Analytical Valuation Model (AVM)
Our AVM is our monthly running evaluation of current property property values for private homes and properties in Denmark
Market leading model to estimate the value of private households
Our Analytical Valuation Model calculates the estimated property value of a specific house. It provides an estimated property value on more than 1.6 million houses in Denmark and places it on a performance measure, where 75% of the houses are estimated within an interval of +/- 20% of the actual value they are traded for. Our model consists of many data components ensuring the precision of the estimates, even if the property has not been sold in recent years
Our analytical valuation model
- ensures an independent estimate
- has evaluated and estimated the property value of Danish properties monthly since 2009 and has continuously been adjusted and optimised
- benchmarks against international standards
How our customers use our analytical valuation model
- As primary lookup to analyse property value or a segmentation criteria, e.g. in real estate, retail and insurance
- As supplementing lookup or as model validation, e.g. of risk modelling and credit evaluation in the financial sector
- As part of business analytics models
- For economic trend analysis in various areas, e.g. in retail
The underlying method for the estimate
The model is built around four model groups that weigh a house’s characteristics in combination with the general price development, comparable houses sold in the neighborhood, and the reduction percentage in the local area
The model builds on a vast amount of data that are collected in four models (A, B, C, D). These four models are eventually combined into the final model which calculates the estimated property value in a weighted price
What is not in the model
Our model is based on available register and statistical data, and sales data on properties sold on regular market terms. Living space not registered in the BBR and data on properties sold on other terms are therefore not included in the model
Model A
An expected sales price is estimated using linear regression
Model B
Based on smoothing and forecasting techniques, the expected sales price is estimated based on previous sales history or the price development in the geographical area
Model C
Through GIS techniques, local property sales data is identified, which is fed into advanced algorithms and a comparable sales price is generated
Model D
Properties currently for sale are estimated based on an expected reduction in the sales price. Model D uses the same techniques as that of Model B
The weighted price weighs the property’s characteristics (Model A) in combination with the price development (Model B), comparable neighbour houses sold (Model C), and the reduction percentage in the local area (Model D). Moreover, the weighted price also considers the historical prices for that specific property, as well as the general increase or decrease in property prices in the concerned postal code
Try yourself: search for an address and see the estimated value
Write the address on a property in Denmark and see how we estimate the address based on our model by visiting Dingeo.dk
Validation of estimates
To ensure the quality of our valuation and the model’s performance, the estimates are validated against the actual value which the property is traded for. For example, model estimates from 01-2018 are viewed against the actual trades in the period 01-02-2018 to 28-02-2018 and so forth
The validation is based on actual trades, where a marker is provided as an insurance that the data quality of the sales price is satisfying. For every sales price, we flag whether the sales price is valid or not based on the following criteria:
- the sales price is validated against the asking price, or the sales price is within 2 standard deviations of the average difference between sales prices and the public valuation of a specific property for a year
- trade date is larger than the year of construction
- the transfer method is ‘free trade’
The validation method does not discriminate on different geographical areas in Denmark, but rather it compares everything. There may be areas where the performance is notably higher than the average and correspondingly also worse
Precision flags
The model’s precision is monitored continuously and validated on a monthly basis based on real sales prices for houses sold in Denmark. We operate with three precision intervals:
Green
Within 0-10% deviation of the estimated price in comparison to the real sales price
Yellow
Between 10-20% deviation of the estimated price in comparison to the real sales price
Red
More than 20% deviation in the estimated price in comparison to the real sales price
Data about sold properties combined with the knowledge about data quality, availability, and model precision are further used to create a marker for our confidence in the model’s estimated market value. In this way, numerous properties will, beside being enriched with an estimated market value, also be accompanied by a marker for how confident we are on the price we have estimated
This marker is developed through an evaluation of our previous precision in our estimates on similar properties, according to type and size within the same postal code as well as the quality of the available data on the specific property. Doing this, every calculated market value will be accompanied by one of three categories for our confidence in the estimate:
Green marker
More than 70% of the sold properties in the groups of properties of the same type, size, and postal code had a deviation of less than 10% between our estimated value and the actual sales price
Yellow marker
More than 70% of the sold properties in the groups of properties of the same type, size, and postal code had a deviation between 11-20% between our estimated value and the actual sales price
Red marker
The rest of the properties, where the criteria for green and yellow are not satisfied