenity capitalization; urban spatial structure and the land share of housing value;
11 +and property valuation, from price indices to automated valuation models. We review each
12 +in turn and state where the present study enters.
13 +
14 +\subsection{The hedonic framework and its identification}
8 15
9 16 The hedonic method has two intellectual roots. \citet{lancaster1966new} recast consumer
10 17 theory in terms of the characteristics embodied in goods rather than the goods themselves,
@@ -12,73 +19,140 @@ while \citet{court1939hedonic} and \citet{griliches1961hedonic} developed the em
12 19 machinery of quality-adjusted price indices. \citet{rosen1974hedonic} unified these ideas
13 20 in an equilibrium model in which the observed price schedule traces out the envelope of
14 21 buyers' bid functions and sellers' offer functions; its gradient with respect to a
15 −characteristic identifies, at the margin, the implicit price of that characteristic. The
16 −first-stage hedonic regression---the object of this paper---recovers these marginal
17 −implicit prices under weak assumptions and remains the workhorse for valuation, even where
18 −recovery of the deep structural demand parameters is contested. A subsequent literature
19 −has clarified exactly what the second stage can and cannot identify:
22 +characteristic identifies, at the margin, the implicit price of that characteristic. A
23 +subsequent literature clarified exactly what the second stage---recovering preference
24 +parameters from the implicit prices---can and cannot deliver. \citet{epple1987hedonic}
25 +formalised the simultaneity problem that plagues second-stage demand estimation;
20 26 \citet{ekeland2004identification} show that nonlinearity of the hedonic price function
21 −aids identification of preferences, \citet{bajari2005hedonic} develop a tractable demand
22 −estimator with unobserved product characteristics, and \citet{kuminoff2010which} document
23 −how sensitive welfare estimates are to specification. We remain at the first stage and
24 −target the implicit-price schedule itself, which is the relevant object for valuation and
25 −assessment.
27 +aids identification of preferences; \citet{bajari2005hedonic} develop a tractable demand
28 +estimator with unobserved product characteristics; and \citet{kuminoff2010which} document
29 +how sensitive welfare estimates are to specification. We deliberately remain at the first
30 +stage and target the implicit-price schedule itself, which is the relevant object for
31 +valuation, assessment and index construction, and which is identified under much weaker
32 +assumptions than the underlying preferences.
26 33
27 −\subsection{Specification of housing hedonics}
34 +\subsection{Functional form and heterogeneity in the implicit prices}
28 35
29 36 A large applied literature studies which attributes belong in a housing hedonic and what
30 −functional form to impose. \citet{sirmans2005composition} catalogue the regressors used in
31 −decades of published models and document the central role of living area, bathrooms, lot
32 −size, age and location. \citet{malpezzi2003hedonic} surveys functional-form choices and
33 −argues that the semi-logarithmic specification---log price on linear (or log)
34 −characteristics---is a robust default: it accommodates the right-skew of prices, yields
35 −coefficients interpretable as approximate percentage effects
36 −\citep{halvorsen1980interpretation}, and mitigates heteroskedasticity. We adopt this
37 −specification throughout. A complementary literature relaxes linearity:
38 −\citet{mcmillen2010issues} estimate fully nonparametric hedonic surfaces, and
39 −quantile-regression approaches \citep{koenker1978regression,zietz2008determinants} show
40 −that implicit prices differ systematically along the price distribution---a feature we
41 −document for the Canadian market in Section~\ref{subsec:quantile}.
37 +functional form to impose. \citet{sirmans2005composition} catalogue the regressors used
38 +in decades of published models and document the central role of living area, bathrooms,
39 +lot size, age and location; their meta-tabulation also shows that bedroom coefficients
40 +are frequently small or negative once floor area is controlled---the pattern we recover
41 +for Canada. \citet{malpezzi2003hedonic} surveys functional-form choices and argues that
42 +the semi-logarithmic specification---log price on linear (or log) characteristics---is a
43 +robust default: it accommodates the right-skew of prices, yields coefficients
44 +interpretable as approximate percentage effects \citep{halvorsen1980interpretation}, and
45 +mitigates heteroskedasticity. We adopt this specification throughout. A complementary
46 +literature relaxes its restrictions in two directions. \citet{mcmillen2010issues}
47 +estimate fully nonparametric hedonic surfaces and show that parametric functional forms
48 +can mask economically meaningful curvature---a concern we address directly with a
49 +quadratic-in-log-area specification that reveals diminishing returns to floor space.
50 +Quantile-regression approaches \citep{koenker1978regression,zietz2008determinants} show
51 +that implicit prices differ systematically along the price distribution, so that a single
52 +conditional-mean coefficient can be an incomplete summary; we document this for the
53 +Canadian market in Section~\ref{subsec:quantile}. \emph{Gap:} these strands are almost
54 +exclusively metropolitan or regional in scope; national-scale evidence on functional form
55 +and distributional heterogeneity, estimated on a single consistent dataset, is rare.
56 +
57 +\subsection{Space in housing models: submarkets, spatial econometrics, fixed effects}
58 +
59 +Because housing is immobile, location is intrinsic to its value, and unobserved local
60 +amenities induce strong spatial dependence in prices. The empirical importance of this
61 +dependence is long established: \citet{dubin1988estimation} showed that ignoring
62 +spatially autocorrelated errors distorts hedonic inference, and \citet{basu1998analysis}
63 +documented pervasive residual autocorrelation in Dallas transactions.
64 +\citet{can1992specification} and \citet{anselin1988spatial} formalised spatial
65 +autocorrelation in hedonic errors, while \citet{pace1998spatiotemporal} and, in the
66 +Canadian context, \citet{dube2013spatiotemporal} developed spatio-temporal weight-matrix
67 +approaches. A parallel strand argues that housing markets are segmented into
68 +\emph{submarkets} within which prices are set: \citet{goodman1998housing} formalise
69 +market segmentation tests, and \citet{bourassa2003submarkets} show that accounting for
70 +submarkets materially improves prediction accuracy---with the practical finding that
71 +simple geographic definitions perform as well as statistically derived ones.
72 +\citet{bourassa2007spatial} and \citet{case2004modeling} compare these approaches
73 +head-to-head and conclude that geographic submarket controls capture most of what
74 +parametric spatial models capture, at a fraction of the complexity.
75 +
76 +Our approach takes this conclusion to its logical end: rather than imposing a spatial
77 +weight matrix, we absorb a fixed effect for each of 1{,}153 FSA neighbourhoods, allowing
78 +the data to assign an arbitrary price level to each area and identifying structural
79 +implicit prices from \emph{within-neighbourhood} variation. This is the housing analogue
80 +of the high-dimensional fixed-effects designs pioneered in labour economics by
81 +\citet{abowd1999high} and now standard in applied microeconomics, with computation
82 +following \citet{guimaraes2010simple} and \citet{correia2017reghdfe} and inference
83 +clustered at the neighbourhood level \citep{cameron2015practitioner}. The choice also
84 +speaks to a methodological debate: \citet{gibbons2012mostly} caution that parametric
85 +spatial-lag models are often hard to interpret causally and that flexible fixed effects
86 +are frequently preferable, while \citet{lesage2009introduction} develop the
87 +spatial-autoregressive alternative. We side with the fixed-effects approach but validate
88 +it directly, testing for residual spatial autocorrelation with Moran's~$I$
89 +\citep{moran1950notes} before and after absorbing the neighbourhood effects.
90 +\emph{Gap:} submarket and spatial-fixed-effect studies are overwhelmingly single-city;
91 +we are not aware of a prior implementation at the scale of an entire national market
92 +with more than a thousand absorbed neighbourhoods.
42 93
43 94 \subsection{Amenity capitalization}
44 95
45 −A large share of the value of location reflects the capitalization of local public goods
46 −and disamenities into prices. Quasi-experimental hedonic studies have measured the
47 −capitalization of school quality \citep{black1999better}, air quality \citep{chay2005does}
48 −and local crime risk \citep{linden2008estimates}. Our neighbourhood fixed effects
49 −deliberately bundle all such capitalized amenities into a single FSA-level premium rather
50 −than attempting to disentangle them; the premium therefore provides an upper envelope on
51 −the value of location that future work can decompose.
96 +Why should a neighbourhood intercept capture economic value at all? The theoretical
97 +answer runs from \citet{tiebout1956pure}, in which households sort across jurisdictions
98 +according to local public goods, through \citet{oates1969effects}, who first measured
99 +the capitalization of taxes and school spending into house prices, to
100 +\citet{roback1982wages}, who embedded amenity capitalization in a general spatial
101 +equilibrium of wages and rents. Quasi-experimental hedonic studies have since measured
102 +the capitalization of specific amenities: school quality \citep{black1999better}, air
103 +quality \citep{chay2005does} and local crime risk \citep{linden2008estimates}, while
104 +\citet{cheshire1995price} and \citet{albouy2016cities} quantify the total land-value
105 +content of local amenities. Our neighbourhood fixed effects deliberately bundle all such
106 +capitalized amenities into a single FSA-level premium rather than attempting to
107 +disentangle them; the estimated premium is therefore an upper envelope on the value of
108 +location that future work can decompose amenity by amenity.
52 109
53 −\subsection{Space in housing models}
110 +\subsection{Urban spatial structure and the land share of housing value}
54 111
55 −Because housing is immobile, location is intrinsic to its value, and unobserved local
56 −amenities induce strong spatial dependence in prices. \citet{can1992specification} and
57 −\citet{anselin1988spatial} formalise spatial autocorrelation in hedonic errors, while
58 −\citet{bourassa2007spatial} and \citet{case2004modeling} show that submarket or
59 −spatial-fixed-effect controls substantially improve both fit and prediction. Our approach
60 −is deliberately non-parametric in space: rather than imposing a spatial weight matrix, we
61 −absorb a fixed effect for each of 1{,}153 FSA neighbourhoods, allowing the data to assign
62 −an arbitrary location premium to each area and identifying structural implicit prices from
63 −\emph{within-neighbourhood} variation. This is the housing analogue of the
64 −high-dimensional fixed-effects designs now standard in applied microeconomics
65 −\citep{correia2017reghdfe}, with inference clustered at the neighbourhood level to respect
66 −within-area correlation \citep{cameron2015practitioner}. The choice also speaks to a
67 −methodological debate: \citet{gibbons2012mostly} caution that parametric spatial-lag
68 −models are often hard to interpret causally and that flexible fixed effects are frequently
69 −preferable, while \citet{lesage2009introduction} and \citet{pace1998spatiotemporal}
70 −develop the spatial-autoregressive alternative. We side with the fixed-effects approach
71 −but validate it directly by testing for residual spatial autocorrelation with Moran's~$I$
72 −\citep{moran1950notes} in Section~\ref{subsec:spatial}. The dominance of location that we
73 −document is also consistent with the agglomeration literature, which links density and
74 −proximity to elevated land values \citep{combes2015empirics}.
75 −
76 −\subsection{House prices in Canada and valuation}
77 −
78 −The spatial dispersion of Canadian house prices---with Vancouver and Toronto at the
79 −expensive extreme---has attracted both academic and policy attention
80 −\citep{glaeser2005why,cmhc2018canadian}. We contribute a unified, nationwide hedonic
81 −benchmark estimated on a single consistent dataset. Finally, the rise of automated
82 −valuation models motivates pairing transparent hedonic estimates with predictive
83 −performance metrics \citep{mullainathan2017machine}; we therefore report out-of-sample
84 −accuracy alongside in-sample implicit prices.
112 +The monocentric tradition of \citet{alonso1964location}, \citet{muth1969cities} and
113 +\citet{mills1967aggregative} predicts that land rents---and hence house prices, holding
114 +structure constant---decline with distance from employment centres, a prediction whose
115 +modern quantitative form is estimated by \citet{ahlfeldt2015density} on uniquely sharp
116 +Berlin data. Our distance-to-metro gradient is the reduced-form counterpart of this
117 +prediction at national scale, and the agglomeration literature reviewed by
118 +\citet{combes2015empirics} supplies the economic mechanism. A related macro-oriented
119 +literature measures how much of housing value is land rather than structure:
120 +\citet{davis2007price} put the U.S. residential land share near 46\% in aggregate,
121 +\citet{knoll2017no} show that land appreciation explains about 80\% of the global
122 +house-price increase since World War~II, and \citet{albouy2016cities} document enormous
123 +cross-city dispersion in land values. On the supply side, \citet{saiz2010geographic}
124 +shows that geographic constraints generate exactly the kind of persistent cross-location
125 +price dispersion that our fixed effects absorb, and \citet{gyourko2013superstar}
126 +formalise the resulting ``superstar city'' dynamics---with Vancouver and Toronto as
127 +textbook candidates \citep{glaeser2005why,cmhc2018canadian}. \emph{Gap:} this literature
128 +measures the land/location share with aggregate or city-level data; a listing-level
129 +decomposition for an entire national market, in which the location share is estimated
130 +directly from within-neighbourhood variation, complements it from the bottom up.
131 +
132 +\subsection{Valuation: price indices, mass appraisal, and machine learning}
133 +
134 +The hedonic regression is one of two workhorses of constant-quality house-price
135 +measurement, alongside the repeat-sales method of \citet{bailey1963regression} that
136 +underlies the \citet{case1989efficiency} index tradition; \citet{hill2013hedonic}
137 +surveys and taxonomises the hedonic index literature. The same estimated surface doubles
138 +as a valuation engine: \citet{clapp2003semiparametric} develops a location-value surface
139 +for automated valuation, \citet{kiel2008location} systematise the ``location, location,
140 +location'' decomposition of house-price determinants, and the mass-appraisal literature
141 +benchmarks prediction accuracy across model classes \citep{mccluskey2013prediction}.
142 +The rise of machine-learning valuation \citep{mullainathan2017machine} has sharpened the
143 +question of what transparency costs in accuracy terms. \emph{Gap:} few studies report
144 +honestly held-out accuracy for a fully transparent hedonic model at national scale; our
145 +out-of-sample results provide exactly that benchmark.
146 +
147 +\subsection{Canadian evidence}
148 +
149 +Canada has a productive hedonic tradition, largely centred on single metropolitan areas:
150 +the Quebec City program of Des Rosiers and coauthors measures the capitalization of
151 +shopping centres \citep{desrosiers1996shopping} and landscaping quality
152 +\citep{desrosiers2002landscaping}, and \citet{haider2000effects} apply spatial
153 +autoregressive techniques to Toronto transactions. Policy work documents the spatial
154 +dispersion of Canadian prices \citep{cmhc2018canadian}, with Vancouver and Toronto at
155 +the expensive extreme \citep{glaeser2005why}. \emph{Gap:} to our knowledge no previous
156 +study estimates a single hedonic system spanning the Canadian market coast to coast at
157 +the listing level; the national scope, the 1{,}153 absorbed neighbourhood effects, and
158 +the validated out-of-sample accuracy are, jointly, the contribution of this paper.
modified paper/sections/methodology.tex +7 −6
@@ -49,12 +49,13 @@ correlation between structural quality and unobserved locational quality
49 49 \citep{can1992specification,bourassa2007spatial}.
50 50
51 51 Including more than a thousand dummies directly is numerically wasteful. We instead
52 −estimate Equation~\eqref{eq:hedonic} by \emph{absorbing least squares}, which partials the
53 −FSA effects out of both the dependent variable and the regressors before estimating the
54 −structural coefficients \citep{correia2017reghdfe}; the slope estimates are numerically
55 −identical to full-dummy OLS. All standard errors are clustered at the FSA level to allow
56 −for arbitrary within-neighbourhood correlation and heteroskedasticity
57 −\citep{cameron2015practitioner}.
52 +estimate Equation~\eqref{eq:hedonic} by \emph{absorbing least squares}, in the
53 +high-dimensional fixed-effects tradition initiated by \citet{abowd1999high}: the FSA
54 +effects are partialled out of both the dependent variable and the regressors before the
55 +structural coefficients are estimated \citep{guimaraes2010simple,correia2017reghdfe},
56 +and the slope estimates are numerically identical to full-dummy OLS. All standard errors
57 +are clustered at the FSA level to allow for arbitrary within-neighbourhood correlation
58 +and heteroskedasticity \citep{cameron2015practitioner}.
58 59
59 60 \subsection{Specification ladder}
60 61 \label{subsec:ladder}
modified paper/sections/results.tex +4 −2
@@ -42,7 +42,7 @@ rising to nearly 15\% in the province-FE model. Half bathrooms attract a small
42 42 for older or more compartmentalised floor plans once total area and full baths are held
43 43 fixed. Bedrooms are economically negligible conditional on living area: holding floor
44 44 space constant, subdividing it into more bedrooms does not raise value---a textbook
45 −hedonic finding that recurs across our samples. Lot information enters positively ($\ln$
45 +hedonic finding \citep{sirmans2005composition} that recurs across our samples. Lot information enters positively ($\ln$
46 46 lot elasticity around $0.03$) but modestly, reflecting both the noisiness of the parsed
47 47 lot field and the fact that, within a neighbourhood, lot variation is compressed.
48 48
@@ -159,7 +159,9 @@ The dashed line is the constant-elasticity estimate.}
159 159 \label{subsec:gradient}
160 160
161 161 Classic urban theory predicts that, holding structure fixed, value declines with distance
162 −from employment centres. We compute the great-circle distance from each listing to the
162 +from employment centres
163 +\citep{alonso1964location,muth1969cities,mills1967aggregative,ahlfeldt2015density}. We
164 +compute the great-circle distance from each listing to the
163 165 nearest of nine major Canadian metropolitan centres and relate it to the location
164 166 component of price (the residual from a structure-only model). Figure~\ref{fig:gradient}
165 167 confirms a pronounced gradient: dwellings within the metropolitan core carry location
modified paper/sections/robustness.tex +5 −3
@@ -48,7 +48,8 @@ full-bathroom premium (right); the dashed line is the OLS estimate.}
48 48 \label{subsec:spatial}
49 49
50 50 A central justification for the neighbourhood fixed effects is that they should absorb the
51 −spatial dependence that pervades raw housing residuals \citep{anselin1988spatial}. We test
51 +spatial dependence that pervades raw housing residuals
52 +\citep{anselin1988spatial,dubin1988estimation,basu1998analysis}. We test
52 53 this directly by computing Moran's~$I$ \citep{moran1950notes} on the residuals, using
53 54 row-standardised $k$-nearest-neighbour spatial weights ($k=10$) on a random sample of
54 55 15{,}000 listings. The structure-only model leaves enormous spatial autocorrelation in its
@@ -117,8 +118,9 @@ attains an out-of-sample $R^2$ of \textbf{0.764} on log price---essentially iden
117 118 its in-sample fit, indicating negligible over-fitting despite the thousand-plus location
118 119 effects. In price levels, after Duan smearing, the \textbf{median absolute valuation error
119 120 is 15.8\%}, the mean is 22.5\%, and \textbf{59\% of held-out dwellings are priced within
120 −$\pm$20\%} of their actual list price (34\% within $\pm$10\%). These figures are
121 −competitive with commercial automated valuation models and establish the transparent
121 +$\pm$20\%} of their actual list price (34\% within $\pm$10\%). These figures are within
122 +the accuracy bands reported in the mass-appraisal and automated-valuation literature
123 +\citep{mccluskey2013prediction,clapp2003semiparametric} and establish the transparent
122 124 hedonic specification as a credible valuation benchmark \citep{mullainathan2017machine}.
123 125
124 126 \input{../results/tables/oos}
125 127