Project
Land Alpha
An early-stage research programme testing whether fragmented public and local data can identify land-market transitions before those changes are fully reflected in market signals.
- Status
- active
- Started
- Source
- Private repository
Land markets may contain information inefficiencies because parcel, transaction, demographic, infrastructure, development, regulatory, and alternative-use information is fragmented and hard to synthesise. Land Alpha asks a narrow, testable question first: can machine learning identify places where meaningful development intensification is emerging, using only information that was knowable at the time?
Approach
The programme works down a hierarchy, from national and regional patterns to local development corridors, before it ever looks at a parcel. Two prediction systems are kept deliberately separate: Development Transition, the probability that a local area intensifies over the next five years, and Land Alpha, the forward excess return relative to comparable markets. The first must be validated before the second is attempted.
Current phase
The historical Development Transition baseline. The team has built the Florida geography, the land-cover truth layer, the point-in-time feature table, and a sequence of experiments: four baseline models tested out of sample (DT-001), a road-access catalyst test (DT-002), and a dated-event test using development applications against placebo events (DT-003). There is no price model, no parcel data, and no application yet, by design.
Research philosophy
Evidence beats narrative. Simple baselines precede complex models. Point-in-time correctness is non-negotiable. Show uncertainty rather than fabricate precision. Preserve raw source artefacts and lineage. Treat failed hypotheses as useful evidence.