Our methodology

1. Raw Data Ingestion

Rental listings are ingested directly from property management systems and first-party feeds. These systems represent the operational layer where rents are created, updated, finalized, and pushed to the public.

• Source systems are systems of record, not aggregators
• Data reflects active pricing and availability states
• Updates flow continuously as listings change

2. Schema Validation

Incoming records are validated against a strict schema to ensure structural integrity before processing. Records failing validation are excluded rather than repaired to prevent propagation of structural errors.

• Required field presence, such as rent, unit type, location, etc.
• Data type enforcement, such as numeric, categorical, temporal
• Referential consistency across unit and property identifiers

3. Data Normalization

Listings originating from different systems often encode the same information differently. Normalization standardizes these attributes into a consistent analytical representation.

• Unit type harmonization across SFR and multifamily
• Bedroom and bathroom standardization
• Geographic normalization to ZIP, city, MSA, and state hierarchies
• Reconciliation to actual U.S. Postal Service addresses
• Rent normalization to monthly effective values

4. Duplicate Resolution

Duplicate listings arise when the same unit appears multiple times across feeds, refresh cycles, or syndication paths.

• Identify candidate duplicates using property identifiers, geospatial proximity, and unit attributes
• Collapse duplicates into a single unit representation
• Preserve the most reliable observation for pricing and availability

5. Listing Reconciliation

When multiple observations exist for the same unit with minor differences, listings are reconciled rather than removed.

• Small rent changes within short time windows
• Slight timing offsets between system updates
• Overlapping listing states

6. Outlier Detection

Outlier detection identifies rent values that materially deviate from expected distributions given market, unit type, and time context.

• Data entry errors
• Misclassified units
• Transitional or non-representative listings

7. Median Adjustment

For repeated or conflicting observations tied to the same unit, values are adjusted toward the median rather than removed outright.

• Reduces leverage of extreme values
• Preserves sample size in sparse markets
• Maintains representative pricing signal

Median-based adjustment is preferred over mean correction due to the heavy-tailed nature of rent distributions.

8. Metric Calculation

Cleaned and reconciled data is aggregated to calculate core rental metrics.

• Median rent
• Rolling median over defined time windows
• Unit-level and market-level distributions

9. Trend Modeling

Trend modeling evaluates directional rent movement over time, operating on stabilized medians to ensure robustness

• Identifying sustained upward or downward movement
• Detecting inflection points
• Separating structural change from transient fluctuation

10. Seasonality Modeling

Seasonal patterns are modeled explicitly so they do not obscure underlying market dynamics.

• Normalization across months
• Comparison of equivalent periods year-over-year
• Isolation of cyclical effects from long-term trends

11. Signal Smoothing

Statistical smoothing techniques are applied to reduce short-term volatility while preserving meaningful price movement.

• Exponential moving averages
• Locally estimated scatterplot smoothing

12. Stabilization

The final step produces a stabilized dataset suitable for benchmarking, modeling, and institutional decision-making.

• Comparable across markets and time
• Resistant to short-term listing churn
• Grounded in real transactional pricing behavior