Valuation Methodology

We believe in complete transparency. Here is exactly how Estiq AVM calculates the estimated market value of your property in Estonia using verified, authoritative datasets rather than speculative portal listing prices.

1. Overview of the Automated Valuation Model (AVM)

The Estiq Automated Valuation Model (AVM) is a proprietary, machine-learning-inspired valuation engine designed strictly for the Estonian residential real estate market. By synthesizing millions of spatial, physical, and temporal data points, the AVM delivers near-instant, mathematically rigorous market estimates. It is built to serve property owners looking for quick price discovery, buyers seeking to verify asking price fairness, and real estate professionals aiming to onboard clients with objective, data-backed valuations.

It is crucial to understand what the Estiq AVM is **not**. It is not a certified physical appraisal (eksperthinnang / hindamisakt), nor does it involve a physical walkthrough of the property by a human appraiser. Instead, it is a highly advanced statistical model that calculates the most probable transaction price of a property under ordinary market conditions, serving as an invaluable baseline for decision making.

2. Authoritative Data Sources

A statistical model is only as credible as the data it ingests. Unlike portal scraped valuations that base calculations on asking prices (which are frequently inflated by 10% to 20%), Estiq relies entirely on confirmed transaction records and official state registries:

3. Step-by-Step Valuation Methodology

Every time an address is submitted, the Estiq AVM executes a precise, multi-step calculation pipeline in less than 200 milliseconds:

1. Address Resolution & Building Matching: The property address is normalized using the Estonian Land Board's ADS system (Address Data System). The exact building is matched in the Ehitisregister (EHR) database to pull structural parameters like construction year, primary building material, and total floors.
2. Comparable Selection Filters: The engine queries the database to find residential sales within a strict historical lookback window of 12 months. Potential comparables are filtered using a double-bound range:
   • • Area Bound: Comparable properties must be within **±15%** of the target property's square meters.
   • • Age Bound: Comparable buildings must be within **±5 years** of the target construction year (or fall into the same architectural building period, e.g. Soviet panel blocks vs. pre-war wooden).
3. Geospatial H3 Weighting (Gaussian Decay): The remaining transactions are mapped onto Uber's H3 hexagonal grid. Closer transactions (in the identical Resolution 9 cell) are given significantly higher weight. The weighting decreases smoothly as a function of geodesic distance using a Gaussian decay formula:
   Weight = e^(-(distance^2) / (2 * σ^2)) * e^(-(time_elapsed) / λ)
   This guarantees that a sale on the adjacent street carries ten times more weight than a sale on the other side of the municipal boundary.
4. Condition Coefficient Adjustments: Historical prices are adjusted based on the property's condition:
   • • Renovated (New / Very Good): Constant baseline.
   • • Satisfactory / Needs Repair: Receives a **-12% to -20%** price coefficient reduction to offset estimated renovation material and labor costs.
5. Energy Class Coefficient Application: Energy efficiency is a huge price driver in Estonia. The model applies a dynamic scaling factor based on the official EHR energy certificate:
   • • Class A–B: Receives a **+5% to +8%** premium.
   • • Class E–F: Receives a **-4% to -7%** adjustment due to increased future heating utility costs.
6. Floor Adjustments: Apartments located on the ground floor or top floor of multi-story buildings without lifts are adjusted by **-3% to -5%** based on historical buyer friction trends.

4. Confidence Intervals Explained

Every valuation generated includes a statistical **Confidence Interval** which measures the reliability of the estimate:

• High Confidence (Interval ±5% to 8%): Occurs in dense urban areas (like central Tallinn, Kesklinn, or Lasnamäe) where there is a high volume of highly similar comparable transactions inside the H3 cell in the last 6 months. Very reliable estimate.
• Medium Confidence (Interval ±9% to 13%): Occurs in suburbs or low-density cities (like Rakvere, Viljandi) where some comparable sales exist, but adjustments are wider due to variance in building age or area.
• Low Confidence (Interval ±14% or more): Occurs in rural counties, unique luxury properties (like Old Town Kesklinn penthouses), or areas with zero transactions in the last 12 months. Requires a physical inspection by a certified broker to verify market value.

5. Limitations & Disclaimer

While the Estiq AVM is incredibly accurate (±6.8% mean absolute error on urban apartments), users must respect its boundaries. It cannot detect structural damage, mold, unrecorded interior upgrades, or unique views (e.g. sea view vs. courtyard).

6. Open Data & Licensing Attribution

Estiq dynamically uses and acknowledges open data provided by the Estonian Land Board (Maa-amet), Building Registry (EHR), Statistics Estonia, and the Police and Border Guard Board (PPA). Traffic statistics and POI parameters are processed using OpenStreetMap contributors under the Open Database License (ODbL). Maps are rendered using CartoDB basemaps and LeafletJS engines under public spatial licenses.