1 Introduction

Sub-divided units (SDUs) have become an emerging accommodation option for the urban poor in Hong Kong. They are formed by sub-division of individual quarters into two or more units for rental purposes to more than one household. “Quarter” refers to the space enclosed by walls in a residential, industrial, commercial and other types of buildings for living or other purposes. It must be internally connected and externally accessible (C&SD 2016a). In general, there are two types of SDUs, observable physical partitions and the provision of independent toilet are their major differences. Tenants have to share the common toilet and access to water supply with other SDU households in the same quarter if their SDUs are without independent sanitation facilities.

The indoor environments of SDUs are perceived to be poor (Lai et al. 2017). Most inhabitants live in overcrowded units, while some even endure dilapidated conditions and absence of basic housing facilities such as independent toilet and window (C&SD 2016a; Ng 2017). The average living space per capita was 47.8 sq ft, which was much below the government standard for overcrowded households (75.3 sq ft) (Yiu et al. 2015; Hong Kong Housing Authority 2017a). SDUs are mostly found in private domestic buildings aged 25 years and above and more SDUs were found in older buildings (Policy 21 Limited 2013). The presence of SDUs imposes extra burden on the ageing building structures. According to Caritas Hong Kong (2015), about 50% of the SDU buildings had at least 11 more units after subdivision, posing safety risks to tenants, buildings and the public.

SDUs are regarded as informal housing due to the unauthorized building works (UBWs) and tenure insecurity. UBWs are commonly found in SDUs, making most of the units illegal, or likely to be illegal. On the other hand, most of the tenancy agreements are not stamped and registered in the Lands Registry. Both landlords and tenants are liable to civil proceedings by the Collector of Stamp Duty of the Inland Revenue Department (Community Legal Information Centre 2017). Enforcement of agreement will also face difficulties as the Court may not accept an unstamped tenancy document as evidence. Albeit being substandard and insecure, SDU market had grown rapidly. It accommodated 171,300 people in 66,900 SDUs in 2013 (Policy 21 Limited 2013). The number of SDUs had increased to 92,700 in 2016, accommodating nearly 209,700 people (C&SD 2018). The growth of SDU market indicates that the formal sector did not adequately meet the demand for affordable housing. Conceptually, SDUs are informal institutions devised to supply housing alternative to meet the shortage of formal affordable housing. According to North (1990), institutional changes emerge when both parties do better with an altered agreement. In SDU context, landlords receive higher aggregate rental income (Lai et al. 2015) while tenants rent shelters at lower costs. It sustains as long as the shortage of affordable housing in the formal sector continues.

There are two major sources for the growing SDU demand. First, 42% of the SDU households were Public Rental Housing (PRH) applicants (C&SD 2016a). The average waiting time was 4.7 years before they could successfully be allocated with rental units (Hong Kong Housing Authority 2018a). Yet, before the allocation, most PRH applicants’ are excluded from the formal housing sector because within the eligibility of PRH, they could hardly afford the market rent of formal private property (Hong Kong Housing Authority 2018b; RVD 2017a). Thus, SDUs become their major affordable choice.

Another source, which has been growing rapidly, is the sandwich class.Footnote 1 They are ineligible for public housing while they cannot afford to own private property. The situation had worsened from 2006 to 2016 as the increase in household income (45%) (C&SD 2016b) could not keep up with that of rent (105%) and property price (264%) (RVD 2017b, c). Households had to spend higher proportion of their income on housing expenses. Some are forced to move into SDUs when they find the formal units unaffordable. The median SDU rent was HK$4200, which was about one-third of the average rent of formal private property (429 sq ft) in 2016 (C&SD 2016a; RVD 2017a). However, the low rent is achieved at the expense of living space and housing facilities. The median SDU size was 111 sq ft, some SDUs even lack basic facilities for sanitation and ventilation (C&SD 2016a). While in most of the previous housing studies, their availabilities were presumed, this paper thus attempts to examine the SDU rent determinants with the focus on the effects of basic facility deficiencies on rent.

Besides, as rental information of informal SDU market is not readily available, high information costs are involved. Observable attributes such as independent toilet, window, floor area, residential services of the buildings are expected to be important rent determinants as households mainly rely on their availabilities when they assess SDU values. For SDUs with shared sanitation facilities, increase in subdivision density is hypothesized to be negatively correlated with the rent as inconvenience, privacy and security concerns of sharing the common toilet with other households escalate.

This is a pioneer attempt to study the SDU rental market in Hong Kong. The results reveal the effects of housing attributes of an inadequate housing market. They are important references to policymakers and non-governmental organizations on social policies and affordable housing design. Also, this study sheds lights on the liveability issues of substandard housing. Households are observed to make trade-offs of housing necessities in order to obtain an affordable home. If the shortage of affordable housing worsens and no timely, feasible and affordable housing alternative is available, their living conditions may continue to deteriorate. They have to squeeze into even smaller units as suggested by Yiu et al. (2015) and more priced housing facilities have to be forgone.

2 Literature review

According to UN-Habitat (2015), characteristics of informal settlements include: (1) inhabitants do not possess tenure security; (2) the neighbourhoods are in lack of basic services and city infrastructure and (3) the housing does not comply with the planning and building regulations. Modalities of informal settlements range from squatters (Mohanty 2006), urban villages (Wang et al. 2009), subterranean housing (Kim 2016), small property right ownership (Webster et al. 2016) and SDUs (C&SD 2015, 2016a, 2018; Policy 21 Limited 2013). Informality can be defined in terms of presence/absence, strength or degree of property rights (Webster et al. 2016). Rights over property can be established by formal and/or informal rules (Barzel 1989) and both rules are elements for the stability of institutional framework (North 1990), SDU is one example of the latter. Emergence of informal settlement can be regarded as institutional change which modify and supplement the formal housing. The positive effects bought by informal settlements include improving urban poor’s resource entitlements and lessening housing starvation (Webster et al. 2016).

To examine the effects of housing attributes, hedonic pricing models have been frequently employed. Theoretically, it stems from Lancaster’s new approach to consumer theory (1966). It suggests that consumers’ utility is derived from the properties or characteristics but not from the goods as direct objects. This new branch of microeconomic theory provided an important foundation for analysis of utility generating function of the goods’ characteristics (Malpezzi 2003). Rosen (1974) interpreted hedonic method theoretically and extended it to the housing market. Goods are valued for their utility bearing attributes and a set of implicit prices are defined by the bundles of characteristics. The consumer and producer’s decisions reach the equilibrium point and they are assumed to maximize their behaviour with joint-envelope function. Hedonic pricing theory rests on the entire set of implicit prices, which guides the decisions of both consumer and producer.

After Rosen’s paper, hedonic pricing method has been widely adopted to examine the effects of individual housing characteristics on housing prices or rental values (Allen et al. 1995; Bizimana 2011; Babalola et al. 2013; Cebula 2009; Kiel and Zabel 2004; Shultz and King 2002; Sirmans et al. 2005). Three major categories of attributes have been frequently studied, they are structural specifics, locational and environmental attributes. Floor area is suggested to be the single most important structural variable across nations (Chin and Chau 2003), which is positively correlated with housing price (Browne et al. 2008; Coulson and Leichenko 2001; So et al. 1997; Wilkinson 1973; Zietz, et al. 2008). Besides, housing units located on upper floors are found to have higher property values (Cebula 2009; Coulson and Lahr 2005; Hui et al. 2007; So et al. 1997). Coulson and Lahr (2005) further suggested that property prices appreciated faster for buildings with more storeys. In Hong Kong, housing units on higher floors are also perceived to be better due to the serenity and better view (So et al. 1997). Yet, in the context of SDUs, which are mostly found in Tong-laus. They are mainly four- to seven-storey buildings without elevators installed. Inconvenience of walking up and down stairs is expected to bring negative effect on the rental values. The actual effect is thus an empirical question.

For number of bathrooms, it generally brings positive effect on the property values under the presumption that at least one bathroom is available (Garrod and Willis 1992; Allen et al. 1995; Carroll et al. 1996; Coulson and Leichenko 2001; Kiel and Zabel 2004; Sirmans et al. 2005). Tse (2002), however, suggested the opposite results with the study of Hong Kong. Households may prefer more living space to more bathrooms as housing units in Hong Kong are generally smaller than the single-family dwellings in the Western countries. Meanwhile, building age is observed to bring negative impact (Kain and Quigley 1970; Babalola et al. 2013; Hui et al. 2007; Zietz et al. 2008). For older buildings, more repair and maintenance costs are incurred. Advancement in building design, electrical and mechanical systems also reduces properties’ usefulness (Chin and Chau 2003). Yet, its effect may be lessened in the context of historical neighbourhood where the cachet effect increases the value of the properties (Coulson and Lahr 2005).

Regarding accessibility, earlier studies used distance to city center as a proxy to examine its effect (Allen et al. 1995; Geoghegan et al. 1997; Hui et al. 2007; Kiel and Zabel 2004; Wilkinson 1973; Zietz et al. 2008). Development of transport networks enables decentralization of business activities. Therefore, some studies began to deploy the availability of transport facilities in property price analysis. Property values decrease with the distance from the Central Business District (CBD) (Hui et al. 2007) and metro stations (Hui et al. 2007; Tse 2002). Proximity to minibus station is also observed to be influential and positively related to the housing prices of the middle-income class (So et al. 1997).

Pollution levels have also been found to exert impact on property values. There has been a general agreement on the negative influence of air pollution (Cebula 2009; Chattopadhyay 1999; Hui et al.; 2007; Ridker and Henning 1967; Zabel and Kiel 2000). For ambient noise level, it is associated with the population density and the traffic volumes. They are perceived to have negative impact on property prices (Chau et al. 2003). Yet, a Hong Kong study shows the opposite results (Hui et al. 2007). Heavy traffic, buzzing streets and construction works are the major noise sources. They are associated with higher accessibility, greater sales and business transactions. The results imply that people in this densely-populated city prefer accessibility to serenity. Regarding the effects of airport proximity, Tomkins et al. (1998) found that the positive impact of improved access outweighed the negative effects by higher noise levels. However, Espey and Lopez (2000) observed the opposite results. Bell (2001) further suggested that high priced homes were influenced by larger reduction in values than low priced homes. Light pollution has also been suggested to cause environmental degradation (Pun et al. 2014) but its effect has yet been explored. This study thus become a pioneer attempt to examine its effect on rent.

Apart from the selection of explanatory variables, previous studies also shed lights on market segmentation of property price analysis. Several types of sub-markets are likely to exist in most markets due to heterogeneity in buyers’ characteristics, affordability and preferences. It is unrealistic to treat the housing market as a single entity (Chin and Chau 2003). Previous studies also suggest that segmentation increases the explanatory power of the hedonic pricing models. Difference in implicit pricing and marginal prices of dwelling characteristics are observed to vary across household profiles (Palmquist 1992; Xu 2008) and property types (Allen et al. 1995). Zietz et al. (2008) also suggested that housing characteristics were valued differently across a given distribution of house prices. Being a submarket of Hong Kong, SDU market operates with its unique informality and non-typical housing conditions. It is expected to have different pricing on dwelling characteristics.

Rental market has rarely been studied in Hong Kong since the data are not available. Thus, previous studies mostly focused on property prices (Chau et al. 2001; Hui et al. 2007; Mok et al. 1995; So et al. 1997). Regarding SDU surveys, Policy 21 Limited (2013) and C&SD (2015, 2016a, 2018) provided estimations on the number of SDUs, their geographical distribution and information about housing conditions and household characteristics. Others are single-district based (Alliance for North District Grassroots’ Rights 2015; Caritas Hong Kong 2014, 2015; Kwai Chung Sub-divided Unit Tenants Union 2014). Tenants’ immediate concerns on housing qualities, water and electricity surcharges and housing policies were reported.

Lai et al. (2015) constructed SDU rental index which revealed the disparity between formal and informal rental values. Meanwhile, Yiu et al. (2015) used quality adjusted rent-to-income ratio to measure the actual change in SDU households’ affordability by controlling the reduction in living space. It reveals that tenants would have to spend even higher proportion of income on rent, increasing from 41.1% to 72.8%, if they kept the same living space two years ago. A recent study attempted to examine the SDU rent (Huang 2017), however, it omitted a very important explanatory variable of SDU market, which is the presence of independent toilet. This could explain why the increase in SDU size has negative impact on rent, which is in contrast with the general agreement from the previous literature. Besides, units for lease listed online may not be SDUs but normal rooms of shared flats. It weakens the explanatory power of the models and the real SDU market conditions could not be fully reflected.

In most of the previous property price literature, housing units were assumed to be equipped with basic housing facilities such as bathrooms and windows. Values of home improvement bought by extra facility were examined (Coulson and Lahr 2005). Yet, under the inadequate conditions of SDUs, the provision of basic facility would have different values to the inhabitants. It measures the value of necessity, instead of value of better convenience. Limited studies were conducted to examine the effect of housing facility deficiency. Babalola et al. (2013) attempted, however, availability of sanitation facility was omitted in the model due to exact collinearity. Kim (2016) conducted rental analysis of Beijing’s informal subterranean housing using the initial listed price and amenities stated online. The results show that ventilation facility and locational attribute are significant rent determinants. However, the attributes are mainly self-report amenities posted on rental website which may not reflect the actual conditions of the units as defects may be hidden by the landlords.

It is expected that the special housing environment of SDUs give rise to distinct implicit pricing of housing attributes. As the rental information is not available, it renders high information and search costs, tenants could only assess the SDU values based on the availability of the observable housing attributes and verbal information by agents. According to Chau and Wong (2016), value of service flow from an office is decomposed into symmetric and asymmetric parts. Symmetric part refers to the quality which is observable by everybody such as accessibility and view, while asymmetric part includes the building quality which can be assessed more accurately by landlord than the prospective tenants. Applying this concept to the SDU market, observable housing facilities, such as window, toilet, floor area and residential services, provide sources of information about the SDU values. Tenants assess the units mainly based on their availabilities. Thus, this study hypothesizes that basic and observable housing facilities are significant SDU rent determinants. When the sanitation facilities are shared among households, increase in subdivision density is expected to cause negative impact on rent. This study aims to provide empirical evidence about the market landscape and the implicit pricing of facility trade-offs of this emerging informal housing market.

3 Method

A property is assumed to be sold as a package of housing attributes (Rosen 1974; So et al. 1997). To estimate the effects of housing attributes on the property values, hedonic pricing model has been frequently adopted (Kim 2016; Leggett and Bockstael 2000; Wallace and Meese 1997; Xu 2008). The marginal attribute prices can be obtained through the parameter estimates of the hedonic pricing function (Chin and Chau 2003). In this study, hedonic pricing model is deployed to examine the SDU rent determinants. Since there are not less than 92,700 SDUs available for selection and they are geographically scattered in different parts of Hong Kong with 52,700 units in Kowloon, 21,900 units in the New Territories and the remaining on Hong Kong Island (C&SD 2018), hedonic model is valid for the rental analysis. The choice is still based on the hedonic model assumptions, except the two foci of this article on their informality and market information availability.

3.1 Selection of housing attributes

13 housing attributes are selected based on previous housing price literature, the inadequate housing conditions of SDUs and the special context of the rental market. They can be grouped into the following four categories:

Structural characteristics include floor area (AREA), building age (AGE), floor level (FLOOR), independent toilet (TOILET), number of window(s) (WIN4), number of SDUs within one housing unit (NSDU), partitioning quality (QIP) and internal conditions (QIINT). Environmental attributes measure the pollution levels from the surrounding environment (QIEXT) and residential services of the building (QISER). Locational attribute measures the accessibility of the SDUs by the availability of public transport facilities (QITRA). Contractual conditions refer to the provision of furniture and electrical appliances (QIFUR) and the measurements of electricity and water charges (QIEW). Descriptions and descriptive statistics of all variables are summarized in Table 1.

Table 1 Variable descriptions and descriptive statistics
Full size table

3.2 Measurements of housing attributes

Regarding the measurements of the housing attributes, some are self-explanatory, e.g. floor area (AREA), building age (AGE) and floor level (FLOOR). SDUs are non-typical housing units in which some basic housing facilities e.g. independent toilet and window are absent. According to Lai et al. (2017), provision of housing facilities is suggested to directly influence the SDU living environment. Thus, their availabilities are included in the rental analysis of this study and positive effects are expected. In the empirical models, the effects of window availability are separately measured in four categories by using three dummy variables, namely WIN4_12, WIN4_34 and WIN4_5OA to clearly reflect the effect of window deficiency and if there is non-linearity in facility effects with the increase in quantity. The base group denotes the SDUs without window, WIN4_12, WIN4_34 and WIN4_5OA measure SDUs with one to two, three to four and five or above windows. The estimates on the three dummy variables thus measure the proportionate difference in rent relative to SDUs without window.

TOILET measures the availability of independent toilet. It is a dummy variable, 1 denotes its presence while 0 represents its deficiency and tenants have to share the common toilet with other tenants in the same housing unit, in which usually only one toilet is provided. The estimate of this attribute is in nature different from the effect of having extra toilet which has been conventionally examined (Garrod and Willis 1992). The number of toilet measured is not least one, but from zero to one, reflecting the inconvenience caused by sharing toilet and access to fresh water. NSDU counts the number of SDUs within one quarter. Here, we could refer a quarter in private domestic building to a housing unit. Although SDUs are separate units, households within one housing unit have to share the common areas like the corridor with each other. Negative effect is expected because there would be more hygiene, privacy and security concerns with the increase in the number of SDUs.

The following seven attributes are measured by quality scores based on the quality assessment scheme of SDUs’ special housing environment. Scores for each attribute are given with reference to the housing conditions during the on-site visits. The measurements include counting the types of residential services provided, number of pollution aspects, level of accessibility as well as the internal and contractual conditions of the SDUs. This method helps to the improve accuracy and objectivity of the data as compared to self-report questionnaires by the tenants and online rental advertisements stated by the landlords. “Appendix” shows the detailed assessment criterion of all quality score attributes.

QIEXT denotes the levels of air, noise and light pollution. The surrounding environment of the SDUs are graded based on the number of the above aspects affecting the SDUs and whether the occupants are affected in the daytime and/or the nighttime. The lowest score is given to SDUs which are with all three aspects of the pollution problems for the whole day while the highest is given to the SDUs which are unaffected by anyone of them. For QISER, it measures the number of items of residential services provided by the building. The assessment criteria include (1) building entrance lock; (2) security service; (3) waste collection; (4) public lighting. As most of the SDU buildings are without property management (Caritas Hong Kong 2015), the above residential services may not be provided. The quality scores range from zero to four based the availability of the above services. More residential services are preferred as they either improve security level or mitigate occupants’ inconvenience.

QITRA is the locational attribute of the SDUs. It ranges from one to three. Highest score, three, is given to SDUs which could be reached from the nearest MTR station within 10 minutes. Two denotes the SDUs which could not be reached from MTR station in 10 minutes but there is public transport available to get to the metro station. Otherwise, one is given to the SDUs with the least accessibility. QIFUR is the provision of furniture and electrical appliances. The quality scores range from zero to two depends on whether none, either furniture or electrical appliances, or both of them are provided by the landlords. Households could save the initial costs if the landlords provide them with furniture and electrical appliances.

QIEW measures the availability of electricity and water meters. Lai et al. (2015) revealed the electricity surcharge problem faced by SDU tenants. Landlords were observed to overcharge utility fees by higher rate or without according to tenants’ actual usage. On average, tenants were overcharged by 2.41%. Installation of meters may help lessen the surcharge problem. Ranging from zero to three, this attribute denote no meter installed, either one is installed, both are installed but with surcharge problem or without surcharge problem. Higher scores are preferred. QIP is the partitioning quality i.e. the shape of the SDUs. The quality is measured by five levels ranging from “unable to place a six feet single bed” to “rectangular shape, with partition for bedroom and living room”. Holding the effects of floor area and other attributes constant, we would like to examine the effect on rents due to partitioning quality difference. Higher scores are given to SDUs with more practical partitioning. For all of the above attributes measured by quality scores, positive signs are expected.

QIINT is included in the model to control the effect of internal conditions of SDUs on rent. As most SDUs are found in old residential buildings, structural problems such as peeling paint, water seepage, exposure of bar tendons are commonly found. The conditions of the interior are expected to affect the rent. Five levels of internal conditions are set based on the presence and seriousness of the above problems. Higher scores are given to SDUs with fewer structural problems. However, due to the fact that internal conditions of the SDUs may be correlated with unobserved factor, this attribute is mainly used as a control variable. It is to avoid the possible omitted variable bias caused by excluding the effect of the internal conditions. Table 2 presents the references for six SDU-specific attributes in the empirical models, including WIN4, NSDU, QIP, QISER, QIFUR and QIEW.

Table 2 References for SDU-specific attributes
Full size table

3.3 Empirical models

Mathematically, the hedonic pricing model is:

Model 1

$$\begin{aligned} \ln Rent_{i} & =\upbeta_{\uptheta} +\upbeta_{1} AREA_{i} +\upbeta_{2} AGE_{i} +\upbeta_{3} FLOOR_{i} +\upbeta_{4} WIN4\_12_{i} +\upbeta_{5} WIN4\_34_{i } \\ & \quad + \,\upbeta_{6} WIN4\_5OA_{i } +\upbeta_{7} TOILET_{i} +\upbeta_{8} NSDU_{i} +\upbeta_{9} QIP_{i} +\upbeta_{10} QIEXT_{i} \\ & \quad + \,\upbeta_{11} QISER_{i} +\upbeta_{12} QITRA_{i } +\upbeta_{13} QIFUR_{i} +\upbeta_{14} QIEW_{i} +\upbeta_{15} QIINT_{i} +\upvarepsilon_{i} \\ \end{aligned}$$
(1)

where lnRenti, natural logarithm of the monthly rent of SDUi; β1–β15, coefficients of housing attributes; i, identification number of SDU in the dataset; εi, error term.

In the model, the dependent variable is natural logarithm of the SDU monthly rent and the independent variables are the housing attributes of the SDUs. Semi-log model is adopted based on the context of this study. Most of the attributes are measured by counting, quality scores or their availabilities. Their marginal effects on rent in percentage term due to a level, a unit change or the presence/absence of housing attributes can thus be clearly expressed by the coefficient estimates. We also perform the Ramsey RESET test to examine whether linear model and/or semi-log model have missed important nonlinearities. Linear model shows the sign of functional misspecification while semi-log model does not. Thus, on the basis of RESET, semi-log model is preferred.

The effect of subdivision density is further studied by Model 2. As mentioned earlier, the availability of independent toilet is the major difference between two types of SDUs (C&SD 2016a). Usually tenants have to share only one common toilet and the access to water supply with other tenants if their SDUs are without independent toilet. Thus, higher subdivision density, i.e. more households sharing one toilet, implies that inconvenience is escalated. They have to wait especially in the early morning and the nighttime. Besides, common facilities in general have poorer hygienic conditions than the private ones. More security concerns also arise with the denser subdivision. Therefore, the effect of subdivision density on rent is expected to be negative. Households are expected to pay more rents to mitigate the inconvenience by renting SDUs with lower subdivision density. In Model 2, we include an interaction term (1-TOILET)*NSDU to examine the effect of subdivision density on rents for SDUs with shared toilet. It is obtained by NSDU (number of SDUs in one housing unit) multiplied by (1-TOILET), in which SDUs without independent toilet will be denoted by “1”.

Model 2

$$\begin{aligned} \ln Rent_{i} & =\upbeta_{\uptheta} +\upbeta_{1} AREA_{i} +\upbeta_{2} AGE_{i} +\upbeta_{3} FLOOR_{i} +\upbeta_{4} WIN4\_12_{i} +\upbeta_{5} WIN4\_34_{i} \\ & \quad + \;\upbeta_{6} WIN4\_5OA_{i} +\upbeta_{7} (1 - TOILET)_{i} +\upbeta_{8} NSDU_{i} +\upbeta_{9} QIP_{i} +\upbeta_{10} QIEXT_{i} \\ & \quad + \;\upbeta_{11} QISER_{i} +\upbeta_{12} QITRA_{i} +\upbeta_{13} QIFUR_{i} +\upbeta_{14} QIEW_{i} +\upbeta_{15} QIINT_{i} \\ & \quad + \;\upbeta_{16} (1 - TOILET)*NSDU +\upvarepsilon_{i} \\ \end{aligned}$$
(2)

where lnRenti, natural logarithm of the monthly rent of SDUi; β1–β16, coefficients of housing attributes; i, identification number of SDU in the dataset, εi, error term.

3.4 Data

Data were collected through face-to-face interviews and on-site visits from July 2015 to April 2016. A questionnaire was designed to collect data about SDUs’ rental details, housing attributes and contractual conditions. Interviews were mainly arranged by two means, referrals by non-governmental organizations and non-scheduled visits. The latter were conducted to mitigate the possible sampling bias from referrals. Private domestic buildings aged 25 and above were randomly selected with reference to the SDU distribution by C&SD (2015), which suggested that 57.7% SDUs were located in Kowloon; 22.4% were in the New Territories and 19.9% were on Hong Kong Island. The reason for confining the building age is that 99.9% of the housing units with SDUs were found in private domestic buildings aged 25 years and above (Policy 21 Limited 2013).

SDUs were identified based on several criteria, including numbers of mailboxes, electricity meters, water meters, door bells, thickened floor screeding, presence of multiple doors (usually with separate room numbers), etc. Although this method was time consuming, it was an effective way to find SDU tenants. This method was also adopted by C&SD (2015, 2016a). Moreover, building records were purchased from the Buildings Department to mitigate the error of self-report data on floor area. The exact SDU size was measured during the visits with reference to the building records. After the data collection, adjustments with regards to the contract commencement date were made. The monthly rent of each SDU was adjusted to October 2014 by using the rental indices published by the Rating and Valuation Department (2017b).

In total, 225 buildings were visited and 71 face-to-face interviews were conducted. The study areas cover three major parts of Hong Kong, namely Kowloon, the New Territories and Hong Kong Island. As most subdivision involves unauthorised building works, which are in breach of the Buildings Ordinance (2012). Their rental arrangements are conducted informally without legal tenancy agreements. The illegality resulted in difficulties in data collection. Some tenants were reluctant to be interviewed or unwilling to provide rental details. Although the sample size is small, this method can thoroughly assess the SDU quality and examine the implicit pricing of housing attributes.

4 Empirical results

The median SDU floor area of the dataset is 89 sq ft, which is only 21% of that of all domestic households in Hong Kong in 2016 (C&SD 2018). On average, each housing unit is sub-divided into 4.63 SDUs. The median monthly rent is HKD3993.7. Table 3 shows the results of the coefficient parameter estimations of Model 1. The adjusted R2 is 0.706, which is reasonably high as 70.6% of the data variance could be explained by the model.

Table 3 Regression results of Model 1
Full size table

As expected, gross floor area (AREA) is a positive and significant SDU rent determinant. One square foot increase raises the SDU rent by 0.2%. For building age (AGE), it ranges from 30 to 71 and the mean value is 47.8. It is not a significant rent determinant though younger buildings are generally conceived to have higher rental values. The reason may be that buildings, which are over 30 years old, are considered as old buildings in Hong Kong context because most of them are reinforced concrete structures with a design life span of 50 years. The housing quality factors beyond the ones included in the models may not vary a lot within the age range.

In the dataset, floor level (FLOOR) does not exert significant effect on rent. As explained in earlier section, the negative effect of walking stairs may mitigate the benefits of better view and less polluted environment on the upper floors. Meanwhile, the results show the significant effect of window deficiency. If the SDU do not have a window, the rental price will be 34.4% lower than those with one to two windows (WIN4_12). The facility effect on rent is increasing at a decreasing rate. There is 8.9% marginal increase in rent if the number of windows raises to three to four (WIN4_34), another 6.2% increase when the SDU are with five or above windows (WIN4_5OA).

Another important result is that if the SDU has an independent toilet (TOILET), the rent will be increased by 22.2%. Unlike the conventional studies which deployed “TOILET” to measure the value of better convenience brought by an extra toilet, this study uses this attribute to measure the value of necessity. In our dataset, we observe that SDU households either have their own independent toilet or share one common toilet with other households in the housing unit. Tenants expressed their worries towards shared sanitation facilities during the on-site visits. All of the respondents did household chores by themselves and various chores require water. Since they are mostly low-income households, they do not opt for paid laundry or household cleaning services as these will further increase their financial burden. It will be much more convenient if they have sanitation facilities inside their own SDUs. There are also security and hygiene concerns, especially for households with children. From the data, 90% of them chose SDUs with independent toilet and their concerns are clearly reflected by the empirical result.

Subdivision density (NSDU), i.e. number of SDUs in one housing unit, is not a significant attribute in Model 1. The reason may be that some SDUs are with independent toilet and direct access to the public corridor or street. Thus less concerns on hygiene and security are resulted. For these households, the number of SDUs within one housing unit may not exert significant influence on their daily lives and thus the rents. Meanwhile, pollution levels of the SDUs are observed to exert significant impact on rent (QIEXT). Less polluted environment is preferred. Holding other attributes constant, SDU rents on average increases by 3.1% for each quality score of QIEXT.

The results also shed light on the importance of residential services (QISER). The provision of entrance lock, security service, waste collection and public lighting, were assessed during the interviews. On average, 11.3% increase in rental value is observed for each quality score of residential services. SDUs are mainly found in buildings like Tong-laus or old shop houses which usually do not have property management. Unlike private housing estates, waste collection, public lighting and common area cleaning services may not be provided. Without waste collection, households have to take the garbage downstairs or even to the waste collection depot. They may also have to clean the common area like corridor and back staircases.

For buildings without entrance lock and security service, anyone could enter the building with ease. Although we do not have information regarding the crime situations of SDUs in Hong Kong, according to Zhang and Jin (2015a, b), high crime rate was observed in group oriented leasingFootnote 2 in China. Three features, namely separation, openness and mobility among tenants are conducive to high crime rate. Security service and entrance lock could help to prevent strangers from entering into the building and the security level could be better maintained. From the in-depth interviews, households expressed their security concerns, yet, trade-offs had to be made as they needed to save on housing expenses. Residential services could help save a lot of inconvenience but some tenants are forced to give them up when they could not afford higher priced SDUs.

On the other hand, partitioning quality (QIP) is not a significantly priced attribute though the respondents expressed that furniture and electrical appliances could be better arranged if the partitioning is more practical. Regarding the accessibility of the SDUs (QITRA), the reason for not being significant could be due to the extensiveness of Hong Kong metro network and the residential buildings with more 25 years of age are mostly located in accessible areas which are near to the metro stations. Little variance is therefore resulted. Provision of furniture and electrical appliances (QIFUR) does not have significant impact on the rental values although interviewees said that their initial move-in costs would be greatly saved if the furniture and electrical appliances are provided.

Regarding measurement of electricity and water charges (QIEW), the effect is not significant though tenants’ usage would be better measured with the meters and it may help to avoid surcharge problem. According to the interviews, it is common that if the meters are not installed, the landlords may overcharge the utilities expenses. With the meter readings, the tenants are more accurately charged though the surcharge problem may still exist. As mentioned in the previous section, internal conditions (QIINT) is served as a control variable in the empirical models to control its effect on rents. Though this variable is not within the focus of this paper, it is also important to examine if QIINT correlates with other explanatory variables in the regression models. We conducted tests with the null hypothesis that the effect of QIINT on other explanatory variable is zero. The regression results do not reject the null hypothesis suggesting that QIINT does not significantly affect other explanatory variables in the empirical models.

The effect of subdivision density is further studied by Model 2. Table 4 shows the regression results. For SDUs without independent toilet (1-TOILET), inconvenience is escalated when tenants have to share the common toilet and the access to fresh water with more households (NSDU). In Model 2, (1-TOILET)*NSDU measures the effect of subdivision density for SDUs without independent toilet. It is significant with coefficient of − 0.062, implying that 6.2% decrease in rent is resulted for one unit increase in subdivision density. Tenants have to pay higher rent for SDUs which are less densely subdivided to mitigate inconvenience, security concern and reduce waiting time for the shared sanitation facilities.

Table 4 Regression results of Model 2
Full size table

5 Research implications and discussion

The findings of this study reveal the rent determinants which cause significant impacts on SDU rents, providing important references for policymakers on affordable housing design and future housing innovations. The determinants are indeed representing the implicit pricing of informal housing attributes. Apart from basic housing necessities such as window and toilet, SDU tenants also value floor area, residential services, and pollution level of the living environment. With limited rental or budget, these represent the housing attribute bundle that is priced by the landlords and the low-income households in Hong Kong. The empirical results also contribute by filling the discontinuity of basic facility effects, i.e. the effects on rents caused by the absence of independent sanitation facilities and window under an inadequate housing environment. These represent the value of necessity to the tenants, which have rarely been studied in previous property price and rental analyses.

Besides, shared toilet facilities is not preferred, especially by households with children. Future housing innovations such as cohousing and social housing which involve shared facilities have to take into account the needs across different household types. In fact, some non-governmental organizations began to provide legal rental units for cohousing by utilizing the vacant properties and acting as principal tenants. It is an innovative solution to help safeguard the lower acceptable bound of living quality and tenure security, while providing affordable housing. It creates values to both property owners and tenants (Light Be 2017). Sanitation facilities are mainly shared between two households in order to uphold the hygiene quality.

Regarding the limitations of this study, difficulties in data collection were encountered as SDU rental information was not readily available. Data were mainly collected by face-to-face interviews through referrals and non-scheduled visits. Yet, the response rate was not high and some tenants were reluctant to provide their rental details. The partial effect of the independent variables on rent and the representativeness of the results may be strengthened if more samples could be collected. Besides, we conducted the omitted variables test to examine if the following set of variables, namely AREA2, FLOOR2, AGE2, make significant contribution to explain the variation of lnRENT. It mainly aims to examine if the decreasing or increasing marginal effects of AREA, FLOOR and AGE on lnRENT are significant. The null hypothesis is that the additional set of variables is not jointly significant. The F-statistic is 1.73 with the p value of 0.18. It does not reject the null hypothesis at 10% significance level. Thus, they are not included in the empirical models.

Moreover, as explained earlier, QIINT is served as a control variable in the regression models to control its effect on rent and to avoid the omitted variable bias caused by excluding it, given that it is not significantly correlated with other explanatory variables in the models. Butler (1982) addressed the omitted variable bias by suggesting that model uses small number of key variables that are costly to produce and yield utility generally would suffice. The regression models of this study consider key SDU variables with reference to previous surveys as well as attributes used in property price and rental literature. The models explain 70.6% and 72.6% of the data variance, which are reasonably high to explain the SDU market landscape. For future studies, fixed effects model could be used to mitigate the possible bias from misspecification.

6 Conclusion

This study is a pioneer attempt to examine the rent determinants of SDUs in Hong Kong. Apart from overcrowdedness, some SDUs even lack basic housing facilities for sanitation and ventilation. The results of this study thus reveal the effects of their deficiencies on rent, which have been rarely studied. It confirms the hypothesis that basic and observable housing facilities have significant and positive impacts on SDU rent. They are mainly structural and environmental attributes, which include floor area, independent toilet, window(s), residential services and external environment. The results provide empirical evidence for the market landscape of Hong Kong informal housing. They are also important references for policymakers and non-governmental organizations on social policies and future affordable housing design. Moreover, this study sheds lights on the effect of subdivision density on SDU rents. For SDUs without independent toilet, rent decreases with the increase in number of households sharing the common toilet and fresh water supply. The significant negative effect reflects the inconvenience as well as the security and privacy concerns faced by the tenants. To relieve SDU households’ rental burden and uphold the basic housing quality, effective and timely housing solutions are needed to increase the supply of formal and affordable housing with basic acceptable living standard, which in turn cool down the overwhelmingly demanded SDU rental market.