About Us |  FAQ  |  Contact Us |  中文版(chinese)

Articles Recommendation

Case Study on Changes in Housing Submarkets in Shanghai, 1994-2005

Zhang Guowu

(Hunan City University, Yiyang, Hunan 413000, China)
 
Abstract: This paper presents a study of housing submarkets in Shanghai. Given the experience of housing reform in China, the paper argues that traditional ways of defining housing submarkets need to be replaced by an alternative which involves a systematic analysis of submarket dimensions specific to the Chinese context. A nested hierarchical housing submarket structure has been developed with appropriate spatial and tenure dimensions to take account of local housing submarket operation. The empirical results from two case study neighbourhoods in Shanghai affirm that submarket analyses will be subject to aggregation bias if they fail to accommodate the existence of submarket structure, and that multi-level equilibria and disequilibria coexist in the submarket complex. The findings are useful in gaining a better understanding of housing market segmentation in urban China.
Key words:
 

Introduction
It is widely accepted that the existence of housing submarkets is of analytical significance (Jones, et al., 2003, Rothenberg, et al., 1991). However, the role of housing submarkets has not yet begun to permeate practical applications partly due to difficulties in defining submarkets. In existing literature, there is a lack of consensus about key empirical dimensions and how to determine housing submarket formation. Most submarket analyses simplify tenure dimensions by focusing on owner-occupation, and treat spatial dimensions in a reductionism way. Furthermore, there are currently many submarket studies which do not recognize the importance of submarket structure.
This paper aims to address some of these problems by proposing an addition to the existing approaches. Specifically, given the experience of housing reform in urban China, an analysis of dimensions specific to the Chinese case is conducted, with a clear distinction between housing supply segmentation and demand differentiation. A nested hierarchical urban housing submarket structure is then developed, with spatial and tenure dimensions necessary to ensure a fuller comprehension of local housing submarket operational processes.
There are many reasons to expect that the housing markets in urban China might be different from the widely studied North American and European market. Experience from Western Cultures is likely to be irrelevant in the case of China. This paper contributes to the ongoing debate in relation to what is a housing submarket, what dimensions determine submarket formation and how it operates, and reduces the information gap by adding to the knowledge of housing market segmentation in urban China by introducing new evidence from the Shanghai market.
The next section proceeds with a brief review of the treatment of housing submarkets in the literature. After a systematic analysis of dimensions contributing to housing submarket formation in Section 3, Section 4 proposes an analytical framework to examine submarket changes in Shanghai based on the development of a nested housing submarket structure. This is followed by section on the data sources and methods. The empirical results are presented in the subsequent sections. The paper concludes with a summary of key findings and policy implications.
The Treatment of Housing Submarket in the Literature: Approaches and Limitations
The concept of the housing submarket was adopted as a working framework in a number of studies of the local housing markets in the 1950s and 1960s (see Fisher and Winnick, 1953, Grigsby, 1963). Subsequently, the term has been widely adopted in the housing literature. Since the 1970s around twenty studies have sought to test their existence, and identify submarkets by using hedonic house-price analysis (for a comprehensive review, see Watkins, 2001).
However, there is considerable imprecision and inconsistency surrounding the conceptualization of housing submarkets (Watkins, 2001). There seems to be a lack of consensus as to what empirical dimensions should be used and how they determine housing submarket formation. Quite often, spatial dimensions have been treated in a reductionist way and tenure dimensions abstracted away. There has also been a lack of clear distinction of dimensions between supply segmentation and demand differentiation. Different sets of dimensions taken from supply and/or demand sides have been used in various attempts to identify housing submarkets. As such, a range of submarket definitions have unsurprisingly emerged. Correspondingly, a number of means and techniques have been employed in various attempts to identify housing submarkets with different conclusions reached (Watkins, 2001).
Furthermore, there appears to be little distinction between homogeneous housing stock segments and submarkets, since in most of the submarket analyses based on either a priori expectation or statistical procedure, housing submarkets include products that are interchangeable from the supply side, with little attention and reference to the possibility of substitution on the demand side. As such, the resulting submarkets may be homogeneous housing stock segments rather than submarkets; while the least-used scheme focuses on the identification of homogeneous demander-groups and ignores the numerous dimensions of housing stock segmentation (see for example, Feitelson, 1993).
Consequently, the role of submarkets has not been embraced in applied research and the concept of housing submarket seems under-theorized (Maclennan and Tu, 1996). There is a strong case to redefine and re-identify the housing submarket, to demonstrate its structure and reveal its boundary. To achieve these aims, it is necessary to conduct an analysis of all possible dimensions contributing to housing submarket formation.    
Housing Market Segmentation in Shanghai
There are numerous influences on housing supply segmentation and on demand stratification. The factors leading to supply segmentation may be conveniently summarized into three categories: hierarchical spatial dimensions related to externalities embedded in complicated urban spatial structure, tenure dimensions, and structural dimensions; whereas the dimensions of demand differentiation vary across societies and over time depending on welfare and institutional arrangements. These are discussed in turn below.
Spatial Dimensions
Location is usually considered the most important of housing characteristics. However, the locational context of housing stock has been treated in a rather reductionist way, and has often been simply and inadequately modelled by using the distance to CBD as a basis for the spatial stratification of housing stock, but this is often of limited value as a locational dimension.
Brueckner et al (1999) classify urban amenities into three categories: exogenous natural amenities, historical amenities, and endogenous modern amenities. For simplification, the hierarchical locational externalities affecting housing value, positive or negative, may be conveniently divided into two parts: externalities above neighbourhood level and ones at neighbourhood level, which are neighbourhood components. The former includes proximity to CBD and to subcenters, transportation networks, recreational areas, and pollution etc. Even though location is fixed, numerous man-made external economies at above neighbourhood level may be dynamic, affecting not only the level of unit costs but also the rate at which they fall/rise over time (Anas et al., 1998). Consequently, the differential effects of externalities, often strongly correlated with location, may dynamically influence housing submarket formation. Therefore, there is a limited basis for pooling housing stock with the same or similar structural characteristics but in different neighbourhoods together to form a submarket, even though these neighbourhoods are the same in terms of all their intrinsic characteristics and at the same distance from the CBD. Additionally, of course, neighbourhoods can differ greatly in their physical, social and economic characteristics.
In addition to the inclusion of accessibility aspects of location, technical progress has been made to add separable amenity terms to the hedonic model to derive rich specifications for econometric modeling of neighbourhood amenities (see, for example, Cheshire and Sheppard, 1995). However, due to the complex and idiosyncratic nature of locational amenities in an urban area, information about a property’s locational characteristics is substantially more difficult to observe and quantify (Dubin et al., 1999). Moreover, numerous potential influences on property value create problems in specifying a parsimonious model. Consequently, multiple regression models designed to estimate property value typically include a limited number of influential locational characteristics. The capitalization of a full range of amenity characteristics may be still incomplete by including a number of observed and measured indicator of externalities.   
Tenure Dimensions
Generally, housing market is dominated by three very distinct tenure types, conferring differing degrees of security, opportunities for mobility, direct and indirect subsidization and potential for capital gain, although the proportion of the public and private rental sector differs in each country according to different welfare arrangement (Smith et al., 1988). Moreover, there are significant regional variations.
Usually, access to homeownership is expensive, and is contingent on an ability to secure a mortgage. Very few households can choose freely without constraints, and the vast majority have their choices limited by local housing market conditions and their position in the labour market. Many cannot access ownership, and have to resort to government aid for social housing. Access to public rental housing is strongly regulated because of the heavy subsidies they entail so private renting is the most accessible form of tenure.
For urban households it is not easy to cross the tenure lines between private rental, public rental, and private ownership; tenure-matching decisions have to be made between owner-occupancy, social rental and private renting that depend not only on the resources at households’ disposal, household needs and preferences, consumption and portfolio investment considerations, but also on welfare institutions. For example, households without local urban identity permits (Hukou registration) in urban China cannot access public rental housing. Moreover, similar to the privatizing policy under the conservative government in the UK, tenants of previously social housing have had the Right to Buy (RTB) at discounted prices much lower than market prices under the policy of ‘selling of former public housing to its sitting tenants’ since the early 1990s.
Tenures constitute the means by which households consume housing and the vehicle by which the state influences that consumption (Ball, 1986). Each tenure, therefore, constitutes a particular way in which housing is provided as a commodity. Tenures thus produce differentiations between households in terms of the physical quantity and quality of the housing in which they live, and its cost. Of the various attributes that characterize the structure of housing provision, tenure is another important axis along which housing stock is stratified, especially in countries where the state’s involvement in housing provision has produced a complex mix of tenure forms.
However, tenure has received relatively little treatment in submarket identification to date. Some studies ignored the tenure difference and treated both owner-occupation and rental sectors together; others have focused on owner-occupation and have excluded rental sectors (see, for example, Bourassa et al., 1999, Goodman and Thibodeau, 1998, 2003, Jones et al., 2003). To fully understand housing market operational processes, it is therefore necessary to include tenure dimensions in the analysis of housing submarkets.
Structural Dimensions
Housing units differ from each other with respect to lot size, number of bedroom(s) and bathroom(s), floor space, storey on which the unit is located, dwelling type, and age of structure, balcony, hardwood floors, as well as various qualitative characteristics. There is little disagreement in the literature about housing supply being segmented along these structural dimensions.
Dimensions of Demand Differentiation
On the demand side, there are different groups of consumers stratified along a number of dimensions. Feitelston (1993) proposed a hierarchical approach to residential demand segmentation, in which at least three types of dimensions can be identified according to the stability of the strata from the household’s perspective. These include variables describing the membership of households in an ethnic, religious, racial, or some institutional group, the socioeconomic situation which determines the resources at households’ disposal, thus their opportunities in the housing market, household life-style choice, and needs.
The sets of dimensions contributing to supply segmentation and demand differentiation may vary across societies and change over time. For example, in the previously socialist welfare housing system in urban China, housing supply was determined by the socialist hierarchical administrative ranks1, characteristics2 and property3 of numerous employers (work-units), in line with their importance to the national development strategy (Zhang, 1997). Residential compounds were generally developed close to work units, and were equipped with appropriate education, medical, and commercial facilities. Since all housing was publicly-owned, there was no tenure difference. As such, housing stock was segmented by spatial and structural dimensions conforming to the socialist hierarchy. Housing demand was stratified by household heads’ Hukou registration, administrative ranks, or its equivalent professional qualifications, and seniority. Household needs and resources were not important in housing consumption. The welfare match was implemented through employers at differing levels.
As a key part of the national economic reform, the market-oriented housing reform in urban China starting in the early 1980s gradually transformed the formula of supply segmentation and demand stratification towards a market-oriented mechanism. Spatial, tenure and structural dimensions interacted together to segment housing supply and household resources and needs became key stratifiers of housing demand. The matching process between segmented supply and stratified demand became increasingly implemented through market mechanism.
 
Housing Submarket Structure, Formation and Operation
Most of the recent work does not recognize the importance of urban housing submarket structure, although a number of previous empirical studies have shown its existence (see, for example, Tu, 1997). To facilitate the identification of submarket and a fuller understanding of housing market operation, a nested hierarchical housing submarket structure is developed for Shanghai (Figure 1). On the supply side, the dimensions are divided into three levels. The first level is the neighbourhood, distinguished by spatial dimensions. It is the unique site, situated in a locality that tends to produce homes with the same or similar structural characteristics and at prices that attend to attract similar household types. The second level of housing submarket structure relates to tenure types, such as private ownership, RTB private ownership, social housing, and free-market rental. The third level is housing structural dimensions, such as number of bedroom(s) and bathroom(s), floor space, dwelling type, balcony etc.
\











 
Figure 1. A Nested Hierarchical Local Housing Submarket Structure
Note: N1, 2 …K refers to neighbourhoods; TPO to Private Owner-occupation sector, TSO to Ownership via RTB, TSR to Social Rental sector, TPR to Private Rental sector; t1, 2… j represents hierarchical housing stock segments differentiated by housing values and structural characteristics in a given neighbourhood. Household strata1, 2 …i refer to homogeneous household groups differentiated by their socio-economic and institutional variables. Similarly, TPOt1, TPOt2 ...TPOtj represent housing stock sub-hierarchy in private owner-occupation sector, TSOt1, TSOt2 ...TSOtj indicate housing stock sub-hierarchy in RTB owner-occupation sector, and so on.
Key dimensions on the demand side include socio-economic status, institutional dimensions, and household income, stage in life cycle, and preferences which stratify urban households which (tend to) live in a neighbourhood into strata. The contribution of each demand dimension to household stratification varies across societies and over time, depending on the ethnic composition, institutional and welfare arrangements, housing policy, and income distribution etc.
The housing submarket structure emphasizes the importance of segmentation of both supply and demand in determining submarket formation, definition, and identification. Spatial, tenure and structural dimensions interact together to segment housing supply; and socio-economic, institutional dimensions, and household income, preference, stages in life cycle come together to differentiate demand. For submarkets to exist it is not sufficient to simply identify homogeneous housing stock segments without making reference to corresponding household groups. It is necessary to stress that different household groups are matched to different classes of housing depending on institutional and welfare arrangements, household resources, needs, and preferences etc. In other words, there must be a clear identification of the differential linkages between different classes of property on the supply side and differentiated groups of households on the demand side (Jones et al., 2004). It is the way in which segmented demand is matched to the differentiated housing stock which is likely to give rise to housing submarkets (Watkins, 2001).
Thus, a housing submarket may be defined as a collectivity of all dwelling units in a neighbourhood with the same potential tenure and similar structural attributes, together with relatively homogeneous potential household demanders for which these housing units are evaluated as a whole as close equivalent.
Housing stock in a given neighbourhood can be arrayed in a hierarchically graded cluster of housing stock according to its structural characteristics. Correspondingly, all the households which actually or potentially reside in this neighbourhood are stratified into hierarchically graded household strata by relevant institutional dimensions, resources at households’ disposal, household size and preferences etc. Different grades of housing stock in the hierarchy house different socio-economic groups of households. It may be a destination for a certain group of households, and in the meantime it may be a starting point for another group of households which intend to move up a housing ‘ladder’.
The housing market in a given neighbourhood typically consists of hierarchical submarkets, and its operation may be viewed as a dynamic match process (rather than choice) between different grades of property on the supply side and different classes of consumer groups on the demand side. The match process arises from both the normal operation of the housing market with respect to both housing supply and demand, and the extent to which governments intervene in the housing market. Therefore, to a certain extent the match mechanism varies across different societies from administrative or welfare match to market match, leading to differing degrees of distortion.
A full understanding of the operation of the housing market will require a grasp of the extent of the match and of changes in the match between household groups and housing product clusters over time. The hierarchical housing submarkets may be in the process of adjustment in which multi-disequilibria and multi-equilibria coexist as a result of frequent disruptions from both the supply and the demand sides.
Data and Methods
This paper applies the above analytical framework to two case study neighbourhoods in Shanghai in order to empirically examine submarket changes during the period from 1994 to 2005, using a multiple case study approach. Each neighbourhood has housing developed by both public and commercial investments, thus, all possible tenure submarkets were included in this study. Prior to 1993 both neighbourhoods were dominated by social rental submarkets. The period is chosen because it saw the post-socialist transition to a market-oriented housing market.  
The empirical section of this paper is based on the analysis of the multifaceted data set of 231 households obtained in March 2006 from the field survey of the two case study neighbourhoods, Gushan and Nancheng, in the east and south of Shanghai respectively. Both are located at the same distance from the CBD, the sub-centre in Pudong New Area, the inner-ring road and the main natural amenity—Huangpu River. On the basis of house price analyses aimed at examining whether price structure is different in different neighbourhoods, the research design would provide empirical evidence to support the argument that housing units with the same or similar structural characteristics and at the same distance to the CBD but in different neighbourhoods cannot be simply pooled together to form a submarket. It also draws analytical conclusions from two independent cases to allow a degree of comparison checking.
The population were about 900 and 350 in Gushan and Nancheng respectively in 1994. Both increased to about 1,100 in 2005, as a result of different availability of land for commercial housing development. Each dwelling was individually coded and entered into a database for stratified sampling. In the Shanghai market, housing stock developed in different periods had different structural characteristics. The population in each neighbourhood was divided into non-overlapping structural groups or strata according to information on built time and the structural types. Subsequently, the established technique of proportionate stratified random sampling was used to select households for interview, to ensure that the sampling fraction was representative of each stratum and that sampled homes were in proportion to the structural compositions of the housing stocks, representing the development processes of each neighbourhood. 118 respondent households were from Gushan and 113 from Nancheng. The survey included information about when and what types of dwellings were built and transacted, when and what types of households moved in, their composition, heads’ age profile, current income, education attainment, occupation, Hukou registration and employment status related to the institutional legacy in China.      
Following the nested submarket structure (see Figure 1), a multiple step procedure was employed to identify housing submarkets optimally in Shanghai. 
Firstly, key housing structural variables, dwelling size, number of room(s) and number of bathroom(s), age (in 2005), were used as independent variables in hierarchical cluster analysis (Hastie et al., 2001) to construct the hierarchy of housing stock segments (t1, 2… , j) in each neighbourhood separately. These variables represent key factors of housing supply at the third—dwelling type level (see Figure 1) in a given neighbourhood which affect household housing consumption.
Secondly, by tabulating the housing stock hierarchy across tenure types, the sub-hierarchies (TPOt1, TPOt2 ...TPOtj in private owner-occupation sector; TSOt1, TSOt2 ...TSOtj in RTB owner-occupation sector; TSRt1, TSRt2 ...TSRtj in social rental sector and TPRt1, TPRt2 ...TPRtj in private rental sector) can be derived, representing housing supply at differing submarket levels in different tenure sectors in each neighbourhood separately.
Thirdly, using lifecycle as an organizing concept, simple statistical techniques were used to identify homogeneous household groups related to the segmented housing stock clusters. Housing submarkets were identified by matching differentiated household groups to their corresponding housing stock clusters, which together constitute housing submarkets at differing levels in different tenure sectors in each neighbourhood separately.
Given the low possibility of, and restrictions on, the modifiability of housing units in the form of flats in multiple-storey and high-rise residential buildings, it is reasonable to assume that housing units in each neighbourhood would stay in their relative housing submarket positions over time, since housing units in the same neighbourhood would be expected to, if possible, appreciate or depreciate at similar rates. As such, cross-sectional information dating back to 1994 on part of households’ housing consumption and their characteristics can be traced and used in cross-sectional analysis to examine the growth and decline of submarkets in different sectors during the study period. 
It is important to note that due to its modest sample size, the dataset may not sufficient to estimate a convincing hedonic regression. However, instead of a direct basis for submarket identification, as in the traditional submarket analyses by testing price differentials, the hedonic models are used only to ascertain whether the price structure is different across neighbourhoods. The observations captured are sufficient to conduct cluster and other statistical analyses at neighbourhood level, allowing the model to be operationalised in a meaningful way.           
Hedonic House Price Estimation
The formulation of regression models requires the construction of a database of appropriate variables which should comprise all significant macro and micro influences on house prices (Adair et al., 1996). Despite the lack of consensus in the literature regarding the variables to be included into the hedonic price function, characteristics in three categories are generally considered appropriate (Bowen et al., 2001): housing structural characteristics (S), social and environmental attributes of the neighbourhoods in which dwelling is located (N), and other locational characteristics such as accessibility to CBD, employment and services etc (L). Given appropriately measured variables in these three categories, it is generally agreed that proper specification of the hedonic price function accordingly expresses the market (cross-sectional) prices of housing units as
\


However, the housing submarket structure model makes it possible to only include variables of housing structural characteristics in the hedonic price analyses, since all housing units in each case study neighbourhood are nested in the same hierarchy of externalities. Externalities contributing to the housing value at neighbourhood level and above, positive or negative, are not unique to any individual property but are common to the group of properties sampled from the same neighbourhood. As such, the variables indicating complicated amenities related to residential locations, which are found to be important in the other hedonic studies, are constant for all the properties sampled from the same neighbourhood, thus were omitted in this analysis. This omission would not bias the results. Instead, the analysis prevents the measurement problems with regard to neighbourhood quality and numerous externalities at differing levels from affecting the accuracy of regression results.
Gloudemans (1990) pointed out that it was important to appreciate that failure to allow for temporal stability can produce unstable regression coefficients. Housing prices in Shanghai were not constant in either nominal or real terms over the period from 1994 to 2005. As the observations of market transactions were 40 and 71 in Gushan and Nancheng respectively, aggregate yearly time variables are utilized to measure changes associated with sales in different time periods, thus capturing the influence of time on housing price, and produce stable regression coefficients.
Many regression forms were carried out to test different models. Table 1 and Table 2 report the regression results for best models in Gushan and Nancheng respectively. They use linear and log-linear forms for cross-validity. It can be seen that house age (Hsag) carried an expected negative coefficient on transaction price. Non-linear forms of explanatory variable Hsag were tried but did not improve the explanatory power of the models as indicated by reduced goodness-of-fit compared to the reported models. 
The only housing structural attribute affecting housing prices is one-room flat (FR1BKM) in Gushan, which carries a negative coefficient, indicating the undesirability of this type of out-of-date dwellings for households’ housing consumption. The remaining housing structural variables in the hedonic models for both neighbourhoods are not found significant. One of the reasons might be that housing depreciation or appreciation over the study period overrides the effects of housing types on house prices. Furthermore, as the number of observations is modest, the structural effects are likely swamped by the aggregate time effects. Moreover,

Table 1. Regression Results for House Price Estimation in Gushan
Dependent Variable   PRICE   LN(PRICE)
Independent Variable Definition Coefficient t-value   Coefficient t-value
FR1BKM one-room flat with kitchen and bathroom -1109.644 -2.975**   -.413 -4.033***
Hsage Flat age -24.675 -2.137*   -.008 -2.377*
Tt94/01 Dummy variable for sale between 1994 and 2001 (1=yes, 0=no) -671.021 -2.198*   -.170 -2.024*
Tt03 Dummy variable for sale in 2003 (1=yes, 0=no) 1097.686 3.032***   .258 2.596*
Tt04 Dummy variable for sale in 2004 (1=yes, 0=no) 2458.581 6.028***   .509 4.543***
Tt05 Dummy variable for sale in 2005 (1=yes, 0=no) 4104.450 5.427***   .759 3.654**
Constant   4191.649 15.318***   8.318 110.629***
R2   0.830     .797  
Adjusted R2   0.800     .760  
F-Statistic   26.941     21.635  
df   32     32  
 

Note:  * p<.05, ** p<.01, *** p<.001; PRICE refers to market transaction price. 

Table 2. Regression Results for House Price Estimation in Nancheng
Dependent Variable   PRICE   LN(PRICE)
Independent Variables Definition Coefficient t-value   Coefficient t-value
HSAG Flat age -133.036 -8.131***   -.030 -8.107***
Ttn97/8 Dummy variable for sale in 1997/98 (1=yes, 0=no) -745.055 -2.631*   -.157 -2.467*
Ttn99 Dummy variable for sale in 1999 (1=yes, 0=no) -1352.582 -3.793***   -.329 -4.103***
Ttn00 Dummy variable for sale in 2000 (1=yes, 0=no) -1459.631 -3.559**   -.450 -4.882***
Ttn01 Dummy variable for sale in 2001 (1=yes, 0=no) -655.969 -3.025**   -.147 -3.012**
Ttn03 Dummy variable for sale in 2003 (1=yes, 0=no) 1704.201 6.633***   .310 5.372***
Ttn04 Dummy variable for sale in 2004 (1=yes, 0=no) 4275.816 13.284***   .716 9.894***
Ttn05 Dummy variable for sale in 2005 (1=yes, 0=no) 3904.895 5.722***   .673 4.384***
Constant   5455.695 39.132***   8.603 274.520***
R2   0.847     .805  
Adjusted R2   0.827     .780  
F-Statistic   42.878     32.018  
df   61     61  
 

Note: * p<.05, ** p<.01, *** p<.001; PRICE refers to market transaction price.

instead of package prices, housing has been transacted in practice on the basis of per-square-meter prices which are usually unrelated to housing types in the Shanghai market.
Though both neighbourhoods are at the same distance from the CBD and the other key amenities, the standard house price in Nancheng (CNY4 ¥ 4,972.774) is nearly 25 percent higher than that in Gushan (CNY ¥ 4,002.145 transacted in 2002). Also, there were six regressors in the Gushan price equations while eight in the Nancheng price equation, for which the coefficients were different. The reasons for these discrepancies might be that the number of transacting properties in Gushan (40) during the study period was smaller than that in Nancheng (71), and transacting time in Gushan was so widespread that some dummy variables indicating transacting time were found significant in the Nancheng but not in the Gushan equation. The differences between coefficients of transaction time dummy variable of the same year reflect that the rates of price appreciation/ depreciation varied spatially in the Shanghai market.
The empirical evidence emphasizes the need for caution in using the distance of housing to CBD as the proxy of spatial dimensions in submarket identification. Since housing units with the same or similar structural characteristics and at the same distance to CBD but located in different neighbourhoods command quite different market values, therefore, they are a substitute for different household groups and may not be pooled together to form a submarket. Because of the different price structures in different neighbourhoods, hedonic models traditionally used to detect submarket existence by analyzing house price differentials may not be sufficient to detect or identify housing submarkets in the case of the Shanghai market.    
The general price model has significant goodness-of-fit measures across the functional forms. The adjusted R2 in both linear and log-linear price model for Gushan and Nancheng are very close to the value of their corresponding R2, with the lowest of either of them .76, implying levels of explanatory power in excess of the 76% of variation in prices. It is worth noting that the estimations are used only to examine whether the price structure is different in different neighbourhoods. The choices of one of the two functional regression forms per se proved to have no effect on the accuracy of the results of housing segment identification, therefore, either one is deemed appropriate for this analysis.
The Identification of Housing Submarkets in Shanghai
The Identification of Homogeneous Housing Stock Segments
There are six different types of single family flats funded by either public or commercial sources in different periods in both Gushan and Nancheng in terms of number of room(s), ranging from one to six.  Six distinct hierarchical levels of housing stock strata emerge in both Gushan and Nancheng (see the ‘Total’ column in Table 3 and Table 4) when the set of independent variables, including dwelling size, number of room(s), number of bathroom(s), and age (in 2005) are entered into group average (GA) method and centroid linkage for hierarchical clustering analysis (Everitt et al., 2001)5.
Housing units in each stratum are sufficiently similar to each other compared to those assigned to different strata, and can potentially be transferred across different tenure sectors on the market. However, the housing unit stratum level labeled at the same grade in different neighbourhoods should not be regarded as equivalent or treated as the same level in the Shanghai market.      

Table 3. Homogeneous Housing Stock Segments in Gushan in 2005
Submarket Level Flat Size
(m2)
Number
of room
Value in 2005 Tenure Total
PO RTBO SR PR
i 138* 6 875 2       2
ii (108-116) 4-5 719-775 8       8
iii (76-97) 4 590-672 11   1   12
iv (56-68) 3 327-505 3 23 1 1 28
v (42-50) 2 214-351 7 17 4   28
vi (21-33) 1 77-193 7 5 17 11 40
Total       38 45 23 12 118
 
Note: * indicates flats with double floor space (fushi Zhuzhai); PO refers to private ownership, RTBO to RTB ownership,
   SR to social rent, and PR to private rent; Value is in thousand CNY.

By tabulating the total hierarchy of housing stock segments across the four different types of tenure in the Shanghai market, the sub-hierarchies of housing stock segments in different tenure sectors are constructed, representing housing supply at differing submarket levels in different tenure sectors in each neighbourhood separately (see the tenure sub-columns in Table 3 and Table 4).

Table 4. Homogeneous Housing Stock Segments in Nancheng in 2005
Submarket Level Flat size   (m2) Number
of room
Value in 2005 Tenure Total
PO RTBO SR PR
I 186* 6 2154 8       8
II 150-162* 5-6 1605-1876 9     2 11
III 118-129 5 1366-1473 14 1     15
IV 97-112 4 1049-1216 27 4 1 2 34
V 62-65 3 628-696 10 15 1 2 28
VI 43 2 416-460 2 5 1 9 17
Total       70 25 3 15 113
 
Note: * indicates flats with double floor space (fushi Zhuzhai); Value is in thousand CNY.

It can be seen that the housing stock compositions in each of the two case study neighbourhoods reveal different patterns. In total, there are more housing units at the lower level of the hierarchy in Gushan, whilst in Nancheng, there is more housing stock clustering at the middle level in the hierarchy. The reasons for the differences may be attributed to: (1) Gushan stock is relatively older, with the development of its oldest properties dating back to the 1950s in comparison with 1986 in Nancheng; (2) as a result of different availability of land for new housing development, commercial housing built after 1996 is only 17.8 percent of the total in Gushan compared to over 60 percent in Nancheng during the same period. Consequently, there are more housing units in RTB owner-occupation sector in Gushan than in Nancheng, which were transferred from previous social rental housing.
Housing in the social rental sector in Nancheng is underrepresented as social housing there is in relatively good condition, and has been privatized to a greater extent under the RTB policy than in Gushan. The difference in the private rental sector is that all private rental flats in Gushan are previously social housing and 91.7 percent are at the lowest submarket level; whilst in Nancheng, 26.7 percent of housing units are from commercial housing stimulated by buy-to-let as an investment option, which are at submarket level II and IV.
Match between Segmented Housing Supply and Stratified Housing Demand
Although there is considerable debate regarding the role of the life cycle in terms of both its impact as an organizing principle for housing consumption and its value in understanding residential mobility, it is an enduring organizing concept for understanding household behaviour and mobility (Clark et al., 1984). The matching analysis is conducted to examine cross-sectional household distribution over the hierarchical submarket levels and household life cycle in Gushan and Nancheng at the beginning and the end of the study period.
The difficulty in establishing more than general relationships between the life cycle and housing consumption is in part related to the varying ways in which life cycle stages have been defined (Quigley and Weinberg, 1977). There appears to be no study attempting to distinguish among alternative definitions. In this study, life-cycle types are defined in terms of household size and age of household head. As suggested by the data, in both neighbourhoods, there were over 75 percent households comprising up to 3 persons, partly because of the ‘one family one child’ policy in China since the 1980s. As such, the households were divided into four categories in terms of household size: single person households, 2-person, 3-person and 4-plus-person households, with head younger than 30, 30 to 50, and 50 plus years old.    
Table 5 shows the match in Gushan and Nancheng respectively in 1994. Both were dominated by larger (three- to four plus person in size) and older households in social rental submarkets at middle to low levels, and single and two-person families were underrepresented in existing tenure submarkets in 1994.

Table 5. Match in Gushan and Nancheng in 1994: Percent of Each Life Cycle Category in Hierarchical Submarkets
    Gushan   Nancheng
Households Life Cycle   PO   RTBO   SR N   SR   N
  v   iv v   iii iv v vi   V VI  
1 Person Households                                
30-50 yrs                   100 1          
2 Person Households                                
18-29 yrs   50             50   2          
30-50 yrs                   100 1          
3 Person Households                                
30-50 yrs             3.8 26.9 26.9 42.3 26   60 40   5
50 + yrs       9.1 9.1     18.2 54.5 9.1 11   60 40   5
4+ Person Households                                
50 + yrs       21.7       30.4 17.4 30.4 23   75 25   4
Total   1   6 1   1 16 18 21 64   9 5   14
  Note: Percentages may not add up to 100 because of rounding.
 Source: Field Survey, Shanghai.

Overall, there was little variation in terms of housing supply and household housing consumption in 1994. This was mainly determined by the housing provision and allocation mechanism in the socialist welfare housing system. Social housing provision was of generally low standard as a result of strict control enforced through the entire process of housing construction, covering housing finance, land administrative allocation, housing standards and design, urban planning control and housing construction (Zhang, 1997). On the other hand, local urban Hukou registration, marital status, seniority and household’s size (indicating crowding condition) were important criteria for urban households to be eligible for welfare housing (Zhao and Bourassa, 2003). Single-person households were under- represented since such people usually lived in employer provided dormitories, or had to live with their parents regardless of their age. Moreover, in 1994, private ownership and private rental submarkets were underdeveloped, as there was no supply of commercial housing on the market in both Gushan and Nancheng until 1996. Correspondingly, no household had a second home for rent on the market.  

Table 6. Match in Gushan in 2005: Percent of Each Life Cycle Category in Hierarchical Submarkets
Household Life Cycle   PO   RTBO   SR  
 
PR  
 
N
  i ii iii iv v vi   iv v vi   iii iv v vi   iv vi  
1 Person Households                                          
   18-29 yrs             50                       50   4
   30-50 yrs       33.3                             66.7   3
   50+ yrs       16.7     16.7       16.7         50         6
2 Person Households                                          
   18-29 yrs       50   50                             2
   30-50 yrs         16.7 16.7     16.7             16.7   16.7 16.7   6
   50+ yrs       5.6   16.7     11.1 38.9 11.1   5.6   5.6       5.6   18
3 Person Households                                          
   30-50 yrs   3.4 10.3 17.2   6.9 3.4   13.8 10.3 3.4       3.4 20.7     6.9   29
   50+ yrs     9.5 4.8     4.8   23.8 19 4.8       4.8 28.6         21
4+ Person Households                                          
   30-50 yrs   7.7 7.7 7.7 7.7   7.7   30.8 7.7                 23.1   13
    50+ yrs     12.5   6.3   6.3   43.8 12.5       6.3 6.3 6.3         16
N   2 8 11 3 7 7   23 17 5   1 1 4 17   1 11   118
% of Total   1.7 6.8 9.3 2.5 5.9 5.9   19.5 14.4 4.2   0.8 0.8 3.4 14.4   0.8 9.3   100
 
Note: percentages may not add up to 100 because of rounding.

Table 7. Match in Nancheng in 2005: Percent of Each Life Cycle Category in Hierarchical Submarkets
Household Life Cycle   PO   RTBO   SR   PR   N
  I II III IV V VI   III IV V VI   IV V VI   II IV V VI  
1 Person Households                                              
18-29 yrs                                   20 20 20 40   5
30-50 yrs         25 25         25               25       4
50+ yrs         100                                   1
2 Person Households                                              
18-29 yrs             50                           50   2
30-50 yrs     12.5 12.5 25 12.5           12.5                 25   8
50+ yrs         11.1 33.3 11.1       11.1 11.1           11.1     11.1   9
3 Person Households                                              
18-29 yrs         100                                   2
30-50 yrs   11.4 2.9 20 37.1 5.7     2.9 2.9 5.7 2.9               2.9 5.7   35
50+ yrs   5   5 25 5       15 30 10     5               20
4+ Person Households                                              
18-29 yrs       50   50                                 2
30-50 yrs   10 35 20 10           20                   5   20
50+ yrs   20       20         20     20   20             5
N   8 9 14 27 10 2   1 4 15 5   1 1 1   2 2 2 9   113
% of Total   7.1 8 12.4 23.9 8.8 1.8   0.9 3.5 13.3 4.4   0.9 0.9 0.9   1.8 1.8 1.8 8   100
 
Note: percentages may not add up to 100 because of rounding.
 

Table 6 and Table 7 report cross-sectional housing consumption in Gushan and Nancheng at the end of 2005 by life-cycle by hierarchical submarket levels respectively. It tells us the match in Gushan and Nancheng respectively in 2005 which is in stark contrast with that in 1994.
It can be seen that there is considerable variability over the life-cycle stage in the proportion of families in single family units as a result of both the legacy of welfare housing mechanisms and market-oriented reform in Shanghai since the 1980s. Two observations can be derived from the tables. First is the expected reversal of owner-occupation and private rental tenancy over the life cycle categorization. There is a higher proportion of private rental tenancy in the younger age category and for single person households, and higher proportions of owner-occupation and social rental tenancy for larger households. Second, there are more large households at an older stage of life cycle than small households at a younger stage of life-cycle, which is underrepresented in subsidized sectors (RTB and social rental sector). This was in part because of the ending of welfare housing allocation in the later 1990s which was based on seniority and crowding conditions.      
In both neighbourhoods on average, households’ permanent income estimated by using earning functions based on human capital theory (Wang, 1995), floor space per person and household heads’ schooling years at all level submarkets in private owner-occupation are higher than those at the same level submarket in RTB owner-occupier, private rental and social housing tenure sectors, with households in social rental sector at the lowest level. Access to RTB and social rental housing at all submarket levels has been strictly regulated by Hukou registration, as all households in these subsidized sectors must have a local urban Hukou registration permit. The above evidence supports the argument that tenure is an important dimension along which housing supply is segmented in the Shanghai market.

Changes in Housing Submarkets in Shanghai
The compositions of housing submarkets in Gushan and Nancheng have undergone significant changes during the period from 1994 to 2005 in both absolute quantity and relative terms, as shown in Table 8. Overall, both neighbourhoods have shifted from the dominance of social rental submarkets at middle to low level in the early 1990s to that of owner-occupier submarkets in 2005 at differing levels.
Despite some new input of social rental housing before it was forcefully stopped in December 1999 (State Council Paper No. 23, 1998), the share of social rental submarkets decreased dramatically. In Nancheng, each level declined to less than one percent of the total in 2005, with the social housing privatized under the RTB programme to a greater extent there than in Gushan. Social rental submarkets at the bottom level in Gushan still accounts for 14.4% of the total in 2005. This can probably be explained by the fact that these housing units were mainly built in the 1950s and the 1960s with shared kitchen and toilet facilities and cannot easily be divided into individual units to be privatised.
Ownership submarkets in Nancheng in 2005 accounted for over 84 percent of the total, which is higher than that in Gushan (around 60 percent). Apart from increasing commercial housing supply over the period, the main reason for this growth was that social housing units in relatively good condition were either transferred to the RTB ownership sector under the RTB program, or resold on the market and transferred to private owner-occupier submarkets. In the meantime, new types of housing submarket—private rental submarkets, mainly at lower levels, began to emerge, accounting for 10 and 13.3 percent of the total in Gushan and Nancheng respectively in 2005.
The vacant flats were not cleared until 2002, indicating that both neighbourhoods had suffered from the market recession during the period from 1996 to 2002 when the Asia Financial Crisis hit this region, and that multi-disequilibria and multi-equilibria may coexist in

Table 8. Growth and Decline of Housing Submarkets in Gushan and Nancheng (1994-2005, % of Total)
  Tenure PO RTBO SR PR Vacant N
Gushan Level i ii iii iv v vi   iv v vi iii iv v vi     iv vi i ii iii  
  2005 1.7 6.8 9.3 2.5 5.9 5.9   19.5 14.4 4.2 0.8 0.8 3.4 14.4     0.8 9.3         118
  2004 1.7 6.8 9.3 1.7 5.9 5.9   19.5 14.4 4.2 0.8 1.7 3.4 20.3     0.8 3.4         118
  2003 1.7 6.8 9.3 0.8 4.2 5.9   20.3 16.1 4.2 0.8 1.7 3.4 22.9     0.8 0.8         118
  2002 1.7 6.8 7.6 0.8 2.5 5.9   18.6 15.3 4.2 0.8 4.2 5.9 22.9       0.8     1.7   118
  2001 1.9 1.9 5.6   1.9 5.6   18.7 15 5.6 0.9 7.5 9.3 25.2       0.9         107
  2000 0.9 0.9 4.7   1.9 5.6   16.8 13.1 3.7 0.9 9.3 11.2 27.1       0.9 0.9 0.9 0.9   107
  1999   0.9 2.8   1.9 2.8   14 8.4 4.7 0.9 12.1 15.9 29       0.9 1.9 0.9 2.8   107
 
 
  1998         2 2   13.1 8.1 2 1 15.2 18.2 35.4       1   2     99
  1997         2     12.1 5.1 3 1 16.2 21.2 37.4           2     99
  1996         2.1     11.3 4.1 1 1 17.5 22.7 40.2                 97
  1995         2.1     8.2 4.1 1 1 20.6 22.7 40.2                 97
  1994         1     6.2 2.1   1 22.7 25.8 41.2                 97
Nancheng Level I II III IV V VI III IV V VI III IV V VI II IV V VI I II III IV  
  2005 7.1 8 12.4 23.9 8.8 1.8 0.9 3.5 13.3 4.4   0.9 0.9 0.9 1.8 1.8 1.8 8         113
  2004 7.1 8.8 12.4 25.7 1.8   0.9 3.5 19.5 7.1   0.9 2.7 7.1 0.9   0.9 0.9         113
  2003 7.1 9.7 12.4 25.7 1.8   0.9 3.5 19.5 7.1   0.9 3.5 8                 113
  2002 5.5 9.2 11 24.8 0.9   0.9 3.7 18.3 7.3   0.9 6.4 8.3         1.8 0.9     109
  2001 4.1 4.1 10.2 18.4     1 4.1 19.4 7.1   1 9.2 10.2         4.1 4.1 2 1 98
  2000   2.9 5.8 14.5     1.4 5.8 18.8 5.8   1.4 21.7 18.8             1.4 1.4 69
  1999   2.9 5.8 11.6       4.3 13 4.3 1.4 2.9 27.5 20.3             1.4 4.3 69
  1998   1.4 4.3 11.6       1.4 10.1 2.9 1.4 2.9 30.4 21.7           1.4 2.9 5.8 69
  1997   1.4 1.4 7.2         5.8 2.9   1.4 34.8 21.7           1.4 7.2 14.5 69
  1996       5.8         5.8 2.9     34.8 21.7           2.9 8.7 17.4 69
  1995                 8.3 5.6     50 36.1                 36
  1994                         58.3 41.7                 36
 
Note: percentages may not add up to 100 because of rounding. Source: fieldwork, Shanghai.

the submarket complex in both Gushan and Nancheng as a result of frequent disruption from both the supply and the demand side.
Conclusions
This paper has developed an alternative analytical framework for the analysis of submarkets in Shanghai based on the development of a nested hierarchical housing submarket structure which reflects the role of space, tenure, dwelling type, variations in household characteristics and preferences in segmenting the market.
Empirically, this paper has applied this framework to areas of Shanghai, demonstrating the importance of the housing submarket structure in understanding local housing submarket operation. Neighbourhoods differ considerably from each other, and cannot be simply distinguished by their distance to main urban amenities. Housing units with the same or similar structural characteristics and at the same distance to CBD and other key urban amenities but in different neighbourhoods command quite different prices and house quite different types of households. The differences in both the supply and demand sides for a given type of housing stock between different neighbourhoods underline the importance of the housing submarket structure, indicating that housing submarket analyses will be subject to aggregation bias if they fail to accommodate the existence of housing submarket structure.
The findings are useful in gaining a better understanding of housing market segmentation in urban China. The submarket analytical framework developed in this study provides a useful basis to examine the dynamic process of submarket formation and operation. It may be used to target submarkets or household groups for a better match between different household groups and housing clusters in urban China over time.
‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑
Notes
1. From the highest to lowest, hierarchical administrative ranks included provincial or ministerial level (Buji), bureau or district prefecture level (Di/Juji), department level (Chuji), branch level (Keji) and section level (Guji).
2. Work units were divided into three categories: government organization/party agency, non-profit institutions (Shiye Danwei, including educational, medical, research and design institutions etc), and enterprises.
3. The property rights of enterprises were divided into state-owned, collective-owned and private owned ones.
4. Chinese Yuan, or called Renminbi, 1$ equal CNY ¥ 8.07 approximately in Dec. 2005.
5. Dendrograms are available upon request.
 
References
Adair, A. S., Berry, J. N. and McGreal, W. S. (1996) Hedonic Modelling, Housing Submarkets and Residential Valuation, Journal of Property Research, 13, 67-83.
Anas, A., Arnott, R. and Small, K. A. (1998) Urban Spatial Structure, Journal of Economic Literature, 36(3), 1426-1464.
Ball, M. (1986) Housing Analysis: Time for a Theoretical Refocus?, Housing Studies, 1(3), 147-166.
Bowen, W. M., Mikelbank, B. A. and Prestegard, D. M. (2001) Theoretical and empirical considerations regarding space in hedonic housing price model applications, Growth and Change, 32(4), 466-490.
Brueckner, J. K., Thisse, J.-F. and Zenou, Y. (1999) Why is Central Paris rich and downtown Detroit poor? An amenity-based theory, European Economic Review, 43, 91-107.
Cheshire, P. and Sheppard, S. (1995) On the Price of Land and Value of Amenities, Economica, 62, 247-267.
Clark, A. V. W., Deurloo, M. C. and Dieleman, F. M. (1984) Housing Consumption and Residential Mobility, Annals of the Association of American Geographers, 74(1), 29-43.
Dubin, R., Pace, R. K. and Thibodeau, T. G. (1999) Spatial Autoregression Techniques for Real Estate Data, Journal of Real Estate Literature, 7, 79-95.
Everitt, B. S., Landau, S. and Leese, M. (2001) Cluster Analysis, Arnold, London.
Feitelson, E. (1993). A Hierarchical Approach to the Segmentation of Residential Demand: Theory and Application, Environment and Planning A, 25, 553-569.
Fisher, E. and Winnick, L. (1953) The Reformulation of the Filtering Concept, Journal of Social Issues, 7, 47-85.
Gloudemans, R. J. (1990). Adjusting for time in computer-assisted mass appraisal, Property Tax Journal, 9, 83-99.
Grigsby, W. G. (1963). Housing Markets and Public Policy, University of Pennsylvania Press, Philadelphia, PA.
Hastie, T., Tibshirani, R. and Friedman, J. (2001). The Elements of Statistical Learning, Data Mining, Inference, and Prediction, Springer.
Jones, C., Leishman, C. and Watkins, C. (2003) Structural Change in a Local Urban Housing Market, Environment and Planning A, 35, 1315-1326.
Jones, C., Leishman, C. and Watkins, C. (2004) Intra-Urban Migration and Housing Sub-markets: Theory and Evidence, Housing Studies, 19 (2), 269-283.
Maclennan, D. and Tu, Y. (1996) Economic Perspectives on The Structure of Local Housing Systems, Housing Studies, 11(3).
Quigley, J. M. and Weinberg, D. H. (1977) Intra-Urban Residential Mobility: A Review and Synthesis, International Regional Science Review, 2(1), 41-66.
Rothenberg, J., Galster, G. C., Butler, R. and Pitkin, J. (1991). The Maze of Urban Housing Markets, The University of Chicago Press.
Smith, L. B., Rosen, K. T. and Fallis, G. (1988) Recent Developments in Economic Models of Housing Markets, Journal of Economic Literature, 26(1), 29-64.
State Council Paper No. 23 (1998). Notice to Further Reform in Urban Housing System and Housing Construction, State Council, Beijing.
Tu, Y. (1997). The Local Housing Sub-market Structure and Its Properties, Urban Studies, 34(2), 337-353.
Wang, Y. (1995) Permanent Income and Wealth Accumulation: A Cross-sectional Study of China Urban and Rural Households, Economic Development and Cultural Change, 43(3), 523-550.
Watkins, C. A. (2001) The definition and identification of housing submarkets, Environment and Planning A, 33, 2235-2253.
Zhang, X. Q. (1997) Chinese Housing Policy 1949-1978: The Development of a Welfare System, Planning Perspectives, 12, 433-455.
Zhao, Y. S. and Bourassa, S. C. (2003) China's Urban Housing Reform: Recent achievement and New Inequities, Housing Studies, 18(5), 721-744.