Empty Homes Part 2
This post looks at the big smoke and tries to get to grips with the city's dizzying house prices and also the idea that property is being bought as an investment object with no real desire to live in it. Although we do not yet have all the data for London, there appears to be evidence of housing being used as a commodity in central London. We use the Borough of Kensington and Chelsea as a case study, for which we estimate there is about £21 billion worth of low-use property. The tests indicate that the more expensive an area the more low-use property it has, that there is a statistically significant 4% difference between the price of low-use homes and the average house price and that luxury homes are 15% more likely to be low-use than would be expected. Overall the suggestion is that for Kensington and Chelsea at least property is perhaps being used as an investment object and not as a place to live.
Empty homes are a big discussion topic when it comes to housing, with many councils offering support for those wishing to bring derelict property back into use, as mentioned in the previous blog there are several hundred thousand empty homes in the country. This article however doesn't look at empty property due to being derelict or in areas of low desirability, it tries to re-frame the discussion slightly questioning whether we are asking the right questions. The concept of low use homes encompasses second homes so as to take a broader view of housing use. If a house is in a nice area and is used every couple of months for a long weekend is it empty? No technically its not, but is it in full occupation? No it isn't either, as such the low-use definition is helpful because it allows the inclusion of underused properties at all levels of the housing market.
Whilst there is a lot of talk and emotion about empty homes and foreign use of "buy-to-leave" it is difficult to really get data on how serious the problem is. Using the methods described in the previous blog post, this article will look briefly at London as a whole. However the main focus will be on Kensington and Chelsea which at over 10,000 low-use homes has comfortably the most of all the boroughs we have received data for. This blog post asks several questions,
- Is there a relationship between price and low use homes?
- Are there patterns in the buying of low-use homes?
- What does it mean for the housing market?
The previous article discussed in more detail about the technicalities of data collection, if some things seem a bit unclear check back to the last post and see if there is an explanation there.
The method is NOT essential for understanding the rest of the article, and some people might find it quite technical. The results are more exciting anyway, so feel free to skip to there if you want.
Defining a low-use home
This case study of Kensington and Chelsea requires a clear definition of what counts as a low-use home. We are using the discounts applied by the borough of which the following are included.
- Empty and Unfurnished: This is property that has been left empty for up to 2 years. Total in borough 1035.
- Empty Works: Property undergoing significant renovation and is currently not fit to live in. Total in borough 34.
- Job related : For jobs where the accommodation is provided and you perform your duties on the property, typically caretakers/ pub landlords etc. Total in borough 1.
- Levy: Properties that have been left empty for more than 2 years, also called long term empty. Total in borough 656.
- Moorings: A boat mooring that is unused. Total in borough 1.
- Second Homes: A second home, there are no minimum living requirements attached to this class. Total in borough 8551.
As this is all to do with homes, house prices are of keen interest. In order to put a value on things we have taken the Land registry data for 2016 and aggregated it to Ward Level. Ward level was chosen instead of LSOA due to the number of sales in a year per LSOA often being too low to provide much statistical validity. Using ward level you lose some granularity but gain quantity, for later work we may use multiple years but that can cause it's own issues as although then there is a large enough quantity, market growth can make comparing prices difficult. When we are calculating the value of the low-value housing we will assume that the housing value is the mean (average) that has been calculated for that ward unless otherwise stated.
The price class of a home
In this analysis we have broken up the market in to 5 price strata based on this definition by Savills, They describe 5 classes in the market: Lower Mainstream, Mid Mainstream, Upper Mainstream, Prime and Super Prime. The document linked is from 2012 so prices have moved around a bit since then (mostly up), what's more they give the prices in terms of £ per square ft. For simplicity we will use the average values as the cutoff which results in the class definitions shown below.
For reference the types of people Savills says live in each of the classes are as follows: Lower and Mid are young professionals, Upper are Downsizers, Prime is typically CEOs and senior bankers, Super Prime are the "International Super-Rich" (genuinely their description).
Exploring the data
We are going to explore three methods of trying to understand how low-use homes fit into the price landscape of London.
- Plot the correlation of ward level emptiness against the median house price of that ward.
- Assuming that homes have the mean value of their ward, do a significance test to see if there is a statistically significant difference between the average house price and the house price of empty homes across the whole borough.
- Using the housing price classification described by Savills, find out if the rates of low-use homes per class are what would be expected given the rates of homes per class in Kensington and Chelsea.
The technique known as Bootstrapping will be used in this analysis as the data is not normally distributed (follow a bell like curve). Bootstrapping is a useful technique as it allows you to make statistical inference without making too many assumptions. However we will be making a couple of assumptions so it is worth stating clearly what they mean.
- The distribution of the value of the sales in an area reflect the distribution of the values of the property in that area.
- The distribution of the value of the low-use homes is identical to the distribution of the value of homes in the area being measured.
The first point means that if 10% of the homes sold last year were £1,000,000 then 10% of all the homes are also worth £1,000,000. Of course this is not totally true but that is part of the magic of the bootstrap, which will generate a distribution round the mean giving an idea of the level of certainty we have. By making this assumption we can get an idea of the total value of the property in each area and an estimated value of the low-use home.
The second point means that all things being equal people will be buying low-use homes in a way that represents the property in the area. Again this means, if 10% of the homes in the Borough/ward/LSOA are worth £1,000,000 then we would expect that 10% of the low-use homes would also be £1,000,000.
Combining the first and second point we can then also say that the distribution of the value of low-use homes in an area should match the distribution of sales of homes in that area.
But how can you say that the distributions are the same? What if it's less? Isn't it dangerous to make such broad assumptions?
No it's not. In all analysis certain assumptions are necessary and although nothing is ever 100% right certain decisions are less risky. Assuming the distributions are the same as the overall is a much smaller assumption than saying it is more or less. What's more we can test to see if these assumptions seem reasonable or not.
The last part of this analysis uses a bootstrapped ratio, which needs a little bit of explanation. If we divide the expected number of low-use homes in Kensington and Chelsea by the sum of the expected number of low-use home in each of the wards, we will see whether the observed distribution of low-use homes does in fact match the distribution we'd expect if people were buying low-use homes evenly across the borough's price strata. However any apparent difference could be coincidence so we will bootstrap the ratio 10,000 times and see what the distribution of the ratio is.
If the expected and actual values are on average the same, the mean of the ratios will be 1 for each of the property classes. If not the ratios will diverge with some classes more than 1 and others less. If that is the case then the assumption that the distribution is the same will not hold and we will know which way it is skewed, which is pretty sweet.
Currently we have received usable data from 22 of London's 33 boroughs, It seems clear that the areas of highest low-use are in central London north of the river. It also seems to be the case areas that are more "desirable" (definitely not a technical term) are more likely to be low-use. A good example of increased low-use in desirable areas is Lambeth which doesn't have much low-use (although the data is still being quality checked as it seems surprisingly low). However many of the areas in Lambeth of highest low use are in new developments along the river such as the buildings behind the old GLA building and the St Georges development. The St Georges is interesting as the St George Wharf Tower, currently locked in a three way battle with the Shard and the BT Tower for the title of London's "Eye of Mordor", was featured in a Guardian article in 2016 for being under occupied. The 22 boroughs that have submitted data have 66,000 low-use homes, 43% more than the 46,000 homes scheduled to be built in London in 2017. This doesn't mean that low-use homes are causing a housing crisis but it does give a handle on how many there are and what they mean.
Map of Lambeth coloured by low-use
This image shows the percentage of low-use homes in Lambeth. Although the data seems to be very low, you can see that the top of the borough has the highest levels of emptiness running along the river.
Low-use homes in Kensington and Chelsea: A case study
The Royal Borough of Kensington and Chelsea (RBKC) has a royal amount of low-use homes. 10278 out of a total of 89280 are classed as low-use, totalling just over 11% of all housing stock. As can be seen in the map large proportions of RBKC have over 10% low-use with Brompton and Hans Town, the area around Hyde Park and Knightsbridge, has some LSOA with over 30 percent low-use homes. The north of RBKC is near the prison Wormwood Scrubs and has larger numbers of council houses and lower property prices, it also has lower numbers of low-use homes. This initial look at the data is similar to what appeared to be happening in Lambeth and London as a whole, which is that the areas that could be called desirable are more likely to have low-use homes than other areas. It should be noted that the waterfront at RBKC is not more low-use than other areas, so there are clearly no hard and fast rules.
Map of Kensington and Chelsea by low-use
So we have seen some basic patterns within RBKC but what of it? To start with it's worth trying to quantify the value of the low-use homes. The figure below left shows a heat map of the value of low-use homes in each ward, the values are in millions. What we can see is that RBKC has an estimated total of about £21 billion of low use homes, Brompton and Hans Town has £5 billion alone.
Looking further at the relationship between low-use properties and price the figure below right shows the correlation between price and percentage low-use. As can be seen there is a positive correlation of 0.67 indicating that wards with higher prices also have more low-use homes. We should stress that this is a correlation and does not mean that low-use homes cause higher prices or vice-versa. There are two notable outliers in the data, one that has an extremely high price and a middling low-use and the other that has a very high low-use. Removing these two outliers actually increases the strength of the correlation to 0.78
So far we have seen that RBKC has a great variation in the amount of low-use homes in each area, that there is a large amount of money invested in those homes and that there appears to be a positive relationship between price and low-useness (it's not a word, I checked). What is now interesting to know is given that there appears to be this price/low-use link is the average price of a low-use home more than the average price of a home and if so is it statistically significant? In short yes it is. The average price of a home in RBKC based on 2016 sales was £2,000,000 whilst the average estimated price of a low-use home was about 4% more expensive at £2,100,000. When a one-sided t-test was performed it showed that the difference was significant (p<0.01). However as the distribution was not normal we performed a bootstrapped test of the difference between the means in this case 100% of the 10,000 bootstrapped samples were larger than RBKC mean. With both the tests being positive (crucially the bootstrap test) we can conclude that the low-use homes are more expensive than regular homes in RBKC.
As a final test we took the bootstrapped ratio of homes in RBKC (explained in the method). By taking the ratio of the expected number of low-use homes per class when measuring at borough level and the estimated number of low-use homes per class when measuring at ward level we found that there was a clear skew. The figure below shows that there are about 15% less lower and mid ranged low-use homes in RBKC and about 15% more prime and super prime low-use homes than would be expected if low-use homes had the same distribution as the borough as a whole. What does this mean? Well it means that individuals are buying prime and super prime property without planning on using it as their main home at significantly greater rates than would be expected. The box representing super prime is much larger than the others and has much larger outliers as there are relatively few compared to the other categories (about 120 instead of thousands), this gives the result much more variability than the others as a change of plus/minus one house has around 10-30 times the effect it would have in the other categories.
At around 66,000 there are a lot of low-use homes in London and the value of these homes is enormous, Kensington and Chelsea alone has £21 billion. Looking at these raw figures it seems clear that the conversation about building new homes and bringing derelict homes back into use needs to develop to take account of low-use at the upper end of the market. At the beginning of the blog we had three questions to try and answer. The first question was about finding the relationship between price and low-use, we found a positive correlation 0.67 between price and low-use. The second question was whether there are any patterns in the low-use homes, using a bootstrapped test we found that low-use homes are 4% (about £100,000) more expensive than the average home in RBKC, and the difference was statistically significant. When we broke homes into price class we found that there were 15% more low-use homes at the luxury end of the market than expected.
The final question is what does it mean for the housing market? This question is harder to judge, it's unclear what is driving the high rates of luxury low-use homes, perhaps it's investment, perhaps it's prestige, or a mix of both. However it does seem that there is at least some degree of housing being used as a commodity. Given the patterns observed it doesn't seem unreasonable to believe that this would push up prices in prime and super-prime property forcing some of those who wish to purchase into a lower housing category resulting in a trickle down of rising prices. In addition the findings raise questions about the construction of new high-end housing developments. High-end property takes up more space but appears to have lower levels of occupation than housing at the other end, Although there may well be a demand for prime and super-prime property if occupation rates are lower it will not satisfy the demand for places to live.
Another question that is very important is, if the low-use homes are investment objects what happens if market conditions change and the investors try to liquidate their assets? Is it possible that a collapse of the luxury housing market acts as a lever for the market causing a collapse everywhere else? If so how much of the market as a low-use home is ok before it becomes vulnerable? with over 10% of homes being low-use RBKC would seem to be at risk if there was a shift in confidence and a desire to get out of the market. This question is definitely beyond the scope of this blog but given what has been observed it is perhaps a question we should be asking.
Low-Use and Democracy
In the general election this week, the electoral ward of Kensington flipped from Conservative to Labour for the first time. Although we are not suggesting that this change is due to low-use homes, many homes were low use before, it is still worth thinking about the effects on local democracy. We have seen that the more expensive areas tend to have higher rates of low-use homes and that poorer areas have higher occupancy rates. In local elections second-home residents can in theory vote in the locations of all their homes as long as they can argue they have a "considerable degree of permanence", according to the electoral commission. However it is likely that owners of low-use homes are less likely to vote in local elections and almost certainly won't vote in a general election. In areas with many low-use homes what does that mean? In marginal constituencies during a general election it could conceivably tip the vote. Whilst in local elections councillors in low-use areas could be representing significantly less voters than councillors in areas with more poorer residents, or even representing the wishes of people who don't really live there. This area is all kinds of difficult to unpick and could be analysed and discussed for ever, but given the very high rates of low-use observed and the unpredictability of recent voting, it might be worth exploring further.
The next post will focus on rural areas, examining if the issues we have seen in London are reflected in tourist destinations, such as St Ives (Cornwall), Windemere (Cumbria) and Aldeburgh (Suffolk).
Other ideas that can be built on are, what is the relationship between number of voters in each LSOA and the low use percentage; What is the relationship between low use homes and retail space? There may be a link or none, but it would be a move towards quantifying the effects and addressing the concerns of people with regards high densities of low use homes.
Unfortunately for this project We cannot release the data due to the licencing rules associated with the FOI (A blog post on this will be coming at some point). The code is available from github, it's kind of messy but it hopefully it is usable.