See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/33754006

Liquidity and Pricing Biases in the Real Estate Market

Article · January 2004

Source: OAI

CITATIONS READS
9 665

1 author:

Zhenguo Lin
Florida International University
53 PUBLICATIONS 720 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Liquidity Risk and Investment View project

Liquidity Risk and Investment View project

All content following this page was uploaded by Zhenguo Lin on 10 August 2014.

The user has requested enhancement of the downloaded file.

 Liquidity and Pricing Biases in the Real Estate Market

By

Zhenguo Lin

A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT

OF THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

(BUSINESS)

at the

UNIVERSITY OF WISCONSIN-MADISON

2004
 i

Table of Contents

Acknowledgements iii

Abstract iv

Chapter 1. Introduction: The Pricing Biases in the Real Estate Market 1

1.1. Appraisal Bias 1
1.2. Sample Selection Bias 2
1.3. Transformation Bias 4
1.4. Aggregation Bias 5
1.5. Liquidity Bias 8
1.6. Valuation Bias 10

Chapter 2. Liquidity and Pricing Biases in the Real Estate Market 18

2.1. Introduction 18
2.2. Real Estate Illiquidity vs. Illiquidity in the Finance Tradition 20
2.3. Measure of Real Estate Illiquidity 22
2.4. Liquidity and Pricing Biases in the Real Estate Market 25
2.4.1. Liquidity Bias 26
2.4.2. Valuation Bias 29

Chapter 3. Liquidity Bias in the Real Estate Market 35

3.1. Introduction 35
3.2. The Model 37
3.3. Liquidity Bias, Real Estate Illiquidity and Holding Period 42
3.4. Empirical Applications 47
3.5. Conclusions 50
 ii

Chapter 4. Valuation Bias in the Real Estate Market 55

4.1. Introduction 55
4.2. The Model 57
4.3. Probability of Sale and Expected Marketing Period 60
4.4. Valuation Bias, Real Estate Illiquidity and Holding Period 62
4.5. Risk Premium and Marketing Period 66
4.6. Empirical Applications 68
4.7. Conclusions 71

Chapter 5. Conclusion: The Risk Premium Puzzle in Real Estate Revisited 79

5.1. Introduction 79
5.2. The Risk Premium Puzzle in Real Estate 80
5.3. Current Valuation Approach to the Estimation of Real Estate Return
and Risk 83
5.4. The Risk Premium Puzzle in Real Estate Revisited 86
5.5. Conclusions 92
5.6. Future Research 95
 iii

Acknowledgements

I would like to thank my committee, Professors Kerry Vandell, Richard Green,

James Hodder, Steve Malpezzi and James Shilling, for their support. In particular, I

would like to recognize my advisor Kerry Vandell because without his guidance and

constant encouragement this dissertation would not have been possible. I consider

myself privileged to have been his student and am grateful for the resources he made

available for my research. I would also like to thank Professors François Ortalo-Magne

and Timothy Riddiough for their numerous discussions that have benefited this research.

I am also grateful to the participants in seminars at California State University at

Sacramento, Fannie Mae, Washington State University at Pullman, Wisconsin, the Asian

Real Estate Society, and the American Real Estate Society and the American Real Estate

and Urban Economic Association for their helpful comments. In addition, I greatly

appreciate Yongping Liang, Kiat-Ying Seah and David Ward for their editorial help, and

special thanks to my colleague and friend, John Lyon for helping me to proofread the

whole dissertation. Finally, I would like to thank my wife, Dr. Yingchun Liu, and my

parents for all they have given me has made possible in the pursuit of my dreams.

Financial support for this dissertation was due in part to fellowships from the

Department of Real Estate and Urban Economics and the Graduate School of the

University of Wisconsin.
 iv

Abstract

This dissertation is the first to relate the micro-analytic foundations of price

dynamics in the real estate market with those describing the real estate transaction

process. Based on the notion that real estate is a heterogeneous good that is traded in

decentralized markets and that transactions in these markets are often characterized by

costly searches, we argue that the most important aspects of real estate illiquidity in both

residential and commercial markets are the time required for sale and the uncertainty of

the marketing period. The randomness of the marketing period violates the fundamental

assumption of immediate execution in the completive market paradigm on which real

estate and modern finance theories are based. We examine how these aspects may bias

the commonly adopted methods of real estate valuation, which are based solely on the

prices of sold properties and implicitly assumes immediate execution. We demonstrate

that there are two pricing biases in the current real estate valuation approach. This

dissertation seeks to provide an approach to correct for these biases. In particular, we

study how real estate illiquidity, the marketing period and its uncertainty, plays a crucial

role in the creation of these two biases. We find that the current valuation approach,

which ignores real estate illiquidity, not only understates real estate risk but also

overstates real estate returns. Our findings can help us to understand the risk premium

puzzle in real estate: observed premiums are too large to be explained by standard finance

theories.
 1

Chapter 1

Introduction: The Pricing Biases in the Real Estate Market

It is well known that there are many bias issues in current real estate pricing

methodologies. The most notable in the literature include appraisal bias, sample selection

bias, transformation bias and aggregation bias. In this dissertation, we will focus on two

additional biases that are related specifically to the real estate transaction process, which

we shall term “liquidity bias” and “valuation bias”. We discuss each of these in turn.

1.1. Appraisal Bias: there are two lines of research in appraisal bias. The first

line is bias in appraisal-based returns. Gilberto (1988) demonstrates that even when

appraisal value is an unbiased estimator, the estimated holding period return is

consistently biased. In particular, when the appraisal errors are serially independent, the

bias is always positive. Geltner (1989a) extends Giliberto’s analysis by considering real

estate performance is measured in the form of the arithmetic mean of a time-series of

holding period returns. He identifies an additional source of bias and finds that the two

types of bias are likely to be of opposite sign, thus possibly offsetting one another,

causing the mean of a time-series of holding period returns to be an unbiased estimate of

real estate performance even if the holding-period return itself is biased. By studying

appraisal data from a commingled real estate fund, Gau and Wang (1990) show that the

bias of the holding-period return can be quite small and thus suggest that the mean of a

time-series of holding period returns could be a substantially biased measure of real

estate returns. Geltner (1991) points out a flaw in Gau and Wang’s empirical analysis;

they mistakenly treat cross-sectional for time-series moments. Therefore, contrary to their
 2

claim, they do not offer empirical support to the hypothesis that appraisal return bias is

small in the holding period return.

The second line of this research is smoothing in appraisal-based returns. Geltner

(1989b, 1991) and Ross and Zisler (1991) demonstrate that, because appraisers have

insufficient confidence to ignore the appraised values of the previous appraisal report, the

appraisal-based returns are less volatile. However, Webb, Miles and Guilkey (1992)

document that transaction-driven portfolio returns have approximately the same volatility

as appraisal-driven returns, although they report that property-specific transaction-driven

returns are more volatile than appraisal-driven returns. By using 3.7 million repeat

purchase and refinance transactions on mortgages bought by Fannie Mae and Freddie

Mac in the period of 1975 to 1993, Chinloy, Cho, and Megbolugbe (1997) test two

hypotheses in their analysis: whether any incentives for appraisals support an underlying

purchase offer with an over-valuation; and, whether appraisal data are smoothed or

exhibit less volatility than purchase data. Their findings indicate that appraisals are

systematically higher than purchase data, a first moment differential, and that appraisal

smoothing does not occur generally. More recently, Lai and Wang (1998) show that the

use of appraisal-based data may result in a higher (not lower) volatility than that of true

returns.

1.2. Sample Selection Bias: there exists a possibility of sample selection bias

when pricing real estate assets is solely based on the sale prices. If the transacted

properties are not a random sample of the whole population, sample selection bias may

result. For example, “starter homes” are generally defined as being relatively small,

having fewer amenities, and selling more frequently than more expansive properties.
 3

Assume a “starter home” and “expensive home” each have a 50% share in the market.

Due to its high frequency of transaction, the “starter home” is likely to have more than a

50% share in the transaction sample, and is therefore likely to be overrepresented in the

sold properties sample. This over-representation of the category of “starter homes” in the

transaction sample may cause the index to misrepresent the population of unsold

properties, and thus sample selection bias results.

A substantial real estate literature has expressed concerns about sample selection

bias in real estate price indices. For instance, Haurin and Hendershott (1991) suggest that

sample selection bias in transaction-based indices may result if the sample of sold

properties transact more frequently than the population of properties. Case, Pollakowski,

and Wachter (1997), Gatzlaff and Haurin (1997, 1998), Meese and Wallace (1997), and

Munneke and Slade (2000, 2001), among others, also show that real estate price indices

based on a sample of transacting properties may be subject to substantial sample selection

bias. Gatzlaff and Haurin (1997, 1998) find evidence of sample selection bias in the

estimation of residential house price indices, and Munneke and Slade (2000) find

evidence of sample selection bias in indices of commercial real estate.

The pioneering work in correcting sample selection bias includes Lee (1978) and

Heckman (1974, 1976 and 1979). Heckman’s two-step approach is most commonly used

to correct the sample selection bias (e.g. Munneke and Slade (2000, 2001), Englund and

Quigley and Redfearn (1999), and Gatzlaff and Haurin (1997, 1998), for example). In

this approach, the first step is to estimate the probability of sale by using a probit model

for the entire sample including transaction and non-transaction properties, and then

calculating the inverse Mills ratio, which is a monotonic decreasing function of the
 4

probability that a property is sold. The second step is to estimate the unbiased price

equation with the inclusion of the inverse Mills ratio by using only the transaction

sample. Sample selection bias is present if the parameter on the inverse Mills ratio is

statistically significant.

Using Heckman’s two-step procedure, a number of papers study sample selection

bias in various real estate price indices. Munneke and Slade (2000) in particular examine

commercial price indices and find the presence of sample selection bias. Gatzlaff and

Haurin (1998) investigate housing indices using the hedonic method and confirm the

existence of sample selection bias. Gatzlaff and Haurin (1997) also study housing indices

by using repeat-sales model and again find sample selection bias. Englund, Quigley and

Redfearn (1999) find a substantial sample selection bias, using an international sample of

all house sales in Sweden during the period 1981 to1993.

1.3. Transformation Bias: the repeat-sales model, first described by Bailey,

Muth and Norse (1963), is widely used to infer to real estate returns through time.

Because of the logarithmic transformation of price relatives, the repeat sales estimators

are essentially equal-weighted cross-sectional geometric averages, while the returns of

equal-weighted portfolios are arithmetic averages of cross-sectional individual asset

returns. Due to Jensen’s inequality, there exists a transformation bias between geometric

averages and arithmetic averages.

Several approaches have been proposed to address this bias problem. Shiller

(1991) suggests arithmetic repeat sales price estimators for equal-weighted and value-

weighted portfolios. Geltner and Goetzmann (2000) propose a nonlinear method that

minimizes the sum of squared residuals directly without taking logs first. More recently,
 5

Goetzmann and Peng (2002) propose an unbiased maximum likelihood that directly

estimates index returns.

1.4. Aggregation Bias: there are two types of aggregation bias. The first type is

spatial aggregation bias. Thomas and Stekler (1983) is an early example to study spatial

aggregation bias. They find evidence that construction activity varies across regions

within the United States. Therefore, they conclude that, prior models of construction,

based on only national time-series data, are subject to aggregation bias. Goodman (1998)

shows that, under a reasonable set of specifications and parameter values for local

housing markets, estimation on geographically aggregated data will result in biased

estimates of true parameters. Guttery and Sirmans (1998) empirically examine

aggregation bias in price indices for multi-family rental properties and find no evidence

of spatial aggregation bias.

The second type of aggregation bias is temporal aggregation bias. Geltner (1993)

shows that, even if a transaction value is identical to the true value of the property at the

time of the transaction, a transaction-based index may still be smoothed because

properties are not always transacted at the end of the period being studied. Dombrow,

Knight and Sirmans (1997) study temporal aggregation bias in repeat-sales indices. They

investigate the two underlying assumptions inherent in the repeat sales model: the

property characteristics must not have changed between sales, and the marginal

contribution of those characteristics to overall house price must be stable across periods.

They find that virtually all repeat-sales indices have bias.

This dissertation focuses on two additional biases, which are closely related to

real estate illiquidity. Based on the notion that real estate is a heterogeneous good that is
 6

traded in decentralized markets and that transactions in these markets are often

characterized by costly searches, we argue that the most important features of real estate

illiquidity are the time required for sale and the uncertainty of the marketing period.

These two features violate the fundamental assumption of immediate execution in the

completive market paradigm on which real estate and modern finance theories are based.

We examine how these features may bias the commonly adopted methods of real estate

valuation, which are based solely on historical prices and the time of sales and implicitly

assumes immediate execution. We show that there are two pricing biases in the current

real estate valuation approach: liquidity bias and valuation bias.

The remainder of this introduction proceeds as follows. We first discuss real

estate price dynamics in the context of the real estate transaction process. We then

demonstrate how liquidity bias and valuation bias may arise within current real estate

pricing methodologies.

Real Estate Price Dynamics: due to the uncertainty of the marketing period in

the real estate market, the formation of transaction prices in the real estate market is very

different from that in the financial market. First, the financial market is a homogeneous

and thickly traded market. At any instant, there are countless buyers and sellers, and

prices are determined by equating supply and demand. On the other hand, the real estate

market is a heterogeneous and thinly traded market and its prices are formed by sellers’

sequential search and their optimal stopping rule: accepting the first price above their

reservation price (e.g. Arnold (1999) and Yavas (1992)). Second, prices exist in the

financial market at any time. Sellers can sell their financial assets at the market price at

any time without waiting. In contrast, real estate prices exist only when there is a current
 7

buyer with an offer price that is at least as high as the seller’s minimum accepted price.

When there is either no buyer or an offer price is below the seller’s reservation price, the

real estate price does not exist and the sellers have to continue to search for the next

buyer. Hence, sellers cannot sell their real estate assets at any given price without

waiting. Due to the nature of stochastic arrival of potential buyers and the uncertainty of

their offer prices, time on market cannot be fully controlled by the sellers. Therefore,

unlike the financial market, real estate investors face not only price risk, as in the

financial market, but also marketing period risk.

Figure 1.1 illustrates the price dynamics in the context of the transaction process

in the real estate market. Here, TH is the time when the real estate asset is placed on the

~
market, and ~
t is a possible marketing period with a sale price PTH + ~t ( ~
t = t1 , t 2, ...) . In each

passing period, the real estate seller faces random arrivals of potential buyers. There are

two possibilities: (1) a successful sale when a potential buyer is present and his/or her

asking price equals or exceeds seller’s minimum accepted price; (2) no successful sale

when there is no buyer at present or an asking price is too low for the seller to accept.

~
Marketing Period t (random)

TH TH + t1 TH + t2 ••• TH + ~
t •••

••• •••
~ ~ ~ ~
t = 0 PTH + ~t ~
PTH + ~t ~ t = t1 PTH + ~t ~
t = t2 PTH + ~t

Return upon A Successful Sale (random)

Figure 1.1
 8

1.5. Liquidity Bias: the current real estate pricing methodologies, such as the

repeat-sales model, are based solely on historical prices (or appraisal values) and the time

of transactions, which have nothing to do with the uncertainty of the marketing period.

However, the uncertainty of the marketing period can be an additional risk to the

investor. For example, suppose there are N similar real estate assets placed on the

market at time TH . For simplicity, assume these assets were bought at the time 0 with an

original price of $1. At any given time T , assume there are n ( n ≤ N ) assets being

successfully sold. Figure 1.2 illustrates the time and the price of each transacted asset,

PTH PT1 PT2 PT3 PTH +T

•••

TH T1 T2 T3 TH + T

Figure 1.2

At time TH + T , we observe that property i has been sold at time Ti with an

average rate of return ( PTi − 1) / Ti ( i = 1,2,..., n) and note that the time of sale Ti is

known. Therefore, looking back, the time of the sale for asset i is completely

deterministic. However, if looking forward at time TH , the time when these assets will be

successfully sold is not known. It can be TH when they are just placed on the market, and

can be TH + t1 , TH + t 2 , and so on. The marketing period ~

t can be 0 , t1 , t 2 , …, and is a
 9

random variable. It is almost impossible for real estate sellers to sell their assets at a

transaction price in a pre-specified time. Hence, at the time the assets are placed on the

market, the sellers face two kinds of risk: price risk and the marking period risk.

Since the time of sales contain no information on how long the properties have

been on the market before they are successfully sold and thus have nothing to do with the

uncertainty of marketing period. Consequently the risk of marketing period cannot be

priced in the current real estate valuation. In the financial market, since the time required

to trade an asset is trivial, the risks associated with uncertain trading periods can be often

ignored. However, marketing period in the real estate market is highly variable and

generally extends to several months or even years. According to the data from the

National Association of Realtors, the average time on the market for residential

properties in the United States was eight to ten months in the early 1990s, when the

market was stagnant, and four to five months in more recent years when the market was

performing well (see Figure 1.3). In the commercial property markets, the average time

on the market may be even longer.

Figure 1.3. Months on Market for Selling a House in US

January 1989 - September 2003
12
10
8
Months

6
4
2
0
1

9
M

M
89

90

91

93

94

95

97

98

99

01

02

03
19

19

19

19

19

19

19

19

19

20

20

20
 10

Since the current real estate pricing methodologies can only estimate price risk

but not the marketing period risk, we need a new approach to measure real estate

performance by incorporating both price risk and marketing period risk.

As we know, historical prices are the data recorded on an “ex-post”, or after the

fact basis of successful sales. In reality, a real estate seller who tries to sell his asset does

not know when it will actually be sold; i.e., “ex-ante”. In order to distinguish our new

approach from the current valuation, we name our new measure the ex-ante measure and

the current measure the ex-post measure. The liquidity bias is defined as the pricing

difference between these two measures. Therefore, the liquidity bias essentially captures

the effect of the uncertainty of marketing period on real estate pricing.

1.6. Valuation Bias: besides liquidity bias, there is another bias that is related to

real estate illiquidity. In the financial market, since investors can sell their assets at

observed market prices almost immediately, a historical price at time t can represent the

price of this asset at that time, meaning that a seller could trade at that price at that time.

However, due to the uncertainty of the marketing period, there are two differences worth

noting in the real estate market.

First, in each time period, only a small portion of properties is sold successfully,

while a large portion of properties is still sitting on the market. The transaction prices at

time t may reflect prices of the sold properties at that time; however, they may not be

able to represent prices of the similar properties that are still waiting on the market.

Second, sellers have often already waited on the market for a long period of time

before selling their real estate assets. Current real estate transaction prices are often the

prices of those assets that have been on the market for some time. Unlike the financial
 11

market, in which the price at time t is often the price of those assets recently put on the

market, there is a substantial time lag between the time when real estate is placed on the

market and when it is sold. Hence, the price of sold properties at time t may not

represent the price of other properties recently placed on the market. In other words,

sellers would be unlikely trade at observable transaction prices of the time when they

place their assets on the market.

The risk of a substantial time lag between the time of sale and time of being put

on the market is important to real estate investors. For example, a household may

experience surprise liquidity shocks including a job loss or a divorce, and facing a

borrowing constraint, must sell its illiquid real estate in a short time period; firms may

have sudden investment opportunities, but face an imperfect and costly external capital

market, and hence have to sell their real estate asset immediately when such an

opportunity arrives; fund managers may face a unexpected increase in withdrawals,

resulting in a need to liquidate their real estate asset. In all these situations, when such a

shock occurs, real estate investors are forced to sell their real estate assets immediately.

How does this “quick sale” have an impact on real estate pricing when such an

excessively high “time lag” is present? In order to answer this question, we need to look

at how transaction prices are formed in the real estate market.

In the discussion above, a real estate transaction price can be observed if and only

if a buyer’s bidding price equals or exceeds seller’s minimum acceptable price (MAP).

Hence, transaction prices are the prices when a bidding price is above the seller’s

minimum acceptable price. In others words, transactions prices are the prices from a

truncated distribution of bidding prices.

 12

Figure 1.4 illustrates the relationship between the distribution of transaction prices

and the distribution of bidding prices. At any time t , suppose that bidding prices are

distributed at between B and A, where Pt is a possible minimum bidding price and Pt is

a possible maximum bidding price. Since a transaction price is the price when a bidding

price is at least as high as seller’s minimum acceptable price, it is distributed at between

O and A.

A
Pt
Observable
MAPt O

B Unobservable
Pt
t

Figure 1.4

There are two points worth noting in Figure 1.4. First, a transaction price is a

bidding price above the seller’s minimum acceptable price. Hence, the expected

transaction price is likely to be higher than the expected bidding price in each point of

time. However, the seller has to face the uncertainty of marketing period to receive a

transaction price. Second, to the investor who experiences a surprise liquidity shock and

has to sell his real estate asset within a short period of time, he is expected to receive a

price from the distribution of bidding prices rather than the distribution of transaction

prices. In other words, a bias may result if his return and risk is directly estimated from
 13

transaction prices. We call this bias as valuation bias. Therefore, the valuation bias can be

regarded as the value of the time on market.

~ Transaction Prices
t ≈0 ~ ~ ~
( Pt transactio n = Pt Pt > MAPt )
Bidding Prices
~
( Pt )
~ Marketing Period Risk
t >0
E[ ~
t]>0
Var ( ~
t)>0

Transaction Prices
Valuation Bias ~ ~ ~ Liquidity Bias
( Pt transactio n = Pt Pt > MAPt )

Figure 1.5

The relationship between liquidity bias and valuation bias can be illustrated in
~
Figure 1.5. With the presence of marketing period risk ( t > 0 ), real estate investors face

not only the risk from transaction prices but also the marketing period risk. The liquidity

bias is due to the fact that the current real estate pricing methods fail to take the

marketing period risk into account, while the valuation bias arises when transaction
 14

prices are applied to the investors who are involved in real estate business and face quick

sales.

This dissertation is the first to combine price dynamics in the real estate market

with explicit modeling of the transaction process. Although the models developed in this

dissertation are based on the illiquid real estate market, the theory and methods in this

study are equally applicable to all markets, in which sales are characterized by the

process of acceptance or rejections of sequential bids associated with uncertainty of time

of sale (e.g. the market for art and private placements of other assets). The difference

between different types of markets is in the parameter values rather than in the essential

behavioral nature of the market.

The remainder of this dissertation proceeds as follows. In Chapter 2, we first

discuss real estate illiquidity and its measurement. We then examine how real estate

illiquidity may affect the two pricing biases. In Chapters 3 and 4, we study how real

estate illiquidity plays a crucial role in the creation of the two biases and show how can

correct for these biases by using marketing period information. In Chapter 5, we revisit

the risk premium puzzle in real estate and conclude with our future research.
 15

Reference:

Arnold M. (1999), “Search, Bargaining and Optimal Asking Prices,” Real Estate
Economics 27, 453-481.

Bailey M., Muth R. and Nourse H. (1963), “A Regression Method for Real Estate Price
Index Construction,” Journal of the American Statistical Association, 58, 933-42

Case B., Pollakowski H. and Wachter S. (1997), “Frequency of Transaction and House
Price Modeling,” Journal of Real Estate Finance and Economics, 173-187

Chinloy P., Cho M., and Megbolugbe I. (1997), “Apprasials, Transaction Incentives, and
Smoothing”, Journal of Real Estate Finance and Economics, 89-111

Dombrow J., Knight J. and Sirmans C. (1997), “Aggregation Bias in Repeat-Sales

Indices”, Journal of Real Estate Finance and Economics, 75-88

Englund P., Quigley J. and Redfearn C. (1999), “Do Housing Transactions Provide
Misleading Evidence About The Course of Housing Values?” Working Paper.

Gan G. and Wang K. (1990), “A Further Examination of Appraisal Data and the Potential
Bias in Real Estate Return Indexes”, AREUEA Journal, 40-48

Gatzlaff D. and Haurin D. (1997), “Sample Selection Bias and Repeat-Sales Index
Estimates,” Journal of Restate Finance and Economics, 33-50

Gatzlaff D. and Haurin D. (1998), “Sample Selection and Biases in Local House Value
Indices,” Journal of Urban Economics, 199-222

Geltner D. (1989a), “Bias in Appraisal Based Returns”, AREUEA Journal, 338-352

Geltner D. (1989b), “Estimating Real Estate’s Systematic Risk from Aggregate Level
Appraisal-Based Returns”, AREUEA Journal, 463-481

Geltner D. (1991), “Smoothing in Appraisal-Based Returns,” Journal of Real Estate

Finance and Economics, p327-345

Geltner D. (1993), “Temporary Aggregation in Real Estate Return Indices”, AREUEA

Journal, 141-166

Geltner D. and Goetzmann W. (2000), “Two Decades of Commercial Property Returns:

A Repeated-Measures Regression –Based Version of the NCREIF Index”, Journal of
Real Estate Finance and Economics, 5-21

Giliberto M. (1988), “A Note on the Use of Appraisal Data in Indexes of Performance

Measure”, AREUEA Journal, 77-83
 16

Goetzmann W. and Peng L. (2002), “The Bias of the RSR Estimator and the Accuracy of
Some Alternatives”, Real Estate Economics, 13-39

Goodman J. (1998), “Aggregation of Local Housing Markets”, Journal of Real Estate

Finance and Economics, 45-53

Guttery R. and Sirmans C. (1998), “Aggregation Bias in Price Indices for Multi-Family
Rental Properties”, Journal of Real Estate Research, 309-325

Haurin D. and Hendershott P. (1991), “House Price Indexes: Issues and Results,”
AREUEA Journal, 259-269

Heckman J. (1974), “Shadow Wages, Market Prices, and Labor Supply,” Econometrica,
679-684

Heckman J. (1976), “The Common Structure of Statistical Models of Truncation, Sample

Selection and Limited Dependent Variables and A Simple Estimator for Such Models,”
Annals of Economic and Social Measurement, 475-492

Heckman J. (1979), “Sample Selection Bias As A Specification Error,” Econometrica,

153-797

Lai T. and Wang K. (1998), “Appraisal Smoothing: The Other Side of the Story”, Real
Estate Economics, 511-535

Lee L. (1978), “Some Approaches to the Correction of Selectivity Bias,” Review of

Economic Studies, 355-372

Meese R. and Wallace N. (1997), “The Construction of Residential Housing Price

Indices: A Comparison of Repeat-Sales, Hednoic-Regression, and Hybrid Approaches,”
Journal of Real Estate Finance and Economics, 51-73

Munneke H. and Slade B. (2000), “An Empirical Study of Sample-Selection Bias in

Indices of Commercial Real Estate,” Journal of Real Estate Finance and Economics, 45-
64

Munneke H. and Slade B. (2001), “A Metropolitan Transaction-Based Commercial Price

Index: A Time-Varying Parameter Approach,” Real Estate Economics, 55-84

Ross S. and Zisler R. (1991), “Risk and Return in Real Estate”, Journal of Real Estate
Finance and Economics, 175-190

Shiller R. (1991), “Arithmetic Repeat Sales Price Estimators”, Journal of Housing

Economics, 110-126
 17

Thomas R. and Stekler H. (1983), “A Regional Forecasting Model for Construction

Activity”, Regional Science and Urban Economics, 557-577

Yavas A. (1992), “A Simple Search and Bargaining Model of Real Estate Markets,”
Journal of the American Real Estate and Urban Economics Association 20, 533-548

Webb R., Miles M. and Guilkey D. (1992), “Transactions-Driven Commercial Rea Estate
Returns: The Panacea to Asset Allocation Models?” Journal of the American Real Estate
and Urban Economics Association, 325-357
 18

Chapter 2

Liquidity and Pricing Biases in the Real Estate Market

2.1. Introduction

Perhaps the most important distinction between real estate and financial assets is

that real estate is a differentiated commodity. The lack of standardization of real estate

assets, including their geographic fixity, implies that they are unlikely to be traded in a

central marketplace. At the very least, trading in the real estate market is much slower

than in centralized financial asset markets. The real estate market is often characterized

by a heterogeneous product, infrequent market transactions, and high information and

transaction costs. Search is necessary in the real estate market because of imperfect

information resulting from, among other considerations, product heterogeneity and

decentralized trading. At any period of time, buyers and sellers must search out one

another to establish a transaction price (e.g. Arnold (1999)). The search process is costly

for both sellers and buyers in terms of time and money. For sellers, search costs consist of

additional holding costs, selling expenses and the opportunity cost of time.

During the search process, sellers receive offers over time from the stochastic

arrival of buyers. Buyers make offers based on the information acquired from their

search. Each time a buyer makes an offer, the corresponding seller will need to evaluate

the benefits of waiting for a potentially better offer and the cost associated with waiting,

and then decide whether to reject that offer or to accept it. If an offer is rejected, the

search continues.
 19

The sequential search and decision process in the real estate market can be best

represented by the search model in labor economic literature (e.g. Morgan and Manning

(1985), Rothschild (1974) and Stigler (1961)). The search model has been extensively

applied to the real estate market since the 1980s (e.g. Green and Vandell (1998), Yavas

(1992), Quan and Quigley (1991), Salant (1991), Read (1988), Lippman and McCall

(1986), and Yinger (1981)). At each point in time, the optimal marketing strategy for

choosing an offer is to accept the first bid above a certain reservation price (optimally set)

and to reject all bids below. 1

Therefore, unlike the financial market where at any point in time sellers can sell

their financial assets almost immediately at the observed market price, real estate sellers

cannot sell their real estate assets without waiting. Oftentimes, a seller has to wait for six

months or more in the residential market and even longer time in the commercial market.

In addition, due to the random arrivals of potential buyers and the uncertainty of the

bidding prices, the waiting time (marketing period) after the property is placed on the

market is purely random. Therefore, there are two distinctive features in the real estate

market: marketing period required before real estate assets can be exchanged for money

(e.g. Lippman and McCall (1986) and Krainer (1997, 1999)) and the uncertainty of the

marketing period (e.g. Anglin (2003)). 2

The time required for a sale in the real estate market is an important element in

discussing the illiquidity of real estate. Next, we first discuss how real estate illiquidity

differs from illiquidity in the finance tradition. We then study how we can measure real

1
A seller’s reservation price depends on his opportunity cost of waiting, information about the
determinants of value of the property, and about the distribution of potential offers; it changes over time
when market conditions change (e.g. Anglin (2003)) or as the sellers update the bid distribution by learning
(e.g. Green and Vandell (1998)).
2
Anglin (2003) characterizes real estate illiquidity as the uncertainty of sale in each marketing period.
 20

estate illiquidity. Finally, we examine how we can quantify the pricing biases of liquidity

bias and valuation bias.

2.2. Real Estate Illiquidity vs. Illiquidity in the Finance Tradition

Modern finance theory is based on the competitive market paradigm (see Jarrow

and Turnbull (1966), and Duffie (1992)), which has two important assumptions. First, all

market orders for sales/purchases have immediate execution. Second, the markets are

perfectly elastic and traders act as price-takers. Liquidity risk occurs when either of these

two assumptions is violated.

The “price-taker” assumption implies that the market is perfectly elastic in both

supply and demand sides and orders have no impact on price. This assumption is widely

thought to be violated in the finance literature. A sudden large purchase should have a

positive impact on price, and a sudden large sale should have a negative impact on price.

Therefore, asset illiquidity in the finance literature often defines the impact of order flow

on price. By this definition, numerous illiquidity measures employed in the literature

include bid-ask spreads (e.g., Amihud and Mendelson (1986), Chordia, Roll and

Subrahmanyam (2001)), transaction costs (e.g. Constantinides (1986), Instefjord (1999)),

liquidity discount (e.g. Subramanian and Jarrow (2001), Huang (2003)), average ratio of

the daily absolute return to the (dollar) trading volume on that day (e.g. Amihud (2002)),

the change in a firm’s stock price associated with its observed net trading volume (e.g.

Breen, Hodrick and Korajczyk (2002)), the probability of information-based trading (e.g.

Easley, Hvidkjaer and O’Hara (1999)), and so on. Asset illiquidity is an elusive concept.

It is not observed directly, but rather has a number of aspects that cannot be captured in a
 21

single measure (see Amihud (2002)). The measures of illiquidity cited above can only be

regarded as empirical proxies that measure different aspects of illiquidity.

There have been two different perspectives in studying financial illiquidity. The

first perspective is that liquidity risk is due to asymmetric information. Large

purchases/sales reveal private information, and cause order imbalance, both of which

impact the prices (e.g. Instefjord (1999), Glosten and Milgrom (1985), Kyle (1985), Chan

and Fong (2000), and Hasbrouck and Seppi (2001)). 3 The second perspective is that

liquidity risk is due to transaction costs associated with immediate execution (e.g. Huang

(2003), Duffie and Ziegler (2003), and Aiyagari and Gertler (1991), and Constantinides

(1986)). Three major sources of such transaction costs are often discussed: brokerage

commissions, bid-ask spreads, and market impact costs.

Due to its heterogeneity, as discussed earlier, the formation of real estate

transaction prices is very different from that of financial prices. Financial price is formed

by equating supply and demand at any interval of time. Since orders for purchases (or

sales) change the demand (or supply) curve and hence affect price, financial asset

illiquidity is measured by the impact of order flow on price. Real estate price is formed

by sellers and buyers’ sequential search together with sellers’ optimal stopping rule;

hence real estate illiquidity can hardly be measured by the impact of order flow on price.

The “immediate execution” assumption implies that assets can be sold

immediately at any time in the market. This assumption may be reasonable in financial

markets, where sellers normally take only a few minutes to trade their assets. However, it

cannot be justified in both commercial and residential real estate markets, where sellers

3
Instefjord (1999), Glosten and Milgrom (1985), Kyle (1985) focus on how trading by informed traders
impact on prices. Chan and Fong (2000), and Hasbrouck and Seppi (2001) study the impact of order
imbalance on prices.
 22

often have to wait for months or even years to sell their assets at transaction prices. Due

to a random mismatch between sellers and buyers, execution delay in the real estate

market is not only substantial, but also uncertain.

Since most work on asset illiquidity in the finance literature maintains the

assumption of immediate execution, and the distinctive aspect of real estate illiquidity is

the uncertainty of execution delay, the existing literature on financial illiquidity does little

to help us fully understand real estate illiquidity, let alone understand the role of real

estate illiquidity in pricing dynamics.

In order to understand the role of real estate illiquidity in pricing, we first discuss

the measure of real estate illiquidity.

2.3. Measure of Real Estate Illiquidity

Considering the marketing period as an aspect of asset illiquidity has been

discussed in the literature. Several illiquidity studies, such as Lippman and McCall

(1986), Brealey and Myers (1988), Greer and Farell (1992), Tosh (1992), Krainer (1997,

1999), and Anglin (2003), define liquidity based on whether the marketing period is

necessary in order to trade at a market price. According to Brealey and Myers (1988),

“liquidity means that you don’t have to accept a discount from a market price if you want

to sell the asset quickly.” Greer and Farell (1992) define liquidity as “the ability to

convert an asset to cash without incurring loss.” Tosh (1992) defines liquidity only as

“the quickness and ease with which an asset may be converted to cash.” Lippman and

McCall (1986) and Krainer (1997,1999) provide a similar definition of liquidity. They

define a measure of illiquidity as the expected waiting time after the asset is placed on the
 23

market. In order to capture the randomness of the marketing period, Anglin (2003)

defines illiquidity as the probability of sale in each marketing period.

We define real estate illiquidity not only as the expected marketing period as

Krainer (1997, 1999), Lippman and McCall (1986) and Tosh (1992) among others do, but

also as the uncertainty of the marketing period as Anglin (2003) does. Next, we discuss

how we can measure real estate illiquidity.

Suppose an investor purchases a real estate asset at time 0 and places it on the

market at time TH . Afterwards, there are stochastic arrivals of potential buyers. t1 is the

waiting time of the first buyer, t 2 is the waiting time of the second buyer, and t i is the

waiting time of ith buyer. Assume the seller’s reservation price at time of the ith buyer’s

arrival is Ptireservation and the buyer’s bidding price is Ptibid . Note that the reservation price is

optimally, not arbitrarily, set by the seller. By the optimal stopping rule, the probability of

a successful sale at time t i given that the real estate has not been sold is

pi = Pr ob( Ptibid ≥ Ptireservation ) (2.1)

Therefore, we can deduce that the real estate asset will be sold at the first offer

with probability of p1 , at the second offer with probability of p 2 (1 − p1 ) , at the third

offer with probability of p3 (1 − p1 − p 2 (1 − p1 )) , and so on. Since pi (i = 1,2,...) is

largely determined by the market, the probability of a successful sale at each offer cannot

be fully controlled by the seller.

 24

Since the seller does not know when a potential buyer will arrive after TH when

the asset is put on the market, the arrival time TH + t i is also stochastic. We thus denote

~
it as TH + ti to emphasize its randomness. Figure 2.1 illustrates the possibility of sale at

time TH + ~
ti ( ~ ~
ti = t1 , t 2, ...t i ,...) , where TH is the seller’s holding time and ti is the seller’s

marketing period. Hence, the marketing time ~

ti of a successful sale not only depends on

the arrival distribution of potential buyers but also on whether buyers’ bidding prices are

higher than the seller’s reservation price. Since both the arrival distribution and seller’s

reservation price are closely related to market conditions, the uncertainty of the

marketing period is beyond the seller’s control.

Holding Period Marketing Period

••• •••

0 TH ( t1 + TH )( t 2 + TH ) (~
ti + TH )

Figure 2.1

We measure real estate illiquidity by the first and second moment of the
~ ~
marketing period TM = ti ,

~
t M = E[TM ] (2.2)

~
σ t2 = Var[TM ]
M
(2.3)
 25

In the next section, we discuss how real estate illiquidity plays a role in real estate

pricing that current valuation methods have ignored.

2.4. Liquidity and Pricing Biases in the Real Estate Market

There are a few papers studying the impact of marketability risk on prices, which

are somewhat related to the violation of immediate execution (e.g. Kahl, Liu and

Longstaff (2003), Dufour and Engle (2000), Longstaff (1995), Silber (1991), and

Stapleton and Subrahmanyam (1979), Brito (1977, 1978), among others). The issue of

how marketability impacts prices is becoming increasingly important not only to the

academic community but also to rating agencies, auditors, regulators, and institutional

investors. There are many situations in which the marketability of a security may be

restricted. For example, the marketability of initial public offering (IPO) shares can be

temporarily restricted for some investors (see Field and Hanka (2001)). This is because

underwriters often pressure investors who are allocated shares in an IPO to not resell the

shares immediately. Another good example is letter stock, which is issued by firms under

SEC Rule 144 and cannot be sold by an investor for a two-year period after it is acquired.

A restricted sale can be treated as an execution delay. For example, if shareholders want

to sell their shares before the end of the restriction period, but have to wait until the end

of the restriction period, execution delay occurs. In this case, the delay can be regarded as

the difference between the time of willingness to sell and the end of the restriction period.

Although marketability risk in the financial literature is often considered in a

situation where a security is restricted from sales in a pre-specified period of time, these

studies suggest that the impact of execution delay on prices can be rather large. For
 26

example, Silber (1991) found that letter stock is 30 to 35 percent less valuable than other

identical unrestricted stock; Longstaff (1995) found a similar result.

In the case of the real estate market, execution delay or marketing period is not

only substantial but also random. Intuitively, the uncertainty of the marketing period

should have an even bigger impact on price if other factors are equal. Therefore, the

current approach, which assumes immediate execution and ignores marketing period risk,

may bias real estate pricing. In other words, real estate illiquidity should be priced in real

estate valuation. 4

In the coming section, we will discuss two pricing biases in the current valuation:

liquidity bias and valuation bias.

2.4.1. Liquidity Bias

Since the current pricing approach is unable to price the risk of the marketing

period, we propose a new approach, ‘ex-ante measure’, that can incorporate both price

risk and marketing period risk. We then demonstrate why ignoring the uncertainty of the

marketing period biases real estate return and risk.

Ex-ante Measure vs. Ex-post Measure

In the previous example, we assumed that a seller places his real estate asset on
~
the market after holding it for TH periods. Suppose the seller receives return RTH + ~t upon

4
In this dissertation, we use the following terms interchangeably: marketing period, execution delay, and
time-on market.
 27

~ ~
successfully selling it at time TH + t ( t = t1 , t 2 ,... ), and note that the marketing period ~
t

~
and the return RTH + ~t are both random.

In the financial market, the difference between the time when an asset is

successfully sold and the time when the asset is placed on the market is trivial. Compared

to the substantial time lag in the real estate market, the time lag in the financial market

can be often ignored and treated as immediate execution. In this sense, a seller in the

financial market faces only price risk.

We define the ex-ante measure as the measure unconditional upon a successful

sale. Hence, ex-ante expected return and risk can be defined as follows,

~ ~
E ex − ante [ RTH + ~t ] = E~ [ E~ [ RTH + ~t ~
t ]] (2.4)
t R

~ ~ ~
Var ex − ante [ RTH + ~t ] = E~ [ E~ [ RTH + ~t − E ex − ante [ RTH + ~t ]]2 ~
t ]] (2.5)
t R

Note that ex-ante expected return and risk are closely related to the uncertainty of the

marketing period ( ~
t ).

We define the ex-post measure as the measure conditional upon immediate

~
execution ( t = 0 ). Hence, ex-post expected return and risk are defined as follows,

~ ~
E ex − post [ R~t +TH ~
t = 0] = E[ RTH ] (2.6)

~ ~ ~
Var ex − post [ R~t +TH ~
t = 0] = E[ RTH − E[ R TH ]] 2 (2.7)
 28

Since the ex-post expected return and risk have nothing to do with the uncertainty

of the marketing time, the ex-post measure only involves price risk. By this definition, the

current valuations of return and volatility are essentially the ex-post return and volatility.

Equations (2.4)-(2.7) indicate that ex-ante expected return and risk are the same as

the ex-post expected return and risk if and only if assets can be sold immediately. In the

presence of the uncertainty of the marketing period, how do they differ from each other?

The following example will illustrate their difference.

Suppose an investor purchases a real estate asset at time 0 and puts it on the

market at period 1. Assume the probabilities of a successful sale at marketing period 0, 1,

and 2 are 1/4, 1/2, and 1/4, respectively. We assume that the return upon a successful sale

at period i is distributed with mean iu and variance iσ 2 , where i = 1, 2, 3 . We can

compute the expected marketing period for this asset as,

E [t ] = 0 × 1 / 4 + 1 × 1 / 2 + 2 × 1 / 4 = 1 (2.8)

Therefore, the expected holding time until sale for the seller is 2 periods – one for

holding the asset and one for the expected marketing period. If the seller can immediately

sell his asset at period 2, by assumption his ex-post return and risk are 2u and 2σ 2

respectively. However, his ex-ante return and risk can be shown to be,

~
E ex− ante [ R1+ ~t ] = 2u (2.9)

~ 1
Var ex − ante ( R1+ ~t ) = 2σ 2 + u 2 (2.10)
2
 29

Therefore, current valuation of real estate risk, which assumes real estate can be

successfully sold at transaction prices immediately, may underestimate real estate risk,

and thus have liquidity bias.

2.4.2. Valuation Bias

In the discussion above, we mainly focused on how the uncertainty of the

marketing period can be an additional risk to a real estate investor, given that there is no

bias in the transaction prices to the investor. In this section, we will examine whether this

assumption holds.

To begin with, we consider a model of house sales. For illustration purposes, we

omit the time subscript and assume that the house market value is,

P = V0 + ε , ε ~ N (0, σ ε2 ) (2.11)

Following Goetzmann and Peng (2003) and Munneke and Slade (2000), we assume

that potential buyers offer a bidding price based on market valuation (2.11), and a seller

decides whether or not to accept an offer based on his reservation price.5 Suppose the

seller’s reservation price is P * , which depends on the seller’s opportunity cost of waiting,

information about the determinants of value of the property, and the distribution of

potential offers, etc.

5
By this assumption, market valuation is essentially the valuation when there is no marketing period risk,
~
i.e., t = 0 .
 30

A transaction price can be observed if and only if the bidding price equals or

exceeds the reservation price:

⎧⎪ P if P ≥ P * ,
P =⎨
T
(2.12)
⎪⎩unobserved if P < P * .

Therefore, a transaction price is the price when market valuation is at least as high

as the seller’s reservation price. As a result, the distribution of transaction prices is a

“truncated” distribution of underlying market valuation.

As discussed earlier, some sellers may have to sell their houses immediately

because of job loss, divorce or other financial distress (e.g. Huang (2003) and Glower,

Hendershott and Haruin (1995)). In this case, they will receive the return and volatility

from the underlying market valuation rather than from the observable transaction prices.

If using the return and volatility estimated from the transaction prices, valuation bias will

certainly occur.

In order to see how serious the valuation bias might be, we consider a simple case

in which the annualized market return is uniformly distributed over [ r , 3r ] , which is

time-invariant. Suppose investors on the market have the same reservation return 2r .

Hence, the distribution of transaction returns is a uniform distribution over [ 2r , 3r ] . As a

result, the return and volatility based on the transaction sample can be expressed as

follows,
 31

rT = 2.5r
r2 (2.13)
σ T2 =
12

However, the investors who have to sell their assets immediately will not receive

those return and volatility. Instead, they will the return and volatility as follows,

rM = 2r
r2 (2.14)
σ M2 =
3

Equations (2.13) and (2.14) yield,

4
rM = rT
5 (2.15)
σM = 2σ T

Therefore, using the return and volatility estimated from the transaction sample not only

overstates the return by 25% but also underestimates the volatility by 50%.6

rT − rM σ −σ M
6
Equation (2.15) implies that = 25%, and T = −50% .
rM σM
 32

Reference:

Aiyagari R. and Gertler M. (1991), “Asset Returns with Transaction Costs and Uninsured
Individual Risk,” Journal of Monetary Economics, 27, 311-331

Amihud Y. (2002), “Illiquidity and Stock Returns: cross-section and time-series effects,”
Journal of Financial Markets, 31-36

Amihud Y. and Mendelson H. (1986), “Asset Pricing and the Bid-ask Spread,” Journal of
Financial Economics, 223-249

Anglin P. (2003), “The Value and Liquidity Effects of A Change in Market Conditions,”
Working Papers, University of Windsor

Arnold M. (1999), “Search, Bargaining and Optimal Asking Prices,” Real Estate
Economics 27, 453-481.

Brealey R. and Myers S. (1988), Principles of Corporate Finance. McGraw-Hill.

Breen W., Hodrick L. and Korajczyk R. (2001), “Predicting Equity Liquidity,”

Management Science, 470-483

Brito N. (1977), “Marketability Restrictions and the Valuation of Capital Assets under
Uncertainty,” Journal of Finance, 1109-1123

Brito N. (1978), “Portfolio Selection in an Economy with Marketability and Short Sales
Restrictions,” Journal of Finance, 589-601

Case K. and Shiller R. (1987), “Prices of Single-Family Homes Since 1970: New Indexes
For Four Cities,” New England Economic Review, 45-56.

Chan K. and Fong W. (2000), “Trade Size, Order Imbalance, and the Volatility-Volume
Relation,” Journal of Financial Economics, 247-273

Chordia, Roll and Subrahmanyam (2001), “Market Liquidity and Trading Activity,”
Journal of Finance, 501-530

Constantinides G. (1986), “Capital Market Equilibrium with Transaction Costs,” Journal

of Political Economy, 842-862.

Dufour A. and Engle R. (2000), “Time and the Price Impact of A Trade,” Journal of
Finance, 2467-2498

Duffie D. (1992), “Dynamic Asset Pricing Theory,” Princeton, NJ: Princeton University
Press
 33

Duffie D. and Ziegler A. (2003), “Liquidation Risk,” Financial Analysts Journal, 42-51

Easley D., Hvidkjaer and O’Hara N. (1999), “Is information risk a determinant of asset
returns?” working papers, Cornell University

Field L. and Hanka G. (2001), “The Expiration of IPO share lockups,” Journal of
Finance, 471-500

Glosten L. and Milgrom P. (1985), “Bid, Ask, and Transaction Prices in a Specialist
Market with Heterogeneously Informed Traders,” Journal of Financial Economics, 14,
71-100

Glower M., Haurin D. and Hendershott P. (1995), “Selling Price and Selling Time: The
Impact of Seller Motivation,” NBER Working Paper No. 5071.

Goetzmann W. and Peng L. (2003), “Estimating Indices in the Presence of Seller

Reservation Prices,” Working Paper, Yale University

Green R. and Vandell K. (1998), “Optimal Asking Price and Bid Acceptance Strategies
for Residential Sales,” Working Paper, Center for Real Estate and Urban Land
Economics, Univ. of Wisconsin - Madison.

Greer G. and Farrell M. (1992), Investment Analysis for Real Estate Decisions, Dearborn.

Hasbrouck J. and Seppi D. (2001), “Common Factors in Prices, Order Flows and
Liquidity,” Journal of Financial Economics, 383-411

Huang M. (2003), “Liquidity Shocks and Equilibrium Liquidity Premia,” Journal of

Economic Theory, 105-129

Instefjord N. (1999), “A Simple Model of Market Liquidity,” Economics Letters, 329-

337

Jarrow R. and Turnbull S. (1996), “Derivative Securities”. Cincinnati, OH: Southwestern

Publishers

Kahl M., Liu J. and Longstaff F. (2003), “Paper Millionaires: How Valuable is Stock to a
Stockholder Who is Restricted from Selling it?,” Journal of Financial Economics, 385-
410

Krainer J. (1997), “Pricing Illiquid Assets,” Ph.D. dissertation, University of Minnesota

Krainer J. (1999), “Real Estate Liquidity,” FRBSF Economic Review, 14-26.

Kyle A. (1985), “Continuous Auction and Insider Trading,” Econometrica, 53, 115-1335.
 34

Lippman S. and McCall J. (1986), “An Operational Measure of Liquidity,” American

Economic Review 76, 43-55.

Longstaff F. (1995), “How Much Can Marketability Affect Security Values?,” Journal
of Finance, 1767-1774

Morgan P. and Manning R. (1985), “Optimal Search,” Econometrica 53, 923-944.

Munneke H. and Slade B. (2000), “An Empirical Study of Sample-Selection Bias in

Indices of Commercial Real Estate,” Journal of Real Estate Finance and Economics, 45-
64

Quan D. and Quigley J. (1991), “Price Formation and the Appraisal Function in Real
Estate Markets,” Journal of Real Estate Finance and Economics 4, 127-146

Read C. (1988), “Price Strategies for Idiosyncratic Goods – the Case of Housing,”
Journal of the American Real Estate and Urban Economics Association 16, 379-395.

Rothschild M. (1974), “Search for the Lowest Price When the Distribution of Prices Is
Unknown,” Journal of Political Economy 82, 689-711.

Salant S. (1991), “For Sale by Owner: When to Use a Broker and How to Price the
House,” Journal of Real Estate Finance and Economics 4, 157-173.

Silber W. (1991), “Discounts on Restricted Stock: The Impact of Illiquidity on Stock

Prices,” Financial Analysts Journal, 60-64

Stigler G. (1961), “The Economics of Information,” Journal of Political Economy 69,

213-225

Stapleton R. and Subrahmanyam M. (1979), “Marketability of Assets and the Price of

Risk,” Journal of Financial and Quantitative Analysis, 1-10

Subramanian A. and Jarrow R. (2001), “The Liquidity Discount,” Mathematical Finance,

447-474

Tosh D. (1992), Handbook of Real Estate Terms, Prentice-Hall.

Yavas A. (1992), “A Simple Search and Bargaining Model of Real Estate Markets,”
Journal of the American Real Estate and Urban Economics Association 20, 533-548.

Yinger J. (1981), “A Search Model of Real Estate Broker Behavior,” American Economic
Review 71, 591-605
 35

CHAPTER 3

Liquidity Bias in the Real Estate Market

3.1. Introduction

In this chapter, we mainly focus on how the uncertainty of the marketing period,

together with the length of the holding period, affects real estate pricing. There is a

substantial real estate literature on time-on-market (TOM), such as Ong and Koh (2000),

Forgey, Rutherford and Springer (1996), Asabare and Huffman (1993), Miller (1978),

Trippi (1977), Belkin, Hempel and McLeavey (1976) and Cubbin (1974). 7 Most of these

studies examined the relationship between TOM and transaction price. Miller (1978),

Trippi (1977), Forgey, Rutherford and Springer (1996), and Asabare and Huffman (1993)

find a positive relationship between time on market and transaction price; i.e. a longer

marketing period results in a higher probability that a relatively higher transaction price

can be obtained. The relationship between marketing period and transaction price is

important in understanding the real estate market, however, the findings in these studies

do not capture the whole picture of the impact of marketing period risk on real estate

pricing. The following example will illustrate this.

In our previous example, we assume a seller holds his real estate for TH period

~
and then places it on the market for sale. Suppose the seller receives return R~t at the time

of sale TH + ~ ~ ~
t ( t = 1,2,... ), where t is the marketing period. To the seller at time TH ,

there are countless possibilities of time when his asset could be successfully traded: his

7
The real estate literature seems to prefer to use the term “time on market” as opposed to “execution
delay” in the finance literature.
 36

~ ~
asset might be sold immediately at TH with return R0 , at TH + 1 with return R1 , at

~
TH + 2 with return R2 , and so on and so forth.

The real estate literature on TOM studied the relationship between the transaction

price at time TH ,i + TOM i and TOM i , where TH ,i is the time when the property i was

placed on the market and TOM i is its marketing period. These studies essentially capture

the relationship between transaction price and the marketing period (the ex-post

measure), but not the relationship between the uncertainty of the marketing period ~
t and

the price at the beginning of the marketing period TH (the ex-ante measure).

As discussed in Chapter 2, our ex-ante measure to evaluate real estate return and

risk is:
~ ~
E ex − ante [ RTH + ~t ] = E~ [ E~ [ RTH + ~t ~
t ]] (3.1)
t R

~ ~ ~
Var ex − ante [ RTH + ~t ] = E~ [ E~ [ RTH + ~t − E ex − ante [ RTH + ~t ]]2 ~
t ]] (3.2)
t R

There are two points worth noting in the ex-ante approach. First, consider the

special case when ~

t = 0 , i.e., the asset can be sold immediately when it is placed on the

market, equations (3.1) and (3.2) then become

~ ~ ~
E ex − ante [ R~t +TH ~
t = 0] = E~ [ RTH ] = E ex − post [ RTH ] (3.3)
R

~ ~ ~ ~ ~
Var ex − ante [ R~t +TH t = 0] = E~ [ RTH − E~[ RTH ]]2 = Var ex − post [ RTH ] (3.4)
R R

In this case, equations (3.1) and (3.2) actually become the ex-post approach to

evaluate real estate return and risk. Therefore, the ex-post approach is a special case of
 37

the ex-ante approach. Second, the ex-ante measure in equations (3.1) and (3.2) account

for both price risk and marketing period risk by incorporating two stochastic processes,

~ ~
t and R~t +TH , into the valuation. The resulting measure is a superior valuation technique

for the real estate market.

The major findings of this chapter can be summarized as follows: First, we

demonstrate that the estimation formula of return and risk based solely on historical

prices and the time of sale is incorrect in the real estate market. Second, the uncertainty

of the marketing period increases real estate risk. When this uncertainty is high, the

increase might be triple or higher than that of current estimations. Third, real estate risk is

closely related to investors’ time horizon. In particular, we find that real estate risk

decreases when the holding period increases. This result is consistent with the

conventional wisdom that real estate is more favorable to long-term investors than to

short-term investors.

The remainder of this chapter is structured as follows: Section 3.2 describes the

model. Section 3.3 studies how real estate illiquidity and holding period affect real estate

return and risk. Section 3.4 discusses empirical applications, and Section 3.5 concludes

with remarks.

3.2. The Model

We consider a discrete model where a seller purchases a real estate asset at time 0,

holds it for TH period, and then places it on the market for sale.8 If the asset is

successfully sold at marketing period t ( t = 0, 1, 2,... ), the seller receives ex-post return

8
For mathematical simplicity, we consider a discrete-time model. However, the findings in this chapter can
be readily extended to a continuous-time model.
 38

~
RTH +t . The probability of sale in marketing period t is denoted by pt ( t = 0, 1, 2, 3,... ),

∞
which characterizes real estate illiquidity and satisfies, ∑p
t =0
t = 1.

Given the discussion in Chapter 2, the main characteristics associated with real

estate illiquidity are the time required for a sale and the uncertainty of the marketing

period. In this model, they can be expressed as follows,

∞
t ] = ∑ tp t
E[~ (3.5)
t =0

∞
t ) = ∑ [t − E[t ]]2 pt
Var (~ (3.6)9
t =0

To better understand how the distribution of a possible sale over marketing period

affects the expected marketing period and its volatility, we consider three special cases.

Case 3.2.1. Constant Conditional Probability of Sale

In this case, the conditional probability (or hazard rate) of selling a similar

property in each market period is constant. Suppose the hazard rate is λ ( λ ≤ 1 ), then the

probability of selling the property in each marketing period t ( t = 0, 1 , 2, ...) is a

geometric distribution,

pt = λ (1 − λ ) t

Since Var ( t ) = ∑ [t − E[t ]] 2 p t = Var ( t + T H ) , and t + T H is a time of sale, we use the following two
9 ~ ~ ~
t =0

terms interchangeably: the uncertainty of marketing period and the uncertainty of time of sale.
 39

Inserting the equation above into equations (3.5) and (3.6), we can obtain,

1
E[~
t ] = −1 (3.7)
λ

1
Var (~
t ) = 2 (1 − λ ) (3.8)
λ

Therefore, a lower hazard rate (i.e., a less liquid market) indicates a higher

expected marketing period ( E[~

t ] ∂λ < 0 ) and higher uncertainty of time of sale

( ∂Var (~ ~
t ) ∂λ < 0 ). When λ = 1 (i.e., a perfectly liquid asset), both E[ t ] and

~
Var ( t ) become zero.

Case 3.2.2. Uniform Probability of Sale

Suppose the probability of sale is constant across all marketing periods. Assume

that N − 1 is the maximum marketing period, then the probability of sale in each

marketing period t ( t = 0, 1 , 2, ..., N − 1) is,

1
pt =
N

We thus have,

N −1
E[~
t]= (3.9)
2

~ N 2 −1
Var ( t ) = (3.10)
12

Equations (3.9) and (3.10) demonstrate that a larger N (less liquidity) implies a

longer expected marketing period and higher uncertainty of time of sale. In particular, the

expected marketing period and the volatility of sale time will both increase at the same
 40

order with N . In other words, when N doubles, the expected marketing period and the

uncertainty of sale time almost double.

Case 3.2.3. Exponential Distribution of Sale

Suppose real estate sales follow the Poisson distribution, then the marketing

period follows an exponential density function with parameter η ,

1
1 − t
η
pt = e , t≥0
η

We thus have,
~
E[t ] = η (3.11)

Var (~
t ) =η2 (3.12)

Therefore, the parameter η represents the expected marketing period.

~
In terms of the ex-post return RTH +t , we assume that it is distributed with mean (TH + t )u

and variance (TH + t )σ 2 for two reasons.

First, empirical results show that the ex-post return increases with holding period.

Figure 3.1 shows the total average return for holding single-family homes in the United

States from 1994Q1 up to 2003Q4 10. From this figure, we can readily see that the ex-post

return in the US residential market increased with holding period.

Second, ex-post risk should increase over time. Case and Shiller (1987) note that

the risk associated with a specific property should be positively related to the length of

time elapsed between transactions of that property. One explanation for this is that both

10
Data source: Office of Federal Housing Enterprise Oversight (OFHEO).
 41

buyers and the seller tend to have better information regarding the true market value of a

given property if the time after the last transaction of that property is relatively small. On

the other hand, after a relatively long holding period, both buyers and the seller tend to

have weaker information regarding the true market value of the property, and thus are

more likely to agree on a price that differs substantially from the market value. Therefore,

the ex-post risk should increase as time passes between transactions.

Figure. 3.1

~
Given the distribution of the ex-post return RTH +t and the probability of sale pt

( t = 0, 1, 2, 3,... ), we next study how real estate illiquidity and holding period affect real

estate pricing.
 42

3.3. Liquidity Bias, Real Estate Illiquidity and Holding Period

First, we consider a simple case in which the holding period TH = 1 . Hence, the

~
ex-post return R1+t is distributed with mean (1 + t)u and variance (1 + t )σ 2 . We can thus

simplify the ex-ante measure in equations (3.1) and (3.2) as follows:

~ ~
E ex − ante [ R1+ ~t ] = (1 + E[ t ])u (3.13)

∞
~ ~ ~
Var ex − ante [ R1+ ~t ] = ∑ [ E[ R12+t ] − 2 E[ R1+ t ](1 + E[~
t ])u + (1 + E[~
t ]) 2 u 2 ] pt (3.14)
t =0

~ ~ ~ ~
Since E[ R12+t ] = Var ( R12+t ) + ( E[ R1+t ]) 2 , Var[ R12+ t ] = (1 + t )σ 2 and

~
E[ R1+t ] = (1 + t )u , we can further simplify equation (3.14) as,

∞
~
Var ex − ante
[ R1+ ~t ] = ∑ [((1 + t )σ 2 + (t − E[~
t ]) 2 u 2 ] pt (3.15)
t =0

Therefore, we have,
~
Var ex − ante [ R1+ ~t ] = (1 + E[~
t ])σ 2 + Var (~
t )u 2 (3.16)

Suppose the distribution of the marketing period is a constant conditional

probability as in case 3.2.1. Inserting equations (3.7) and (3.8) into (3.13) and (3.16), we

can have,

~ 1
E ex − ante [ R1+ ~t ] = u
λ (3.17)
~ 1 (1 − λ ) 2
Var ex − ante
[ R1+ ~t ] = σ 2 + u
λ λ 2
 43

Note that the seller’s expected holding period until sale is 1 λ .11 Hence, the seller’s

annualized ex-ante return ( u ex − ante ) and variance ( (σ ex− ante ) 2 ) are

u ex − ante = u
1 (3.18)
σ ex − ante = σ 2 + (1 − λ )u 2
λ

Without considering the marketing period risk, the current approach would

suggest that the annualized return and variance should be u and σ 2 . Thus, the current

1
approach underestimates real estate risk by (1 − λ )u 2 .
λ

From this simple example, we demonstrate that the assumption of the current

approach where the uncertainty of the marketing period can be ignored results in the

underestimation of real estate risk. We next examine how the uncertainty of the

marketing period together with the holding period affects real estate return and risk in a

more general setting.

Intuitively, not only does real estate illiquidity affect pricing, holding period

should be also intimately related to real estate pricing. Conventional wisdom suggests

that real estate is more favorable to long-term investors than to short-term investors.

However, to our knowledge, beyond appealing to the notion of higher transaction costs,

not much work has been done to explain why real estate risk will be mitigated for

11
TH ~
The seller’s holding period is periods, and his expected marketing period is E[t ] . Hence, the total
~ ~ 1
expected holding period until sale is ( TH + E[t ] ). In this case, TH = 1 and E[ t ] = − 1 (see equation
λ
~ 1
(3.7)), hence TH + E[ t ] = .
λ
 44

relatively long-term investors. Amihud and Mendelson (1980) provide a theoretical basis

for the argument that financial assets with higher illiquidity risk are allocated to

portfolios with longer expected holding periods, and vice versa. Atkins and Dyl (1997)

and Collett, Lizieri and Ward (2003) found empirical evidence supporting this

proposition. One purpose of this chapter is to seek the rationale why real estate is more

favorable to long-term investors.

In this model, we assume that the seller’s holding period TH is exogenously

given. In reality, the holding period may not be fully controlled by the seller, especially

during an unexpected liquidity shock. Our intention for this assumption is to distinguish

short-term investors from long-term investors and to study how investment time horizon

affects real estate pricing. Doubtlessly, an investor with a large TH is likely to be a long-

term investor.
~
Given the assumption that the ex-post return RTH +t is distributed with mean

(TH + t )u and variance (TH + t )σ 2 , we have,

~
E[ RTH +t ] = (TH + t )u (3.19)

~
Var ( RTH +t ) = (TH + t )σ 2 (3.20)

Inserting equations (3.19) and (3.20) into (3.1) and (3.2), we can obtain Theorem
3.1 as follows. 12

12
Proof can be found in the appendix.
 45

Theorem 3.1: Suppose an investor purchases a real estate asset at time 0 and

markets it in period TH . Assume the probability of being sold at marketing period

∞
~
t = 0,1,2,..., ∞) is p ~t ( ∑ p ~t = 1 ) with the ex-post return RTH + ~t , which is
~
t (~
~
t =0

distributed with mean (TH + ~ ~

t )u and variance (TH + t )σ 2 . Then

(1) The ex-ante expected return is the same as the expected ex-post
return: u ex − ante = u ;
(2) The ex-ante variance is higher than the ex-post variance, specifically,

Var (t ) 2
(σ ex − ante ) 2 = σ 2 + u (3.21)
TH + E[t ]

Four conclusions can be drawn from Theorem 3.1. First, based on our new ex-

ante approach, real estate risk can be decomposed into two components: price risk and

liquidity risk. If we confine real estate risk to the price risk from the current approach, we

always understate real estate risk. This underestimation becomes more serious when the

uncertainty of marketing period Var (~

t ) is relatively large and /or when the expected

annualized ex-post return u is high. As a result, using risk estimated from historical

prices and the time of transactions, which have nothing to do with the uncertainty of

marketing period, will underestimate real estate risk, and thus lead to liquidity bias.

Second, we implicitly assume that the seller faces either no liquidity shock or no

borrowing constraints; hence he can always wait for the “best” buyer. Huang (2003)

considers an investor who holds an illiquid asset having to liquidate his asset immediately

by a discount price when a liquidity shock occurs. If this is the case, a higher ex-ante risk

and lower ex-ante return may lead to an even higher liquidity bias.
 46

Third, ex-ante return is the same as the return from the current approach.

However, this result holds only when the ex-post return series increase linearly over

holding period until sale, as we have assumed. If the market faces a downturn, this

assumption is likely to be violated and the annualized ex-post return should decrease over

time; hence the ex-ante return will be less than the return estimated from the current

approach.

Finally, a longer holding period indicates lower liquidity risk, hence lower real

estate risk and lower liquidity bias, ceteris paribus. This result provides a rationale for our

common perception that real estate is more favorable to long-term investors than to short-

term investors.

Table 3.1 summarizes how the distributions of marketing period affect real estate

risk. For example, when the marketing period follows a constant conditional probability

of sale, the higher the conditional probability, the less the liquidity risk, and therefore the

less the real estate risk. If the marketing period is distributed at an exponential
 47

distribution, then the longer the expected marketing period (notice that E[~
t ] = η ), the

higher the real estate risk, given other things equal.

3.4. Empirical Applications

Theorem 3.1 demonstrates that the current approach of using ex-post variance to

measure real estate risk has a bias problem. The ex-post variance always underestimates

real estate risk. In the meantime, Theorem 3.1 also provides a formula to correct for this

bias, i.e.,

Var (t ) 2
(σ ex − ante ) 2 = σ 2 + u (3.22)
TH + E[t ]

Equation (3.22) tells us that, in order to correct for the bias, besides ex-post return,

ex-post risk and holding period, we need to know the distribution of the marketing period.

Bond and Hwang (2004) investigate a number of assumptions about the distribution of

times to sale, such as the normal, chi-square, gamma and Weibull distributions and find

that the exponential density function explains the data better than the others.

By assuming the marketing period is distributed at the exponential distribution,

we next look at how marketing period risk together with investment time horizon affect

real estate risk in both residential and commercial property markets.

Case I: the US residential property market

We first consider the US residential property market. We choose annual return of

5.7% and standard deviation of 1.7%, which are based on the OFHEO’s US home price

index during the period 1996Q1 to 2003Q3.

 48

Table 3.2 illustrates by how much the ex-ante variance exceeds the ex-post

variance under various scenarios of expected marketing period and holding period. From

this table, we can readily see that, first, if the expected time-on-market is zero, the real

estate becomes a liquid asset. Hence, real estate risk is completely from price risk and the

ex-ante variance is the same as the ex-post variance. Second, the degree of

underestimation of using the ex-post variance for real estate risk increases with the

expected marketing period and decreases with the holding period. Third, if the expected

time-on-market is eight months, the ex-ante variance will be three times higher than the

ex-post variance if the holding period is only one year and 47% higher if the holding

period becomes 10 years.

Table 3.2.
Liquidity Bias between Ex-ante Approach and Ex-post Approach
The case of the US residential property market

The table above shows by how much the variance of the returns after taking the uncertainty of marketing
period into account is greater than that given by the current estimation. For example, if an investor holds a
property for 10 years and the expected marketing period is 8 months, the risk faced by the investor is 47%
higher than that given by the current estimation. (Note: the ex-post return and the ex-post risk used in this
table are from OFHEO).
 49

Therefore, the current approach of using the ex-post variance for the ex-ante

variance can seriously underestimate real estate risk, especially when the expected

marketing period is high and the holding period is relatively short.

Case II: the UK commercial property market

Bond and Hwang (2004) are the first to apply this model to the UK commercial

property market. Their findings are surprisingly similar to ours (see Table 3.3). For

example, they find that an investor with a holding period of 10 years and for a property

with an average marketing period of eight months, the total risk faced by the investor is

48% higher than that given by the current approach; while in the case of the US

residential property market discussed earlier, the total risk is 47% higher. If the holding

Table 3.3.
Liquidity Bias between the Ex-ante Approach and Ex-post Approach
The case of UK commercial property market

The table above shows by how much the variance of the returns after taking the uncertainty of marketing
period into account is greater than that given by the current estimation. For example, if an investor holds a
property for 10 years and the expected marketing period is 10 months, the risk faced by the investor is 74%
higher than that given by the current estimation. This table is from Shawn and Hwang (2004).
 50

period is 1 year, the total risk becomes 307% higher than that of the current estimation,

while the total risk in the US residential property market is 300% higher.

A general view in the academic community is that the liquidity risk is trivial when

investment time horizon is large. Hence, little has been studied on the size of the risk

associated with marketing period and its uncertainty. However, the results above strongly

suggest that the liquidity risk can be substantial even when the investment time horizon is

large. Therefore, the uncertainty of the marketing period should be certainly priced in the

real estate market.

3.5. Conclusions

In this chapter, we have argued that the risk of real estate assets comes not only

from price risk, but also from marketing period risk. Incorporating these two risks in the

valuation of real estate return and risk, we have shown that the estimation formula for

return and volatility (the ex-post measure) in the financial market is not applicable to the

real estate market. In addition, we have demonstrated that the uncertainty of the

marketing period increases real estate risk. When this uncertainty is high, the increase

might be triple or higher than that of current estimations. Furthermore, we have

demonstrated that the longer the holding period, the lower the real estate risk. This result

is consistent with the common perception that real estate is more favorable to longer-term

investors.

The theoretical model explored in this chapter focuses primarily on the ex-ante

and ex-post measure of an individual property. However, under normal conditions, these

results also hold when the concept is extended to a portfolio of several properties.
 51

Challenges still remain. For example, this model treats the ex-post return and

volatility as both exogenously given and hence independent of the probability of sale. In

a more general equilibrium framework, the ex-post return and volatility should be state-

dependent (Clarke R. and Silva H. (1998)) and so should the probability of sale (Krainer

(2001)). Anglin (2003) discusses how both sale price and probability of sale are

correlated and vary over market conditions. Moreover, we assume that the ex-post return

linearly increases over holding time. This assumption is likely to be violated, especially

when a market faces a downturn. We hope to address these challenges in our future

research.
 52

Appendix. Proof of Theorem

This appendix contains proof for Theorems 3.1.

Proof of Theorem 3.1

By the definitions of ex-ante expected return and variance (i.e., equations (3.1)
and (3.2)), and equations (3.19) and (3.20), we have,

∞ ∞
~ ~
E ex − ante [ R~t +TH ] = ∑ E[ Rt +TH ] pt =u ∑ (t + TH ) pt
t =0 t =0 (A3.1)
~
= u (TH + E[ t ])
∞
~ ~ ~
Var ex − ante ( R~t +TH ) = ∑ E[ R~t +TH − E ex − ante [ R~t +TH ]]2 pt
t =0

∞
~ 2 ~
= ∑ E[Rt +TH ] pt − E ex − ante [ R~t +TH ] 2 (A3.2)
t =0

∞
~
= ∑ ([t + TH ]σ 2 + [t + TH ] 2 u 2 ) pt − E ex − ante [ R~t +TH ] 2
t =0

Inserting equation (A3.1) into equation (A3.2) yields,

~ ~ ~
Var ex − ante ( R~t +TH ) = (TH + E[ t ])σ 2 + Var ( t )u 2 (A3.3)

Hence, the annualized ex-ante return and variance are,

u ex − ante = u

Var (~
t) 2 (A3.4)
(σ ex − ante ) 2 = σ 2 + u
TH + E[~t]

Q.E.D.
 53

References:

Amihud Y. and Menselson H. (1980), “Dealership Market: Market-Making with

Inventory,” Journal of Financial Economics, 8, 31-53.

Anglin P. (2003), “The Value and Liquidity Effects of A Change in Market Conditions,”
Working papers, University of Windsor.

Asabare P. and Huffman F. (1993), “Price Concessions, Time on Market, and the Actual
Sales Price of Homes,” Journal of Real Estate Finance and Economics 6, 167-174.

Atkins A. and Dyl E. (1997), “Transactions Costs and Holding Period for Common
Stocks,” Journal of Finance, 309-325

Belkin J., Hempel D. and McLeavey D. (1976), “An Empirical Study of Time on Market
Using Multidimensional Segmentation of Housing Markets, “ Journal of the American
Real Estate and Urban Economics Association 4, 57-75.

Bond S. and Hwang S. (2004), “Liquidity Risk and Real Estate: A Quantitative Approach
to Assessing Risk”, Working paper, University of Cambridge.

Case K. and Shiller R. (1987), “Prices of Single-Family Homes Since 1970: New Indexes
For Four Cities,” New England Economic Review, 45-56.

Clarke R. and Silva H. (1998), “State-Dependent Asset Allocation”, Journal of Portfolio

Management, 57-64

Collett D., Lizieri C. and Ward C. (2003), “Timing and the Holding Period of
Institutional Real Estate,” Real Estate Economics, 2003, 205-222

Cubbin J. (1974), “Price, Quality, and Selling Time in the Housing Market,” Applied
Economics 6, 171-187.

Forgey F., Rutherford R. and Springer T. (1996), “Search and Liquidity in Single-Family
Housing,” Real Estate Economics 24, 273-392

Huang M. (2003), “Liquidity Shocks and Equilibrium Liquidity Premia,” Journal of

Economic Theory, 105-129

Krainer J. (2001), “A Theory of Liquidity in Residential Real Estate Markets,” Journal of

Urban Economics, 32-53

Miller N. (1978), “Time on the Market and Selling Price,” AREUEA Journal 6, 164-174.

Ong S. and Koh Y. (2000), “Time On-market and Price Trade-offs in High-rise Housing
Sub-markets,” Urban Studies, 2057-2071
 54

Trippi R. (1977), “Estimating the Relationship between Price and Time of Sale in
Investment Property,” Management Science, 23, 838-842
 55

CHAPTER 4

Valuation Bias in the Real Estate Market

4.1. Introduction

In Chapter 3, we studied how the uncertainty of the marketing period contributes

an additional risk to a real estate investor. In this chapter, we will study the relationship

among valuation bias, the uncertainty of the marketing period and the holding period.

As we know, in order to receive a transaction price, a real estate seller has to face

the uncertainty of marketing period. However, to the investors who have immediate cash

needs and are forced to sell their real estate assets immediately, they will receive the

prices from the distribution of bidding prices instead of the prices from the distribution of

transaction prices. Therefore, a valuation bias will arise if their return and volatility are

directly estimated from transaction prices.

Given the earlier discussion, a transaction price can be observed if and only if a

bidding price equals or exceeds the reservation price:

⎧⎪ P if P ≥ P * ,
P =⎨
T
(4.1)
⎪⎩unobserved if P < P * .

Where:

P = V0 + ε , ε ~ N (0, σ ε2 ) (4.2)

Equation (4.1) can be rewritten as,

P T = V0 + [ε ε ≥ P * − V0 ] (4.3)
 56

Similar properties with different sellers may transact very differently. For a

variety of reasons, some sellers may have to lower their reservation prices in order to sell

their houses more quickly than others. Equation (4.3) implies that the distribution of

transaction prices varies over reservation prices. Therefore, strictly speaking the return

and risk in the real estate market are “seller-specific”.

As in Goetzmann and Peng (2003), Munneke and Slade (2000), Englund, Quigley

and Redfearn (1999), and Gatzlaff and Haruin (1998), we assume a potential buyer offers

a bidding price based on market valuation. Hence, a bidding price is the price from the

distribution of market valuation, which is not “seller-specific”.

As we know, the only prices can be observed in the real estate market are

transaction prices and biding prices below seller’s reservation price cannot be observed.

As a result, market valuation cannot be fully observed in the real estate market. How can

we obtain unobservable market valuation from the market?

With this question, we further rewrite equation (4.3) as follows,

P T − V0 = ε ε ≥ P * − V0 (4.4)

The left-hand term in equation (4.4) can be regarded as the valuation bias between

market values and transaction prices.

Since the reservation price P * is closely related to marketing period (TOM), the

higher the reservation price, the longer the expected marketing period, we can thus

rewritten (4.4) as:

P T − V0 = [ε ε ≥ f (TOM )] (4.5)

Holding other factors constant, intuition suggests a higher reservation price not

only displays a larger valuation bias between the observable transaction price and the
 57

market valuation, but also has a longer marketing period. Therefore, we should expect

that a higher marketing period should have a higher valuation bias.

Normally, we cannot observe the reservation price; however, information on

marketing period is readily available. Equation (4.5) suggests that we may use the

information of transaction prices and time-on-market to estimate market return and

volatility. One objective of this chapter is to seek this relationship.

The remainder of this chapter is organized as follows. Section 4.2 specifies the

model. Section 4.3 examines the relationships among reservation price, probability of

sale and expected marketing period. Section 4.4 studies the relationship among valuation

bias, expected marketing period and holding period. Section 4.5 discusses how the risk

premium relates with the expected marketing period. Section 4.6 provides empirical

applications. Section 4.7 makes concluding remarks.

4.2. The Model

Due to the heterogeneous nature of real estate assets and their geographical

specificity, the transaction process in the real estate market is often characterized by high

search costs and infrequent trading. Buyers arrive randomly and a transaction occurs only

when a potential buyer is present and his or her bidding price is high enough for the seller

to accept.

The typical assumption of buyers’ stochastic arrival is the Poisson process with

constant arrival rate in each point in time. Following Gatzlaff and Haurin (1998), Salant

(1991), Haurin (1988), and Lippman and McCall (1986), we assume that a potential seller
 58

receives bid prices from potential buyers at a rate of one per period (units of time can be

made arbitrarily small). 13

Regarding the distribution of bid prices, Arnold (1999) and Sirmans, Turnbull and

Dombrow (1995) assume that the bid distribution is over [ p, p ] with density

function f ( P bid ) , where p ( p ) is the minimum (maximum) bid price. 14 The assumption

of a time-invariant distribution of offer prices may be too simple, and cannot be justified

for two reasons. First, the underlying value of a specific property normally increases over

time. According to OFHEO’s repeat-transactions home price index (estimated using data

from Fannie Mae and Freddie Mac), home prices rose by an annual average of 7.75

percent over the past three years. Hence, underlying home values should also increase

over time. Second, as discussed earlier, after a relatively long holding period, the buyers

tend to have weaker information regarding the true market value of the property, and thus

are more likely to agree on a price that differs substantially from the market value.

Therefore, the underlying risk should increase as time passes between transactions (Case

and Shiller 1987). Taking these two facts into account, we assume that the distribution of

bidding prices vary over time. In particular, we assume that it is distributed over

[τ p + P0 ,τ p + P0 ] at time τ , where P0 is the original price purchased at time 0.

Yavas (1992) and Read (1988) assume the density function f ( p ) is uniformly

distributed. For technical simplicity, we adopt the same assumption here. Hence, buyer’s

at time τ is distributed as,

bid
bid price Pτ

13
The assumption is made for technical simplicity. Anglin (2003), Arnold (1999), Glower, Haruin and
Hendershott (1998) and Miceli (1989) assume that an arrival rate is λ per period. A more complicated
model would allow buyers to respond to sellers’ asking price, i.e., a higher asking price implies a lower
arrival rate, but that is beyond the scope of this analysis.
14
p can be regarded as the seller’s asking price.
 59

⎧ 1
⎪ , P bid ∈ [τ p + P0 ,τ p + P0 ]
f ( Pτ ) = ⎨ ( p − p )τ
bid
(4.6)
⎪0,
⎩ otherwise

The seller decides whether to accept an offer or not based on the reservation price

for the property. 15 An each point in time, the optimal marketing strategy for the seller is

to accept the first bid above the reservation price, and to reject all bids below. Like the

buyers’ bid price, we allow a time-varying reservation price and denote this as pτ* .

Therefore, in any period, there exist two possibilities. First, a transaction occurs when a

buyer arrives with a bid price Pτbid ≥ pτ* and the seller accepts that offer and sells the

property for that price. Second, no transaction occurs if an offer price Pτbid < pτ* . If there

is no transaction, no deal is reached and the seller will continue to search for the next

buyer. Thus, real estate observable transaction prices (denoted by PτT ) must be in the

range of [ pτ* , pτ + P0 ] , and in order to trade at an observable transaction price, the real

estate seller experiences the uncertainty of sale in each period. 16

Holding other things equal, the higher the underlying market value of a property,

the higher the seller’s reservation price. Accordingly, we assume that the seller’s

reservation price also increases with time. For analytical tractability and without loss of

15
To solve reservation prices is not the focus of this analysis. We treat it as given. Lippman and McCall
(1986, 1976 a) and DeGroot (1970) have some discussions on this.
16
In the real estate market, most sellers sell their assets by exercising the optimal stopping rule: to accept
all bids above the reservation price and to reject all bids below. Henceforth we assume observable
transaction prices are the prices that are at least as high as sellers’ reservation prices.
 60

generality, we further assume that, for a particular seller, at time τ his reservation price

pτ* is p *τ + P0 .

4.3. Probability of Sale and Expected Marketing Period

Given the discussion above, in any period there are two possibilities: a transaction

or no transaction. If a buyer’s offer is too low for the seller to accept, the seller has to

wait and continue to search for the next buyer. In this model, the probability of having a

p − p*
transaction in each period given the property is not sold can be obtained as .
p− p

~
Hence, the probability of sale at marketing period t ( t = 0, 1, 2, 3... ) (denoted by TM ) is a

geometric distribution,

~
Prob( TM = t ) = π (1 − π ) t (4.7)

Where,

p − p*
π= (4.8)
p− p

And π is the hazard rate of sale.

Equations (4.7) and (4.8) highlight two interesting facts. First, the probability of a

successful sale in each period not only depends on the seller’s reservation price ( p * ), but

also depends on the dispersion of buyers’ valuation, ( p − p) , and a lower reservation

price indicates a higher probability of sale. Therefore, the reservation price plays an

important role in the determination of matching probability between sellers and buyers.
 61

Second, sellers cannot sell their asset at a predetermined time with certainty. Since
~
p * = p rarely occurs in the real estate market, Prob(TM = t ) < 1 holds for all t . In other

words, the time required for sale in the real estate market is a random variable (e.g. Trippi

1977).

Expected Marketing Period

Having derived the probability of a successful sale in each period, we next study

how long the seller is expected to wait on the market. By definition, expected marketing

period can be expressed as follows,

∞
~
E[TM ] = ∑ tπ (1 − π ) i (4.9)
t =0

Simplifying (4.4) yields,

~ 1
E[TM ] = − 1 (4.10)
π

Equation (4.10) indicates that the probability of sale is uniquely related to the expected

marketing period through the hazard rate π .

Taking Equations (4.10) and (4.8) together, we can conclude that real estate

sellers sell their assets immediately only when they accept whatever price the buyer can
~
offer ( p * = p , hence π = 1 and E[TM ] = 0 ).

In the next section, we focus on the relationship among valuation bias, the

uncertainty of the marketing period and the holding period.

 62

4.4. Valuation Bias. Real Estate Illiquidity and Holding Period

Suppose rM and σ M are the period return and volatility from market valuation

and rT and σ T are the period return and volatility based on the transaction prices. We

formally define valuation bias as,

return bias = rT − rM (4.11)

volatility bias = σ M − σ T (4.12)

As before, we assume that an investor purchases a real estate asset at time 0 with

a price P0 , holds it for TH periods, and then put it on the market for sale. In each

marketing period t (t = 0, 1, 2, ...) , conditional upon a transaction being observed, the total

return from this transaction can be estimated as,

~T PTTH +t − P0
RTH +t = (4.13)
P0

and,

⎧⎪ PTbid
H +t
, if PTbid
H +t
≥ p * (TH + t ) + P0
P T
T H +t =⎨ (4.14)
⎪⎩unobserved, if PTbid
H +t
< p * (TH + t ) + P0

Hence, PTTH +t is uniformly distributed over [(TH + t ) p * + P 0 , (TH + t ) p + P0 ] . As a result,

the average period return from this transaction is,

~
~ PT
rTH +t =
T
(4.15)
P0

~
Here, P T is uniformly distributed over [ p * , p]
 63

Similarly, we can obtain the average period return from the underlying market

valuation as follows,
~
~ PM
rTH +t =
M
(4.16)
P0

~
Where: P M is uniformly distributed over [ p, p ] . Unfortunately, we can only observe

~ ~
market values when P M ≥ p * and cannot observe them when P M ∈ [ p, p * ) . In other

words, we cannot estimate market return and risk solely based on the transaction prices.

However, the relationship between the market valuation and transaction prices can be

established by the use of marketing period information. We summarize this finding in the

following theorem. 17

Theorem 4.1. Market Valuation, Transaction Prices and Marketing Periods

~
1. rT − rM = 3E[TM ]σ T (4.17)
~
2. σ M − σ T = E[TM ]σ T (4.18)
~
Where: E[TM ] is the expected marketing period.

The left-hand terms in equations (4.17) and (4.18) are actually the return bias and

the risk bias, respectively. Hence, Theorem 4.1 illustrates the relationship between

valuation bias and the expected marketing period. Through this relationship, we can use

the information of transaction return ( rT ), transaction volatility ( σ T ) and the expected

~
marketing period ( E[TM ] ) to impute the underlying market return ( rM ) and volatility

( σ M ) as follows,

17
Proof can be found in the appendix.
 64

~
rM = rT − 3E[TM ]σ T (4.19)

~
σ M = (1 + E[TM ])σ T (4.20)

Therefore, we can estimate the unobserved market valuation by using both observable

transaction prices and marketing period (see Figure 4.1).

Valuation Bias

Transaction Prices ( rT , σ T )
~ ~~
(Pt transaction = Pt Pt > MAPt )
~
t ≈0 Market Valuation ( rM , σ M )
(at the absence of illiquidity)

Marketing Period Risk

E[~
t ]>0
~
Var(t ) > 0
Observable

(Theorem 4.1)
Observable Unobservable

Figure 4.1

Several conclusions can be drawn from Theorem 4.1. First, valuation bias

disappears when assets can be traded at observable transaction prices with immediate
 65

~
execution (i.e., E[TM ] = 0 ). In other words, there is no valuation bias in the financial

market when using observable transaction prices for market valuation.

Second, unlike liquidity bias, the holding period plays no role in valuation bias.

However, the expected marketing period plays an important role in the determination of

valuation bias. In particular, valuation bias increases when the expected marketing period

increases, which is consistent with the common perception that a longer expected

marketing period implies a higher reservation price and hence a larger deviation of

transaction prices from the market valuation.

Third, market return is always lower than the return estimated from the

transaction prices and market volatility is always higher than that estimated from the

transaction prices. Hence, evaluating real estate performance based solely on the

transaction sample is likely to result in an overstated risk-adjusted return.

Fourth, valuation bias increases with the dispersion of transaction return, σ T .

When σ T = 0 , valuation bias disappears. The intuition behind this is obvious. σ T = 0

implies that all possible bid prices collapse to one price in each point in time. Put

differently, the distribution of possible bidding prices, which becomes a single price, is

the same as that of transaction prices, and so no bias results.

Finally, we cannot conclude that real estate submarket A is better than real estate

submarket B just by looking at their return and risk estimated from transaction prices; the

marketing period information should be also considered. For example, suppose the return

and volatility based on the transaction sample are the same in both submarket A and

submarket B (i.e., rTA = rTB and σ TA = σ TB ), and assume the expected marketing period in

~ ~
submarket A is much higher than that in submarket B (i.e., E[TMA ] >> E[TMB ] ). By
 66

Theorem 4.1, we can readily have σ MA >> σ MB and rMA << rMB . Therefore, the real estate

market in submarket B is better than that in submarket A . However, just looking at the

return and volatility based on the transaction prices, we would conclude that they are the

same.

In the financial market, the marketing period is trivial, and consequently, its

market return and volatility are the same as the return and volatility estimated from

transaction prices. In the real estate market, however, the marketing period is substantial,

and the market return and volatility can be very different from the return and volatility

based on the transaction prices. As a result, the notion that real estate has extremely high

risk-adjusted returns could be misleading, especially to the investors who face quick

sales.

4.5. Risk Premium and Marketing Period

Holding everything else constant, sellers with different financial situations may

have very different selling strategies. Sellers who are recently divorced, or face a job loss

likely have to sell their houses more quickly than others. Examples of seller motivations

that have been studied in the literature include: properties owned by relocated sellers

(Turnbull, Sirmans and Benjamin, 1990); vacant properties (e.g. Zuelke, 1987);

foreclosure properties (e.g. Forgey, Rutherford and Vanbuskirk, 1994); and time-

constraints on sale (e.g. Glower, Hendershott, and Haurin 1998).

We next discuss how sellers’ motivations affect their expected return and

volatility. Consider two types of sellers: type A has a liquidity constraint with a shorter

expected marketing period, denoted by t , and type B does not have any financial
 67

distress with a longer expected marketing period, denoted by T ( T > t ). Suppose that the

expected period return and volatility that types A and B receive are denoted by (rA , σ A )

and (rB , σ B ), respectively. We summarize our findings in the following Theorem. 18

Theorem 4.2. Risk premium and marketing period

T t
1. rB − rA = 3 ( − )σ M > 0 (4.21)
1+ T 1+ t
T −t
2. σ A − σ B = σM > 0 (4.22)
(1 + T )(1 + t )

Where: σ M is the volatility from market valuation.

Two points are worth noting. First, type B sellers are expected to experience a

longer time on the market and thus have a higher marketability risk. However, they are

also expected to receive a higher expected return and a lower volatility in compensation.

The institution behind this result is quite clear. Type A sellers who are motivated to sell

quickly have a lower reservation price and accept earlier, and hence receive lower offers.

Those sellers who are not motivated to sell quickly will have a higher reservation price

and will only accept offers that are relatively high, even if this means an extended wait.

This result is also consistent with recent findings by Huang (2003). Huang studies an

equilibrium in which agents face surprise liquidity shocks and invest in liquid and illiquid

assets. He finds that the illiquid asset generates a higher expected return to compensate its

holders for marketability risk.

Second, type A sellers with financial distress have to sell more quickly than type

B sellers. However, they have to give up a higher expected return and a lower volatility.

18
Proof can be found in the appendix.
 68

In other words, frequency of transaction is closely related to the return and volatility of

the transaction properties. The systematic differences in return and volatility among

different sellers in the real estate market suggest that properties exhibiting different

selling motivations, hence different transaction frequencies, constitute a different price

behavior that should be analyzed separately; or if not analyzed separately, the analytical

methodology should take into account the systematic differences in the frequency of

transaction (see “Heterogeneous Sellers” in the appendix).

4.6. Empirical Applications

In this section, we discuss how we can empirically correct for valuation bias. The

required information in our correction method includes transaction prices and the

marketing period of each transaction property, both of which are readily available in the

market.

For simplicity, we assume that sellers in the real estate market are

homogeneous.19 Hence, there is only one uniform reservation price in the market and

every seller accepts an offer only when the offer price is at least as this price.

Suppose the estimated return and volatility from the transaction prices are r̂T and

)
σ T , respectively, and average marketing period is TˆM . From equations (4.19) and (4.20),

we can estimate the unobservable market return and volatility as follows,

rˆM = rˆT − 3TˆM σˆ T (4.23)

σˆ M = (1 + TˆM )σˆ T (4.24)

19
The case of heterogeneous sellers with different reservation prices is discussed in the appendix.
 69

Equations (4.23) and (4.24) indicate that, given the same return and volatility

estimated from transaction prices, the underlying market return and volatility can be quite

different. We next look at how the average marketing period affects valuation bias in

both residential and commercial property markets.

Case I: the US residential property market

As in Chapter 3, we choose annual return of 5.7% and standard deviation of 1.7%

from the OFHEO’s US home price index during the period 1996Q1 to 2003Q3. Table 4.1

illustrates how market return and volatility vary over the average marketing period. As

discussed earlier, market return and volatility are essentially the return and volatility at

the absence of real estate illiquidity, i.e., marketing period ~

t = 0.

Table 4.1.
Valuation Bias: The case of the US residential property market

M a rk e t V a lu a tio n V a lu a tio n B ia s
Tˆ M
(M o n th s ) M a rk e t R e tu rn M a rk e t V o la tility R e tu rn B ia s V o la tility B ia s
0 5 .7 % 1 .7 % 0 .0 % 0 .0 %
2 5 .2 % 2 .0 % 0 .5 % 0 .3 %
4 4 .7 % 2 .3 % 1 .0 % 0 .6 %
6 4 .2 % 2 .6 % 1 .5 % 0 .9 %
8 3 .7 % 2 .8 % 2 .0 % 1 .1 %
10 3 .2 % 3 .1 % 2 .5 % 1 .4 %
12 2 .8 % 3 .4 % 2 .9 % 1 .7 %
Market return and volatility are the return and volatility received by the sellers who sell their real estate
assets immediately. The table above shows how we can estimate market return and volatility from the
return and volatility estimated from a transaction sample, given the expected marketing period. For
example, suppose the transaction return and volatility are 5.7% and 1.7%, respectively (data source:
OFHEO), and assume the expected marketing period for these return and volatility is 8 months, then the
market return and volatility are 3.7% and 2.8%, respectively. The valuation bias is simply the difference
between the market return/volatility and the transaction return/volatility.
 70

From Table 4.1, we can see that a higher average marketing period implies a

lower market return and higher market volatility and hence a higher valuation bias, given

the same transaction return (5.7%) and volatility (1.7%). When the average marketing

period is over six months, the market return is over 26 percent lower than the transaction

return, and the market volatility is more than 53 percent higher than the transaction

volatility. Therefore, in the presence of a high average marketing period, there is a

significant bias problem for the return and volatility estimated from a transaction sample

when an investor faces a quick sale. As we know, in the residential market, the average

marketing period is about eight to ten months when the market is “cold”, and about four

to five months when the market is “hot”.20

Case II: the US commercial property market

The annual return and volatility of the National Council of Real Estate Investment

Fiduciaries (NCREIF) “NCREIF “transaction” index during the 1978-1998 period are

9.2% and 4.3%, respectively.21 We thus choose annual return of 9.2% and standard

deviation of 4.3% as the transaction return and volatility in the commercial market.

Table 4.2 illustrates the relationship between average marketing period and

valuation bias. Similar to Table 4.1, we can see that a higher average marketing period

implies a lower market return and higher market volatility, given the same transaction

return (9.2%) and volatility (4.3%). When the average marketing period is over eight

months, the market return is over 53 percent lower than the transaction return, and the

20
In the past three years, the real estate market in U.S. can be characterized as a ‘hot’ market where
appreciation was high and marketing periods were short (4.5 months). In the early 1990s, however, the
market was depressed and the average marketing period was about 9 months. Krainer (2001) has a good
discussion on “hot” and “cold” real estate markets.
21
See Geltner and Goetzman (2000).
 71

market volatility is more than 65 percent higher than the transaction volatility. Given the

fact that average marketing periods in commercial markets are often longer than those in

residential markets, valuation bias could be a more serious problem when using the return

and volatility estimated from the transaction prices.

Table 4.2.
Valuation Bias: The case of the US commercial property market

Market Valuation Valuation Bias

TˆM
(Months) Market Return Market Volatility Return Bias Volatility Bias
0 9.2% 4.3% 0.0% 0.0%
2 8.0% 5.0% 1.2% 0.7%
4 6.7% 5.7% 2.5% 1.4%
6 5.5% 6.4% 3.7% 2.1%
8 4.3% 7.1% 4.9% 2.8%
10 3.1% 7.8% 6.1% 3.6%
12 1.8% 8.5% 7.4% 4.3%
Market return and volatility are the return and volatility received by the sellers who sell their real estate
assets immediately. The table above shows how we can estimate market return and volatility from the
return and volatility estimated from a transaction sample, given the expected marketing period. For
example, suppose the transaction return and volatility are 9.2% and 4.3%, respectively (data source:
NCREIF), and assume the expected marketing period for these return and volatility is 10 months, then the
market return and volatility are 3.1% and 7.8%, respectively. The valuation bias is simply the difference
between the market return/volatility and the transaction return/volatility.

4.7. Conclusions

This chapter documents a potential bias in estimating return and volatility solely

from the sample of sold properties. We consider a model in which the buyer’s bid price is

determined by the market valuation, and a transaction occurs if and only if the bid is at

least as high as the seller’s reservation price. Our major findings can be summarized as

follows:
 72

First, we formally show that a longer expected marketing period implies a larger

valuation bias. Second, we find that, ceteris paribus, a higher expected marketing period

implies a lower risk and a higher expected return. The intuition behind this result can be

illustrated by a seller who is compensated for giving up some degree of marketability.

This finding also suggests that we cannot make a conclusion that real estate submarket A

is “better” than real estate submarket B just by looking at their return and risk, which are

estimated solely from a transaction sample; we also need to consider their marketing

period information. Third, we obtain a “closed form” relationship between the expected

marketing period and the valuation bias. Based on this relation, we propose a correction

method to estimate market return and volatility by using both transaction prices and

widely available marketing period information. Fourth, unlike in liquidity bias, the

holding period plays no role in valuation bias. Finally, by ignoring the marketing period

in the real estate market, the findings that show real estate to have extremely high risk-

adjusted returns could be misleading.

We have to point out that the assumption of the bidding process and arrival rate

being exogenously given may be too simple. Our model assumes the arrival rate is one

per period. Allowing a time-varying arrival rate may be more appropriate because it

could recognize real estate cycles as well as potential buyers responding to seller’s asking

prices. We hope to address these challenges in future research.

 73

Appendix. Proofs of Theorems

This appendix contains proofs for Theorems 4.1 and 4.2 and the discussion of

heterogeneous sellers.

Proof of Theorem 4.1

Based on equations (4.15) and (4.16), the transaction return (annualized) and the

market return (annualized) can be expressed as:

~
~ PT
rTH +t =
T
(A4.1)
P0

~
where, P T is uniformly distributed over [ p * , p]

~
~ PM
rTH +t =
M
(A4.2)
P0

~
where, P M is uniformly distributed over [ p, p ] .

Hence, we have,

~ ( p + p)
P
rM = E[ ] = (A4.3)
P0 2 P0

~
P T ~T p* + p
rT = E[ P ≥ p ]=
*
(A4.4)
P0 2 P0

~ ( p − p) 2
P
σ M2 = Var[ ]= (A4.5)
P0 12 P02
~
PT ~ ( p − p* )2
σ T2 = Var[ P ≥ p* ] = (A4.6)
P0 12 P02

Equations (4.8) and (4.10) yield,

 74

~ p* − p
E[TM ] = (A4.7)
p − p*

Given,

( p + p) ( p + p* ) p* − p ( p − p* )2
= − 3( ) (A4.8)
2 P0 2 P0 p − p* 12 P02

and equations (A4.4) to (A4.6), we have,

~
rM = rT − 3E[TM ]σ T (A4.9)

Similarly, we have,

( p − p) 2 p* − p ( p − p* )2
= [1 + ] (A4.10)
12 P02 p − p* 12 P02

Given equations (A4.5), (A4.6) and (A4.7), we have

~
σ M = (1 + E[TM ])σ T (A4.11)

Q.E.D.

Proof of Theorem 4.2

Given the expected marketing time t for type A and T for type B, and equations

(A4.9) and (A4.11), we have,

rA − rM = 3tσ A (A4.12)

σ M − σ A = tσ A (A4.13)

rB − rM = 3Tσ B (A4.14)
 75

σ M − σ B = Tσ B (A4.15)

Equations (A4.13) and (A4.15) yield,

1
σA = σM (A4.16)
1+ t

1
σB = σM (A4.17)
1+ T

From equations (A4.16) and (A4.17), we have,

T −t
σ A −σ B = σM > 0 (A4.18)
(1 + T )(1 + t )

From equations (A4.12), (A4.14), (A4.16) and (A4.17), we have,

3T
rB = rM + σM (A4.19)
1+ T

3t
rA = rM + σM (A4.20)
1+ t

Therefore,

T t
rB − rA = 3 ( − )σ M > 0 (A4.21)
1+ T 1+ t

Q.E.D.

Heterogeneous Sellers

In reality, different sellers coexist in the real estate market. Sellers with different

financial situations, as discussed earlier, receive different transaction returns and

 76

volatilities. Suppose there are N different types of sellers on the market, the average

marketing period for type i ’s sellers ( i = 1,2,..., N ) is TˆMi . Assume the frequency of type

i ’s sellers is given by g i , and denote the return and volatility for type for type i ’s sellers

based on the transaction sample as r̂Ti and σˆ Ti , respectively. Then, we can estimate the

market return and volatility in two approaches. The first approach is to apply equations

(4.19) and (4.20) to each type of sellers and estimate them separately. The second

approach is to take into account the differences in the frequency of each motivation and

estimate them as follows,

∑ σˆ i
T gi
σˆ M = N
i =1
(A4.22)
1
∑ 1 + Tˆ
i =1
i
gi
M

N N
3TˆMi
uˆ M = ∑ rˆTi g i − σˆ M ∑ g (A4.23)
i =1 i =1 1 + T
ˆi i
M
 77

Reference:

Anglin P. (2003), “The Value and Liquidity Effects of A Change in Market Conditions,”
Working Papers, University of Windsor

Arnold M. (1999), “Search, Bargaining and Optimal Asking Prices,” Real Estate
Economics, 27:453-482

Case B. and Shiller R. (1987), “Prices of Single-Family Homes Since 1970: New Indexes
For Four Cities,” New England Economic Review, 45-56.

DeGroot M. (1970), Optimal Statistical Decisions. New York: McGraw-Hill.

Englund P., Quigley J. and Redfearn C. (1999), “Do Housing Transactions Provide
Misleading Evidence About The Course of Housing Values?” Working Paper.

Forgey F., Rutherford R. and Vanbuskirk M. (1994), “Foreclosure Status and Its Effect
on Residential Selling Price,” Journal of Real Estate Research 9, 313-318.

Gatzlaff D. and Haurin D. (1998), “Sample Selection and Biases in Local House Value
Indices,” Journal of Urban Economics, 199-222

Geltner D. and Goetzmann W. (2000), “Two Decades of Commercial Property Returns:

A Repeated-Measures Regression –Based Version of the NCREIF Index”, Journal of
Real Estate Finance and Economics, 5-21

Glower M., Haurin D. and Hendershott P. (1998), “Selling Price and Selling Time: The
Impact of Seller Motivation,” Real Estate Economics, 719-740

Goetzmann W. and Peng L. (2003), “Estimating Indices in the Presence of Seller

Reservation Prices,” Working Paper, Yale University

Haurin D (1988), “The Duration of Marketing Time of Residential Housing,” AREUEA

Journal 16, 396-410

Haurin D. and Hendershott P. (1991), “House Price Indexes: Issues and Results,”
AREUEA Journal, 259-269

Huang M. (2003), “Liquidity Shocks and Equilibrium Liquidity Premia,” Journal of

Economic Theory, 105-129

Krainer J. (2001), “A Theory of Liquidity in Residential Real Estate Markets,” Journal of

Urban Economics, 32-53

Lippman S. and McCall J (1976a), “The Economics of Job Search: A Survey,” Economic
Inquiry 14, 155-89.
 78

Lippman S. and McCall J. (1986), “An Operational Measure of Liquidity,” American

Economic Review 76, 43-55

Miceli T. (1989), “The Optimal Duration of Real Estate Listing Contracts,” Journal of
the American Real Estate and Urban Economics Association 17, 267-277

Munneke H. and Slade B. (2000), “An Empirical Study of Sample-Selection Bias in

Indices of Commercial Real Estate,” Journal of Real Estate Finance and Economics, 45-
64

Read C. (1988), “Price Strategies for Idiosyncratic Goods – the Case of Housing,”
Journal of the American Real Estate and Urban Economics Association 16, 379-395

Salant S (1991), “For Sale by Owner: When to Use a Broker and How to Price the
House,” Journal of Real Estate Finance and Economics 4, 157-173.

Sirmans C., Turnbull G. and Dombrow J. (1995), “Quick House Sales: Seller Mistake or
Luck?” Journal of Housing Economics, 230-243.

Trippi R. (1977), “Estimating the Relationship between Price and Time of Sale in
Investment Property,” Management Science, 23, 838-842.

Turnbull G., Sirmans C. and Benjamin J. (1990), “Do Corporations Sell Houses for Less?
A Test of Housing Market Efficiency,” Applied Economics, 1389-98

Yavas A. (1992), “A Simple Search and Bargaining Model of Real Estate Markets,”
Journal of the American Real Estate and Urban Economics Association 20, 533-548.

Zuehlke T. (1987), “Duration Dependence in the Housing Market,” Review of Economics

and Statistics 69, 701-704
 79

Chapter 5

Conclusion: The Risk Premium Puzzle in Real Estate Revisited

5. 1. Introduction

Virtually all real estate return and volatility are estimated solely from observed

transaction prices (or appraisal values) and the time of transactions. This dissertation has

demonstrated that this estimation method has two potential problems. The first is

liquidity bias that comes from the fact that transaction prices and the time of sales contain

no information about the uncertainty of marketing period; hence, the marketing period

risk is not priced. The second is valuation bias, which is due to the fact that real estate

seller who faces a quick sale can only receive a price from market valuations rather than a

price from transaction prices, and transaction prices are the prices from a “truncated”

distribution of the market valuations. 22

Numerous studies, based on the return and volatility estimated from transaction

prices, have shown that real estate has historically displayed much higher returns and less

volatility than other asset classes, such as stocks and bonds, and real estate’s Sharpe ratio

is at least as three times as that of stocks and bonds. The premiums in real estate are too

large to be explained by standard finance theories (Shilling (2003)).

In this concluding chapter, we revisit the puzzle. We find that current real estate

valuation techniques, which ignore real estate illiquidity, not only understate real estate

risk but also overstate real estate returns, and thus overestimates real estate’s Sharpe

ratio. Our findings can help us to understand the risk premium puzzle in real estate.

22
Appraisal values are essentially estimated from observable transaction prices, since they are obtained by
appraisers based on transaction prices.
 80

The remainder of this chapter proceeds as follows: Section 5.2 discusses risk

premium puzzle in real estate. Section 5.3 reviews how we currently evaluate real estate

return and risk. Section 5.4 revisits the risk premium puzzle by taking the two biases into

consideration. Section 5.5 draws conclusions. In Section 5.6, we describe our on-going

research agenda to extend the work in this dissertation.

5.2. The Risk Premium Puzzle in Real Estate

Both early and recent studies have concluded that real estate not only has

extremely low volatility, but also has extremely high risk-adjusted returns. During the

period 1978 to 1998 (see Table 5.1), the standard deviation of the National Council of

Real Estate Investment Fiduciaries (NCREIF) returns was 3.66%, which is less than one-

fifth of that for common stocks (20.82% for the large cap stocks and 40.04% for the small

cap stocks), and less than a half of that of long-term bonds (8.32% for corporate bonds

and 8% for government bonds). In terms of risk-adjusted return, NCREIF’s Sharpe ratio

(1.47) is more than six times that of bonds and at least three times that of both the large

cap stocks (0.41) and the small cap stocks (0.35).23

Some have argued that the returns to real estate are less volatile than those of

common stocks because real estate returns, such as the NCREIF returns, are estimated by

using both appraised data and transaction prices, and appraisals tend to smooth real estate

values and make them less volatile (e.g. Geltner 1991, Clayton, Geltner and Hamilton

2001, Geltner, MacGregor and Schwann 2003). After excluding appraisal data in the

NCREIF property sample, Geltner and Goetzman (2000) have constructed a transaction-

23
The Sharpe ratio, by definition, can be computed as ( r~ − r f ) / σ ~r , where three-month T-Bill rate was
chosen as an approximation to the risk-free interest rate r f .
 81

based NCREIF index. They have found that the standard deviation of the transaction-

based NCREIF returns slightly increases from 3.66% to 4.26% in the period 1978 to

1998; however, it is still very low compared to those of stocks and bonds. In addition, the

annual return of the transaction-based NCREIF index is 9.2%, and its Sharpe ratio (1.27)

is still extremely high.

Table 5.1
Annual Returns and Volatility in Real Estate and Other Assets

(1978-1998)

Sharpe
Series Return Std Dev Ratio
Large Stocks 12.26% 20.82% 0.41
Small Stocks 17.8% 40.04% 0.35
LT Corp Bonds 5.74% 8.32% 0.24
LT Govt Bonds 5.13% 8.00% 0.17
IT Govt Bonds 5.24% 5.44% 0.28
NCREIF 9.12% 3.66% 1.47
Source: Ibbotson Associates

Table 5.2
Annual Returns and Volatility in Real Estate and Other Assets

Earlier Periods
 82

This pattern also holds for earlier periods when common stocks did not perform

well. Firstenberg, Ross and Zisler (1988) found that the standard deviation of stock

returns was over five times greater than that of real estate returns, and the standard

deviation of bond returns was three times that of real estate returns, while the average

returns of real estate were slightly higher than that of both stocks and bonds. Hoag (1980)

developed a property index based on a sample of 463 unleveraged properties valued at

over $540 million and found its return to be 14.2%, compared to common stock returns

of 3.7% reported by Ibbotson and Sinquefield (1982) for the corresponding time periods.

Zerbst and Cambon (1984) found the return of Commingled Real Estate Funds (CREF) to

be 14.0%, compared to 6.5% for common stocks of the same period. By examining the

standard deviation of returns, they also found that real estate returns appeared far less

volatile than those for common stocks and corporate bonds. For example, the standard

deviation of CREF returns for 1973-1981 was 4.7%, compared to 21.2% for common

stocks and 7.8% for corporate bonds in the same period. In addition, several other studies

consistently found similar results for the comparable time period. In particular, in the

period 1947 to 1978, the standard deviation (3.5%) of real estate was about one-fifth that

of stocks (18%) and real estate’s Sharpe ratio (1.31) was about three times that of stocks

(0.38). During the period 1960 – 1973, real estate’s Sharpe ratio (1.75) was even higher

and about eight times that of stocks (0.21) (see Table 4.2).

This raises the question, is there a risk premium puzzle in real estate? 24 Or do the

current valuation methods misprice real estate’s risk-adjusted return? To answer this

question, we first review how we currently evaluate real estate return and risk.

24
Some have argued that real estate has such a higher risk-adjusted return because it often involves huge
transaction costs. This argument has been challenged by Kallberge, Liu and Greig (1996).
 83

5.3. Current Valuation Approach to the Estimation of Real Estate Return and Risk

Virtually all financial and real estate estimates of risk and return are based on

historical prices. For example, in a given period of time, suppose the price of one asset at

time t is Pt ( t = 0, 1, 2, ..., T ) . Then the simplest and most commonly used formula for

estimating the return and risk of this asset are

T
Pi − Pi −1
∑
i =1 Pi −1
rˆ =
T

T
Pi − Pi −1
∑( i =1 Pi −1
− rˆ) 2
σˆ =
T −1

Due to the heterogeneity of real estate assets in both residential and commercial

markets, we cannot directly apply the formula above to estimate real estate return and

volatility. The simplest approach is to compose a real estate price index as the average of

the values of the properties being studied. A more sophisticated approach is represented

by regression-based indices of property values over time. The commonly used regression-

based models include the repeat-sales model, the hedonic model, the hybrid model and

the appraisal-based model. We briefly discuss each of these models and their

shortcomings.

The Repeat Sales Model: The repeat-sale model was first proposed by Bailey,

Muth and Nourse (BMN) (1963) and was popularized by Case and Shiller (1987). The

BMN repeat-sale model estimates a house price index by analyzing data where all units
 84

have been sold at least twice. This method produces the growth index in sale prices over

time by regressing, using ordinary least squares, the change in log price of each unit on a

set of dummy variables, with one dummy for each time period in the sample, except for

the first sale. BMN argued that if the log price changes of individual houses follow an

independent, identically distributed noise term, then the estimated index is the best linear

unbiased estimate of the percentage growth index in sales prices. Case and Shiller (1987)

argued that the house-specific component of the change in log price is probably not

homoscedastic but that the variance of this noise increases with the time interval between

sales. They suggest using generalized least squares to mitigate the heteroscedasticity

problem. Archer, Gatzlaff and Ling (1996) extended the standard repeat-sales model by

including a location dummy variable, which can be used to estimate annual house price

appreciation in each submarket, relative to overall market appreciation. The repeat-sales

model has the advantage of requiring no information on the characteristics of the unit, so

the data is widely available. It has a number of shortcomings as well, however. For

example, this method utilizes only a fraction of potential information on the market (the

units have to have been sold at least twice), it yields only estimates of price changes, and

no information on price level can be derived from this method. This method also

implicitly assumes that there is no significant change in the quality or quantity of the

units (or their neighborhoods) between sales.

The Hedonic Model: The hedonic model can overcome some the drawbacks of

the repeat-sales model. The basic idea of the hedonic model (see Rosen (1974)) is that

housing units, although different, can be decomposed into a bundle of measurable

housing characteristics. The hedonic model hypothesizes that property value is a

 85

function of measurable housing characteristics. An example of a simple hedonic model is

a regression of log (house price) on property characteristics. The independent variables

represent the individual characteristics of the housing unit, its lot, and its locational

amenities, and the coefficients are the implicit prices of these characteristics. The

regression results provide us with an estimated implicit price for each housing

characteristic, so we can deduce the price of any housing unit if given its housing

characteristics. The choice of a functional form of hedonic equation has been discussed in

Halverson and Pollakowski (1981). There are shortcomings of the hedonic model as well.

First, the hedonic model requires all the information of housing characteristics for the

units, and obtaining such a dataset is costly. Second, the coefficients of the underlying

hedonic model are assumed to remain constant. In other words, the implicit price of each

housing characteristic does not change over time. This assumption may not hold when

the time period of the data sample is relatively long (e.g. Green and Malpezzi (2003)).

Third, there exists the problem of omitted variables (e.g. Gatzlaff and Haurin (1998), and

Green and Malpezzi (2003)).

The Hybrid Model: The hybrid model takes advantage of the strengths of the

repeat-sales and hedonic models, estimating these two models simultaneously by

imposing a constraint that price changes over time are equal in both models (e.g. Case

and Quigley (1991), Quigley (1995), and Gatzlaff and Haurin (1998)).

The Appraisal-based Model: The appraisal-based model uses both transaction

prices and appraised values to construct real estate price (e.g. Clapp and Giaccotto

(1992)). The advantage of the appraisal-based model is being able to use the entire

sample, since appraised values are available for the whole sample. However, appraisal
 86

smoothing and appraisal bias are two serious problems with appraisal data (e.g. Genlter

(1991), Chinloy, Cho and Megbolugbe (1997), Clayton, Geltner and Hamilton (2001),

and Geltner, MacGregor and Schwann (2003)).

Both the real estate pricing models and the estimations of real estate return and

risk essentially use the information of historical prices and the time of sale. These

methods implicitly assume that real estate assets could be sold immediately at a historical

price at any time, as in the financial market. This assumption is violated in the real estate

market. In our previous chapters, we have demonstrated that there are two pricing biases

in the current valuation approach: liquidity bias and valuation bias.

In the next section, we revisit the risk premium puzzle by taking the two biases

into consideration.

5. 4. The Risk Premium Puzzle in Real Estate Revisited

For simplicity, we consider a real estate market where sellers are homogeneous

with the same reservation price and hence the same expected marketing period. Suppose

real estate return and volatility estimated from transaction sample are denoted by rT and

σ T , respectively, and the average marketing period is dented as τ . From equations

(4.19) and (4.20) in Chapter 4, the underlying market return and volatility when

marketing period ~
t = 0 , can be estimated as follows,

rM = rT − 3τσ T (5.1)

σ M = (1 + τ ) × σ T (5.2)
As mentioned earlier, some investors may face surprise financial distresses or

other better investment opportunities and have to sell their illiquid real estate within a
 87

short time period. To those investors, in order to have a quick sale, they have to lower

their reservation prices and accept a lower transaction return and higher risk. The degree

of such liquidity shocks can vary over investors. Some may have to sell their real estate

assets right away, and others may be able to wait a little bit longer. For simplicity, we use

θ ( 0 ≤ θ ≤ 1 ) to characterize the degree of such shock, and assume that the expected

marketing period is θ τ when such a shock occurs. Therefore, a lower θ implies a higher

liquidity shock. In particular, “ θ = 0 ” means that the sellers have to sell their assets

immediately and “ θ = 1 ” indicates that the sellers have no liquidity shocks.

Given a liquidity shock of θ and by applying Theorem 4.1, the investor’s

expected return and risk can be estimated as follows,

rT ,θ = rM + 3θ τσ T ,θ (5.3)

σM
σ T ,θ = (5.4)
1+θ τ

Where: rM and σ M are the market return and volatility, respectively

Inserting equations (5.1) and (5.2) into (5.3) and (5.4) yields,

3 τ (1 − θ )
rT ,θ = rT − σT (5.5)
1 + θτ

1+τ
σ T ,θ = σT (5.6)
1+θ τ

Since the return and volatility estimated from equations (5.5) and (5.6) do not take

marketing period risk into consideration, they are actually the ex-post return and

volatility. In Chapter 3, we have demonstrated that the “true” return and volatility in the
 88

real estate market are the ex-ante return and volatility, which account for not only price

risk but also marketing period risk. By applying Theorem 3.1, we can thus have,

u ex − ante = rT ,θ (5.7)

θ 2τ 2
(σ ex − ante ) 2 = σ T2,θ + rT ,θ (5.8)
TH + θτ

Where: TH is the investors’ expected holding period, and the marketing period is

assumed to follows an exponential distribution.

Suppose r f is risk-free rate, then current estimation of real estate’s Sharpe ratio

(ex-post) is,

rT − r f
Sharpe Ratio ex − post = (5.9)
σT

In other words, the current estimation implies that real estate’s Sharpe ratio is the same

across investors.

After taking marketing period risk and investor’s liquidity shock into account, the

new estimation of real estate’s Sharpe ratio (ex-ante) becomes,

ex − ante
u ex − ante − r f
Sharpe Ratio = (5.10)
σ ex − ante

From equations (5.5), (5.6), (5.7) and (5.8), we can conclude that u ex − ante and

σ ex− ante vary over investor’s liquidity shock θ and holding period. Hence, unlike the

current estimation, real estate’s Sharpe ratio varies over investors. The difference

between these two measures can be summarized in the following theorem.25

25
Proof can be found in the appendix.
 89

Theorem 5.1 (1) u ex − ante ≤ rT , σ ez − ante > σ T ; (2) Sharpe Ratio ex − ante < Sharpe Ratio ex − post ;

and lim Sharpe Ratio ex − ante = Sharpe Ratio ex − post .26 In addition, real estate’s Sharpe
τ →0

ratio is “investor-specific”. In particular, (3) the higher the investment time

∂Sharpe Ratio ex − ante

horizon, the higher the real estate’s Sharpe ratio, i.e., > 0 ; (4)
∂TH

the lower the liquidity shock, the higher the real estate’s Sharpe ratio, i.e.,

∂Sharpe Ratio ex − ante

< 0.
∂θ

Two points are worth noting in Theorem 5.1. First, the return estimated from a

transaction sample is overstated, however, the volatility is underestimated. Hence, the

current estimation of real estate’s Sharpe ratio is always overestimated. Second, holding

all else constant, a higher expected holding period implies a higher real estate’s Sharpe

ratio. In addition, investors who experience financial distresses and have to sell their real

estate quickly will have relatively lower Sharpe ratios than others.

In order to see how the expected holing period together with liquidity shocks

affect real estate’s Sharpe ratio, we consider both residential and commercial markets.

Case I: the US commercial property market

As before, we still use the annual return (9.2%) and volatility (4.3%) of the

NCREIF “transaction” index in the period 1978-98. We assume that the marketing period

~
26
Throughout this dissertation, we assume the ex-post return RTH + t is distributed with mean (TH + t )u
and variance (TH + t )σ . If it is distributed at a persistent return, we can show that (2) of Theorem 5.1
2

also holds.
 90

follows an exponential distribution and the expected marketing period is 8 months.

Suppose θ = 1, 3/4, 1 / 2, 1 / 3, 1 / 4, 1 / 6, 0 and the expected holding period ranges from 1

year to 30 years. We examine how real estate’s Sharpe ratio varies under different

scenarios.

Table 5.3 illustrates the degree to which the liquidity factor θ together with

sellers’ expected holding period could impact on real estate’s Sharpe ratio.

Table 5.3

Real Estate’s Sharpe Ratio: the case of the US commercial property market

The table above shows how real estate’s Sharpe ratio varies over investors’ holding period and liquidity
constraint factor. The liquidity constraint factor can be regarded as the degree flexibility the sellers can wait
on the market. When it is 0, sellers have to sell immediately. When it is 1, the sellers can wait and choose
anytime to maximize their Sharpe ratio. In this table, we choose the annual return and volatility as 9.2%
and 4.26%, respectively (data source: NCREIF), and the risk-free interest rate as 3.74%. Hence, the Sharpe
ratio estimated by the current approach is 1.27, which is seriously overstated, especially to the investors
with shorter-time horizon and higher liquidity constraints.

From Table 5.3, we can clearly see that, real estate’s Sharpe ratio substantially

varies over investment time horizon and seller’s liquidity shock. In particular, real
 91

estate’s Sharpe ratio increases with the expected holding period and decreases with the

liquidity-constraint (a lower θ implies a higher liquidity-constraint). Second, the current

approach, which ignores the marketing period risk, can seriously overstate real estate’s

Sharpe ratio. For example, for an investor intending to keep a property for ten years, his

Sharpe ratio can be as low as 0.07. In his best scenario, i.e., with no liquidity shocks, his

Sharpe ratio is 0.72. However, the Sharpe ratio estimated by the current approach is 1.27,

which is highly overstated.

Case II: the US residential property market

As in Chapter 3, we choose annual return of 5.7% and standard deviation of 1.7%

from the OFHEO’s US home price index during the period 1996Q1 to 2003Q3. Since the

average marketing periods in commercial markets are often longer than those in

residential markets, we assume that the expected marketing period in the residential

market is 6 months.

Table 5.4 demonstrates how the liquidity factor θ together with sellers’ expected

holding period impact on the real estate’s Sharpe ratio in the residential market. The

findings are very similar to those in the commercial market. For example, suppose an

investor intends to hold a property for ten years. His Sharpe ratio can be as low as 0.19

and as high as if 0.51. However, his Sharpe ratio estimated by the current approach is

1.15. Therefore, ignoring real estate illiquidity can seriously overstate real estate’s Sharpe

ratio.
 92

Table 5.4

Real Estate’s Sharpe Ratio: the case of the US residential property market

The table above shows how real estate’s Sharpe ratio varies over investors’ holding period and liquidity
constraint factor. The liquidity constraint factor can be regarded as the degree of flexibility the sellers can
wait on the market. When it is 0, sellers have to sell immediately. When it is 1, the sellers can wait and
choose whatever time maximizes their Sharpe ratio. In this table, we choose the annual return and volatility
as 5.7% and 1.7%, respectively (data source: OFHEO), and the risk-free interest rate as 3.74%. Hence, the
Sharpe ratio estimated by the current approach is 1.15, which is seriously overestimated, especially to the
investors with shorter-time horizon and higher liquidity constraints.

5.5. Conclusions

This dissertation documents the potential pricing biases in current real estate

valuation, which implicitly assumes real estate assets can be sold immediately without

waiting. The assumption of immediate execution may be reasonable in the financial

market where the time to trade an asset is trivial; however, it is certainly not valid in the

real estate market where marketing period is not only uncertain but also substantial. The

uncertainty of marketing period by all means exposes an investor additional risk,

however, little work has been done to study the size of this risk.
 93

This dissertation attempts to do so formally by proposing a new measure of ex-

ante return and variance to capture both price risk and liquidity risk faced by real estate

investors. Our findings suggest that the risk of the ex ante returns is substantially higher

than that given by the current estimations. For example, in the case of US residential

market, to short-term investors (one year holding period) the actual risk can be as high as

four times that of current estimations; and to long-term investors (10 years holding

period), although the risk of marketing period can be amortized over the longer holding

period, the actual risk can be still forty-seven percent higher than that of current

estimations. In the case of UK commercial market, Shawn and Hwang have found a

similar result. In particular, the actual risk faced by short-term investors is about 307%

higher than that of current estimation, and to long-term investors, it is still forty-eight

percent higher than that of current estimation.27 Therefore, the current real estate

valuation, which assumes real estate can be sold immediately and ignores the uncertainty

of marketing period, can seriously underestimate real estate risk.

In addition, our findings also suggest that investors with longer investment time

horizons will be less affected by real estate illiquidity, not because they do not trade

frequently but because the additional risk caused by the uncertainty of marketing period

is amortized over the longer holding period. This result is consistent with the common

perception that real estate is more favorable to long-term investors.

After correcting for “liquidity risk”, another bias still exists in the current real

estate valuation, because transaction prices are the prices from a “truncated” distribution

of market valuation. Holding other things equal, a higher reservation price implies a

longer expected marketing period and a larger deviation of transaction prices from the
27
The average marketing period is assumed to be eight months here.
 94

market valuation, we conjecture that there is a positive relationship between the expected

marketing period and valuation bias.

In Chapter 4, we formally examine how the expected marketing period affects the

valuation bias by considering a model in which the buyer’s biding prices are based on

market valuation and a transaction prices occurs if and only if a bidding pride equals or

exceeds the seller’s reservation price. We derive a “closed-form” relationship among

market return/risk, transaction return/risk and marketing period. We find that, a longer

expected marketing period implies a larger valuation bias, ceteris paribus. In particular,

investors with a higher expected marketing period are expected to receive a higher return

and lower risk in compensation. Moreover, our results suggest that, at the absence of real

estate illiquidity (marketing period is close to zero), the underlying market return is much

lower than transaction return and the underlying market risk is much higher than

transaction risk.

Finally, we conclude that the current valuation of real estate return and risk,

which borrows from finance theory by ignoring real estate illiquidity, not only

understates real estate risk but also overstates real estate returns. The findings of this

dissertation can help us to understand the “risk premium puzzle” in real estate.

There are several points worth noting in this dissertation. First, this dissertation is

the first to formally study how marketing period plays a role in real estate pricing. We

show that the current approach of estimating return and risk in the financial market is not

applicable to the real estate market, and a new estimation procedure has been provided.

Second, our findings indicate that real estate return and risk can vary substantially over

investment time horizons, seller’s liquidity shock and expected marketing period.
 95

Therefore, unlike the financial market, real estate return and risk are investor-specific.

For instance, if investors are likely to have short-term investment horizons and face

liquidity shocks, their actual risk can be many times higher than that of current

estimation, while their return can be much lower. Third, our analysis shows that

marketing period and its uncertainty play a crucial role in real estate pricing. This

suggests that, without proper modification it could be misleading to apply in simplistic

fashion finance theories, such as mean-variance portfolio theory and CAPM, to the real

estate world. Finally, throughout this dissertation we treat holding period as given.

However, a more realistic situation may be one where an investor has a certain

expectation about holding the property but there would be variation around this time

depending on different factors. Therefore, it would be beneficial for future research to

consider a more complex model to allow the holding period to become a random

variable.

5.6. Future Research

Our next project is to explain why the real estate portfolio-allocation puzzle

exists. That is, when the historical data on the performance of commercial property is

analyzed, the risk-return profile of real estate [even adjusting the problem of transaction

costs] is often seen to be far more appealing than bonds and stocks. Yet is it known that

institutional investors’ holding of property are well below what may be obtained from a

mean-variance portfolio analysis based on historical data. For example, it has been found

that the top 200 public and private pension funds have allocated only 3.6% to real estate
 96

in 1986, and more recent studies show that the large pension funds hold only between

3.5% and 4% of their assets in real estate and most smaller pension funds have nothing in

real estate. In addition, in the United Kingdom data from Russell/Mellon CAPS shows

the average allocation to real estate over the last 15 years was 2.5%. However, academic

studies of mixed-asset portfolios, by contrast, often suggest optimal allocation to real

estate ranges from 10% to 20% or more (Firstenberg, Ross and Zisler (1988), and

Kallberg, Liu and Greig (1996)).

Motivated by this gap puzzle between theory and practice and seeking to better

understand how the marketing period risk of real estate illiquidity impact on asset

allocation, we will consider a model in which an investor makes optimal portfolio choices

between illiquid real estate and liquid assets when facing surprise liquidity shocks. We

assume that the investor has to sell his portfolio for cash when such a liquidity shock

occurs.

In the traditional portfolio choice paradigm, one of the most fundamental

assumptions is that assets can be traded at transaction prices with immediate execution.

This assumption is certainly violated when a portfolio includes illiquid real estate assets.

As a result, the traditional portfolio theory may not be applicable to the portfolio that

includes a real estate asset.

How does the uncertainty of marketing period affect the optimal allocation to the

real estate asset? Suppose a surprise liquidity shock occurs at time t 0 . There are two

options for the investor to sell his real estate asset. First, sell the real estate asset

immediately at the underlying market return and market risk; Alternatively, place the real

estate asset on the market by transaction prices but facing uncertainty of marketing
 97

period. If choosing the first option, the real estate will be sold immediately; however,

from our previous discussion the investor will face a much lower return and much higher

risk. If choosing the second option, the investor is expected to receive a higher return,

however, he has to borrow money against the unsold real estate until it is successfully

sold and face additional marketing period risk.

After taking into account the uncertainty of marketing period, this dissertation has

shown that, treating real estate as a liquid asset by assuming it can be traded instantly at

observable transaction prices not only overstates real estate returns but also understates

real estate risk, and consequently exaggerates the benefit of real estate in a portfolio. As a

result, the optimal allocation to real estate is expected to be lower than that of a liquid

asset. In addition, we show that the optimal allocation to real estate decreases when

surprise liquidity shocks increase and the expected holding period decreases. Our

preliminary findings suggest that directly applying traditional portfolio theory without

considering real estate illiquidity is the cause of real estate portfolio-allocation puzzle.
 98

Proof of Theorem 5.1

From equations (5.5), (5.6), (5.7) and (5.8), we have,

u ex − ante < rT (A5.1)

σ ex − ante > σ T (A5.2)

Hence,
Sharpe Ratio ex − ante < Sharpe Ratio ex − post (A5.3)
Since
lim u ex − ante = rT
τ →0
(A5.4)
lim σ ex − ante = σ T
τ →0

Therefore, lim Sharpe Ratio ex − ante = Sharpe Ratio ex − post (A5.5)

τ →0

Given equation (5.8), we have,

∂σ ex − ante
<0 (A5.6)
∂TH
Thus,
∂Sharpe Ratio ex − ante
>0 (A4.7)
∂TH
Since we have,
∂σ ex − ante
<0 (A5.8)
∂θ
∂u ex − ante
And >0 (A5.9)
∂θ
Accordingly,
∂Sharpe Ratio ex − ante
<0 (A5.10)
∂θ
Q.E.D.
 99

Reference:

Archer W., Gatzlaff D., and Ling D. (1996), “Measuring the Importance of Location in
House Price Appreciation,” Journal of Urban Economics, 223-353

Bailey M., Muth R. and Nourse H. (1963), “A Regression Method for Real Estate Price
Index Construction,” Journal of the American Statistical Association, 58, 933-42

Case B. and Quigley J. (1991), “The Dynamic Of Real Estate Prices,” The Review Of
Economics and Statistics, 50-58

Case K. and Shiller R. (1987), “Prices of Single-Family Homes Since 1970: New Indexes
For Four Cities,” New England Economic Review, 45-56.

Chinloy P., Cho M., and Megbolugbe I. (1997), “Apprasials, Transaction Incentives, and
Smoothing”, Journal of Real Estate Finance and Economics, 89-111

Clapp J. and Giaccotto C. (1992), “Estimating Price Indices for Residential Property: A
Comparison of Repeat Sales and Assessed Value Methods,” Journal of The American
Statistical Association, 300-306.

Clayton J., Geltner D. and Hamilton S. (2001), “Smoothing in Commercial Property

Valuations: Evidence from Individual Appraisals,” Real Estate Economics, 337-360

Firstenberg, Ross and Zisler (1988), “Real Estate: The Whole Story”, Journal of
Portfolio Management; 14(3), 22-34.

Gatzlaff D. and Haurin D. (1998), “Sample Selection and Biases in Local House Value
Indices,” Journal of Urban Economics, 199-222

GeltnerD. (1991), “Smoothing in Appraisal-Based Returns,” Journal of Real Estate

Finance and Economics, p327-345.

Geltner D. and Goetzmann W. (2000), “Two Decades of Commercial Property Returns:

A Repeated-Measures Regression –Based Version of the NCREIF Index”, Journal of
Real Estate Finance and Economics, 5-21.

Geltner D., MacGregor B. and Schwann G. (2003), “Appraisal Smoothing and Price
Discovery in Real Estate Markets,” Urban Studies, 1047-1064

Green R. and Malpezzi S. (2003), ”A Primer on U.S. Housing Markets and Housing
Policy,” Urban Institute Press for the American Real Estate and Urban Economics
Association

Halverson R. and Pollakowski H. (1981), “Choice of Fundamental Form for Hednonic

Price Equations,” Journal of Urban Economics 10, 37-49
 100

Hoag J. (1980), “Toward Indices of Real Estate Value and Return”, Journal of Finance,
May.

Ibbotson R. and Fall C. (1979), “The United States Market Wealth Portfolio”, Journal of
Portfolio Management, Fall.

Ibbotson R. and Sinquefield R. (1982), “Stocks, Bonds, Bills and Inflation: The Past and
the Future”, The Financial Analysts Research Foundation

Kallberg J., Liu C. and Greig W. (1996), “The Role of Real Estate in the Portfolio
Allocation Process,” Real Estate Economics, 359-377

Kelleher R. (1976), “ How Real Estate Stacks Up to the S&P 500”, Real Estate Review,
Summer.

Quigley J. (1995), “A Simple Hybrid Model for Estimating Real Estate Price Indexes,”
Journal of Housing Economics, 1-12

Rosen S. (1974), “Hedonic Prices and Implicit Markets,” Journal of Political Economy,
177-201.

Shilling J. (2003), “Is There a Risk Premium Puzzle in Real Estate?” Real Estate
Economics, 501-525

Zerbst R. and Cambon B. (1984), “Historical Returns on Real Estate Investments”,

Journal of Portfolio Management, Spring.

View publication stats