UQO Working Papers · WP05 · v1.1
Airbnb, Residential Rents, and Housing Market Pressure: A Hedonic and Spatial Econometric Analysis
- Year
- 2026
- First release
- October 2026
- Latest revision
- October 2026
- Publication status
- Working paper
The Airbnb–rent correlation in Quebec is mostly between neighbourhoods
With city fixed effects one more Airbnb within 500 m means rents 0.4% higher; with forward-sortation-area fixed effects the coefficient falls to 0.08% and is indistinguishable from zero under Conley or neighbourhood-clustered standard errors.
Abstract
This paper examines the relationship between short-term rental (STR) activity and residential rents in Quebec, Canada, using a 2026 cross-section of 3,456 Airbnb listings and 8,303 residential rental listings with precise coordinates. For each rental listing the number of Airbnb listings within 250 m, 500 m, 1 km and 2 km (Haversine distances) is counted and hedonic rent equations are estimated with building-type and location fixed effects. With city fixed effects, one additional Airbnb listing within 500 m is associated with monthly rents about 0.4% higher (0.3–0.5% across specifications); the per-listing coefficient decays with distance, the association is larger at the upper quantiles of the conditional rent distribution, and it survives spatial lag and spatial error models (total impact 0.40%).
Two further results qualify this picture. First, heteroskedasticity-robust standard errors overstate precision: Conley spatial HAC and neighbourhood-clustered standard errors are about three times larger, although the city-fixed-effects estimate remains significant. Second, with fixed effects for forward sortation areas (postal-code neighbourhoods), the coefficient falls to 0.08% per listing and is not distinguishable from zero under spatial inference, with the same pattern for logarithmic and extensive-margin exposure measures. The Airbnb–rent association in these data is therefore predominantly a between-neighbourhood phenomenon, consistent with STR supply concentrating in high-amenity, high-rent neighbourhoods rather than with a within-neighbourhood causal effect.
Spatially blocked cross-validation shows that the apparent advantage of tree-based machine-learning models over the linear hedonic model disappears when test neighbourhoods are unseen, and a hedonic model of Airbnb nightly prices finds no premium attached to city-level rents. The paper discusses the implications for the interpretation of cross-sectional STR–rent correlations and for Quebec's registration-based regulation.
Key findings
8,303
Quebec rental listings matched to 3,456 Airbnb listings
+0.4%
rent per additional Airbnb within 500 m, city fixed effects
0.08%
same coefficient with neighbourhood (FSA) fixed effects, insignificant
≈ 3×
Conley and FSA-clustered errors relative to HC1 errors
0.40%
total impact per listing in the spatial lag model
0.65 → 0.43
random-forest R² from random to FSA-blocked folds, below OLS at 0.47
Scientific contributions
- First listing-level STR–rent analysis for a Canadian market
- Exact distance-based matching of 8,303 rental listings to 3,456 Airbnb listings yields exposure counts at four radii (mean 7.92 within 500 m, 44% zeros) and hedonic rent equations with building-type, city and neighbourhood fixed effects.
- The identification problem is made measurable
- Moving from 187 city to 289 forward-sortation-area fixed effects cuts the per-listing coefficient from 0.0039 to 0.0008; log and extensive-margin exposures behave the same way, and within neighbourhoods any Airbnb presence is associated with 2.8% lower rents.
- Spatial inference triples the standard errors
- Conley spatial HAC (1–10 km) and FSA-clustered standard errors are 0.0007 against an HC1 error of 0.0002, taking the city-FE t-statistic from 17 to about 5.5 and the FSA-FE one to 1.1–2.0.
- Spatial lag and error models confirm the city-FE estimate
- Kelejian–Prucha GMM gives ρ = 0.134 and λ = 0.544; LeSage–Pace impacts are 0.0034 direct, 0.0005 indirect and 0.0040 total, equal to OLS.
- Machine learning's edge is neighbourhood interpolation
- Random-forest and gradient-boosting R² fall from 0.65 under random five-fold cross-validation to 0.43 and 0.41 under FSA-blocked folds, below the linear hedonic model at 0.47.
- Airbnb prices do not respond to city rents
- The scraped price field is bimodal (26.5% of listings above $1,000); on the plausibly priced subsample the entire-home and size premia reappear, the superhost discount reported in the earlier version disappears, and the city-rent coefficient is 0.003 (s.e. 0.052).
Methodology
Methodology (5)
Hedonic rent equations with nested fixed effects
Log monthly rent on Airbnb exposure, structural controls and building type with city fixed effects (Model 1c) and forward-sortation-area fixed effects (Model F2), including log and any-Airbnb exposure variants.
Spatially robust inference
HC1, city-clustered, FSA-clustered and Conley spatial HAC standard errors with 1–10 km cutoffs, compared side by side for both fixed-effects designs.
Spatial lag and spatial error models
Kelejian–Prucha GMM on a k-nearest-neighbour weight matrix built on projected kilometre coordinates, with White-robust errors and LeSage–Pace direct, indirect and total impacts.
Quantile regressions and buffer decomposition
Conditional quantiles of rent, per-listing coefficients by radius and by ring, subsample, trimming and leave-one-city-out checks (0.0038–0.0040).
Spatially blocked machine-learning benchmark
Random forest, gradient boosting and OLS compared under random and FSA-blocked five-fold cross-validation on 4,370 listings with 181 neighbourhood groups.
Research data
- Airbnb listings, Quebec
- 5,000 listings scraped in early 2026, of which 3,456 with a displayed price; coordinates, property type, rating, reviews, superhost status, capacity and amenities; price field bimodal and partly total-stay prices.
- Residential rental listings
- 8,303 Realtor.ca rental listings scraped on 8 January 2026 (from 8,356 raw) with monthly asking rent, coordinates, bedrooms, bathrooms, building type, interior size and postal code mapped to 333 forward sortation areas across 253 municipalities.
- Spatial exposure measures
- For every rental listing and radius of 250 m, 500 m, 1 km and 2 km, six Airbnb metrics (count, density, mean nightly price, entire-home share, and more) computed by chunked vectorised Haversine distances, plus city-level aggregates.
Reproducibility
- Twelve numbered Python scripts; steps 04–12 and the paper rebuild in under one minute from the five committed parquet datasets (seed 42)
- New script 12 (FSA fixed effects and Conley/cluster inference) and a run log of the 2026-10-17 pipeline in results/logs/
- 18 LaTeX table fragments and 17 figures regenerated from results; readable variable labels generated from src/latex_tables.py
- Full LaTeX source of the 59-page paper; pre-revision PDF, review report and changelog archived in docs/
Research artifacts
Version history
| Version | Date | Description | File |
|---|---|---|---|
| v1.1current | October 2026 | Current manuscript as published on www.spboucher.ai (revised edition after the 2026-10-17 scientific review for papers 1–17). | PDF · 59 p. |