How we measure accuracy
We test our model by predicting the past: scoring every district using only data that existed at a given date, then measuring how well those scores predicted subsequent price growth.
The test: out of sample prediction
In December 2013, we re-ran our scoring engine using only signals that existed at that date, with no data from 2014 onwards. We called this a pseudo-2013 snapshot.
We then asked: did districts with high scores in December 2013 actually see stronger price growth over the following 2 and 5 years?
This is an out of sample test. The scoring engine had no knowledge of what happened after its prediction date. Unlike backtests that use the same data to build and evaluate a model, this method directly mirrors real-world predictive accuracy.
Results
better than chance
Top-10% scoring districts had a 50.8% chance of being top-10% price performers over the next 2 years. Random chance would give 10%.
signal appears quickly
Spearman rank correlation ρ = +0.557 at 2 years, comparable to the 5-year signal (ρ = +0.554). The model identifies fast-moving areas early.
of weak areas declined
62% of bottom-decile districts saw price declines in 2022 to 2024, validating the model as a risk filter, not just a growth finder.
Score quartile vs actual returns (2013 to 2015)
Districts sorted by December 2013 score, split into quartiles. Average annual price appreciation measured over the following 2 years.
Annual CAGR (compound annual growth rate). Source: ONS HPSSA median prices. n = 2,518 districts.
Test methodology
Disclaimer: Past performance does not guarantee future results. Trajectory scores are statistical indicators derived from publicly available data, not financial advice. Property investment carries risk including loss of capital. Always seek independent financial advice before investing.
Want to understand how the score is built? Read the full methodology