Housing Returns across the Income Distribution
2026 Real Estate and Urban Economics Conference
24 September 2026
Capital gains rise by 0.017 pp per income-rank point
Annualized real capital gains are close to linear in income rank across the whole distribution.
0.017 pp per rank point, or 1.36 pp between the 90th and 10th percentiles
Binned means of annualized real capital gains by income rank.
Location fixed effects absorb the gradient
Y_{i}=\beta_0+\beta_1\text{Income Rank}_{i} + X_{i}\Gamma + \epsilon_{i}
Property and buyer characteristics and purchase- and sale-year fixed effects leave most of the coefficient intact.
Municipality fixed effects remove two-thirds, postcode fixed effects four-fifths. Postcode-by-year fixed effects leave a coefficient statistically indistinguishable from zero.
A Gelbach decomposition attributes 88 percent of the movement in the coefficient to location, against 6 percent for timing and 5 for buyer characteristics.
Coefficient on income rank across specifications, 95 percent confidence intervals. Controls are cumulative. Location fixed effects as in Goldsmith-Pinkham and Shue (2023) and Kermani and Wong (2026). Timing-by-location effects absorb local cycles (Ferreira and Gyourko 2023).
The yield gradient offsets the capital-gain gradient
The total-return gradient is statistically insignificant and stable across specifications.
Using an internal-rate-of-return measure that nets out carrying costs it is −0.031.
Capital gains and rental yields move in opposite directions across locations (Amaral et al. 2025).
Additive measure: capital gain plus a location rental yield, imputed at postcode and municipality level from the 1999 register cross-section of rents as rent per square meter over price per square meter. Controls added progressively.
Choice sets are larger for higher-income buyers
Share of same-year transactions a buyer could finance under maximum borrowing.
High-growth areas are municipalities in the top quintile of average capital gains.
Credit constraints alone do not pin down who buys where
Credit changes who owns
Looser credit lets constrained households buy. Ownership responds, prices barely move (Kaplan et al. 2020).
Credit changes what it costs
With a fixed stock, looser credit raises every bid. Prices absorb the shock and the ranking of bidders is unchanged (Greenwald and Guren 2025).
The composition of buyers did not change
Lownc_{imt} = \alpha_m + \gamma_t + \sum_{k=1998,\, k\neq 2003}^{2010} \beta_k (HighGrowth_m \times \mathbf{1}_{t=k}) + X_{it}\Gamma + \epsilon_{imt}
Event-study coefficients for high-growth municipalities relative to others. 95 percent confidence intervals. High-growth municipalities are the top fifth by 1997–2002 price growth. The 2016 macroprudential tightening in Copenhagen and Aarhus shows the same pattern in reverse.
Location accounts for 88 percent of the explained movement
A Gelbach (2016) decomposition attributes the shrinkage in the coefficient to groups of controls independently of their order.
88% location, under postcode fixed effects. Timing 6 percent, buyer characteristics 5, property characteristics 1.
A statement about what accounts for the gradient statistically, not about what causes it.
High-price, high-appreciation markets have low rental yields
Yields are imputed at the postcode and municipality level from the 1999 register cross-section of unit-level rents, as rent per square meter over price per square meter
Within Copenhagen the gradient is slightly negative
Within the Capital region, higher income predicts a slightly lower gain, −0.004 per rank point with postcode and timing fixed effects.
The national gradient reflects exposure to high-growth locations and not better within-market performance.
Average annualized capital gain by income rank, capital region and rest of Denmark.
Interpretation: constraints plus inelastic supply
With a fixed stock of dwellings allocated to the highest bidders, a proportional increase in borrowing capacity raises every bid and the clearing price but leaves the ranking of bids unchanged. Composition can move only where supply responds. The evidence is consistent with this reading and inconsistent with a simple credit-access account. We do not isolate a single cause.
Inverse housing supply elasticity (Guren et al. 2021). Fixed supply and persistent price growth in high-price cities (Gyourko et al. 2013). Left: by municipality. Right: by the income rank of buyers, national and regional prices. The variable is an inverse elasticity.
Risk does not account for the gradient
Moving from the 10th to the 90th percentile raises the standard deviation of log house-price growth by about 0.008. At a benchmark Sharpe ratio of 0.36 (Doeswijk et al. 2020), the implied premium is at most 21 percent of the baseline gap. ¨
Total returns under the internal rate of return
Left: implied annual rental yield. Right: total return. The IRR nets out maintenance, property tax and transaction costs.
IRR gradient across specifications
Imputed capital gains for single transactions
Capital gains imputed for properties that sold once, by income rank, binned means and regression coefficients.
Maximum purchase price by income rank
Municipality-level capital gains by buyer rank
Homeownership by income
Aggregate house-price growth
External validity: US ZIP codes by 2000 price level
