Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Sage Publications, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to
Urban Studies
Abstract
1. Introduction
Tilak Abeysinghe is in the Department of Economics, National University of Singapore, 1 Arts Link,
Singapore, 117570, Singapore. E-mail: tilakabey@nus.edu. sg.
Jiaying Gu is in the Department of Economics, University of Illinois at Urbana-Champaign, 41 0 David
Kinley Hall, 1 407 W. Gregory, Urbana-Champaign, Illinois, 61 801 , USA. E-mail: gu1 7@illinois.edu.
Singapore's private housing sector has also Secondly, an intertwined housing price
grown over the years. Although originally structure shapes the Singapore housing
intended to serve mainly the high-income market now. There are three housing sub-
earners, growing incomes have opened up the markets in Singapore: the new HDB flats
private residential market to a large segment market, the HDB resale flats market and the
of the population. Quite contrary to some private property market. As the prices of the
perceptions that the private housing market latter two segments are market determined,
caters primarily to foreigners, data from the they tend to be affected by a similar set of
Real Estate Information System (REALIS) macroeconomic conditions and price move-
show that between 1997 and 2005 about 82 ments stay closely aligned, although private
per cent of the private housing stock was houses command a much higher price.3 As for
owned by Singapore citizens and permanent new HDB flats, since 1990, the government
residents and close to 90 per cent of newhas revised its objective from affordability to
buyers of private housing are local people.quality. Although the new flats are still sold
Even the rental demand for private housing at a subsidised price, the price is partially
by foreigners has not been that high; in 2003,pegged to the resale market price (Ng and
owner-occupation of private housing was Chow, 2004). In fact, despite the dominance
about 88 per cent (DOS, 2005). Governmentof the public housing segment in Singapore,
policies have also been shifting to rely more the private housing market has come to domi-
on the private sector to meet housing needs. nate the price trends of the housing market in
With a higher population target of 5.5-6.5general. A causality analysis within a cointe-
million residents, the demand for housinggration framework as discussed in Rajaguru
will continue to increase. With aspirationsand Abeysinghe (2008) clearly shows that the
to upgrade to private housing rising highprivate housing market leads in setting the
(DOS, 2006) and several episodes of house price trends in the Singapore housing market.
price escalations in the past, housing afford- In this context, a measure of housing
ability has become a hot issue among the affordability would be a useful indicator
potential buyers. for policy-makers. The literature shows that
There are several factors in operation inresearchers have used different measures
this regard. First, some major policy changesof housing affordability. Some common
have led to an increasing interaction betweenmeasures include a ratio of housing cost to
the public and private housing sectors. Forcurrent income or mortgage payments to
example, compulsory savings in the Centralcurrent income (Keare and limenez, 1983;
Provident Fund (CPF) that could only be with-Kamath, 1988). A measure of accessibility
drawn for the purchase of public housing havedefined by the Australian National Housing
been allowed for private housing purchasesStrategy (NHS, 1991) assesses the households
since 1981. Furthermore, the HDB resale mar- ability to afford a downpayment. Considering
ket has been deregulated since 1989 to allow Singapore's upgrading phenomenon, Ong
HDB-dwellers to purchase private property.and Sing (1999) suggested a modified mea-
The provision of subsidised funds and the sure of affordability, which they termed a
removal of policy restrictions have bridgedthreshold upgradeability index'. This index
the public and private sectors and directlyconsiders the scenario that a household in the
triggered the upgrading trend in the Singaporepublic housing sector qualifies for upgrading
housing market. Upward housing mobility is to private housing if the resale of the HDB flat
well noted in the local literature (Lee and Ong,generates enough cash for a downpayment
2005; Yuen et al> 2006; DOS, 2006). and the household's current income level is
policy implications. W = £
(2)
2. House Price and Lifetime ^rp(i+1 (1+r) )
r
Income
uses personal savings for buying a house or by extending the amortisation period, AT,
for a downpayment, this still involves a cost which amounts to using up more of lifetime
in terms of forgone interest earnings. For earnings for interest payments.
simplicity, we can set the house price equal Because of the difficulty of estimating
to a fixed- rate N- year mortgage quantum L lifetime income and permanent income, in
(loan), which requires an annual repayment practice Rt / Yt and Pth / Yt are often used
of R . If the mortgage rate is r and house price to assess housing affordability. However, as
is Ph then we have Quigley and Raphael (2004) have argued,
the use of current income ( Yt ) may produce
N 1
a misleading picture of affordability depend-
Ph=L = Ry£(l + rTn=R(-^ 1
n= 1 ^ ing on which income level is picked on the
age-income distribution. Gan and Hill (2009)
This shows that Ph is a stock measure and advocate using Pth / Yt by measuring it over
R is a flow measure. Therefore, we obtain a the entire distribution and taking the aver-
meaningful assessment of long-term hous-age instead of just focusing on the median
ing affordability only if we compare thehouse price and median income. As we shall
house price with a stock measure like lifetimesee in Figure 1, the use of an unfiltered cross-
income (wealth) and mortgage payment withsectional distribution may still distort the
a flow measure like permanent income. affordability measure.5
Lifetime income and permanent income Moffitt (1982, 1984) devised a method
are well established concepts in economicsfor constructing lifetime income based on
under the life cycle and permanent income observed household income by age. Although
hypotheses of consumption (Modigliani andMiron (1984) also estimated lifetime income
Brumberg, 1954; Friedman, 1957). Under assuming constant income growth rates,
the permanent income formulation, lifetimeMoffitťs method is more promising. We adapt
income or wealth ( W) is defined as the cur- Moffitťs method to construct aggregate mea-
rent income plus the discounted present valuesures of lifetime income of Singapore house-
of expected future incomes, where income isholds based on some limited data available.
broadly defined to include both labour andIn Singapore, time-series data on mortgage
non-labour incomes. Permanent income ( Yp payments
) are hardly available. Therefore, we
is the annuity value of the wealth. For illustra-focus only on comparing house price with
tive purposes if income ( Y) is earned over A+ 1measures of lifetime income to assess housing
periods, we can write affordability.
O I I I I I I I I I
the Government of Singapore, provided usCM
uuuuuuuuu
with unpublished data on household income
by age.6 These data span over 13 years (1990, m o m c LnounoLH
§ J. vi i vi i vi i vi J
val to 60-64 to have equal five-year intervals.c') oo N h ve m m Tf Tf
uuuuuuuuu
We obtained the data for the three income
uuuuuuuuu
We simply refer to these as lower, median and
upper income quantiles.7 o m c m o m o m o
cr-)
Ideally, we need a proper panel dataset to O I I
O V£> VC I H ^ H VÛ
I I I I I I I
'05 5 I I I I I I I I I
for every birth cohort. The problem is pre- > 0' ,-'CH^H'Oh'OH
r-i vc vû m m ^ ^ m m
rO
sented in Table 1. In the table, birth cohorts 03
UUUUUUUUU
03
are indicated by Cxx-xx. For example C66-70 "O
mc: m o m o m o m
CD K hh^voinin^^m
refers to the sample group that was born in E os I
r-yCH^)H'ÛH'ÛH
I I I I I I I I
o
1966-70. As highlighted in the table, incomes u
»-H h^^min^^cofo
UUUUUUUUU
c
for the cohort C66-70 are available only over
E m o m o m o m o m
the 20-39 age range. We need, therefore, a way o un i> vo vo m m ^t1 co
on I I I I I I I I I
to fill in the missing income points in order to c
ON
f-H
,-^C
vc vo
H^H'ÛH'ûh
m m m co
UUUUUUUUU
get complete income streams for each cohort. Q_
t
There are many cohorts alive in one particu- O c inoinomomo
_c
lar year. Therefore, the age-income profile in o
on I I I
ON ^C r-H VC> r-H V£) VO r-H VC>
I I I I I I
u t-H 'c vomin^^romfN
a given year will be a misrepresentation of the u uuuuuuuu
bimodal, which stands counter to the hump- us to assume fixed cohort effects in the regres-
shaped age-income curve predicted by the life- sion model that follows.9
cycle model.8 Since various cohorts stay pooled After arranging the available data in an
in a given cross-section, the life-cycle compo- unbalanced panel format, we can use the
nent and the cohort effect remain mixed. As following regression of income to generate
an economy progresses, cohorts may differ a complete income profile from age 20 to 64
from each other due to changing education for each cohort in our sample
and economic opportunities and other factors.
Deaton (1994, 1997) regarded these differences
log Yit- H+ax Age i +a2 Age] +ß'xt+ ^
as a cohort effect which shifts the life-cycle Y,Y,Cohort!+£li
age-income profile upward if each successive
generation becomes better- off. If we plot the where, i - 1,2, ..., A is the the zth age; t - 1,
age-income profile by different cohorts, as in 2, ..., Tis the ith year; and j= 1, 2, ..., / is the
Figure 2, the hump shape emerges. Figure 2 jth cohort; in a balanced datasetj = A-z+řand
also shows that the age-income profiles for / = A+T-l. Further, x is a vector of variables
different cohorts are roughly parallel to each to capture business cycle effects; and Cohort
other, especially for later cohorts. This allows represents cohort dummies.10
Note that equation (3) comes underThe whatcohort coefficients, on the other hand, are
is known as the age-period-cohort (APC)easier to handle. Although they are estimated
analysis, especially in the medical statistics
from past data, it is usually the case that young
literature (Holford, 2006). A purely dummy
people project their income profile by looking
variable approach to the three effects leads to income profile of their elders in similar
at the
an unidentified model because of the perfect
professions. Therefore, if necessary, the cohort
collinearity among the three time variables
coefficients of the young generations can be
(period - age = cohort). Although statistical
calibrated to shift the income profile up or
methods are being devised to solve the down
iden- to produce some scenarios of afford-
tification problem (McKenzie, 2006; ability
Yang for policy purposes.
et al. , 2008), Heckman and Robb (1985) For the computation of our housing afford -
argued against such mechanical approaches
ability index, we need the lifetime income in
that disregard the underlying subject mat- terms. However, our computations
nominal
ter. In this spirit, we tried to capture
alsothe
shed light on lifetime income inequality
which
period effect that represent business cycle should be measured in real terms. We
used the annual consumer price index over
effects common to all households in a given
period by using a composite GDP index 1990-2007
of to deflate the nominal income
Singapore's trading partners and Singapore's
figures to obtain real incomes. In all, therefore,
unemployment rate as x variables inwe
equa-
ran six regressions to obtain our results, the
tion (3). These variables did not improve the income quantiles each in both nominal
three
goodness of fit of the model, so we proceeded
and real terms. For illustrative purposes, we
to estimate (3) without ß'xt . report some results for the median income
quantile. Based on 153 observations and
A disadvantage of having the x variables
in the model is that we have to predictusing
theirthe mid-points of age-groups as age,
future values well over 30 years to generate
the estimated regression for median nominal
income is11
future income data for younger generations.
growth (cohort) effect. Thirdly, the move- discounted PV in equation (5) is computed,
ment from A to C represents the longitudinal we have to use a fixed r since future interest
movement resulting from the combined rates are not known at age a.
life-cycle and growth effects. Since we can We choose a = 30 in equation (5) as roughly
compute the lifetime income from each the age at which a household considers hous-
curve in Figure 3, we can construct a cohort ing affordability. We obtain the discounted PV
income profile that would provide us with a of the income stream from age 30 to 64 using
time-series of lifetime incomes. two discount rates, 5 per cent, which has been
After generating a complete income the average prime lending rate during the
profile for each cohort from the regres- observation period, and 8 per cent; to assess
sion method, we can compute the lifetime the impact on affordability when the mort-
income or wealth relevant for a housing gage rate goes up. Although we can compute
purchase as: W for different choices of a, computations
based at age 30 provide sufficient information
A Ý for policy analysis.
K=Y a ^ - - +K a~l i
a ^ /i + , „y-a a~l i With a = 30 we have to generate accumu-
i=a(l /i + , r) „y-a (5)
lated savings from age 20 to 29 to obtain W
from equation (5). We can estimate savings
where, Wa is the wealth at age a for a house-
of young households as 5 - ý :s , where s¿ is
hold in a given cohort; y. is the estimated
the savings rate for i = 20, 21, ..., 29. Singapore
(expected) income at age i ; and r is has
the dis-
conducted a household expenditure sur-
count rate.
vey since 1972/73 at five-year intervals (see
The first term on the right-hand DOS,
side 2009a).
of From the age-expenditure and
equation (5) is the discounted present value
age-income distributions published in these
(PV) of the expected income stream from age we can compute expenditure rates
surveys,
a to A and the second term is accumulated
(expenditure/income) of young households
savings at age a- 1 or the initial wealth. The for lower, middle and upper income quan-
second term can easily be computed using tiles. The savings rate is then obtained as one
the standard recursive formula
minus the expenditure rate. After computing
the
where C is consumption expenditure).13 spline interpolation method in the SAS
Since
software package to obtain a time-series of
APS is available annually, we use the predicted
values from these regressions to estimate sav-at the annual frequency.
wealth
ings rates at the annual frequency. In 1972/73
Before we proceed to constructing a hous-
ingvery
the savings rates of young households were affordability index, it would be useful to
small (only 3 per cent for the lower income
shed some light on lifetime income distribu-
tion.
group). As a result, the predicted savings For this we set a - 20 and A - 64 and r
rates
= 0.05
from APS become zero and negative when wein equation (5). Since earning starts
at age 20, the second term on the right-hand
move to years before 1970. We set the negative
savings rates to zero. side of equation (5) becomes zero at age 19.
The next issue is what interest rate to be Figure 4 presents real lifetime income of
used to accumulate savings. More than 70 households at different quantiles by birth
per cent of the financial assets of Singaporeyear of household heads. It is interesting
to see that real lifetime incomes of cohorts
households are held in currency and deposits
and in the CPF (DOS, 2009b). Therefore, we
born before the 1960s were stagnant for all
use the weighted average of the interest rates
the three quantiles. With the rapid growth of
independent Singapore since the 1960s, the
for savings deposits and CPF ordinary account
lifetime income of later cohorts also grew
deposits as the appropriate interest rate. The
rapidly. Growth of lifetime income slowed
share of currency and deposits has fluctuated
down for cohorts born after 1975. These
around 56 per cent over the years; therefore,
we use the weights 0.56 and 0.44 for the two
cohorts entered their working age after the
interest rates respectively. With savings andmid 1990s when the Singapore economy
interest rate series in hand, we can accumulate
entered a turbulent period starting with the
savings from age 20 to 29 by working through Asian financial crisis in 1997. What is most
striking to note is that the income inequal-
the recursion Wa_, = (1 + rfl_2 )Wa_2 + Sa_x by
assuming Wļ9= 0- ity, even when measured in terms of lifetime
Note that, because of the grouping of theincome (instead of the usual annual income),
income data to five-year age-intervals, our
has not only widened but the gradients of
wealth measure from equation (5) providesthe income profiles of the upper and lower
a time-series at five-year intervals centred income
at groups have moved in opposite direc-
the mid-year of each five-year birth cohort. tions for the cohorts born after 1975. Even
Since W moves smoothly over time, we apply
for the median income category, the lifetime
Figure 4. Real lifetime income by birth year of household head at different income quantiles.
income has not shown much growth sinceto properties which are beyond their means.
the Asian financial crisis. The question is how to define an optimal
cut-off point. If we can define a standard non-
housing expenditure share of income that is
3. The Housing Affordability Index
uncorrelated with housing expenditure, then
(HAI)
we can work out the residual lifetime income
and then obtain the HAI. HAI = 1 in this case
We define our housing affordability index
(HAI) as is a natural cut-off point, HAI<1 implying a
housing-induced reduction of non-housing
W
expenditure. Results in Abeysinghe and Choy
HAI aytayt
t = - ttjjf-
h h
housing affordability index for the 30-year to define a cut-off value for our HAI. In
Figure 5. Average price of private housing and HDB resale flats (quarterly).
cent), five-room ($340000; 23 per cent) and across income groups. If we use the
property
executive ($415000; 8 per cent) flats. 2.86 cut-off value mentioned earlier, the mag-
Figure 5 shows the average price nitude
levelsof the numbers in Table 2 indicates that
for both the private and public residential
private properties become unaffordable even
properties. Both price series follow for
similar
the first-time buyers of the 75th income
trends and turning-points with a sustained
percentile when property prices escalate as in
the mid
large price gap of similar magnitude over the1990s and in 2007. In fact, the house-
hold expenditure survey (DOS, 2009a) shows
years. Since 1980, the average annual growth
of private residential property prices wasaverage monthly household income of
that the
about 1 1 per cent, with prices increasing,
private flat owners in 2007/08 was S$ 1 6 3 1 1 and
on a year-on-year basis, by 102 per that
centforinlanded property owners it was S$20 427.
1981Q1, 47 per cent in 1994Q3 andThese
31 perincome levels fall into the top 20 per cent
cent in 2007Q4. The price of HDB resale
andflats
top 10 per cent income groups respectively.
has followed similar cycles. Therefore, the lower affordability we observe
Table 2 presents the HAI (ignore
forHAI-
the 75th percentile is not surprising; private
adjusted for the moment) by income quantiles
properties are only for the top income groups.
for private and public residential properties
The same computations in Table 2 for aver-
computed for the 30-year-old age-group at HDB resale flats show a much bet-
age-priced
ter affordability picture. Although the price
the 5 per cent discount rate. Figure 6 highlights
the cyclical movements of the HAI forbubble in the mid 1990s led to an erosion of
private
properties for the three income quantiles.
the Since
affordability of HDB resale flats, even for
the cohort income profile moves smoothly,
the low income group the HAI remains about
cycles in HAI are primarily determined by average-priced
4 for the flats; in other words, their
lifetime
gyrations of property prices. The largest dropincome is about four times the aver-
in HAI across income groups occurred age
in price
the of a resale HDB unit or 23 per cent
early 1980s and housing affordability of their lifetime income (see the bottom row
never
recovered to the pre- 1980 levels. The
of price
Table 2). It should be noted that housing
bubble in the mid 1990s resulted in another grants provided by the government to qualify-
substantial erosion of affordability of private ing low- income groups are not accounted for
Table 2. Housing affordability index for 30-year-olds by income quantile (r = 5 per cent)
Year Lower Median Upper Lower Median Upper Lower Median Upper
1975 2.8 5.0 8.0 - - -- - -
1976 2.7 4.9 7.7 - - -- - -
19 77 2.7 5.0 8.1 - - -- - -
1978 2.7 5.0 8.2 - - -- - -
1979 2.3 4.2 7.0 - - -- - -
1980 1.3 2.5 4.2 - - -- - -
1981 0.8 1.5 2.6 - - -- - -
1982 0.8 1.5 2.6 - - -- - -
1983 0.7 1.3 2.4 - - -- - -
1984 0.8 1.4 2.7 - - -- - -
1985 1.0 1.8 3.3 - - -- - -
1986 1.2 2.1 4.0 - - -- - -
1987 1.0 1.9 3.6 - - -- - -
1988 1.1 1.9 3.6 - - -- - -
1989 1.0 1.8 3.4 - - -- - -
1990 1.0 1.7 3.2 1.2 2.1 3.8 6.0 10.6 19.5
1991 1.0 1.7 3.1 1.1 2.0 3.6 6.4 11.2 20.3
1992 0.9 1.6 3.0 1.1 1.9 3.5 6.4 11.0 19.9
1993 0.8 1.3 2.4 0.9 1.6 2.9 4.5 7.8 14.0
1994 0.6 1.0 1.8 0.7 1.2 2.1 3.8 6.6 11.7
1995 0.5 0.9 1.6 0.6 1.1 1.9 3.3 5.7 10.0
1996 0.5 0.9 1.5 0.6 1.1 1.9 2.5 4.3 7.6
1997 0.6 1.0 1.8 0.8 1.3 2.3 2.6 4.5 7.8
1998 0.8 1.5 2.5 1.1 2.0 3.4 3.4 5.9 10.2
1999 0.9 1.5 2.6 1.2 2.0 3.4 3.7 6.4 11.0
2000 0.8 1.4 2.4 1.1 1.8 3.1 3.8 6.6 11.1
2001 1.0 1.7 2.8 1.3 2.2 3.7 4.4 7.6 12.7
2002 1.1 1.9 3.2 1.4 2.5 4.1 4.7 8.2 13.6
2003 1.1 2.0 3.3 1.5 2.7 4.4 4.6 8.0 13.3
2004 1.1 2.1 3.4 1.5 2.8 4.6 4.4 7.9 13.1
2005 1.1 2.0 3.4 1.4 2.7 4.5 4.4 8.2 13.7
2006 1.0 1.9 3.2 1.3 2.5 4.2 4.3 8.3 14.0
2007 0.8 1.6 2.7 1.0 2.0 3.3 3.8 7.7 13.0
Ave 1980-2007 0.9 1.6 2.9 1.1 2.0 3.4 4.3 7.6 13.1
Price-income ratio 111 63 34 91 50 29 23 13 8
percentage, 1980-2007
Figure 6. HAI of private property for 30-year-old age-group at different quanti les.
income. Despite the emphasis in the literature price escalation have led to a substantial ero-
on the importance of focusing on long-runsion of housing affordability.
income measures, housing affordability is A natural question to ask then would be,
usually assessed based on short-run measures with more than 90 per cent homeownership
that compare mortgage payments or house(both public and private), why should there
price with current incomes. Difficulty of com-be any concern about property price escala-
puting lifetime income due to data constraintstions if higher prices mean higher wealth for
has been the main reason for not venturing into Singaporeans? Although there is no question
long-run measures of housing affordability. that a higher price means a higher value of
Many developed countries publish survey- the housing stock, how this translates into a
based data on household income by age of'wealth effect' is what matters for the aggre-
household heads. Based on these data for gate economy. Abeysinghe and Choy (2007)
Singapore, we have presented a regressionhave examined in detail the wealth effect of
procedure to predict the entire income pro-
property prices on consumption in Singapore
file over the working life of different birth
and found that the wealth effect is very much
absent. In the absence of cheaper suburbs
cohorts. We have then presented a procedure
to obtain lifetime income or wealth for each which offer quality living, the only way for
birth cohort as the discounted present value Singapore residents to unlock property val-
of the future income stream plus accumulatedues is, apart from emigrating, to downgrade
savings at the age at which a house purchase to smaller units. This does not seem to be
occurs. We measure housing affordability by happening extensively (DOS, 2006) and
the ratio of lifetime income to house price so explains why the 'housing wealth effect' on
that the increase in the index implies improv-consumption is insignificantly small. Instead,
ing affordability. This measure is likely to be Abeysinghe and Choy (2007) observe the
better than the ratio of mortgage payments presence of a negative and significant 'price
to permanent income because the monthlyeffect' of house-price escalations on con-
mortgage payment can be reduced by increas- sumption expenditures, leading to a fall in
ing the amortisation period that does notthe average propensity to consume (see also
amount to improved affordability in the Ludwig and Slok, 2002; Phang, 2004). This
long run. Although our focus is on long-run fall is present even in the different income
housing affordability, our measure of wealthquantiles that we examined in this exercise to
also takes into account the short-run afford- compute the savings rates. With the 99-year
ability through accumulated savings. Despite lease system, even the bequest value of higher
the initial computational involvement, theproperty prices is likely to be dominated by
procedure can easily be automated for ready the negative 'price effect'.
updating of the affordability index. Overall, there seems to be a 'paradox of
Our estimates of lifetime income in housing price' in operation here. If in aggre-
Singapore for three income quantilesgate
showthe 'price effect' of high property prices
that even the lifetime income inequality
outweighs
has the wealth effect, it is important
been increasing rapidly, especially sincethat
the property prices do not escalate to erode
Asian financial crisis. In fact, despite the affordability. The average growth
housing
substantial growth of the economy, the rate of lifetime income for cohorts born after
lower
income quantile has seen a drop in their1960
realfor the median quantile has been about
4-5 per cent which has also been the average
lifetime income. As for housing affordability,
growth rate of per capita disposable income
our index shows that past episodes of house
6. The
since 1975. Property prices should fall in (gross) income data defined in the
household expenditure survey refer to regular
line with this trend. Although it is difficult
income from work or employment, as well
to avoid property price cycles, policies could
as income received from rental, investment
be devised to reduce the amplitude of these
(for example, interest and dividends) and
cycles. In this regard, it is worth question-
other sources such as pensions and cash
ing why one should let the private housing
contributions received from relatives. Irregular
market - that accounts only for about 20orper extra-ordinary receipts like proceeds from
cent of the housing stock - dominate the price
sale of properties, or one-off payments such as
trends of the entire housing market and erode
lump-sum CPF withdrawals, insurance claims
housing affordability. and Economic Restructuring Shares from the
government, as well as rebates and waivers on
Notes rent and utilities for HDB flats are not included.
7. Note that many developed countries publish
1. It is common in Singapore to use the term income by age annually; therefore, our
'public housing' to refer to public-sector- methodology is not constrained by the
provided flats owned by private individuals. data. Although Singapore does not publish
These flats are built by the government and sold these data, they can be obtained from the
to Singaporeans at subsidised rates, but subject Department of Statistics on a regular basis.
to a 99-year lease term. The main government
8. The bimodal income pattern in a cross-section
agency responsible for public housing is the is likely to result from women withdrawing
Housing and Development Board (HDB). from the labour force or taking no-pay leave
2. The HDB concessionary mortgage rate is during child bearing and rearing years.
pegged at 0.1 per cent above the Central
9. We provide plots for the median quantile
Provident Fund Ordinary Account interest only because the three income quantiles share
rate. In 2008, the HDB concessionary interest similar shapes.
rate was 2.6 per cent, 3 points lower than the
10. Going by the standard practice, we use log Y
average interest rate of banks and financial to reduce heteroscedasticity and skewness of
companies on housing loans for 15 years income data and also to provide coefficients
(www.hdb.gov.sg; www.mas.gov.sg). As of that represent the growth effect on income.
2008, to be eligible for the HDB concessionary The quadratic age function is well justified by
loan, monthly household income should not the observed hump shape of the age-income
have exceeded S$8000. profile (see Attanasio and Browning (1995) and
3. The demand for HDB resale flats arises Attanasio and Guglielmo (1995) for the hump
primarily from those households who cannot
shape in UK and US income data respectively).
afford private housing, who do not qualify1 for
1 . Note that, in order to have more observations
new HDB flats, who want to avoid the long for age-groups, we interpolated 1991-94 values
waiting time for a new HDB flat and who are
assuming constant income growth rates. We
concerned about the location. did not do the same for 1996 because of the
4. A review of literature on various social distorted growth rates resulting from the Asian
and macroeconomic consequences of financial crisis in 1997.
homeownership can be found in Dietz and 12. Given the nature of the data and the type of
Haurin (2003). regression, a residual analysis for diagnostics
5. Note that the main objective of Gan and Hill will be of little help in this case.
(2009) was to offer another affordability index 1 3 . These disaggregate savings rates and APS show
based on a criterion similar to value-at-risk. An a close correspondence. The R2 values for
informal exercise that tries to use permanent lower, middle and upper income regressions
income to measure housing affordability in were 0.82, 0.87 and 0.94 respectively.
the UK can be found at: http://boards.fool. 14. According to DOS (2006), in 2005 about 85 per
co.uk/Message.asp?mid=l 1698976. cent of those who shifted to larger units had
outstanding HDB loans exceeding $100 000. Deaton, A. (1994) Saving, growth, and aging
Relevant payments include an upgrading cost, in Taiwan, in: D. A. Wise (Ed.) Studies in the
applicable for the HDB flats that are affected Economics of Aging, pp. 331-357. Chicago, IL:
by HDB's Main or Lift Upgrading Programme Chicago University Press.
and an upgrading levy, applicable to the flats Deaton, A. (1997) The Analysis of Household
in an upgrade precinct. Surveys: A Microeconometric Approach to
15. We do not report the results at the 8 per cent Development Policy. Baltimore, MD: Johns
discount rate to conserve space. The results Hopkins University Press.
show that the average HAI values over 1980- Dietz, R. D. and Haurin, D. R. (2003) The social
2007 remain above 3 for HDB flats and that and private micro-level consequences of
for private properties fall to an unaffordable homeownership, Journal of Urban Economics,
value of 2.3, even for upgraders. 54, pp. 401-450.
DOS (Singapore Department of Statistics) (2005)
Home ownership and equity of HDB households
Acknowledgement 2003. Occasional paper, Singapore Department
of Statistics, Government of Singapore.
The authors would like to thank six anonymous
DOS (2006) Housing mobility between 1995 and
referees of the journal for their insightful
2005. Occasional paper, Singapore Department
comments. Thanks are also due to Basant Kapur,
of Statistics, Government of Singapore.
Cheolsung Park, Kim Jong Hoon and Lim Kim
Lian for their constructive comments on an DOS (2009a) Report on the household expenditure
earlier version of the exercise and to Chua Mui survey 2007/08. Singapore Department of
Statistics, Government of Singapore.
Hoong, Janadas Deven and a number of readers
DOS (2009b) Statistics Singapore newsletter.
of a newspaper article on the topic by the authors
Singapore Department of Statistics, Government
for their valuable suggestions and comments.
of Singapore, September.
Friedman, M. (1957) A Theory of the Consumption
References Function. Princeton, NJ: Princeton University
Press.
Abeysinghe, T. and Choy, K. M. (2007) TheFullerton, D. and Rogers, D. L. (1993) Who bears
Singapore Economy: An Econometric Perspective. the lifetime tax burden7. Brookings Institution,
London: Routledge. Washington, DC.
Attanasio, O. and Browning, M. (1995) ConsumptionGan, Q. and Hill, R. (2009) Measuring housing
over the life cycle and over the business cycle, affordability: looking beyond the median,
American Economic Review, 85, pp. 1 1 18-1 137. Journal of Housing Economics, 18, pp. 115-125.
Attanasio, O. P. and Guglielmo, W. (1995) Is con-Gans, J. and King, S. (2004) The housing lifeline:
sumption growth consistent with intertemporal a housing affordability policy, Agenda, 11,
optimization? Evidence from the consumer pp. 143-155.
expenditure survey, Journal of Political Economy, Goodman, A. C. (1988) An econometric model
103, pp. 1121-1157. of housing price, permanent income, tenure
Bogdon, A. S. and Can, A. (1997) Indicators of choice and housing demand, Journal of Urban
local housing affordability: comparative and Economics, 23, pp. 327-353.
spatial approaches, Real Estate Economics , 25,Goodman, A. C. and Kawai, M. (1982) Permanent
pp. 43-80. income, hedonic prices, and demand for hous-
Bohlmark, A. and Lindquist, M. J. (2006) Life-cycle ing: new evidence, Journal of Urban Economics,
variations in the association between current12, pp. 214-237.
and lifetime income: replication and exten-
Haider, S. and Solon, G. (2006) Life-cycle varia-
sion for Sweden, Journal of Labor Economics ,tion in the association between current and
24, pp. 849-896. lifetime earnings, American Economic Review,
Deaton, A. (1985) Panel data from time series 96, pp. 1308-1320.
Heckman, J. and Robb, R. (1985) Using lon-
of cross-sections, Journal of Econometrics, 30,
pp. 109-126. gitudinal data to estimate age, period, and