Free online reading
Table of Contents
2.1. Problem Statement
2.2. Research Objectives
2.3. Key Limitations
2.4. Research Strategy
3. Literature Review
3.2. Overview of North American Research
3.3. The Use of Regression Analyses
3.3.2. ‘Macro’ Factors
3.3.3. ‘Micro’ Factors
3.4. Research into the London Build to Rent market
3.4.1. Research Reports
3.4.2. General Discourse - UK
5.1. Sample Area
5.3. Data Collection
6. Research Results
6.2. Interpreting the Results
6.3. Key Findings
6.3.1. Amenity Spaces
6.3.2. Stepwise and Best Subsets Analyses
6.3.3. Hedonic Pricing Model
6.3.4. Other Findings
7.1. Further Research Required
Table of Figures
Figure 1 Results of Knight Frank Tenant Survey (2017)
Figure 2 Build to Rent good practice guide from Trust for London (2017)
Figure 3 Sirman et al.’s most recurring characteristics
Figure 4 Hypothetical relationship between sugar consumption and obesity
Figure 5 Distribution of t-values
Figure 6 Results of multiple regression analysis
Figure 7 Number of occurrences of amenities within the database
Figure 8 Best Subsets results
Figure 9 Stepwise results
Figure 10 Hedonic Pricing Model results
This paper analyses the effects of including additional amenities in London’s Build to Rent buildings on rental values. Three amenities were analysed: concierge services, gymnasiums and residents’ lounges. A total of 68 Build to Rent buildings across Greater London were analysed and 149 datasets were collected. The research is thought to be of importance due to the increasing number of Build to Rent buildings under construction and in the development pipeline in the capital, many of which will include onsite amenities. A review of academic literature revealed that the effects of including onsite amenities on rental value has not been determined for the London market. It also found that some consultants, architects and commentators are making unsupported suppositions as to the values that these amenities add, usually by way of reference to the US and wider North American markets, where ‘multifamily housing’ has been established for some decades.
A multiple regression analysis was undertaken and, from this, a hedonic pricing model was produced. This methodology follows a well-established route which has been widely employed across the Atlantic and in the Asian property markets, but does not appear to have been applied in any academic capacity to London or the UK. The analysis seeks to compare and contrast the different contributory factors which comprise the overall rental value of a property, thereby attributing an individual value to each factor. The findings of the research were that amenities had no statistically significant impact on rental values, all other things considered. The research results also highlighted the importance of restricting further analyses of a similar nature to smaller submarkets, as well as proving the efficacy and utility of the multiple regression methodology.
The following research has been undertaken to address a question which is often asked of me in my professional capacity as a Build to Rent property consultant and, more importantly, to fulfil the graduation requirements of the MSc Real Estate Development course at the University of Westminster.
The research project was initially inspired by the various consultants, architects and Build to Rent commentators whom I encountered speaking at conferences, writing in property industry magazines and advising on projects with which I was involved. It seemed that without exception, it had been taken as read that including amenities in buildings would increase rental values. Through my own initial research, I was unable to find any robust analysis or academically-reviewed evidence that supported these claims.
My discussions with my dissertation supervisor, Christopher Pittman – to whom I am very grateful for the advice, insight and support he has given – confirmed that this was a worthwhile avenue of enquiry.
The research that has been undertaken over the past 12 months has been time consuming and the writing of this dissertation has been more so; I am therefore also grateful to my partner, Alexandra McGregor for the patience, support and love that she has shown in this somewhat stressful period.
Further thanks are owed to my two bosses, Charles Seifert and Debra Yudolph, whose kindness and generosity has made the pursuit of my higher education possible. Their continuing belief in and dedication to mine and my colleagues’ professional development is truly exemplary.
Finally, I also owe a debt of gratitude to my parents, friends and family whose wise counsel and kind words have always, and will surely continue to, serve me well.
At present, little is known in respect of the value that onsite amenities add to Build to Rent developments in London. It seems, however, that much is assumed.
As a fledgling sector, Build to Rent (or ‘PRS’) has presented an opportunity for developers and consultants to make unsupported suppositions regarding the extent to which amenities drive value without recourse to peer-reviewed academic literature. Examples of this include publications by Jeanette Veldkamp (Veldkamp, 2015) of Chapman Taylor who states that added amenities are integral to the offering of Build to Rent buildings and “are part of the PRS life-style formula”, as well as Artur Carulla of Allies and Morrison Architects (Carulla, 2014) who identifies the inclusion of “congregational spaces” as the design measure most likely to add value to schemes.
As noted above, the issue with statements such as these is that there is a deficiency of valuable, usable and reliable research relating to the London Build to Rent market from which to draw such conclusions. Research into this area of study has been undertaken in the USA and Canada since the 1970s (see, for example, Sirmans and Benjamin, 1989), but there does not appear to be a single academic publication examining the London market. However, it is important to note that a number of commercial publications dealing with this question have been published in the last two years and these are discussed further in the Literature Review below.
The question is an important one for developers, investors, operators and lenders. Research undertaken by Sam Long at Molior (Long, 2016) found that the completion of new-build units in London “increased by more than 50% over the first three quarters of 2016 – almost 2,900 Build to Rent units completed between January and September and half of those were in Q3 alone.” The appetite for Build to Rent development is increasing amongst institutional investors such as M&G and Legal & General, as well as developers, operator-developers such as Essential Living, Fizzy Living and Go Native, and housing associations such as Notting Hill and A2 Dominion; the same report by Long (Long, 2016) found that 8,799 Build to Rent units were under construction in 86 developments across London in November 2016.
Furthermore, with the recent publication of the government housing white paper, ‘Fixing our broken housing market’ (CLG, 2017a) and the potential implementation of the policies outlined therein, it appears that local authorities may be actively incentivised to encourage the delivery of Build to Rent products in the near future.
The significance of this is that many future Build to Rent schemes are likely include at least one amenity provision, such as a gym, residents’ lounge or concierge service, yet investors and operators are not aware of the value that such provisions will bring. Indeed, this presents a very different proposition to the Build for Sale market; not only will the operators and investors be liable for paying the running costs of these schemes, but it will directly impact their return on investment. Unlike a scheme that is built to be sold where variable service charges will cover the running costs of such amenities, investors are required to examine the cost-benefit of including amenities to ensure that they are maximising their return.
In short, common wisdom is that added value can be created in Build to Rent schemes through the inclusion of amenities. However, there does not appear to be any reliable academic research supporting this viewpoint. The data to draw such conclusions is obtainable and available, however it seems that this very important question, which could affect the direction of the infant sector, remains unaddressed. A quantitative study analysing the relative rental values of properties with and without such amenities would go some way to answering this question.
2.1. Problem Statement
The expressed aim of the research is to analyse the value, if any, that different amenities add to Build to Rent schemes in London with a view to determining the impact on rental values for each amenity type. By the conclusion of this research, it is hoped that a percentage uplift and an average increase in rental vales can be determined for each amenity type. As a secondary research project, this information could be interpolated with the average cost of running the amenities in order to reach a conclusion as to the cost-benefit of their inclusion.
The key question to be addressed is: ‘do amenity spaces increase rental values in Build to Rent schemes in London and, if so, to what extent?’
2.2. Research Objectives
The first stage is to determine the limitations of the current literature addressing this subject. The literature review at chapter 2 below goes some way to completing this objective.
The second objective is to determine the rental value of a large selection of new-build schemes across London. It is thought that over 138 datasets will be required and this will also include information regarding location, amenity offering, specification and a number of other factors, which are further detailed at Chapter 3 below.
The third objective is to input the data into a multiple regression model in order to determine the relative value of each factor. The model is designed to compare and contrast the features of each building in order to identify correlations in value. Once enough data has been provided, the model is able to place a percentage or monetary value on each factor and establish a degree of confidence.
2.3. Key Limitations
It is also important at this stage to identify the anticipated key limitations of a piece of research such as this. Perhaps the greatest limitation is the scope of the model. While it is possible to identify a large number of variables which effect the rental value of a property, it is not possible to identify them all. Some variables are impossible to quantitatively account for, such as attractiveness of the branding, skill of the letting agent and quality of marketing.
For the purposes of this research, it is assumed that all properties are let to a willing buyer by a willing seller and both parties are not subject to any extenuating circumstances, as well as in possession of all information at the transaction date. Clearly, in the real world, this is not always the case and the model is not able to account for such eventualities.
Another limitation which relates particularly to the cost-benefit analysis is the failure to account for the capital cost of installing such amenities; rather the cost-benefit analysis will deal only with the ongoing running costs.
One key limitation of this research is the fact that there is a distinct lack of relevant literature addressing the topic in question. This does present an opportunity to undertake an original piece of research, however it deprives the conclusions of the research of a robust, supportable context within which to fall. In turn, this diminishes the strength of the study as it can only be viewed in isolation, rather than as part of a wider body of scholarship.
Finally, this research does not seek to determine whether amenities increase the likelihood of tenants staying in the Build to Rent scheme. Decreasing turnover of tenants is a primary consideration for all Build to Rent operators and investors as it maximises profit by minimising rent loss and void liabilities (BPF, 2015).
2.4. Research Strategy
As discussed and outlined above, the research pertaining to the value of amenities in London PRS schemes is very limited and, in order to find an answer to the proposed research question, it will be necessary to collect new and relevant data. The study to be undertaken will, therefore, be a quantitative study.
The use of multiple regression analyses and hedonic pricing models is widespread in real estate academia. Monson (Monson, 2009) explores the necessary considerations and approach to be taken with this model. The model provides a means of separating out the individual components that comprise the rental value of any particular property through drawing correlations between schemes sharing the same characteristics.
Monson highlights the various perceived failings of traditional property valuation techniques, such as discounted cash flow valuations, the weaknesses of which had been highlighted at the time of writing his paper in 2009 by the recent worldwide property market crash. Monson (2009, p. 62) points to broad-based assumptions such as discount rates, capitalisation rates and market cycles, the failure to accurately model which played a part in increasing “irrational exuberance fuelled by inexpensive capital chasing deals […] as a result, the market value of real estate far outweighed the true, or intrinsic, value”. Monson proposes the techniques of multiple regression analyses and hedonic pricing models as viable alternatives to the traditional techniques and suggests that they worthy of serious consideration.
By way of introduction to the methodologies of regression analysis and hedonic pricing models, Malpezzi, Ozanne and Thibodeau (1980) provide a good analogy by comparing housing to a basket of goods. The basket of goods may be small or large and is made up of a variety of different products, each of which contribute to the overall value of the basket. In the words of Monson (2009, p. 63), “By collecting information about many different bags and the items contained therein, regression analysis can be used to determine the contributing affect (or correlation) each of those items has on the overall price. This technique also helps to determine which item(s) most significantly impact the price”
According to Malepezzi et al.’s theory, the valuation of real estate is no different and by collecting information about the different components which make up a property’s intrinsic rental or sales value, the same methodology can be applied to calculate the value of each of those components. These components are either characteristics of the property or building (B), or characteristics of the location (L). The final price (P) can therefore be stated to be a function of these variables, as well as any unknown quantities (x). The formula for calculating the final sales or rental price of a property is therefore:
illustration not visible in this excerpt
This is useful for the purposes of this paper as it allows for the isolation of single characteristics and, provided that enough data is inputted into the model to result in a statistically significant result, a value can be ascribed to each of the characteristics analysed.
Thus, the strategy for this research will be based on the collection of sufficient data and sufficient input variables to account for all characteristics that comprise rental values. The main two items requiring consideration in respect of the research strategy are:
(i) rental and property data collection; and
(ii) construction of the model
The exact number of variables required to reach a statistically significant conclusion is, theoretically, a mathematical question as well as a valuation question. It is often speculated that the valuation of property is both a science and an art form (Warren-Meyers, 2015), however there are a number of factors which are universally accepted to contribute to the overall price: number of bedrooms, number of bathrooms, condition, age, etcetera. Malpezzi (2002), through his research into hedonic pricing models, has drawn up a list of variables considered to be good practice for the inclusion into any future models. This has been used primarily as the guidance for the collection of data and is discussed in greater detail in Chapter 3, ‘Methodology’.
Unlike house prices, there is currently no open-source, centralised, readily-accessible source for rental prices in the UK. Various companies collect this data, such as tenant referencing services (HomeLet, ReferenceMyTenant, etc.) and letting agents (Foxtons, Carter Jonas) but either they use this for their own internal analysis or do not put it to any statistical use. HomeLet do publish a monthly rental index using their data, but the individual data points setting out the rent, type of property, number of bedrooms, age of the building and any amenities, is not available to the general public.
Rental data can be collected ‘manually’ by conducting telephone interviews with lettings agents having first identified which schemes in the sample area are will contain properties that are useful for the analysis; in this instance, build to rent schemes in London, particularly those with amenity spaces, will be required. This method will ultimately result in enough data being collected to be able to draw an analysis but has the drawbacks of being potentially unreliable, vastly time-consuming, and time sensitive. This last point is pertinent; the London rental market is relatively turbulent at present and therefore the data must be collected quickly and within the same period to ensure that the effects of market forces on the results are negated.
HomeLet’s (July 2017) aforementioned Rental Index shows that between July 2016 and July 2017, the average rent in London dropped by just £4 per calendar month from £1,568 to £1,564. This statistic alone belies the fluctuations in the interim period where average rents dropped to £1,508 in December 2016, £1,497 in January 2017 and steadily climbed back to £1,546 in March before dropping again to £1,502 in May. Therefore, rents could have varied by up to 4.5% depending on which point in the year data is collected. Clearly, the fundamental characteristics of the property – and therefore its intrinsic value – will not have changed during that period, but the forces of the market will have acted to drive prices up or down. In order to eliminate the potential for the results of this research to be skewed by such forces, a ‘snapshot’ of rental prices is required.
Some property research companies, although mostly not in possession of verified rental data in the same way that referencing companies and letting agents are, do provide up-to-date information on achieved rents for schemes across London. The most comprehensive collection of data appears to be Molior’s ‘Build to Rent Database’. The database provides rental values for all Build to Rent buildings in London with 20 or more units, as well as detailed information on amenity provision, size and specification of the units and location.
It is important to take into account four main considerations when using Molior’s rental data for an analysis such as this:
(i) Molior do not update their data on a monthly basis. Typically, Molior will gather all information in May and then update it, adding rental data for new schemes, every year. This means that at the time of writing, there may be information available for more Build to Rent schemes in London that has not been included in the analysis. It also means that the results of this research can be considered to be dated May 2017, rather than August 2017.
(ii) Molior collect their information via telephone interviews with lettings agents. This does have a number of drawbacks, particularly the issue of reliability which appears more pertinent when one considers the subscribership of Molior. Molior’s website homepage claims that 95% of the “top residential developers and agents” use the service. Given that one letting agent is typically awarded a contract by a Build to Rent developer for all of the lettings in a new scheme, it is notionally in the agent’s interest to make it seem that high rents are being achieved in any scheme for which that they have responsibility for the lettings so as not to jeopardise the prospect of winning future business from London’s “top residential developers”.
(iii) Due to the infancy of the Build to Rent sector, there are only a relatively small number of schemes from which to gather data. As discussed further in Chapter 3 below, 68 schemes across Greater London have been used, but multiple data points from each building have been collected. This may have obvious consequences on the results but, on the basis that Molior provide the most comprehensive database of Build to Rent schemes, is unavoidable.
(iv) The rental values provided in Molior’s Build to Rent database are building-specific and not property-specific. This means that a certain element of data ‘richness’ is lost and the analysis is unlikely to perform as well as may be hoped. The reasons for this are further elucidated below however, in summary, factors such as floor area, the floor on which the apartment is situated, and furnishing cannot be accounted for, despite the fact that they are likely to influence the rental value.
Despite these drawbacks, Molior presents the best-quality, most comprehensive and widely-trusted database of rental values which is currently available for schemes in London’s Build to Rent sector.
In order to analyse the data, a statistical software package is required. Microsoft Excel does provide the user with the option of some statistical tools through the Data Analysis plug-in. However, this software is somewhat basic and does not allow for more advanced analyses such as hedonic regression. A specialist software package such as Minitab or IBM’s SPSS should therefore be used; the former has been selected for both ease of use and familiarity. Once the data has been collected, the analysis is as simple and instantaneous as selecting the type of regression analysis to be undertaken from a drop-down menu. The interpretation of the results requires some understanding of the outputs, as will be further discussed at Chapter 4.
3. Literature Review
There is little to no academic, peer-reviewed research addressing the drivers of value in in the London Build to Rent market, particularly in respect of amenity spaces. One plausible explanation is the relative infancy of the Build to Rent sub-market.
The British Property Federation (2015) refer to Build to Rent as “the UK’s newest housing sector” and, along with other notable industry players, have recently campaigned to have Build to Rent recognised as an independent planning use class. To this end, the government have recently consulted with the wider industry (Department for Communities and the Local Government, 2017b), albeit unsuccessfully. It is reasonable to surmise that due to the youth of the sector, academic research has not yet been considered viable, desirable or relevant to undertake.
Conversely, in the USA, the Build to Rent market is far better established and it is known as ‘multifamily housing’. Colliers International (2016a) identified the US multifamily sector as “among the hottest asset classes through the post-financial crisis recovery” and found that gross yields averaged 5.6% in Q3 2016, the lowest of any property type and reflective of strong investor confidence in the market.
Doan (1997) identifies the years 1971-3 as the inception period for the asset class in the USA, during which 966,000 units were delivered. Across the Atlantic, interest in Build to Rent, both commercial and academic, has been strong for some 45 years now and this is reflected in the output of academic literature, some of which attempts to address the key question at hand.
3.2. Overview of North American Research
There is a wealth of research devoted to rental values in multifamily buildings and the factors that influence the same; many of the research papers cited below make use of regression techniques. For example, the work of Sirmans et al. (1989) demonstrates the wide range of factors that influence rental values of multifamily flats in Lafayette, Louisiana including basic factors such as the number of bedrooms, to amenities such as a swimming pool ($14 per month), covered parking ($50 per month) and included utilities ($32 per month). In turn, this research draws on the work of Guntermann and Norrbin (1987) which examined the impact of number of bedrooms, size, location and amenity provision on the rental values of houses and flats in Phoenix, Arizona.
Other research has also been undertaken by Marks (1984) who, in the course of analysing the effects of rent control on the price of housing, found that apartments in Vancouver, Canada, were substantially dependent on on-site amenities such as covered parking and laundry rooms, as well as situational amenities such as distance to transport and the age of the building using a hedonic index.
More specialised research into just one or two variables which affect rental values or house prices has also been undertaken. For example, Li and Saphores (2011) built on the research of others such as Des Rosiers et al. (2002), Kestens et al. (2004) and Donovan and Butry (2008) to establish that increases in urban tree canopy cover and local green spaces did not significantly impact rental value in multifamily housing.
3.3. The Use of Regression Analyses
At this stage, it is worth considering the prolific use of regression modelling to inform analyses on factors affecting the sales and rental values of properties. Much of the body of research is devoted to the former, but the same principles apply when considering its application to the latter.
The groundwork for the analysis of the value of goods using a regression model was laid by the work of Court (1939), an economist for the Automobile Manufacturer’s Association of Detroit. Court was interested in price indices for vehicles but criticised contemporary procedures as insubstantial, noting that “Passenger cars serve so many diverse purposes that such a single, most important specification can not [sic] be found like rated tonnage in the case of trucks. The simple method is inapplicable, but why not combine several specifications to form a single composite measure?” (p. 107). In simple terms, Court combined the different factors that make up the characteristics of any vehicle into a statistical model to produce a single index of “usefulness and desirability” (Goodman, p. 292).
This work was built upon by the economist Zvi Griliches in the early 1960s (Goodman, p. 291) who popularised the methodology along with Kevin Lancaster. Lancaster – as a result of his article “A New Approach to Consumer Theory” in the Journal of Political Economy (1966) – is often attributed the honour of having developed the theory that the price of a good is a product of its individual, inherent characteristics. These characteristics can themselves be valued and the overall price or value of any good is therefore the sum of its contributory components. His work did apply this theory to housing, as well as to broad, macroeconomic topics such as the demand for money and financial assets (Corsini, p. 5).
Further research was undertaken by Sherwin Rosen (1974, p. 35), who was able to conclude that “[w]hen goods can be treated as tied packages of characteristics, observed market prices are also comparable on those terms. The economic content of the relationship between observed prices and observed characteristics becomes evident once price differences among goods are recognized as equalizing differences for the alternative packages they embody”. Put simply, his research was able to determine that it is statistically and theoretically robust to assume that the individual characteristics of goods can be priced and that the total price of a good can be determined by adding up the individual values of its characteristics. He further surmised that differences between prices between two goods can largely be explained by differences in the makeup of their characteristics.
Rosen’s research also highlighted the deficiency of the model in dealing with the dual forces of supply and demand, surmising that “price differences generally are equalizing only on the margin and not on the average. Hence, estimated hedonic price-characteristics functions typically identify neither demand nor supply.” (p.35). This means that Rosen’s model generally works well in a market equilibrium, but in times where demand is rapidly outstripping supply – or vice-versa – inconsistencies are likely to occur. The same is also true when applied to markets within which various different sub-markets exist, as has been examined and extrapolated by other researchers such as Limsombunchai, Gan and Lee (2004) and Sirmans et al. (2005).
Sirmans et al. (2005) conducted a large-scale review of recent studies that used regression modelling to estimate house prices in the USA. One of the aims of the research paper was to compare and contrast the results of the different analyses in order to identify trends and conclude whether findings were consistent across regions. Early in their paper, they state:
One caveat in using hedonic pricing models is that the results are location specific and are difficult to generalize across different geographic locations. On the other hand, comparing studies across areas may at least establish those characteristics that are consistently valued (either positively or negatively) by homebuyers. Because of this, hedonic pricing models are generally used to gain insight into the workings of a particular market. (Sirmans et al, 2005, p.2)
Indeed, their findings concluded amongst other things that the effect of square footage on selling price did not vary a great deal across regions. “The greatest effect was in the Southwest and the lowest average effect was in the Midwest” (p. 37). Furthermore, the total size of the lot was generally consistent across all regions and, interestingly, the value of an additional bathroom was generally between 10-12% in the Northeast and Southwest regions. As demonstrated by these results, the caveat outlined in the introduction can be interpreted as to apply to vast geographical regions, such as the Midwest, rather than to sub-markets in individual cities.
However, Limsombunchai, Gan and Lee (2004) state in their research – which seeks to compare the performance of the hedonic pricing model against artificial neural network theory, an artificial intelligence model “designed to replicate the human brain’s learning process” – that “it is generally unrealistic to deal with the housing market in any geographical area as a single unit. Therefore, it seems more reasonable to introduce geographical information or location factor into a model that allows shifts in the house price level” (p. 194). This assertion is backed up by reference to the conclusions of Fletcher et al. (2000) who modelled housing sub-markets in Portland, Oregon. Their findings were that there was a significant relationship between location and property value. This appears to be self-evident to anyone au fait with a basic knowledge of the London housing market but their research also provided useful adjustments which can be made to dampen the effects of geographical variations on the overall reliability of the results.
These factors can be considered ‘micro’ factors, which also include attributes of the building and/or flat, as well as locational factors such as proximity to transportation links and shopping facilities. ‘Macro’ factors on the other hand are those which relate to wider economic forces. These tend to have a cumulative rather than an immediate or innate effect on the rental or sales value of a property and thus studies which aim to analyse the effects of such factors tends to be longitudinal in nature.
3.3.2. ‘Macro’ Factors
Examples of literature analysing the influence of such ‘macro’ factors include Aspergis and Rezitis’ (2003) study analysing the impact of variables such as loan rates, inflation, employment, and money supply on house prices within the context of Greece’s membership of the Economic and Monetary Union of the European Union. An error correction vector autoregressive model (ECVAR) was used to determine this, which was largely altered and refined by the authors of the paper, but which is based on and similar in character to a hedonic regression model.
Other research projects into ‘macro’ factors include Baffoe-Bonnie’s (1998) work assessing “The dynamic impact of macroeconomic aggregates on housing prices and stock of houses” which analysed the interrelationship of national trends and regional factors, such as employment growth and prevailing interest rates. Glindro et al. (2008) and Prashardes and Savva (2009) conducted similar macro-economic analyses using regression models in the Asia-Pacific region and Cyprus respectively.
3.3.3. ‘Micro’ Factors
In contrast to the somewhat constrained focus of the macro-economic research cited above, the analysis of ‘micro’ factors has been varied in spread and wide-ranging in breadth. As stated above, the hedonic regression method has been employed and found to be useful not only in determining the overall predicted rental value or sales price of an individual property, but also in determining the effect of individual characteristics. This is essentially the crux of this piece of research: to extricate the individual values added (or otherwise) by the amenities in Build to Rent buildings from the rest of the components that make up the rental value.
Indeed, the hedonic regression methodology has been used to determine the effects of environmental factors such as noise and pollution; the studies of Mieszkowski et al. (1978) and Uyena et al. (1980) looked at the impact of noise pollution caused by proximity to airports, whereas Damm et al. (1980) used similar techniques to analyse the likely effects of the noise caused by the Washington District Metro on house prices. Ridkar and Henning (1982) were more holistic in their focus but paid special attention to the effects of air pollution on house prices, concluding that the value of property in the St Louis metropolitan area could increase by a total of $82,790,000 if sulfation levels across the city were reduced by 0.25mg.
In addition to this, the research of Chattopadhyay (1999), found that buyers were prepared to pay a premium for properties in areas with reduced air pollution, specifically a reduction in the level of sulphur dioxide and particulate matter (PM-10) in the immediate atmosphere. Similarly, research into the premium that buyers were willing to pay for increased levels of water quality has been undertaken Leggett and Bockstael (2000); their findings were that areas with lower concentrations of coliform bacteria had significantly higher property values but did not address whether this factor could be considered a cause or effect. Ketkar (1992), however, identified that the effect of the presence of a hazardous waste site in a municipality was a 2% reduction in house prices.
Like environmental factors, situational factors relating to the property’s location, local amenities, transport links and retail amenities can be considered to be types of ‘micro’ factor. One of the few studies that has been undertaken using this methodology and relating to property values in the UK is that of Forrest, Glen and Ward (1996). They sought to compare the effects of a new urban transit system, the Manchester Metrolink, on house prices before and after its construction and implementation. Property and neighbourhood characteristics as well as location were used as variables in the study and, in contrast to many of the claims pertaining to light transport systems’ effects on property prices overseas, did not find any significance in their results which indicated a house price increase.
However, Tse (2002) was able to determine that proximity to the Mass Transit Railway, as well as sea views, had significant effects on property values in Hong Kong. Tse used as variables: presence of local amenities, age of buildings, floor area, floor level, aspect, view, and transport accessibility.
In a similar vein, Clark and Herrin’s study (2000) – which found that the school district and thus the quality of education provided therein had a significant impact on property values – can also be said to have considered the relative impact of a property’s location and neighbourhood on its value. Clark and Herrin (2000) also studied the impact that a building’s age has on sales prices for the apartments contained within; this, while a ‘micro’ factor can be considered a different sub-division of ‘micro’ factor, as explained below.
The locational and environmental factors outlined above, including air quality, pollution levels, noise levels, and distance to transportation and good schools can be further distinguished from the characteristics relating to the buildings and apartments themselves. A well-constructed, brand new, luxury apartment building with a host of modern amenities could be built in an area with high levels of pollution, poor transport links and low-quality drinking water; it stands to reason that it would not be as valuable as a building situated in an area benefitting from the opposite and this is borne out by the research cited above.
As discussed in the above, the rental value of a property (P) can be stated to be a function of its building characteristics (B) – such as age and presence of amenities – and its locational characteristics (L) as well as unknown quantities (x):
illustration not visible in this excerpt
Examples of research into building ‘micro’ factors include that of Goodman and Thibodeau (1995), who analysed the effect that the age of a building has on house prices. An analysis of over 8,500 transactions for single family units in Dallas was undertaken and, interestingly, they found that the relationship between age and house price was non-linear. This result was attributed to complicating factors, such as “depreciation and the vintage effect” (p.40). This contrasted to the findings of Clark and Herrin (2000) who found that the age of a building was negatively correlated with value. Other research into this area, such as the work of Kain and Quigley (1970, p. 539) found that a new building in St Louis was likely to sell for $3,150 more than an “otherwise identical one that is 25 years old. Monthly rent decreases by about $2.82 per month for each increase of ten years in age of structure”. Other research (Rodriguez & Sirmans, 1994; Straszheim, 1975) reaches the same conclusion as that of Clark and Herrin, as well as Kain and Quigley.
3.4. Research into the London Build to Rent market
Clearly, the age and focus of this research does not lend itself to direct interpolation to the London Build to Rent market in 2017 and thus, it is also important to analyse the scope of recent, UK-focused publications as well. As noted above, there are currently no academic papers of the type focusing on the North American market outlined above, but there have been an increasing number of commercial publications over the past year.
3.4.1. Research Reports
In 2016, Johnny Morris of Countrywide presented a piece of research conducted by his firm entitled ‘Paying for more: Premiums in the private rented sector’ (Morris, 2016). Using an unknown methodology, their research identified that there is a rental premium applicable for new-build flats in the UK, which has varied over time according to the wider market. In 2015, this premium was estimated to be approximately 20%. Furthermore, they circulated a questionnaire which found that 52% of renters would not pay for a ‘premium service’, including amenity areas.
Knight Frank publish an annual report entitled ‘The Knight Frank Tenant Survey’ (Knight Frank, 2016). Last year’s publication found, using rental data from Dolphin Square in Pimlico, SW1, that smaller flats achieve a higher value per square foot than larger flats. It also found that 33% of the renters surveyed rated “good amenities” as an important factor when choosing a property. This was further broken down by amenity type, with 33% of those surveyed identifying a gym as important down to just 6% for a communal cinema room. In the most recent publication of the survey (Knight Frank, 2017), tenants have been subdivided into socio-demographic types based on their age, income and demographic profile. The results of the survey are reproduced below (pp. 6-7):
illustration not visible in this excerpt
Figure 1 – Results of Knight Frank Tenant Survey (2017)
Knight Frank’s most recent tenant survey is based on the answers of 10,218 respondents across the UK and was undertaken by YouGov; reportedly “Knight Frank analysis of more than 1.5 million data points from the survey highlights the key characteristics and priorities for tenants across the country” (Knight Frank, p.3). Although tenants claim that they will pay additional rent for amenities, the question remains: does the inclusion of such amenities translate into higher rental values for Build to Rent landlords?
Colliers International (2016) have also published a research report entitled ‘Build-to-Rent: Colliers Rental Insights 2016’ which, like this paper, aimed to establish the impact of amenities on rental premiums and concluded that professionally managed blocks achieve a premium of c. 9.6% over the base rent, while gyms achieve c. 11.7% and concierge services achieve just c. 2%. The data used was from 45 schemes all across the United Kingdom. It is not known what methodology was used to draw these conclusions and, as will be discussed in further detail in Chapter 3 below, it is not thought that 45 data points is sufficient to drawn a statistically significant or meaningful conclusion. However, this research provides a strong indication that developers, investors and property consultants are beginning to recognise the commercial and academic importance of addressing this topic.
3.4.2. General Discourse - UK
It appears that much of the discourse surrounding the question centres on the comparison of the emerging British Build to Rent sector with the established US market. It seems to be accepted that there are lessons to be learnt from the successes of American multifamily housing which enjoys primacy amongst the 37% of households in the private rented sector in the USA (US Census Bureau, 2016). By way of comparison, the UK private rented sector comprises 20% of households (DCLG, 2016) although this figure is anticipated to rise to 24% of households by 2021 (Knight Frank, 2017). Jeanette Veldkamp (2017), a director at the architects Chapman Taylor, states that “The Multifamily housing market in the US is now well established; the offer is very different from 10 years ago. With the UK Build to Rent market emerging, this is a good time to appraise what we can learn from the more mature market in the US”
Veldkamp goes on to state that:
Multifamily Housing offers institutional investors diversification from other real estate sectors in that it provides a safe, steady cash flow, but the product has to be right. The uplift from a regular rental development to Multifamily Housing, reportedly a whopping 20% higher with its added amenities and services, is one reason why this product is so attractive for investors and developers.
To attract investors, the product must appear safe. Therefore, layouts, amenities, and the standard of finishes should look familiar in order to forecast future returns.
A premium of up to 20% is said to be achievable by Veldkamp if a “regular rental development” is transformed to “Multifamily Housing” which is defined by the “added amenities and services” necessary to “forecast future returns”. It would appear that this statement is unsubstantiated and therefore can be seen as a good example of the received wisdom which developers and investors are at risk of following without an evidence-based rationale.
Further evidence of this can be found in the self-published works of other designers and consultants such as Carl Wood (2017) of project managers Turner and Townsend who states that “increase in amenity provision and decrease in unit sizes” is a market differentiator that “future players in the market” may have to embrace in order to demonstrate their innovation, particularly over existing interests. It is accepted that statements such as this do not go so far as to quantify the hypothetical rental premium created by amenity areas, but it is widely held that amenities do drive value in some form.
Barton and Breen (2016) agree with the notion that they can act as a differentiator for prospective tenants, stating “for PRS projects, amenities and communal areas will require more consideration. Amenity spaces, such as gyms, business areas, cafes and crèches may be viewed more favourably by potential tenants. Areas that can be hired for private functions will also be popular, and communal outdoor spaces – roof gardens, green areas, quiet corners – will be important for families.”
However, Jordan Perlman, co-founder at Newgrounds Architects – “the boutique architecture practice behind two of London's biggest build-to-rent schemes” – is quoted in Property Week (19/08/2016) as having said that “‘areas where people can socialise, make friends and feel they live in a genuine community is essential if they are staying over the long run […] But amenities like this pose their own management issues and the extent to which they add value to a development has yet to be determined.’” This demonstrates that there is scepticism in some quarters surrounding the perceived value of amenity spaces in Build to Rent schemes.
Indeed, this a view shared by Richard Berridge, Director at Blackbird Real Estate, a specialist Build to Rent consultancy practice. Berridge, also quoted in Property Week (31/05/2017) believes that “Renting is not, by and large, a premium business except, perhaps, on a ‘boutique’ scale. But many operators are bringing some very large schemes to the market expecting premium rents and for those rents to continue to be driven upwards by the sophistication of the offer.” Sophistication can be taken to mean ‘additional extras’ which differentiate the Build to Rent sector from the Buy to Let sector and Berridge further believes that “when the buildings become older and competition becomes more intense in this sector, the novelty will wear off and it will simply be down to value for money and great service. Security of tenure, customer service and rapid response to issues will continue to be worth more than [the Buy to Let] sector can offer”.
It is worth noting at this point that this viewpoint has notable similarities to guidance which has been published in a more official capacity, such as that of industry bodies like the British Property Federation. The BPF’s ‘Build to Rent – Welcome to the UK’s Newest Housing Sector’ (2015) also identifies security of tenure as a key offering to assist investors in “keeping their buildings fully-occupied with satisfied tenants” - reportedly their “primary motivation”. However, “good onsite amenities” are also proclaimed to be viable means to achieve this end. While this paper does not hope to address the supposition that amenity provisions help to reduce tenant turnover and decrease vacancy rates, it is important to illuminate all possible reasons that developers and investors may have for including amenity provisions and thus it is worth highlighting.
The Trust for London (2017) published a report aimed at “public- and private-sector players [in order to] identify opportunities and stumbling blocks for a sector that is still trying to define itself” (p.3). The report offers good practice guidance for architects and developers, part of which has been reproduced below:
Abbildung in dieser Leseprobe nicht enthalten
Figure 2 – Build to Rent good practice guide from Trust for London (2017)
It is interesting to note that secure tenancies are mentioned once again, as is the provision of “communal social space” of the kind which Jordan Perlman (Property Week, 31/05/2017) believes is yet to be tested from a long-term value perspective. The non-qualified inclusion of concierge services is also of interest, particularly considering the large ongoing costs associated with such a provision. To provide 24-hour concierge coverage, at least four members of staff would be required. The salaries and additional costs – such as national insurance and pension or VAT – associated with four members of staff are likely to exceed £100,000 per year, assuming that they at least the London Living Wage. To incur such costs for little or no benefit is likely to significantly affect returns, particularly in lower-value areas or smaller schemes without the capacity to absorb such expenditure.
The quantity surveyors, construction consultants and property advisors Arcadis, sum up the balance which developers and investors should seek to strike when considering amenity spaces in Build to Rent schemes. In their paper ‘Build to Rent: Pushing the Boundaries” (Arcadis, 2017, p. 21), it is stated that “driving additional revenue streams can add value to the overall PRS offer but these need to be driven by a cost vs. revenue analysis. In some instances, the location of the development will deliver a basic level of amenity; an adjacent gym and fitness facilities, restaurants, serviced offices etc. but the difference in costs / ft² to insert a leisure facility (especially with a pool) catering or restaurant facilities into the development itself, should not be underestimated.”
Arcadis (Arcadis, 2016) seem to urge developers and investors to consider amenities from a long-term operational perspective, highlighting the additional expenditure that they are likely to incur as a result of them: “the real impact on revenues needs to be understood and reflected in the investment business case. The US market for instance requires many schemes to have expensive leisure amenities included to differentiate from other schemes despite there being significant under-utilisation of them. Similar conscious decisions might still be made here in the UK to de-risk occupancy but a solid understanding of capital and operational expenditure impact relative to revenues is critical.” (Arcadis, p. 21).
It is therefore clear that much research has been undertaken into the US Build to Rent – or multifamily – market, at least in part due to its maturity and as a response to commercial necessities. As a well-established industry in which billions of dollars are invested each year and have been since the early 1970s (Doan, 1997), time and money have driven the creation of a strong body of academic research.
Much of this research uses the same methodology as this paper, seeking to collect rental data from a locality, supplementing it with additional information regarding the ‘micro’ factors which influence values, and drawing conclusions as to what impact each of these components of value has on the total rental price by way of a hedonic regression analysis. A large amount of work has been done in the USA and throughout the world to bring this methodology to the fore as one of the most accurate and statistically robust approaches for analysing property values.
However, in the UK, it appears that despite a marked rise in commentary and interest in the Build to Rent market, such an analysis is yet to be undertaken in respect of London’s private rented sector. Consultants, designers, architects and industry bodies have offered views and suppositions on the value that amenity spaces add to buildings, however much of this seems to be drawn from hypothetical parallels to the US multifamily sector and little seems to be based on statistics or evidence. There is, therefore, a gap in the collective knowledge which this paper will seek to address.
As outlined above, particularly at Section 1.3 ‘Research Strategy’, an analysis of this type is very reliant on quality and quantity of data. In order to glean useful, robust results, a large amount of data points must be collected and each individual rental value must be supplemented with a wide range of other variables. Indeed, one of the main drawbacks of an analysis of this breadth and scale is the sheer amount of rich data that needs to be collected and the practicalities of acquiring it. Usually, all of the data necessary for such an analysis will not be found from one source, as was certainly the case with this project. Furthermore, much of the additional data that would strengthen and enrich the analysis is currently unavailable. This has implications as to the reliability of the results, but provides a good deal of scope for further research to be undertaken in the same field using the same methodology.
The various techniques, difficulties, approaches and methodologies used to reach the final research results have been elucidated below. Particular focus is given to: sampling; the selection of variables; modelling and further explanation of the theory of regression analysis; data collection; and the difficulties encountered during the research.
In order to answer the research question, ‘to what extend does the provision of amenity spaces in London’s Build to Rent schemes increase rental value’, it appears to be self-evident that the sample should be restricted to London’s Build to Rent schemes. However, the term ‘London’ allows scope for interpretation, as does, to some extent, ‘Build to Rent’. It is therefore necessary to specify which rents will be collected and analysed in further detail.
5.1. Sample Area
London can be interpreted as at least four different, overlapping areas: the City of London – i.e. the ‘Square Mile’; Inner London as defined by the London Government Act 1963, which comprises the innermost 12 boroughs and is geographically similar to the former County of London (Saint, 1989); Greater London, which comprises the 32 London boroughs under the administration of the GLA; and, less commonly, the area within the M25 motorway.
Much of London’s Build to Rent market is based in Outer London as demonstrated by the number of construction starts in Q1/Q2 2017: 1,145 Build to Rent units began construction in the 20 Outer London boroughs, compared with just 140 in the Inner London boroughs (Molior, 2017). Of the 18,888 starts since 2009, just over 70% or 13,724 of these have been in Outer London locations (ibid). The reasons for this are likely to lie in land economics as land in outer locations is typically cheaper (Trust for London, p.18), which allows for less expensive build costs. As rental values are more restricted by domestic economic factors, namely household incomes, employment rates and wage growth, the extremely high values sometimes achieved in London’s Build for Sale market are more difficult to achieve. This is a somewhat simplified explanation but goes some way to explaining the discrepancies in rental yields between inner and outer boroughs, and indeed between London and the rest of the UK. The map below produced by LendInvest displays average rental yields by area with the darkest red indicating high yields (above 4.5%) and lighter red indicating low yields (less than 3.0%):
For the purposes of this research, a sample of Build to Rent buildings from both inner and outer boroughs (i.e. Greater London) has been used. This will increase the sample size and should therefore ensure that the results are more accurate, however it is recognised that there are some additional problems created by including data from a number of different submarkets in one analysis (Limsombunchai et al., 2004; Fletcher et al., 2000).
Molior’s Build to Rent database displays 324 different Build to Rent buildings in the capital, however also includes sites with pre-planning, planning permission, planning withdrawn, in the application stages or under construction; clearly these are of no use as they are currently not let so have been excluded. Once this have been eliminated, the following 110 properties are displayed:
These have been further reduced manually during the data collection stage to a sample of 68 buildings, giving 149 separate rental values. This is because data exists for studios, one bedroom units, two bedroom units and three bedroom units in each scheme. The buildings that were excluded either did not have any rental data available or were let out by Housing Associations at local housing allowance rates, rents which are pegged to housing benefits and therefore do not constitute open market evidence.
The number of variables required is a mathematical question, as well as a valuation question. The rental value of a property is made up of a number of different factors; if you asked someone with a basic understanding of property values, they might be likely to list size, number of bedrooms, number of bathrooms, location and finish as the drivers of value. A property professional would be likely to agree that these are the main drivers, however may also list factors such as: age of the building, height of the building, floor that the property is situated on, private amenity spaces, building services including shared amenity provision, as well as certain intangible factors such as ‘trendiness’ of the area, scope for renovation, and potential for value uplift in the future.
Sirman, Macpherson and Zietz (2005) undertook a review of 125 regression pricing models from analyses in the USA and summarised the most frequently occurring variables, along with the number of times they were found to be statistically significant or otherwise. Below are the results of this research reproduced in tabular format and taken from Corsini (2009, p.9):
Abbildung in dieser Leseprobe nicht enthalten
Figure 3 – Sirman et al.’s most recurring characteristics
Clearly, not all 20 of these variables are applicable to rental properties in London and therefore some should not be modelled. Lot size refers to the acreage of grounds appurtenant to the property and square feet refers to the gross internal area. In the first instance, the information regarding lot size is obsolete as we are analysing new build apartments which are not sold with appurtenant outdoor space. Secondly, square footage is not available as the data deals only with rents at a building level, rather than an apartment level. Brick refers to the construction materials and, again, is not thought to be pertinent to an analysis of London’s rental product. Other factors which are more specific to the USA are air-conditioning, basement, deck and pool. Very few properties, newly-built apartments or otherwise, benefit from these amenities and their inclusion would likely result in a statistical irrelevance. Time on the market is interesting and certainly pertinent to properties offered for sale, however as we are looking at rents on a building-wide level, again this can be discounted.
The list of variables that were decided as most influential on a property’s rental value were:
1. Number of bedrooms
2. Approximate age – in years since the completion of the building
3. Number of bathrooms
4. Bath ratio – this was included to guard against any false equivalency that may result from the direct correlation between bedrooms and bathrooms
5. Finish – rated out of five
6. Proximity to Underground/ Overground/ National Rail – in miles
7. Proximity to Shopping Hub – in miles
8. Travel time to Oxford Circus – in minutes
11. Residents’ Lounge
The last three variables, concierge, gym and resident’s lounge, will form the basis of the answer to the research question. These three were chosen due to their ubiquity in the Build to Rent pipeline; as outlined in the literature review developers and architects often talk of concierge services being included as standard, along with “areas where people can socialise” (Property Week, 19/08/2016) such as residents’ lounges. Gyms are also often included as they are often near the top of the list of amenities for which tenants state they will pay additional rent, particularly if they already have a gym membership; they are often seen as a key part of the “full suite of add-ons” (Property Week, 11/05/2017)
As eleven variables – or ‘predictors’ – have now been identified which will influence the ‘response’ – or rental value – it is now possible to ascertain the amount of data that will be required. Green (1991) identifies a number of factors which will influence the size of the sample required including expected effect size, desired correlation and number of predictors. The minimum sample size recommended by Green for testing multiple correlation is 50 + 8m where m is the number of independent variables. To test individual predictors, 104 + m samples are required. Therefore, where m is 11, the sample size should be larger than 115 to 138 in order to obtain statistically significant results.
5.3. Data Collection
The rental data, as outlined above, is available from Molior’s Build to Rent database. The database provides a wealth of information regarding each scheme and this information is collected from before planning permission has been granted in many cases. Molior tracks the progress of a building from design to construction to occupation, providing data on the developer, consultant teams, architects, number of units, unit mix, links to planning documents via the local authority’s planning portal, and construction updates. The commercial property equivalents are CoStar and Glenigan, although the former focuses primarily on rental values and the latter primarily on construction progress; Molior provides detailed information on both.
The name of the scheme, postcode and rental values for one bedroom units, two bedroom units, three bedroom units and four bedroom units were collected from Molior. Studios were not taken into account due to the difficulty in modelling zero bedrooms in the regression analysis. The letter ‘S’ instead of a number would produce an error result as the software can understandably only model numerical values. Inputting zero would impact the value of the constant and likely skew the results very heavily. Also taken from Molior was the data for the age of the building.
The rental values given by Molior are in three formats: an upper value, a lower value and a median value. The median format was used for the purposes of this research, with one exception outlined below.
The number of bathrooms is not given by unit type on Molior’s database and this therefore required manual research. Rental particulars and advertisements were of assistance when attempting to ascertain whether a two or three bedroom unit in any given scheme had one bathroom or two. Often, the investment brochures and building brochures were available online and typical floorplans are often included, allowing one to surmise how many bathrooms is the norm for each unit type. Where buildings have two bedroom units, some with just one bathroom and others with two, it was found that Molior’s upper and lower ranges often tallied exactly with the prevailing asking rents. In other words, the two bathroom units were often achieving the upper rents and the one bathroom units the lower rents. In these instances, it was necessary to confirm the rental values manually as well, either by reference to the online property portals (such as Zoopla and Rightmove) or, preferably, by having a short conversation with the letting agents.
Quality of finish for each building was estimated using a similar methodology; rental particulars and advertisements were found, along with brochures outlining the fixtures, fittings and specification of the properties. It was necessary to use qualitative judgement to come to a quality rating and the range was 0 to 5 stars, with increments of 0.5. A ‘standard’ finish unit would naturally be awarded a quality rating of 2.5 stars and an outline specification was formulated to correspond to this:
Flooring - tongue and groove timber laminate flooring in the open plan kitchen/reception room and hallway, with carpets to the bedroom and white, floor-to-ceiling ceramic tiles to the bathroom.
Fixtures - contemporary white three piece suite to the bathroom with fiberglass bath, ceramic WC and ceramic single sink, Howden’s-style kitchen with black, marble-effect laminate worktop and white acrylic units, Lamona or similar appliances, uPVC double glazed windows.
M & E – video entry phone system, combination gas boiler (individual), data points in bedroom and living room.
While it is recognised that this is not scientific in approach, it was felt that quality of finish is an important variable and some effort should be made to quantify and rationalise its impact on rental values.
Proximity to shopping hub was calculated by reference to Google Maps. Instead of an as-the-crow-flies measurement, the walking function was selected to realistically calculate the distance that tenants would have to travel between their flats and the shopping hub. ‘Shopping hub’ was decided to constitute either a reasonably-provisioned, town centre-sized high street or a medium sized shopping centre. This does not take into account the quality, attractiveness or utility of such a provision and further subcategorization may be considered to be desirable by some.
Proximity to London Underground / Overground / National Rail was calculated using the same methodology as proximity to shopping hub. Again, this does not take into account the quality of the public transportation service available from the station and it may also be unfairly weighted against those properties in south London, which benefit from an increased bus service instead of a large number of Underground stations. In order to balance against this potentially harmful variable, ‘Travel time to Oxford Circus’ has been included as well.
This independent variable has been calculated using the web application ‘CityMapper’, which analyses all available public transportation routes and finds the quickest journey to a destination. Oxford Circus was chosen as it is widely considered to be the centre of the West End, but due to the fact that many people work in the City and Canary Wharf, as well as all over the capital, it can be considered somewhat arbitrary.
The provision of a gym, a concierge service and/or residents’ lounges were the final variables collected. Once again, this information was gathered from a mixture of references to Molior, as well as independently verifying the data against rental particulars, brochures and conversations with lettings agents.
A total of 149 rental values were collected, along with the corresponding independent variables for each data point. The data was inputted into Minitab, a specialist statistical analysis software programme, and various analyses were undertaken. At this point, it may be useful to clarify some of the terms that will be used in the remainder of this chapter and the following chapter.
Multiple Regression Analysis – the type of statistical analysis referred to throughout this paper which is predicated on the ‘basket of goods’ supposition. A basic regression analysis, known as a ‘least squared’ analysis, is familiar to anyone who has studied a graph plotting a dependent variable (typically ‘y’) against an independent variable (‘x’). For example, the rate of childhood obesity in a number of different areas (y) could be plotted against the average consumption levels of sugar (x), producing the following graph:
illustration not visible in this excerpt
Figure 4 – Hypothetical relationship between sugar consumption and obesity
On the strength of this graph taken in isolation, one might say with a reasonable amount of certainty that increased sugar consumption leads to increased obesity levels. However, a multiple regression analysis also takes into account a number of other contributing factors. These factors may serve to shed further light on the results or change the conclusion. In the example above, access to outdoor play facilities, number of greengrocers within a 1 mile radius, prevalence of underactive thyroid conditions and genetic disposition to Cushing’s syndrome might also be taken into account to form a multiple regression analysis. The correlation between sugar consumption and obesity levels would almost certainly change as a result of this consideration of other contributory factors. Frost (2012) describes how researchers once believed that there was a causal relationship between drinking coffee and heart disease; after undertaking a multiple regression analysis, they discovered that coffee drinkers are more likely to be smokers and were able to determine that it was in fact smoking that led to the increase in heart disease, not coffee.
Hedonic Regression / Pricing Model – produced as a result of a multiple regression analysis, a hedonic model is an accurate predictive model whereby the independent variables are manually inputted and an estimated price is produced. For example, a hedonic model would be able to estimate the rental value of a property if the number of bedrooms, bathrooms, amenities, distances and travel times were inputted. It is important to note that there is typically a substantial margin for error, relative to how comprehensive and robust the model is deemed to be. This margin for error is quantified by the model, allowing a range to be produced, as well as single price.
Stepwise Regression - a tool for regression analysis whereby a number the independent variables are ‘cycled’ for inclusion in the model. The tool determines whether the inclusion of each variable helps to make the model more reliable or whether it reduces accuracy. If the accuracy is reduced beyond a lower threshold, the variable is eliminated and the next variable is tested. Typically, two cycles are used to test each variable against the others.
Best Subsets – similar to the stepwise regression tool, but instead of a single model being produced as a result, a number of different models with different reliability scorings are provided, allowing the researcher to pick which model should be used themselves.
For the purposes of this research, a basic multiple regression analysis will be produced. This will indicate the reliability of the model and each variable’s influence on the rental price. It will also demonstrate which of the variables have no notable effect on the model and the relative strength of each factor in influencing the rental value.
A stepwise regression will also be used in order to find the most statistically significant variables and this will then be compared to a best subsets analysis to see which other variables were statistically significant. Finally, a hedonic exercise will be undertaken to compare the price produced by the model against a known rental value by inputting the independent variables.
6. Research Results
The purpose of the research was to determine the extent to which the inclusion of amenity areas increased rental values in Build to Rent buildings in London. As described above, data from 64 buildings was collected including rental values and amenity provision, as well as various other independent variables such as number of bedrooms, number of bathrooms, distance to local services and travel time to Central London. In total, 1,788 data points creating 149 complete datasets of 11 independent variables and one dependent variable (i.e. rental value).
A multiple regression analysis was undertaken, as well as a stepwise regression and best subsets analysis, the results of which were hoped to be able to produce an accurate hedonic pricing model that could be used to estimate the rental value of a property providing all other variables were known.
6.2. Interpreting the Results
Statistical software of this nature often produces results which are not immediately coherent or comprehensible. The language of statistics seems to be inherently complex and therefore, the data tables produced as a result of the analyses and reproduced further below will benefit from some additional context:
R-Squared Value – also stated as R², the R-squared value “is equal to one minus the ratio of the sum of squared estimated errors (the deviation of the actual value of the dependent variable from the regression line) to the sum of squared deviations about the mean of the dependent variable” (Sykes, 1993, p.23). In simple terms, it can be described as ‘goodness of fit’. A higher R-squared value means that more data points fall within the model’s predicted range and it is sometimes also given as a percentage. For example, an R squared value of 0.5214 would mean that 52.14% of the variation of the dependent variable is explained by the regression. The remaining 47.86% is explained either by other, unaccounted-for independent variables, or x. The equation outlined earlier in this paper states x to be unknown quantities or ‘noise’.
Coefficient – the multiplier for each independent variable. To take an earlier example, if the coefficient of ‘number of greengrocers within a 1 mile radius’ was -4.1, the effect of two greengrocers within a mile on the dependent variable – i.e. percentage of obese population – would be -8.2%.
T-Value and P-Value – the dual indicators of reliability for each independent variable. In the words of Runkel (Runkel, 2016), “The t-value measures the size of the difference relative to the variation in your sample data. Put another way, T is simply the calculated difference represented in units of standard error. The greater the magnitude of T (it can be either positive or negative), the greater the evidence against the null hypothesis that there is no significant difference.” The graph below shows a standard bell curve distribution:
Abbildung in dieser Leseprobe nicht enthalten
Figure 5 – Distribution of t-values
If the variable has a t-value close to zero it means that it falls within a standard range of distribution, demonstrating that it is unlikely to be of statistical significance. The chances of the data being in the region from 2.8 upwards (or -2.8 downwards) are very small, as illustrated by the maroon shading and by the number 0.005712. This is the probability in decimal form, or p-value, that the variable’s impact on the regression analysis is a mistake. As the probability is so small, this means the variable is statistically significant.
6.3. Key Findings
The findings from the multiple regression analysis have been reproduced in their raw format below. The following paragraphs will analyse the findings in further detail.
illustration not visible in this excerpt
Figure 6 – Results of multiple regression analysis
6.3.1. Amenity Spaces
The first thing to note is the R squared value. This is 0.5777 or 57.77% meaning that the model estimates that it can explain 57.77% of the variance between rental values from the data inputted. While this is perhaps lower than might be hoped for, it is not statistically insignificant and the findings should therefore have a good degree of weight. However, it is perhaps indicative that more data is necessary in order to draw a stronger conclusion.
The three variables in which we are most interested are the concierge service, the gym and the residents’ lounge. They have t-values of -0.25, -0.13 and -0.33 respectively. On the standard distribution bell curve, these data points would therefore fall in the centre. This means that their effect on rental value is insignificant to the extent that it is not possible to ascertain whether the independent variables have influenced the dependent variable to any degree.
In order to explore the further reasons for this, an additional analysis of the data was carried out. While 149 data points may be mathematically sufficient to carry out such an exercise (Green, 1991), it is interesting to note how many of the datasets actually included gyms, concierge services or residents’ lounges:
illustration not visible in this excerpt
Figure 7 – Number of occurrences of amenities within the database
The blue bar represents the number of occurrences with the orange indicating the number of non-occurrences. Residents’ lounge occurred 20 times, gym occurred 17 times and concierge service occurred 50 times. Of the three, the concierge service would theoretically be the most likely to have a statistical significance on the basis that it occurs in over a third of datasets. However, the strength of correlation between concierge services and rental values is lower than the residents’ lounge, despite having 2.5 times more occurrences.
It is also interesting to note that each of the amenity spaces had negative values, albeit very small negative values. This would suggest that they are detrimental to a property’s rental value, when the other independent variables remain constant. However, the ‘standard error’ for each of the variables indicates the degree to which the model predicts the average variance could fall. The standard error coefficient is £219 for the concierge service, £309 for the gym and £304 for the residents’ lounge. This means that each of these variables could have a positive effect on rental value in certain instances and the model recognises this. However, this is again constrained by the large p-values and small t-values of the results for amenity spaces, which indicate that the variance – positive or negative – could also be largely explained by chance rather than the variable itself.
In short, the results indicate that the effect of adding amenity spaces into Build to Rent schemes in London is largely indiscernible. From the data, the most likely effect on rental values is that they slightly decrease rent but the standard error is fairly high, meaning that they can also increase rent within a certain range, as well. However, any effect which the model determined that the amenity spaces have on rent could also be explained by chance. For the gym, there is an 89.5% chance that the variance is explained by chance and for the concierge service and residents’ lounge, this figure is 80.0% and 74.3% respectively.
This means three things: firstly, more data is needed to reduce the probability that any results given in the future can be attributed to chance; secondly, from the data provided, the quantitative value of amenity spaces on rent cannot be stated with statistical certainty; and thirdly, enough datasets were collected to theoretically result in statistically significant results meaning that it is likely that amenity spaces do not have any substantial impact on rental values.
The hypothesis that can be taken forward from this is therefore rental values are not notably affected by the provision of amenity spaces and any future research which uses more data will also conclude that amenity spaces do not alter the rent in any statistically significant way.
6.3.2. Stepwise and Best Subsets Analyses
The results of the stepwise and best subsets analyses will help to determine which of the remaining variables were the most useful for explaining the rental value in each of the datasets analysed by the model.
illustration not visible in this excerpt
Figure 8 – Best Subsets results
As number of bedrooms had the highest t-value and lowest p-value in the regression analysis, this indicates that it was the least likely to have influenced the dependent variable by chance (probability of 0.000). Number of bedrooms was therefore chosen as the continuous predictor, the variable which the model always keeps in the stepwise analysis. As is demonstrated by the findings, the inclusion of all variables results in the joint-strongest correlation, although this is a very marginal outcome. The inclusion of 7 or 8 independent variables also resulted in similarly strong R squared values. It is interesting to note that in each of the four other strongest analyses, the missing variables are all amenity variables. In other words, the effect of including the amenity spaces to enhance the reliability of the model is almost negligible.
It is expected that the stepwise analysis will result in a similar outcome, with the amenity spaces being eliminated from the final model on the basis that they bring very little statistical significance. For this analysis it is necessary to assume a minimum level of reliability for the variables. This is known as an ‘alpha’ and 0.15 was entered, meaning that only those variables which have a probability of meaningfully affecting the dependent variable of above 85% would be included.
illustration not visible in this excerpt
Figure 9 – Stepwise results
The variables found to be the most reliable by the stepwise function are:
- Number of bedrooms
- Ratio of bathrooms to bedrooms
- Proximity to public transport
- Proximity to shopping hub
- Travel time to Oxford Circus
It can be said that each of these variables has a significant impact in explaining the variance of the dependent variable, from 57.45%. In other words, number of bedrooms, bathrooms, finish, etcetera can all go a significant way to quantifying the intrinsic rental value of at least 57.45% of any of the buildings.
6.3.3. Hedonic Pricing Model
As a final test of the model, the following independent variables were put in:
illustration not visible in this excerpt
Figure 10 – Hedonic Pricing Model results
In short, a two bedroom flat with one bathroom and finished to a reasonable standard in a one year old building with no amenities was assumed. The flat is located 0.3 miles from a shopping hub and railway station, resulting in a total travel time to Oxford Circus of 30 minutes. This produced a rental value of £1,547.
A dataset was identified from amongst the samples which corresponded with these inputs in order to compare the accuracy of the results. The dataset was for a two bedroom flat in Lewisham with one bathroom, no amenities, a travel time of 32 minutes to Oxford Circus and in a five year old building. The rent is £1,530 which indicates a good degree of accuracy from the model.
In order to further analyse the impact of amenity spaces on the rental value, the same independent variables were used, however a full suite of amenities was assumed. The rent dropped from £1,547 to £1,350. It should once again be noted that the model’s high degree of scepticism regarding the usefulness of the amenity areas in predicting the rental value means that this cannot be considered significant.
6.3.4. Other Findings
It is interesting to note the factors which did substantially affect the rental values of properties in the analysis undertaken. Using the results of the stepwise regression analysis above, the four factors with the smallest p-values – all of which were 0.000, meaning that the probability that the correlation between them and the dependent variable can be ascribed to chance are nil – were: number of bedrooms, ratio of bathrooms, finish and travel time to Oxford Circus.
The most significant coefficient of these four variables was ratio of bathrooms to bedrooms which had an average £1,691 effect on rent. As this is a ration, it does not mean that for every bathroom added to an apartment by a developer, an increase of the same amount can be expected, but rather for every ratio of 1:1, this can be expected. For a two bedroom property with one bathroom – i.e. a ratio of 2:1 – half of this amount would be added as rental value by the model.
However, when looking at this, or any other variable, in isolation the constant value of minus £1,202 also needs to be taken into account, as well as the other variables in the model. As such, it is not particularly helpful or accurate to state that a one bedroom flat with a 1:1 bathroom ratio would achieve £1,307 (-1,202 + 1,691 + 818 [value per bedroom]) as this does not take into account the rent to be subtracted for distance to public transportation and travel time to Oxford Circus, nor does it account for quality of finish; it makes assumptions which are not possible to make. In brief, the model must be looked at as a whole.
Further to the above, the second most significant coefficient with the smallest p-value is number of bedrooms. All things considered and being equal, the value per additional bedroom was found to be £818 per month. While this is technically statistically significant, the individual datasets collected during the course of this research show that this clearly will not always be held to be true; a one bedroom flat and two bedroom flat in an outer London location are not likely to vary by such a great difference and therefore further attention should be given to the ‘standard error’ identified by the model which, in the case of number of bedrooms, is £126. This means that most datasets will fall within a range of £126 either side of £818 per additional bedroom. This goes some way to achieving a variance more in line with the outer London market, as does the standard error for ratio of bathrooms, which is £393, but highlights the difficulties encountered by attempting to find coefficients and averages which fit both the outer and inner London submarkets, and indeed all other submarkets within those locations.
Travel time to Oxford Circus is also of significance and interest. Remembering that this was measured in minutes using the journey planning tool City Mapper, its negative coefficient seems to make sense. The coefficient is minus £39.75, meaning that for every minute away from Oxford Circus a property is located, its rental value falls by almost £40. To express it another way, the difference between two identical properties, one in Oxford Circus and one forty minutes’ travelling time from Oxford Circus, is £1,590 per month. Unlike number of bedrooms and bathrooms, this variable inherently factors in and measures the location of a property respective to the other datasets. It is therefore far less likely to be inaccurate when considering the London market as a whole and the effect of sub-markets skewing the results is lessened.
Lastly, the effect of quality on rent is also substantial. The coefficient for quality of finish is £494 and it is thought that this has useful implications for developers and investors. It is important to remember that this is a qualitative analysis, however if developers are able to increase the quality of their product from 2.5 stars to 3.5 stars, this would result in an average uplift of £494 per month. Over the course of a year, this would translate to an additional £5,928 and, discounted at 5% into perpetuity, this results in an additional value of £118,560. Therefore, if the costs of undertaking the works to enable a property to increase from 2.5 to 3.5 stars are less than this, it may be considered a worthwhile investment at the initial development stage.
However, consider the fact that the property will require redecoration and refitting during this period (i.e. perpetuity) and the additional investment therefore has limited efficacy. Assuming that a property needs to be entirely redecorated and refitted every seven years, the year’s purchase at a 5% discount rate will substantially decrease and the costs would have to total less than £34,302 to make the investment worthwhile. Again, there is a standard error of £123, so these figures ought not be taken entirely at face value, however this provides a potentially interesting new approach to development appraisal in the Build to Rent sector.
In the first regression analysis, other findings included: rental values decrease by £42.30 per month for every year the building has been completed, thus going some way to quantifying the ‘new build premium’; and rental values decrease by £463 per month for every mile a property is located from a London Underground, London Overground or National Rail station. These findings were ultimately excluded by the stepwise regression analysis as they were not determined to be of the utmost statistical significance, but they are still of some interest.
There are a significant number of limitations in respect of the research results, as evinced by the R-squared value of 0.5777. The model is unable to explain 42.33% of the variance in rental values and it is thought that this is primarily a result of the lack of data and the quality of the data used. Further limitations can be explained by inherent deficiencies in the methodology, some of which were discussed at chapter 2 above and notably include the difficulties in distinguishing between diverse submarkets.
Research from Cluttons (Cluttons, 2017) indicates that within their definition of prime central London (PCL) – which comprises mainly the West End, the eastern parts of Kensington & Chelsea, along with some southern parts of Camden – there are sixteen distinct residential submarkets, such as Holland Park, Knightsbridge, Mayfair and Chelsea. If it can be determined that there are sixteen main submarkets across the parts of three boroughs, the implication is that there are dozens more across the entirety of the 32 boroughs included in this research.
The research of Bourassa et al. (Bourassa et al., 2003) found – in answer to their research question “Do housing submarkets really matter?” – that there were significant differences in values as a result of distance to the central business district (CBD) and quality of local neighbourhood. Whilst the former has been partially accounted for in this model, the latter, quality of neighbourhood, has not been quantified in any meaningful way. The research of Goodman and Thibodeau (Goodman and Thibodeau, 2003) found that a hierarchy of submarkets should be created in such models and developed parameters for creating this. This tallies with the findings of Bajic (Bajic, 1985) who theorises that the hedonic model should be fitted differently for different submarkets and factors such as the quality of local schools should also be taken into account.
This highlights some of the deficiencies of the approach taken, along with some of the other independent variables that ought to be included in future iterations of the model. Other such variables which could not be included in this model were listed throughout the course of the research with the hope and expectation that data collected in the future will be complete enough to enable their inclusion, thus improving the accuracy of the model. These variables are summarised below:
- Size of the property – this will require individual rental data for properties rather than building rents
- Floor/storey – again, this requires individual rental data so that the variance in rents between apartments on lower and upper floors can be measured
- View/aspect – as with floor/storey, this is likely to have an impact on rental value but requires individual rental data. It also requires some degree of qualitative analysis to determine desirability of view.
- Furnished/unfurnished – it is assumed that this will also impact rental values but may not in all cases. Again, this requires individual rental data
- Parking – this is likely to be of greater importance in certain submarkets, particularly those in outer London
- Other amenities – including, co-working spaces, cinema rooms, bike storage and repair facilities, broadband, roof terraces, etcetera. Much more data is needed and a substantial number of schemes which include these amenities are thought to be required in order to produce statistically significant results.
- Quality of management – this is very difficult to accurately quantify as it will again require a degree of qualitative analysis. Resident satisfaction could be used as a proxy and further variables such as presence of onsite management, response rate and presence of dedicated maintenance teams could also be factored in if enough data was collected.
Two final points relate to the limitations presented by the independent variables which were included. Firstly, the use of ‘Travel time to Oxford Circus’ as a proxy for ‘Convenience of access to the central business district’ is somewhat flawed in a city such as London. Whilst this independent variable did produce a statistically significant coefficient, it neglects the fact that many Londoners do not work or even spend time in the West End. For many, access to the City or Canary Wharf is more important. For others who work locally to their homes, access to central London is not a daily necessity and is therefore less likely to influence the amount that they spend on rent. Once again, this highlights the importance of analysing submarkets in a greater degree of detail and isolation.
Secondly, as highlighted earlier, the classification of quality of finish was subjective and based on the interpretation of the researcher rather than any quantitative factor. Whilst a logical approach was taken and a standard specification was drawn up to identify the ‘mid-point’ thus providing a yardstick against which all buildings could be measured, the individual interpretation of another researcher could vary substantially.
The methodology and statistical model used has returned a number of interesting and worthwhile results. The coefficients of the variables that were tested – i.e. number of bedrooms, presence of amenities, etcetera - are known within the limits of the standard error returned and therefore these ‘items’ in the ‘basket of goods’ can be attributed a rental value. However, the degree of probability with which the variable can be determined to have not influenced the independent variable by chance – or p-value – also needs to be considered in this context, as this gives an indication of the reliability of the individual results.
The three independent variables with which this paper is primarily concerned are the presence of concierge services, gyms and residents’ lounges. As determined by the regression analyses undertaken, none of these variables can be determined to have had a statistically significant impact on rental values. The likelihood that the presence of any of the three amenities in a building affected rental values by chance is between 74.3% and 89.5%, depending on the amenity. This means that the results relating to amenity spaces are not statistically significant. The theory that including amenities in buildings will increase, or even affect, rental values can therefore be understood to be disproved as it has been determined that there is no significant relationship between the independent and dependent variables.
In contrast, a number of the other independent variables were found to have very significant relationships with the dependent variable. For instance, the factors which most substantially impacted rental value were number of bedrooms, ratio of bathrooms, quality of finish and, when looked at cumulatively, travel time to Oxford Circus. The results of the research appear to show that adding an additional bedroom increases rental value by £818 per month on average. The value of a 1:1 bathroom ratio was found to be £1,691 per month. Therefore, if this ratio reduced to 2:1 whilst the number of bedrooms increased by one, then the average uplift would be just £55 per month.
It seems obvious that the premium achieved by a two bedroom unit over a one bedroom unit in the same building would not be either £55 per month or £818 per month, depending on whether it had an additional bathroom or not. However, this interpretation belies the effects of standard deviation, which were £126 for bedrooms and £393 for the ratio of bathrooms. These margins for error is fairly substantial and is indicative of three interrelated key constraints: firstly, the inherent limitations of the model, which can only be used to find average variance; secondly, the fact that by using average rental values from across Greater London limit the utility of the results when applied to a single building in a single location are ; and thirdly, the range of rental values which were taken from submarkets across inner and outer London means that the standard error was always likely to be very high.
The rental values analysed in the data ranged from £790 per month for a one bedroom flat in Barking to £14,300 per month for a three bedroom flat in Mayfair. The range of the data is therefore £13,510, which explains the standard error coefficient returned. The independent variable ‘Travel time to Oxford Circus’ only had a standard error of £7.94 or 18.30% and it is thought that this because it inherently accounts for the effects of different submarkets. It is therefore further hypothesised that further analyses of this nature would benefit from being focused exclusively on submarkets, rather than seeking to derive insights into Greater London as a whole.
While the individual coefficients – or ‘prices’- for each variable – or ‘good’ – may be of limited use, this research has concluded that amenity spaces do not appear to significantly or notably impact on rental values. It also illuminates the hierarchy of factors which are most important in determining rental value: ratio of bathrooms to bedrooms, number of bedrooms, distance from Oxford Circus and quality of finish. As discussed further above, the findings in respect of quality of finish could be used by developers and investors to determine whether additional investment in a rental product would be profitable or worthwhile. Additionally, the hedonic pricing model can also be used to determine likely differences in the rental values between two sites with different travel times to Oxford Circus. Furthermore, it was also found that rental income gradually diminishes as a building ages and these findings can be used in cash flow modelling for Build to Rent investments as a factor which counterbalances rental growth over a given period.
Perhaps the most useful finding of this research is that this methodology can be of use to developers and investors, provided that enough good quality data is collected. Statistically significant results can be found by using regression analyses and creating hedonic pricing models. This, surely, has to be a more scientific and robust approach to guide investment and development decisions than relying on the unsupported presumptions of some consultants and architects or assuming that the lessons that have been learnt in overseas markets are directly transferrable to the London market.
7.1. Further Research Required
In order to improve on these results and to ensure that future analyses using the same methodology are more useful, three key considerations should be taken into account. Firstly, in order to improve the reliability and ‘goodness of fit’ of any regression analyses, more data is needed. At present, not enough data is available for Build to Rent buildings in London, primarily because there are so few. It is understood that all readily available data relating to relevant Build to Rent buildings was used in this analysis but this amounted to just 149 individual datasets; the minimum required to produce any statistically significant results, as specified by Green (1991), is 138 datasets. If more datasets were used, it follows that the results are likely to be more significant.
Secondly, this data should be of good quality and relate to individual properties rather than buildings. This will help to improve the R-squared value of the results as more of the factors which explain the makeup of a rental value will be accounted for in the model. Additional variables such as the floor on which the property is located, the net internal area and whether it benefits from furnishings are all of interest to developers and investors and are thought to account for at least some of the 42.33% of variance in rental values which was unable to be explained by this model. More amenities could be analysed to determine if they impact on rental values and the impact of onsite management resources could also be assessed, giving operators an insight into whether it is cost-effective to situate costly staff in a building on a day-to-day basis.
Lastly, these analyses should be limited to distinct submarkets or areas where rental values are broadly similar in order to reduce the standard deviation of the coefficient and improve utility. As the range in rental values is extremely large across Greater London, this limits the reliability of the model and reduces the usefulness of the coefficient for each variable. As not enough data on Build to Rent buildings in currently available, it is thought that data relating to Build for Sale buildings could be used as well. Any model using data from both Build to Rent and Build for Sale buildings should also include an additional independent variable to account for the tenure of the building and, theoretically, it should be possible to determine whether professional management by an institutional landlord increases rental values.
In summary, the results show that amenities do not have a statistically significant impact on the rental values of Build to Rent buildings in London, although the limitations and deficiencies of the data and methodology are recognised. In order to clarify these findings, future research is needed and additional, richer data should be used. It is hypothesised that even though the findings are somewhat limited, the fact that enough data was used to theoretically produce a significant result means that any future research with more data is also likely to result in the same result. This should, of course, be tested. Further findings were that location has a strong influence on results and future research should therefore be limited to submarkets and that the multiple regression methodology is potentially of great use for Build to Rent investors and developers who require confirmation and quantification for theoretical property appraisals.
Apergis N..and Rezitis A. (2003) ‘Housing prices and macroeconomic factors in Greece: prospects within the EMU’, in Applied Economics Letters, 10, 799–804
Arcadis Group (2016) Build to Rent: Pushing the Boundaries, London: Arcadis Group
Bajic, Validimir (1985) ‘Housing-Market Segmentation and Demand for Housing Attributes: Some Empircal Findings’, in Real Estate Economics, 13:1, pp. 58-75
Barton, J. and Breen, G. (2016) ‘Cost Model: Private Rented Sector’, Available at: https://www.building.co.uk/cost-model-private-rented-sector/5080297.article (Accessed 12/08/2017)
Berridge, R. (2017) ‘Are amenities always the answer for build-to-rent?’, in Property Week, 31st May 2017, Available at: http://www.propertyweek.com/opinion/are-amenities-always-the-answer-for-build-to-rent?/5089528.article (Accessed: 18/02/2017)
Bonnie J. B (1998) ‘The dynamic impact of macroeconomic aggregates on housing prices and stock of houses: A national and regional analysis’, Journal of Real Estate Finance and Economics, 17:2, pp.179-197
British Property Federation, The (2015) Build to Rent: Welcome to the UK's newest housing sector, Available at: http://www.bpf.org.uk/sites/default/files/ resources/BPF-Build-to-Rent-Welcome-to-the-UKs-newest-housing-sector.pdf (Accessed: 18/02/2017).
Carulla, A. (2014) Designing for the private rented sector, Available at: http://prsupdate.co.uk/2014/10/designing-for-the-private-rented- sector/ (Accessed: 18/02/2017).
Chattopadhyay, S. (1999) ‘Estimating the Demand for Air Quality: New Evidence Based on the Chicago Housing Market’, in Land Economics, 75:1, pp. 22-38
Clark, D. E. and Herrin, W. E., (2002) ‘The Impact of Public School Attributes on Home Sale Prices in California’, in Growth and Change, 31
Cluttons (2017) Residential Market Outlook: Autumn 2017, London: Cluttons
Colliers International (2016) 2016: Spotlight on Multifamily, Available at: http://www.colliers.com/-/media/files/marketresearch/unitedstates/2016-research-reports/2016-multifamily-spotlight-report.pdf (Accessed: 18/02/2017).
Colliers International (2016b) Build-to-Rent: Colliers Rental Insights 2016, London: Colliers International
Corsini, K. R. (2009) Statistical Analysis of Residential Housing Prices in an Up and Down Real Estate Market: A General Framework and Study of Cobb County, GA, Georgia: Georgia Institute of Technology
Countrywide [Johnny Morrison] (2016) ‘Paying for more: Premiums in the private rented sector’, The PRS Forum: Conference Papers [Held 22nd February 2016 at Bishopsgate Conference Centre]
Court, A. T. (1939) ‘Hedonic Price Indexes with Automotive Examples’, in The Dynamics of Automobile Demand, New York: General Motors
Damm, D., Lerman S. R., Lemer-Lam E. and J. Young (1980) ‘The Response of Urban Real Estate Values in Anticipation of The Washington Metro’ in Journal of Transport Economic Policy, September, 315 – 336.
Department for Communities and Local Government (2016) English Housing Survey: Private rented sector, 2015-16, London: Her Majesty’s Stationery Office
Department for Communities and the Local Government (2017a) Fixing our broken housing market [Government White Paper], London: Her Majesty’s Stationery Office.
Department for Communities and the Local Government (2017b) Planning and Affordable Housing for Build to Rent: A consultation paper, London: Her Majesty’s Stationery Office.
Des Rosiers F., Theriault M., Kesten Y., Villeneuve P. (2002) ‘Landscaping and house values: an empirical investigation’, Journal of Real Estate Research, Vol. 23(1/2), pp. 139–162.
Doan, M. C, (1997) American Housing Production, 1880-2000: A Concise History, Lanham: University Press of America
Donovan G. H., Butry D. T. (2010) ‘Trees in the city: valuing street trees in Portland, Oregon’, Landscape and Urban Planning, Vol. 94, pp. 77–83.
Fletcher, M., Gallimore P. and J. Mangan (2000) ‘The Modeling of Housing Sub-markets’, in Journal of Property Investment & Finance, 18(4).
Forrest D., Glen J. and Ward R. (1996) ‘The impact of a Light Rail System on the Structure of House Prices: A Hedonic Longitudinal Study’, in Journal of Transport Economics and Policy, 30:1, pp. 15-29
Frost, J. (2013) ‘A Tribute to Regression Analysis’, on Minitab Blog, Available at: http://blog.minitab.com/blog/adventures-in-statistics-2/a-tribute-to-regression-analysis (Accessed: 12/08/2017)
Glindro E. T., Subhanij T., Szeto J. and Zhu H. (2008) ‘Determinants of house prices in nine Asia-Pacific economies’, Bank for International Settlements Working Papers No. 263
Goodman, A. C. (1998) ‘Andrew Court and the Invention of Hedonic Price Analysis’, in Journal of Urban Economics, 44, pp. 291-298.
Goodman, A. C., and T. Thibodeau. (1995) ‘Age-Related Heteroskedasticity in Hedonic House Price Equations’, in Journal of Housing Research, 6(1), pp. 25–42.
Goodman, A. C., and T. Thibodeau. (2003) ‘Housing Market Segmentation and Hedonic Prediction Accuracy’, in Journal of Housing Economics, 7(2), pp. 121-143.
Green, S. B. (1991) ‘How many subjects does it take to do a regression analysis?’, Multivariate Behavioral Research, 26, pp. 499-510.
Guntermann, K. L. and Norrbin, S. (1987) ‘Explaining the Variability of Apartment Rents’, Real Estate Economics, Vol. 15: pp. 321–340.
HomeLet (2017) July 2017 HomeLet Rental Index [Website], Available at https://homelet.co.uk/homelet-rental-index (Accessed 12/08/2017)
Kain, J. and Quigley, J. (1970) ‘Measuring the Value of Housing Quality’, in Journal of the American Statistical Association, 65, pp. 532-548.
Kestens Y., Thériault M., Des Rosiers F. (2004) ‘The impact of surrounding land use and vegetation on single-family house prices’, Environment and Planning B, Vol. 31, pp. 539–567.
Ketkar, Kusum. (1992) ‘Hazardous Waste Sites and Property Values in the State of New Jersey’, in Applied Economics, 24. 647-59
Knight Frank (2016) The Knight Frank Tenant’s Survey 2015/16, London: Knight Frank
Lancaster, K. (1966) ‘A New Approach to Consumer Theory’, in the Journal of Political Economy, 74:2, pp. 132-157
Leggett, C. and Bockstael, N. (2000) ‘Evidence of the Effects of Water Quality on Residential Land Prices’, in Journal of Environmental Economics and Management, 39:2, pp. 121-144
Li, W. and Saphores, J. (2011), ‘A Spatial Hedonic Analysis of the Value of Urban Land Cover in the Multifamily Housing Market in Los Angeles, CA’, Urban Studies Journal, Vol. 49, No. 12, pp. 2,597-2,615
Limsombunchai V., Gan C. and Lee M. (2004) ‘House price prediction: hedonic price model vs. artificial neural network’, American Journal of Applied Sciences, 1:3, pp. 193-201.
Long, S. (2016), Quarterly Analysis: PRS, November 2016, London: Molior.
Long, S. (2017), Quarterly Analysis: PRS, August 2017, London: Molior.
Marks, D. (1984), ‘The effect of rent control on the price of rental housing: an hedonic approach’, in Land Economics 60, 81-94.
Mieszkowski, P. and A. M. Saper (1978) ‘An Estimate of the Effects of Airport Noise on Property Values’ in Journal of Urban Economics, 5: 425 – 440.
Monson, M. (2009) ‘Valuation using Hedonic Pricing Models’, Cornell Real Estate Review, Vol. 7, pp. 62-7
Malpezzi, S., Ozanne, L. and Thibodeau, T. (1980), Characteristic Prices of Housing in Fifty-Nine Metropolitan Areas, Washington DC: The Urban Institute
Malpezzi, S. (2002), ‘Hedonic Pricing Models: A Selective and Applied Review’, in Housing Economics and Public Policy, Oxford: Blackwell
Prashardes P. and Savva C. S. (2009) “Factors affecting house prices in Cyprus: 1988-2008.” Cyprus Economic Policy Review, 3:1, pp.3-25
Property Week (2016) ‘Built to Run Build to Rent?’, in Property Week, 19th August 2016, pp. 36-7
Ridker, R. and J. Henning (1982) ‘The Determinants of Residential Property Values with Special Reference to Air Pollution’, Review of Economics and Statistics, 46: pp. 246 – 257.
Rodriguez M., Sirmans C. F. (1994) ‘Quantifying the value of a view in single-family housing markets’, in Appraisal Journal, 62, pp. 600–603
Rosen, S. (1974) ‘Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition’, in Journal of Political Economy, 82:1, pp. 34-55
Runkel, P. (2016) ‘What Are T Values and P Values in Statistics?’, on Mintab Blog, Available from: http://blog.minitab.com/blog/statistics-and-quality-data-analysis/what-are-t-values-and-p-values-in-statistics (Accessed 12/06/2017)
Saint, A (1989) Politics and the people of London: the London County Council (1889–1965), London: Bloomsbury Publishing
Sirmans, G. S. and Benjamin, J. D. (1989) ‘Determining market rent: the value of amenities, services and external factors’, Journal of Real Estate Research, Vol. 4, No. 2, pp. 33-43
Sirmans, G. S., Macpherson and Zietz (2005) ‘The Composition of Hedonic Pricing Models’, in Journal of Real Estate Literature, 13:1, pp. 3-43
Straszheim, M. R. (1975) An Econometric Analysis of the Urban Housing Market, National Bureau of Economic Research: NBER Research
Trust for London (2017) Making the Most of Build to Rent, London: Future of London
Tse, R. Y. C. (2002) ‘Estimating Neighbourhood Effects in House Prices: Towards a New Hedonic Model Approach’, in Urban Studies, 39:7
US Census Bureau (2015) American Community Survey, 1-Year Estimates, Washington DC: US Census Bureau (updated 10/2016)
Uyeno, D., Hamilton S. W. and A. J. G. Biggs, 1993. The Density of Residential Land Use and The Impact of Airport Noise. J. Transport Economics and Policy, January, XXVII(1).
Veldkamp, J. (2015) ‘How to capitalise on the opportunity of PRS’, Available at: http://www.chapmantaylor.com/en/insights/article/uk-residential-how-to-capitalise-on-the-opportunity-of-prs/en/ (Accessed: 18/02/2017).
Warren-Myers, Georgia. (2015). Editorial: ‘Implications of Valuation "Art and Science" in decision making’. Journal of Property Investment and Finance. 33. 10.1108/JPIF-01-2015-0001.
Wood, Carl (2017), ‘Build to rent: hitting the sweet spot’, Available at: http://www.turnerandtownsend.com/en/insights/build-to-rent-hitting-the-sweet-spot/ (Accessed 12/08/2017)