Summary

End-to-end DuckDB analytics on Land Registry data: SQL aggregates and baselines (national vs outward-area PMRs), Python for evaluation, and a clear pre-2025 / 2025 holdout split. 2025 mean absolute percentage error (national vs area): 17.51% vs 17.32%. Scope: This is a deliberate baseline—no hedonic features, a static PMR assumption, and coarse geography. It demonstrates reproducible SQL pipelines, honest out-of-sample metrics, and how far simple structure can go before you need richer modelling. Repository: github.com/bdevan5/property_prices