This stage tests whether tightness today foretells price changes tomorrow. The dependent variable is future price growth; the key input is current tightness, with demand, supply, and the sold-price level alongside it.

Variables

  • Tightness θ: search demand divided by listed supply, as defined above, the key predictor being tested.
  • Search demand, listed supply, sold transactions: the counts behind tightness and the price, carried as inputs.
  • Sold price: the Land Registry sold price for the area and year, a coarse area-level proxy weighted across the local authorities in each travel-to-work area.
  • Annual price growth: the year-on-year change in the area’s sold price, and its one- and two-year-ahead values, the outcomes the lead-lag test predicts.

Data source

HM Land Registry sold prices, carried in the Rightmove listings feed, aggregated to the travel-to-work area, joined to the tightness series. Sold prices populate a majority of sale listings; the price is a weighted-median proxy rather than a hedonic index.

Summary statistics of the analysis variables

VariableRoleMeanSDMinimumMaximum
Tightness θ (key predictor)input3.601.731.0417.71
Search demand (enquiries)input52,399115,0501891,566,171
Listed supply (listings)input13,28123,11650327,575
Sold transactions (count)input3,3325,9501105,341
Sold price (£)input£225,238£84,857£72,327£563,293
Annual price growthdependent2.2%6.7%-32.9%21.0%
Price growth, one year aheaddependent2.0%6.9%-32.9%21.0%
Price growth, two years aheaddependent4.7%10.4%-48.0%28.5%

Grain: one observation per travel-to-work area and year, 2016 to 2025. Price and the growth outcomes are defined on the sale market; the forward-growth variables thin toward the end of the window as the horizon runs past 2025.

Method

A panel of areas by year. Price growth a set number of years ahead is regressed on current log tightness, with area and year fixed effects and standard errors clustered by area; the coefficient is read across horizons. A reverse regression checks that the lead does not run the other way, and an out-of-sample nowcast tests whether the signal lowers forecast error.