University of Southern California Dissertations and Theses

Essays on housing market behavior analysis within the international context

(USC Thesis Other)

Essays on housing market behavior analysis within the international context

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON HOUSING MARKET BEHAVIOR ANALYSIS WITHIN THE INTERNATIONAL CONTEXT by Diehang (Della) Zheng A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PLANNING AND DEVELOPMENT STUDIES August 2007 Copyright 2007 Diehang (Della) Zheng ii Dedication I dedicate this work to my family, especially … my parents for their unconditional support and care and Christian for his understanding and patience. iii Acknowledgments My deepest gratitude is to my advisor, Dr. Yongheng Deng, for his dedication to work, many years of guidance, and generous support. I truly appreciate Dr. Peter Gordon for his warm encouragements, and the time he devoted to providing skillful comments and helpful advice. Dr. Delores Conway is one of the best teachers that I have had. She helped me tremendously both inside and outside academia. Special thanks go to Dr. Stuart Gabriel for his insightful comments and constructive criticisms which hold me to a high research standard. To the faculty and staff of the School of Policy, Planning, and Development (SPPD), Marshall School of Business, Department of Economics, and Department of Mathematics assisted and encouraged me in various ways during my course of studies. I am especially grateful to Dr. Tridib Banerjee, Dr. Raphael Bostic, Dr. Cheng Hsiao, Dr. Jaksa Cvitanic, and Dr. Fernando Zapatero. I also thank the students I was privileged to teach and from whom I learned. I appreciate the great support from Pacific Legal Foundation that funded parts of the research in this dissertation. iv Table of Contents Dedication ii Acknowledgements iii List of Tables v List of Figures vii Abstract viii Chapter 1. Optimal Pricing Strategy under Price Dispersion: New Evidence from the Tokyo Housing Market 1 Introduction 1 Theoretic Model. 3 Data 12 Empirical Results 25 Conclusions 37 Chapter 2. An Early Assessment on Residential Mortgage Termination in China 40 Introduction 40 Current Residential Mortgage Market in China 43 Option Theory and the Proportional Hazard Model 47 Data 50 Empirical Results 62 Conclusions 75 Chapter 3. Impact of Rent Control on Mobile Home Prices in California 78 Introduction 78 Mobile Home Industry 81 Model Setup 84 Data 88 Empirical Results 98 Conclusions 113 Bibliography 116 Appendices 123 Appendix A 123 Appendix B 133 v List of Tables Table 1.1 Listing and delisting –yearly counts 17 Table 1.2 Mean and std. deviation of selected continuous variables 19 Table 1.3 Mean and std. deviation of selected time varying covariates 20 Table 1.4 Frequency of selected discrete variables 24 Table 1.5 OLS regressions on price for the full sample 27 (a) Dependent variable - logarithm of list price, at initial listing 27 (b) Dependent variable - logarithm of list price, at delisting 28 Table 1.6 OLS regressions on price for the sold units 29 (a) Dependent variable - logarithm of list price, at initial listing 29 (b) Dependent variable - logarithm of list price, at delisting 30 Table 1.7 Time on the market: proportional hazard models 33 Table 1.8 Time on the market: proportional hazard models 35 Table 2.1 Descriptive statistics for mortgage loans - means and standard deviations at origination 53 Table 2.2 Descriptive statistics for mortgage loans – means and standard deviations for time varying covariates 54 Table 2.3 Descriptive statistics for mortgage loans -frequency of loans by major categorical covariates and by payoff types 56 Table 2.4 Mortgage choice and early termination 60 Table 2.5 Proportional hazard estimates for mortgage prepayment and default 63 Table 2.6 Prepayment and default schedule for different mortgage maturity groups 66 (a) Prepayment schedule for different mortgage maturity groups 66 (b) Default schedule for different mortgage maturity groups 66 Table 2.7 Proportional hazard estimates for mortgage prepayment and default, extended models (part a) 67 vi Table 2.8 Proportional hazard estimates for mortgage prepayment and default, extended models (part b) 73 Table 3.1 Annual growth rates (1984-2002) 92 Table 3.2 Descriptive statistics – means and standard deviations of continuous covariates 93 Table 3.3 Descriptive statistics – frequencies of categorical covariates 96 Table 3.4 GLS estimate for policy change indicators 99 Table 3.5 GLS estimates for logarithm of price growth rate, by census tract with different characteristics 105 (a) GLS estimates for logarithm of price growth rate, by census tract household income 105 (b) GLS estimates for logarithm of price growth rate, by census tract proportion of elderly population 106 (c) GLS estimates for logarithm of price growth rate,, by census tract median household income and proportion of elderly population 107 Table 3.6 Resale price comparison for an average structure in Los Angeles 109 vii List of Figures Figure 1.1 Demographic and economic indicators for Tokyo, 1994-2002 13 (a) Year to year percentage change in population and net immigrants 13 (b) CPI (2000=100) 13 (c) Unemployment rate 13 (d) Average monthly household income (yen) 13 Figure 1.2 Properties’ time on the market, by status 18 Figure 1.3 Plots of submarkets 21 (a) Standard deviation in sales prices vs. average sales price 21 (b) Number of lists vs. number of transactions 22 Figure 3.1 Mobile home transactions (1982-2002) 90 (a) Total traded square footage 90 (b) Total traded value (constant $, base=1996) 90 (c) Number of transactions 91 Figure 3.2 Logarithm price growth rate by income and age sub-groups 103 (a) Logarithm of price growth rate by income group 103 (b) Logarithm of price growth rate by age group 103 Figure 3.3 Comparison in indexes: no rent control vs. change in rent control policy (adoption of rent control without decontrol) 108 Figure 3.5 Simulation results: resale price comparison for Los Angeles county 111 (a) Resale price comparison 111 (b) Resale price comparison-median household income 112 (c) Resale price comparison-age community 112 (d) Resale price comparison-median household income and elder proportion 113 viii Abstract This dissertation consists of three essays on urban economics and housing market, emphasizing on market behavior within an international context. Chapter One comprises an analysis of the optimal selling strategy in a market with price dispersion using the central Tokyo condominium resale market list data from 1994 to 2002. The optimal pricing strategy is chosen to maximize the return from search. Higher price dispersion leads to higher reservation and optimal asking prices, which in turn results in higher expected sales prices. Under the assumption that the offering prices follow a normal distribution, market price dispersion can increase the probability of a successful transaction and/or speed up the sale process for the overpriced properties. Chapter Two includes a discussion of the transitional residential mortgage market in China using a unique micro dataset depicting the mortgage loan history in Beijing. While the option theory failed to explain prepayment and the default behavior in current Chinese residential mortgage markets, other non-option theory related financial economic factors play major roles. Short term mortgage borrowers are more likely to prepay. Many borrowers choose to pay off mortgage debts in bear market. Unemployment rate is positively associated with mortgage prepayment rate. Borrower’s characteristics, such as income, education, marital status, etc., are significant in determining prepayment and default behaviors, hence may be used as an effective tool to screen potential high risk borrowers. ix Chapter Three examines the impact of mobile home parks rent control on mobile home resale prices by using transaction data from seven California counties between 1983 and 2003. The imposition of rigid rent control (rent control without vacancy decontrol) leads to higher growth rates in resale prices; while a flexible regime, or rent control with vacancy decontrol, results in lower growth rates in resale prices. It suggests that the imposition of rigid rent control will lead to the capitalization of future rent savings when a coach is sold. That is, the buyer will not only pay for the coach but also for the net present value of the expected savings associated with the future of legally constrained rent obligations to the landlord. 1 Chapter 1 Optimal Pricing Strategy under Price Dispersion: New Evidence from the Tokyo Housing Market 1. Introduction Violation of “the law of one price” is common even for homogenous products. Price dispersion is prevailing and persistent. Most sellers are only aware of the market offering price distribution rather than a single determined price according to the demand-supply model. After Stigler’s seminal paper (1961), search model, as the agents’ solution to price dispersion, has been widely discussed, especially in labor economics, commodity pricing, and optimal auction strategy, among other subject areas. In a market with offering price dispersion, the seller can set a take-it-or-leave price, or state a starting price for negotiation. 1 In either case, she needs to establish the asking price before she receives a response or counter-offer from the potential buyers. Together with the well-discussed reservation price, the optimal asking price is an important component of an optimal selling strategy in order to maximize the seller’s return from the search. Recent literature acknowledged the importance of the asking price. However, limited attention has been paid so far to the role of offering price dispersion per se. 1 The case of auction is not discussed here. 2 The housing market provides us with a natural laboratory to examine the effects of price dispersion. A huge volume of list and transaction data has been documented. Sellers can search for transaction in the submarkets with relatively homogenous properties and lower price dispersions, or in the submarkets with heterogeneous properties and higher price dispersion. Hence, it is feasible to check the impact of price dispersion across different submarkets. Although we cannot directly observe the seller’s reservation price, other observable factors and search outcomes, such as list prices, sales prices, and the property’s time on the market, enable us to study the impact of price dispersion. We assume that the seller understands the offering price distribution. Within the search model framework, a seller first sets her reservation price and asking price, aiming at maximizing the expected return from the search. As shown in the next section, the reservation price is determined by search cost, seller’s minimum required return or the opportunity cost, and the market conditions, including the offering price distribution and the offer arrival rate. A seller will accept an offer only if it is above her reservation price. On the other hand, the buyer’s offering price will never be higher than the seller’s asking price. Therefore, the reservation price and asking price serve as the floor and ceiling respectively for the buyer’s offering price and the final sales price if there is a successful transaction. I use the Tokyo condominium resale market data as the empirical example to investigate the impact of price dispersion on (1) the asking price, (2) sales price, and (3) time on the market. 3 I find that (1) consistent with previous studies, the reservation price is related to the expected return from search; the optimal asking price is determined by the market conditions and the expected return from search; (2) citeris paribus, the greater price dispersion allows the seller to set higher reservation price and higher optimal asking price; (3) for normally distributed offering prices, price dispersion leads to higher sales price and quicker sale of overpriced properties. This chapter is organized as following. Section 2 comprises a brief literature review and a simple search model. Section 3 introduces the Tokyo condominium resale market dataset. Section 4 shows the results of our empirical study. Section 5 concludes this chapter. 2. Theoretical Model Under the assumption of neoclassical microeconomics, in the market for a homogenous product, with a large number of identical rational sellers and buyers, perfect information, neither transaction cost nor capacity constrains, the single Nash equilibrium price is a perfectly competitive price. However, this “ideal” market is hard to find in reality. For most goods, not a “single price” but a range of prices are observed. 2 It has generated many theoretical models in trying to explain how different market conditions could lead to equilibrium price dispersion. 3 A lot of 2 The examples are, but not restricted to, automobile (Stigler, 1961), retail store (Lach, 2002; Pratt et al. 1979; Sorensen, 2000), securities (Garbade and Silber, 1976; Hamilton, 1987), housing market (Leung et al. 2006), insurance (Schlesinger and Schulenburg, 1991; Berger et al. 1989; Mathewson, 1983; Dahlby and West, 1986; Seog, 2002), air ticket (Borenstein and Rose, 1994), etc. 3 The possible explanations are spatial competition (Hotelling, 1929; Butters, 1977; Shilony, 1977), heterogeneity of sellers or consumers (Braverman, 1980; Diamond, 1987; Salop and Stiglitz, 1977; Reinganum, 1979; Wilde and Schwartz, 1979; MacMinn, 1980; Reitman, 1991; Postel-Vinay and Robin, 2002), product differentiation (Perloff and Salop, 1985), search friction (Reinganum, 1979; 4 discussions have also been carried out to address the agents’ solution to price dispersion as various search models. The reservation price is the key element of the traditional search model. For a specific market, the reservation price implies the expected return from the search and search duration. However, most transactions start with the asking price rather than the unobservable reservation price. Clearly, the selling strategy is composed of both the “hidden” reservation price and the “exposed” asking price. Its outcomes are the sales price and search duration, or the time on the market. For a rational seller, the goal of her strategy is to sell the goods at the price as high as possible and as quickly as possible, which is represented by the expected return from the search. Researchers started to understand the importance of the asking price. Stull (1978) as well as Guasch and Marshall (1985) developed search models in which asking prices must be set, but they did not consider the case whereby transaction prices may be below list prices. Chinloy (1980) assumed that the seller’s reservation price is a constant fraction of the asking price. Horowitz (1992) as well as Chen and Rosenthal (1996) showed that the asking price serves not only as the resource allocation mechanism, but also as the upper bound of the transaction price. Yavas and Yang (1995) looked at both the asking price and the time on the market, but they did not explicitly address the importance of price dispersion. Chen and Rosenthal (1996a, 1996b) showed that the asking, or ceiling, price is the incentive for potential buyers to incur search costs. Arnold (1999) demonstrated from a search-and- Sorensen, 2000), imperfect information (Nelson, 1970; Varian, 1980; Burdett and Judd, 1983), and seller’s capacity constrains (Dana, 1999; Arnold, 2000), etc. 5 bargaining model that the asking price influences the rate at which potential customers arrive. The optimal asking and reservation prices are characterized. Haurin et al. (2006) showed the seller’s rational selection of asking price and reservation price, but they did not take discount rate into consideration and also overlooked the opportunity cost. In the multistage search model, I assume the buyers’ offering prices follow distribution with cdf F(p) and arrive according to a Poisson process with parameter λ . The continuous discount rate is γ per period. If an offer of price p is accepted, the seller’s expected payoff is Sp c = − , where c is the seller’s cost of search to bring in current offer. Note that a p P ≤ because the asking price, a P , serves as the price ceiling (see Horowitz, 1992; Chen and Rosenthal, 1996; Arnold, 1999). If the offer is not accepted, the payoff is { } / (max ' , ' ) Wb c e E p cW γλ − =− + − , where b is the value of the property’s second best use, or the opportunity cost of sale, and ' p is the amount of the next offer. Then the value of having an offer in hand is { } ( ) { } / ( ) max , max , ( ') Op SW p cb c e E Op γλ − == −−+ (1.1) The reservation price is the unique price such as making the decision between sale or wait indifferent, i.e. ( ) r SP W = . Thus () / (') r Pb e EOp γλ − =+ , or the difference between the reservation price and the value of the property’s second best use is the discounted expected payoff from future search. Therefore, maximizing the return from search is equivalent to maximizing r P with a specific b. Insert above 6 formula into equation (1.1) and yield ( ) { } ( ) () max , ( ) r EO P E p P E c =− ; we can further obtain {} // 0 / 0 max , ( ) = ( ) ( ) ( ) / a r ra rr P P ra PP Pb e pPdFp e c b e PdFp pdFp PdFp C γλ γ λ γλ λ ∞ −− ∞ − =+ −  ++ + −    ∫ ∫∫ ∫ (1.2) If the total cost of search for every period is fixed as C, then the average search cost for each offer is / cC λ = . Note that the offer arrival rate, λ , is a decreasing function of the asking price, a P . 4 Equation (1.2) shows that the reservation price has two components. One is the value of the property’s second best use other than sale, and the other is the discounted expected return from search. Note that no search will happen if the expected payoff from sale is less than the value of the property’s second best use; or, equivalently, no search will happen if the expected return from search is negative. Hence, r Pb ≥ . Equation (1.2) reveals that the higher opportunity cost (from the 4 The potential buyers decide to inspect only if the transaction price is possibly below their reservation price ( r B ). The sales price discount rate is defined as / s a PP θ = . If the buyer is aware of the prevailing sales price discount rate ( θ ), and decide to inspect the property only if the asking price satisfies / ar PB θ ≤ , hence higher asking price will turn down more potential buyers and lead to lower offer arrival rate. On the other hand, if we assume the offer follow normal distribution, when the dispersion of the offering price becomes greater, keeping all the other variables constant, the potential buyers can be expressed as () 1 ( ) 1 a a a P P dF p F P θ θ µ θ σ ∞  − =− =−Φ   ∫ , which is positively related to σ ; or 0 λ σ ∂ > ∂ . 7 second best use) is associated with higher reservation price. 5 Besides that, higher seller’s discount rate, γ , 6 lower offer arrival rate, λ , and higher search cost per period, C , will all lead to lower reservation price. 7 On the other hand, when the value of above factors are all given, reservation price is a function of asking price, a P . For a well informed seller, she understands that her return-maximizing search problem is essentially a reservation-price- maximizing problem, as explained above. The solution is implied by the following first order condition: () * 22 * 1( ) () / a a C FP G p C P γ λ λ λλ ∂  −=− + −  ∂  (1.3) where 0 () () () () a r ra P P ra PP G p PdFp pdFp PdFp ∞ =+ + ∫∫ ∫ , and ( ) () / Gp C λ − represents the expected return from search. The decision on optimal asking price is related to the market condition (offering price distribution, offer arrival rate, and cost of search), seller’s discount rate, and the expected return from search. Note that the value of the property’s second best use is not considered in the selection of optimal asking price. 5 Genesove and Mayer (1997, 2002) proved that equity status and loss aversion both play significant roles in property selling. We can extend the value of the property’s second best use as a mix of rational financial valuation, effect of equity constrain, and sentiment factors (loss aversion), etc. 6 Glower et al. (1988) argued that sellers’ level of motivation to sell is important as well. 7 The impatient seller is represented by higher discount rate. She will consequently select lower asking price and lower reservation price, and will enjoy higher offer arrival rate and higher probability of match, and hence it is possible for her to sell faster. 8 Price dispersion and the optimal selling strategy The implication of higher price dispersion to the seller’s optimal strategy can be examined by the following derivatives, r P σ ∂ ∂ and * a P σ ∂ ∂ . Let’s first look at () Gp σ ∂ ∂ : () () 00 () () () () 1- ( ) ( ) a r a P P ar P a r ar pdFp PdFp p P dF p P P Gp FP FP σ σσσ ∞  ∂+ −−   ∂ ∂ ∂  ==+ ∂∂ ∂∂ ∫∫ ∫ . According to formula (2), () () / / () / () / r Gp C P e Gp C e γλ γλ λ λ σσ σ − − ∂− ∂ ∂ =−+ ∂∂ ∂ . Plug the calculated () Gp σ ∂ ∂ into r P σ ∂ ∂ and rearrange we can obtain () ( ) 2 // / 1( ) () / 1() a P a r r eFP e GpC C P eFP γλ γ λ γ λ σ σ λ γλ λ σ ∂ −− ∂ ∂ ∂ −   −+ − + ∂   = ∂− (1.4) According to formula (3), () * 22 2 * * () / () a a a Cr C Gp C P fP P λ λ λλ λ σσ σ    +− ∂   ∂ ∂    −=− + ∂∂ ∂ ∂       . Plug in the formula of () Gp σ ∂ ∂ and r P σ ∂ ∂ and rearrange to obtain: () ( ) ()() 2 / *4 / * */ * / () ( ) / 1 22 ( ) / 1() ()1 ( ) 1 ( ) 1() r a r a ar r a r FP e G P C C CGPC C PFPe P fP e F P F P P eFP γλ γλ γλ γ λ γλ γγ λ λλ λγλ λ γ σλ λ σ − − − −  −+  ∂∂  ⋅⋅ − − − + +  ∂∂ − ∂  = ∂ ∂ −−− ∂ − (1.5) The numerator is: 9 () () () () / *4 / 2 *4 () ( ) / 1 22 () / 1() 1 ( ) / 2 ( ) / 1 r a r a FP e G P C C CGPC C PFPe GP C C GP C PB γλ γλ γγ λ λλ λγλ λ γ σλ λλ γ γ λλγ λ σλ − −  −+   ∂∂   ⋅⋅ − − − + +  ∂∂ −   ∂∂    =⋅ ⋅ − + − − −      ∂∂ −   , where / () r B eFP γλ − = . Note * 0 a P λ ∂ < ∂ , 0 λ σ ∂ > ∂ , and ( ) () / GP C λ − >0. Further assume the discount rate between two expected offers is confined to / 0.5 1 e γλ − ≤< . It is easy to show / 2 1 B γ λ ≤ − , 8 then numerator > 0. On the other hand, the denominator is () () / 2 / * * ()1 ( ) 1 ( ) 0 1() ar r a r fP e F P F P P eFP γλ γλ γ λ λ − − ∂ −−− ∂ > − . Therefore, * 0 a P σ ∂ > ∂ . As to r P σ ∂ ∂ expressed in equation (1.4), note 0 a P σ ∂ > ∂ , ( ) () / 0 GP C λ − > , and 0 λ σ ∂ > ∂ , hence 0 r P σ ∂ > ∂ . So both reservation price and asking price increase with price dispersion. Expected sales price and time on the market The sales price is the only way for the seller to realize her return from search. The expected sales price is a conditional expectation based on the acceptance of the 8 For function () 1() x x fx eF p − = − , where () Fp is a cdf function with 0()1 Fp ≤≤ . () () 2 1()1 () 0 1() x x eF p x fx x eF p − − −− ∂ => ∂ − , if ( ) ()1 1 x eF p x − − < , which includes the range 01 x < < . When we define / x γ λ = , then x e − is the discount rate. The range in the discount rate / 0.5 1 e γλ − ≤≤ is equivalent to 0 0.693 x ≤ ≤ . Hence the maximum value of () f x is obtained with 0.693 x = at 1.386 . 10 offer above the reservation price. Following McCall (1970), the offers are assumed to be independent, then () () () ( | ) 1() a ra P a PP sr r pdF p P dF p EP E p p P FP ∞ + =≥= − ∫∫ (1.6) Taking derivative in terms of σ on both sides yields: [] 2 () 1 ( ) () () () () 1() 1() aa ra r PP s aa rr ar r PP P r EP F P P fP P pdF p P dF p P dF p FP FP σ σσ ∞  ∂−∂ ∂ =+− +   ∂∂−∂ −  ∫∫ ∫ (1.7) As showed previously that higher price dispersion results in higher asking price and higher reservation price, or 0 r P σ ∂ > ∂ and 0 a P σ ∂ > ∂ , we can in turn state that higher price dispersion will consequently lead to higher sales price as well, or () 0 s a EP σ ∂ > ∂ . In general, price and duration are two related aspects. Although the sellers’ goal is to sell the good for as high the price as possible and as quickly as possible, higher price is generally associated with longer time on the market. Belkin et al. (1976) pointed out that time on the market is an important descriptor of market behavior. Ho (2003) showed that liquidity is a desirable feature. Benveniste et al. (2001) demonstrated that improving liquidity can raise the market value of illiquid property assets. Consistent with the conventional search model, the probability of match regarding each offer is 1 ( ) r M FP = − . Following geometric distribution, the probability of sale at nth offer is 1 (1 ) n M M − − , and the expected number of offers 11 before sale is 1 () En M = . As mentioned above, if we assume there are λ offers arriving in each period, then the expected number of periods in the market is () 11 () 1() r EN M FP λλ == − (1.8) The reservation price, r P , determines the probability of match for each offer. The higher reservation price leads to the longer expected time on the market as the result of lower probability of match, as shown in () 2 () ()/ 1 () 0 rr r EN fP F P P λ ∂  =− >  ∂ . On the other hand, when the offer arrival rate is a decreasing function in a P , the higher asking price results in longer time on the market due to the lower offer arrival rate as stated above, or () () 2 () 1/ 1 ( ) 0 r aa EN FP PP λ λ ∂∂  =− − >  ∂∂ . Assume that the offering price follows normal distribution (, ) N µ σ , then higher variance ( 2 σ ) indicates fatter tail, or () 2 0 if () / 0 if P p Fp z P µ µ σϕ µ σ < >  − ∂∂=−  ><  , where 2 p z µ σ − = . So the impact of price dispersion on the time on the market is, () { } [] () ( ) {} 2 2 2 () 1/ 1 ( ) ( )/ = 1/ 1 ( ) 0 if = 0 if rr r r r r EN FP FP P zFP P P λσ σ µ ϕλ σ µ µ ∂  =− ∂ ∂  ∂ −  −−  <>   ><  (1.9) 12 The price dispersion facilitates a quicker sale of “overpriced” properties when the seller’s reservation price is higher than its market value, if they can be sold at all; on the contrary, it slows down the successful transaction of “under-priced” properties. This section has demonstrated a simple theoretical model of pricing strategy characterized with both reservation price and asking price. The hypotheses are (1) higher price dispersion leads to higher reservation price and higher asking price; (2) property’s sales price increases with offering price dispersion; and (3) for overpriced properties, higher offering price dispersion is associated with relatively shorter time on the market. I use Tokyo condominium resale market data to test above hypotheses 9 as shown in the following two sections. 3. Data Tokyo Condominium Market: Post-Economy-Bubble The empirical study is the condominium resale market in central Tokyo metropolitan area from 1994 to 2002, or the post-bubble period. By focusing on the post-bubble era, the market under study is exempt from the unusual speculation activities during the bubble economy. 9 For the first hypothesis, we did not conduct the direct test on reservation price. 13 (a) Year to year percentage change in population and net immigrations -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00 1994 1995 1996 1997 1998 1999 2000 2001 2002 Percentage population Net I mmigration ratio (b) CPI (2000=100) 90 92 94 96 98 100 102 104 106 108 110 1994 1995 1996 1997 1998 1999 2000 2001 2002 (c) Unemployment rate 2 3 4 5 6 1994 1995 1996 1997 1998 1999 2000 2001 2002 Percentage (d) Average monthly household income (yen) 530,000 540,000 550,000 560,000 570,000 580,000 590,000 600,000 1994 1995 1996 1997 1998 1999 2000 2001 2002 Figure 1.1 Demographic and economic indicators for Tokyo, 1994-2002 14 In the study period, Tokyo has experienced a mild population change within 1 percent every year (Figure 1.1(a)). Fluctuations in CPI are also limited (Figure 1.1(b)), so the case of price adjustment resulting from inflation could be excluded (Fishman, 1992; Diamond, 1993). The recession after the bubble economy was partially represented by the increasing unemployment rate (Figure 1.1(c)), which expedited since 1997, slowed down after 2001. This statistic was around 5 percent at the end of 2001, 2 percentages higher comparing to that of January 1994. The average monthly household income still increased stably until 1997 (Figure 1.1(d)), attributed to Japanese lifetime employment conventions; and then declined for about 10 percent by the end of 2001. An important institutional characteristic of Japanese property market is that the seller is in principle expected to make her property vacant before she posts it onto the market. That is to say that the seller is not able to sell her property if it is occupied by her or other tenants. Moreover, the Japanese existing housing market has relatively lower turnover rate comparing to the United States existing home market because of high cost. 10 Housing transaction is subject to a series of taxes accounted for around 2 percent of housing value, besides the capital gain tax. Government subsidized loans also discriminate against existing housing buyers with a lower limit on the absolute value of loans and by prohibiting lending to the buyers of units older than ten years. 10 [Kanemoto (1997)] The number of existing house transactions per household in Japan in 1992 was one-tenth that in the United States in the same year. 15 Submarkets, price dispersion, and market thickness Based on the properties’ key characteristics, we can divide the listed units into different submarkets. For the central Tokyo condominium market, the important features are: geographic area segment, if having train station within walking distance, unit’s size range, and age of the structure. Properties with sizes between 25 and 85 square meters are categorized as “family-type” condominiums, more than 80 percent of the listed properties fall into this category; those smaller than 25 square meters are mostly studios; while the units with size above 85 square meters are high-end condominiums. Age of the structure is another important characteristic. The homeowners who prefer new condominiums, called as “the new property runners”, tend to switch from current units, which are generally around 4 years old, to new properties. Because of depreciation, homeowners may have to pay higher maintenance for the structures older than 10 years; hence those units are expected to sell at lower prices. By local regulation, older structures (around 22 years) may be subject to demolishing and rebuilding, which have significant impact on the market price. Tokyo condominium transaction data Recruit Co., a publisher of information magazines on jobs, entertainment, automobiles and travel, also publishes Shukan Jutaku Jouhou (Weekly Housing Information). 11 This is a weekly magazine containing house/condominium sellers’ 11 Recruit Co. publishes Shukan Jutaku Jouhou in seven areas in Japan, including the Tokyo Metropolitan Area. 16 advertisements. The advertisements are classified into four categories: new detached houses, resale detached houses, new condominiums, and resale condominiums. Each advertisement includes the location and a brief description of the property, the asking price, and the name of the seller or the broker. The coverage of this Recruit Co. data set is comprehensive, especially for the resale condominiums. 12 According to Shukan Jutaku Jouhou, there were 91,037 condominium properties listed in central Tokyo metropolitan area from January 1994 to June 2002. During this period, 37,110 were sold, 51,442 were cancelled without sale, and 2,485 were still in the market (censored) till the end of June 2002 (Table 1.1). Each record contains the dates of listing and delisting, initial asking and delisting prices, ward, distance to major/minor train station, average access time to metropolitan sub-centers, unit size, top floor or ground floor, date of construction, real estate agent type (small, middle, or big), if structure is steel framed ferroconcrete, and if eligible of government financing, etc. Table 1.1 indicates that the average list price decreased slowly since 1994, but dropped considerably after 1997, 13 and touched the bottom around 2001. In line with the change in property market and the macroeconomics, the average time on the market increased slightly in 1997 and 1998. Then the slowdown in the decline of property prices helps shortening the duration in the market after 1999. As the 12 In the central Tokyo area (23 special wards), the Jutaku Tochi Toukei Chousa (Housing and Land Survey) of the General Administration Agency of the Japanese Government estimates that there were 9,333 transactions of condominium resale in 1998, while Recruit Co. reported 10,636. The difference may come from difference in definition, sampling errors, etc. 13 This change is consistent with the time of decrease in average income, which occurred after 1997. Please refer to Figure 1.1(d). 17 economy starts picking up, more units have been put into the market, and more are finally sold too. Table 1.1 Listing and delisting –yearly counts year Units listed Units de- listed Units sold Average initial list price (10 thousand yen, 2000) Average time on the market (week) 1994 9,435 7,096 3,033 3,866 13.21 1995 8,513 8,909 3,848 3,343 12.75 1996 9,841 9,296 4,385 3,295 12.31 1997 10,634 10,347 4,128 3,234 13.05 1998 10,518 10,430 3,886 3,180 13.15 1999 11,004 11,081 4,325 3,081 12.57 2000 11,887 11,587 4,692 3,087 12.56 2001 12,916 13,572 5,836 3,091 11.79 2002 6,289 6,234 2,977 3,273 7.36 Note: Statistics for 2002 end on June. I deleted the properties with extreme initial asking and delisting prices, that is, the 99 th and first percentile in the sample. Because this study is about the resale condominium market, new units are excluded. Since single family house market can behave quite differently from condominium market, all single family house observations are deleted. Besides that, less than 1 percent units have stayed in the market for over one year, or 52 weeks, these are excluded as well. Following the submarkets definition discussed above, 14 I divided the whole sample into 64 sub- samples. By constraining the minimum size of valid sub-sample as 25, less than 200 observations were dropped, or 21 sub-samples were left out. In our final dataset, 14 In the first part of this section, we discussed that central Tokyo condominium market could be divided into submarkets by several important features, including geographic area segment, if having train station within walking distance, unit’s size range, and age of the structure. 18 there are 84,352 observations, including 34,969 sold, 47,213 cancelled without sale, and 2,170 remaining in the market. 15 Figure 1.2 shows a histogram plot for the properties’ time on the market measured by the number of weeks. For the sold units, the time on the market follows lognormal distribution. The properties have the highest tendency to be sold at around 4 weeks after initial listing. 16 More than 90 percent of the units de-listed without sale are cancelled within the first 28 weeks, or within about half a year after initial listing. The sellers are more likely to cease their lists at around week 4 (end of first month), week 14 (after three months), and week 27 (after six months). Figure 1.2 Properties’ time on the market, by status 0 1000 2000 3000 4000 5000 6000 0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 Week Frequency Sold Cancelled without sale Censored Notes: Here only shows the units staying in the market up to 52 weeks. 15 There are 43 submarkets in the final sample. 16 The time required by the list information spreading to the public defers the sale. 19 According to Case and Quigley (1991), the important factors determining a property’s market value are structural characteristics, location, and the time of transaction. I use the sold units’ records, regress the transaction price on the property’s size and age, and its location measured by commuting time, for each month in our sample. Because units in small complex may have different market behavior compared to the units in big complex, I treat them as two subgroups. 17 Then at the second stage, I apply the coefficients estimated from the above regressions back to each listed property in the sample to obtain its estimated market value. The ratio between the list price and the estimated market value serves as the indicator of “overpricing”. Table 1.2 Mean and std. deviation of selected continuous variables All Sold Cancelled without sale Other No. of weeks on market 12.3 10.5 13.8 8.4 (9.4) (8.1) (10.0) (7.6) Initial list price (10 thousand yen) 3,259 3,101 3,366 3,465 (1,620) (1,499) (1,680) (1,914) List price at delisting (10 thousand yen) 3,174 3,021 3,287 4,451 (1,567) (1,454) (1,637) (2,818) 185 188 183 185 No. of months after construction when listed (93) (91) (94) (116) Size (square meters) 59.3 58.8 59.4 65.5 (19.8) (18.8) (20.3) (22.2) No. of observations 84,352 34,969 47,213 2,170 Note: Standard deviations are in parenthesis. Table 1.2 presents the means and standard deviations of selected continuous variables. The average duration on the market is 12.3 weeks, or around 3 months. 17 The regression results are in Appendix A. 20 The sold units have shorter duration which is on average 10.5 weeks, about 3 weeks less than the cancelled ones. The delisting price is around 2 percent lower than the initial asking price in general. Both the initial asking price and the final delisting price of the sold units are averagely 8 percent lower than the cancelled ones. The average age of the units is 188 months (15.4 years) at listing. The sold units are older compared to the cancelled ones by just 5 months. The average size is 59 square meters. Cancelled without sale properties have slightly bigger size, and those still in the market are the biggest among the three status groups. 18 Table 1.3 Mean and std. deviation of selected time varying covariates At initial listing All Sold Cancelled without sale Other Japanese Nikkei 225 index 16,660 16,754 16,843 11,153 (3,241) (3,251) (3,081) (437) 481,225 481,442 481,925 462,498 Average monthly household income (2000 yen) (9,643) (9,449) (9,126) (1,466) 1.134 1.137 1.14 0.949 Recruit residential (condominium) price index: Tokyo special district (23 wards) (0.18) (0.18) (0.18) (0.01) No. of observations 84,352 34,969 47,213 2,170 At delisting All Sold Cancelled without sale Other Japanese Nikkei 225 index 16,385 16,549 16,512 10,966 (3,337) (3,355) (3,190) Average monthly household income (2000 yen) 480,574 4809,31 481,154 462,207 (10,055) (9,900) (9,597) 1.116 1.122 1.119 0.947 Recruit residential (condominium) price index: Tokyo special district (23 wards) (0.18) (0.18) (0.18) No. of observations 84,352 34,969 47,213 2,170 Note: Standard deviations are in parenthesis. 18 The statistics for the “other” or the censored lists indicate that as the market starts picking up, the more recent lists are relatively bigger and more expensive. 21 I merged the transaction records with city level macroeconomic indicators and condominium market information. Table 1.3 lays out the means and standard deviations of selected time varying covariates. Due to the economy slowing-down in our sample period, Japanese Nikkei 225 index is on average higher at initial listing compared with the index at delisting. This difference is more significant among the cancelled without sale properties. The average monthly household income was also higher at listing than at de-listing following the macroeconomic trend. The sold units experienced lower drop in this indicator than the cancelled lists (510 yen vs. 771 yen). Condominium price index is also generally higher at initial listing than at delisting. Figure 1.3(a) Standard deviation in sales price vs. average sales price (submarkets) 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 1000 2000 3000 4000 5000 6000 7000 8000 Average sales price Std Dev. in sales price 22 In order to find the submarkets with higher price variance, I calculated the average sale price and sale price standard deviation for each submarket. Figure 1.3(a) is the scatter plot of these two variables. The observed pattern is that more expensive markets usually have higher variance in the sale price, with a log-linear relationship. 19 Hence the observed variance in the submarket can be decomposed into two components: one is the “predictable variance” associated with the average price, another is the additional “abnormal variance” which cannot be explained by the average price level. I define “market with higher price dispersion” as the submarkets with observed sale price variance higher than the expected value calculated from the average sale price level. Figure 1.3(b) Number of lists vs. number of transactions (submarkets) 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 Number of lists Number of transactions 19 The regression results (for submarkets) is: Sales price std. dev = -5,155.74 + 759.68×ln(Average sales price) (Adj. R 2 = 0.88) (t) (17.81) 23 Figure 1.3(b) is the scatter plot of number of lists and number of transactions from all the submarkets. The two variables display a linear relationship. As mentioned earlier in this section, market thickness could be represented by the number of transactions. When the number of lists is perfectly correlated with the number of transactions, the former can be also used as a proxy for the market thickness. I define the submarkets with more than 1,000 lists as thick markets. Table 1.4 provides the frequencies of selected discrete variables. Area I, Area II, and Area III present 26.8 percent, 44.3 percent, and 28.9 percent of the central Tokyo condominium market separately. The transaction rate is the lowest in Area I (36.4 percent), and highest in Area III (41.2 percent). Majority of the listed properties have train station within walking distance (98.6 percent), but they also have lower transaction rate compared to the rest (41.4 vs. 48.0 percent). 33.5 percent of the sellers ever adjusted their list price; among them, 97.9 tend to decrease the list price, while only 2.1 percent tend to increase the list price. 20 The transaction rate for the price adjusted properties is slightly lower than those without any price adjustment units (40.8 vs. 41.8 percent). 20 As Horowitz (1992) mentioned, the increase in list price rarely happens. It may happen when the seller obtain multiple offers at the same time. Under that condition, auction theory should be applied. 24 Table 1.4 Frequency of selected discrete variables All Sold Cancelled without sale Other Area Area I - Central Business 22,609 8,234 13,711 664 (26.8) (36.4) (60.6) (2.9) Area II - Southwest 37,350 15,382 21,009 959 (44.3) (41.2) (56.2) (2.6) Area III - Northeast 24,393 11,353 12,493 547 (28.9) (46.5) (51.2) (2.2) Train station within walking distance 83,165 34,399 46,630 2,136 (98.6) (41.4) (56.1) (2.6) Ever adjust list price 28,217 11,502 14,546 2,169 (33.5) (40.8) (51.6) (7.7) Ever adjust list price - increase 584 278 278 1 (0.7) (47.6) (47.6) (0.2) Ever adjust list price - decrease 27,633 11,224 14,241 2,168 (32.8) (40.6) (51.5) (7.8) Market with higher price dispersion 72,409 30,311 40,360 1,738 (85.8) (41.9) (55.7) (2.4) Thick market 78,079 32,491 43,644 1,944 (92.6) (41.6) (55.9) (2.5) Real estate agent category Big 52,647 19,625 31,642 1,380 (62.4) (37.3) (60.1) (2.6) Middle 15,580 7,490 7,739 351 (18.5) (48.1) (49.7) (2.3) Small 16,125 7,854 7,832 439 (19.1) (48.7) (48.6) (2.7) Delisting season Spring 22,381 9,575 12,729 77 (26.5) (42.8) (56.9) (0.3) Summer 21,846 9,445 12,382 19 (25.9) (43.2) (56.7) (0.1) Fall 19,965 7,478 10,850 1,637 (23.7) (37.5) (54.3) (8.2) Winter 20,160 8,471 11,252 437 (23.9) (42.0) (55.8) (2.2) Size (square meter) Less than 25sq.m. 2,436 849 1,548 39 (2.9) (34.9) (63.5) (1.6) 25-85sq.m. 74,527 31,470 41,261 1,796 (88.4) (42.2) (55.4) (2.4) More than 85sq.m. 7,389 2,650 4,404 335 (8.8) (35.9) (59.6) (4.5) Structure age 1-4 years 8,093 2,764 4,983 346 (9.6) (34.2) (61.6) (4.3) 4-10 years 13,445 5,452 7,545 448 (15.9) (40.6) (56.1) (3.3) 10-22 years 45,805 19,696 25,334 775 (54.3) (43.0) (55.3) (1.7) More than 22 years 17,009 7,057 9,351 601 (20.2) (41.5) (55.0) (3.5) No. of observations 84,352 34,969 47,213 2,170 Note: Column percentages are in parenthesis for “all” in column 1; row percentages by “sold”, “cancelled without sale”, and “other “are in parenthesis in column 2-4. 25 Following our definition of market with higher price dispersion as those with higher observed sale price variance than the expected variance estimated from the average sales price level, 85.8 percent lists fall into this category. 41.9 percent of the lists in the submarkets with higher price dispersion are sold, while only 39 percent of the rest are sold. Taking the submarkets with more than 1,000 lists as thick market, 92.6 percent lists are in this category. The transaction rate is higher in the thick submarkets than the rest (41.6 vs. 39.5 percent). The small and middle size agents have less than 40 percent of the market share, but manage to sell almost half of their lists, which is more than 10 percent higher than the transaction rate of the big agents’. More properties are listed in spring and summer than in fall and winter. Fewer transactions are completed in fall than in other seasons. Majority of the lists are “family-type” condominiums, with size between 25 and 85 square meters. They also have relatively higher transaction rate compared to both studios and luxury units (42.2 vs. 34.9 and 35.9 percent, separately). More than half of the listed properties are between 10 and 22 years old, which also have highest transaction rate (43 percent). The newest group, between 1 and 4 years old structures, presents the lowest transaction rate (34.2 percent). 4. Empirical Results The theoretical model shows that higher price dispersion leads to higher reservation price and higher asking price, which in turn results in higher sales price 26 (equation (1.4), (1.5), and (1.7)). For overpriced properties, higher offering price dispersion is associated with relatively shorter time on the market (equation (1.9)). List prices at initial listing and delisting Although the reservation price is not observable, list price is widely documented. According to equation (1.3), the selection of optimal list price is related to the market condition (including offering price distribution, offer arrival rate, and cost of search) and the expected return from search. Equation (1.5) shows the optimal list price increases with price dispersion under certain discount rate constraint, such as 0.5 1 discount rate between two offers ≤≤ . As an important outcome of search, the expected sales price defined in equation (1.6) is also shown increasing with price dispersion (equation (1.7)). The following analysis examines if the list price and the sales price are higher with higher price dispersion. Both list price and sales price is closely related to the property’s fundamental value as estimated quality adjusted market price. Besides that, the market conditions such as price dispersion and market thickness, together with the seller’s behavior like if she ever adjusts list price and which real estate agent she selects can all have impact on the list price selection and the final sales price. Table 1.5 lays out the OLS regression results of the list price for the full sample. Panel (a) shows the results at initial listing, and panel (b) shows the results at delisting. Model 1 in panel (a) focuses on the quality adjusted price estimated from the previous step. The property’s fundamental value is highly significant and 27 positively related to the initial asking price. Model 2 adds in price dispersion by looking at the sales price standard deviation in the corresponding submarket. It is also significant. 1 percentage increase in the sales price standard deviation leads to 0.12 percentage increase in the initial list price. This pattern has been kept throughout the listing period. It confirms our hypothesis that the asking price is higher in the market with greater price dispersion. Table 1.5 OLS regressions on price for the full sample (a) Dependent variable - logarithm of list price, at initial listing Model 1 Model 2 Model 3 Model 4 Model 5 intercept 0.17 -0.75 -0.81 -0.80 -0.77 (11.32) (-43.21) (-46.15) (-45.48) (-43.63) 0.98 0.89 0.91 0.91 0.90 Logarithm of predicted price at listing (536.31) (441.87) (423.17) (421.79) (419.10) 0.12 0.11 0.11 0.11 Logarithm of sales price dispersion in the corresponding submarket (88.52) (77.69) (77.96) (78.13) Thick market 0.06 0.06 0.06 (21.62) (21.35) (20.90) Ever adjust list price 0.03 0.03 (20.55) (19.57) Small size agent -0.03 (-14.72) Middle size agent -0.02 (-7.87) R square 0.774 0.793 0.794 0.795 0.796 Number of observations 84156 84156 84156 84156 84156 As Lazear (1986) pointed out that prices are more or less depending upon the thinness of the market, model 3 also looks at the thick market effect. After controlling both market value and price dispersion, model 3 reveals that sellers tend to take the advantage of the liquidity in the thick market by pricing their property 28 0.06 percent higher. 21 Model 4 includes seller’s marketing behavior. Some sellers tend to post a higher experimental price at the beginning and adjust it later on when they do not understand the market that well (Taylor, 1999; Chade and Serio, 2002; Sass, 1988). Model 5 also considers the effect of different types of real estate agents. Both small size and middle size agents are less aggressive than the big size agents. Table 1.5 OLS regressions on price for the full sample (b) Dependent variable - logarithm of list price, at delisting Model 1 Model 2 Model 3 Model 4 Model 5 intercept 0.08 -0.78 -0.85 -0.86 -0.83 (5.48) (-44.59) (-47.65) (-48.30) (-46.64) 0.99 0.91 0.92 0.93 0.92 Logarithm of predicted price at delisting (531.94) (437.62) (419.51) (420.94) (418.12) 0.11 0.10 0.10 0.10 Logarithm of sales price dispersion in the corresponding submarket (83.75) (72.73) (72.91) (73.09) Thick market 0.07 0.07 0.07 (22.43) (22.78) (22.39) Ever adjust list price -0.03 -0.03 (-19.69) (-20.55) Small size agent -0.02 (-12.83) Middle size agent -0.01 (-6.19) R square 0.775 0.793 0.794 0.795 0.796 Number of observations 82013 82013 82013 82013 82013 The results in panel (b) are quite similar to those in panel (a). The delisting price is still positively related to the estimated quality adjusted fundamental market value, the submarket’s price dispersion, and thick market indicator. But the ex ante less informed sellers who “post high and adjust later” are not necessarily leaving the 21 Thick market is associated with higher offer arrival rate, which in turn leads to a higher reservation price in equation (2). When we assume offer arrival rate is a linear function of asking price, i.e. * a P λ ∂ ∂ is a constant, higher officer arrival rate is related to higher asking price by equation (3). Under this set of assumptions, we expect to observe higher asking price and higher sales price in the thick market. 29 market with a higher price. Again, units listed with big size agents may be also de- listed at a higher price. As mentioned above, only around 41 percent listed properties are finally sold. Table 1.6 shows the results of the same set of regressions but only on the sold units. Most of the factors have similar effects compared to the regression results as for the full sample, including the impact of estimated quality adjusted market value, the submarket’s sales price dispersion, thick market indicator, and if the seller understands market well enough. It confirms our hypothesis that higher sales price is associated with higher price dispersion. 1 percentage increase in the sales price standard deviation can lead to about 0.10 percentage increases in the sales price. Table 1.6 OLS regressions on price for the sold units (a) Dependent variable - logarithm of list price, at initial listing Model 1 Model 2 Model 3 Model 4 Model 5 intercept 0.15 -0.83 -0.91 -0.89 -0.89 (6.53) (-30.59) (-33.10) (-32.36) (-32.16) 0.98 0.90 0.91 0.91 0.91 Logarithm of predicted price at listing (342.44) (288.51) (277.54) (275.95) (273.63) 0.12 0.11 0.11 0.11 Logarithm of sales price dispersion in the corresponding submarket (59.47) (52.58) (53.00) (53.09) Thick market 0.07 0.07 0.07 (16.30) (15.88) (15.80) Ever adjust list price 0.04 0.04 (16.65) (16.59) Small size agent 0.00 (-1.08) Middle size agent 0.01 (2.93) R square 0.771 0.792 0.793 0.795 0.795 Number of observations 34895 34895 34895 34895 34895 30 Table 1.6 OLS regressions on price for the sold units (b) Dependent variable - logarithm of list price, at delisting Model 1 Model 2 Model 3 Model 4 Model 5 intercept 0.00 -0.90 -0.98 -0.99 -1.00 (0.01) (-33.75) (-36.31) (-36.84) (-36.82) 1.00 0.92 0.93 0.94 0.94 Logarithm of predicted price at delisting (353.68) (296.93) (285.76) (286.43) (284.17) 0.11 0.10 0.10 0.10 Logarithm of sales price dispersion in the corresponding submarket (56.75) (49.81) (49.87) (49.93) Thick market 0.07 0.08 0.08 (16.70) (16.99) (16.98) Ever adjust list price -0.03 -0.03 (-11.79) (-11.67) Small size agent 0.00 (0.63) Middle size agent 0.01 (4.14) R square 0.782 0.800 0.802 0.803 0.803 Number of observations 34895 34895 34895 34895 34895 For the sold units, the middle size agents outperform the small and big size agents with significantly higher sales price. Although the big agents have greater client base, they do not necessarily understand the local market that well. Consequently, they are likely to persuade their clients “posting high and adjusting later”. Unfortunately, the possible outcome is that their clients have to stay longer in the market (as shown later in this section) without guarantee of a better price. 22 Time on the market: likelihood of sale Property’s time on the market is another important outcome of search. Equation (1.8) expresses the expected time on the market as a function of offer arrival rate, seller’s reservation price, and the offering price distribution. If assume 22 Sirmans et. al. (1995) showed that quick sold houses’ prices are not significantly different from those of the houses stay in the market for a considered normal marketing time. 31 the offering price follows normal distribution, equation (1.9) shows that the expected time on the market for an overpriced property with reservation price higher than its market value can be sold faster in the market with higher price dispersion, if it can be sold at all. When we assume that most of the mis-priced properties are tended to be actually overpriced, then averagely shorter time on the market or higher probability of sale should be expected in the markets with higher price dispersion. Our empirical models are estimated based on the Cox Partial Likelihood approach (Cox, 1975). The hazard function of the Cox model is defined as the product of a baseline hazard function and a set of proportional factors such that () () ( ) ( ) ( ) 0 ;exp', ij j i j j ij j ij j ht z t h t z t β = j = 1, 2, (1.10) 23 where () 0jij ht is a baseline hazard function that describes the overall shape of the time on the market of the listed properties, i.e., list termination risks by sale or cancellation. j indicates sale (if j = 1), withdraw without sale or censored lists (if j = 2, or 3). The hazard rate of termination is the probability that a property sells at any given time t , given that it has not sold or cancelled prior to t . () jij zt is a vector of proportional factors capturing time-varying or time- invariant covariates. In our empirical example, ( ) jij zt includes (1) the logarithm of 23 Green and Shoven (1986) are among the first to apply it to study mortgage prepayments due to interest rate movements. Since then, researchers have developed more sophisticated and realistic applications of the Cox proportional hazard model to study mortgage termination behaviors (See Schwartz and Torous, 1989, and Deng et al., 2000, for more recent applications). In housing market analysis, Zuehlke (1987) employed a Weibull hazard model to examine the relationship between probability of sale and market duration in housing markets. Kluger and Miller (1990) tried to develop a liquidity measure for real estate based on the Cox proportional hazard technique. Genesove and Mayer (2003) used the Cox proportional hazard model and find that the sellers subject to nominal losses will stay longer on the market. 32 list price, serving as the determinant of offer arrival rate; (2) degree of market dispersion, represented by the indicator of “market with higher price dispersion”, defined as a dummy variable to indicate if the submarket has abnormal price variance than expected from its average price level; 24 (3) seller’s behavior such as the degree of overpricing, and if she ever changes the list price; (4) real estate agent type, representing different search intensity, and the extent of local market knowledge; and (5) economic and housing market conditions, such as thick market indicator, month/year of delisting, stock exchange index, average household income, house price index, etc. Most of those factors can affect offer arrival rate, and in turn influence property’s time on the market. Table 1.7 presents estimates of six different models. In model 1, as expected, the logarithm of list price is significant and negative. It is consistent with the previous discussions: higher the list price, longer the time on the market, or the lower likelihood of sale. Following Leung et al. (2002) and Anglin et al. (2003), both the list price and the deviation of list price from the property’s market value determine time on the market. Model 2 focuses on the effect of the overpricing indicator, i.e. the ratio between list price and the estimated hedonic price. As expected, this indicator is negative and significant. For an overpriced property, it has to stay longer in the market before sale if it could be sold. 24 In section 3, we showed that the standard deviation of sales price and the average sales price should follow log-linear relationship. 33 Table 1.7 Time on the market: proportional hazard models Dependent variable: number of weeks in the market Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 -0.38 -0.19 -0.17 -0.16 -0.10 Logarithm of list price (-33.45) (-14.70) (-13.14) (-12.09) (-8.08) List price/hedonic price -0.97 -0.80 -0.83 -0.86 -0.83 (-42.78) (-31.75) (-32.35) (-32.96) (-31.67) 0.12 0.06 0.07 Market with higher price dispersion (7.73) (3.16) (3.59) Thick market 0.17 0.19 (6.92) (7.66) Ever adjust list price -0.65 (-55.72) -Log (likelihood) 368496 368082 367972 367941 367917 366278 SBC 368502 368087 367982 367957 367938 366304 Note: t-statistics are in parenthesis. Model 3 studies the effects of the list price and the overpricing indicator simultaneously. Previous relationships are both held. Model 4 further includes the indicator of whether submarket has abnormal price variance than expected from its average price level. 25 If most of the sellers overprice their properties if they do not price at the market value, according to equation (1.5), the submarket with higher price dispersion will help them to sell faster. The “market with higher price dispersion” dummy is significant in model 4. The properties listed in the market with higher price dispersion are more likely to be sold. It is consistent with our theoretical discussion. Model 5 extends model 4 by including market thickness indicator. Thick submarket, which has more sellers and buyers, provides higher liquidity to the 25 Since the expected variance which associated with the average price in the submarket is implied by the list price. In another word, the variable, “logarithm of list price” assumes both price effect and the expected submarket price variance effect. The inclusion of “market with higher price dispersion” dummy is the proxy of higher abnormal variance in the submarket. 34 properties by increasing the offer arrival rate. This indicator is positively related to the likelihood of sale. Model 6 adds the indicator of if the sellers “ever adjusted list price”. Those “listing high adjusting later” sellers may experience difficulty in selling due to lower offer arrival rate. They have to reduce the list price if they want to sell. On the other hand, it is likely that the other sellers who understand the market better tend to price their properties more realistically in order to ensure high enough offer arrival rate. Those sellers do not need to adjust their list price that often but rather hold a constant list price. The indicator of if the sellers “ever adjusted list price” is negatively related to the likelihood of sale. Table 1.8 lists another set of extended proportional hazard model estimates. Model 7 adds in the effects of different types of real estate agents categorized by the size. Listing with middle size agent has the highest likelihood of sale, while listing with big size agent may face the longest time on the market. Model 8 also considers the seasonality factor. Properties are more likely to be sold in March, June, and November, and the least likely to be sold in January. 35 Table 1.8 Time on the market: proportional hazard models Dependent variable: number of weeks in the market Model 7 Model 8 Model 9 Model 10 -0.10 -0.10 -0.13 -0.18 Logarithm of list price (-7.64) (-7.78) (-9.97) (-13.52) List price/hedonic price -0.82 -0.82 -0.81 -0.52 (-31.38) (-31.23) (-30.38) (-19.17) 0.06 0.06 0.04 0.05 Market with higher price dispersion (3.43) (3.44) (2.11) (2.95) Thick market 0.19 0.19 0.21 0.15 (7.67) (7.66) (8.52) (6.14) Ever adjust list price -0.64 -0.64 -0.64 -0.65 (-54.71) (-54.16) (-54.16) (-54.69) Small size agent 0.07 0.07 0.06 0.06 (5.16) (5.31) (4.68) (4.79) Middle size agent 0.18 0.18 0.17 0.18 (13.00) (12.90) (12.85) (13.34) De-list month February 0.26 0.26 0.26 (9.82) (10.04) (9.82) March 0.34 0.34 0.36 (13.08) (13.24) (13.99) April 0.28 0.28 0.30 (10.58) (10.61) (11.10) May 0.27 0.27 0.27 (10.06) (9.89) (10.07) June 0.32 0.31 0.32 (12.22) (11.57) (12.06) July 0.21 0.21 0.23 (7.44) (7.35) (8.13) August 0.14 0.13 0.12 (4.92) (4.55) (4.20) September 0.16 0.15 0.15 (5.83) (5.29) (5.31) October 0.23 0.22 0.21 (8.66) (8.23) (7.97) November 0.31 0.30 0.27 (11.33) (10.81) (9.83) December 0.23 0.22 0.19 (8.09) (7.77) (6.64) Note: t-statistics are in parenthesis. 36 Table 1.8 Time on the market: proportional hazard models (Cont.) Dependent variable: number of weeks in the market Model 7 Model 8 Model 9 Model 10 De-list year 1995 -0.17 (-6.85) 1996 -0.05 (-1.96) 1997 -0.27 (-11.07) 1998 -0.40 (-15.92) 1999 -0.34 (-13.77) 2000 -0.27 (-11.26) 2001 -0.21 (-8.79) 2002 -0.13 (-4.59) After 1997 -0.56 (-28.79) Logarithm of Nikkei 225 Index -0.27 (-6.73) -1.51 Logarithm of average monthly household income (2000 yen) (-3.78) -0.93 Recruit residential condominium price index: Tokyo special district (23 wards) (-17.02) -Log (likelihood) 366194 366055 365812 365132 SBC 366231 366149 365948 365247 Note: t-statistics are in parenthesis. Model 9 further includes the delisting year indicator to capture the general trend in condominium market from 1994 to 2002. Japanese bubble economy, which is directly related to real estate market, burst in 1991. In the following two years, Japanese people experienced a painful economic hard-landing. The economy slowdown mitigated from 1993, but the 1997 Asian Financial Crisis worsened the economic condition again even before it reached the picking up point. From 2001, 37 the economy finally started to show some signs of improvement. The condominium resale market followed the ups and downs of the macro economy as other aspects in the society. The healthier macroeconomic condition is associated with stronger economic activities, including all sorts of investments, which implies a higher potential buyers’ arrival rate, λ , as in equation (1.5). Instead of using the simple delisting year dummies, Model 10 explicitly includes the time-varying macroeconomic and market condition indicators, including Nikkei 225 index, average monthly household income, and condominium price index for Tokyo special district. We treat Tokyo house market before and after 1997 Asian Financial Crisis differently. Stock market is a competitor of housing market for the limited capital in the society, hence less housing investment can be observed with bull stock market. To buy an existing house may not be the top choice for most of Tokyo residents if they have enough funding. When the average income increases, more buyers will shop for new properties or even single family houses rather than to purchase existing homes. In general, when housing becomes more expensive, the likelihood of sale decreases due to affordability issue. 5. Conclusions Economists have noticed price dispersion for a long time. It is the price dispersion that requires selling strategy, which is composed of the reservation price and the asking price. Recent literature acknowledged the importance of asking price. 38 However, very limited attention so far has been paid to the impact of price dispersion per se. In our multistage search model, reservation price is the unique price making the seller indifferent between accepting and declining the current offer. It serves as the floor of buyers’ offering price and final sales price. This reservation price is determined by the opportunity cost of sale, search cost, seller’s discount rate, and the market conditions such as the offer arrival rate and the offering price distribution. On the other hand, the optimal asking price is chosen to maximize the return from search. Rather than being related to the reservation price, the selection of optimal asking price is related to the market conditions (offer arrival rate, offering price distribution, and cost of search), seller’s discount rate, and the expected return from search. Moreover, the asking price serves as the ceiling of buyers’ offering price and the final sales price. The higher price dispersion leads to higher reservation price and higher optimal asking price, which in turn results in higher sales price. If the offering prices follow normal distribution, price dispersion also facilitates quicker sale of overpriced properties with reservation price higher than the market value. In reality, sellers have different reservation prices due to different reasons, such as various opportunity costs and discount rates. Those different reservation prices realize themselves as a range of sales prices and different time on the market, which reinforce the existence of price dispersion. 39 Our example on Tokyo condominium resale market provides empirical evidence for the theoretical hypotheses. 1 percentage increase in the submarket’s sales price standard deviation can lead to 0.11 and 0.10 percent increase in the initial list price and the final sales price separately. Although the overpriced properties will stay in the market longer with lower probability of match due to higher reservation price or/and lower offer arrival rate as the result of higher list price, market price dispersion can increase the probability of successful transaction and/or facilitate quicker sale if there is a successful transaction in the end. Moreover, the sellers who do not understand the market well enough are more likely to list their properties significantly higher at the beginning, and reduce the list prices later on. They stay in the market longer and sell their properties about 3 percent lower than those who have better knowledge of the market. Therefore, those measures helping the agents understanding the market better can effectively improve market efficiency. 40 Chapter 2 An Early Assessment on Residential Mortgage Termination in China 1. Introduction The first residential mortgage loan in China was issued by the China Construction Bank (CCB) in 1986. During the next twelve-year period, mortgage market in China grew very slowly. By the end of 1997, total outstanding mortgage balance in China was only around RMB Yuan 22 billion. In 1998, the State Council of the People’s Republic of China published several administrative laws for broadening housing reform and expediting housing construction. Residential mortgage lending began to expand at an accelerating rate since 1998 in line with reforms aiming to end state-controlled welfare housing system. 26 In 1999, China's residential mortgages loans to individual households exceeded RMB Yuan 126 billion, doubled the previous year’s level. By August 2002, total outstanding balance of the residential mortgages reached RMB Yuan 763 billion, increased by 27 percent compared to the balance at the beginning of 2002, 34 times compared to the balance at the end of 1997. More than half of the newly issued real estate loans during 2002 are residential mortgage loans. 27 The residential mortgage market becomes a 26 Prior to 1998, over ninety percent of the urban residential housing units in China were developed and owned by state-owned enterprises. These housing units were leased to the employees at very low rent as part of welfare for the state-run enterprises employees’ and collective owned enterprises employees’. Under the PRC State Council 1998 Administrative Law, state-owned enterprises will no longer be allowed to allocate welfare housing to employees after December 31, 1999. 27 See People’s Daily, 11/22/2002. 41 financial engine for the booming residential housing development and sustained economic growth in China. Recently, there are active debates among policy makers, scholars and experts in the banking industry about the necessity and feasibility of developing mortgage- backed security (MBS) market in China. One of the key feasibility conditions for setting up a MBS market is the ability to manage the duration risks of mortgages. It is well known that mortgage loans are exposed to prepayment and default risks, which in turn create uncertainty about the duration of the securities backed-up by these mortgage loans, and hence create difficulty in pricing MBS. A reasonable understanding in mortgage initiation and duration is the prerequisite of MBS market development. Despite the rapid growth of the residential mortgage market and potential of developing MBS market in China, there have been very limited empirical studies on the performance of this newly developed and important financial market sector in China, largely attributing to the immature regulatory environment and sparse mortgage data. This is one of the first rigorous empirical studies about the residential mortgage choice and performance in China based on a unique micro data set of residential mortgage loan history collected by a major residential mortgage lender in China. This study addresses following questions: (1) What are major determinants to Chinese borrowers’ prepayment and default decision? (2) Who are the high risk borrowers in the Chinese residential mortgage market? and (3) To what extent 42 efficiency and/or equity of the current residential mortgage market in China may be improved? Unlike the developed market, the “option theory” 28 does not play a significant role in determining mortgage prepayment and default in China. On the other hand, financial factors, macroeconomic environment, and risk sharing mechanism are crucial to borrowers’ decision. For example, equity position, stock market investment opportunities, household income, consumers’ expectation, and the possible construction period risks are among the major determinants driving mortgage prepayments in China. Moreover, borrowers’ characteristics, such as borrower’s age, occupation, job position and education can serve as important indicators to separate high risk borrowers from the low risk population. Borrowers’ initial choices on mortgage amount/loan to value ratio and loan term reveals the possible premature termination in the mortgage lifetime. Introducing a risk-based pricing to the residential mortgage market can improve the efficiency of the mortgage market and enhance the mortgage credit availability to the much needed population, such as young and lower-income households and blue-collar workers in China. The remaining of the paper is organized as following: section 2 discusses the institutional background of mortgage prepayment and default in China, section 3 28 The “option theory” developed by Black and Scholes (1973) and Merton (1973) has been adopted widely to explain mortgage prepayment and default risks in the United States and other developed countries. A detailed discussion of the option theory and its application to mortgage valuation will be discussed in the following section. 43 reviews related literatures and discusses the econometric model; section 4 describes the mortgage loan data set and section 5 discusses estimation results. Conclusions and policy implications are discussed in section 6. 2. Current Residential Mortgage Market in China The current residential mortgage market in China is dominated by four major lenders – Industrial and Commercial Bank of China (ICBC), China Construction Bank (CCB), Bank of China, and Agricultural Bank of China. By the end of 2002, ICBC’s total outstanding balance was RMB Yuan 258 billion, which accounted for about 36% of the market share. Currently there are three categories of residential mortgages in China – individual account housing loans, authorized housing loans, and combined housing loans. Individual account housing loans refer to loans granted to individual buyers to facilitate housing purchases. The funding sources of the individual account housing loans are the credit funds of the bank. Authorized housing loans refer to loans to individuals who buy ordinary houses granted by the bank with the authorization of the public reserve fund management department, according to the prescribed requirements, and with the public reserve deposits as the source of funds. A combined housing loan refers to a loan granted by the bank to the same borrower to facilitate the purchase of an ordinary house for self use. The funding sources of combined housing loan include both the public reserve deposit and the credit funds. 29 29 Refer to Bank of China documents for a more detailed discussion. 44 Basic requirements. The mortgage loan amount shall not exceed 80% of the evaluated value or the purchase price of the house, and payment to income ratio should not exceed 70%. Applicants should provide documents for other assets including tax return, bank statement on saving’s account, proof of vehicle ownership and its value, proof of stock market investment, and proof of other property value. Total other assets to loan ratio should be greater than or equal to 25%. The mortgage term shall not exceed 30 years for a RMB mortgage loan, and borrower’s age plus mortgage term should not exceed 65 years. Guarantees. The lender determines the types of guarantee for a housing loan according to the specific conditions, which may include: collateralized through the property of the borrower or the property of a third party; guaranteed by a third party with joint asset account; combined guarantee; and the borrower purchasing commercial credit insurance. Mortgage interest rate and payment. Mortgage interest rates are determined by the People’s Bank of China. Starting from June 19, 1999, mortgage rates for all long term mortgages (loan term is greater than 5 years) should follow 6-month bank legal lending rate without floating range set by the People’s Bank of China. The spread between the long term (more than 5 years) and short term (5 years or less) mortgage rates should be 27 basis points. If a new interest rate is published by the People’s Bank of China, the mortgage rates will be adjusted starting from the first of January in the following year. Repayment may be made by equal installments or by progressive installments. 45 An important institutional characteristic of current Chinese mortgage market is that all the loans with the same mortgage maturity have the same mortgage rate, which is unrelated to LTV ratio and loan amount. Therefore, the mortgage’s real cost is essentially determined by the choice of maturity. Loan Application Procedure. Applicants fill out an application form and provide relevant documents; the bank carries out eligibility investigation. Upon approval, the bank and the borrower sign a mortgage contract. The borrower then opens a mortgage account for mortgage payment. Most Chinese are reluctant to have a debt. There is an old Chinese saying “free of debt reduces your burden”. As reported by Beijing City Survey Organization, there are more than 75 percent residents who are aware of the existence of personal loans. Among them, less than 10 percent ever applied for a loan. Among the 48 applicants in the survey, 30 applied for mortgage loans, and 16 applied for credit card loans, and 8 applied for auto loans. The reasons of loan application are: the convenience of life or work (56.5%); personal loan is one kind of investment (34.8%); enjoying life on credit out of confidence about future financial status (26.1%); accepting this new lifestyle (34.8%); need for credit (21.7%). 30 The motivation of prepayment in China is quite different from that in the United States or other developed countries. All residential mortgages in China are adjustable rate mortgages (ARMs). Once the Central Bank (the People’s Bank of China) announces a rate adjustment, this new rate will be applied to all existing mortgage loans (with term longer than one year) starting from the beginning of the 30 See Beijing Municipal Bureau of Statistics, September 2002 report. 46 following year. Due to the processing time and the immaturity of the mortgage market, virtually all prepayments observed in the sample are early payoff rather than refinance. However, another motivation related to the presale properties cannot be neglected because of the popularity of presale practice in the housing market in China. Many Chinese mortgage borrowers use mortgage as an instrument to share the construction period risk with the bank. If the developers eventually deliver the properties according to the presale contract, some borrowers might choose to pay off the debts as soon as they can. On the contrary, in case the developer fails to deliver the properties, the borrowers will default the loans. In other words, the mortgage borrower has a put option to sell the poorly constructed house to the bank at a price set by the remaining balance of the loan. The number of default cases in China is quite small. To most Chinese, purchase a dream home financed by mortgage is not only an investment, but also a necessity of life. Though for some borrowers who do not have stable income (such as those self-employed businessmen), and who purchase the house for investment purpose, shocks in the housing market or changes in mortgage rate will affect their ability of making on-time payment. However, majority of the default case occurred in the residential mortgage market in China are related to presale property where developers fail to deliver the housing units according to the presale contract. In such case, lender takes the loss if the value of the housing units recovered by the bank is less than the outstanding loan balance. 47 3. Option Theory and the Proportional Hazard Model There has been a large volume of studies focusing on risks and performance of mortgage lending in the United States. The existing literature on economic behavior of residential mortgage borrowers has reached consensus at least in the following two areas: first, the option theory developed in the finance literature (Black and Scholes, 1973, and Merton, 1973) provides an important theoretical framework to analyze mortgage borrowers’ prepayment and default behavior in the United States; second, the proportional hazard model developed by Cox (Cox, 1972) provides an important analytic tool for analyzing the dynamics and duration of mortgage termination by prepayment and default. Findley and Capozza (1977), Dunn and McConnell (1981), Buser and Hendershott (1984) and Brennan and Schwartz (1985) were among the first to apply option theory to the mortgage valuation. Since then, the option theory becomes the predominant theoretical framework in analyzing mortgage borrower’s prepayment and default behavior in the United States. According to the option theory, in the absence of transaction costs, a rational borrower can maximize her welfare by refinancing her mortgage when the call (prepayment) option is “in-the-money” (that is when the prevailing market rate of mortgage drops below the existing mortgage coupon rate). Similarly, a borrower should default the mortgage loan if the put (default) option is “in-the-money” (that is when the current market value of the house, which serves as collateral of the mortgage debt, drops below the current 48 market value of the remaining mortgage balance). Hendershott and Van Order (1987) and Kau and Keenan (1995) provided thorough surveys on these theoretical literatures. Schwartz and Torous (1989), Deng (1997) and Stanton and Wallace (1999) among others demonstrated empirically the importance of the financial option values to borrowers’ exercising of prepayment or default options based on historical pool or loan level mortgage data in the United States. These empirical literatures found strong evidence that the market value of the call (prepayment) option is statistically significant and positively associate with mortgage termination by refinance; and the market value of the put (default) option is statistically significant and positively associated with mortgage default risk. The empirical literature also found that mortgage borrowers may not ruthlessly exercise the prepayment or default options as predicted by the option theory. In other words, other non-financial option related factors, such as transaction costs of refinance, borrower’s credit worthiness, household’s income and wealth, unemployment risks, divorce rates, etc., also serve as important determinants to trigger or deter the borrowers’ decision on prepayment and default. (See Stanton, 1995, Quigley and Van Order, 1995, for discussions on impacts of transaction costs and trigger events in mortgage prepayment exercise.) Green and Shoven (1986) are among the first to apply the Cox proportional hazard model to study the effect of interest rates on mortgage prepayment. Since then, researchers have developed more sophisticated and realistic applications of the 49 Cox proportional hazard model to study mortgage termination behaviors (See Schwartz and Torous, 1989, Deng, Quigley and Van Order, 2000, and Deng and Quigley, 2002, for more recent applications.) The hazard function of the Cox model is defined as the product of a baseline hazard function and a set of proportional factors such that () () ( ) ( ) ( ) 0 ;exp', ij j i j j ij j ij j ht z t h t z t β = j = 1, 2, (2.1) where () 0jij ht is a baseline hazard function that describes the overall shape of the mortgage termination risks by borrowers’ prepayment or default decision; ( ) jij zt is a vector of proportional factors capturing time-varying or time-invariant covariates. These covariates reflect market values of the financial options as well as other financial/economic market variations and mortgage borrowers’ characteristics; j indicates prepayment (if j=1) or default (if j=2) event. Deng et al. (2004) tested if option theory is applicable to Chinese residential mortgage market, and concluded that it is still not. But they omitted a potentially important variable, mortgage term. For instance, a shorter term loan borrower may just needs the mortgage instrument to improve her short-run financial liquidity. Whenever this necessity disappears, she may prepay. On the contrary, most of the longer term loan borrowers expect much longer mortgage holding period due to their credit constrain for the payment or investment consideration. They are much less likely to prepay as quickly as the shorter term loan borrowers. In another word, the 50 expected holding period affects mortgage instrument choice at initiation; the instrument choice at initiation will reveal the expected holding period in turn. 4. Data The empirical analysis is based upon micro mortgage data with loan history information collected by a major residential mortgage lender in Beijing, China. The original dataset includes 75,536 single-family mortgage loans issued between March 1998 and October 2002. All loans are adjustable rate mortgages. 31 Most of them are constant payment mortgage loans. 32 The mortgage history period ends in October 2002. For each loan, the available information includes the year and month of origination and termination (if a loan has been closed), appraisal value of the property at origination, original loan amount, initial loan-to-value ratio, mortgage contract interest rate, term to maturity, and indicators of prepayment or default. The dataset also provides valuable information about the borrowers’ characteristics, including household monthly income, borrower’s age, marital status, education, occupation, and job position. Following Deng, Quigley and Van Order (2000), we compute a time-varying path of current equity to market value ratio (i.e. the ratio between the contemporaneous equity value and the market value of the property) for each loan starting from its origination till termination (or censored point), by using Beijing real 31 Decisions on mortgage rate adjustments are made by the People’s Bank of China. Once a rate adjustment is announced, the new rate will be applied to all existing mortgage loans on the market, without margin and caps, starting on January 1 st of the following year. 32 There are two methods of payment are adopted in Chinese mortgage market: constant payment mortgage (CPM) and constant amortization mortgage (CAM). 51 estate indices together with property value at loan origination and contemporaneous market value of the remaining mortgage balance. More specifically, the ratio of equity to market value, E of the property i in the kth month since purchase is () ,,, , , , ,, 1 , , , 1 iii i i ii i i ii ii ii ik im k ik ik k ik i TM k i im k t t k MV E M I MC I P V m τ τ τ τ τ + + − + = + − =  =    = + ∑ (2.2) where M is the current market value of the property, C is the original purchasing value of the property (at time τ), I τ and I τ+k are house price indexes at time τ and k month thereafter, respectively, V is the current value of the mortgage, TM is the mortgage contract term, P is the monthly mortgage principal and interest payment, and m is the current market rate of the mortgage. We also compute a time-varying covariate of call (prepayment) option (i.e., the present value of the differences in remaining monthly payments calculated using the mortgage note rate and the contemporaneous market rate) for each loan observation. 33 However, since all residential mortgage loans issued in Beijing are adjustable rate mortgages (ARM) without margin and caps, our data analysis indicates that call option virtually has no economic value to Chinese mortgage borrowers. So far, there has been no refinance-driven prepayment reported in 33 See Deng, Quigley and Van Order (2000) for a discussion of computing the time-varying covariates of call option value for each loan. 52 Chinese mortgage market. Therefore, at least for now, the option theory cannot explain the observed prepayment behavior in Chinese residential mortgage market. We opt not to include the call option covariate in our empirical analysis. In addition, we match macroeconomic variables including 5 year mortgage rate, 20 year mortgage rate, and Shanghai Stock Exchange Index (SSEI). The first variable indicates the cost of short term mortgage; the difference of the first two variables provides the relative expensiveness of long term mortgage loans. Shanghai Stock Exchange Index is the proxy of alternative investment opportunities. Our analysis is confined to level-payment mortgage loans with 5 year, 10 year, 15 year or 20 year maturities. There are only 313 loans originated in 1998, among which 29 were defaulted due to a development project dispute. Such abnormally high default rates caused by a project dispute will bias our estimation of borrower’s behavior. Therefore we decide to exclude the 313 loans originated in 1998 from our analysis. The upper and bottom 1 percentile in terms of annual household income are also excluded. 34 Also, we deleted the observations with extreme house price to annual household income ratio. 35 The final sample contains 64,181 loan records, among which, 4,497 loans (about 7 percent) were prepaid, 191 loans (about 0.3 percent) were defaulted, and 59,493 (about 93 percent) were still active at the end of data collecting period. 34 The cutting point for upper 1 percentile for annual household income is 624,000 RMB, while the bottom 1 percentile is 14,400. 35 This ratio is constrained as less than or equal to 30. 53 Table 2.1 Descriptive statistics for mortgage loans - means and standard deviations at origination All Prepay Defaulted Other * Loan-to-value ratio (LTV) 69.56 65.07 74.99 69.88 (13.67) (15.94) (7.01) (13.43) Original loan amount 407,511 378,711 1,278,115 406,893 (381,786) (328,611) (709,186) (380,766) House price 576,739 571,786 1,719,876 573,443 (519,555) (456,707) (1,003,115) (517,711) Household annual income 97,948 98,867 257,344 97,366 (88,779) (88,235) (163,757) (88,016) Borrower's age 35.00 37.00 37.00 35.00 (7.60) (8.35) (7.34) (7.52) Number of observations 64,181 4,497 191 59,493 Note: Standard deviations are in parentheses. * Other includes matured mortgages as well as those outstanding at the end of the data collecting period. Table 2.1 presents means and standard deviations of the continuous covariates measured at mortgage origination. The statistics suggest that borrowers with higher loan to value (LTV) ratios at origination may carry higher default risks. On the other hand, those loans prepaid have lower LTV than the rest of the loans in the pool. Such observation may suggest that less liquidity constrained borrowers in China are likely to payoff their mortgage earlier. Moreover, the loans prepaid are associated with lower original loan amount and house price, while the loans defaulted are associated with higher original loan amount as well as house price. The latter also have higher average household annual income. Hence most of defaults happened during this period are related to some higher-end projects. Both of prepayment and default risks are more likely to be carried out by relatively older borrowers. 54 Table 2.2 Descriptive statistics for mortgage loans – means and standard deviations for time varying covariates All Prepay Defaulted Censored Current equity to market value ratio 0.30 0.35 0.25 0.30 (0.14) (0.16) (0.07) (0.13) Slope of yield curve 3.14 2.96 2.94 3.15 (0.41) (0.20) (0.17) (0.42) Unemployment rate (%) 1.10 0.96 0.92 1.11 (0.31) (0.26) (0.23) (0.31) Shanghai stock exchange index 1819 1871 1963 1814 At Loan Origination (227) (237) (184) (226) Current equity to market value ratio 0.38 0.43 0.31 0.38 (0.15) (0.16) (0.08) (0.14) Slope of yield curve 3.86 3.64 3.59 3.88 (0.13) (0.42) (0.45) (0.00) Unemployment rate (%) 1.50 1.45 1.42 1.50 (0.04) (0.12) (0.14) (0.00) Shanghai stock exchange index 1516 1619 1714 1508 At Loan Termination (39) (179) (84) (0.00) Number of observations 64,181 4,497 191 59,493 Note: Standard deviations are in parentheses. * Other includes matured mortgages as well as those outstanding at the end of the data collecting period. Table 2.2 shows the means and standard deviations of time varying covariates, including the calculated current equity to market value ratio, slope of yield curve, 36 unemployment rate, and Shanghai Stock Exchange Index. As we expected that current equity to market value ratios are lower at origination and higher at termination for all loans due to mortgage amortization process. Besides that, the statistics also indicate that loans that were eventually defaulted are associated with the lowest equity ratios at origination and the loans prepaid are associated with the highest equity ratio at origination as well at termination. This is consistent with the LTV statistics in Table 2.1. Moreover, a positive term structure scenario (increasing 36 It is defined as the ratio between the 5 year CD yield and the checkable deposit yield. 55 in slopes of yield curve) is observed during the sample period. On the other hand, the unemployment rate is rising, and Stock Index is declining. Table 2.3 presents the number of loans in the sample stratified by major categorical covariates (in eight separate panels) and by loan status (in columns). It also presents percentage share by prepayment, default and other (censored) within each sub-categories (these percentage figures are reported in the parentheses in columns 2 to 4), as well as percentage share for each sub-categories (these percentage figures are reported in the parentheses in column 1). Panel 1 of Table 2.3 reports the frequency statistics separated by mortgage terms. About 33 percent borrowers selected the shorter term mortgages which are less than or equal to 10 years, while 67 percent chose the longer term mortgages. The shorter term borrowers have comparatively higher prepayment rate while the longer term borrowers experienced higher default rate. Those mortgages with 5 year maturity have significantly higher prepayment rate than the rest. 56 Table 2.3 Descriptive statistics for mortgage loans -frequency of loans by major categorical covariates and by payoff types All Prepay Defaulted Other * Mortgage term 5 years 4,055 620 5 3,430 (6.32) (15.29) (0.12) (84.59) 10 years 17,382 1,565 25 15,792 (27.08) (9.00) (0.14) (90.85) 15 years 12,147 758 46 11,343 (18.93) (6.24) (0.38) (93.38) 20 years 30,597 1,554 115 28,928 (47.67) (5.08) (0.38) (94.55) Income group Low 2,981 205 1 2,775 (4.64) (6.88) (0.03) (93.09) Med-low 3,541 219 1 3,321 (5.52) (6.18) (0.03) (93.79) Median 4,227 248 4 3,975 (6.59) (5.87) (0.09) (94.04) Med-high 8,907 621 9 8,277 (13.88) (6.97) (0.10) (92.93) High 44,525 3,204 176 41,145 (69.37) (7.20) (0.40) (92.41) Origination year 1999 4,290 498 5 3,787 (6.68) (11.61) (0.12) (88.28) 2000 18,361 1,839 113 16,409 (28.61) (10.02) (0.62) (89.37) 2001 25,258 1,913 67 23,278 (39.35) (7.57) (0.27) (92.16) 2002 16,272 247 6 16,019 (25.35) (1.52) (0.04) (98.45) Age group Age ≤30 20,079 1,108 36 18,935 (31.28) (5.52) (0.18) (94.30) 30 ≤Age ≤35 16,573 1,056 46 15,471 (25.82) (6.37) (0.28) (93.35) 35 ≤Age ≤40 13,740 1,008 45 12,687 (21.41) (7.34) (0.33) (92.34) Age ≥40 13,789 1,325 64 12,400 (21.48) (9.61) (0.46) (89.93) Number of observations 64181 4497 191 59493 (to be continued) 57 Table 2.3 Descriptive statistics for mortgage loans -frequency of loans by major categorical covariates and by payoff types (continued) All Prepay Defaulted Other * Marital status Single 35,169 2,602 155 32,412 (54.80) (7.40) (0.44) (92.16) Married 29,012 1,895 36 27,081 (45.20) (6.53) (0.12) (93.34) Occupation Business and trade 14,135 939 24 13,172 (22.02) (6.64) (0.17) (93.19) Social service 6,109 424 21 5,664 (9.52) (6.94) (0.34) (92.72) Self employment 25,737 1,955 123 23,659 (40.10) (7.60) (0.48) (91.93) Education and research 6,238 358 5 5,875 (9.72) (5.74) (0.08) (94.18) Others 11,962 821 18 11,123 (18.64) (6.86) (0.15) (92.99) Job position Manager 39,046 2,820 159 36,067 (60.84) (7.22) (0.41) (92.37) Clerk 12,760 888 17 11,855 (19.88) (6.96) (0.13) (92.91) Technician 8,631 529 5 8,097 (13.45) (6.13) (0.06) (93.81) Others 3,744 260 10 3,474 (5.83) (6.94) (0.27) (92.79) Education Primary school 2,182 155 6 2,021 (3.40) (7.10) (0.27) (92.62) Secondary school 21,554 1,369 61 20,124 (33.58) (6.35) (0.28) (93.37) College 40,445 2,973 124 37,348 (63.02) (7.35) (0.31) (92.34) Number of observations 64181 4497 191 59493 Note: Column categorical percentages are in parentheses in column 1; row percentages by prepayment, default and other type are in parentheses in columns 2-4. *Other includes matured mortgages as well as those outstanding at the end of the data collecting period. 58 Panel 2 indicates that about 69 percent of all the borrowers are from the high income households. The high income borrowers also have the highest prepayment rate at 7.2 percent. The median-high income borrowers have the second highest prepayment rate at 7.0 percent, slightly higher than the low income borrowers; whereas the median income households are the most reluctant to prepay. The default risk is generally less than half percent to all of the sub-categories, while the rate associated with high-income households is above 10 times higher than the low and median-low income groups, and 3 to 4 times higher than the median and median- high income groups. Panel 3 reports the frequency statistics by loan origination years. The residential mortgage market in China took off rapidly since the 1998 administrative laws for broadening housing reform and expediting housing construction issued by the State Council of the People’s Republic of China. Only 4,290 loans (less than 7 percent) in the final sample were originated in 1999. Newly issued residential mortgage loans were more than quadruple to 18,361 (about 30 percent) in 2000, almost 6 times to 25,258 (39 percent) in 2001. The frequency statistics also indicate that loans originated in 1999 have higher prepayment rates (11.6 percent) than loans originated after 1999 (10.0 percent for loans originated in 2000, 7.6 percent for 2001, and 1.5 percent for 2002). Default risk in residential mortgage lending in China is quite low. For all groups, the default risk is less than one percent. Loans originated in 2000 have the highest default rate (0.6 percent) which is more than 5 times compared to the loans originated in 1999, 59 more than twice compared to the loans originated in 2000, and more than 16 times compare to those originated in 2002. Over three-quarters of the borrowers belong to age cohort under 40. Panel 3 indicates that borrowers over 40 years-old are more likely to pay off their loans earlier than those under 40. Almost 10 percent of the borrowers over 40 years-old cohort prepaid, while only 6 percent of borrowers under 40 prepaid. The younger the borrower, the lower prepayment as well as default rates. About 55 percent of the borrowers are single, who have slightly higher prepayment rate and significantly greater default rate compared with the married borrowers. The default rate associated with the single borrowers is more than quadruple of the default rate among the married borrowers. About 40 percent of the borrowers are self-employees, 22 percent are in business and trade, 10 percent are in education and research, another 10 percent are in social service, and the rest 19 percent belong to others. 37 Among these groups, self employees have both higher prepayment and default risks. As to the latter, self employees have about 6 times the default risks compare to those in education and science. White-collar workers are the majority of the borrowers. 38 Over 60 percent of the borrowers are managers who have relatively higher default rate (0.41 percent) than clerks and technicians (0.13 percent and 0.06 percent, respectively). 37 ‘Others’ here includes employees in government, finance and insurance, posts and telecommunications, army, real estate and construction, agriculture, industry, and water, etc. 38 ‘Other’ here refers to borrower who is army man, farmer, or inoccupation. 60 Finally, Borrowers with higher education are more likely to take advantage of mortgages. Over 63 percent of borrowers have college education, and just 3.4 percent of borrowers never go beyond primary schools. The higher educated borrowers also have slightly higher prepayment and default rates. As mentioned above, in the current Chinese mortgage market, all the loans with the same maturity enjoy the same mortgage rate. Hence the latter is the essential determinant of mortgage cost. To a rational borrower, borrowing at the regulated maximum LTV but borrowing short can effectively improve his purchase power with relatively low cost. On the other hand, the risk of a mortgage loan is associated with its monthly payment. A more risk-averse borrower will either borrow longer or/and borrow less, i.e. with lower LTV for a specific property. Table 2.4 Mortgage choice and early termination LTV Loan maturity = 60 months (percentage) Prepayment (percentage) Default (percentage) Low 60.3 6.9 6.88 0.03 Med-low 64.4 5.0 6.18 0.03 Median 66.8 4.2 5.87 0.09 Med-high 67.4 4.8 6.97 0.10 Income group High 71.3 6.9 7.20 0.40 Age ≤30 70.4 5.0 5.52 0.18 30<Age ≤35 70.0 5.5 6.37 0.28 35<Age ≤40 69.9 6.1 7.34 0.33 Age group Age>40 67.6 9.5 9.61 0.46 Single 70.6 5.9 7.40 0.44 Marital status Married 68.3 6.8 6.53 0.12 Table 2.4 shows the initial LTV and mortgage maturity choices as well as prepayment and default rate after mortgage origination for different types of borrowers. The first column displays the average LTV ratio, and the second column 61 displays the percentage of borrowers choosing short-term mortgage (= 120 months) in the corresponding income or demographic group. Higher income borrowers are more likely to borrow more and borrow relatively shorter, indicating that they are cost-minimizer. They are also financially savvy to take advantage of the even-mortgage-rate among the loans with the same maturity. As mentioned before, they have the highest prepayment rate compared to all the other income groups. It is likely that the higher income borrowers use a short term mortgage to help them with temporary liquidity needs or to share the construction risk with the lender; once the needs disappear or the property is usefully delivered as in the presale contract, they are likely to payoff their mortgages. On the other hand, disliking borrowing, the low income borrowers borrow less, borrow short, and payoff their mortgages as early as they can. Younger borrowers are comfortable to “live with the debt”. They have averagely higher LTV ratio and are the least likely to assume short term loans; while the older borrowers are more financially conservative by borrowing less and borrowing short. Similar to the low income borrowers, the older borrowers tend to payoff their mortgages earlier. Married couples are more conservative compared with the single borrowers. They borrow less and are more likely to borrower short. As the more stable economic units, they are less likely to prepay or default. Those different patterns reveal that different borrowers have different concerns at loan origination, which will further illustrate themselves later in the mortgage’s lifetime. 62 5. Empirical Results The empirical models are estimated based on the Cox Partial Likelihood approach (Cox, 1975). Table 2.5 presents estimates of two basic models of mortgage prepayment and default. As we discussed in previous section, currently all residential mortgages in China are adjustable rate mortgages (ARMs). Once the Central Bank (The People’s Bank of China) announces a rate adjustment, this new rate will be applied to all existing mortgage loans on the market without caps starting from the beginning of the following year. As a result, the “call option” has virtually no value to Chinese mortgage borrowers. This contradicts to the conventional wisdom in the existing mortgage literature which considers the “call option” value as a dominant factor driving prepayments in the U.S. residential mortgage market. (See, for example, Kau, Keenan, Muller and Epperson, 1990, for a theoretical analysis on the adjustable rate mortgages in option theory framework. Quigley, 1987, and Stanton and Wallace, 1995, analyze the impacts of interest rate on adjustable rate mortgage termination and valuation. Cunningham and Capone, 1990, and Calhoun and Deng, 2002, provide empirical evidence of the association between the “call option” value and the ARMs prepayment behavior in the U.S. market.) In fact, all prepayments observed in the sample are early payoff rather than refinance. Therefore, we exclude the “call option” value from the determinants of prepayment in our model. 63 Table 2.5 Proportional hazard estimates for mortgage prepayment and default Model 1 Model 2 Prepay Default Prepay Default Current equity to market value ratio 2.75 -4.11 (19.11) (-2.44) Slope of yield curve -0.28 0.06 -0.27 0.02 (-5.39) (0.22) (-5.19) (0.06) Loan to value ration (LTV) > 70 -0.48 0.09 0.06 -0.44 (-14.88) (0.57) (1.23) (-1.67) Log value of original loan amount -0.02 1.74 0.11 1.68 (-1.32) (19.69) (5.45) (18.45) Unemployment rate (%) 0.05 0.02 0.05 0.03 (28.12) (3.20) (27.66) (3.28) -Log likelihood 44871 1757 44700 1753 Note: t-ratios are in parentheses. Model 1 focuses on the key determinants traditionally used by the lending industry to control interest rate risk and credit risk. These covariates include slope of yield curve, initial loan-to-value ratio, local unemployment rate, and loan characteristics such as log value of original loan amount. Initial loan-to-value ratio is served as proxy for measuring borrower’s credit worthiness; yield curve slope is the proxy for alternative investment opportunity; and local unemployment rate is served as proxy for macro economic environment and consumers’ confidence about the economy and their financial wellbeing. Estimates from model 1 indicate that borrowers who have poor credit are less likely to prepay the mortgage. This is consistent with the liquidity constrain argument widely discussed in the existing mortgage literature (See, for example, Archer, Ling and McGill, 1996, and Deng, Quigley and Van Order, 1996). Borrowers who apply for higher LTV ratio typically are constrained by limited liquid assets and hence less likely to prepay. 64 The slope of yield curve discloses the relationship between leverage in housing and opportunities in stock market investments. 39 When the yield curve getting flatter, borrowers in China choose to pay off the current mortgage debt rather than take a long position in the long term bond market. Therefore, as indicated in model 1 that prepayment risk is negatively related to the slope of yield curve. This slope is statistically insignificant in the default function, indicating that residential mortgage borrowers in China are reluctant to consider default as a financial choice. Log value of original loan amount is significant and positively associated with default risk. Unemployment rate is highly significant in determining prepayment risk and less so for default risk. Contrast to previous findings in the existing literature about the residential mortgage borrowers’ behavior in the United States, 40 the empirical estimates from model 1 indicate that prepayment risk increases as unemployment rate rises. Unemployment rate is a macro variable indicating the strength of the macro economic environment. It also reflects Chinese borrowers’ confidence towards their future income and financial safety and soundness. To most Chinese households, housing is a basic necessity rather than luxury of living. Chinese borrowers tend to pay off their mortgage debt when they feel uncertain about future income and financial safety and soundness. In other words, when Chinese households feel uncertain about their future wealth, they will choose to invest safe assets (housing) rather than risky assets (such as stocks and 39 Steep yield curve implies higher return for investing in long term capital market, and vice versa. 40 Deng, Quigley and Van Order (1996) (2000), among others, found that prepayment risk declines as unemployment risk increases. This is due to liquidity constrains faced by many borrowers during the weak economy. 65 bonds). This is quite different to what we have learnt in the U.S. mortgage market that in general unemployment rate is negatively associated with prepayment risk. Model 2 included time varying “current equity to market value ratio” into consideration. Similar to LTV, it is also a measurement of borrower’s credit worthiness. The difference is that LTV ratio is a snapshot at loan origination; while “current equity to market value ratio” is real time updated information. The latter has higher power over the outdated static LTV ratio during amortization procedure. As observed in model 2, the initial loan to value ratio is not statistically significant with the presence of current equity to market value ratio. Log value of original loan amount is significant and positively associated with both default risk and prepayment risk in model 2. These findings suggest that borrowers of jumbo loans are more likely to consider housing as a luxury good or investment instrument rather than necessity of living. Jumbo loan borrowers in China are risky borrowers and lenders should take precaution when they approve jumbo loans. Table 2.6 lists the prepayment and default schedules for different mortgage maturity groups separately. 5-year loans have the highest prepayment rate at 15.3 percent, about 3 times of the rate among the 20-year loans. Longer the mortgage term, lower the prepayment rate. Default rate is generally quite low. For all mortgage term groups, default rate is below 0.5 percent. It is higher among the longer term subgroups, i.e. equal to or longer than 15 years. Default rate among the shorter term mortgages is less than half of the rate among the longer term mortgages. Hence, the 66 expected holding period affects borrowers’ mortgage choices at the first place. For instance, a borrower who only needs short term liquidity or a home buyer just wants to share the construction risk with the bank are more likely to borrow a short term loan, and tend to payoff his mortgage whenever the liquidity issue disappears or when the property is successfully delivered. Table 2.6 Prepayment and default schedule for different mortgage maturity groups (a) Prepayment schedule (b) Default schedule Model 3 in Table 2.7 extends model 2 by controlling for mortgage maturity. Compared with the reference group including the loans with 5 year term, longer term is associated with lower likelihood of prepayment. For default, the impact of mortgage maturity is not statistically significant. Most of the key determinants reported in model 2 are quite robust, except for current equity to market value ration in the default model, which becomes insignificant after the inclusion of mortgage term indicators. It implies that, based on the current data, Chinese mortgage Term Number of observations Prepaid cases Percentage of prepayment 5 years 4,055 620 15.3 10 years 17,382 1,565 9.0 15 years 12,147 758 6.2 20 years 30,597 1,554 5.1 Total 64,181 4,497 7.0 Term Number of observations Default cases Percentage of default 5 years 4,055 5 0.12 10 years 17,382 25 0.14 15 years 12,147 46 0.38 20 years 30,597 115 0.38 Total 64,181 191 0.30 67 borrowers are not considering the financial put option value as a factor that drives their default decisions. Table 2.7 Proportional hazard estimates for mortgage prepayment and default, extended models (part a) Model 3 Model 4 Model 5 Prepay Default Prepay Default Prepay Default Current equity to market value ratio 2.14 -0.78 1.88 -0.97 2.17 -1.28 (-12.48) (-0.42) (10.80) (-0.51) (17.91) (-1.22) Slope of yield curve -0.27 -0.004 -0.21 -0.50 -0.21 -0.43 (-5.1) (-0.02) (-3.88) (-1.38) (-3.93) (-1.20) Loan to value ration (LTV) > 70 -0.02 -0.18 -0.08 -0.30 (-0.51) (-0.64) (-1.73) (-1.07) Log value of original loan amount 0.1 1.68 0.00 1.82 0.01 1.82 (-4.98) (-18.89) (-0.14) (17.21) (0.41) (17.75) Unemployment rate (%) 0.05 0.02 0.02 0.04 0.02 0.05 (-27.9) (-2.69) (11.98) (3.46) (11.90) (3.56) Mortgage term 10 years -0.34 -0.44 -0.33 -0.40 (-6.42) (-0.83) (-6.21) (-0.74) 15 years -0.42 0.55 -0.43 0.57 (-6.67) (-1.00) (-6.74) (1.03) 20 years -0.41 0.54 -0.40 0.51 (-6.7) (-0.97) (-6.51) (0.91) Long term mortgage (>5 years) -0.34 0.25 (-6.75) (0.51) Origination year 2000 0.36 2.30 0.33 2.40 (5.09) (3.99) (4.71) (4.22) 2001 0.80 3.18 0.77 3.24 (8.31) (4.92) (8.04) (5.07) 2002 0.66 3.91 0.63 3.91 (4.89) (4.48) (4.69) (4.51) Income group Med-low 0.08 -0.44 (0.84) (-0.31) Median 0.13 0.45 (1.37) (0.40) Med-high 0.30 0.28 (3.65) (0.27) High 0.34 -0.44 (4.02) (-0.43) 0.25 -0.58 Median-high and High income borrowers (4.89) (-1.27) Borrowers with Age ≥40 0.24 0.20 0.24 0.16 (7.04) (1.22) (7.12) (1.00) Married -0.04 -0.64 -0.03 -0.63 (-1.12) (-3.30) (-0.92) (-3.22) (to be continued) 68 Table 2.7 Proportional hazard estimates for mortgage prepayment and default, extended models (part a) (cont.) Model 3 Model 4 Model 5 Prepay Default Prepay Default Prepay Default Education Secondary school 0.14 -0.23 0.14 -0.22 (1.64) (-0.53) (1.60) (-0.50) College 0.24 -0.82 0.23 -0.81 (2.82) (-1.94) (2.75) (-1.91) Occupation Business and trade 0.01 -0.54 0.01 -0.54 (0.23) (-1.71) (0.28) (-1.73) Social service 0.01 0.62 0.01 0.59 (0.09) (1.91) (0.12) (1.84) Self employment -0.04 0.43 -0.04 0.39 (-0.94) (1.66) (-0.94) (1.51) Education and research -0.20 -0.32 -0.19 -0.34 (-3.06) (-0.63) (-3.03) (-0.67) Job position Manager 0.29 0.22 0.30 0.25 (4.44) (0.68) (4.56) (0.76) Clerk 0.35 0.41 0.28 -0.35 (4.89) (1.01) (3.65) (-0.63) Technician 0.27 -0.44 0.36 0.55 (3.56) (-0.79) (4.96) (1.34) Shanghai stock exchange index a -0.20 0.37 -0.20 0.36 (-13.05) (6.17) (-12.99) (6.03) -Log likelihood 44674 1744 44458 1683 44463 1696 Note: t-ratios are in parentheses. a Shanghai Stock Exchange Index is defined as Shanghai Stock Exchange Composite divided by 100. Model 4 in Table 2.7 further extends model 3 by including borrower characteristics, financial market trend, and origination year indicators, etc. Mortgage loans originated after 2000 tend to have much higher prepayment risk compared to those originated in 1999. Such trend continues in 2001, and then slightly declines in 69 2002. 41 During the period from 1998 to 2001, there have been several major regulations 42 published regarding new policy of Beijing housing reform and the development of residential mortgage system. These policy changes led to swift shifts in the practice of mortgage origination process and hence the performance of the mortgage loans originated thereafter. Borrower’s characteristics have different effects on prepayment and default risks. Those characteristics include borrower’s household income, age, marital status, education, occupation, and job position. The median-high and high income borrowers are more likely to prepay compared to the other borrowers. Households with higher income have more liquid asset, consequently, are more capable of paying off their loans. Borrower’s age is important to determine prepayment risk as the borrowers more than 40 years old are more likely to prepay compared with the other younger borrowers, as the latter have longer horizon before retirement; moreover, the younger generation prefers consumption on credit and they are indeed more comfortable to “live with debt”. Neither income nor age is significant in determining default risk. Single borrowers have higher default risk compared to the married couples. In general, family is a more stable social unit than single individual, and married borrowers prefer more stable monthly expense stream. 41 Our loan history data is censored at October 2002. Therefore the mortgage pool originated in 2002 in our sample has not yet reached its prepayment peak compared to the loans originated earlier with longer span of duration. 42 For example: The People’s Bank of China Bulletin on Expanding Credit Available to Residential Mortgage Lending and Supporting Residential Housing Construction and Consumption, April 7, 1998; The Office of Beijing Housing System Reform Bulletin (98) No. 265 on Policies Regarding Sales and Pricing of Public Housing Units to Employees in 1999; Bulletin on Further Improving Sales of Public Housing Units, Feb. 10, 1999; Beijing Housing Financing Center Bulletin (99) No. 117 on Adjusting Policies Regarding Residential Insured Mortgage Lending in 1999; etc. 70 Borrowers with college degree have higher prepayment risk but lower default risk. Educators and researchers have lower propensity to prepay their mortgage loans, probably due to their relatively stable income during their entire career. Borrowers in education and research fields are the least likely to prepay with stable financial status. White-collar workers, such as managers and clerks tend to prepay loan faster than blue-collar workers such as technicians, while others (which include freelance workers, military service personnel, and farmers) is the borrower group with least prepayment risk. Both occupation and job position seem not significant in determining default risk. As mentioned above, stock exchange has been attached more and more importance in Chinese people’s financial considerations as the fast growing alternative investment opportunity to the traditional deposit. Shanghai Stock Exchange index (SSEI) is significant in both the prepayment and default models. Borrowers’ decision on mortgage is a financial decision on investment portfolio choice. SSEI is negative and highly significant in the prepayment function – indicating that bear market drives Chinese households to reallocate their assets from stock market to payoff their mortgage debts. One the other hand, SSEI is positive and significant in the default function, namely bull market is associated with higher default risk, implying that households who stop paying their mortgage may choose to reallocate their assets from housing to stock market. Model 5 is the results after re-group the borrowers regarding their mortgage terms and income level. For current residential mortgage loans, there are only two 71 mortgage rate, one for the short term mortgage (5 years and less), another for the long term mortgage (more than 5 years). The prepayment risks attached to the long term mortgage groups in model 4 are similar, but all significantly lower than the prepayment risk of the short term mortgages. So we divide the mortgage loans into short term and long term two groups with cutting of point as 5 years. For the income groups, median-low and median income borrowers do not behave significantly different from the low income borrowers, while the median-high and high income borrowers are different from the others in the similar pattern. Accordingly, we divide the sample into “median-high to high income” and the rest, which includes low, median-low, and median income borrowers. Besides that, only the current equity to market value ratio is kept as the real time updated equity status while LTV ratio as the snapshot at loan origination is dropped. The results after re-grouping mortgage term group and income group are similar to the results in model 4. Model 6 in Table 2.8 extends model 5 to test whether different borrowers groups respond differently to the macro economic shocks. As discussed above, borrowers in different income and age groups can behave quite differently in prepayment decision. Long term and short term mortgage borrowers can have very different concerns especially towards mortgage prepayment according to previous models. Hence, I include the interactions of borrower income group to the slope of yield curve, unemployment rate, and stock index, as well as the interactions of these macro economic variables with long term mortgage maturity choice, and borrower’s age group in the further extended prepayment model. On the other hand, borrowers’ 72 marital status has significant impact on their default decision. So I include the interactions of borrowers’ marital status to those macro economic variables in my extended default model. The results of model 6 suggest that, in prepayment function, slope of yield curve turns out to be significantly positive, which serves as the proxy of people’s expectation of future economic situation. Jumbo loan borrowers have lower prepayment risk. Married borrowers are less likely to prepay. After considering borrowers’ reactions to the macroeconomic shocks, rather than the borrowers in education and research fields, those with jobs in social service are the least likely to prepay compared with the others such as borrowers in business and trade as well as the self employed borrowers. With better financial knowledge and more investment opportunities, the higher income borrowers treat mortgage as one component of their portfolio. When better long term investment opportunity emerges, they borrow the mortgage and invest their money for the higher return in the future. When the yield curve is flat, they reallocate the money from other investments to payoff their mortgage. Long term mortgage borrowers lock into longer loan horizon hence relatively more sensitive to the return of long term investment opportunities. Flat yield curve also drives them to prepay. 73 Table 2.8 Proportional hazard estimates for mortgage prepayment and default, extended models (part b) Model 6 Model 6 Prepay Default Prepay Default 1.34 -2.26 Job position Current equity to market value ratio (10.48) (-1.95) Manager 0.29 0.15 Slope of yield curve 1.40 -0.10 (4.36) (0.44) (16.17) (-0.30) Clerk 0.35 -0.31 Log value of original loan amount -0.09 1.92 (4.52) (-0.57) (-3.66) (18.63) Technician 0.45 0.62 Unemployment rate (%) 0.07 0.05 (6.16) (1.51) (24.94) (3.74) Shanghai stock exchange index a -0.69 0.25 Long term mortgage (>5 years) -24.28 -0.06 (-37.28) (3.98) (-12.22) (-0.12) Interaction with Median-high and High income borrowers Origination year Slope of yield curve -2.49 2000 0.16 1.85 (-6.85) (2.21) (3.14) Shanghai stock exchange index 0.32 2001 1.25 2.61 (5.03) (12.53) (3.98) Unemployment rate (%) -0.08 2002 1.26 3.42 (-5.57) (9.29) (3.96) Interaction with Longer mortgage maturity c Slope of yield curve -4.88 17.41 -0.81 (-13.38) Median-high and High income borrowers (8.35) (-1.75) Shanghai stock exchange index 1.31 Borrowers with Age ≥40 7.90 0.12 (21.30) (6.39) (0.77) Unemployment rate (%) 0.15 Married -0.10 -22.92 (10.19) (-3.18) (-2.14) Interaction with Borrowers with Age ≥40 Education Slope of yield curve 0.13 Secondary school -0.09 -0.56 (1.05) (-1.06) (-1.28) Shanghai stock exchange index -0.31 College 0.16 -0.98 (-10.41) (1.94) (-2.29) Unemployment rate (%) -0.02 Occupation (-3.42) Business and trade 0.05 -0.58 Interaction with Married borrower (1.04) (-1.83) Slope of yield curve -6.34 Social service -0.87 0.13 (-11.22) (-12.09) (0.36) Shanghai stock exchange index 1.27 Self employment -0.03 0.32 (5.54) (-0.76) (1.26) Unemployment rate (%) 0.18 Education and research -0.17 -0.36 (3.40) (-2.70) (-0.71) -Log likelihood 37395 1576 Note: t-ratios are in parentheses. a Shanghai Stock Exchange Index is defined as Shanghai Stock Exchange Composite divided by 100. b Long term mortgage dummy takes value 1 if mortgage term equal to or longer than 5 years, and 0 otherwise. 74 The higher income borrowers are less sensitive to the fluctuations in the stock market because of their diversified investment besides the stocks traded in Shanghai Exchange. Interestingly, the long term mortgage borrowers behave quite differently from other borrowers towards the stock market shocks. The long term mortgage borrowers are more likely to prepay in the bull market as the risk-averse agents who are likely to reallocate the assets from risky investment to payoff their mortgage. The borrowers over 40 years old may heavily invest in the stock market, and hence are more sensitive to SSEI fluctuations. They are more likely to pull the money out from the stock market and payoff their mortgage during bear market. As mentioned above, the long term mortgage borrowers are risk averse. Higher unemployment rate signals greater economic and financial uncertainties, which in turn results in higher prepayment rate among long term mortgage borrowers. On the other hand, both higher income borrowers and borrowers more than 40 years old are more likely to have less prepayment rate during higher unemployment period, especially the borrowers over 40 years old. The possible reason is that the economic downturn may be more likely to have material impact on the higher income households and the people over 40 years old, either as less economic gain or significant economic loss. In terms of default function, current equity to market ratio is marginal significant and negatively related to the default risk. The married borrowers have lower default risk than the single borrowers when the yield curve is flat. The married borrowers are relatively more sensitive to the change in unemployment and stock 75 exchange index. It gives the impression that the married borrowers are more ruthless in default, but this conclusion is questionable given the fact that a lot of defaults in the study period are project related. 6. Conclusions The residential mortgage market in China is a newly emerged sector of the capital market. The new residential mortgage market is evolving rapidly with the swift housing system reform currently carried on in China. The fast growth and the accelerating importance of the residential mortgage sector becomes a financial driver for the booming residential housing development and sustained economic growth in China. The distinct Chinese features of the residential mortgage market makes real estate finance a very attractive research topic. Financial call option is currently infeasible due to imperfect market circumstances; while the put option is in general “out-of-money” to the borrowers because of the steady increases of the property values in the housing market during the sampling period. Option theory apparently fails to explain the prepayment and default behavior in the residential mortgage market in China. On the other hand, other non-financial-option related social-economic factors and borrower characteristics play major roles in explaining the prepayment and default behavior in China. In general, borrowers choose to pay off mortgage debts in the bear market and when the yield curve is flat. The current extremely low deposit 76 rate in China makes saving no longer a rational option for long term investment to many Chinese. Stock market provides Chinese households a viable locale to benefit from the higher return investment in the capital market. Therefore, stock market’s fluctuations have significant impact on mortgage borrowers’ prepayment and default decisions. Many Chinese borrowers tend to be “uncertainty averse” such that when unemployment rate rises, borrowers will reallocate their investment portfolio to safe assets by paying off their mortgage debts. This contradicts to the borrowers’ behaviors observed in the residential mortgage markets in the United States and other countries. Generating a loan from the bank might also be taken as a mechanism to share construction period risk by the borrowers, especially in the terms of the presale projects. Some short term loan borrowers will obtain a mortgage even they can payoff the purchase price at the very time, in order to avoid the financial risk once the developer fails to deliver the housing as contract. Thus lenders shall pay enough precautions to the short term loan borrowers. The reform of housing and housing finance system in China bring along swift changes in many housing and finance related policies and regulations, which influence households’ decisions. The factor of policy changes has proved to be one of the critical determents in our model for mortgage prepayment risk. Borrower’s initial mortgage choice reveals his expectation in holding period. Long-term loan borrowers are less likely to prepay. Loans with lower loan to value 77 ratio are more likely to be prepaid. Borrower’s characteristics are found to be significant in determining borrower’s prepayment behaviors, hence may be used as an effective tool for screening across loan applicants and for determining who the potential high risk borrowers are. These findings have important policy implications. Median-high to high income borrowers, borrower over 40 years old, as well as white-collar workers are more likely to prepay their mortgage debts. Younger households, blue-collar workers, and married borrowers are less likely to prepay. On the other hand, jumbo loan borrowers and single borrowers are more likely to default. Moreover, different borrowers have different concerns and react to macroeconomic and financial shocks differently. The higher income borrowers are more sensitive to cost, while the long term mortgage borrowers are more sensitive to risk. The borrowers over 40 years old may be more financially conservative according to their effort to avoid debt. Hence, adopting a risk-based pricing in residential mortgage lending in China will not only improve the efficiency of the market, but also enhance the credit available to the most needed households, i.e., the younger households, blue-collar workers, lower income households, and help them become homeowners. 78 Chapter 3 Impact of Rent Control on Mobile Home Prices in California 1. Introduction The social and economic impacts as well as the legal implications of price controls in the rental housing market have been debated extensively among economists and legal scholars. Most economists agree that “a ceiling on rents reduces the quantity and quality of housing available” (Alston, Kearl, and Vaughan, 1992). 43 While some renters benefit from lower rents, there are also costs (Olsen, 1972). 44 Severe rent controls can cause undersupply in the apartment rental markets, and increase the search costs for tenants (Arnott and Igarashi, 2000). Below-market prices, if enforced, create opportunities for non-price rationing. Potential renters are obliged to invest time and effort chasing down access to the limited supply. These resource expenditures do not accrue to any sellers and cannot be converted to product and thus become a “deadweight” efficiency loss. Additional efficiency losses come from renters failing to adjust their housing consumption even though their circumstances change so they can hold on to their rent controlled units. (See Gyourko and Linneman, 1989, and Glaeser and Luttmer, 2003, for a thorough discussion of such effects for the case of New York city.) In addition, higher-income households are likely have an advantage in tracking valuable 43 Alston, Kearl, and Vaughan (1992) reported that from their survey of a stratified random sample of 1,350 economists employed in the United States, more than three-quarters of the respondents agreed with the above statement. 44 One seminal study of rent control in New York City estimated that the costs to property owners are two times the benefits to renters (Olsen, 1972). 79 information on available units. Thus, equity gains from rent controls have seldom been realized by those most in need such benefit, suggesting that rent control has not been a policy useful for lower-income households that seek affordable housing (Quigley, 2002). However, some economists agree that a well-designed rent control regime could improve the unrestricted equilibrium of an imperfect market (Arnott, 1995). Rent control is preferred when the market distortion is the unavailability of insurance against a sharp, unanticipated rent rise. Furthermore, economists disagree on the net social effect of rent controls, for example, whether rent controls might affect increased homelessness by reducing the supply of rental units (Tucker, 1989, Quigley, 1990, HUD, 1991, Early and Olsen, 1998). Rent control policies have caught the attention of legal scholars and have been extensively litigated. Legal scholars have debated whether some rent control regulations have violated the Fifth Amendment’s Takings Clause. In Yee v. City of Escondido, Cal. (90-1947), 503 U.S. 519 (1992), the Supreme Court affirmed a Superior Court decision rejecting the argument that the ordinance effected a physical taking by depriving park owners of all use and occupancy of their property and granting to their tenants, and their tenants' successors, the right to physically permanently occupy and use the property. The Supreme Court ruled that the argument that rent control ordinances benefit current but not future property owners has “nothing to do with whether it causes a physical taking.” Since then, some legal scholars have nevertheless argued that certain rent controls constitute a regulatory taking because the regulation has unfairly singled out some property owners to bear a 80 burden that should be borne by the public at-large. (See Rubinfeld, 1992, and Radford, 2004, for a discussion.) A unique aspect of the mobile home industry is the separate ownership of the coach and the pad where the coach is located; i.e., the mobile homeowner owns only the coach unit and rents a pad in a mobile home park on which she parks her (not-so- mobile) home. If the rent of the pad is constrained to be below market rents, coach owners are hypothesized to be able to sell the coach for more than it would be worth without rent control, thereby capitalizing the rent savings arising from rent control. The framework of separate ownership of coach unit and pad provides a unique opportunity to explicitly test the economic impact of rent control policy. Several studies by Hirsch (1988), Hirsch et al (1988) and Hirsch and Rufolo (1999) have analyzed capitalization of the effects of rent control on coach unit values. Using straightforward hedonic modeling techniques and relatively small samples, the authors demonstrated that the savings from pad rent controls are capitalized. In this study, we propose an extended economic model using a large sample of about 200,000 mobile home transaction records from January 1983 to May 2003 from seven counties in California. 45 The coach transaction records include information on each coach’s quality (brand, length, and width), age, address, original sale price, last sale price, last sale date, etc. This study provides a more explicit and comprehensive measure of the economic impacts of rent controls through an analysis of micro transactions data. 45 The data are maintained by the California Department of Housing and Community Development (HCD). 81 The remainder of this chapter is organized as follows: section 2 summarizes the unique characteristics of the mobile home market and rent controls in such markets. Section 3 describes our methodology for measuring rent control capitalization. Section 4 discusses the data; section 5 presents the empirical results. The last section offers concluding remarks. 2. Mobile Home Industry In 2003, almost 9 million of the nation's 121 million housing units were mobile homes. This was 7.4 percent of the housing stock, up from 6.5 percent of the housing stock in 1987. Of the more than 338 thousand mobile homes added to the U.S. housing stock in 1999, more than 10 percent were added in the western states. As the number of Americans living in mobile homes increases, more attention is being paid to mobile home parks, their management and associated public policies. Among the issues of great interest are the effects of various types of mobile home park rent control. The imposition of rent controls in many California jurisdictions has been shown to explain declining shipments of mobile homes to California (Hirsch and Rufolo, 1999). More up-to-date time series data for California also show that from 1983-2003 (through the month of May), the number of mobile homes subject to rent controls increased. Figures 3.1(a) – (c) show this by various mobile home descriptors including Total Traded Square Feet, Total Traded Value (constant dollars), and the Number of Transactions. 82 Rent control policies and by-laws vary in scope and severity. Flexible rent control regimes allow vacancy decontrol while more rigid regimes permit rent increases tied to one of a variety of cost-of-living indices. Rent control regulations in many cities have gone through cycles of control and decontrol, in some cases resulting in various vintages of the stock being grandfathered. A visible result is that some cities have neighborhoods with well-kept free-market rental housing adjacent to poorly maintained rent controlled housing. Rent control systems are most well known for having been imposed on traditional multi-family rental housing units where the landlord owns the land and the building and rents individual units to tenants using formal or informal leasing arrangements. Legal systems have evolved to define the property rights of landlords and tenants. Rent controlled systems often “piggy-back” on these systems as security of tenure is deemed critical if landlords are permitted to increase rents when a unit is vacated. 46 In contrast, in mobile home parks, the site (“pad”) is rented to the tenant who either acquires the mobile home from a prior tenant or buys a new mobile home which is then assembled on-site. The curious distinction from the more common rent control of multi-family apartment units is that in a mobile home park the landlord owns the land (the pad) and the tenant owns the improvements (the coach). If there is no vacancy decontrol (ability on the part of the landlord to adjust the pad rent to the market rent when the coach is sold), the net present value of the anticipated future rent savings should be capitalized into the sales price of the coach if it stays in place. 46 In more rigid rent control systems, controls are tied to the unit rather than the tenant. 83 And usually it does as mobile homes are seldom moved once they are located on a pad. This unique aspect of rent control for mobile home parks has been the subject of some empirical analysis. Rent control systems for mobile home parks provide an opportunity to investigate the economic impact of the rent control policy on the tenants of mobile parks (i.e., the mobile coach owners). Rapidly rising housing and land prices in California explain rising pad rents in many of the state’s mobile home parks. As a consequence, renters in many jurisdictions responded by launching efforts to have rent controls enacted into law. In 2003, ninety seven California cities and eight counties had some sort of mobile home rent control. In a series of papers published in the late 1980s, Hirsch and his colleagues (Hirsch, 1988, and Hirsch and Hirsch 1988) examined the impacts of rent controls on mobile homes in California. As alluded to previously, their analysis rests on the insight that ownership of mobile home living space is divided between the owner of the coach and the owner of the land, which is then leased to coach owners. Therefore, a coach atop a rent-controlled pad can be expected to sell at a premium. Hirsch et al’s studies contain estimates of these values. Basing their analysis on a relatively small sample of observed transactions from the mid-1980s, they estimated that sales prices were boosted by 32 percent because of rent controls, other things equal. An updated analysis by Hirsch and Rufolo (1999) focused on a single mobile home park in Oceanside, California. A hedonic regression based on a sample of 90 84 mobile home sales over the period 1986-1992 found that, other things equal, rent controls explain a price premium of eight percent. More recently, Quigley (2002) presented an economic analysis of mobile home rent control based on a single mobile home park in San Rafael, California. The study documented the arms-length sales of 40 mobile homes in that park during a three-year period. The study estimated the average value of the rent control premium varied from $16 to $160,000. Price premiums for coaches enjoying rent control benefits in this study averaged an impressive 366 percent. 3. Model Setup This section lays out the methodology adopted to model the capitalization of rent control. The approach to measuring housing price appreciation has advanced as a consequence of work by Bailey, Muth and Nourse (1963), Case and Shiller (1989), Case and Quigley (1992) and Harding, Rosenthal and Sirmans (2005) among others. In this paper, we employ repeat sales price indexes in the mobile home market in California to examine the impact of variation in rent control policies among jurisdictions, while controlling the socio-economics of the neighborhoods in question and of the physical attributes of the coaches themselves. As discussed in previous sections, due to the separate ownership of the mobile home coach and the pad where the coach parks, coach owners in a jurisdiction with rigid rent control can capitalize the anticipated savings from controlled rents when they sell their units. Our model extends the work by Bailey, 85 Muth and Nourse (1963) and Case and Shiller (1989) by decomposing the mobile home repeated sale house price differential into mobile home property value appreciation and the capitalization of the mobile home park rent control premium. 47 Consider a mobile home coach which is observed to have been purchased and sold in periods t andt τ + , respectively. The purchase price is ' 2 12 ' 1 tt x tt PAx e e γ αα = (3.1) Where A is the parameter to capture the common factors’ effect on all the units under study; 1t x is a vector of the property’s hedonic characteristics; 2t x is a vector of binary regulatory attributes, e.g. different rent control regimes; 1 α and 2 α are shadow prices for hedonic characteristics and regulatory attributes; and it γ is a price index parameter capturing aggregate price appreciation. Let the price vary continuously with time, at timet τ + , the selling price is ' 2 12 ' 1 tt x ttt PPx e e τ τ γ βτ β τ ττ ++ ++ = (3.2) If we assume the characteristics and their shadow prices did not change between the two transactions, then the price growth comes only from a change in the level of market prices. () tt tt PPe τ γ γ τ + − + = (3.3) This is exactly the repeat house price index model developed by Bailey, Muth and Nourse (1963) and Case and Shiller (1989). 47 Our decomposition process is similar to the model proposed by Harding, Rosenthal and Sirmans (2005) in which they modify the repeated sales house price index model by decomposing the housing price appreciation into the components of capital depreciation and impact of housing maintenances. 86 However, suppose the binary regulatory attributes change from 2 x to * 2 x , as a consequence of the adoption of a rent control ordinance during the time between the two transactions. If we still hold the shadow price vector 2 α constant, the price growth can be decomposed into property value appreciation and the impact of change in regulatory attributes. *' ' 22 2 ()() tt xx tit PPe e τ γγ α τ + − − + = (3.4) Taking logarithms and rearranging equations (3.3) and (3.4), we have, ln( ) t tt t t P P τ τ τ γ γε + ++ =− + (3.5) and * 22 ln( ) '( ) t tt t t P xx P τ τ τ γ γα υ + ++ =− + − + (3.6) where t τ ε + and t τ υ + are random error terms. For the case of mobile homes, provided all the other characteristics except for the rent control policy are time-invariant, as well as the shadow prices, the problem of selecting the proper specification of functional form on how qualitative and quantitative characteristics determine housing price growth has been eliminated. This elimination enhances the reliability of the estimates for the price index, γ . For a sample of mobile homes that are transacted at various time periods, we obtain ln( ) ' ( ) ( ) t tt t P DRC RC P τ τ γαωτ + + =+ − + (3.7) 87 where D is a matrix with elements of -1, 0 and 1, determined by both purchase and sale of each property; 48 γ is price index vector; RC is a vector of indicator variables for various rent control regimes, 49 and the modified repeated sales house price index model expressed in equation (3.7) allows us to decompose the capitalization of any rent control premium from property value appreciation. Redfearn (2005) has pointed out that a source of significant bias in aggregate indexes is the ignorance of local dynamics. Case and Quigley (1992) acknowledged that it is appropriate to control statistically for the varying characteristics of properties in inferring price trends. Clapham et al. (2004) compared the stability of the repeat sales index and the hedonic index and found that character-based hedonic indexes appear to be substantially more stable than repeat-sales indexes. Hence we can modify equation (3.7) ln( ) ' ( ) ( ) t tt t P DRC RC Z P τ τ γαθµτ + + =+ − + + (3.8) where θ is the parameter; Z is a vector of control variables capturing the location effect (measured by county), as well as hedonic effects (coach structure such as single-, double-, or triple-width, size, and neighborhood quality, etc.). This additional vector measures the proportional effects of location and the coach’s amenities on the price growth rate. Harding et al. (2005) noticed the impact of depreciation and maintenance on existing house resale price. Depreciation is usually age-related, and at a non-linear 48 -1 indicates the property was purchased in period t; and 1 indicates the property was sold in period t; 0 for all the other periods. 49 1 indicates the property is under rent control; 0 means no rent control policy adopted. 88 rate (Shilling et al (1991), Lee at al (2005)). Following Harding et al., the age change of the unit between the two transactions, denoted τ , should be included in the repeat sales model. The non-linear depreciation function specification can avoid the perfect collinearity issue noted by Bailey, Muth and Nourse (1963). 50 Therefore, equation (3.8) can be rewritten as: ln( ) ' ( ) log( ) ( ) t tt t P DRC RC Z P τ τ γαθρτµτ + + =+ − + + + (3.9) Where log( ) τ is a vector of logarithm age change for the properties between the two transactions, and ρ is the elasticity of house price appreciation with respect to the change in age between the sales dates. 4. Data In this research used twenty years of transactions data for mobile homes in seven counties of California (Los Angeles, Orange, Riverside, San Bernardino, San Diego, Santa Clara and Ventura). After geo-coding and matching with census tracts, there were 201,228 records from the period beginning of January 1983 and ending in May 2003. As we are interested in creating hybrid repeat-sale indexes, our focus is on the coaches with two transactions, the first purchase and one resale, thus records for single sales were removed from the data set 51 . Other records were eliminated because of incomplete city or mobile home park (Park ID) information. Outliers 50 The index indicator variable t D is perfectly collinear with the age difference variable τ if we assume the depreciation is linear and include τ into repeat sale index calculation directly. 51 For those properties with more than two transactions, only one resale record is included in the analysis. 89 were also removed. 52 These modifications left a 20-year population of 137,221 sales in the data set. Each record in the original data set is a mobile home transaction and includes the county, city, address, mobile home park name and ID, the original sale price and resale price, the dates of sale (year, month, and day), and measures of other attributes of the unit such as model year, manufacturer, width, length, and type (single, double, or triple) and so on. These data were combined with census data by census tract, including median household income, changes in median household income, vacancy rate, proportion of the elderly ( ≥ 65 years), unemployment rate, and proportion of households with public assistance income, etc. Finally, city-specific mobile home park rent control information was incorporated in the merged data set, including an indicator of the nature of the rent control regime. 53 There are 201 municipalities in the seven counties included in this study, among which 49 cities have a rent control ordinance. As many as 48,664 transactions in the data set (35.5 percent) were actually in jurisdictions where there was rent control. Among those, around half were under rigid rent control. From Figures 3.1(a) – (c), we can see that the share of transactions as measured by square footage, value and number has increased significantly over the twenty-year period under study. 52 Records with incomplete geographic information and the top and bottom 1 percent of records based on the variables Original Sales Price (Constant $), Resale Price (Constant $), Size and Average Annual Growth Rate were removed from the data set. We also explored the impact of focusing on properties where 0.5 < (Constant Resale Price/Constant Original Sales Price) < 3.0. We found that the results were robust to the imposition of this constraint. See Table 1 for descriptions of the variables. 53 For cities with rent control, dummy variables were including starting in the quarter in which rent control was implemented. We also included a dummy variable that indicates whether or not there is vacancy decontrol. No vacancy decontrol reflects a more rigid rent control regime. 90 Figure 3.1(a) Total traded square footage (in thousands) 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Without rent control Rent control with vacancy decontrol Rent control w/o vacancy decontrol Figure 3.1(b) Total traded value (constant $, base=1996) (in millions) 0 50 100 150 200 250 300 350 400 450 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Without rent control Rent control with vacancy decontrol Rent control w/o vacancy decontrol 91 Figure 3.1(c) Number of transactions 0 2,000 4,000 6,000 8,000 10,000 12,000 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Without rent control Rent control with vacancy decontrol Rent control w/o vacancy decontrol The share of rent controlled units among the total mobile home transactions in the seven counties grew from almost thirteen percent in the early 1980’s to more than forty percent in 2003; moreover, the growth of those under rigid rent control outpaces the growth of transactions of units under mild rent control since the middle 1990s. The change in the share of the square footage traded and the value traded attributable to rent-controlled units is of the same order. Table 3.1 presents the annual growth rates in rent-controlled units among the transactions as measured by square footage, value and number. The growth rates were 13.02, 13.19 and 13.17 percent, respectively. Clearly, rent control policies in the mobile home marketplace in California are taking on increasing importance. 92 Table 3.1 Annual growth rates (1984-2002) Total units traded (square footage) Total traded value (constant $) Number of units transacted Year All Rent control group Rent control with vacancy decontrol Rent control without vacancy decontrol All Rent control group Rent control with vacancy decontrol Rent control without vacancy decontrol All Rent control group Rent control with vacancy decontrol Rent control without vacancy decontrol 1984 24.11% 44.39% 71.27% 24.19% 25.43% 49.46% 76.55% 31.17% 24.88% 47.47% 77.27% 24.51% 1985 13.78% 28.20% 33.25% 22.97% 14.89% 25.09% 29.07% 21.47% 13.27% 26.38% 30.48% 21.88% 1986 6.02% 15.01% 22.56% 6.53% 6.22% 17.08% 24.15% 10.25% 5.82% 14.86% 24.02% 4.10% 1987 8.82% 13.45% 14.54% 12.06% 8.75% 15.34% 16.62% 13.95% 8.75% 15.81% 14.79% 17.24% 1988 12.16% 28.74% 26.48% 31.72% 13.64% 26.75% 24.29% 29.50% 10.32% 29.43% 26.99% 32.77% 1989 11.00% 32.23% 42.40% 19.32% 12.13% 33.10% 41.92% 23.66% 9.10% 31.37% 44.57% 14.08% 1990 -12.01% -1.28% -0.53% -2.41% -16.67% -6.13% -3.09% -9.85% -10.54% 0.57% -0.42% 2.22% 1991 -17.81% -11.20% -10.39% -12.45% -28.81% -24.19% -22.93% -25.84% -16.41% -8.76% -9.31% -7.87% 1992 0.78% 2.37% -0.68% 7.20% -16.46% -15.87% -18.15% -12.74% 1.84% 5.80% 2.96% 10.31% 1993 16.91% 40.11% 29.10% 56.21% -9.64% 16.53% 5.94% 30.13% 15.13% 37.59% 27.49% 52.60% 1994 18.61% 28.53% 25.32% 32.41% 1.15% 15.13% 5.10% 25.61% 16.68% 25.72% 22.48% 29.75% 1995 8.90% 10.56% 3.36% 18.79% 1.93% 2.15% -8.12% 11.14% 8.66% 10.46% 3.62% 18.48% 1996 -6.86% -15.09% -26.30% -3.93% -1.31% -8.23% -19.90% 0.22% -7.22% -15.23% -26.21% -3.98% 1997 7.15% 5.33% 4.97% 5.61% 19.61% 23.17% 24.15% 22.60% 7.27% 5.77% 6.25% 5.39% 1998 17.33% 11.96% 7.09% 15.66% 31.45% 21.54% 13.19% 26.43% 16.32% 12.32% 6.87% 16.65% 1999 6.86% 1.40% 10.69% -5.13% 16.72% 7.77% 23.57% -0.52% 5.67% 0.03% 10.21% -7.38% 2000 1.73% 8.44% 0.52% 14.93% 24.90% 40.64% 11.44% 59.67% 0.33% 6.08% -0.36% 11.66% 2001 5.66% 5.19% 14.96% -1.82% 8.51% 3.23% 20.28% -4.52% 6.40% 6.15% 14.00% 0.09% 2002 3.20% -0.98% -8.17% 5.05% 12.16% 8.07% 5.30% 9.66% 2.02% -1.54% -8.31% 4.42% Average annual growth rate 6.65% 13.02% 13.71% 13.00% 6.56% 13.19% 13.13% 13.79% 6.22% 13.17% 14.07% 13.00% 93 All Los Angeles Orange Riverside San Bernardino San Diego Santa Clara Ventura Original price (constant $, 1996) 40,775 40,696 45,362 39,691 36,940 38,102 46,410 38,699 (21150) (22183) (21313) (20029) (18324) (19407) (25154) (18691) Resale price (constant $, 1996) 34,118 33,207 38,982 28,567 22,085 30,925 54,441 42,687 (26201) (24331) (26981) (25116) (17834) (22767) (31546) (28392) 43,274 46,923 47,193 33,149 36,650 39,991 52,492 48,194 Median household income (constant $, 1996) (13524) (14011) (12135) (9862) (11097) (10462) (15100) (11262) 0.067 0.070 0.052 0.086 0.090 0.059 0.057 0.050 Proportion of households with public assistance income (0.047) (0.046) (0.039) (0.040) (0.053) (0.045) (0.044) (0.033) Proportion of persons ≥ 65 years old 0.132 0.103 0.117 0.244 0.115 0.151 0.114 0.122 (0.094) (0.055) (0.088) (0.160) (0.067) (0.083) (0.079) (0.052) Unit's size (sq. ft.) 1,127 1,094 1,199 1,160 1,093 1,103 1,158 1,113 (396) (411) (370) (422) (383) (389) (406) (376) Unit's age 19.23 19.10 19.28 16.01 19.09 20.75 18.82 20.85 (9.15) (9.39) (9.15) (8.51) (9.24) (9.07) (8.55) (8.61) Sample size 137,221 31,404 23,934 12,912 20,827 28,551 13,423 6,170 Note: Standard deviations are in parenthesis. Table 3.2 Descriptive statistics – means and standard deviations of continuous covariates 94 Table 3.2 provides descriptive statistics for the continuous covariates for the complete data set as well as for each county. The original sale price varies from an average of $36,940 (San Bernardino) to $46,410 (Santa Clara); while the average resale price varies from $22,085 (San Bernardino) to $54,441 (Santa Clara). The average original sale price and resale price of the seven counties are $40,775 and $34,118 respectively. Mobile homes are less likely to keep up with rapid residential land market appreciation. Declines in the values of mobile homes are more common than for “stick-built” houses because the land component is often not part of the selling price. 54 Census tract variables including Median Household Income, Proportion of Households with Public Assistance Income, and Proportion of Persons ≥ 65 Years Old were collected and also reported in Table 3.2. Other census tract-level variables, including Changes in Median Household Income, Vacancy Rate, and Unemployment Rate, were collected but were ultimately not of significance in the analysis so their summary statistics are not reported here. Riverside county had the lowest median household income ($33,149) accompanied by the highest elderly population (24.4 percent). Santa Clara county was at the other extreme with the highest median household income ($52,492) and was tied with two other counties for the second lowest proportion of elderly population (11.4 percent). Perhaps surprisingly, Los Angeles county had the lowest proportion of elderly at 10.3 percent. San Bernardino and Riverside counties had the 54 A more suitable comparable for mobile home sales would be sales of stick-built houses on leased land. 95 highest proportion of households benefiting from public assistance (9.0 and 8.6 percent, respectively) while Ventura and Orange counties had the lowest (5.0 and 5.2 percent, respectively). We ultimately used the census information to segment the data set. Specifically, high- (or low-) income is defined by whether the census tract median household income is above (or below) the median census tract household income in the corresponding county. High- (or low-) elderly population proportion is defined by whether the census tract older population proportion is above (or below) the median census tract elderly population proportion in the corresponding county. The smallest mobile homes are in San Bernardino and Los Angeles counties as measured by Unit’s Size (1,093 and 1,094 square feet) while the largest are in Orange county (1,199 square feet). Ventura and San Diego counties have the oldest units (20.85 and 20.75 years, respectively) while Riverside county has the youngest (16.14 years). 96 Table 3.3 Descriptive statistics – frequencies of categorical covariates All Los Angeles Orange Riverside San Bernardino San Diego Santa Clara Ventura Single 35,787 9,157 3,532 3,513 6,741 7,820 3,325 1,699 (26.1) (29.2) (14.8) (27.2) (32.4) (27.4) (24.8) (27.5) Double 94,110 20,394 18,997 8,338 13,265 19,457 9,412 4,247 (68.6) (64.9) (79.4) (64.6) (63.7) (68.1) (70.1) (68.8) Triple 7,324 1,853 1,405 1,061 821 1,274 686 224 (5.3) (5.9) (5.9) (8.2) (3.9) (4.5) (5.1) (3.6) Rent Control 48,664 8,828 1,092 7,865 8,959 10,871 5,948 5,101 (35.5) (28.1) (4.6) (60.9) (43.0) (38.1) (44.3) (82.7) Rent control with vacancy decontrol 24,341 7,035 1 5,720 3,401 4,745 252 3,187 (17.7) (22.4) (0.0) (44.3) (16.3) (16.6) (1.9) (51.7) Rent control without vacancy decontrol 24,323 1,793 1,091 2,145 5,558 6,126 5,696 1,914 (17.7) (5.7) (4.6) (16.6) (26.7) (21.5) (42.4) (31.0) 20,980 5,926 1 4,862 2,833 4,399 228 2,731 Adoption of rent control with vacancy decontrol (15.3) (18.9) (0.0) (37.7) (13.6) (15.4) (1.7) (44.3) 21,723 1,415 1,041 1,942 5,148 5,445 5,063 1,669 Adoption of rent control without vacancy decontrol (15.8) (4.5) (4.3) (15.0) (24.7) (19.1) (37.7) (27.1) Sample size 137,221 31,404 23,934 12,912 20,827 28,551 13,423 6,170 Note: Percentages to corresponding full or county samples are in parenthesis. 97 Table 3.3 provides descriptive statistics for the sample’s categorical covariates. Double-width mobile homes are the most common as indicated by a market share of almost 70 percent for the type “Double”. The second biggest share is for single-width or the “Single” category at a market share of 26 percent; triple-width or the “Triple” category represent the luxury high-end mobile homes and have a five percent share of the market. The market share of double-width in Orange county is about 15 percentage points higher than in Los Angeles, while the market share of single-width is about 15 percentage points lower perhaps reflecting stage of the evolution of the mobile home market as well as the role of mobile homes as a housing choice when the homes were originally sold and the mobile home parks developed. As noted earlier, a number of qualitative (“dummy”) variables were created which are critical to the analysis. First, communities are identified in which rent control of mobile home parks was present. Second, the policy is classified as rigid (by-laws which do not permit vacancy decontrol) or flexible (by-laws which permit vacancy decontrol), depending on whether or not vacancy decontrol is permitted. Then transactions within these jurisdictions were identified accordingly. Third, the first transaction after the adoption of a rent control policy is also indicated. The results of these classifications for the aggregate data set also appear in Table 3.3. The prevalence of rent control ordinances varies dramatically among the seven counties. While in the aggregate, less than five percent of the transactions in Orange county were in rent-controlled parks, 82.7 percent of the transactions in 98 Ventura county were in rent-controlled parks. As noted earlier, there was significant variation in these percentages between 1983 and 2003 with an increasing number and share of mobile home units being regulated by rent control policies over time. Of the 35.5 percent of mobile home transactions in our data set that are within rent controlled jurisdictions, roughly half of them are in jurisdictions with flexible rent control regimes and the rest are in jurisdictions with rigid rent control regimes. 55 5. Empirical Results We had a much larger and more comprehensive sample available to us than was employed in Hirsch (1988), Hirsch et al (1988, 1999) or Quigley (2002). As noted, we focused on 137,221 observations collected from over twenty years of mobile home transactions between 1983 and the early part of 2003. The ultimate transaction data base included only repeat- or multiple-sales along with descriptive information about the coach. Each sale was geo-coded permitting the census tract variables including Median Household Income (constant $, 1996), Proportion of Households with Public Assistance Income and Proportion of Persons ≥ 65 Years Old to be appended to each record. The first two are proxies for local amenity values as well as demand, while the third is a proxy for one of the components of demand 55 Rent control regimes may become less rigid or more rigid through time as policies evolve due to the changing economic and political environment. We were able to identify regimes which are currently rigid (no vacancy decontrol permitted) or flexible (vacancy decontrol permitted). In order to ascertain whether today’s regime accurately reflected the nature of the regime since 1983, we surveyed every city in our data set (201). We received 55 responses in total with 20 of them from cities with rent control in place. The results of the survey supported the approach taken. 99 for mobile home units -- as many older households choose mobile homes as a cost- effective housing choice in retirement. Table 3.4 GLS estimate for policy change indicators Impact of Rent Control Impact of Rent Control together with Other Effects (1) (2) (3) (4) Orange County -0.04 -0.04 -0.04 (-8.48) (-7.36) (-7.16) Riverside -0.27 -0.27 -0.27 (-46.08) (-44.32) (-44.50) San Bernardino -0.29 -0.29 -0.29 (-56.24) (-54.78) (-54.14) San Diego -0.03 -0.02 -0.03 (-5.80) (-4.68) (-5.23) Santa Clara 0.31 0.32 0.33 (51.79) (50.59) (53.98) Ventura 0.37 0.37 0.37 (42.55) (42.84) (42.12) Double 0.20 0.21 0.21 (55.68) (51.11) (50.90) Triple 0.29 0.29 0.29 (40.37) (39.51) (39.71) -1.37 -1.33 -1.30 Proportion of households with public assistance income (-40.06) (-35.32) (-34.44) Log of unit's age (quarter) -0.02 -0.02 (-7.57) (-6.96) -0.25 -0.26 Log of unit's age (quarter) × indicator of model before 1976 (-19.60) (-20.30) 0.02 0.02 0.02 Under rent control without vacancy decontrol (1.72) (2.45) (2.08) -0.01 -0.01 -0.007 Under rent control with vacancy decontrol (-1.54) (-0.72) (-0.82) 0.15 0.12 0.12 Adoption of rent control without vacancy decontrol (28.77) (23.19) (22.97) 0.0038 Adoption of rent control without vacancy decontrol × No. of quarters under rent control (30.44) -0.11 -0.09 -0.09 -0.08 Adoption of rent control with vacancy decontrol (-20.70) (-16.53) (-16.69) (-15.80) 0.412 0.496 0.497 0.499 R square Sample size 137,221 137,221 137,221 137,221 The working hypothesis was that because rigid rent control policies allow coach owners to pass on future pad rent savings to subsequent owners of the coach, 100 prices of coaches in those communities will increase more rapidly or decrease less rapidly than the prices of coaches in communities without rent control or flexible rent control. Thus, if rent control or the rigidity of the rent control regime influences the rate of change of coach prices, the relevant estimated coefficient will be positive and significant. Moreover, the biggest beneficiary is the first generation of mobile home owners when the rent control policy is/was adopted. The following coach owners are not able to realize a comparable benefit because they have paid for most of the premium for rent control upfront. Table 3.4 presents the results of GLS 56 estimation of equation (3.7), (3.8), and (3.9). In model (1), mobile homes values are positively influenced if rigid rent control policies are adopted but those will decrease when mild rent control policies are adopted. This set of relationships still holds when the proportional effects of 56 GLS is used to address the problem of heterogeneity. We estimate the rent control impact by decomposing the capitalization of rent control premium from a repeated-sales mobile home price index model using a three-stage GLS estimation approach. In the first stage, we use OLS to estimate the following mobile home price index model which decomposes the rent control premium capitalization from the mobile home price appreciation: ln( ) ' ( ) log() () t tt t P DRC RC Z P τ τ γαθρτµτ + + =+ − + + + , where ( ) ln i is natural logarithm function, t P τ + is the resale price after τ periods since the original sale at t , t P is the original sales price at t , D is a matrix composed of -1, 0 and 1; RC is a vector of indicator variables to capture the rent control premium capitalization; Z is a vector of hedonic characteristics capturing the location effect (measured by county), and coach structure characteristics, such as single, double, or triple-width, size, and neighborhood quality, etc., and µ is the normal distributed error term with variance varying by the time span between two transactions. In the second stage, we estimate a diffusion process of the mobile home price index by regressing the square term of residues, 2 ε , collected from the first stage OLS regression on a quadratic function of time span between the two sales, such that () 2 2 ετβτ γτ ω =+ + , where τ is the duration between original sale and resale (measured in quarters), and ω is a normally distributed error term. In the third stage we re-estimate the mobile home price index model using a GLS estimation approach weighted by the square root of diffusion variance, () ˆ ε τ , estimated from the second stage. Fore readability, the estimated mobile home price indices are not reported in the table. 101 other hedonic factors have been considered in model (2). Riverside and San Bernardino have the lowest growth rates, while Ventura outperforms all the other counties. The larger luxury units experience higher appreciation than smaller ones 57 . The coaches in the neighborhoods with a higher proportion of households with public assistance suffer from lower growth rates. For rent control policies, the first generation mobile home owners under rent control can benefit from the adoption of rigid rent control ordinances (rent control without vacancy decontrol), but experience a loss from mild rent control (rent control with vacancy decontrol). For the following generations of tenants, rigid rent control policy is no longer a significant factor in determining the resale price, while a mild rent control regime may continue contributing to lower resale price. Model (3) further extends model (2) by adding the depreciation factors, “log of unit’s age” and “log of unit’s age (quarter) × indicator of model before 1976”. The price appreciation slows down upon the aging of the property. According to the history of mobile home, Congress passed National Manufactured Housing Construction and Safety Standards Act in 1974; then in June 1976, the Federal Manufactured Home Construction and Safety Standards (or the “HUD Code”) became law. In terms of mobile home design, 1970 is a cutting point: before 1970, the mobile homes are more like big boxes; after 1970, they start getting closer to traditional house. The presence of the interactive variable, “log of unit’s age × indicator of model before 1976”, takes the obsolescence into consideration. Both 57 “Size” is highly correlated with “Double” and “Triple” dummy variables, hence excluded from the model. 102 factors are significant and negative, showing that depreciation/obsolescence should be considered in the repeated sale price index model. Because rigid rent control practically freezes real rents paid, in a market where demand increases in the long run such as in the market for mobile homes, the longer this restriction is in force, the bigger would be the price gap, between the controlled units and the uncontrolled units. Model (4) replaces the simple dummy variable “under rent control without vacancy decontrol” by the interactive variable “under rent control without vacancy decontrol × number of months under rent control at time of transaction”, and also replaces the dummy variable “adoption of rent control without vacancy decontrol” by the interactive variable “adoption of rent control without vacancy decontrol × number of months under rent control at time of transaction”. The two interactive variables are both significant, and keep the same sign as their simple dummy counterpart in model (3). In another words, rigid rent control pushes up mobile homes’ resale prices. The adoption of rigid rent control has the greater impact, which is to say that the first generation mobile home owners can capitalize the stabilization of future real rents by charging a higher resale price. For any following owners, it appears that exceptionally low real rent under rigid rent control may still lead to relatively higher resale prices. 103 Figure 3.2 Logarithm price growth rate by income and age sub-groups 58 (a) Logarithm of price growth rate by income group (b) Logarithm of price growth rate by age group 58 These figures present the box plot of the logarithm of price growth rates by various sub-groups. The boxes contain the middle 50 percent of the data. The upper edges of the boxes indicate the 75 th percentile of the data set, and the lower edges indicate the 25 th percentile. The lines in the boxes indicate the median value of the data. The ends of the vertical lines indicate the minimum and maximum data values. The logarithm of price growth rate is defined as: ln( ) t t P P τ + , where ( ) ln i is natural logarithm function, t P τ + is the resale price after τ periods since the original sale at t , t P is the original sales price at t . 104 From a policy perspective it might be valuable to know which income-age group can benefit the most from a rent control policy, and also, who might suffer the most. Figure 3.2(a) and 3.2(b) present the plots of LogRatio by different median household income census tracts and different proportion of elderly population census tracts. The differences between higher and lower income communities, as well as older and younger communities are visible. Hence, we refine the full sample and re- estimate the index with proportional location and hedonic factors for each income/age group, in order to estimate rent control’s impact under different scenarios. The re-estimated results for different median household income census tract groups are presented in Table 3.5(a). Adoption of rigid rent control leads to higher increase in resale price for coaches in wealthier communities, but homeowners in poorer communities suffer less from the adoption of mild rent control. On the other hand, the future homeowners in wealthier neighborhoods can enjoy relatively greater benefit from rigid rent control compared to a the future homeowners in poorer communities, but the latter have marginal gain from the mild ret control which is not applicable to the former. 105 Table 3.5(a) GLS estimates for logarithm of price growth rate, by census tract household income Higher median household income Lower median household income Orange County 0.04 -0.05 (4.97) (-6.49) Riverside -0.15 -0.33 (-16.28) (-37.18) San Bernardino -0.21 -0.33 (-28.04) (-42.19) San Diego -0.05 0.00 (-7.66) (0.06) Santa Clara 0.198 0.385 (16.62) (44.95) Ventura 0.38 0.37 (22.62) (33.09) Double 0.17 0.23 (26.06) (44.17) Triple 0.24 0.33 (23.66) (30.40) -2.03 -0.81 Proportion of households with public assistance income (-23.94) (-16.61) Log of unit's age (quarter) -0.03 -0.02 (-5.87) (-4.97) -0.234 -0.256 Log of unit's age (quarter) × indicator of model before 1976 (-12.22) (-14.57) 0.06 0.004 Under rent control without vacancy decontrol × No. of quarters under rent control (4.43) (0.29) -0.02 0.02 Under rent control with vacancy decontrol (-1.59) (2.12) 4.36E-03 4.13E-03 Adoption of rent control without vacancy decontrol × No. of quarters under rent control (19.86) (27.13) -0.11 -0.04 Adoption of rent control with vacancy decontrol (-14.26) (-5.75) R square 0.44 0.54 Sample size 52,921 84,300 Table 3.5(b) provides the estimates for census tracts with higher and lower proportions of elderly population. Younger neighborhoods benefit from the adoption of rigid rent control ordinances more than the older neighborhoods, while the latter suffer less from new mild rent control policies. Clearly, younger mobile home 106 owners have a longer anticipated benefit period. The following generations of mobile home owners in older neighborhoods are relatively better off from both rigid and mild rent control ordinances and slightly better off under the former. Interestingly, obsolescence depreciates the property value less in the elderly neighborhood than in younger neighborhood. Table 3.5(b) GLS estimates for logarithm of price growth rate, by census tract proportion of elderly population High proportion of elder Low proportion of elder Orange County -0.03 -0.04 (-4.00) (-5.11) Riverside -0.36 -0.20 (-40.19) (-24.89) San Bernardino -0.40 -0.19 (-52.23) (-25.54) San Diego -0.08 0.03 (-11.14) (4.04) Santa Clara 0.34 0.34 (39.71) (38.10) Ventura 0.33 0.41 (26.89) (33.32) Double 0.22 0.19 (35.66) (35.24) Triple 0.32 0.23 (31.97) (20.75) -0.95 -1.17 Proportion of households with public assistance income (-15.02) (-24.73) Log of unit's age (quarter) -0.02 -0.03 (-4.46) (-5.93) -0.16 -0.33 Log of unit's age (quarter) × indicator of model before 1976 (-8.82) (-17.95) 0.08 -0.04 Under rent control without vacancy decontrol × No. of quarters under rent control (5.66) (-3.54) 0.05 -0.03 Under rent control with vacancy decontrol (3.43) (-3.24) 3.14E-03 4.99E-03 Adoption of rent control without vacancy decontrol × No. of quarters under rent control (20.86) (22.83) -0.01 -0.14 Adoption of rent control with vacancy decontrol (-0.96) (-18.10) R square 0.49 0.52 Sample size 69,547 67,674 107 Table 3.5(c) GLS estimates for logarithm of price growth rate,, by census tract median household income and proportion of elderly population High income & high proportion of elder High income & low proportion of elder Low income & high proportion of elder Low income & low proportion of elder Orange County 0.10 0.01 0.02 -0.10 (8.45) (0.78) (1.93) (-9.57) Riverside -0.16 -0.11 -0.31 -0.33 (-8.12) (-10.16) (-25.45) (-25.87) San Bernardino -0.45 -0.10 -0.32 -0.33 (-32.63) (-10.77) (-28.81) (-27.43) San Diego -0.09 -0.01 0.02 0.01 (-8.76) (-1.08) (1.46) (1.09) Santa Clara 0.35 0.19 0.44 0.34 (16.29) (12.84) (36.68) (27.50) Ventura 0.36 0.38 0.43 0.33 (11.70) (19.15) (28.24) (19.13) Double 0.19 0.16 0.22 0.23 (17.27) (19.40) (30.75) (31.19) Triple 0.28 0.17 0.33 0.30 (18.41) (12.05) (24.33) (16.54) -0.53 -2.05 -0.51 -0.87 Proportion of households with public assistance income (-3.25) (-20.88) (-6.58) (-13.25) Log of unit's age (quarter) -0.06 -0.01 -0.04 -0.02 (-8.26) (-2.23) (-6.24) (-2.36) -0.16 -0.25 -0.21 -0.34 Log of unit's age (quarter) × indicator of model before 1976 (-5.52) (-9.61) (-9.10) (-12.45) 0.15 -0.03 0.001 0.01 Under rent control without vacancy decontrol × No. of quarters under rent control (6.93) (-1.61) (0.06) (0.77) 0.02 -0.01 0.06 0.003 Under rent control with vacancy decontrol (0.66) (-0.54) (3.66) (0.19) 2.68E-03 7.70E-03 3.62E-03 4.93E-03 Adoption of rent control without vacancy decontrol × No. of quarters under rent control (10.41) (18.62) (18.99) (18.71) 0.03 -0.15 0.003 -0.08 Adoption of rent control with vacancy decontrol (2.81) (-14.45) (0.32) (-6.51) R square 0.41 0.49 0.53 0.55 Sample size 24,826 28,095 44,721 39,579 In Table 3.5(c), we show the results for further refined subsamples by both median household income and proportion of elderly population at the census tract level. When we consider income and age at the same time, age is a higher order factor than income: both younger communities, with higher or lower income, enjoy 108 greater benefit from the adoption of rigid rent control than the older communities. Homeowners in younger communities with higher median household income are the biggest beneficiaries. For the newly adopted mild rent control policies, homeowners in older communities with higher median household income might be the only beneficiaries, and all the rest will suffer, especially the ones in younger communities with higher median income. The future homeowners in the former community will continue benefiting from rigid rent control but to a lesser extent. Figure 3.3 Comparison in indexes: no rent control vs. change in rent control policy (adoption of rent control without decontrol) 30 40 50 60 70 80 90 199201 199203 199301 199303 199401 199403 199501 199503 199601 199603 199701 199703 199801 199803 199901 199903 200001 200003 200101 200103 200201 200203 200301 Quarter Adoption of rent control without decontrol No rent control Adoption of rent control without decontrol 109 Figure 3.3 presents the simulated comparison of indexes between units in the jurisdictions without rent control and units in hypothetical jurisdictions which adopted rigid rent control in the beginning of 1993. 59 The price index of the latter diverges from the former soon after the adoption of the policy. Table 3.6 Resale price comparison for an average structure in Los Angeles The following Tables present the estimated resale prices for the average mobile home units (in each subgroup) under different rent control regimes in Los Angeles County. The average unit is defined as the one with average structure (double, triple), average age, in census tract with average proportion of households with public assistance income, etc. We assume that rent control ordinance effective date is January 1, 1993 for the units with adoption of rent control between the first and the second transactions. By using the average units' characteristics and the parameters estimated in Table 3.5, we can simulate the resale prices as presented here. Rent control regime Resale price comparison for an average 19 years old structure No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol Resale price $39,515 $40,301 $39,515 $45,969 $36,380 Price increase -$1,229 -$443 -$1,229 $5,225 -$4,363 Growth rate (percentage) -3.02 -1.09 -3.02 12.82 -10.71 Growth out from rent control (percentage) - 1.93 0 15.84 -7.69 In Tables 3.6, we report the estimated resale prices for the average mobile home units (in each subgroup) under different rent control regimes in Los Angeles County. The average unit is defined as the one with average structure (double, triple), average age, in census tract with average proportion of households with public assistance income, etc. We assume that rent control ordinance effective date is January 1, 1993 for the units have rent control adopted between the two transactions. 59 The estimated repeated sale price index is included in the Appendix B. 110 By using the average units' characteristics and the parameters estimated in Table 3.5, we can simulate the resale prices. Table 3.6 exhibits the estimated resale price for the average mobile home in Los Angeles County at the aggregate level under different rent control regimes. We find that if the unit was under a rigid rent control regime before the first purchase, its real growth rate in price would be 1.9 percentage points higher than comparable units in a market without rent control or units under a flexible rent control regime. In contrast with an unchanged rent control policy environment, the adoption of rent control between the first and second transactions leads to a greater and more significant change in prices. More rigid rent control leads to a growth rate of 15.8 percentage points more than the units in a market without rent control, or an increase in coach value of $5,225. In contrast, flexible rent control results in a growth rate 7.7 percentage points lower than the coaches in a market without rent control, or a loss in value of $4,363. Figure 3.4 (a)-(d) is the graph presentation of the simulation results. 60 Consistent with the regression results, the adoption of rigid rent control has the greatest impact on the properties’ price in the wealthier and younger communities, and the least effect on the properties’ price in the wealthier and elder communities. But the bottom line is that the adoption of rigid rent control leads to inflated price in all neighborhoods. 60 The table presentation is in Appendix B. 111 Figure 3.4 Simulation results: resale price comparison for Los Angeles county The following figures present the estimated resale prices for the average 19 years old mobile home units in each subgroup in Los Angeles County, under different rent control regimes. The average unit is defined as the one with average structure (double, triple), in census tract with average proportion of households with public assistance income, etc. Figure 3.4(a) Resale price comparison $3 0,0 00 $3 2,0 00 $3 4,0 00 $3 6,0 00 $3 8,0 00 $4 0,0 00 $4 2,0 00 $4 4,0 00 $4 6,0 00 $4 8,0 00 No rent contro l Under rent co ntrol without vacancy decontrol Under rent contro l with vacancy d econtro l Ado ption o f rent contro l witho ut vacancy decontrol Ado ption o f rent co ntrol with vacancy d econtro l 112 Figure 3.4(b) Resale price comparison-median household income $30,000 $32,000 $34,000 $36,000 $38,000 $40,000 $42,000 $44,000 $46,000 $48,000 No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol High Median Household Income Low Median Household Income Figure 3.4(c) Resale price comparison-age community $30,000 $35,000 $40,000 $45,000 $50,000 $55,000 No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol High Proportion of Elderly Population Low Proportion of Elderly Population 113 Figure 3.4(d) Resale price comparison-median household income and elder proportion $30,000 $35,000 $40,000 $45,000 $50,000 $55,000 No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol High Income & High Proportion of Elder High Income & Low Proportion of Elder Low Income & High Proportion of Elder Low Income & Low Proportion of Elder 6. Conclusions Mobile homes are usually owned by households that pay a periodic rent for use of the land in a mobile home park (a pad) on which the coach is located. Because mobile homes tend not to be mobile but rather fixed on the pad on which they are initially located, and since the pads are rented, economic theory suggests that the imposition of rent control will lead to the capitalization of future rent savings when a coach is sold. That is, the buyer will not only pay for the coach but also for the net present value of the expected savings associated with the constrained future pad rent obligations to the landlord. Our results support the hypothesis, based on a more extensive and timely data set than had been employed in similar prior studies in California. Results from our 114 extensive analysis suggest that lower income group receives a slightly larger first time rigid rent control premium (adoption of rigid rent control increases the coach value by 17.3 percent for higher income group and 17.6 percent for lower income group, respectively). On the other hand, the higher income group benefits more from continuous rigid rent control policy while the lower income group will have marginal gain from mild rent control. The higher income group also has to face greater losses from the adoption of flexible rent control (coach value decrease by 9.6 percent for higher income group and 4 percent for lower income group). Younger owners on average can get about a 20.6 percent premium for adoption of rigid rent control while elderly owners on average benefit from a 13.7 percent premium which is 7 percentage points lower than the former. But the elderly owners are able to gain from both rigid and flexible rent control policies which adopted before the first transaction through 8.4 and 5.1 percent increase respectively in their coach value, while the younger owners may have to suffer a decrease between 3 to 4 percent. Moreover, the younger owners also have to experience a big drop in the coach value for about 12.4 percent if a flexible rent control is adopted. When we combine the two factors of income and age, higher income younger owners are the biggest winners from the adoption of rigid rent control which brings them a roughly 34 percent increase in coach value. In contrast, higher income elderly owners on average benefit the least from adoption of rigid rent control with a 9.8 percent increase in coach value, but they are the only group enjoying significant price growth at averagely 14 percent under rigid rent control. On the contrary, lower 115 income elderly owners are the only group enjoying mild price growth under flexible rent control at around 6 percent. Comparing to the older communities, the younger owners suffer significant drop in the coach value when flexible rent control is adopted. 116 Bibliography Alston, Richard M., J. R. Kearl, and Michael B. Vaughan (1992), “Is There a Consensus among Economists in the 1990’s?” American Economic Review, 82, 203- 209. Anglin, Paul M., Ronald Rutherford, and Thomas M. Springer (2003), “The Trade- off Between the Selling Price of Residential Properties and Time-on-the-Market: The Impact of Pricing Setting,” Journal of Real Estate Finance and Economics, 26(1), 95- 111. Archer, W. R., D. C. Ling, and G. A. McGill (1996), “The Effect of Income and Collateral Constraints on Residential Mortgage Terminations,” Regional Science and Urban Economics, 26(3-4) 235-261. Arnold, Michael A. (1999), “Search, Bargaining and Optimal Asking Prices,” Real Estate Economics, 27(3), 453-481. Arnold, Michael A. (2000), “Costly Search, Capacity Constrains, and Bertrand Equilibrium Price Dispersion,” International Economic Review, 41(1), 117-131. Arnott, Richard and Masahiro Igarashi (2000), “Rent Control, Mismatch Costs and Search Efficiency,” Regional Science and Urban Economics, 30, 249-288. Arnott, Richard (1995), “Time for Revisionism on Rent Control?” The Journal of Economic Perspectives, 9(1), 99-120. Bailey, Martin J., Richard F. Muth, and Hugh O. Nourse (1963), “A Regression Model for Real Estate Price Index Construction,” Journal of the American Statistical Association, 58, 933-942. Belkin, Jacob, Donald J. Hempel, and Denies W. McLeavey (1976), “An Empirical Study of Time on Market Using Multidimensional Segmentation of Housing Markets,” AREUEA Journal, 4(2), 57-75. Benveniste, Lawrence, Dennis R. Capozza, and Paul J. Seguin (2001), “The Value of Liquidity”, Real Estate Economics, 29(4), 633-660. Berger, Lawrence A., Paul R. Kleindorfer and Howard Kunreuther (1989),"A Dynamic Model of the Transmission of Price Information in Auto Insurance Markets", Journal of Risk and Insurance, 56(1), 17-33. Black, F. and M. S. Scholes (1973). “The Pricing of Options and Corporate Liabilities,” Journal of Political Economy, 81, 637-654. Borenstein, Severin and Nancy L. Rose (1994),"Competition and Price Dispersion in the U.S. Airline Industry," Journal of Political Economy, 102(4), 653-683 Braverman, Avishay (1980), “Consumers Search and Alternative Market Equilibria,” Review of Economic Studies, 47(3), 487-502. 117 Brennan, M. J., and E. S. Schwartz (1985), “Determinants of GNMA Mortgage Prices,” Journal of the American Real Estate and Urban Economics Association 13, 209-228. Burdett, K. and K. Judd (1983), "Equilibrium Price Distributions," Econometrica, 51, 955-970. Buser, S. A., and P. H. Hendershott (1984), “Pricing Default Free Mortgage,” Housing Finance Review, 3, 405-429. Butters, Gerard R. (1977), “Equilibrium Distributions of Sales and Advertising Prices,” Review of Economic Studies, 44(3), 465-491. Calhoun, A. C., and Y. Deng (2002), “A Dynamic Analysis of Fixed- and Adjustable-Rate Mortgage Terminations,” Journal of Real Estate finance and Economics, 24(1-2), 9-33. Case, Bradford and John M. Quigley (1991), “The Dynamics of Real Estate Prices,” Review of Economics and Statistics, 73(1), 50-58. Case, Karl E. and Robert J. Shiller (1989), “The Efficiency of the Market for Single Family Homes,” American Economic Review, 79(1), 125-37. Chade, Hector and Virginia Vera de Serio (2002), “Pricing, Learning, and Strategic Behavior in A Single-sale Model,” Economic Theory, 19(2), 333-353. Chen, Yongmin and Robert R. Rosenthal (1996a), “Asking Prices as Commitment Devices,” International Economic Review, 37(1), 129-155. Chen, Yongmin and Robert R. Rosenthal (1996b), “On the Use of Ceiling-Price Commitments by Monopolists”, RAND Journal of Economics, 27(2), 207-220. Chinloy, Peter T. (1978), “An Empirical Model of the Market for Resale Homes,” Journal of Urban Economics, 7(3), 279-292. Clapham, Eric, Peter Englund, John M. Quigley, and Christian L. Redfearn (2004), “Revisiting the Past: Revision in Repeat Sales and Hedonic Indexes of House Prices”, unpublished working paper. Cox, D. R. (1972), “Regression Models and Life-Tables,” Journal of the Royal Statistical Society, B 34(2), 187-220. Cox, D. R. (1975), “Partial Likelihood,” Biometrika, 62, 269-276. Cunningham, D., and C. Capone (1990), “The Relative Termination Experience of Adjustable to Fixed-Rate Mortgage,” Journal of Finance 45(5), 1687-1703. Dahlby, Bev and Douglas S. West (1986), "Price Dispersion in an Automobile Insurance Market," Journal of Political Economy, 94(2), 418-438. Dana, James D. Jr. (1999), “Equilibrium Price Dispersion under Demand Uncertainty: The Roles of Costly Capacity and Market Structure,” RAND Journal of Economics, 30(4), 632-660. 118 Deng, Y. (1997), “Mortgage Termination: An Empirical Hazard Model with Stochastic Term Structure,” Journal of Real Estate Finance and Economics, 14(3), 309-331. Deng, Y., and J. M. Quigley (2002), “Woodhead Behavior and the Pricing of Residential Mortgage,” USC Lusk Center working paper 2003-1005. Deng, Y., J. M. Quigley, and R. Van Order (1996), “Mortgage Default and Low Downpayment Loans: The Cost of Public Subsidy,” Regional Science and Urban Economics, 26(3-4), 263-285. Deng, Yongheng, John M. Quigley, and Robert Van Order (2000), “Mortgage Terminations, Heterogeneity and the Exercise of Mortgage Options,” Econometrica, 68(2), 275-307. Diamond, Peter (1987), “Consumer Differences and Prices in a Search Model,” Quarterly Journal of Economics, 102(2), 429-436. Diamond, Peter (1993), “Search, Sticky Prices, and Inflation,” Review of Economic Studies, 60(1), 53-68. Dunn, K. B., and J. J. McConnell (1981), “Valuation of Mortgage-Backed Securities,” The Journal of Finance, 36, 599-617. Early, Dirk W. and Edgar O. Olsen (1998), “Rent Control and Homeless,” Regional Science and Urban Economics, 28, 797-816. Findley, M. C., and D. R. Capozza (1977), “The Variable Rate Mortgage: An Option Theory Perspective,” Journal of Money, Credit and Banking, 9, 356-364. Fishman, Arthur (1992), “Search Technology, Staggered Price-Setting, and Price Dispersion,” American Economic Review, 82(1), 287-298. Garbade, Kenneth D. and William L. Silber (1976), “Price Dispersion in the Government Securities Market,” Journal of Political Economy, 84(4), 721-740. Genesove, David and Christopher J. Mayer (1997), “Equity and Time to Sale in the Real Estate Market,” American Economic Review, 87(3), 255-269. Genesove, David and Christopher J. Mayer (2001), “Loss Aversion and Seller Behavior: Evidence from the Housing Market,” Quarterly Journal of Economics, 116(4), 1233-1260. Glaeser, Edward L. and Erzo F.P. Luttmer (2003), “The Misallocation of Housing Under Rent Control”, The American Economic Review, 93(4), 1027-1046. Glower, Michel, Donald R. Haurin, and Patric H. Hendershott (1988), “Selling Time and Selling Price: The Impact of Seller Motivation,” Real Estate Economics, 26(4), 719-740. Green, Jerry and John B. Shoven (1986), “The Effect of Interest Rates on Mortgage Prepayment,” Journal of Money, Credit and Banking, 36(1), 41-58. 119 Guasch, J. Luis and Robert C. Marshall (1985), “An Analysis of Vacancy Pattern in the Rental Housing Market,” Journal of Urban Economics, 17(2), 208-229. Gyourko, Joseph and Linneman, Peter (1989), “Equity and Efficiency Aspects of Rent Control: An Empirical Study of New York City,” Journal of Urban Economics, 26(1), 54-74. Hamilton, James L. (1987), "Market information and price dispersion: Unlisted stocks and NASDAQ," Journal of Economics and Business, 39(1), 67-80. Harding, John P., Stuart S. Rosenthal, and D.F. Sirmans (2005), “Depreciation of Housing Capital, Maintenance, and the Gains from Homeownership Estimates From a Repeat Sales Model,” unpublished working paper. Haurin, Donald R. (1988), “The Duration of Marketing Time of Residential Housing,” Journal of the American Real Estate and Urban Economics Association, 16(4), 396-410. Haurin, Donald R., Jessica L. Haurin, Taylor Nadauld, and Anthony Sanders (2006), “List Prices, Sale Prices, and Marketing Time: An Application to U.S. Housing Markets,” working paper. Hendershott, P. and R. Van Order (1987), “Pricing Mortgages: An Interpretation of Models and Results,” Journal of Financial Services Research, 1, 77-111. Hirsch, Werner Z. and Anthony M. Rufolo (1999), “The Regulation of Immobile Housing Assets under Divided Ownership,” International Review of Law and Economics, 19, 383-397. Hirsch, Werner Z. and Joel G. Hirsch (1988), “Legal-Economic Analysis of Rent Controls in A Mobile Home Context: Placement Values and Vacancy Decontrol,” UCLA Law Review, 35, 399-467. Hirsch, Werner Z. (1988), “An Inquiry into Effects of Mobile Home Park Rent Control,” Journal of Urban Economics, 24, 212-226. Ho. Michael H.C. (2003), “Liquidity and Price Differentials: Evidence in the Hong Kong Residential Re0sale Market”, Housing Studies, 18(5), 745-763. Horowitz, Joel L. (1992), “The Role of the List Price in Housing Markets: Theory and an Econometric Model,” Journal of Applied Econometrics, 7(2), 115-129. Hotelling, Harold (1929), “Stability in Competition,” Economic Journal, 39, 41-57. Kanemoto, Yoshitsugu (1997), “The Housing Question in Japan,” Regional Science and Urban Economics, 27, 613-641. Kau, J. B., and D. C. Keenan (1995). “An Overview of the Option-Theoretic Pricing of Mortgages,” Journal of Housing Research, 6(2), 217-244. Kau, J. B., D. C. Keenan, W. J. Muller, and J. F. Epperson (1990), “The Valuation and Analysis of Adjustable Rate Mortgages,” Management Science, 36(12), 1417- 1431. 120 Kluger, Brian D. and Norman G. Miller (1990), “Measuring Residential Real Estate Liquidity,” AREUEA Journal, 18(2), 145-159. Lach, Saul (2002), “Existence and Persistence of Price Dispersion: An Empirical Analysis,” Review of Economics and Statistics, 84(3), 433-444. Lazear, Edward P. (1986), “Retail Pricing and Clearance Sales,” American Economic Review, 76(1), 14-32. Leung, Charles Ka Yui, Youngman Chun Fai Leong, and Ida Yin Sze Chan (2002), “TOM: Why Isn’t Price Enough?” International Real Estate Review, 5(1), 91-105. Leung, Charles Ka Yui, Youngman Chun Fai Leong, and Siu Kei Wong (2006), “Housing Price Dispersion: An Empirical Investigation,” Journal of Real Estate Finance and Economics, 32(3), 357-385; MacMinn, R. D. (1980), “Search and Market Equilibrium,” Journal of Political Economy, 88, 308-15. Mathewson, G. Frank (1983), "Information, Search, and Price Variability of Individual Life Insurance Contracts," Journal of Industrial Economics, 32(2), 131- 148. McCall, John J. (1970), “Economics of Information and Job Search,” Quarterly Journal of Economics, 84(1), 113-126. Merton, R. C. (1973), “Theory of Rational Option Pricing,” Bell Journal of Economics and Management Science, 4, 141-183. Nelson, Phillip (1970), “Information and Consumer Behavior,” Journal of Political Economy, 78(2), 311-329. Olsen, Edgar O. (1972), “An Econometric analysis of Rent Control”, The Journal of Political Economy, 80(6), 1081-1100. Perloff, Jeffrey M. and Steven C. Salop (1985), “Equilibrium with Product Differentiation,” Review of Economic Studies, 52(1), 107-120. Postel-Vinay, Fabien and Jean-Marc Robin (2002), “Equilibrium Wage Dispersion with Worker and Employer Heterogeneity,” Econometrica, 70(6), 2295-2350. Pratt, John W., David Wise, and Richard Zeckhauser (1979), “Price Differences in Almost Competitive Markets,” Quarterly Journal of Economics, 93(2), 189-211. Quigley, J. M. (1987), “Interest Rate Variations, Mortgage Prepayments and Household Mobility,” Review of Economics and Statistics, 69, 636-643. Quigley, J., and R. Van Order (1995), “Efficiency in the Mortgage Market: The Borrower’s Perspective,” Journal of the American Real Estate and Urban Economics Association, 18(3), 237-252. Quigley, John M. (1990), “Does Rent Control Cause Homeless? Taking the Claim Seriously,” Journal of Policy Analysis and Management, 9, 89-93. 121 Quigley, John M. (2002), “Economic Analysis of Mobile Home Rent Control: The Example of San Rafael, California,” unpublished working paper. Radford, R. S. (2004), “Why Rent Control Is Still a Regulatory Taking?” unpublished working paper. Redfearn, Christian L. (2005), “Measure Globally, Aggregate Locally: The Interplay of Housing Market Dynamics and Price Index Construction”, unpublished working paper. Reinganum, Jennifer F. (1979), “A Simple Model of Equilibrium Price Dispersion,” Journal of Political Economy, 87(4), 851-858. Reitman, David (1991), “Endogenous Quality Differentiation in Congested Markets,” Journal of Industrial Economics, 39(6), 621-647. Rosenthal, S.S, J.V. Duca, Stuart Gabriel (1991), “Credit Rationing and the Demand for Owner-occuped Housing”, Journal of Urban Economics, 30, 48-63. Rubinfeld, Daniel L. (1992), “Regulatory Takings: The Case of Mobile Home Rent Control,” Chicago-Kent Review, 67(3), 923-929. Salop, Steven and Joseph Stigliz (1977), “Bargains and Ripoffs: A Model of Monopolistically Competitive Price Dispersion,” Review of Economic Studies, 44(3), 493-510. Sass, Tim R. (1988), “A Note on Optimal Price Cutting Behavior under Demand Uncertainty,” Review of Economics and Statistics, 70(2), 336-339. Schlesinger, H. and J. Schulenburg (1991), “Search Costs, Switching and Product Heterogeneity in an Insurance Market,” Journal of Risk and Insurance, 58(1), 109- 119. Schwartz, Eduardo S. and Walter N. Torous (1989), “Prepayment and the Valuation of Mortgage-Backed Securities,” Journal of Finance, 44(2), 375-392. Seog, S. Hun (2002), “Equilibrium Price Dispersion in the Insurance Market,” Journal of Risk and Insurance, 69(4), 517-536. Shilony, Yuval (1977), “Mixed Pricing in Oligopoly,” Journal of Economic Theory, 14(2), 373-388. Sirmans, C.F., Geoffrey K. Turnbull, and Jonathan Dombrow (1995), “Quick House Sales: Seller Mistake or Luck?” Journal of Housing Economics, 4(3), 230-243. Sorensen, Alan T. (2000), “Equilibrium Price Dispersion in Retail Markets for Prescription Drugs,” Journal of Political Economy, 108(4), 833-850. Stigler, George J. (1961), “The Economics of Information,” Journal of Political Economy, 69(3), 213-225. Stull, William J. (1978), “The Landlord’s Dilemma: Asking Rent Strategies in a Heterogeneous Housing Market,” Journal of Urban Economics, 5(1), 101-115. 122 Taylor, Curits R. (1999), “Time-on-the-Market as a Sign of Quality,” Review of Economic Studies, 66(3), 555-578. Tucker, W. (1989), “America’s Homeless: Victims of Rent Control,” Heritage Foundation Backgrounder, 685. US Department of Housing and Urban Development (HUD) (1991), Report to Congress on Rent Control. Office of Policy Development and Research. Varian, H. (1980), "A Model of Sales," American Economic Review, 70(4), 651-659. Wilde, Louis L. and Alan Schwartz (1979), “Equilibrium Comparison Shopping,” Review of Economic Studies, 46(3), 543-554. Yavas, Abdullah and Shiawee Yang (1995), “The Strategic Role of Listing Price in Marketing Real Estate: Theory and Evidence,” Real Estate Economics, 23(3), 347- 368. Zuehlke, Thomas W. (1987), “Duration Dependence in the Housing Market,” Review of Economics and Statistics, 69 (4), 701-709. 123 Appendix A: Hedonic regression results (1) For small complexes (Number of units <= 30) Month intercept Access Size Age of building (months) N R 2 Feb-94 7.85 -0.02 0.02 0.00 35 0.71 (32.62) (-3.09) (7.48) (-1.95) Mar-94 7.76 -0.03 0.02 0.00 71 0.76 (42.99) (-4.80) (13.44) (-2.81) Apr-94 8.19 -0.03 0.02 0.00 50 0.74 (33.73) (-3.26) (10.64) (-3.64) May-94 8.00 -0.03 0.02 0.00 62 0.78 (41.06) (-4.69) (12.97) (-4.11) Jun-94 7.71 -0.02 0.02 0.00 59 0.83 (43.93) (-3.53) (15.12) (-3.33) Jul-94 7.83 -0.02 0.02 0.00 66 0.79 (45.11) (-4.64) (15.03) (-2.30) Aug-94 7.36 -0.01 0.02 0.00 50 0.78 (31.03) (-1.50) (11.87) (-1.67) Sep-94 7.66 -0.02 0.02 0.00 82 0.79 (57.24) (-4.30) (17.24) (-2.43) Oct-94 7.61 -0.02 0.02 0.00 71 0.77 (42.71) (-3.34) (14.56) (-2.43) Nov-94 7.94 -0.03 0.02 0.00 75 0.78 (39.04) (-4.74) (13.37) (-3.19) Dec-94 7.80 -0.02 0.02 0.00 82 0.71 (46.32) (-4.33) (13.66) (-2.84) Jan-95 7.21 -0.01 0.02 0.00 54 0.76 (28.68) (-0.89) (12.60) (-1.64) Feb-95 7.84 -0.02 0.02 0.00 60 0.83 (39.23) (-3.81) (13.55) (-4.68) Mar-95 7.66 -0.02 0.02 0.00 68 0.78 (44.93) (-4.03) (14.98) (-2.74) Apr-95 7.40 -0.02 0.03 0.00 49 0.78 (37.15) (-2.77) (11.29) (-3.77) May-95 7.38 -0.02 0.02 0.00 58 0.84 (32.10) (-2.53) (15.00) (-2.79) Jun-95 7.16 -0.01 0.02 0.00 79 0.77 (43.08) (-2.75) (14.45) (-1.66) Jul-95 7.40 -0.02 0.02 0.00 61 0.79 (34.79) (-3.00) (13.30) (-1.06) Aug-95 7.93 -0.03 0.02 0.00 52 0.76 (29.01) (-3.50) (10.06) (-3.05) Sep-95 7.94 -0.02 0.02 0.00 65 0.79 (34.91) (-3.87) (11.30) (-3.57) Oct-95 7.88 -0.03 0.02 0.00 73 0.82 (47.03) (-6.69) (15.02) (-3.94) Nov-95 7.98 -0.03 0.02 0.00 78 0.84 (46.08) (-6.06) (16.00) (-3.70) Dec-95 7.52 -0.02 0.02 0.00 65 0.72 (31.21) (-3.06) (11.33) (-2.30) 124 For small complexes (Number of units <= 30) cont. Month intercept Access Size Age of building (months) N R 2 Jan-96 7.33 -0.01 0.02 0.00 56 0.82 (29.65) (-1.55) (12.69) (-2.83) Feb-96 7.47 -0.02 0.02 0.00 112 0.84 (63.73) (-4.78) (22.45) (-6.76) Mar-96 7.52 -0.02 0.02 0.00 89 0.79 (46.28) (-4.63) (17.24) (-4.74) Apr-96 7.60 -0.02 0.02 0.00 101 0.77 (46.47) (-5.08) (17.05) (-2.41) May-96 7.70 -0.02 0.02 0.00 80 0.80 (35.78) (-3.82) (13.60) (-3.29) Jun-96 7.59 -0.02 0.02 0.00 89 0.83 (45.41) (-4.34) (19.37) (-4.44) Jul-96 7.50 -0.02 0.02 0.00 75 0.78 (34.93) (-2.51) (14.43) (-3.50) Aug-96 7.73 -0.03 0.02 0.00 68 0.72 (30.19) (-3.27) (11.25) (-2.59) Sep-96 7.59 -0.02 0.02 0.00 77 0.79 (49.54) (-3.64) (15.48) (-3.94) Oct-96 7.75 -0.02 0.02 0.00 108 0.79 (48.18) (-4.71) (18.56) (-4.26) Nov-96 7.86 -0.03 0.02 0.00 86 0.86 (46.74) (-5.49) (20.70) (-4.32) Dec-96 7.57 -0.03 0.02 0.00 67 0.82 (35.78) (-4.79) (16.26) (-1.46) Jan-97 7.64 -0.01 0.02 0.00 90 0.80 (40.56) (-2.73) (16.49) (-4.76) Feb-97 7.70 -0.02 0.02 0.00 95 0.80 (55.08) (-5.26) (17.13) (-6.12) Mar-97 7.85 -0.02 0.02 0.00 80 0.80 (41.31) (-4.54) (15.94) (-5.61) Apr-97 7.68 -0.02 0.02 0.00 75 0.80 (44.39) (-3.40) (15.93) (-6.89) May-97 7.47 -0.02 0.02 0.00 73 0.79 (31.09) (-2.20) (15.60) (-2.71) Jun-97 7.55 -0.02 0.02 0.00 69 0.88 (40.13) (-2.59) (19.62) (-4.50) Jul-97 7.47 -0.02 0.02 0.00 83 0.83 (44.44) (-4.54) (17.92) (-3.67) Aug-97 7.44 -0.01 0.02 0.00 42 0.74 (24.89) (-1.94) (8.95) (-3.18) Sep-97 7.63 -0.03 0.02 0.00 66 0.81 (42.80) (-5.20) (14.87) (-4.51) Oct-97 7.60 -0.02 0.02 0.00 90 0.83 (53.24) (-5.37) (18.56) (-3.77) Nov-97 7.58 -0.01 0.02 0.00 63 0.76 (34.24) (-2.19) (12.86) (-2.72) Dec-97 7.57 -0.02 0.02 0.00 53 0.82 (38.97) (-3.26) (14.46) (-4.77) 125 For small complexes (Number of units <= 30) cont. Month intercept Access Size Age of building (months) N R 2 Jan-98 7.77 -0.02 0.02 0.00 49 0.81 (36.51) (-3.92) (10.88) (-4.22) Feb-98 7.88 -0.03 0.02 0.00 72 0.75 (34.86) (-3.93) (10.73) (-3.72) Mar-98 7.88 -0.03 0.02 0.00 90 0.76 (43.91) (-4.70) (14.48) (-5.36) Apr-98 7.47 -0.01 0.02 0.00 81 0.79 (54.88) (-3.20) (15.61) (-4.78) May-98 7.51 -0.01 0.02 0.00 75 0.79 (40.99) (-1.11) (14.88) (-5.33) Jun-98 7.74 -0.02 0.02 0.00 71 0.75 (45.55) (-3.97) (13.02) (-3.53) Jul-98 7.45 -0.01 0.02 0.00 79 0.80 (38.17) (-1.95) (16.42) (-5.46) Aug-98 7.95 -0.02 0.02 0.00 48 0.70 (32.23) (-3.38) (8.34) (-4.65) Sep-98 7.58 -0.02 0.02 0.00 75 0.82 (44.12) (-4.27) (14.26) (-5.15) Oct-98 7.68 -0.03 0.02 0.00 91 0.82 (50.60) (-5.12) (18.45) (-5.08) Nov-98 7.66 -0.02 0.02 0.00 91 0.73 (42.46) (-3.52) (12.68) (-4.78) Dec-98 7.92 -0.03 0.02 0.00 81 0.80 (45.50) (-5.41) (13.88) (-5.57) Jan-99 7.23 -0.02 0.02 0.00 69 0.89 (51.51) (-4.01) (22.11) (-5.67) Feb-99 7.53 -0.02 0.02 0.00 98 0.73 (46.62) (-3.64) (15.62) (-3.21) Mar-99 8.07 -0.04 0.02 0.00 81 0.77 (50.54) (-6.91) (14.96) (-6.55) Apr-99 7.61 -0.02 0.02 0.00 104 0.81 (46.20) (-3.51) (18.17) (-5.86) May-99 7.73 -0.01 0.01 0.00 78 0.83 (43.31) (-2.09) (14.84) (-7.63) Jun-99 7.70 -0.02 0.02 0.00 90 0.79 (48.75) (-4.37) (13.53) (-6.83) Jul-99 7.85 -0.03 0.02 0.00 90 0.77 (46.54) (-5.89) (15.41) (-5.57) Aug-99 7.66 -0.02 0.02 0.00 65 0.84 (41.74) (-3.23) (15.55) (-4.96) Sep-99 7.57 -0.01 0.02 0.00 92 0.84 (50.76) (-2.46) (17.60) (-9.15) Oct-99 7.76 -0.03 0.02 0.00 76 0.78 (40.69) (-4.62) (14.50) (-5.24) Nov-99 8.08 -0.04 0.02 0.00 70 0.77 (38.10) (-5.13) (12.05) (-5.43) Dec-99 7.59 -0.03 0.02 0.00 71 0.78 (43.27) (-4.58) (14.31) (-4.76) 126 For small complexes (Number of units <= 30) cont. Month intercept Access Size Age of building (months) N R 2 Jan-00 7.77 -0.02 0.02 0.00 84 0.77 (52.89) (-4.39) (14.48) (-6.01) Feb-00 7.84 -0.02 0.02 0.00 105 0.76 (58.74) (-5.07) (16.09) (-7.22) Mar-00 7.95 -0.03 0.01 0.00 129 0.72 (51.45) (-4.96) (16.29) (-6.68) Apr-00 7.55 -0.02 0.02 0.00 101 0.81 (49.51) (-4.59) (16.61) (-5.52) May-00 7.98 -0.03 0.02 0.00 108 0.81 (44.57) (-6.01) (18.30) (-7.08) Jun-00 7.48 -0.01 0.02 0.00 112 0.77 (53.64) (-4.18) (15.96) (-6.15) Jul-00 7.90 -0.04 0.02 0.00 69 0.72 (35.55) (-4.87) (10.84) (-5.70) Aug-00 8.09 -0.04 0.02 0.00 95 0.77 (51.63) (-7.31) (15.38) (-5.58) Sep-00 7.74 -0.02 0.02 0.00 73 0.68 (35.92) (-3.43) (8.77) (-5.75) Oct-00 7.65 -0.03 0.02 0.00 99 0.80 (43.02) (-5.11) (16.34) (-5.09) Nov-00 7.76 -0.03 0.02 0.00 119 0.81 (57.90) (-6.84) (20.39) (-6.46) Dec-00 7.68 -0.02 0.02 0.00 86 0.76 (38.74) (-3.25) (13.28) (-5.89) Jan-01 7.80 -0.03 0.02 0.00 148 0.74 (60.13) (-7.39) (18.34) (-6.64) Feb-01 7.73 -0.03 0.02 0.00 118 0.73 (48.71) (-5.36) (15.27) (-5.96) Mar-01 7.94 -0.03 0.02 0.00 126 0.81 (59.16) (-7.27) (19.80) (-8.51) Apr-01 7.61 -0.03 0.02 0.00 114 0.79 (58.38) (-7.16) (18.60) (-5.42) May-01 7.52 -0.02 0.02 0.00 129 0.76 (61.16) (-6.36) (16.13) (-5.62) Jun-01 7.75 -0.03 0.02 0.00 120 0.80 (55.86) (-7.42) (17.71) (-6.76) Jul-01 7.99 -0.03 0.02 0.00 86 0.77 (44.19) (-5.32) (12.36) (-7.48) Aug-01 7.76 -0.03 0.02 0.00 116 0.80 (59.37) (-6.55) (18.55) (-6.96) Sep-01 8.01 -0.04 0.02 0.00 108 0.79 (55.82) (-7.43) (16.58) (-6.33) Oct-01 7.67 -0.03 0.02 0.00 138 0.84 (62.69) (-7.09) (22.50) (-6.22) Nov-01 7.71 -0.03 0.02 0.00 113 0.81 (51.38) (-5.42) (17.11) (-6.30) Dec-01 7.62 -0.02 0.02 0.00 111 0.80 (48.34) (-4.86) (18.08) (-5.77) 127 For small complexes (Number of units <= 30) cont. Month intercept Access Size Age of building (months) N R 2 Jan-02 8.15 -0.03 0.01 0.00 137 0.82 (69.09) (-7.70) (17.95) (-10.91) Feb-02 7.74 -0.02 0.02 0.00 151 0.79 (56.24) (-5.55) (19.94) (-6.68) Mar-02 7.50 -0.02 0.02 0.00 135 0.83 (56.67) (-5.58) (20.08) (-6.30) Apr-02 7.62 -0.02 0.02 0.00 129 0.81 (53.47) (-4.84) (20.54) (-6.87) May-02 7.74 -0.02 0.02 0.00 138 0.85 (60.24) (-6.55) (17.44) (-7.15) Jun-02 7.68 -0.03 0.02 0.00 118 0.79 (51.53) (-5.66) (16.11) (-5.09) 128 Hedonic regression results (2) For bigger complexes (Number of units > 30) Month intercept Access CBD Size Age of building (months) N R 2 Jan-94 7.78 -0.03 -0.02 0.02 0.00 19 0.87 (27.92) (-2.43) (-0.08) (9.04) (-0.64) Feb-94 7.35 -0.01 0.23 0.02 0.00 91 0.81 (55.31) (-2.97) (3.80) (18.05) (-2.03) Mar-94 7.61 -0.02 0.19 0.02 0.00 235 0.84 (100.70) (-7.78) (5.03) (31.80) (-6.19) Apr-94 7.65 -0.02 0.18 0.02 0.00 167 0.78 (66.85) (-4.52) (3.56) (22.28) (-4.93) May-94 7.70 -0.02 0.13 0.02 0.00 169 0.81 (71.37) (-5.66) (2.71) (25.43) (-3.42) Jun-94 7.94 -0.02 0.16 0.02 0.00 247 0.80 (92.68) (-8.28) (3.92) (27.39) (-7.22) Jul-94 7.74 -0.02 0.16 0.02 0.00 214 0.80 (82.65) (-7.52) (3.76) (27.45) (-5.79) Aug-94 7.84 -0.03 0.03 0.02 0.00 149 0.80 (62.93) (-6.44) (0.69) (22.99) (-4.00) Sep-94 7.87 -0.02 0.11 0.02 0.00 264 0.78 (91.05) (-9.38) (2.81) (27.17) (-6.36) Oct-94 7.75 -0.02 0.14 0.02 0.00 242 0.81 (91.16) (-8.04) (3.47) (29.74) (-6.24) Nov-94 7.70 -0.02 0.17 0.02 0.00 229 0.79 (72.21) (-7.47) (3.49) (25.83) (-4.50) Dec-94 7.89 -0.03 0.10 0.02 0.00 248 0.78 (88.46) (-9.54) (1.99) (27.31) (-5.81) Jan-95 7.62 -0.02 0.06 0.02 0.00 236 0.73 (70.01) (-5.59) (1.28) (22.15) (-4.10) Feb-95 7.72 -0.03 0.10 0.02 0.00 230 0.77 (85.90) (-8.78) (2.58) (25.15) (-5.42) Mar-95 7.58 -0.02 0.12 0.02 0.00 331 0.79 (89.89) (-8.22) (3.23) (32.07) (-6.65) Apr-95 7.55 -0.02 0.19 0.02 0.00 215 0.79 (68.36) (-4.79) (4.10) (25.71) (-6.42) May-95 7.78 -0.03 0.04 0.02 0.00 178 0.73 (56.54) (-6.47) (0.69) (18.97) (-4.89) Jun-95 7.60 -0.02 0.08 0.02 0.00 283 0.76 (85.00) (-8.04) (2.03) (28.19) (-5.03) Jul-95 7.83 -0.03 0.08 0.02 0.00 203 0.75 (65.27) (-7.38) (1.52) (20.55) (-6.25) Aug-95 7.65 -0.02 0.08 0.02 0.00 225 0.74 (73.58) (-6.93) (1.68) (21.79) (-5.89) Sep-95 7.52 -0.02 0.14 0.02 0.00 270 0.77 (79.49) (-7.47) (3.29) (27.31) (-6.34) Oct-95 7.68 -0.02 0.11 0.02 0.00 270 0.76 (77.81) (-7.29) (2.50) (26.20) (-7.60) Nov-95 7.57 -0.02 0.12 0.02 0.00 324 0.79 (95.45) (-8.85) (3.40) (32.21) (-8.28) Dec-95 7.48 -0.02 0.18 0.02 0.00 236 0.76 (68.95) (-5.29) (3.94) (24.70) (-6.41) 129 For bigger complexes (Number of units > 30) cont. Month intercept Access CBD Size Age of building (months) N R 2 Jan-96 7.49 -0.02 0.10 0.02 0.00 159 0.76 (58.05) (-5.08) (1.71) (21.10) (-3.14) Feb-96 7.49 -0.02 0.14 0.02 0.00 377 0.77 (88.95) (-7.54) (3.98) (31.57) (-9.10) Mar-96 7.34 -0.02 0.15 0.02 0.00 276 0.80 (85.45) (-6.22) (3.84) (29.61) (-8.15) Apr-96 7.59 -0.02 0.15 0.02 0.00 273 0.81 (90.81) (-8.39) (3.87) (29.57) (-10.65) May-96 7.61 -0.02 0.13 0.02 0.00 253 0.78 (81.89) (-8.38) (3.13) (27.18) (-8.73) Jun-96 7.45 -0.02 0.11 0.02 0.00 243 0.76 (74.00) (-6.61) (2.69) (25.69) (-6.02) Jul-96 7.55 -0.02 0.14 0.02 0.00 255 0.78 (78.39) (-7.56) (3.51) (26.35) (-6.92) Aug-96 7.38 -0.02 0.06 0.02 0.00 272 0.83 (83.74) (-7.71) (1.66) (33.10) (-4.79) Sep-96 7.51 -0.02 0.12 0.02 0.00 241 0.78 (71.32) (-6.53) (2.80) (26.38) (-6.16) Oct-96 7.64 -0.02 0.14 0.02 0.00 371 0.75 (97.51) (-8.33) (3.97) (29.24) (-12.37) Nov-96 7.54 -0.02 0.14 0.02 0.00 275 0.76 (78.63) (-6.94) (3.32) (26.59) (-7.72) Dec-96 7.55 -0.02 0.01 0.02 0.00 228 0.79 (77.36) (-6.64) (0.20) (26.55) (-8.61) Jan-97 7.61 -0.02 0.07 0.02 0.00 255 0.69 (71.92) (-6.75) (1.70) (21.44) (-5.34) Feb-97 7.69 -0.03 0.12 0.02 0.00 302 0.77 (83.98) (-9.18) (3.09) (27.98) (-7.87) Mar-97 7.37 -0.02 0.10 0.02 0.00 319 0.78 (80.12) (-6.56) (2.78) (31.82) (-5.62) Apr-97 7.67 -0.02 0.06 0.02 0.00 288 0.77 (80.60) (-7.53) (1.39) (27.12) (-10.18) May-97 7.43 -0.02 0.12 0.02 0.00 214 0.81 (70.85) (-7.18) (3.05) (27.14) (-6.58) Jun-97 7.42 -0.02 0.17 0.02 0.00 243 0.77 (66.40) (-5.99) (4.17) (26.64) (-5.12) Jul-97 7.56 -0.02 0.11 0.02 0.00 300 0.76 (80.80) (-6.38) (2.82) (27.42) (-8.90) Aug-97 7.73 -0.03 0.07 0.02 0.00 163 0.80 (62.99) (-6.93) (1.52) (22.82) (-7.50) Sep-97 7.63 -0.02 0.09 0.02 0.00 190 0.76 (63.75) (-6.84) (1.99) (20.25) (-5.24) Oct-97 7.47 -0.02 0.15 0.02 0.00 341 0.77 (74.65) (-5.49) (4.16) (30.02) (-8.36) Nov-97 7.48 -0.02 0.09 0.02 0.00 224 0.79 (73.44) (-7.99) (2.07) (26.81) (-7.06) Dec-97 7.32 -0.02 0.09 0.02 0.00 202 0.74 (61.84) (-4.99) (1.82) (22.53) (-5.98) 130 For bigger complexes (Number of units > 30) cont. Month intercept Access CBD Size Age of building (months) N R 2 Jan-98 7.47 -0.02 0.10 0.02 0.00 205 0.77 (61.59) (-6.03) (2.10) (23.12) (-5.61) Feb-98 7.64 -0.03 0.11 0.02 0.00 253 0.78 (73.36) (-8.20) (2.89) (26.28) (-6.88) Mar-98 7.41 -0.02 0.09 0.02 0.00 231 0.80 (68.96) (-5.27) (2.04) (29.13) (-7.51) Apr-98 7.51 -0.03 0.06 0.02 0.00 277 0.77 (68.05) (-7.11) (1.55) (28.73) (-7.09) May-98 7.58 -0.03 0.03 0.02 0.00 247 0.78 (75.39) (-7.67) (0.61) (25.73) (-7.97) Jun-98 7.40 -0.02 0.10 0.02 0.00 210 0.79 (67.75) (-5.81) (2.26) (25.82) (-5.90) Jul-98 7.45 -0.02 0.08 0.02 0.00 257 0.81 (70.65) (-6.71) (2.02) (30.31) (-7.86) Aug-98 7.40 -0.02 0.05 0.02 0.00 167 0.75 (61.57) (-4.11) (1.05) (18.95) (-5.94) Sep-98 7.63 -0.03 0.05 0.02 0.00 204 0.77 (67.10) (-7.08) (1.10) (22.76) (-8.72) Oct-98 7.47 -0.02 0.13 0.02 0.00 272 0.77 (83.11) (-7.24) (3.57) (26.06) (-8.11) Nov-98 7.31 -0.02 0.06 0.02 0.00 214 0.82 (68.30) (-5.18) (1.53) (28.62) (-6.67) Dec-98 7.54 -0.03 0.15 0.02 0.00 236 0.80 (71.11) (-7.73) (3.85) (26.84) (-9.28) Jan-99 7.36 -0.02 0.08 0.02 0.00 210 0.81 (70.34) (-4.84) (1.95) (27.66) (-7.96) Feb-99 7.40 -0.02 0.19 0.02 0.00 328 0.74 (74.77) (-5.03) (5.02) (25.81) (-10.16) Mar-99 7.51 -0.02 0.15 0.02 0.00 263 0.79 (76.81) (-6.83) (3.89) (25.65) (-10.23) Apr-99 7.45 -0.01 0.14 0.02 0.00 281 0.75 (72.65) (-4.54) (3.28) (26.27) (-10.32) May-99 7.45 -0.02 0.10 0.02 0.00 251 0.82 (77.78) (-6.94) (2.42) (29.11) (-9.68) Jun-99 7.51 -0.02 0.15 0.02 0.00 245 0.75 (71.47) (-5.43) (3.48) (23.81) (-9.10) Jul-99 7.60 -0.03 0.10 0.02 0.00 283 0.79 (71.74) (-7.66) (2.57) (28.06) (-11.84) Aug-99 7.47 -0.02 0.14 0.02 0.00 149 0.74 (50.95) (-3.97) (2.18) (16.94) (-6.54) Sep-99 7.37 -0.01 0.16 0.02 0.00 302 0.76 (76.51) (-4.52) (4.34) (25.66) (-11.15) Oct-99 7.47 -0.02 0.13 0.02 0.00 202 0.77 (61.95) (-5.43) (2.93) (21.24) (-8.34) Nov-99 7.50 -0.02 0.09 0.02 0.00 268 0.80 (73.68) (-7.15) (2.45) (27.58) (-8.55) Dec-99 7.55 -0.02 0.17 0.02 0.00 246 0.71 (69.49) (-4.77) (3.60) (21.83) (-8.43) 131 For bigger complexes (Number of units > 30) cont. Month intercept Access CBD Size Age of building (months) N R 2 Jan-00 7.52 -0.02 0.06 0.02 0.00 211 0.71 (61.29) (-4.88) (1.33) (18.51) (-9.19) Feb-00 7.46 -0.03 0.12 0.02 0.00 266 0.76 (70.50) (-7.47) (2.98) (24.70) (-9.63) Mar-00 7.53 -0.02 0.19 0.02 0.00 387 0.72 (79.50) (-6.67) (5.09) (25.71) (-13.37) Apr-00 7.52 -0.02 0.12 0.02 0.00 241 0.78 (69.53) (-6.35) (2.86) (23.55) (-10.47) May-00 7.55 -0.02 0.06 0.02 0.00 269 0.72 (67.35) (-6.68) (1.21) (22.12) (-8.25) Jun-00 7.44 -0.02 0.13 0.02 0.00 297 0.74 (72.82) (-5.72) (2.90) (24.95) (-10.43) Jul-00 7.61 -0.02 0.09 0.02 0.00 178 0.84 (66.15) (-6.41) (2.08) (22.84) (-11.78) Aug-00 7.74 -0.03 0.09 0.02 0.00 275 0.77 (74.34) (-9.13) (2.23) (25.05) (-11.13) Sep-00 7.67 -0.03 0.08 0.02 0.00 211 0.76 (60.89) (-6.60) (1.78) (19.79) (-10.24) Oct-00 7.58 -0.02 0.12 0.02 0.00 275 0.72 (68.16) (-6.29) (2.78) (24.22) (-7.95) Nov-00 7.61 -0.02 0.11 0.02 0.00 314 0.72 (69.43) (-6.29) (2.33) (23.48) (-10.53) Dec-00 7.67 -0.03 0.11 0.02 0.00 222 0.71 (63.92) (-7.38) (2.08) (20.08) (-8.73) Jan-01 7.37 -0.02 0.13 0.02 0.00 316 0.70 (69.98) (-5.87) (3.13) (23.84) (-8.81) Feb-01 7.49 -0.02 0.18 0.02 0.00 354 0.70 (77.11) (-6.77) (4.42) (23.22) (-10.67) Mar-01 7.61 -0.02 0.10 0.02 0.00 342 0.74 (79.57) (-7.19) (2.66) (25.15) (-12.32) Apr-01 7.53 -0.02 0.13 0.02 0.00 342 0.71 (77.65) (-6.83) (3.38) (24.94) (-9.64) May-01 7.66 -0.02 0.13 0.02 0.00 358 0.72 (85.24) (-8.86) (3.63) (26.04) (-10.63) Jun-01 7.47 -0.02 0.15 0.02 0.00 321 0.74 (81.31) (-6.12) (3.83) (25.52) (-10.93) Jul-01 7.48 -0.02 0.14 0.02 0.00 237 0.71 (61.31) (-4.90) (2.85) (20.28) (-9.30) Aug-01 7.57 -0.02 0.11 0.02 0.00 346 0.70 (77.56) (-6.43) (2.73) (21.95) (-10.96) Sep-01 7.62 -0.03 0.07 0.02 0.00 340 0.72 (79.36) (-9.01) (1.82) (25.35) (-9.27) Oct-01 7.58 -0.02 0.18 0.02 0.00 392 0.76 (83.01) (-7.61) (4.93) (29.88) (-12.92) Nov-01 7.65 -0.03 0.19 0.02 0.00 303 0.72 (73.49) (-8.14) (4.75) (22.41) (-10.61) Dec-01 7.81 -0.03 0.10 0.02 0.00 280 0.71 (68.65) (-7.86) (2.26) (19.59) (-12.51) 132 For bigger complexes (Number of units > 30) cont. Month intercept Access CBD Size Age of building (months) N R 2 Jan-02 7.42 -0.02 0.11 0.02 0.00 254 0.74 (65.43) (-6.79) (2.31) (22.96) (-9.62) Feb-02 7.54 -0.03 0.14 0.02 0.00 362 0.75 (73.15) (-7.81) (3.71) (26.63) (-10.35) Mar-02 7.69 -0.03 0.11 0.02 0.00 354 0.76 (79.72) (-9.29) (2.99) (26.94) (-12.45) Apr-02 7.79 -0.03 0.05 0.02 0.00 305 0.74 (77.72) (-9.29) (1.33) (24.01) (-12.13) May-02 7.78 -0.03 0.05 0.02 0.00 316 0.74 (72.89) (-9.76) (1.28) (24.41) (-11.28) Jun-02 7.75 -0.03 0.12 0.02 0.00 300 0.71 (74.32) (-7.69) (3.01) (23.94) (-11.78) 133 Appendix B. Appendix B.1: Repeat Sale Price Index (1983=100) Quarter Index Quarter Index 1983-1 100.00 1993-2 58.48 1983-2 107.14 1993-3 56.30 1983-3 112.53 1993-4 49.77 1983-4 113.92 1994-1 52.57 1984-1 105.22 1994-2 47.16 1984-2 115.09 1994-3 45.90 1984-3 116.42 1994-4 44.73 1984-4 118.23 1995-1 48.10 1985-1 102.09 1995-2 42.55 1985-2 116.29 1995-3 42.11 1985-3 119.58 1995-4 42.28 1985-4 117.48 1996-1 46.38 1986-1 100.33 1996-2 40.11 1986-2 118.04 1996-3 41.99 1986-3 120.02 1996-4 40.23 1986-4 119.34 1997-1 49.83 1987-1 103.74 1997-2 39.89 1987-2 117.26 1997-3 42.01 1987-3 119.52 1997-4 45.05 1987-4 116.73 1998-1 54.24 1988-1 105.98 1998-2 46.12 1988-2 117.34 1998-3 47.43 1988-3 120.51 1998-4 47.90 1988-4 119.52 1999-1 55.66 1989-1 101.32 1999-2 50.52 1989-2 117.61 1999-3 49.73 1989-3 121.61 1999-4 51.00 1989-4 118.95 2000-1 57.39 1990-1 98.91 2000-2 53.56 1990-2 117.02 2000-3 56.88 1990-3 115.69 2000-4 59.59 1990-4 107.37 2001-1 60.17 1991-1 86.67 2001-2 59.13 1991-2 97.43 2001-3 61.05 1991-3 96.54 2001-4 60.41 1991-4 87.31 2002-1 60.46 1992-1 77.71 2002-2 61.16 1992-2 79.22 2002-3 64.55 1992-3 74.68 2002-4 65.23 1992-4 67.51 2003-1 61.23 1993-1 59.60 2003-2 66.74 134 Appendix B.2 Resale Price Comparison for an Average Structure in Los Angeles The following Tables present the estimated resale prices for the average mobile home units (in each subgroup) under different rent control regimes in Los Angeles County. The average unit is defined as the one with average structure (double, triple), average age, in census tract with average proportion of households with public assistance income, etc. We assume that rent control ordinance effective date is January 1, 1993 for the units with adoption of rent control between the first and the second transactions. By using the average units' characteristics and the parameters estimated in Table 3.5, we can simulate the resale prices as presented here. (a) Resale Price Comparison for an Average Structure in Los Angeles, Different Income Communities High Median Household Income Low Median Household Income Resale price comparison for an average 19 years old structure No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol Resale price $39,194 $41,707 $39,194 $46,660 $35,037 $34,731 $34,731 $35,594 $40,977 $33,327 Price increase -$3,940 -$1,427 -$3,940 $3,526 -$8,097 -$761 -$761 $102 $5,485 -$2,165 Growth rate (percentage) -9.13 -3.31 -9.13 8.17 -18.77 -2.14 -2.14 0.29 15.45 -6.10 Growth out from rent control (percentage) - 5.83 0 17.31 -9.64 - 0 2.43 17.60 -3.96 (b) Resale Price Comparison for an Average Structure in Los Angeles, Different Age Communities High Proportion of Elderly Population Low Proportion of Elderly Population Resale price comparison for an average 19 years old structure No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol Resale price $43,557 $47,135 $45,704 $49,394 $43,557 $36,305 $34,727 $35,104 $44,327 $31,495 Price increase $1,078 $4,655 $3,225 $6,914 $1,078 -$2,650 -$4,228 -$3,851 $5,372 -$7,460 Growth rate (percentage) 2.54 10.96 7.59 16.28 2.54 -6.80 -10.85 -9.89 13.79 -19.15 Growth out from rent control (percentage) - 8.42 5.05 13.74 0 - -4.05 -3.08 20.59 -12.35 135 (c) Resale Price Comparison for an Average Structure in Los Angeles, Different Income and Age Communities High Income & High Proportion of Elder High Income & Low Proportion of Elder Resale price comparison for an average 19 years old structure No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol Resale price $38,284 $44,607 $38,284 $42,614 $39,594 $39,094 $39,094 $39,094 $53,192 $33,606 Price increase -$5,778 $544 -$5,778 -$1,449 -$4,469 -$2,933 -$2,933 -$2,933 $11,164 -$8,422 Growth rate (percentage) -13.11 1.23 -13.11 -3.29 -10.14 -6.98 -6.98 -6.98 26.56 -20.04 Growth out from rent control (percentage) - 14.35 0 9.83 2.97 - 0 0 33.54 -13.06 Low Income & High Proportion of Elder Low Income & Low Proportion of Elder Resale price comparison for an average 19 years old structure No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol No rent control Under rent control without vacancy decontrol Under rent control with vacancy decontrol Adoption of rent control without vacancy decontrol Adoption of rent control with vacancy decontrol Resale price $35,842 $35,842 $38,130 $41,425 $35,842 $33,113 $33,113 $33,113 $40,330 $30,596 Price increase -$2,221 -$2,221 $66 $3,362 -$2,221 -$454 -$454 -$454 $6,763 -$2,971 Growth rate (percentage) -5.84 -5.84 0.17 8.83 -5.84 -1.35 -1.35 -1.35 20.15 -8.85 Growth out from rent control (percentage) - 0 6.01 14.67 0 - 0 0 21.50 -7.50

Abstract (if available)

Abstract This dissertation consists of three essays on urban economics and housing market, emphasizing on market behavior within an international context. Chapter One comprises an analysis of the optimal selling strategy in a market with price dispersion using the central Tokyo condominium resale market list data from 1994 to 2002. The optimal pricing strategy is chosen to maximize the return from search. Higher price dispersion leads to higher reservation and optimal asking prices, which in turn results in higher expected sales prices. Under the assumption that the offering prices follow a normal distribution, market price dispersion can increase the probability of a successful transaction and/or speed up the sale process for the overpriced properties.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on price determinants in the Los Angeles housing market

PDF

Three essays on aging, wealth, and housing tenure transitions

PDF

Three essays on agent’s strategic behavior on online trading market

PDF

Risks, returns, and regulations in real estate markets

PDF

Behavioral approaches to industrial organization

PDF

Three essays on the microeconometric analysis of the labor market

PDF

Essays on understanding consumer contribution behaviors in the context of crowdfunding

PDF

Three essays on industrial organization

PDF

Theoretical and data-driven analysis of two-sided markets and strategic behavior in modern economic systems

Asset Metadata

Creator Zheng, Diehang Della (author)

Core Title Essays on housing market behavior analysis within the international context

School School of Policy, Planning, and Development

Degree Doctor of Philosophy

Degree Program Planning

Publication Date 06/27/2007

Defense Date 04/26/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag house price,mobile home park,mortgage,OAI-PMH Harvest,prepayment and default,rent control,search

Place Name California (states), China (countries), Japan (countries), Tokyo (city or populated place), USA (countries)

Language English

Advisor Deng, Yongheng (committee chair), Conway, Delores (committee member), Gabriel, Stuart A. (committee member), Gordon, Peter (committee member)

Creator Email diehangz@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m556

Unique identifier UC1192509

Identifier etd-Zheng-20070627 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-507178 (legacy record id),usctheses-m556 (legacy record id)

Legacy Identifier etd-Zheng-20070627.pdf

Dmrecord 507178

Document Type Dissertation

Rights Zheng, Diehang Della

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu