Probabilistic Machine Learning for Parcel Forecasting at National Scale

1. Problem

Most public-facing home valuation products collapse an uncertain path into one number. A single value is easy to display, but it does not communicate downside risk, upside potential, or forecast uncertainty.

Homecastr produces parcel-level forecast distributions: P10, P50, and P90 trajectories at 1- to 5-year horizons for U.S. parcels. Building this required solving several problems that shaped the architecture:

• Per-parcel calibration. Most probabilistic forecasts are calibrated in aggregate. Making an 80% prediction interval actually contain 80% of outcomes for each geography and horizon, across parcels ranging from $50K rural lots to $10M urban condos, is a harder problem. Miscalibration compounds across horizons and is difficult to detect without structured evaluation infrastructure.
• Cross-jurisdiction generalization. Every U.S. county publishes property data in a different schema with different identifiers, vintages, and suppression rules. Building one model that generalizes across these sources required a canonical panel design that absorbs schema heterogeneity at the data layer rather than the model layer.
• Regime sensitivity. The model learns transition dynamics from historical data. When the current macro environment diverges from training history, as during the 2021 to 2022 rate shock, forecasts can degrade silently. The system needs to detect this and flag it rather than serve overconfident yet inaccurate distributions.
• Trajectory coherence. Forecast paths need to look like realistic extensions of a parcel's historical curve, not flat or linearized extrapolations. Independent per-horizon regressors produce trajectories that lack the curvature, momentum, and volatility present in actual price histories. This motivated a generative architecture that samples jointly coherent multi-year paths.

2. Why Naive Approaches Fail

The challenges in Section 1 ruled out simpler approaches. Each limitation below directly motivated a component of the current architecture (Section 4).

Point Forecasts Miss the Distribution

A point forecast cannot tell a homeowner whether their downside risk is 5% or 25%. Post-hoc prediction intervals wrapped around a point estimate assume symmetric, fixed-width uncertainty, but real property value distributions are heavy-tailed and vary by geography and price segment. Calibrated intervals require learning the distribution directly, which motivated the diffusion decoder described in Section 4.

Independent Horizons Produce Incoherent Paths

Training separate regressors for each forecast horizon (year 1, year 2, through year 5) produces trajectories that are not jointly coherent: a parcel might show 10% growth in year 2 but only 3% cumulative by year 3. A stochastic process model generates paths that respect temporal dependencies and look like realistic extensions of observed price histories. Critically, this enables the model to forecast corrections - growth in years 1-2 followed by mean reversion in year 3 - which independent horizon regressors cannot represent.

3. Data System

The pipeline ingests county assessment rolls from multiple jurisdictions, including Texas statewide via TxGIO, Florida statewide via DOR, Harris County via HCAD, and New York City via DOF RPAD. These are joined with ACS census-tract demographics and FRED macroeconomic series covering mortgage rates, CPI, unemployment, and other indicators. All sources are standardized into a canonical parcel-year schema that absorbs differences in identifiers, vintages, and suppression rules at the data layer. Source-specific characteristics and schema details are documented in the technical note.

Transformed features are stored across PostGIS/Supabase (geospatially indexed), GCS (Parquet bulk storage), and Redis (low-latency feature retrieval).

Figure 1. End-to-end system architecture. Source data flows through a canonical panel into the model stack. The FT-Transformer produces a trend estimate; the Diffusion Decoder generates calibrated distributions conditioned on both the backbone output and spatial token paths. Results stream to GCS as a write-ahead buffer before bulk upsert to Supabase.

4. Model Architecture

Why a Generative Architecture

The failure modes described in Section 2 each map to a specific architectural response. Incoherent horizons motivated joint trajectory sampling via the diffusion decoder. Missing distributional shape motivated learning the full predictive distribution rather than post-hoc intervals. And unobserved neighborhood dynamics motivated the inducing token mechanism.

Current Production Architecture

The current system has three components, trained jointly. An FT-Transformer backbone [1], a self-attention encoder over heterogeneous tabular features, outputs a deterministic trend prediction and a context embedding. Shared inducing tokens [2] with learned persistence capture neighborhood-level dynamics that are not identifiable from parcel features alone. A diffusion decoder [3] using DDIM sampling [4] generates the full predictive distribution conditioned on the backbone's trend estimate and the inducing token paths.

The model predicts year-over-year log-growth changes rather than absolute levels, which stabilizes learning across parcels with different price scales. The diffusion loss uses min-SNR weighting [5] so that the model learns distributional shape rather than collapsing to conditional means, and is normalized independently per horizon to prevent forecast attenuation at longer lead times. Architecture details, equations, and training diagnostics are in the technical note.

AI-Native Interaction Layer

Traditional real estate platforms rely on rigid filters and map bounds. Homecastr surfaces its probabilistic outputs via an AI-Native Interaction Layer designed for both human and agentic consumption:

• Semantic Omnibar: A natural language gateway that parses complex queries (e.g., "Show me areas in Austin with >15% upside and highly-rated schools"), maps them to neighborhood geometries, and instantly retrieves the relevant forecast distributions.
• Model Context Protocol (MCP) API: Homecastr operates as an MCP server, exposing forecast distributions, appreciation outlooks, and comparable market data as standardized tools. This allows external reasoning models (like Claude or custom agentic swarms) to pull raw probabilistic context directly into their context windows for advanced downstream analysis.

5. Evaluation

We evaluate each model candidate on genuinely held-out origin years using an expanding-window protocol. Macro features are lagged by one year relative to the origin to prevent look-ahead contamination.

Calibration Diagnostics

Before a model candidate enters production, it is evaluated against the diagnostic targets below [6]. These thresholds are not strict pass/fail gates. Actual results routinely fall outside these ranges, particularly during regime breaks. They serve as reference points that guide development decisions rather than automated acceptance criteria. A model that improves point accuracy but severely degrades calibration does not ship. These same diagnostics are re-run continuously as described in Section 7.

We classify calibration failures into three common cases. Forecast attenuation: step deltas drop to near-zero beyond the first year, producing flat trajectories. Miscalibrated dispersion: prediction intervals are too narrow (overconfident) or too wide (uninformative) to match realized outcome rates. Mean bias: systematic over- or under-prediction, typically caused by regime drift. Each case has a separate diagnostic and a corresponding remediation path.

Backtest Results

Each row reports the longest verifiable horizon for that origin year — the furthest point at which we can compare the model's forecast against realized values. MdAE is the median absolute percentage error of the P50 forecast versus realized values. Model Wins shows the fraction of parcels where the model beats a naive persistence baseline. Bold values meet horizon-adjusted diagnostic targets (MdAE < 10% × √h, Coverage 65–95%, Wins > 50%). Full per-horizon breakdowns are in the technical note.

6. Serving

The serving layer separates three concerns:

7. Monitoring

Continuous Calibration

The calibration diagnostics from Section 5 are not one-time. As new actuals arrive, the same tests run against the production model to detect calibration drift. If interval coverage, tail accuracy, or horizon scaling degrades below its target threshold, the model is flagged for retraining.

The pattern of which tests fail is diagnostic. Coverage dropping while anchor integrity holds suggests miscalibrated dispersion. Anchor integrity failing suggests mean bias from regime drift. Horizon scaling failing while short-term metrics hold suggests forecast attenuation. Each pattern maps to a different remediation path.

8. Limitations

Appraisal vs. transaction prices: The model trains on county-appraised values, not recorded sale prices. Appraised values can lag market movements, smooth over short-term volatility, and diverge from actual transaction prices in rapidly changing markets.

ACS self-reporting: Census tract demographics from the American Community Survey are self-reported survey estimates, not administrative records. They carry sampling error and non-response bias, particularly in small geographies and hard-to-reach populations.

Data coverage: Not all U.S. jurisdictions publish property assessment data in machine-readable formats. Prediction intervals are wider in data-sparse regions because the input signal is weaker.

Regime sensitivity: The model learns transition dynamics across its training history. Sudden macro regime breaks (e.g., the 2021 to 2022 rate shock) can cause systematic bias when the current regime diverges significantly from historical patterns.

Attribution gap: Feature attributions are computed from the deterministic backbone, not the full stochastic pipeline. There is a gap between the sum of explained drivers and the calibrated trajectory.

9. Roadmap

References

Interested in the engineering challenges behind Homecastr? Connect on LinkedIn · About the team