Momentum signals: evidence and methodology
How we compute each signal, what the backtest shows, and why we show signals individually rather than combining them into a single score.
1. What we test and why
For each area (outcode), we compute a set of economic signals every quarter using only data that was publicly available by the end of that quarter. We then measure how well those signals predicted actual house-price growth over the following two years (8 quarters), using the rank correlation (Spearman IC) between signal percentiles and realised forward price growth across all areas.
Every signal uses a publication-lag offset so that no future data leaks into the signal at quarter T. The backtest covers every quarter from 2008Q4 (the earliest quarter where enough scored signals had history) through to the most recent quarter with 8 quarters of forward price data.
The purpose of this page is to let you judge the evidence directly. We have not smoothed the numbers or selected a flattering sample window.
2. Per-signal backtest results
Sorted by average rank correlation (mean IC) across all tested quarters. A positive IC means the signal pointed in the right direction on average; a negative IC means the opposite. The era columns separate the 2008-11 financial crisis period from the post-2012 recovery and expansion.
| Signal | Quarters | Mean IC | % correct |
|---|---|---|---|
| Rent growth | 28 | +0.117 | 100.0% |
| Sales volume | 72 | +0.040 | 72.2% |
| Migration pull | 36 | +0.036 | 36.1% |
| Crime trend | 45 | +0.021 | 57.8% |
| Affordability headroom | 72 | +0.011 | 52.8% |
| Supply pressure | 72 | +0.007 | 63.9% |
| Rail usage growth | 72 | +0.000 | 54.2% |
| Claimant rate trend | 36 | -0.022 | 41.7% |
| Possession stress | 72 | -0.031 | 40.3% |
| Planning activity | 9 | -0.037 | 22.2% |
| Deprivation shift | 14 | -0.104 | 0.0% |
| Sold-price momentum | 72 | -0.187 | 11.1% |
On price momentum during crises: the sold-price momentum signal (4-quarter rolling price growth) shows a negative IC in the 2008-11 period. This is expected behaviour: in a sharp market downturn, areas that had recently seen the strongest price growth tended to fall harder than laggards — the momentum reversal effect documented in property cycles. The signal recovers a positive (though weak) reading in the post-2012 era.
Note on polarity: Lower = better signals (crime, claimants, possession stress, supply pressure) are scored so that a high percentile means a favourable reading. The IC is measured after that orientation, so a positive IC means the signal correctly identified areas where conditions were improving.
3. Why we do not publish a single score
We built a combined score and tested it using the same methodology as the individual signals: for each quarter, we computed a weighted-average percentile across all scored signals and measured its rank correlation with 2-year forward price growth.
Before running the backtest we set a pre-committed acceptance gate: the composite score would be published only if it achieved a mean IC above 0.02 and was positive in more than 60% of quarters (excluding the seven COVID-disrupted quarters of 2020Q2 through 2021Q4, which were excluded from weight estimation because the forced-suspension of normal market activity makes them uninformative for signal weighting).
The gate failed. Excluding COVID quarters, the composite scored a mean IC of -0.003 and was positive in 54% of quarters. The individual signals show meaningful variation — some genuinely predictive, some not — but combining them into a single number removed that signal and added noise. We therefore show the signals individually so you can see exactly what each one is doing.
Composite score IC per quarter
Shaded quarters (2020Q2-2021Q4) were excluded from the gate calculation. Horizontal reference at 0 and +0.02 (the gate threshold).
Pre-committed gate (not met): mean IC > 0.02 AND >60% of quarters positive (excluding COVID). Actual result: mean IC -0.003, 54% positive. The composite score is retained for internal research purposes but is not shown as a product feature.
4. Publication lags
Each signal is offset by a lag so that the data assigned to quarter T was genuinely available by the end of T. This prevents look-ahead bias. A lag of 1 quarter means we use data from the previous quarter.
| Signal | Lag (quarters) | Reason |
|---|---|---|
| Rent growth | 0 | ONS PIPR published without material lag relative to quarter end |
| Sales volume | 1 | Same source as price momentum (Land Registry PPD) |
| Migration pull | 4 | ONS internal migration estimates published ~13 months after period end (4 quarters) |
| Crime trend | 0 | data.police.uk publishes crime data ~1 month after period end (rounds to 0 quarters) |
| Affordability headroom | 4 | ONS affordability ratios published annually the following March (4 quarters) |
| Supply pressure | 3 | MHCLG housing supply statistics published with approximately 3 quarter lag |
| Rail usage growth | 4 | ORR station usage statistics published approximately 4 quarters after the reference year |
| Claimant rate trend | 0 | DWP claimant count published same month as the reference period |
| Possession stress | 1 | MoJ court data published with approximately 1 quarter lag |
| Planning activity | 1 | MHCLG planning statistics published with approximately 1 quarter lag |
| Deprivation shift | 0 | Published as discrete editions (2015, 2019, 2025); forward-filled between publications (lag 0 at each release) |
| Sold-price momentum | 1 | Land Registry PPD registration typically lags transaction date by ~1 quarter |
5. A note on price-momentum measurement
The measured rank correlation for the sold-price momentum signal is additionally dampened by a structural overlap: the 4-quarter rolling price used as the signal and the 8-quarter forward window used as the outcome share a transition zone in the post-signal quarters, which mechanically compresses the IC toward zero in periods of mean-reverting prices, particularly during the 2008-11 crisis era.
Signal data is updated quarterly after each Land Registry PPD release. See individual area pages for current signal readings.
Browse areas table →