The chart below each heading overlays the cumulative return of one factor across every city that has it, all rebased to 0% at January 2017 so divergence reflects market conditions rather than differences in dataset length.
Jump to any factor — the count shows how many of the covered cities have it.
Quality-adjusted price appreciation that cannot be explained by changes in what is being sold or how characteristics are priced. The closest thing to a pure market return.
Z-score of each area's trailing-12-month median price/sqm against the city-wide distribution. Rising = expensive areas pulling away from cheap ones (gentrification or polarisation); falling = convergence.
Septile bucket of each transaction's price/sqm relative to the trailing-12-month distribution, scored from -3 (cheapest) to +3 (dearest). Rising = the most expensive deals are commanding a growing premium over hedonic predictions; falling = price convergence.
U-shaped vintage premium centred on each city's mid-century cohort. Positive cumulative return means stock far from that cohort (older period homes and recent builds) has outperformed mid-century stock.
Encoding is city-specific (centred on band 4 for London/NYC, band 2 for Singapore, band 3 for Seoul and Tokyo) because each city's "middle" era is different. The series share an interpretation but not a unit, so read this chart for direction and timing rather than absolute magnitude.
These factors appear in some city models but not all of them. Each chart shows only the cities where the factor is present.
Repricing of property size: how much the market is paying for an additional square metre over time. A rising line means space has become progressively more expensive.
Encoding differs by city. Taipei uses a linear total_floor_area (price per extra sqm); Paris and Seoul use log_floor_area so the per-sqm premium is non-linear in size; Singapore uses log_floor_area_sq_c80, a squared-log deviation from the typical 80sqm 4-room HDB flat. Read the chart for direction and timing of size-premium shifts, not for absolute magnitude. London is omitted because its selected size factor is now inv_floor_area (an inverse small-unit-premium encoding) and NYC because its size factor is log_units_res (residential units per building), not floor area.
Non-linear room-count premium: whether dwellings with more or fewer rooms than the typical layout command a premium, holding floor area constant. Positive cumulative return means room count has grown more valuable relative to open space.
All four model room count as a U-shape, centred on a different typical layout per market (band 2 for London, Taipei and Tokyo, band 4 for Singapore's 4-room HDB flats), so read the chart for direction and timing rather than absolute magnitude. NYC and Paris do not select a room factor.
Linear ageing effect: how the market reprices a building per additional year of age, holding construction-era cohort constant. Captures ongoing depreciation versus renovation cycles.
NYC, Singapore, Seoul and Tokyo all use a linear building_age (price effect per additional year of age), so this comparison is directly apples-to-apples. London and Paris encode age primarily through Construction Period bands rather than a separate linear factor.