Elsevier

Journal of Public Economics

Volume 115, July 2014, Pages 131-145
Journal of Public Economics

Estimating the distortionary effects of ethnic quotas in Singapore using housing transactions

https://doi.org/10.1016/j.jpubeco.2014.04.006 Get rights and content

Highlights

  • I study the ethnic housing quotas in Singapore using housing transactions.

  • I matched 500,000 names in phonebooks to ethnicities to construct ethnic proportions.

  • Transaction outcomes are different between constrained and unconstrained blocks.

  • Selection effects cannot fully account for these differences.

  • Thin housing markets and inelastic preferences present challenges to desegregation.

Abstract

Desegregation is a key policy issue in many countries. I investigate a residential desegregation program in Singapore — the ethnic housing quotas. I show that choice restrictions imposed on apartment blocks above the quota limits (constrained) could have distortionary effects, causing price and quantity differences for constrained versus unconstrained blocks. I test these predictions by hand-matching more than 500,000 names in the phonebook to ethnicities, to calculate ethnic proportions at the apartment block level. I can then investigate differences for constrained and unconstrained blocks close to the quota limits and test for sorting around the limits. I find that price differences are between 3% and 5%. Quantity effects are economically significant, translating to longer time-on-market durations. Selection cannot fully explain these results. My results point to challenges in achieving desegregation using quantity restrictions.

Introduction

Around the world, there are many policies encouraging gender and ethnic diversity in education, public and private employment, politics, and housing.1 Countries commit large amounts of money in their budgets to encourage diversity.2 There could be unintended distortionary costs introduced by these policies. There is a large literature on the redistributive benefits of these policies, but there is less on distortionary costs (Holzer and Neumark, 2000). I investigate potential policy-induced distortionary costs using a natural experiment in Singapore — the ethnic housing quota policy.

The quota policy was introduced in 1989 to prevent further segregation among the three major ethnic groups in Singapore — Chinese (77%), Malays (14%) and Indians (8%) (Singapore Department of Statistics, 2000). The policy is a set of limits on Chinese, Malay and Indian proportions that determine which ethnic groups are “segregated” in an area.3 In areas above the quota limit, sellers from the non-segregated group cannot sell to buyers from the segregated group because this transaction increases the ethnic proportion of the segregated group farther above the quota limit. I quantify the distortionary effects of these policy restrictions by comparing housing transaction outcomes for constrained areas subject to the restrictions and comparable unconstrained areas. To my knowledge, this is the first paper to investigate distortionary effects of a residential desegregation policy using housing transactions.4

To do this, I develop a conceptual framework that delivers testable predictions on how prices and quantities are expected to differ for apartment blocks that are above the quota limit (constrained) versus blocks below the quota limit (unconstrained). By restricting the choices of whom non-segregated sellers and segregated buyers can transact with, the model shows that these choice restrictions on the demand- and supply-side can lead to differences in transaction outcomes for constrained blocks versus unconstrained blocks.

The model has two important features: buyers have segregation preferences5 and housing markets are thin because housing attributes are heterogeneous along many dimensions (Arnott, 1989). Therefore, a housing unit in a given location with a vector of attributes could have few (or no) units sharing similar attributes. Because of segregation preferences, segregated buyers are willing to pay more than non-segregated buyers to live in constrained blocks. So, non-segregated sellers who cannot sell to segregated buyers have to accept a lower price to attract non-segregated buyers. Because housing markets are thin, segregated buyers who can only buy from segregated sellers of constrained blocks, may be willing to pay a higher price to live in constrained blocks if their most preferred unit is in the constrained block, their second most preferred choice is in an unconstrained block and the two choices are imperfect substitutes.

In the paper, I show that under certain assumptions, the model predicts the following price and quantity effects. First, on average, prices will be higher for Chinese-constrained blocks versus comparable unconstrained blocks, but prices will be lower for Malay- and Indian-constrained blocks. Second, fewer units will be sold in constrained blocks versus comparable unconstrained blocks. The model also highlights two mechanisms that have opposite price effects (segregation preferences versus thin markets).

I test these predictions on prices and quantities using data on two transaction outcomes, prices and the proportion of units in an apartment block that was sold during my sample period. The main identification challenge is that whether a quota binds or not is correlated with unobserved housing quality. To circumvent the problem that quota-constrained and quota-unconstrained locations are not comparable, my strategy is to identify kinks in the outcome variable that coincide with kinks in the policy rule while controlling flexibly for ethnic proportions. The identification strategy is similar in spirit to the regression kink design (Card et al., 2012).

This research design requires many observations above and below the quota limits. However, many desegregation policies impose strict upper limits so that few or no housing areas are above these limits.6 By contrast, when the quota policy was implemented in Singapore in 1989, the Housing Development Board (HDB) did not want to evict owners in apartment blocks that were quota-constrained and they also wanted to minimize the number of households that would be affected. Therefore, they allowed all transactions that involved buyers and sellers of the same ethnicity because these transactions did not make housing areas more segregated. One benefit of analyzing housing transactions in my context is that there are many transactions both below and above the limits.7

This empirical strategy also needs data on the running variable used to determine the quota status. For the ethnic quotas in Singapore, the running variable of interest would be the ethnic proportions at the apartment block level. Since many of these policies are highly contentious, it is often hard to find public data of the running variable or even public data of the quota limits. 8 I circumvent these data issues by hand-matching more than 500,000 names to ethnicities using the Singapore Residential Phonebook. This allows me to calculate ethnic proportions for more than 8000 apartment blocks. I combined this data with outcomes for more than 35,000 housing transactions that I downloaded from the HDB website. While I do not have administrative data used by HDB to determine the quota status of each block, I show in the paper that my proxy calculated using the phonebook is a valid measure of ethnic proportions.

An important identification assumption is that individuals cannot “precisely sort” around the quota limits so that variation in the treatment status around the policy limit is “as good as randomized” (Lee and Lemieux, 2010). I test for discontinuities in the density of the running variable and do not find evidence of sorting using Chinese and Indian proportions (McCrary, 2008). For Malays, the sorting pattern is not consistent with households trying to manipulate treatment assignment. I return to this in Section 6.

I find price effects that are comparable to the literature on diversity-enhancing policies and larger quantity effects. On average, transaction prices of Chinese-constrained units are 5% higher than observably comparable unconstrained blocks. The average prices are 3% lower for Malay- and Indian-constrained blocks. Additionally, units in constrained blocks tend to be harder to sell. These effects are economically significant, translating into units being on the market 1.4 to 3.7 months longer (the median duration in this market is 42 days). I show that the results above cannot be fully explained by selection.

These estimated differences in prices and quantities between constrained and unconstrained blocks are consistent with my model. They suggest that choice distortions due to demand- and supply-side choice restrictions are significant, leading to differences in transaction outcomes. Further, I use the opposite price effects predicted in the model to disentangle the two mechanisms discussed above. I find evidence that segregation preferences are important for all three quotas. This suggests that location preferences are inelastic because of segregation preferences. I also find support for supply-side constraints and thin markets for the Chinese quotas.

Together, these estimated policy effects on prices and quantities of sold housing units imply that transaction values of constrained blocks are 12% to 21% lower than the transaction values of comparable unconstrained blocks. My calculations suggest that more than 70% of the impact on transaction values is due to reductions in the quantity domain. Understanding these effects on transaction values is important because the ease of sale affects household mobility, housing transactions have spillover effects on the broader economy and housing transaction taxes are an important source of tax revenue.9

These results point to distortionary effects from imposing quantity restrictions. An important lesson is that diversity-enhancing policies can exacerbate existing frictions, especially in the housing market. This contrasts against a literature on second-best choices suggesting that distortions from diversity policies could offset existing imperfections in the market, leading to efficiency gains.10 It is exactly because the housing market is thin that there are already few transactions relative to the housing stock. Restricting non-segregated sellers from transacting with segregated buyers could exacerbate, rather than mitigate this problem, leading to even fewer units being sold for constrained blocks.

My findings have policy implications beyond Singapore. Many countries have residential desegregation programs, and some include quota limits.11 These results suggest that it might take a long time to achieve desegregation goals. The fact that many blocks are still above the quota limits sixteen years after the policy was introduced, and that there are still meaningful differences in transaction outcomes between constrained and unconstrained blocks suggest the combination of inelastic preferences and thin markets is important, presenting challenges to achieve desegregation. In a separate paper that uses quasi-experimental variation from the quota policy to structurally estimate preferences and simulate the first best allocation of ethnic groups to neighborhoods, Wong (2013) finds that sixteen years after the policy, only 30% of neighborhoods are within one standard deviation of the first best allocation.

One caveat of the analysis is that I do not have pre-policy data so, I cannot simulate a counterfactual of no policy. My identification strategy compares transactions for constrained versus unconstrained blocks, holding the supply of housing units and the distribution of preferences fixed at the post-policy equilibrium. In the paper, I discuss possible confounders due to general equilibrium adjustments and show that they cannot fully explain my results. Moreover, the preceding discussion shows that finding meaningful differences in transaction outcomes for constrained versus unconstrained blocks so many years later is important.

The remainder of the paper proceeds as follows. In Section 2, I provide some background on Singapore and the ethnic quotas. I describe the data in Section 3, provide a theoretical and empirical framework in 4 Conceptual framework, 5 Empirical framework, discuss results in Section 6 and conclude in Section 6.1.

Section snippets

Background

Singapore is a multi-ethnic country with a population of 4.5 million (Singapore Department of Statistics, 2006). The three major ethnic groups are the Chinese (77%), the Malays (14%) and the Indians (8%). The Chinese have the highest median monthly income (S$2335), followed by the Indians (S$2167) and the Malays (S$1790).

Public housing is the most popular choice of housing in Singapore with 82% of the resident population living in public housing (Housing Development Board, 2006). The units are

Data

Table 2 lists the summary statistics of the full dataset. The analysis only focuses on the public housing market which represents 82% of the citizens and permanent residents in Singapore. There are 8007 blocks and 35,744 resale transactions. Online Appendix includes more details on how the sample was created.

Conceptual framework

My empirical analysis compares transactions in constrained and unconstrained blocks. This section lays out a theoretical framework to deliver testable predictions for two transaction outcomes: number of sold units (quantity) and prices. I focus on describing treatment effects for the Chinese block quota limits only. The effects for the other quota limits are similar.

I use a static, discrete choice framework with heterogeneous households choosing heterogeneous HDB units.18

Empirical framework

The thought experiment in the theoretical framework compares constrained versus unconstrained blocks that are otherwise conceptually identical. My empirical framework operationalizes this thought experiment by comparing transaction outcomes of constrained versus unconstrained units, restricting the analysis to blocks close to the quota limits, controlling flexibly for polynomials of ethnic proportions (an important attribute that drives sorting behavior) and a set of observables that can

Results

In this Section, I first describe the main results on transaction outcomes, then, discuss take-away lessons from these estimates. Finally, I discuss major threats to identification of the treatment effects and argue that alternative stories cannot explain my results.

Conclusion

Many desegregation policies take the form of quotas but it is hard to find the data to evaluate these policies credibly because they are either not available publicly or there are not enough observations close to the quota limits. This paper uses a hand-collected dataset to study the impact of the ethnic housing quotas in Singapore, one of many residential desegregation programs around the world.

I develop a model that shows that distortionary effects can arise when segregated and non-segregated

References (67)

  • R. Arnott

    Housing vacancies, thin markets, and idiosyncratic tastes

    J. Real Estate Financ. Econ.

    (1989)
  • D. Autor et al.

    Housing market spillovers: evidence from the End of rent control in Cambridge Massachusetts

  • S. Banhardt

    Near and dear? Evaluating the impact of neighbor diversity on inter-religious attitudes

    (2009)
  • P. Bayer et al.

    A unified framework for measuring preferences for schools and neighborhoods

    J. Polit. Econ.

    (2007)
  • L. Beaman et al.

    Female leadership raises aspirations and educational attainment for girls: a policy experiment in India

    Science

    (2012)
  • R. Benabou

    Workings of a City: Location, Education, and Production

    Q. J. Econ.

    (1993)
  • S.T. Berry

    Estimating discrete-choice models of product differentiation

    RAND J. Econ.

    (1994)
  • S.T. Berry et al.

    Automobile prices in market equilibrium

    Econometrica

    (1995)
  • J. Boisjoly et al.

    Empathy or antipathy? The impact of diversity

    Am. Econ. Rev.

    (2006)
  • L. Boustan

    Racial residential segregation in American cities

  • D. Card et al.

    Tipping and the dynamics of segregation

    Q. J. Econ.

    (2008)
  • D. Card et al.

    Nonlinear policy rules and the identification and estimation of causal effects in a generalized regression kink design

  • Chay, K., Fairlie, R., 1998. Minority business set-asides and black self-employment. University of California,...
  • R. Chetty et al.

    Adjustment costs, firm responses, and micro vs macro labor supply elasticities: evidence from Danish tax records

    Q. J. Econ.

    (2011)
  • A. Corns et al.

    Can affirmative action be cost effective? An experimental examination of price-preference auctions

    Am. Econ. Rev.

    (1999)
  • D.M. Cutler et al.

    The rise and decline of the American ghetto

    J. Polit. Econ.

    (1999)
  • Dutch Refugee Council

    Housing for refugees in the European Union

  • P.-A. Edin et al.

    Ethnic enclaves and the economic success of immigrants — evidence from a natural experiment

    Q. J. Econ.

    (2003)
  • P. Edin et al.

    Settlement policies and the economic success of immigrants

    J. Popul. Econ.

    (2004)
  • D. Epple et al.

    Estimating equilibrium models of local jurisdictions

    J. Polit. Econ.

    (1997)
  • S. Gabriel et al.

    Household location and race: estimates of a multinomial logit model

    Rev. Econ. Stat.

    (1989)
  • A. Gavazza

    The role of trading frictions in real asset markets

    Am. Econ. Rev.

    (June 2011)
  • I. Hendel et al.

    The relative performance of real estate marketing platforms: MLS versus FSBOMadison.com

    Am. Econ. Rev.

    (2009)
  • Cited by (19)

    View all citing articles on Scopus

    I wish to thank my advisers, Professors Esther Duflo, Amy Finkelstein, Panle Jia and Bill Wheaton for their guidance and support. I am also deeply indebted to Professors Peter Diamond, Fernando Ferreira, Michael Greenstone, and Joe Gyourko, two anonymous referees and the editor for their helpful comments. I have benefited from the conversations with Professor Chua Beng Huat, Jessica Cohen, Greg Fischer, Raymond Guiteras, Kam Lee Ching, Trang Nguyen and Wong Liang Kit. I would like to thank the Singapore Housing Development Board for their permission to use their data. This study would not have been possible without the support from the MIT Shultz Fund. I am also grateful for the financial support from the Zell-Lurie Real Estate Center. Lee Hye Jin and Chen Ying provided excellent research assistance. All remaining errors are my own.

    View full text