How Bricklane uses technology to predict rental income
Automating the process of comparative valuation
Hello, I’m Ali from Bricklane’s Data team.
Bricklane is building technology that aims to make investing in the UK residential property market easier and more efficient. This is achieved by Bricklane’s unique ability to purchase and service nearly any property in the UK, be it a full block housing estate, a farm house or a single urban flat.
Our approach is to construct portfolios of properties, which investors can earn returns from the rental income. Like other types of investment, a portfolio is measured by its returns. This is normally expressed as a yield — which describes the earnings for a given investment over a set period of time.
As you can imagine, yield is therefore an important metric to us. We track it for the existing properties in our portfolios to understand how they are performing. Importantly, we also try to predict it before buying a property.
This approach is common in the real estate industry. To calculate the yield, we need to know how much a property will cost to buy, and how much annual rent we will be able to collect for it. The first number is easy — the price of a property for sale is almost always on the listing advert. Finding the second number is more challenging, either the property isn’t to be rented out so it’s unknown, or if it is then the current rent isn’t published. In both cases, we have to predict it.
At Bricklane, a significant part of this process is driven by our technology. This article will describe the first of two ways we approach predicting rental yield.
Comparative valuation
Predicting the potential rental value of a property is achieved by creating a valuation for it. This is frequently performed manually by estate agents and private buy-to-let investors.
A manual valuation works by using the property you are valuing as a starting point (referred to as the subject) and finding similar properties nearby being advertised to rent. A similar property is one of comparable floor area, bedroom count, condition and age etc. Once you have 3 or 4 of these, the average rent is used as the prediction.
There are various problems with this manual approach. Firstly, it’s a time consuming exercise — try doing it for hundreds of potential properties in a given area!
Secondly, the “similar” properties you choose might not be that similar. One could have a nicer bathroom than your subject property, another could be in need of complete redecoration.
Automating valuations
At Bricklane, we’ve been building a database of listings and properties for several years. We used this to develop an automated valuation function, here’s how it works:
Our database has longitude and latitude points we can use to define property locations. Using this point for our subject as the centre, we find all the nearest properties that have been advertised for rent within a certain radius and match a similar set of features.
There are usually several results currently being advertised, however we also include historic listings, limited to a certain age. This is because we’ve found that rental prices move on a relatively slow, month-by-month basis — a 3 bed flat in Kings Cross advertised 2 months ago is likely to ask for the same rent if it came back on the market today.
Once we have the results, we take a median average of the rents to produce our predicted rental value. This is really fast — it takes milliseconds to generate this for a property, and not much longer to generate it for thousands of properties, or millions in Bricklane’s case.
Business impact
This is pretty powerful, by giving a rental yield estimate automatically and near instantly, it means our Property team can easily sort through thousands of properties and pick the best candidates to look at further.
Whilst the predictions won’t be perfect, the benefit of immediately screening a massive number of properties efficiently and accurately far outweighs the poor results for a small number of outliers.
When combined with other techniques to filter out properties that won’t produce good yields, it’s clear how technology can have a huge impact on being able to invest in residential property at scale.
This is possible due to both the knowledge Bricklane has built up, and the unique data sets we’ve created over time. With these, my colleagues and I are able to produce deep insights into the residential property market, and use these to drive our investment strategy and day-to-day operations.
Using machine learning
You might be wondering why we didn’t use machine learning to solve this problem straight away. I am a massive fan of approaching an analytical problem like this with a simple solution first, before using more sophisticated methods like machine learning algorithms.
A simple approach allows you to quickly build an understanding of the problem and its pitfalls. This means you are more likely to have a better understanding of the machine learning algorithm’s behaviour and performance when you build one. This point becomes even more pronounced if you aim to use a black box method such as Neural Networks.
You can read more about how I approach this valuation problem with machine learning in my next article, so stay posted and subscribe to the Bricklane Tech blog for all updates!
