How accurate is Bricks&Logic?

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.

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

Table 01 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.

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 (Table 2).

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.

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.

It’s all in the micro detail! We generate a uniquely evolving price estimate for each property depending on its location and type.
Chart 01 shows how properties in a similar location can move very differently over time.

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.