Resource Allocation to Housing Investment: Comments and Further Results. The Demand for Housing in Developing Countries: The Case of Korea

Transcription

1 Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized World Bank Reprint Series: Number 183 Bertrand Renaud Resource Allocation to Housing Investment: Comments and Further Results The Demand for Housing in Developing Countries: The Case of Korea (with James Follain and Gill-Chin Lim) Determinants of Home Ownership in a Developing Economy: The Case of Korea (with Gill-Chin Lim and James Follain) Reprinted with permission from Economic Development and Cultural COange, vol. 28, no. 2 (January 1980), pp , copyrighted by the University of Chicago; Journal of Urban Economtics, vol. 7 (1980), pp ; and Urban Studies, vol. 17 (1980), pp

2 Resource Allocation to Housing Investment: Comments and Further Results* In a recent paper, Burns and Grebler (B-G) presented one of the first systematic investigations of the place that investment in residential construction occupies in the total output of an economy at various levels of development., Their analysis is particularly welcome because of the frequent lack of realism observed in formulating housing plans. Burns and Grebler show that there is a systematic nonlinear relationship between investment in residential construction and the per capita output of a country. This relationship is sufficiently stable and precise that one may ask under what condition their findings can be used as a norm. The term "norm" is used here in the sense of a simplified procedure capable of producing meaningful orders of magnitude to gauge planned levels of housing investment in an economy, before proceeding into a detailed analysis of the housing sector. To establish the conceptual foundations of their analysis, B-G reviewed three types of arguments concerning the proper size of housing in total output. First, they reviewed the argument of the "housers," who are "convinced that better dwellings and neighborhoods are the most effective and direct means of improving the human condition" (p. 96). The problem with this approach is that it can quickly substitute the goals of the planners to the wishes of households and lead to improper standards and the imposition of consumption patterns which intended beneficiaries might not choose if given alternatives. Since households do not have a housing problem as much as an income problem, the only recourse is a better knowledge of the structure of housing demand. At the aggregate level, if one overestimates the total volume of resources which "should" be available to the housing sector, there is a good chance that a few "high quality" (i.e., "high standards") projects will be produced and that they will crowd * This article draws on a recent study of the Korea housing market conducted for the World Bank. However, the author is solely responsible for this article, which may not be quoted as representing the views of the World Bank or its affiliated organizations, Special thanks are due to Leland Burns and Leo Grebler, who kindly made thci r original data available for further analysis. 1 Leland S. Burns and Leo Grebler, "Resource Allocation to Ihousing Investment: A Comparative Analysis," Eqconomioic Developmemtt and Cultural Clhange 25, no. I October 1976): (hereafter referred to by page number in text). to 1980 by The University of Chicago. 00l3-0079U80/ $

3 Economic Developmnent and Czultural C7hange out other more desirable forms of housing investment, the end result being mostly misallocation of resources within the sector with no overall increase in the total volume of investment. The second type of argument relates to the "productivity theory of housing," for which the authors themselves have found only limited empirical support or consistent evidence. A possible reasoni is that the theory is composed of two distinct elementts which shlouldl probably not be mixed. One strand consists in saying that better housing will improve health and the productivity of iabor, increase educational ettorts, anld reduce crime. Those are not unreasonable expectations; but such improvements could very well be achieved with limited investmenit in ind(ividual housing and major efforts in neighborhoodi improvements in roads, water supply, drainage, and sewerage as well as other utilities and public services, all of which will be accou.nted for under "other construction" or "social overhead" in national accouints. The other strand of the productivity theory of housing is hardly distinguishable from the third type of argument that housing im'estment is closely related to general dcveloprnent. Burns and Grebler define "the optimum split tetween housing in'estnlicnt as the point where the marginal contribution of housing to national ic-i nc just equals the decrease in the contribution oflalternative sectors resultl;ig from an additional investnment in housing" (p. 98). In other words, one wouild expect the level of investment in housing to be closely related to the opportuniity cost of capital in the country andl the expected returni on housing project. (which is the Keynesian concept of marginal efficiency of investmenit applied to housing). The share of housing investmnent in total output will increase when the scarcity of capital becomes less intenise. In that respect, it must be kept in mind that nonresi(lential conistruction could yield higher returns to capital than housing while providing just the same kinds of indirect employment andi income benefits so often mentioned as arguments in favor of housing programs. Using a cross-section sample of 39 countries Nwith a large representation of more developed economies, B-G report the following orders of magnitude for the ratio of investment in residenitial housing to GD)P, which they call the shalre of h1ousin1g in total output (SHTO): for $85_350 per capita GDP (in U.S. dollars), SHITO =- 2.75; for $380-1,250, SHITO () 4.42; andl for $1,300-3,000, SH1TO = Thcese are figures from their sample for the period (B-G, table 1, p. 103) in 1970 diollar eqtuivalents. They could hardly be utsed tlirectly for ainy country because of Neveral1 sources of variations in the level of SEHTO arotund these imagnitudes and also because of the measurement errors inhietenit in the dlata base used. Graphically, the results for the 39 couintries yiekl a rather smo)tli curve (fig. 1). An obvious test of the nonilinear relationiship established by B-G is to analyze the behavior of SHTO in a single country over time. In that 390

4 6 4 - t 0o( 1.() M0C FIG. 1.-Diagram showing the relationship between SHTO and GDP per capita. The arrows reflect the change in the value of SHTO and (GDP between the two sample periods and for a given country. Source.- Leland S. Burns and Leo Grebler, "Resources Allocation to Housing Investment: A Comparative Analysis," Economic Development and Cultural Change 25, no. I (October 1976' 1 I 1

5 Economic Developmeent and Ctultural Chlange case, there is no particular problem in the measurement of either GDP or GNP because they can be calculated in internally consistenl constant domestic prices. It is well knowni that international comparisons based on GNP converted into dollars at official exchange rates are biased. The research by Kravis et al. has shown that levels of output in dollar terrms overstate the gap between low- and high-income countries. 2 However, corrections for such distortions were not possible for the B-(i crosscountry sample. An analysis based on Korea would be instruclive in cross-checking the original fin(dings because it is a country which has experienced a very high rate of growth over the last 15 years. It is also a country where housing problems are quite serious, offering a good test for the stability of the nonliniear relation bcween h1ousin)g invcstnient and per capita output. Burns and Grebler rely on economic and demographic variables to explain SHTO. In a first test, we canl compare the Korcani results with the B-G cross-sectionial sample, relyinig exclusively on per capita GNP as an explaniatory variable. Table 1 sumniarizcs the performance ot' the Korean housing sector over the period , avcragcdl by 5-year periods corresponding to the successive economic planis. Also includsed are the oflicial projections of the current fourtlh 5-year plani and the sample means reporte(d by 13-C for the period in their paper. It can he seen that in Korea the beha\ ior of SHTO con rornoi very well to the nonlitnear relationship identitied by B-G.l BecauLse the stronigest finding of the B-G0 paper is that of a nlonllinear relationship between SHTO and per capita GDP, Ae have compared four common nonlinear models for both Korea and their sample (lata. An additional reason for ignorintg demograp;hic \ail i.les in this first series of tests is their lack of significanice in the second-period ( ) 2 The World 3ailk and the U.N. Statistical Office have been sponsouing a project on international comparison under the direction of Irving Krawis (see Irving KrlxaNki, Zoltanl Kenessey, Alan I leston, and Robert Sumnners, A.S'stten of International Compurison of (;i-os.% Protlduct anid Purchasing Power [Balfimrorc andi London: Johlins l lokins University Press. 1975] and Intcrnational (Comparison.s of 'Rta Product aniid( Pturchasing Poiver [Baltimore and ILondon: Johns Hopkins tnr.e i.itv Press, 19)781), -3 Care must be exercised when using the B-(; equation -. ith dollar values. All Korean GNP values are in constant 1970 wons onx ceried into Li.S. dollars at the 19)70 exclhange rate. The results are draiiarc.0r cll'. oifrerent from current woriis at the current exchange rate, as cani he seeni for selecte(d Nears of Korea per capita (INP: (inlp iii Current l)ollars (NI I at ( ontant Plrices (in isl ') i $iq) ) It mulst he ke;pt in midr(i that the tus. dollar is niet a coinstaint ya.rdlstick and has lost 4X,X8 ; of it'- donestic value ibctween 1970 (presumnied date for the 134(i samiiple) and V. 392

6 TABLE I SHARE OF HousiNG IN TOTAL OUTPUT: KOREA COMPARED WITH BURNS AND GREBLER SAMPLE Per Capita SHTO SHTO SHTO Population Urbaniza- GDP Projected Projectedt Projected Growth tion (in 1970 SHTO (Burns- (Korea, (Korea, Period Rate* Indicator* Dollars)* Actual* Grebler) eq. [51) eq. [31) Postwar transition ( ) First Plan ( ) Second Plan ( ) Third Plan ( ) Fourth Plan ( )... 59t 2.20t 865t 3.60t Sample means for group of 39 countries SOURCES.-Projections for the Fourth Plan were obtained from Republic of Korea, Bureau of Statistics, Economic Planning Board, Thze Fourth Five- Year Economic Development Plan, (Seoul, 1976). Other data on Korea were obtained from Republic of Korea, Bureau of Statistics, Economic Planning Board, Korea Statistical Yearbook, various years. Data for the Burns and Grebler 39-country sample were obtained from Leland Burns (personal communication, February 1978). NOTE.-SHTO = share of housing in total output. "5-yr average. iplan projections. Q~ 5a1 CD F. '.Q F.'

7 Economic Development and Cuiltural Change equation reported by B-G; their significance in the pooled data analysis reported by B-G is entirely due to the first-period sample. Four equation forms are used: (a) a quadratic form to conform with the B-G findings; (b) a reciprocal transformation implying an upper limit to the value of SHTO; (c) a logarithm-inverse transformation implying an upper limit for SHTO and also an inflection point and a curve pa;ising through the origin; and, finally, (cl) a double log regression. The results presented in table 2 confirm the B-G choice of a quadratic equationi (e.g., eq. [2]) for the 39-country sample. For Korea the results are statistically quite good for all models, but the quadratic form (eq. [1]) is not acceptable because it yields results inconsistent with the actual behavior of SHTO: The value of SHTO would peak at the unrealistically low income level of $375 because of the restricted range of income values. On the other hand, the two nonlinear equations embodying a limiting value for SHTO are quite satisfactory. The reciprocal transformation (eqq. [3] and [4]) yields limiting values for SHTO that are almost identical for Korea anid the 39-country sample at 5.4. The logarithm-inverse transformation (eqq. [5] and [6]) yields results that are also consistent with actual patterns observed for Korea and the cross-section sample. In particular, the limiting value for Korea of 7.58 is within the range of the highest levels of housing investment reported by B-G. In the B-G cross-sectional sample, the better fit of the quadratic equation can be attributed to the single U.S. observation wherc GDP per capita is much larger than for the rest of the sample and SHTO is low and equal to 4%,. Using the fitted equations for Korea yields the interesting result that by any of three standards the Korean Fourth Plan has selected a target value of SHTO of 3.60 which is low. It is two 'tandtrd errors or 36F below the predicted value of 4.90 derived from equation (3) (table 2). There is the possibility of serious imbalances in the housing market if Korean planners who can affect the level of investment in housing adhere to this target level of SHTO. The equations estimated until now yield very useful insights into the dynamics of residential construction in Korea. It remains to be seen whether the demographic variables used by B-CG could improve the model further. The results of alternative formulations of the nonlinear relationship suggestedi by B-G are presented in table 3. It is quite clear that the omission of demograplhic variables from the model involves no loss of statistical significance: When per capita GNP is omitted, the national rate of population growth is the most significant variable (see cqq. [2] and [3], table 3); when per capita GNP is incluided, all demog.aphic variables become insignificant. Alternatively, one may ask whether economic variables more closely associzited with the concept of the marginal efficieic) of investment in lhousing could yield improved statistical results. In addition to the per capita GNP variable which influences the demand for housing, we con- 394

8 TABLE 2 ALTERNATIVE MODELS OF THE NONLINEAR RELATIONSHIP BETWEEN SHTO AND GNP PER CAPITA Limiting Functional Form a b c Value SHTO R2 SEE F-Value Y= a+bx+ cx2: Eq. (1). Korea Maximum (.022) (.0001) j-3.85 x= $375 Eq. (2). 39 countries y= (.83) (.326) x= $1,570 Y= a-b xr: Eq. (3). Korea a= (17.36) Eq. (4). 39 countries a= (.08) In Y=~a-bx: Eq. (5). Korea = (6.47) (IJ162)* Eq. (6). 39 countries e'= (.018) (1.337)* In Y=~a+bIn x: Eq. (7). Korea None (.106) (1.188)* Eq. (8). 39 countries None a (.052) (1.372)* NOTE.-SHTO = share of housing in total output; SEE = standard error of estimate. *When In Y is used, the reported value of SEE must be converted to antilog values for comparisons with the other equations.

9 TABLE 3 EFFECT OF DEMOGRAPHIC VARIABLES Intercept URB URB2 DPOP DPOP2 I GNP R2 SEE F-Value a Eq. (1). SHTO ** -1.06** (3.28) (2.49) (.45) Eq. (2). SHTO ** 2.81** (5.56) (4.79) (.99) Eq. (3). SHTO ** ** (9.87) (6.90) (1.08) (3.34) (1.46) Eq. (4). SHTO *** (8.71) (4.62) (3.44) (6.83) (1.27) (7.80) NoTE.-URB = annual rate of growth of cities over 100,000 divided by the average rate of growth in national population; DPOP = annual rate of increase of national population; SEE standard error of estimate; SHTO share of housing in total output; standard errors are given in parentheses. "*Significant at 95%. level. ***Significant at 99% level.

10 Bertrand Renaud sidered two variables reflecting the supply of capital: the ratio of domestic savings to GNP (DOMSAV) and the rate of interest observed on the unregulated money markets (UMM), also referred to as the noninstitutional money markets. This rate of interest could be considered an acceptable proxy for the actual cost and scarcity of capital in a given period and is monitored by the Bank of Korea. In recent years it has fluctuated around 3.4% per month for the third quarter :sf the year when inflation has been about 10%,-15% yearly. In the absence of a significant mortgage market for housing in Korea, the interest rate on the curb market gives an indication of the financial constraints placed on housing investment during a given construction season. It can be seen from equation (3) in table 4 that both supply variables DOMSAV and UMM can be found significantly related to the level of housing investment with the right signs. However, wlheni they are introduced jointly with per capita GNP as in equation (6), they are found insignificant. In addition, their signs are unstable and are often contrary to expectations, a situation which may be caused by the high correlation between GNP and DOMSAV of.917. It is very likely that a more sophisticated model of housing finance would be effective in explaining the behavior of SHTO. However, since we are presently interested in a simple and robust procedure, we can conclude that the earlier models estimated in table 2 based only on the value oi' per capita GNP fully meet our objective and provide useful information. Thus, the analysis presented for Korea supports the earlier findings of B-G. In particular, the existence of a nonlinear increase of SHTO as the economy expands is supported. Whenever one is reviewing the housing sector of a country, it should prove useful to estimate a country-specific equation. However, two important considerations should be added to this discussion. They relate to the possible underreporting of housitng investment in low-income countries where squatters and illegal housinlg units are likely to be both significant and unreported in official statistics. In that case, the nonlinear relationship between SHTO anid GNP per capita may be an artifact. Another factor which may also influence the SHTO curve is the nature of the climate in the country studied. First, if the value of SHTO at very low levels of income is underestimated because of the important role played by squatter houising and because of the possible underreporting ol housing construction, what could be the magnitude of the bias? An order of magnitude can be estimated for the year This was the peak year in terms of squatter appearances in Seoul, where about 66f(' of all illegal units were concentrated in Korea. During that year, abolit 18,500 net nevk illegal units appeared in Seoul and 140,000 units were ollicially reported built arolind the country. 4 If we 4 The figures for the number of illegal units are derived from data compiled by the Seoul Special City Government. The share of illegal units found in Seoul is based on data reported in E. Mills and B. N. Song, "Korea's UJrbanization and Urban Problems," Korea Development Institute Working Paper no (Scoul, 1977), table A

11 00 TABLE 4 EFFECT OF VARIABLES REFLECTING FINANCING CONDITIONS Observa- Intercept 1 GNP DOMSAV I DOMSAV UMM R2 SEE F-Value tions (N) Eq. (1). SHTO *** -6.53*** (1.43) (5.81) (4.62) I Eq. (2). SHTO *** * (.28) (2.03) (.55) Eq. (3). SHTO ** -. 96*** (1.06) (1.02) (.25) Eq. 14). SHTO *** (1.41) (3.26) (.27) Eq. (5). SHTO *** C (.64) (3.47) (.22) Eq. (6). SHTO *** i4 (1,67) (8.84) (5.78) (.23) NoTE.-DOMSAV = ratio of domestic savings to GNP; UMM = unregulated money markets; SEE share of housing in total output. *Significant at 90('% level. **Significant at 95%, c level. ***Significant at 991c level. standard error of estimates; SHTO=

12 Bertrand Renaud assume that in Korea the value of a squatter unit is one-third that of a legal unit, the omission of squatter units would bias the SHTO curve downward for Korea in 1968 by about 6,74', (i.e., SHTO in 1968 could have been 3.35% instead of 3.21%-).5 This is not a very large difference compared with the standard error of the estimate in the regressions presented earlier. A more important factor is the possible role of climate in affecting the SHTO curve. Most of the countries in the B-G sample are in the temperate zone and have an important cold season. A comparison of SHTO for Korea and Taiwan, which are two countries with very similar economic growth patterns and relatively comparable cultures, suggests that climate may have an important impact oni the value of SHTO. At comparable levels of per capita GNP, the value of SHTO is systematically lower for Taiwan than for Korea (SHTO is 2% instead of 3%1,). Because of the rigorous climate, the scarcity of land, and the cost of construction materials (granite or cement rather than clay), the relative cost of housing is significantly hligher in Korea. Because of the very different relative prices between housing and nonhousing goods, the values of SHTO differ marlcedly between the two countries andl follow difterenit paths. If the climatic differential suggested by the conmparion of Korea and Taiwan is generalizable-and it should be -the implicationis of' the B-G analysis are very clear: The volume of residential conistruction activities will be the smallest in tropical counltries at the lowest level of per capita GNP. On the other hand, the lignificance of the housinlg sector will be tlhe greatest in temperate or cold countries at medium- to high-income levels. Because recent analyses show that the income elasticity of demand for housing is less than one in less developed countries (1.I)Cs) as well as in developed couritries,4 there appear to be important limiitaltiolns to relying on housinig as a stimulant to low-income tropical LDC economlies, as often advanced by those whom B-G have calledi "housers." Otlher avenuies for raising welfare and income levels may be niore pronmising, such as provision of other types of urban cer% ices which generate greater extcrnalities as well as employment opportunitics in the construction industry. World Bank BE:R1RA'.M) RE NAUDO 51The assumption that a s.quatter unit is worth one-third the value of a legal unit is basedl on unpublished survey data collected by the Institutc for Urban Studies of Yonsci ULiecrsity of' Seoul in 1970 on construction cost, anvd mnarket Nalues. If anmthing, it is overstating, the relative value of illegal hiousinig in conmparison witlh legal units. 6 See J. Follain, G. C. Lim, and B. Retiatid, "The Demiand for I lousilng in Korea," Journal oj Urban EcotnomiCs (in press). 399

13 JOURNAL OF URBAN ECONOMICS 7, (1980) The Demand for Housing in Developing Countries: The Case of Korea' JANES FOLLAIN The Urban Institute GILL-CHIN LIM Northwestern Unitersity AND BERTRAND RENAUD The World Bank Received April 21, 1978; revised September 20, 1978 This paper presents the results of an analysis of urban housing demand for Korea talking into account the most recent findings of housing demand analysis concerning specification and aggregation biases. In order to obtain correctly specified demand functions, a procedure based on a model of the housing market originally proposed by Muth is used. Drawing on the detailed land information available in Korea, this procedure permits the calculation of an individual price per unit of housing services for each household. The results show conclusively that both the income and price elasticity of the demand for housing services in Korea are comparable to those found in the United States: the income elasticity is smaller than one and the price elasticity is negative and smaller than one in absolute value. Given the number of countries found within the per capita income range between Korea ($700) and the United States ($7800), the finding that these two countries have comparable demand elasticities is of major significance: in the absence of good national estimates, the order of magnitudes found here would be used for other country analyses. I. INTRODUCTION The need for accurate estimates of the response of housing expenditures to changes in income and prices is even greater in developing countries 'The research reported in this paper is based on a larger study of the Korean housing market organized by the World Bank under the overall responsibility of Bertrand Renaud. However, responsibilities have varied for each section of the analysis and the names of the authors are listed accordingly in the resulting papers. Primary credit is due to James Follain and Gill-Chin Lim in this paper. Helpful comments were receivecd from several staff members of the World Bank and the Urban Institute, in particular, Gregory Ingram and Larry Ozanne. The authors are also grateful for the comments reoi,tived from Tong-Hun Lee. The project has also benefited from Raymond Struyk's contributic,ns to the design and the initial analyses of the survey. This study may not be quoted as representing the views of the World Bank and its affiliated organizations or the views of the Urban Institute /80/ $02.00/0 Copyrnght by Academic Press, Inc. All nghts of reproduction in any forn reserved.

14 316 FOLLAIN, LIM, AND RENAUD than in richer countries because housing problems are more severe and resources very scarce. Reliable elasticity estimates are needed for better forecasts of housing demand and the sound formulation of urban policies. They are central to the analysis of the incidence of property taxes on which so many local governments are or could be relying extensively. Further, in land-scarce countries such as Korea, it is important to develop better perspectives on patterns of city expansion or on the magnitude of suburbanization under the compounded impact of high rates of income and population growth. The results presented here are of interest for two additional reasons. First, the analysis takes into account the most recent findings on demand analysis which have shown that some of the previous uncertainty on the value of the income elasticity of the demand for housing services can be explained by a series of specification and aggregation problems (see Polinsky [17], Campbell and Smith [1, 23], and Lee and Kong [8]).2 In order to obtain correctly specified demand functions this paper uses a procedure based on a model of the housing market originally proposed by Muth. Drawing on the extremely detailed land information available in Korea, this procedure permits the calculation of an individual price per unit of housing services for each household. Alternative specifications of the demand equations allow the testing of the directions of the biases created by aggregation and specifications errors as recently discussed in the U.S. literature. Second, and more importantly, this analysis provides for the first time a consistent basis for comparison of the characteristics of demand under sharply different economic conditions: the per capita GNP in the United States was $7867 and it was only $710 in Korea in 1976 when the survey data for this study were collected. The results fully confirm the findings presented in recent U.S. articles concerning the direction of various biases which can affect the estimates of the demand parameters. They also show conclusively that both the income and the price elasticity of the demand for housing services in Korea are comparable to those found in the United States: the income elasticity of demand for housing is smaller than one and the price elasticity is negative and smaller than one in absolute value. The presentation of the results is organized as follows; First, Section II briefly describes the data base used for the survey and the characteristics of the housing market in Korea. Section III reviews ibriefly the state of research on the estimation of housing demand in the United States, and describes the method chosen for the specification of housing demand. 2 In order to avoid repetition of previous work the reader is referred to these four recent papers for a comprehensive review of the state of the art in the analysis of the demand for housing.

15 HOUSING IN KOREA 317 Section IV explains how the theoretical model has been implemented and describes in vurn the specification of the price per unit of housing services, household income, transportation costs, the sociodemographic determinants of housirng demand, and the alternative forms of the estimating equation. Section V presents the results of the analysis. Section VI draws the conclusirons of the study. II. THE KOREAN CONTEXT Korean housing standards are still extremely low in spite of the very high rate of increase in income over the last 15 years when the average growth rate of GNP has been around 10%1o per year. At the time of the 1975 census the ratio of urban households to housing units was 1.278, indicating that in cities 27.8% of the households cannot afford to rent separate dwelling units and can only find rooms to rent. This situation is due to the very high rate of urban growth 3 as well as the destruction of a significant proportion of the housing stock during the Korean War. Between 1955 and 1975 the population living in cities over 50,000 grew from 5.3 to 16.8 million. The analysis reported in this paper is based on the Family Income and Expenditure Survey regularly conducted by the Bureau of Statistics of the Economic Planning Board of the Korean government, which has been combined with a Special Housing Survey conducted in July The sample size is 1293 households with a reported monthily average income of 75,500 ($156) and an average of 4.9 household members. Forty-eight percent of these households own their dwellings. The sample average size of a dwelling is pyeong (613.5 square feet). These orders of magnitude were found consistent with the results of the Korean housing censuses. The sample survey shows that the median Korean household devotes about 15% of its income before tax to housing exclusive of such items as utilities, water charges, and home furnishings. The rent-to-income ratio by income group declines as income increases, as shown in Table 1. This decline is very marked and suggests that rental expenditures are not very responsive to income changes. For example, the ratio of the percentag'. change in rent between the 30,000- to 39,000-won income class and the 80,000- to 90,000-won class is only Yet it is impossible to conclude that demand is very inelastic with respect to income in the absence of 3 Over the period South Korea experienced the highest rate of urbanization in the world for countries with more than 15 million people in See Renaud [191, 4 The original questionnaire design was prepared by Raymond Struyk at the time of a review of the Korean housing policy for the fourth 5-year plan, A full description of the regular housing and expenditures surveys based on monthly diaries and of the special housing survey can be found in Renaud et al. [20].

16 318 FOLLAIN, LIM, AND RENAUD TABLE I Rent and Rent/Income Ratios by Income Class for Entire Sample' Income class (Won)b R/ Y R N Zero or not reported , , , ,000-49, ,000-59, ,000-69, ,000-79, ,000-99, , , , Average I (entire sample) Average II (excluding incomes less than 30,000) an number of sample households in the income class; R = monthly rental expenditure (zero or inputed); Y monthly current income. bthe 1976 foreign exchange rate was 484 Wons to one U.S. dollar. information on price elasticity and without controlling for other factors affecting the demand of a given household. 5 In most developed countries households can be classified into owneroccupants and renter-occupants with no further breakdown. In Korea, however, seven types of tenure exist: In addition to homeovwnership, the rental arrangements are in order of frequency: chonsei, security deposit with monthly payment, declining chonsei, pure rental, free housing, and government housing, These various rental arrangements are distinguished from one another by their payment schemes, terms of contract, and sometimes eligibility criteria. 6 In the two cases of declining chonsei and rent with security deposit, a household payment agreement has been converted 5 1t would not be correct to conclude from such a simple calculation that the demand for housing is extremely inelastic because other,factors influencing the demand of a given household are not being controlled. For instance, the rent-to-income ratio could decline as it does in Table 1 even if the true income elasticity were greater than one in a case where the price elasticity is less than one and prices vary in such a way that high-income households face a lower price per unit of housing services than low-income households. 6iA full analysis of the determinants of home ownership and reliance on chonsei arrangements is presented in a separate paper by [.im et al. [91. Home ownership in Korea is not widespread (48.3% in our sample) because of the limited availability and terms of mortgage money and absence of tax benefits. Borrowing from the extremely active unregulated money

17 HOUSING IN KOREA 319 TABLE 2 Rent and Rent/Income Ratios by Tenure for Entire Sample Tenure R/Y R N Owner , Free housing , Government housing ,222 9 Chonsei (0.130)0 8,164 (290) Security deposit , Declining Chonsei , Rent , 'The rent/income ratio for Chonsei is seriously distorted by the figure for the income group 1-29, 999, where the value of the ratio is 5.24! Elimination of observations relating to that group only yields a mean ratio for the rest of the population of into an imputed monthly rental charge. 7 In the other five cases no additional computation was required since households were asked to convert their rental arrangements into an imputed monthly rental charge. It may seem strange that owners would know precisely such a number. In Korea, however, it is quite believable because owners frequently sublet part of their homes so that they are quite aware of tlk- rental values of their homes. The estimates of imputed rent by occupants of government housing and free housing are more suspect, Therefore, we report results obtained with and without this relatively small part of the sample. Table 2 lists the seven tenure arrangements encountered in the sample as well as the average rent and the average rent to income ratio for each tenure arrangement. These imputed rental charges-either calculated or reported--are used as the measure of housing expenditures in the housing demand equations. markets is not possible because the rates are very high (between 3 and 5% monthly) and the terms of maturity very short (typically a maximum of 20 months). The chonsei is based on this capital scarcity: instead of a monthly rent the households give a deposit varying generally between 250,000 and 600,000 wons to the landowner, who places the funds on the money markets or uses them; the total deposit is returned in whole (in the case of a pure chonsei) or in part at the end of the lease. 1'he pure rent, Western style, is used only by the poorest households. n 7 For a declining <honsci the imputed rent is estimated as R 2 [(A -D t I)i + D]/1, where A = amount of original ch/onwei, D = monthly deduction from chonsei, n = period, i = curb market interest rate. For security deposits with monthly rent, R is equal to the monthly rent to which is added the deposit multiplied by the monthly curb rate. The unregulated money market monthly rate for the sample households is estimated by takisg the ratio of imputed rent to chonsei deposit in the sample chonsei data. The average value of i /month is very consistent with the regular monetary market surveys conducted by the Bank in Korea.

18 320 FOLLAIN, LIM, AND RENAUD III. THE CHOICE OF METHOD The model of demand estimated in this paper is one of a general family of models of housing markets in which the central units of measurement are units of "housing services" and "housing stock." Given some minimum amount of operating inputs, each unit of housing stock produces one unit of housing services per period. Housing services refer to the sum of all services, inclusive cf neighborhood attributes, provided by a housing unit during some period of time such as space, privacy, availability and dependability of utilities, and other features of the unit which provide comfort and pleasure. Given the price per unit of housing services (p), the rent of a dwelling unit measures the units of housing services (q) produced by a dwelling since rent equals p * q. Similarly the value of a dwelling unit equals the price per unit of housing stock (P) times the units of housing stock embodied in the dwelling unit (Q). The principal developers of this family of models are Mills [11] and Muth [14]. The most specific model which underlies the estimates reported is that developed by Muth [15, 16]. The details of the model are discussed below; here, an important reason for employing the model is noted. One of the most salient features of the model is that the price per unit of housing stock (and subsequently, housing services) can be constructed from the price per unit of land, the prices per unit of structure, and the shares of housing expenditures attributable to land and structure inputs. Since such data are available in Korea, a price term can be included in the demand for housing equations we estimate. As long as a unit of existing housing stock and an identical unit of new housing stock produce the same quantity of housing services, this measure serves as a measure of the price of both new and existing housing services. We feel this is a reasonable assumption which is commonly employed in studies of housing markets. Including the price term not only allows inferences to be made about the price elasticity of demand but also ensures that the estimates of the income elasticity are not biased because of the omission of a relevant variable (Polinsky [17], Strassheim [24]). An added advantage is that the procedure yields individual estimates of the price per unit of services for each individual household when past studies have frequently employed citywide indices. Polinsky [17] has shown that if demand is price inelastic, and prices decline with distance from the center of a city, then use of a citywide index rather than a more precise measure of the price faced by each household yields biased estimates of the income elasticity. The model employed has both a supply side and a demand side. The supply side is crucial to the construction of a price per unit of housing services. The key relation on the supply side is the production function for housing services: q = q*(q, 0) = q*[q(l, N), 01, (1)

19 HOUSING IN KOREA 321 where q measures the units of housing services per period; Q measures the units of housing stock; L measures the units of land inputs; N measures the units of structure inputs; 0 measures the units of operating inputs. We assume operating inputs are proportional to the units of housing stock, so that q is directly related to L and N. This new production function, q = q(l, N), is assumed to be homogenous of degree 1. Profit maximization behavior by suppliers of housing in a comp;titive market will satisfy the cond;tions PLIP = ql (2) PN/P = qn, (3) q = q(l, N) (4) where ql and qn are the marginal products of land and nonland, respectively. Muth [15] shows that the logarithmic differentials (e.g., q* = d log q) of (2), (3) and (4) can be arranged as - knl* + knn* + P* = GRL* (5) kll* - kln* + Up* = upn, (6) q*-kll*-knn* = O, (7) where a is the elasticity of substitution of land for structures in the production of housing. On the demand side, consumer responses to changes in p and y are derived from the standard utility maximization problem. Max U(q, X) subject to pq + X = Y, (8) where U(q, X) is the household utility function and X is the quantity of all goods other than housing (the price of X is the numeraire and equals l); Y is household income. Differentiating the first-order conditions for this problem yields the fourth and final condition for our system. q*= Efp* + (5) where cp and cy are elasticities of demand with respect to price and income, respectively.

20 322 FOLLAIN, LIM, AND RENAUD Because the survey data provide rental expenditures we can solve (5), (6), (7), and (9) for (pq)*, which yields the two key results: (pq)* = (1 + EP)(kLPL* + kvnp) + cyy (10) p * = klp + knp * ( 11) Equation (10) expresses changes in the demand for housing as a function of changes in the prices of land and nonland inputs and changes in income. Equation (11) simply says that any percentage change in the price per unit of housing services resulting from changes in the price of land or nonland inputs is a weighlted average of the percentage changes in PL and PN- The expression in the second set of parentheses in (10) equals p * by (11) and suggests the following two-step procedure for estimating e 1 p and cy First, calculate p* for each household using an equation based upon (II) from the survey data; second, substitute the calculated price p into Eq. (10) and estimate it using ordinary least squares (OLS). The problems associated with implementing this procedure are discussed in the next section. IV. IMPLEMENTATION OF THE MODEL Equations (10) and (11) raise five problems regarding the implementation and estimation of the demand for housing: 1. Choosing the appropriate approximation of (11) which permits the computation of the unobservable price per unit of housing services from PL, PN, kl, and k,. 2. Choosing the appropriate income concept and examining the sensitivity of the results to a variety of measures. 3. Adjusting for short-term variations in housing prices due to market disequilibria. 4. Adjusting income for transportation costs. 5. Identifying the principal sociodemographic determinants of the demand for housing. IK1. Calculation of the Price per Unit of Housing Services Equation (11) shows precisely how a given percentage change in the price per unit of land/or nonland inputs produces (in a static equilibrium sense) a percentage chanige in the price of housing. Unfortunately, Eq. (I 1) does not allow the calculation of the price of housing services because it is expressed in terms of percentage changes (for which we do not have data) and not of actual prices (for which we do have data). The absolute level of p from PL, PN, ki, and kn can be approximated with an equation based on

21 HOUSING IN KOREA 323 Eq. (I 1). The obvious first choice which is eventually used is lnp = kl I 1 PL + kn InPN. (12) This formulation is exact if the elasticity of substitution is unity, the value associated with the Cobb-Douglas production function. Although the Cobb-Douglas function is a frequently employed function, uncertainty concerning the value of a does exist. If anything, it is likely to be less than one (Muth [161). If it is less than one, our estimates of a are probably upward biased. This bias results because our measure of p would understate the true change in p given some change in the price of land. This suggests caution in the estimates of a if they are large in absolute value. As seen below, the estimates are significantly below unity, so the bias does not alter our ultimate conclusion that demand is price inelastic. Although we do not have the exactpl orpn faced by each household, an excellent estimate is available because the Korean government keeps systematic records of land prices and publishes every year three land prices for each neighborhood or Dong. 8 The chosen measure of PL faced by a household is the average of the high, middle, and low prices of land per pyeong in the Dong in which the household resides. The housing construction cost per pyeong is derived from the construction cost records of the Korea National Housing Corporation. The data are based upon the nonlandl costs of constructing a particular type of apartment unit in eleven cities. 9 For the five cities in the sample not represented in these unpublished data, we assigned the cost figures provided for the nearest city. Because of the close geographic proximity of these cities this is probably not a serious error, The theoretical justification for this a;)proach is that competitive forces eventually eliminate significant differences in construction costs between two nearby areas as long as transportation costs and localized market powers are not too great. This reasoning is confirmed in the analyses where estimates computed with and without these five cities are very similar. The factor shares, kl and kn, are calculated as kl = lpll/r; kn = I - kl, (13) 8 A Dong is an administrative subdivision of Korean cities. In Seoul, the number of households in one Dong varied between 1000 and 8000 households in 1975; a typical figure would be 3000 for Seoul. The official Korea Appraisal Board has been maintaining records of land prices by Dong since at least 1962 in most cities. This rather remarkable monitoring system is used in various ways in the implementation of land policies. See Doebele and Hwang [5] for an up-to-date review of Korean urban land policies, and Mills and Song [121 for an analysis of these data. 9 These cities are Seoul, Busan, Incheon, Chuncheon, Daejon, Jeonju, Daegu, Masan, Chungju, Weonju, and Suwon.

22 324 FOLLAIN, LIM, AND RENAUD where i is the interest rate used to convert the land and nonland prices from prices per unit of stock to rentals per unit of stock per period. i is set equal to /month. Selection of this value is based upon two considerations. One, the rental values of the apartments built by the Korean Construction Company are between and of the total cost of construction per unit. This should be adjusted downward since the construction data do not include land costs. Two, kl typically equals about 0.33 for Korean homes. 10 From this value, the rental value of land can also be calculated since the definition of kl implies i = klr/lipl. Inserting the averages for PL, L, kl, and R of 63729, 12.9, 0.32, and 8251, respectively, into the equation yields a value for i slightly over This number may appear low in comparison to the 3.26% per month interest rate which can be earned in Korea's curb market. However, a rent-to-value ratio equal to does not mean landlords earn only 0.28% per month; it is perfectly consistent with the fact that a large part of Korean landlords' profits comes in the form of capital gains on the real estate which they own. Consider the following demonstration of this point. The rate of return can be expressed as NI R+ra-X / >, -,(14) where This is equal to NI = net income per month, V = value of assets, R = rent per month, ra = rate of appreciation of V per month, X = operating costs, property taxes and depreciation. NI R-X + ra = i + CG, where CG is the rate of return attributable to capital gains. If i = 0.003, then the landlord needs capital gains of about 2% per month in order to earn a return competitive with that in the curb market. (One would probably expect the return earned outside of the curb market to be less 10 In a survey of recent housing units the share of land in total cost was found to vary between 40%o for single-unit houses and 24% for apartments. See "Survey Report on Financing Status of Private Housing Funds," Korea Industrial Development Research Institute " 1976).

23 HOUSING IN KOREA 325 than the curb rate since the curb market is likely to be more risky.) It is not uncommon for land values to increase by 2% per month; for instance, land values averaged about 3.2% increase per month in Suwon during the period 1968 to IV2. Measurement of Income The choice of the appropriate measure of household income has been a source of considerable difficulty in practically all housing demand estimations. Many argue that for the estimation of the long-run income elasticity the appropriate measure is probably not the current rate of income which is easily measured, but rather the long,run expected income (the permanent income) which is not directly measurable. Choosing the appropriate variable is difficult but important because estimates of the income elasticity obtained using current period incomes are downward biased if the permanent income hypothesis is true (see Theil [25]). Given the nature of the survey data, two alternatives for r.nasurement of the permanent income are used in addition to the monthly income reported in the survey. First, total household consumption expenditures can be substituted for the measure of household disposable current income. The reasoning is that the level of consumption is likely to be a good proxy for permanent income if, according to the permanent income theory, consumption is proportional to permanent income [6]. We use the log linear specification so the coefficient Of C0nsLuniption is the estimate of E,. An alternative which has often been proposed is to estimate the parameters of the demand equation using grouped household data av the unit of observation, e.g., SMSA. We choose to aggregate by Dong. This aggregation procedure is used under the assumption that the nonpermancnt or transitory components of current income will cancel out when households are grouped bv factors unrelated to transitory income. This aggregation imparts an upward bias to the income elasticity estimates if rental expenditures are tightly grouped and the distribution of rental expenditures within Dongs is significantly tighter than their distribution throughout the sample. We have little information about the severity of this problem except that Dongs are thought to be significantly more heterogeneous than their counterparts (e.g., census tracts) in the United States. If the grouping produces estimates of the income elasticity significantly below unity, then knowledge of the upward bias will only strengthen the findiing that demand is inelastic with respect to income. IV.3. Adjustment for Shiort-run,.xcess Detmantid Variations in the currenit mnarket price of housing scrvice.h are linked not only to variations in its long-run deternmiiianiis buit also to short-run fluctuations caused by excess demand. Since our goal is to estimate the

24 326 FOLLAIN, LIM, AND RENAUD elasticity of demand with respect to the long-run price of housing, we introduce an additional variable into the demand equation which controls for short-run variations in the price of housing associated with excess demand. The variable (S) is a measure of the housing shortage in a particular city and equals the ratio of the number of households minus the number of housing uinits to the number of housing units in a particular city based on the 1975 census. We expect the sign of S to be positive, when statistically significant. In addition, if the price elasticity cp is estimated with and withoout S5 the estimates of e. when S is included are expected to be algebraically smaller. The reasoni would be that failure to control for short-run variations causes variation in the long-run price of housing to be overstated, and consequently, estiniates of Er, to be biased downward. This procedure is analogous to efforts which include the vacancy rate as an explanatory variable in equations explaining the supply of housing (see de Leeuw and Ekanem [4]). Estimates are computed with and without S. 'rhe differences in the estimates of fp and Ey with and without S are small bult consistent with our interpretation. IV.4. Adjustment of Income for Tranisportation Expenditures Muth, Mills, and others argue that the appropriate inicome concept in a demand equation for housing is income minus transportation costs whei ever the margitnal costs of transportation witlh respect to distance are positive- -an eminnently reasonable assumption (see Muth [13]). Precise measurements of the costs of transportztio'ii expenses (e.g., gasoline, bus fare, etc.) and the opportunity costs of travel time are required. Since we were unable to find a satisfactory Korean study of the value of travel time, we adopted the simplified but probably reasonable assumption that transportation costs are proport:ronal to travel time from the central city for which data were collected. IV.5. Sociodemographic Determinants of the Denmand for Housing Some variables which were considered for inclusion in the demand equation included aiverage age of household members, number and ages of children in the household, number of subfamilies in the household, sex and occupation of the household head, and number of people in the household. Household size was finally selected for inclusion in the complete demand equation because it performs most consistently. Since household size is often correlated with the other determinants conisidered, we suggest that household size be interpreted as a proxy for some of the sociodemographic variables which undoubtedly influence the demand for housing but whose effects are difficult to sort out precisely. Past research shows that the introductioni of sociodemographic variables often leads to somewhat smaller income elasticities [8, 18],

25 HOUSING IN KOREA 327 V. THE EMPIRICAL RESULTS Essentially two types of equations are estimated. The first one uses the household as the unit of observation and includes either one of the two income measures. The equation is: In R = a 0 + a, ln Yi + a 2 In P + a 3 S + a 4 In HS + a, In T where R = imputed or explicit monthly rental expenditures, Y, = current disposable monthly income, Y2=total monthly consunmption expenditures, S the housing shortage variable (see above), HS = the number of people in a household, T = travel time to central city. (i = 1, 2), (15) The second type of equation uses averages of the variables by Dong as the unit of observation: ln R = ao + al In Y± + a 2 ln P +a 3 S+ a4 HS Fa + 5 T, (16) where bars above the variables indicate that they have been averaged over all sample households in a particular Dong. For example, one observation would be R, =, ' -) 11Li where L is the inumber of households in the sample residing in the ith Dong, and Ry is the rental expenditure of the jth household in the ith Dong. The household income and expenditures survey on which the housing survey is based covered households located in 17 cities so that various groupings of observations could be made on the basis of city size as well as more common characteristics such as homeowners versus renters or highincome versus low-income households. A large number of equations were estimated but only the most important results are reported here. Two types of nationwide samples are reported as groups I and 2. Group I consists of all households in the original sample except (a) those in Jeju, an island off the Southerm Coast; (b) those with reported zero income; and (c) those with a calculated land share kl not between zero and one." 1 Only demand "The city of Jeju has been deleted because it is on an island 100 km off the Southern Coast of Korea for which transportation costs may be substantial in the case of bwlding materials, Unfortunately, this is one of the cities for which local building costs data were unavailable. In the case of households with reported income of zero and/or calculated value of the land share outside the interval zero-one, deletion was necessary because of measurement errors.

26 328 FOLLAIN, LIM, AND RENAUD equations based on individual observations are reported for group 1. The second type of sample, labeled group 2, is a sample slightly reduced in size by withdrawing from the group I sample the households presumed to be benefiting from government housing subsidies as well as households residing in cities without construction cost data. The basic results for the two national samples are reported in Tables 3, 4, and 5. In addition demand estimations based on group 2 were estimated on interesting subsamples such as owners versus renters, households residing in Seoul (population, 7.5 million), households residing in Busan and Taegu (two other cities of over I million people), and households residing in cities other than the three largest. By estimating demand functions for such subsamples we can test for statistically significant differences in housing demand over groups which are of major significance to policy formulation. The results for these subsamples are presented in Tables 6, 7 and 8. Their discussion follows that of the analyses based on the two national samples (groups 1 and 2). V.l. Demand Results Based on the Nationw4ide Samples (Grotps I and 2) The discussion of the natiouial results is divided into five parts. The first four parts deal with tile estimates of the elasticity of demand with respect to (a) income; (b) price, (c) household size, and (d) transportation expenditures. They are followed by an overall assessment of the demand equation findings. TABLE 3 Estimates of Demand la for Group lb. Dependent variable Constant Y, Y2 P S HS T R 2 R Note. N = Standard errors are below the estimates.,'demand equation I has the household as the unit of observation. bgroup I includes all observations in the sample except those with zero reported incomes, calculated kls not between zero and one, as well as households residing in Jeju. 'Yi = current disposable income; Y2 = consumption expenditures; P = price per unit of housing services; S =housing shortage ratio of number of households minus number of dwellings to number of dwellings; HS = household size; T = travel time to center of city; cp = price elasticity equal to regression estimate minus one.

27 HOUSING IN KOREA 329 TABLE 4 Estimates of Demand la for Group 2 b c Depeiident variable Constant Y 1 Y2 P S HS T R 2 e , Note. N 896. Standard errors are below the estimates. "Demand equation I has the household as the unit of observation. bgroup 2 includes all observation in group I minus those in cities without construction data and those residing in government housing. 'See footnote c, Table 3, for definitions of parameters. Dependent TABLE 5 Erstimates of Demand Equation II" for Group 1 6, variable Constant Y 1 P S HS T R 2 fp I ,10 (0.17) (0.16) (0.17) (0.17) (0.07) (0.37) (0.07) Note. Standard errors are below the estimates. 'Demand equation II uses averages by Dong as the unit of observation. bc;roup I includes all observations in tthe sample except those with zero reported incomes, calculated k 1 values not between zero and one, and those residing in Jeju. csee footnote c, Table 3, for definitions of parameters. (a) Income elasticity estimates (EY). The estimates of the income elasticity equal the estimates of the coefficients of the income term since all variables are in logarithmic form. The estimates of E, obtained using national data are given in Tables 3, 4, and 5. Three major aspects of the results should be emphasized. First, the -tiniates of cy are less than one for the three measures of income used: tl.,- estimates of Ey using household data and Y,, the reported income in the month of July 1976, range from 0.16 to 0.20 (lines 3 and 4 of T'ables 3 and 4). The estimates of Ey obtained using YT (grouped disposable income) both equal Second, the estimates are all highly significant in the sense that the ratio of the estimate of its standard error is always greater than 8 with household data and about 3.4 with grouped data. Third, the estimates vary with the income measure

28 330 FOLLAIN, LIM, AND RENAUD Dependent R TABLE 6 Estimates of Demand Equation ia for Spcial Groups: Owners vs Rentersb variable Constant Y 1 Y2 P S HS T R 2 cp 1. Owners (a) , (0.04) (0. 1) (0.01) (0.08) (0.04) (b) (0.04) (0.08) (0.008) (0.07) (0.04) 2. Renters (a) (0.03) (0.09) (0.007) (0.06) (0.03) (b) (0.04) (0.09) (0.007) (0.06) (0.03) Note. N (owners); N = 481 (renters). Standard errors are below the estimates. ademand equation I has the household as the unit of observation. bsee footnote c, Table 3, for definitions of parameters. R Dependent TABLE 7 Estimates of Demand Equation Pa for Special Groups: Low vs High Incomeb variable Constant Y 1 Y2 P S HS T R 2 e 1. Low income (a) , (b) High income (a) (b) Note. N (low income); N = 415 (high income). Standard errors are below the estimates. 'Demand equation I has the household as the unit of observation. blow-income households are those below the mean. High-income households are those above the mean. 'See footnote c, Table 3, for definitions of parameters.

29 HOUSING IN KOREA 331 TABLE 8 Estimates of Demand Equation la for Special Groups: Seoul; Busan and Daegu; All Other Cities in the Samplesb R Dependent variable Constant Y 1 Y2 P S HS 7' R 2 ep 1. Seoul (a) (0.04) (0. I1) (0.06) (0.05) (b) (0.04) (0.10) (0.06) (0.05) 2. Busan and Daegu (a) (0.04) (0.18) (0.02) (0.09) (0.05) (b) , (0.05) (0.07) (0.02) (0.07) (0.05) 3. All cities except Seoul, Busan, and Daegu (a) (0.05) (0.12) (0.01) (0.09) (0.07) (b) (0.06) (0.12) (0.01) (0.0)9) (0.07) Note. N (Seoul): N = 264 (Busan and D)acgu);.V = 216 (all other cities except above). 'Demand equation I has the household as the tunit of observation. 1 See footnote c, Table 3, for definitions of parameters. used. The estimates of EY using Y 2 and Y 1 our two proxies for permanent income, are about three times larger than the estimates of Ey using Y. The findings of this analysis of Korean housing demand are remarkably similar to those of several recent studies of the income elasticity of the demand for housing in the United States (Carliner [2], Roistacher [21], Mayo [10], Polinsky [17], Smith and Campbell [23], and Lee and Kong [8]. The behavior of the alternative income elasticity estimates obtained when the three different income variables are used is quite consistent with the theoretical expectation that estimates of Ey using proxies for permanent income are larger than those using current income n fact the ratio of one to three found for Korea is much larger than the U.S. and U.K. findings, where "on tha average, the permanent income elasticity estimates of these studies are 50% higher than the current income elasticity estimates," as summed by Polinsky 117]. One likely reason for the wider gap is that the Korean current income figure which had to be used is the income figure for the month of July 1976 eported in the diaries collected in the regular income and expenditures s,urvey. A monthly income averaged over a 12-month period would have been much better but could not be estimated. The low elasticity based on current income may be attributed to this problem.

30 332 FOLLAIN, LIM, AND RENAUD (b) Estimates for the price elasticity (ep). The estimates obtained indicate that the Korean housing demand for housing is price inelastic. For the nationwide samples (Tables 3, 4, and 5) the estimates range from to ' 3 The two cases of positive elasticity estimates arise when grouped data are used, and in these two cases the standard errors are large enough to find the two estimates not significantly greater than zero. In his 1977 paper Polinsky [17, p. 5] also finds that the use of grouped prices will bias the price elasticity toward zero. It must also be noted that, in the rest of the equations estimated, the price elasticities have larger absolute values when the better proxy for permanent income is used. On the basis of the estimates obtained using nationwide samples of individual households, the price elasticity of the Korean housing demand is within the range to This is consistent with the sketchy evidence already presented on the value of EP for Korean households [12], and the fact that the estimates are in agreement with the most precise study to date of the value of fp-the Experimental Housing Allowance Program [10]. (c) Estimates of household size elasticity (EHS). The coefficient of HS is the elasticity of rental expenditures with respect to household size. This number, EHS, is estimated to be 0.36 and 0.14 when the income measures are Y 1 and Y 2, respectivelv, and the household is the unit of observation. The estimates are at least three times the value of their standard error. When the average by Dong is the unit of measurement, EHS is not significantly different from zero. This is probably due to the fact that average household size does not vary much by Dong. The estimates of the coefficient of HS are consistent with the hypothesis that the demand for housing is relatively insensitive to the size of the household. This suggests that the urban dernand for housing will not be much affected by future charges in average family size. (d) Elasticity wvith respect to tranisportation expenditures (ET). Estimates of the coefficient of T, the travel time to the center of the city, indicate that the elasticity of the demand for housing with respect to T is quite small: et is less than The precision of the estimates is very difficult to assess. On the one hand, most of them have the correct sign. On the other hand one would expect their magnitude to be closer in absolute value to the coefficient of income. The fact that the estimates of Ey are typically four or more times larger than the estimates of ET suggests that the proxy for transportation expenditure is picking up only part of the households' transportation expenditures, that is, the out-of-pocket travel expenses which are directly a function of travel time. The other component of transportation expenditures, the opportunity cost of travel time, is a function of income, not " 3 When reading the regression results one must keep in mind that the coefficient of the price variable p is equal to I + ep.

31 HOUSING IN KOREA 333 distance. So, if more precise estimates of ft are to be had, Korean estimates of the value of travel time must be made first. (e) Assessment of the overall equation. The demand equations using the national groups are very good. This assertion is supported by three kinds of evidence. First, the signs of the estimated coefficients are as expected, and the estimates are almost always significant. Second, the explanatory power of the estimated equations is reasonable for equations estimated using a cross section of households as a data base. The values of R 2 for the full equation range from 0.25 to 0.48 with the household as the unit of observation. R 2 reaches a high of 0.56 when the average by Dong is the unit of observation and Y 2 is the income measure. Third, the explanatory power and overall pattern of results are fully consistent with and quite comparable to those obtained in other studies of the demand for housing which used household data. V.2. Analysis of Special Subsamples Additional demand equations have been estimated for groups of households which may be of interest to policymakers. The comparison of the results across groups also sheds light on the assumption that the parameters are constant across groups as implicitly assumed in the national group estimates. (a) Owners vs renters. Table 6 presents estimates of the household demand for owners and renters. The results indicate that the income elasticiky is larger for owners than for renters: f, is estimated to be 0.62 (0.21) for owners when Y2 (Y 1 ) is the income measure while E, is estimated to be only 0.42 (0.12) for renters. The differences in the estimates are statistically significant, and are consistent with those reported by OeeLeeuw for owners vs renters in the tunited States [4]. A comparison of the estimates of the other coefficients suggests that the behavior of the two groups is similar in some respects but not in others. The price elasticity estimates are quite similar for both groups. The renters are slightly more responsive to household size, but less responsive to transportation expenditures. Finally, consider the estimates of E,, for renters and owners with the national estimates of group 2. C(omparable national estimates, of Ey are 0.17 and 0.58, Both are about midway between the estimates of Ey for owners and renters. What is more important to note is that the results obtained using these subgroups; are quite consistent with the contention that E, is less thanl. unity. (b) Low vs high-tncome households (TFable 7). National estimates of fy and Ep are important for some government policy considerations, but at other times policy development is aimed at the lower-income households. By separating group 2 by income at the mean sample income more precise

32 334 FOLLAIN, LIM, AND RENAUD estimates of fy and Ep for low-income households may be obtained. The estimates indicate that low-income households are much less responsive to changes in current income than high-income households (Y,), but not significantly different in their response to changes in c-onsumption expenditures, the proxy for permanent income (Y 2 ). Since the estimates of sy are biased downward when current income is used as an income measure, the estimate of 0.02 for fy obtained with Y 1 is probably much too low. The alternative estimates of fy with consumption expenditures as the permanent income proxy are probably a better guide to the responses of low- and high-income households. The conclusion is that Korean low-income households behave essentially in the same way as the higher-income households. Price elasticity estimates are also quite similar. In fact, the estimates of fp with Y 2 as the income measure are identical (-0.27). The similarity continues upon comparison of the estimates of EHS and ET. On the whole then, the fact that the estimates for the two subgroups are ftot significantly different from each other and are not much different from the national estimates suggests that little is lost by pooling households with different incomes. (c) Grouping by size of cities. Three subgroups of group 2 have also been analyzed separately. They are Seoul by itself, Busan and Daegu, and the other remaining cities in the sample. Seoul is the largest city in Korea, (over 7 million residents), Busan and Daegu are the next two largest (over 1 million), and the third group of cities range from about 100,000 to slightly over a million. Estimating separate demand equations for these three subgroups provides some insight into the potential significance of city size for the demand for housing. Statistically significant differences arise among the estimates presented in Table 8. The income elasticity estimates (with Y 2 as the income measure) is largest in Busan-Daegu (0.68) and smallest in the group of smaller cities (0.37). The differences do not conform, however, to a simple positive relationship with city size, nor do they revterse the earlier claim that demand is income inelastic. The most significant differences concern the estimates of EP. The estimates for Seoul (-0.12 and -0.17) are quite consistent with the national estimatcs. Estimates of ep for Busan-Daegu and the smaller cities are, however, three to four times smaller (algebraically). VI. CONCLUSIONS Given the significant number of countries found within the income range between Korea's income at the time of the survey ($710 in 1976) and the U.S. figure ($7870) the finding that these two countries have strikingly comparable demand elasticities is of major importance. Lacking good local

33 HOUSING IN KOREA 335 estimates of the price and income elasticities of the demand for housing policy analyses in other countries, one should rely on income elasticity estimates smaller than one and on price elasticity estimates also smaller than unity in absolute value. The best point estimate is 0.6 given that the estimates obtained with the two proxies for permanent income are the most precise. While further refinements of the findings presented in this paper are desirable, these first results give reason for optimism in that available econometric models developed for the analysis of cities in the United States could perform well in Korea and other countries. REFERENCES 1. J. Campbell and B. Smith, The demand for housing: A new look at urban empiricism, Paper presented at the Western Economic Association Meeting (1976). 2. G. Carliner, Income elasticity of housing demand, Ret. Econ. Statist., 55, (1973). 3. F. DeLeeuw, The demand for housing: A review of cross-section evidence, Amer. Econ. Re'. (February 1971). 4. F. DeLeeuw and N. F. Ekanem, The supply of rental housing, Amer. Econ. Rev., 61, (1971). 5. W. Doebele and M. C. Hwang, Land policies in Korea: With special reference to land development, mimeo (1977). 6. M. Friedman, "A Theorv of the Consumption 'unction," Princeton Univ. Press. Princeton N.J. (1957). 7. T. Hi. Lee, Housing and permanent income: 'Tests based on a three-year reinterview survey, Rev. Econ. Statist., (November 1968). 8. T. H. Lee and C. M. Kong, Elasticities of housing demand, Southern Econ. J., 44, (1977). 9. G. Lim, J. Follain, and B. Renaud, The determinants of home ownership and rental arrangements in Korea, manuscript. 10. S. Mayo, "Housing Allowance Demand Experiment, Housing Expenditures and Quality," Part 1, "Report on Housing Expenditures under a Percent of Rent Allowance," Chap. 4, Abt Assoc., Boston (January 1977). 11. E. S. Mills, An aggregative model of resource allocation in a metropolitan area, Amer. Fcon. Ret'., (May 1967). 12. E. S. Mills and B. N. Song, "Korea's Urbanization and Urban Problems ," K.D.I. Working Paper No. 7701, Korea Development Institute, Seoul (1977). 13. R. Muth, "Cities and Housing," Univ. of Chicago Press, Chicago (1969). 14. R. Muth, Demand for non-farm housing, in "The Demand for Durable Goods" (A. Harberger, Ed.), pp , University of Chicago Press, Chicago (1960). 15. R. Muth, The derived demand for a productive factor and the industry supply curve, Oxford Econ. Pap. (July 1964). 16. R. Muth, Derived demand for urban residential land, Urban Studies, (1971). 17. A. M. Polinsky, Demand for housing: A study in specification and grouping, Econometrica, 45, (1977). 18. A. M. Polinsky and D. T. Ellwood, "An Empirical Reconciliation of Micro and Grouped Estimates of the Demand for Housing." Discussion Paper 567, Harvard Institute of Economic Research (August 1977). 19. B. Renaud, "Economic Fluictuations and Speed of Urbanization: A Case Study of Korea, ," World Bank Staff Working Paper No. 270 (November 1977). 20. B. Renaud, G. Lim, and J. Follain, "Housing in Korea," to appear. 21. A. Roistacher, Short-run housing responses to changes in income, Amer. Econ. Rev. Pap.

34 336 FOLLAIN, LIM, AND RENAUD Proc., (February 1977). 22. P. Rydell, Measuring the supply response to housing allowance, Rand Pqp. Ser. (January 1976). 23. B. Smith and J. M. Campbell, Jr., Aggregation bias and the demand for housing, Paper presented at the Econometric Society (December 1976). 24. M. Strassheim, Estimation of the demand for urban housing services from household interview data, Rev. Econ. Statist., 55, 1-8 (1973). 25. H. Theil, "Principles of Econometrics" Wiley, New York (1971).

35 Urban Studies (1980) 17, / S02.00 ( 1980 Urban Studies Determinants of Home-ownership in a Developing Economy: the Case of Korea Gill-Chin Lim, James Follain and Bertrand Renaud [Received May 1979] It is often suggested that increasing the rate of home- considerable controversy, since incorrect measureownership is one of the major goals of national hous- ments of income are likely to result in biases in ing policy. The rationale is that greater home- estimated coefficients. For this reason, more recent ownership would improve the general level of housing research has attempted to test the permanent income quality and facilitate savings and wealth accumula- hypothesis using several alternative specifications tion of households. However, despite a pheno- (Kain and Quigley, 1972; Carliner, 1974; Struyk, menal expansion of literature on housing market 1974). None the less, there have been few attempts behavior in recent years, studies on the determinants to evaluate the relative importance of current and of home-ownership have received relatively little permanent income in a systematic way. scholarly attention in both developed and develop- These studies have also suffered from a lack of an ing countries. On the one hand, in advanced explicit consideration of the impact of housing countries the amount of research on home-ownership market conditions on the rate of home-ownership. falls far behind that on demanid for housing services, In an effort to evaluate the effect of marlcet conditions and much work is needed to refine the existing on tenure decisions, Struyk (1976) has estimated a analytical framework. On the other hand, in de- model of an aggregate demand for owner-occupancy veloping countries studies on household tenure using data from 39 cities. Alterniatively, Carliner decisions are virtually nion-existent. (1974) has used dummy variables categorising size Previous studies on home-ownership conducted in of communities in an attempt to explain the effect of the United States have usually focused on the estima- regional differences on home-ownership. None of tion of the effect of household socio-economic char- these studies, however, have measured the effect of acteristics such as income, race, age, sex, and marital the long-run housing price on individuals' tenure status of household head and househiold size on decisions in an explicit manner. tenure decisions (Lee, 1963; Maisel, 1966; Kain and Another aspect ofteni omitted from home-owner- Quigley, 1972; Carliner, 1974; Struyk, 1974). Among ship studies is the discussion of the role of household these household socio-economic variables, income mobility. Despitethesalientimportanceofmobility raises one of the most difficult problems in estima- in teniure decisions, only Kain and Quigley (1972), tion. Most of the earlier studies uised current income amonig the studies llmenltioned above, have offered an as a measurement for household income. However, analysis of the relationship between home purchase the specification of the income variable inivolves a and household mobility. Gill-Chin Limn is Assistant Professor, the Technological Intstitute, Northihvsicrin University; Janmes Follaitz is economnist, the Federal Home Loan Banik of San Franzcisco; and Bertrand Rentaud is Ifead, Urban Affairs Divisioln, OECD. Authior's Note: The research reported in this paper is based on a larger study of the Korean housing market organised by the World Bank unler the overall responsibility of Bertrand Renaud. However, responsibilities have varied for each section of the analysis and the names of the authors are listed accordingly in the resulting paper. Primary credit is duc to Gill-Chiin LIim in this paper. Gill-Clhin Lim is grateful to Linda Lawrence for her research assistance. This study may not be quoted as representing the views of the World Bank and its affiliated organisations, 13

36 14 GILL-CHIN LIM, JAMES FOLLAIN AND BERTRAND RENAUD The purpose of this paper is to present an em- ownership and its estinmation procedures. The fourth pirical analysis of the determinants of home-owner- section reports on the results of the empirical ship in a developing economy by employing a analysis. The paper concludes with a summary of methodological framework which overcomes some the results and discusses the implications of the of the deficiencies of existing studies discussed above. analysis. A substantial inmprovement in empirical estimation is made possible by the detailed housing survey and Home-ownership and Alternative Forms of Rental land value information available for Korea. More specifically, the present analysis attempts to con- g i o tribute to the study of household tenure choice in The analysis reported in this paper is based on the several ways. First of all, this is the first serious Family Income and Expenditure Survey regularly analysis of household tenure decision behavior in conducted by the Bureau of Statistics of Economic developing countries. The study provides a basis Planning Board of the Korean Government and a for making observations on major similarities and Special Housing Survey conducted in July differences in household tenure choice between The sample size is 1293 households with a reported developed and developing countries. It can also monthly average income of 75,500 Won (US $150) serve as an important and useful guideline for hous- and an average of 4.9 household members. ing policy-making in developing economies. Second, Perhaps Korea is the country with one of the most the study offers an analysis of the relative importance sophisticated systems of rental housing market in of current income and permanent income. Third, an terms of the availability of options open to renters. explicit consideration of the effect of market condi- In most countries households can be classified into tions on individual decision units is provided by owner-occupants and renter-occupants and no further employing a long-run price of housing services for an breakdown is necessary. In the Korean Housing individual household and an indicator of housing Survey, however, six alternative forms of rental arshortage. Finally, the study includes a measure of rangementsareidentified. Theyareclhonsei,security household mobility to analyse its relationship to deposit with monthly payment, declining chonsei, tenure choice. free housing, government housing, and monthly rent. The organisation of this paper is as follows. In These different types of rental arrangements are disthe following section a brief description of home- tinguished from one another by their payment ownership and various types of rental arrangements schemes, terms of contracts, and sometimes eligibility in Korea is presented. The third section explains criteria. Selected features of owner households and the specification of a model of the demand for home- six types of renter households are reported in Table 1. Table 1 Selected Features of owner households and six types of renter houtseholds Months in Samiiple Household Male-headed Number current No. (") Size Age of lhead household working dwelling Owner 624 (48.3) Chonsei 357 (27.6) Security deposit with monthly rent 114 (8.8) Declining chonsei 70 (5,3) Free housing 13 (1.0) Government housing (free) 9 (0.7) Monthly rent 96 (7.4) Unclassified 10 (0.8) Total 1293 (100) The original questionaire design was prepared by Raymond Struyk at the time of a review of thle Korean housing policy for the fourth five-year plan For details of the income and expenditure survey and the housing survey, see Renaud, Lim and Follain (forthcoming).

37 DETERMINANTS OF HOME-OWNERSHIP IN KOREA 15 Traditionally, home-ownership has been the most is automatically deducted from the deposit. Thereimportant form of tenure in Korea. However, the fore, the actual rent paid by a renter is the imputed ratio of urban home-ownership has continued to monthly value of the initial deposit plus the monthly dwindle in recent years. In 1960, the rate of home- payment. ownership in urban areas was 62.0 per cent and in Declining chonsei, which constitutes 5.4 per cent of 1970 it dropped to 48.4 per cent (Economic Planning all households, is another variant of chonsei. This Board, 1960, 1970). In our survey about 48.3 per type of renter makes a one-time deposit from which cent of the households are home-owners. The rapid a certain anmount of due is deducted every month. rate of urbanisation, an enormous number of Any fund remaining at the end of occupancy is refugees from the north, massive destruction of the returned to the renter. Therefore, the actual amount housing stock during the Korean war, and rising of monithly rent equals the monthly deduction plus costs of land and capital inputs are often considered the imputed monthly value of the balance of the to be responsible for decreasing urban home-owner- deposit. ship. In addition, it should be noted that two of the A very small fraction of rental housing (in our major institutional factors which have encouraged survey I per cent) is offered free. Usually, free prohome-ownership in the United States and other vision of housing is practised between close relatives developed countries do not exist in Korea. Presently, or friends. Sometimes private companies also offer few mortgages are available for households through free housing to their eniployees. In some other an official financial sector and the national govern- instances, a renter is required to offer some form of ment does not offer any tax concessions to home- service to the landlord in lieu of monetary payments. owners. The Korean government created the Both local and national governments in Korea Korean Housing Bank 10 years ago to finance hous- own a small number of free or rental housing units ing construction, but the amount of mortgage funds for government officials. In our survey only 9 out from the Bank is very meagre. Although it is possible of 1293 households are reported as free government for households to borrow from the so-called un- housing occupants. Each governmental unit managregulated money market, the interest rates are un- ing these houses has a set of eligibility criteria. favourably high and the term of maturity is limited to a very short period, usually ranging between 1 and 2 years. The standard form of housing market transaction is, therefore, the full payment of the The theoretical basis of the model of tenure choice price of a dwelling unit at the time of purchase. is essentially the same as that of the demand for Chonsei is by far the most important type of rental housing services. Therefore, a general model of arrangement in Korea. The survey shows that tenure choice takes the probability of home-ownerchonsei renters account for 27.6 per cent of all house- ship as a function of household income, size, comnholds and 54.2 per cent of renters. Under this position, stage of life cycle, and the price of housing arrangement, a renter makes a lump sum deposit of services. Since we hypothesise that the tenure chonsei (usually translated as 'key' money) at the decision is also affected by short-run conditions of beginning of occupancy which is fully refunded at the the housing market and the mobility of households, end of the contract period. With this fund the land- we add two n-tore explanatory variables. Thus lord usually makes an investment of his choice. symbolically, Therefore, the imputed rent of chonsei is roughly PO-f (Y HC P S M) equivalent to the interest on the initial deposit. P f ( H Security deposit with monthly rent covers 8.8 per where P0 is the probability of home-ownership, Y cent of all households in the survey. This type of is some measure of income, HC is household charrental arrangement is a mixture of chonsei and acteristics, P is a measurement of the price per unit monthly rent. At the beginning of occupancy a of housing services, S is an indicator of housing renter makes a certain amount of deposit which is shortage in a city, and M is a measure of mobility. refunded when he moves out. But a renter is also Obtaining a reliable measurement of permnanent responsible for paying a monthly due. In case a income is crucial to the study, because one of our renter is unable to pay the monthly rent, the amount primary concerns is the evaluation of the relative US 17/1--B

38 16 GILL-CHIN LIM, JAMES FOLLAIN AND BERTRAND RENAUD strengths of current income and permanent income. The price per unit of housing services (P) is con- For current income we simply calculated disposable structed from a neoclassical framework which asincome by subtracting monthly tax payment from sumes profit maximising suppliers and utility maxireported monthly income. For permanent income, misilng consumers. 3 The price variable is intended we initially developed two measurements. One to reflect long-run market conditions. is a predicted permanent income using a multiple Housing markets are affected not only by the regression model and the other is consumption as a long-run determinants but also by short-run condiproxy for the permanent income. In our preliminary tions such as excess demand. To control for the experimentation, the latter proved to be a much effect of short-run variations in housing markets, we better measurement than the former. While the include an indicator of housing shortage (S). The former showed a very low level of significance and, housing shortage variable is obtained by dividing the frequently, incorrect signs, the latter was consistently Table 2 significant and always showed expected signs. 2 In this study we report only the results of the Variableslist estimation using disposable current income (Y,) and Dependent variable consumption as permanent income (Y 2 ). PO: Home-ownership dummy = 1 if the household is Several variables are selected to describe house- an owner hold characteristics; household size (HS), age of the Independent variables head of a household (AGEH), male-headed house- Y 1 : Monthly current disposable income in logarithm hold dummy (MALE), the number offamily members Y 2 : Monthly consumption in logarithm working (NWORK), the number of members under AGEH: Age of head in logarithm 6 years of age (N6), the number of members under 20 MALE: Sex of head dummy = 1 if male years of age (N20), and an elderly subfamily dummy NWORK: Number of members working A The reason we include two variables for N6: Number of members under 6 (SFAM). years of age N20 Number of members under 20 years of age the number of children is to estimate the effect of the SFAM: Subfamily dummy = I if there is elderly subfamily age of children on tenure decision in greater detail. P: Price per unit of housing services in logarithm Since we are interested in the estimation of the S: MON: Length Housing of shortage occupancy in percentage in the dwelling in logarithm relative importance of various household char- of months acteristics, the multicollinearity problem cannot be overlooked. In order to avoid this problem, the cor- number of households less the number of housing relation coefficient matrix and the changes in the units by the number of housing units in a city. estimated coefficients among alternative specifica- Finally, a household's length of occupancy in a tions having a different combination of variables dwelling (MON) reported in the survey is used as an were carefully examined. It was found that the indicator of mobility. household size and the number of members under The list of variables and their definitions are pre- 20 are highly collinear, having a correlation co- sented in Table 2. efficient of approximately We also note(' that Several multivariate statistical techniques are the regression estimates were highly sensitive to available for the analysis of a dichotomous dependchanges in the specification. The corrective pro- ent variable. We have opted for ordinary least cedure we adopted is the specification of two sets of squares (OLS) estimation, despite the recognised equations which separate the two variables. econometric problems of the procedure. 4 There are 2 We have obtained similar results for an analysis of demand for housing services which are reported in Follain, Lim and Renaud (forthcoming). 3 The price per uniit of housing services is calculated by the following equation: InP = klinpl+knlnpn where P is the price per unit of housing services, kl is the share of the value of total output spent on land input, PL is the price per unit of land, kn is the share of the value of total output spent on non-land inputs, and PN is the price per unit of non-land inputs. For detailed mathematical procedures in deriving this equation, see Follain, Lim and Renaud (forthcoming). 4 Some of the well-known drawbacks are: (I) error terms are likely to involve heteroscedasticity; (2) estimated probabilities may lie outside 0-1 range; (3) the linearity assumption ma t misspecify the true relationship; (4) the linear additive assumption cannot explain the multidimensional interaction effects among independent variables.

39 'DETERMINANTS OF HOME-OWNERSHIP IN KOREA 17 several reasons for this choice. First, as shown by Results of Empirical Analyses some Monte Carlo studies of linear additive func- The model described in the previous section is tions, the loss of efficiency in OLS is rather trivial estimated for five subgroups of the sample. The for large samples (i.e., greater than 100) compared household income and expenditure survey and the to generalised least squares (GLS) (Goldfeld and special housing survey cover households in 17 cities Quandt, 1972; Smith and Cicchetti, 1972). Second, in Klorea. The grouping of the sample allows us to the mean value of the dichotomous dependent vari- observe variations in estimates among cities of differable in our study falls between 0.3 and 0.5, thus, ent size. The first group consists of the entire sample substantially reducing the occurrence of the pre- except Jeju, a city on an island off the southern dicted value going outside of the 0-1 interval. Third, the lack of previous the revousstuieson ackof studies on tenure enue coic choice in in coast I iisfrwic.. (Table 3). The h second oenhusn group is composed oprto of 11 cities developing for which countries the Korean does Housing not Corporation warrant the use of provides construction cost information 6 (Table 4). more complicated and expensive tools such as GLS The third group is Seoul, the largest city of the and logit analysis. 5 The advantages of the more sophisticated techniques can be best appreciated only nation abe the fourthgro isrbusa and afte soe fatues esental o deermiant of Daegu xu'lich are the second and third largest cities, a riar ou- respectively (Table 6). Finally, the fifth group contenure choice in developing countries sists of eight cities of smaller size with construction lined. cost information 7 (Table 7). Table 3 Home-ownership regression models for all cities but Jeju* () () (mi) (IV) (V) (VI) Constant Yi (0.02) (0.02) Y (0.03) (0.03) (0.03) (0.03) HS ,,, (0.05) (0.05) (0.05) AGEH (0.06) (0.06) (0.06) (0.05) (0.06) (0.05) MALE (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) NWORK (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) N (0.02) (0,02) (0.02) (0.02) (0.02) (0.02) N (0.01) (0.01) (0.01) SFAM (0.05) (0.05) (0.04) (0.04) (0.04) (0.04) P (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) S (0.004) (0.004) MON (0.01) (0.01) R * Standard errors of estimates are in narentheses. 5 The main advantage of GLS is its ability to handle heteroscedasticity. The problem of linearity assumption remains. See Goldberger (1964). The logit analysis best qualifies for S-shaped true relationship. See Nerlove and Press (1973). For a comparison of OLS with other tools, see Jack Goodman (1976). 6 These cities are Seoul, Busan, Incheon, Chuncheon, Daejeon, Jeonju, Daegui, Masan, ChUngjU, Wconju and Suweon. The construction cost information was used to calculate the price of non-land inputs. 7 These cities include all in the third group except Seoul, Busan and Daegu.

40 18 GILL-CHIN LIM, JAMES FOLLAIN AND BERTRAND RENAUD Table 4 Home-ownership regression models for eleven cities* () (II) (llf) (IV) (V) (VI) Constant YI (0.02) (0.02) Y , (0.03) (0.03) (0.03) (0.03) HS (0.05) (0.05) (0.05) AGEH (0.06) (0.06) (0.06) (0.06) (0.06) (0.06) MALE (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) NWORK (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) N (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) 1N (0.01) (0.01) (0.01) SFAM (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) I' (0.07) (0.06) (0.06) (0.06) (0.06) (0.06) S (0.005) (0.005) MON (0.01) (0.01) R * Standard errors of estimates are in parentheses. Table 5 Home-ownership regression models for Seoul* (I) (11) (III) (IV') (V) (VI) Constirit Y (0.04) (0.04) Y (0.05) (0.05) (0.05) (0.05) HS (0.08) (0.08) (0.08) AGEH (0.10) (0.10) (0.10) (0.10) (0.10) (0.10) MALE (0.09) (0.08) (0.09) (0.08) (0.08) (0.08) NWORK (0.04) (0.04) (0.03) (0.03) (0.04) (0.03) N , (0.03) (0.03) (0.03) (0.03) (0.03) (0.C3) N (0.02) (0.02) (0.02) SFAM (0.07) (0.07) (0.07) (0.07) (0.07) (0.07) P (0.11) (0.10) (0.10) (0.10) (0.10) (0.10) MON (0.02) (0.02) R * Standard errors of estimates are in parentheses.

41 DETERMINANTS OF HOME-OWNERSHIP IN KOREA 19 Table 6 Honme-owniership r egressiont models for Busan and Daegut pooled* (I) (It) ([11) qv) (V) (VI) Constant Y (0.3) 0.03 (0.3) Y (0.05) 0.30 (0.05) 0.25 (0.06) 0.29 (0.05) HS (0.09) (0.10) (0.09) AGEH n.05 (0.11) (0.11) (0.11) (0.10) (0.11) (0.10) MALE (0.09) (0.09) (0.09) (0(08) (0.09) (0.08) NWORK (0.04) (0.04) (0.04) (0.03) (0.04) (0.03) N (0.04) (0.04) (0.04) (0.03) (0.04) (0.03) N (0.02) (0.02) (0.02) SFAM (0.08) (0.09) (0.08) (0.07) (0.08) (0.07) P (0.14) (0.14) (0.13) (0.12) (0.13) (0.12) S (0,01) (0.01) MON., (0.02) (0.02) R * Standard errors of estimates are in parentheses. Table 7 Fotome-owvnershlip regression models for eight cities of smaller size* (I) (il) (111) (IV) (V) (VI) Constant Y (0.06) (0.05) Y (0.06) 0.10 (0.06) 0.14 (0.06) 0.09 (0.06) HS (0.11) (0.11) 0.35 (0.11) AGEH 0.24 (0.13) 0.12 (0.1) 0.24 (0.13) 0.06 (0.13) 0.13 (0.13) (0.13) MALE (0.11) (0.11) (0.11) (0.10) (0.11) (0.10) NWORK (0.06) (0.07) (0.06) (0.06) (0.06) (0.06) N N20 (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) SFAM (0.03) (0.03) (0.03) (0.10) (0.10) (0.10) (0.10) (0.10) (0.10) P S (0.13) (0.12) (0.12) (0.12) (0.12) (0.12) (0.01) (0.01) MON (0.03) (0.03) 1J * Standard errors of estimates are in parentheses.

42 20 GILL-CHIN LIM, JAMES FOLLAIN AND BERTRAND RENAUD The results are generally consistent throughout effect of Y 2 is roughly three times stronger than that five groups. Although we are careful in interpreting of Y 1. The result strongly suggests a significant and the result of standard errors because of possible systematic difference between the effects of the two heteroscedastic error terms and inconsistent estimates income measurements for Korean households. This of standard errors, it is noteworthy to observe that is contrasted to the findings of a study for the UJS most key variables are statistically significant. The (Struyk, 1974) which showed stronger effccts of signs of the coefficients are found to be correct in permanent income only for the younger families. most cases and those variables with unexpected signs are usually accompanied by standard errors large enough to accept the null hypothesis of zero co- Houselhold Characteristics efficient. R square values range between 0.19 and Of the seven household characteristic variables, the 0.49, the average being This is a very high size of household (HS) turns out to have the most level of explanatory power, because R' values of significant and influential effect on the probability of regressions using individuial data rarely exceed 0.30 due to the discrepancy between home-ownlership. the ranges All of signs correspond to a priori the expectations and the coefficients are statistically actual and predicted values of a binary dependent sigivcant at the 1 per cent level in most cases. The variable. Comparable studies conducted in the magnitude of the coefficient covers a range of 0.20 United States report R's of much smaller magnitude. 8 Th reuts fo Bua an Dag pold,n the to 0.54 with a mean of This e r B indicates s fr n that on the average group doubling of eight cities the of household smaller size size, are less with other consistent things being equal, will increase the probability of than others. Some of the results for these two.4 grous groups th cntrdic reultsforothr contradict the results goup for other Wehome-ownership groups. We the by variation It is also in the observed size of that the coefficient among note that the sample sizes for the two groups are much different groups of samples is relatively small. We smaller than others and primarily concentrate on the consider the stability of the coefficient as an evidence interpretation of the result for the first tlhree groups. for universal importance of the household size variable in tenure decision in Korea. Inconie Three other variables which characterise householis' size and composition are also found signifi- Both current incomne (Y 1 ) and consumption as a cant. Except for a few equations for Seoul the effect measure of permanent income (Y 2 ) prove to have of the number of members under 6 years of age (N6) positive and significant influence on the probability is significantly different from zero at least at the 5 per of owning a home. Except for two cases, Y 1 is cent level. All signs are negative and the coefficient statistically significant at least at the 5 per cent level lies between and with the average of for a two-tailed test. The statistical significance of In contrast to N6 the numbc-i of members Y2 is even much higher than that of Y 1. In most under 20 (N20) shows positive signs for all equations. equations Y 2 is significant at the 1 per cent level. It is statistically more significant than N6. All co- In terms of magnitudes of the coefficients, Y 2 also eflicients are significant at least at the 5 per cent shows strong-r effect on the probalbility of home- level. The variable's effect on the probability of ownership. The coefficient of Y 1 ranges between home-ownership ranges between 0.05 and The 0.03 and 0.12 with a mean value of Since the average value of the coefficient is income variables are in logarithmic forms, the co- The interpretation of the coefficients of N6 and efficient of 0.06 for Y 1 means that a 100 per cent N20 needs some caution because they are estimated increase in current inconme will result in the increase by tw~o differenitly specified models. The coof the probability of home-ownership by For efficienit of for N6 means a household with a Y 2 we obtained a range of 0.10 to 0.30 and a mean child utnder 6 years of age is 6 per cent less likely to of O.19. Thisimpliesthatifahousehold doublesits be a home-owner than a houselhold with the same permanent income, the probability of owning a home size and characteristics but without a child under 6 is increased by The result indicates that the years of age. The degree of the net effect of an 8 Struyk (1974) reports R 2 of for his OLS estimation. Kain and Quigley (1975) obtained for their OLS.

43 DETERMINANTS OF HOME-OWNERSHIP IN KOREA 21 additional child depends on the family size. For simply explain that the existence of a subfamily, exa.mple, if a household of three members gains a which is defined as the parents of the head of the child under 6 years of age, the net effect on the household, increases the probability of home-ownerprobability of home-ownership equals which ship because of increased demand for home-owneris the sum of the effects of HS (0.143) and N6 ship. An alternative interpretation can be derived (-0.06). For a household of four members, it is from the wealth aspect of the existence of a subabout Therefore, for a given size of a house- family. It has long been a social convention in hold the presence of children under 6 years con- Korea that the first son of a household lives with stitutes a disincentive for home-ownership, but the his parents in their house 1 ' even after his marriage net effect of an additional member under 6 years is and inherits the house. The inheritance of the house positive. Sirnce the coefficient of N20 is estimated is often made while they live together. If this is the in a model without HS, the value of 0.07 simply case SFAM is not a cause for the increased proreflects the effect of an additional member under 20 bability of home-ownership, but a predetermined years and a direct comparison of the coefficients of factor for a higher rate of ownership. It scems that N6 and N20 cannot uncover the relative strength of both are reasonable interpretations, but we hesitate the two variables. To conmpare the relative strength to make any conclusive remarks because of the lack of N6 and N20, the net effect of N20 is computed of information on the wealth of the households. We using a model which includes both HS and N20.9 believe this subject can be better investigated by a For a household of four it is about The result future study with detailed information on wealth indicates that the effect of N20 is stronger than N6. status of households. A reference to a couple of particular aspects of Age of the head of household (AGEH) shows an household composition and real estate practice expected sign for 27 out of 30 equations. The coseems useful to understand the result. First, in most efficients with expected signs are mostly significant Korean households children stay with their parents at the 5 per cent level. Three cases wilh unexpected until they graduate from college or get married. signs are not statistically significant. The average Therefore, pressure to own a house increases with value of the coefficient for the entire sample is about the increase in the number of grown-up children For Seoul it is The size of the co- The number of children in their early childhood has efficient covers a rather wide range ( ). This a smaller effect on the probability of home-owner- may be due to multicollinearity. Although we found ship, because children under 6 years of age can no large correlation coefficient between AGEH and always share a room with their parents.10 Second, other variables, it is possible that AGEH is pairwise as already described, a full payment of the cost of a independent and yet a linear combination of two or house at the time of purchase is a standard real more of other variables. estate practice in Korea. The accumulation of the The studies done for the US generally show that a capital equal to the full value of a house takes a female-headed household is less likely to be a homenumber of years of thrift, preventing many couples owner all other things being equal. The evidence with children from becoming home-owners. presented here suggests that it could well be a reverse The presence of a subfamily dummy (SFAM) in Korea. The dummy variable for male-headed shows positive and significant effects on home- household (MALE) shows a negative sign for 26 out ownership for all groups except the group of eight of 30 equations. Eleven cases with negative sign and cities of smaller size. FoI- the first four groups the four cases with positive sign are statistically invariables are significant at the 5 per cent level in significant. The size of the coefficient is quite large. most cases. The magnitude of the coefficient is The simple average from all equations is Two about There are a couple of alternative ways alternative explanations can be offered. It may be of interpreting the rcsult for SFAM. First, we may simply due to a bias in sampling or the result may 9) This model is not presented because of the problem involving multicollinearity. For the purpose of prediction, however, it is still useful. ) For a more detailed analysis of determinants of residential crowding, see Follain, Lim and Renaud (1979). 11 Studies on household composition in Korea indicate that a large proportion of Korean households are compuised of more than two generations. See Kwon, Lee, Chang and Yu (1975).