How Accurate is Bricks&Logic?

Back to articles
Apr 6 - 2020
6 min read

We provide millions of estimated prices, but just how accurate are they?

The UK Land Registry House Price Index

One of the major advantages of Bricks&Logic is that we are able to give you very specific data on how the estimated value of a property has moved over time depending on its location and type.
The only real comparison to this price estimate is provided by the UK Land Registry House Price Index which provides an index for every Local Authority

The Repeat-Sales Index

A traditional method of house price analysis is called the ‘Repeat-Sales method’ which looks at multiple sales of a single property to examine how the price changed over time. When aggregated across many properties, this creates an index summarising the aggregate change.

This index can be used to ‘predict’ property values when you have a previous sale and are waiting for the next one. How much did it sell for in the past, and what has been the index change for the area since then?

For this blog, we compared the Bricks&Logic forecast for a group of properties with forecasts generated by the Repeat-Sales method using Land Registry Local Authority indices for the same properties.

We looked at all sales in London postcodes between August and December 2019. There were 10,029 sales where both Bricks&Logic and the repeat-sales method using Land Registry data could produce a forecast.

How does Bricks&Logic compare to other analysis?

We found that the median forecast error for Bricks&Logic was 20% lower than forecasts obtained using Land Registry data (Figure 1.)

Figure 1 - A breakdown of the forecast error on 10,029 sales in London
Forecast Error RangeBricks&LogicLand Registry
< 10%56%48%
< 15%73%64%
< 20%84%76%

Figure 1 shows that using the Bricks&Logic forecast, 56% of the 10,029 forecasts were within 10% of the actual price at the time of sale. Using Land Registry data, only 48% of forecasts fell within 10% of the actual sale price.

Bricks&Logic confidence

Given the quality of our data and our ability to examine each property individually, we are able to define a confidence level for our forecasts. We split this confidence into High, Medium and Low and we are able to ask what confidence we had in our individual predictions at the time of the last sale (Figure 2.

Figure 2 - A breakdown of the error by confidence at the time of testing.
 B&L Confidence
Error SizeHighMediumLow
< 10%62%53%46%
< 15%79%70%62%
< 20%89%81%73%

So when our analysis produced a High Confidence property estimate, 62% of properties within the test period had an actual sale price within 10% of the estimated value.

Additional Coverage

Using the repeat-sales method has a further limitation, in that you can only forecast the price of a property if it has sold at some point in the past.
Bricks&Logic uses an attribute-based forecast so we can forecast the value of a property in any area that we have appropriate data for.
So, non-price data which helps us build the attribute model can then allow us to estimate the value in the absence of any historical sales data.
This allowed us to generate forecasts for an additional 30% of properties that the repeat-sales method could not.

What makes the Bricks&Logic forecast so much better?

It’s all in the micro detail! We generate a uniquely evolving price estimate for each property depending on its location and type.
We have used this to show that properties in a very similar location can move very differently over time.

Below (Figure 3) is a 10 year chart of the Bricks&Logic estimated value for two different types of property in Lambeth compared to the Land Registry Index.

Figure 3 - Property values over time (Lambeth, London).


By creating micro-level models for millions of individual London properties, Bricks&Logic provides not only the most comprehensive data on property descriptions and attributes but also demonstrably more accurate value estimates based on that data.

Try it today.