Technical Whitepaper

Kubera AVM Methodology

Circle Rate Anchoring, Stochastic Market Calibration, and Geospatial Regression for Automated Property Valuation Across 453 Indian Cities

1. Introduction

Property valuation in India presents unique challenges absent in Western markets. Each of India's 28 states and 8 union territories maintains its own registration portal with independently defined circle rates (also called guideline values, ready reckoner rates, or jantri rates depending on the state). These rates use different units (Rs/sqft, Rs/sqm, Rs/sq.yd, Rs/marla, Rs/katha, Rs/are), are updated at different frequencies, and often bear little relation to actual market prices.

The gap between government-notified rates and market reality varies dramatically — from near-zero in newly notified zones to 400%+ in premium urban localities. Kubera's central innovation is treating this gap as a measurable, calibratable signal rather than an obstacle: the circle rate provides a stable, legally anchored floor, and a proprietary market calibration model captures the locality-specific dynamics that drive actual transaction prices.

2. Three-Pillar Architecture

The Kubera AVM computes property value through three independent, verifiable data layers that are combined via a proprietary calibration model:

3. Circle Rate Anchoring

Circle rate anchoring is the foundational principle of the Kubera AVM. Rather than estimating property value from scratch using comparable sales (which are sparse, noisy, and often unreported in Indian markets), Kubera starts with the one data point that is legally guaranteed to exist for every locality: the state-notified circle rate.

We source rates directly from 20 state government registration portals, normalizing all values to a common unit (Rs/sqft). For states that publish separate rates for land and apartments, the AVM automatically selects the most appropriate rate type for the property being valued.

Zone-level spatial resolution is achieved through a point-in-polygon lookup against state-level GeoJSON files containing Survey of India village boundary polygons enriched with circle rate data. For any GPS coordinate, the system returns the exact government rate for that zone using a spatial index.

4. Stochastic Market Calibration

The market calibration layer captures the gap between government-notified rates and actual market prices. Indian property markets exhibit significant variation in this gap — a premium urban locality may trade at 3-4x the circle rate, while a newly notified area may trade close to it.

Kubera uses a proprietary stochastic model calibrated against verified real-world transaction data across 178 cities. The model employs log-space estimation techniques to remain robust against outliers and naturally captures the multiplicative structure of Indian property pricing.

For cities with insufficient market data, a hierarchical Bayesian fallback applies: state-level priors serve as the first fallback, with an all-India prior as the hyper-prior. This multi-tier approach ensures coverage across 453 cities — even in markets with limited data.

4.1 Zone-Level Adjustment

Within a city, market dynamics vary by zone. Premium zones (high circle rate relative to city median) tend to exhibit different market-to-government-rate ratios than peripheral zones. Kubera applies a zone-level adjustment that captures this intra-city variation using a non-linear regression model fitted on ground-truth data.

5. Geospatial Regression: 11-Factor Location Intelligence

Beyond circle rates and market calibration, Kubera scores every zone across 11 geospatial factors computed from real infrastructure data. These factors are amenity-proximity and infrastructure-density measures computed at the zone level.

Location factor weights are learned from ground-truth transaction data and formulated as dimensionless scores. This design ensures they transfer across cities without requiring per-city re-training — a critical property for scaling to 453 cities with varying data density.

6. Spatial Data Infrastructure

Kubera maintains a comprehensive spatial data pipeline built on Survey of India (SOI) village boundary shapefiles covering 26 states and union territories. Each village polygon is enriched with circle rate data from the corresponding state portal, creating a seamless spatial lookup system.

The spatial index uses an R-tree structure for efficient query performance. For any GPS coordinate, the system identifies the containing polygon, resolves nested zones, and returns the zone-specific circle rate — enabling real-time valuation at query time.

7. Validation & Accuracy

Model accuracy is evaluated using cross-validation across calibrated cities and validated against known bank valuations for specific properties.

The ~18% median error reflects the inherent uncertainty in Indian property markets, where transaction prices are influenced by factors not captured in public data (negotiation, undisclosed amenities, builder reputation). For context, professional human valuers in India typically achieve 10-15% accuracy, while portal-based “estimates” often exceed 30%.

8. Limitations & Caveats

• 1.Valuations are estimates for personal reference and do not constitute a registered valuation under RERA or any state registration act.
• 2.Circle rates may be outdated between government notification cycles (typically annual, but varies by state).
• 3.Market calibration depends on transaction data density; smaller cities with limited data use state-level priors with wider uncertainty bounds.
• 4.The model does not account for builder reputation, floor level, interior fit-out, or legal encumbrances — these must be assessed independently.
• 5.Under-construction properties are not separately modeled; the condition multiplier provides a coarse adjustment only.