UQO Working Paper No. 5· 51 pages
Airbnb, Residential Rents, and Housing Market Pressure: A Hedonic and Spatial Econometric Analysis
Nearby Airbnb activity is priced into Quebec rents
Is short-term rental density associated with higher residential rents? Each additional Airbnb listing within 500 m corresponds to roughly +0.4% monthly rent.
Simon-Pierre Boucher — contact@spboucher.ai
+0.4%
monthly rent per additional Airbnb listing within 500 m (0.3–0.5% across specifications)
8,303 + 3,456
cleaned rental listings and Airbnb listings across Quebec
0.93% → 0.06%
monotonic spatial decay of the association from 250 m to 2 km
0.0047 at τ = 0.90
quantile coefficient at the top of the rent distribution, ~25% above OLS
ρ̂ = 0.137***
spatial lag parameter; coefficient survives SAR/SEM with mild attenuation (0.0034)
0.0038–0.0040
leave-one-city-out coefficient range, including dropping Montreal
Abstract
This paper investigates the relationship between Airbnb short-term rental activity and residential rents in Quebec, Canada, using cross-sectional microdata on 3,456 cleaned Airbnb listings and 8,303 rental listings. For every rental unit, Airbnb exposure is measured within 250 m, 500 m, 1 km, and 2 km buffers via Haversine distances, and a hedonic pricing framework augmented with spatial econometric techniques quantifies the conditional association between nearby Airbnb presence and monthly rents.
Baseline estimates indicate that an additional Airbnb listing within 500 m is associated with a statistically significant rent increase of approximately 0.3–0.5%, controlling for dwelling characteristics, building type, and city fixed effects — about $6–10 per month at the median rent, the same order of magnitude as Berlin's quasi-experimental evidence. The per-listing association decays monotonically with distance, from 0.93% at 250 m to 0.06% at 2 km.
Quantile regressions show the association is strongest at the upper tail of the rent distribution (0.0047 at the 90th percentile, about 25% above OLS), and the coefficient survives spatial autoregressive and spatial error models estimated by GMM with only mild attenuation. The paper explicitly cautions that these cross-sectional associations should not be read as causal effects, and discusses implications for housing affordability and short-term rental regulation.
Key Results
Canadian evidence on the Airbnb–rent link
Fills a documented gap in Canadian and Quebec evidence on short-term rentals and housing costs, benchmarking magnitudes against Berlin, Los Angeles, and Boston quasi-experiments.
Multi-radius spatial exposure measurement
Six exposure metrics (count, density, mean price, entire-home share, mean rating, superhost share) computed at four Haversine buffer radii yield 24 exposure variables per rental listing.
Distance-decay gradient documented
The per-listing rent association falls monotonically from 0.93% at 250 m to 0.06% at 2 km, consistent with a genuinely local neighborhood-level channel.
Robustness across estimators and samples
The coefficient stays positive and significant at 1% in every check: SAR/SEM spatial models, city-clustered errors, ring decomposition, trimming, and leave-one-city-out including Montreal.
Complementary Airbnb pricing model
A hedonic model of nightly prices shows short-term rental pricing is driven by listing characteristics, with a superhost discount that contrasts with premia in prior work — discussed, not hidden.
Candid identification discussion
Endogeneity and identification limits are addressed head-on: cross-sectional associations are explicitly framed as conditional capitalization gradients, not causal effects, with seven stated limitations.
Data
Airbnb listings (Quebec)
Roughly 5,000 scraped listings cleaned to 3,456, with nightly price, coordinates, property type, rating, review counts, and superhost status.
Realtor.ca rental listings
8,356 scraped listings cleaned to 8,303, with monthly rent, coordinates, bedrooms, bathrooms, building type, and unit size, covering the province of Quebec.
Spatially merged exposure dataset
For each rental, six Airbnb exposure metrics at four buffer radii (250 m to 2 km) computed via chunked vectorized Haversine distances, plus city-level aggregates for 153 cities.
Methodology
Hedonic rent regressions
Log monthly rent on Airbnb exposure plus dwelling controls, building type, and city fixed effects (Models 1a–1e), estimated by OLS with HC1 robust errors.
Spatial econometrics (SAR/SEM)
Spatial lag and spatial error models with row-standardized KNN(5) weights, estimated by Kelejian–Prucha GMM via PySAL/spreg; ρ̂ = 0.137, λ̂ = 0.539.
Quantile regression
Estimated at τ ∈ {0.10, 0.25, 0.50, 0.75, 0.90} plus a fine grid, tracing the exposure gradient from 0.0037 to 0.0047 across the rent distribution.
Machine-learning benchmark with SHAP
OLS, LASSO, Elastic Net, Random Forest, and GBM under an 80/20 split (RF test R² = 0.71), with SHAP attribution confirming the exposure variable's role.
Extended robustness programme
Nine checks: alternative radii and exposures, subsamples, trimming, city-clustered standard errors, size control, log(1+count) form, ring decomposition, and leave-one-city-out.
Reproducibility
- Eleven-script numbered Python pipeline from raw data inspection to extended robustness; fully deterministic (seed 42), most steps run in under 15 seconds on Apple Silicon.
- Committed processed parquet files make steps 04–11 and the entire paper reproducible without the raw scraped data, which cannot be redistributed.
- Rebuilt under a strict no-result-changes policy: 3/3 merged parquets value-identical and 14 of 16 tables byte-identical to the originals, with residual float-level diffs documented.
- All 32 references added in the scholarly upgrade verified via Crossref/JMLR DOIs, with per-reference justification; 56 references in total.
- The 51-page LaTeX paper builds via latexmk with zero unresolved references; 13 LaTeX table fragments and 16 publication PDF figures regenerate from the pipeline.