Journal of Banking and Finance 92 (2018) 1–12

Contents lists available at ScienceDirect

Journal of Banking and Finance

journal homepage: www.elsevier.com/locate/jbf

House price, loan-to-value ratio and credit risk

Xun Bian a, Zhenguo Lin b,∗, Yingchun Liu c
a
Department of Accounting, Economics, Finance and Real Estate, College of Business and Economics, Longwood University, United States
b
Hollo School of Real Estate, College of Business, Florida International University, United States
c
Department of Finance, Insurance, Real Estate and Law, College of Business, University of North Texas, United States

a r t i c l e i n f o a b s t r a c t

Article history: Real estate transactions are often established through financing. We study the effect of financing on prop-
Received 13 December 2017 erty prices. We show that properties can transact at prices well above their collateral values. Therefore,
Accepted 5 April 2018
the commonly used loan-to-value (LTV) ratio suffers a bias that can significantly understate credit risk.
Available online 22 April 2018
This bias is exacerbated when mortgages are originated with longer terms, at higher LTV ratios, or when
Keywords: sellers possess stronger bargaining power. Furthermore, this bias is larger under aggressive lending prod-
Asset prices ucts, e.g. interest-only loans and mortgages allowing negative amortization. Our simulation results sug-
Bargaining gest that many mortgages originated at the peak of the housing bubble are, in fact, “under water” at
Mortgage financing origination. In particular, the loan amount of a 30-year mortgage at a 95% LTV can be 15% greater than
Loan-to-value ratio the collateral value of the property, suggesting the mortgage is already deep “under water” at origina-
Credit risk tion. These findings call into questions underwriting and risk control practices in mortgages and other
collateralized debts.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction Housing Administration (FHA) program or Veteran’s Loan Guaran-

The loan-to-value (LTV) ratio, calculated as the loan amount as
a percentage of the transaction price of the collateralized prop-
erty, is arguably the most widely used indicator for measuring fi-
nancial leverage and assessing credit risk in the United States and
around the world.1 Because the property acquired by the borrower
serves as security for the loan, as the LTV ratio increases, credit
risk escalates. The reasoning is twofold. First, a higher LTV ratio in-
creases the probability of negative equity (i.e., property value dips
below the mortgage balance). As a result, defaults become more
likely. Second, in the case of default, a higher LTV ratio means
the collateralized property is less likely to bring sufficient proceeds
at a foreclosure sale to cover the outstanding loan balance, past
due payments and other foreclosure costs incurred by the lender.
Therefore, loss given default will be greater. Relative to most other
asset classes, real estate is deemed more “collateralizable”, and
ownership of real estate are often accompanied by relatively high
LTV ratios. For example, conventional home loans in the U.S. can be
originated with a LTV ratio of up to 80% without credit enhance-
ments. With private mortgage insurance (PMI) or under the Federal
∗
Corresponding author.
E-mail addresses: bianx@longwood.edu (X. Bian), zlin@fiu.edu (Z. Lin),
yingchun.liu@unt.edu (Y. Liu).
1
In Australia and New Zealand, the Loan-to-value ratio may be abbreviated as
LVR.
tee Program, the LTV ratio can be well above 90%. Given the thin
cushion between the transaction price and the amount of debt, ef-
fective credit risk assessment and control critically rely on the LTV
ratio to accurately reflect true financial leverage. Specifically, the
validity of using the LTV ratio as a credit risk indicator hinges on
the assumption that transaction price is an unbiased measure of
a property’s collateral value. In this paper, we show that financing
and bargaining, which are two signature components of most real
estate transactions, cloud the validity and reliability of transaction
price serving that purpose.
Real estate transactions are often done through bargaining, and
resulted transaction prices are usually financed through mortgage
debt. In this study, we explore the price formation of a mortgage-
financed property in a bilateral bargaining game. We show that a
property can transact at a price that is well above its collateral
value, and the LTV ratio suffers a bias that understates credit risk.
Our results indicate that the common practice of using the LTV ra-
tio to gauge financial leverage is inaccurate. Consequently, credit
risk control strategies based on LTV ratios, which are currently
widely used in mortgage industries, are problematic. Our findings
are critical for several reasons. First, the scale of the residential
mortgage market is vast. According to Federal Reserve Bank of
New (2017), housing debt (mortgage balance plus housing equity
lines of credit (HELOC)) in the United States is estimated to be
$9.14 trillion in the second quarter of 2017. The scale of mort-
gage sector globally is of course much larger. Inaccurate measures

https://doi.org/10.1016/j.jbankfin.2018.04.006
0378-4266/© 2018 Elsevier B.V. All rights reserved.
2 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12
of financial leverage and ineffective control of credit risk of home
loans have severe consequences, an example of which is the 2007
housing market meltdown and the slowdown of the global econ-
omy many years after. Second, the bias we have identified in this
study also exists in many other collateralized debts (e.g. commer-
cial mortgages, auto loans). Therefore, our results call into ques-
tions underwriting practices and risk control strategies of those
debt instruments as well.
The deviation of transaction prices from collateral values stems
from the fact that transaction price capitalizes not only the col-
lateral value of the property but also transition-specific factors
such as financing and relative bargaining powers. For example, low
mortgage rates or other attractive financing terms, ceteris paribus,
reduces the cost of ownership and, as a result, boosts buyers’ will-
ingness to acquire real estate. Buyers’ eagerness to buy, through
price bargaining, translates into higher transaction prices. Addi-
tionally, a stronger position of the seller increases the degree to
which financing is capitalized and further raises transaction price.
It is worth noting that neither financing nor relative bargaining
strength is intrinsic to the property. Therefore, neither contributes
to the property’s collateral value. Ebbs and flows of the economy
drive mortgage rates up and down and cause various lending prod-
ucts to come and go. Housing markets oscillate between being
“hot” and “cold”, and strong bargaining power of either buyers or
sellers never perpetuates. Therefore, prices formulated under favor-
able financing terms and/or during a “hot” market tend to overesti-
mate collateral values.2 Consequently, the LTV ratio contains a bias
that underestimates credit risk. In this study, we derive a closed-
form solution of the bias contained in the LTV ratio. Our simula-
tion results suggest such a bias can be quite substantial. For exam-
ple, a 30-year fixed rate mortgage (FRM) originated at an 80% LTV
may, in fact, have its loan amount exceeding the collateral value of
the property. That is, the loan is essentially “under water” at orig-
ination. Under higher leverages and with more aggressive lending
products, the degree to which a loan is “under water” at origina-
tion increases dramatically. For instances, at a LTV ratio of 95%, a
30-year FRM may disburse a loan amount that is over 115.01% of
the collateral value of the property. In addition, a 30-year 80%-LTV
interest-only loan may disburse a loan amount that is 100.05% of
the collateral value of the property.
The bargaining-financing model we construct also allows us to
explore the impact of bargaining and financing on the liquidity of
real estate (i.e., time-on-market). To our knowledge, this is an area
that has not been examined in the literature. Sellers of real estate
conduct costly searches for buyers, and the benefit of search de-
pends on the seller’s potential gain from trade, which is ultimately
determined by transaction price. Therefore, if transaction price is
jointly determined by bargaining and financing, so should liquidity.
In our model, the seller optimizes her search time by taking into
account the impact of both bargaining and financing. We find that
stronger seller’s bargaining power and an increase in credit supply
incentivize sellers to search more extensively and wait longer.
We contribute to existing literature in several ways. First, our
model incorporates three critical elements of real estate transac-
tions: search, bargaining and mortgage financing, into a unified
2
Appraisers, who are employed to provide unbiased opinions of home values,
are supposed to protect lending institutions by properly measuring collateral values.
Unfortunately, many appraisers were not fulfilling their duty in pre-crisis transac-
tions. This was happening for two reasons. First, during the housing boom, com-
petition incentives appraisers to cater to loan offers to increase their probability of
winning future business. For example, Conklin et al. (2018) show that more com-
petition leads to more inflated appraisals (measured using at-price appraisals). Sec-
ond, residential real estate is appraised using the sales comparison method, under
which an appraiser anchors his or her estimate on transaction prices of comparable
sales. If prices of comparable sales were inflated, it will lead to an inflated appraised
value.
framework to enable a holistic perspective on real estate transac-
tion and to gain in-depth understanding of how property prices
and liquidity are determined. We show that transaction prices can
be influenced by financing arrangements, which are unrelated to
the fundamental value of a property. Because of that, transaction
prices often fail to serve as an unbiased estimate of collateral
value, and the LTV ratio can significantly understate credit risk. Our
study is the first to identify and quantify such a bias contained in
the widely used LTV ratio for credit risk control. Additionally, we
contribute to the literature by examining factors that influence the
magnitude of such bias. Specifically, we show that the bias is larger
when a mortgage is originated with a longer term, at a higher LTV
ratio, or when the seller possesses stronger bargaining power rela-
tive to the buyer. Furthermore, this bias is larger under aggressive
lending products, such as interest-only loans and mortgages that
allow negative amortization. In addition, our analysis sheds light
on an important but under-investigated question: How do bar-
gaining and financing influence liquidity? We are able to derive a
closed-form solution of the optimal time-on-market that captures
the impact of both bargaining and financing.
The second contribution of our study is that our theoretical
framework adds new insights to the understanding of real estate
market cycles. Our model can offer explanations why the pervasive
use of aggressive lending products was a crucial contributing fac-
tor that fueled the dramatic escalation of U.S. housing prices dur-
ing the early years of the 20 0 0s. It is worth noting that our ex-
planation is different from ones offered by previous studies, which
generally suggest that aggressive lending products boost housing
demand by enabling risky borrowers who would previously have
difficulties of obtaining a loan to become homeowners. This addi-
tional demand bids up housing prices. We, on the other hand, sug-
gest an alternative channel through which aggressive lending prod-
ucts can incite a bubble. Having features that allow borrowers to
delay repayments (e.g. interest-only (IO) loans and mortgages with
negative amortization), aggressive lending products can be more
attractive to some borrowers than traditional home loans. Through
price bargaining, the value-added loan is split between the buyer
and seller and translates into a higher transaction price. We show
that aggressive lending products, even when only taken by high-
quality borrowers rather than risky borrowers and produce no ad-
ditional demand, can inflate house prices.
The rest of the paper is organized as follows. Section 2 reviews
related literature. In Section 3, we construct a model of real estate
bargaining with mortgage financing and analyze the joint effect of
bargaining and financing on property price formation and liquidity.
In Section 4, we simulate our model under various scenarios to
present economic significance about the impact of bargaining and
financing. Section 5 provides concluding remarks.
2. Literature review
The effect of mortgage financing on property transaction prices
has been examined by many empirical studies. It is well known
that how a property is financed can have a substantial influ-
ence on its transaction price. Zerbst and Brueggeman (1977),
Brueggeman and Zerbst (1979) and Colwell et al. (1979) compare
residential property transactions financed through Federal Hous-
ing Administration (FHA) loans or Veteran Affair (VA) loans, under
which discount points are paid by sellers, to transactions financed
through conventional mortgages. The studies find that sellers shift
the cost of discount points to FHA buyers and VA buyers by ad-
justing prices upward. Sirmans et al. (1983) study house prices re-
sulted from transactions in which assumable mortgages are used.
They show that houses with assumable loans tend to sell at higher
prices than the ones financed with conventional mortgages carry-
ing higher interest rates. Rosen (1982) studies creative financing ar-
 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12
rangements, such as teaser rates, assumable mortgages, and seller
financing at below-market rates. He finds that the advantages of
creative financing are capitalized in housing prices leading to a
“creative financing premium”. More recent studies investigate the
role of subprime mortgages in home prices and how they cause
housing bubble conditions where the demand for subprime lend-
ing fueled lenders’ willingness to extend loans to more risky buy-
ers, which in turn helped to further fuel the housing bubble and
eventually led to the 20 07–20 09 housing crash due to borrowers’
defaults for various economic and behavioral reasons (e.g. Pavlov
and Wachter, 2011; Collins et al., 2015, and Seiler, 2015a, 2015b,
2018).
Our study relates to this stream of literature by formally exam-
ining the effect of mortgage financing on prices. First, our model
provides a theoretical framework to help understand price forma-
tion of a mortgage-financed property. Although there is empirical
evidence that advantageous financing arrangements are capitalized
in transaction prices, the exact mechanism of such capitalization
is less clear. Our model shows exactly how the capitalization of fi-
nancing into prices is realized. Furthermore, our analysis goes be-
yond prices by looking at two other important issues. First, we
study how transaction price can loop around and impact loan per-
formance. For example, a transaction price substantially above the
property’s collateral value may imply escalated likelihood of mort-
gage default. More importantly, the increased credit risk is not fully
reflected in the widely used LTV ratio, and, as a result, lenders bear
that risk but are uncompensated for. Second, we examine the ef-
fect of financing and bargaining on liquidity, which is largely over-
looked in the literature. We are the first to derive a closed-form
solution of the optimal time-on-market that captures the impact
of both bargaining and financing. In addition, our findings can also
explain recent empirical findings in the finance literature. For ex-
ample, Min and Yang (2009) find that loss-given-default is posi-
tively related to the LTV ratio. We show that when the original LTV
ratio is higher, the transaction price is more likely to be inflated. In
a distressed housing market when house prices decline, high-LTV
loans are of course more likely to default and produce a greater
loss-given-default (because the collateral value is overstated due
to financing). Cowan and Cowan (2004) focus on default correla-
tion (i.e. when defaults occur, how likely are they going to happen
at the same time?). Our findings can explain why subprime mort-
gages have a high default correlation. In our model, high credit risk
has nothing to do with idiosyncratic borrower characteristics (e.g.
FICO scores and household income). Instead, risky loans are made
due to a systematic flaw in mortgage underwriting (e.g. the LTV
ratio understating credit risk). Therefore, when house prices start
to fall, many defaults are going to happen at the same time.
Our paper also contributes to the large body of literature
on hedonic pricing models. Most hedonic pricing models do
not explicitly take into account financing terms. Hansz and
Hayunga (2016) show that only two studies control for financing
terms in hedonic pricing models in the literature from 20 0 0 to
2014. Our findings contribute to the thin literature on the effect of
financing on home price formation and highlight the importance
of incorporating financing terms in hedonic pricing models.
Our study is also related to the literature on non-cooperative
bargaining theory, which has been heavily influenced by the sem-
inal work of Rubinstein (1982). Since Rubinstein (1982), the bar-
gaining literature has been extended in various ways. In general,
bargaining outcomes may be influenced by (1) outside options,3
(2) the rules of the bargaining process,4 and (3) information and
3
See Binmore (1985).
4 8
See Binmore et al. (1986), Baliga and Serrano (1995) and Krishna and Ser-
rano (1996).
3
market structures.5 Our model extends non-cooperative bargaining
theory in a different direction. We show that there exists a feed-
back loop between bargaining and financing. On the one hand, the
outcome of bargaining (e.g. transaction price) is affected by avail-
able financing arrangements, and more attractive financings lead
to higher prices. On the other hand, transaction price determines
how much can be financed (e.g. the loan amount) and the level
of credit risk undertaken by the lender. Such an intertwined rela-
tionship between financing and bargaining has not been analyzed
before, and our study attempts to fill this gap.
3. The model
3.1. The setup
Real estate market distinguishes itself from financial markets
in its high degree of illiquidity. Unlike participants in the finan-
cial market who can readily buy or sell a security at its quoted
price, the seller in the real estate market must search for a de-
sirable buyer. During the search process, the seller receives offers
over time from a stream of potential buyers whose timing of ar-
rival and offering prices are both stochastic in nature.6 Suppose
that tj is the time interval between the arrival of the ( j − 1 )th
and jth buyers, then the random arrival time of the Nth buyer is

TN = Nj=1 t j . At time TN , the seller accumulates N offers.7
Suppose that seller’s second-best use of the property is R;8 and
for buyer j ( j = 1, 2, 3, . . . , N ), we denote Vj as his private value of
the property. We begin by considering the simple case of a cash
transaction, in which mortgage financing is not used and buyer j
pays cash for his offer price Pj if his offer is accepted. In this case,
his gain and the seller’s gain from this transaction are respectively
Bj = V j − Pj (1)
and
S = Pj − R (2)
Therefore, the total gain from trade, j , is the summation of Bj
and S ,
 j = Bj + S = V j − R (3)
Among the N buyers, some may find other appropriate proper-
ties and, as a result, exit the bidding. We denote ζ as the percent-
age of bidders who are still interested in the property, thus (1 − ζ )
is the percentage of exiting offers. Other things being equal, a
higher ζ implies fewer exiting offers and a smaller supply of simi-
lar properties, hence a tighter housing market. Therefore, ζ N is the
5
See Myerson and Satterthwaite (1983), Samuelson (1984), Abreu and
Gul (20 0 0) and Fuchs and Skrzypacz (2010) for bargaining under information asym-
metry. See Rubinstein and Wolinskey (1990) and Corominas-Bosch (2004) for bar-
gaining under different market structures.
6
Search has been studied extensively in the real estate literature. Some recent
examples include He et al. (2018) and Cheng et al. (2015). Our model differs from
these two previous studies in several ways. Our model takes a holistic view of
the real estate transaction process by modeling its three key elements in a unified
framework: search, bargaining and mortgage financing. We model both the seller’s
and the buyer’s payoffs in a bargaining game, which endogenously determines the
equilibrium price and, in term, the benefit of search. Additionally, our model also
includes mortgage financing, which affects both parties’ payoffs and the seller’s
search strategy. In contrast, both He et al. (2018) and Cheng et al. (2015) assume
that offer prices follow a certain distribution and, without bargaining, sellers make
take-it-or-leave-it offers. In other words, He et al. (2018) and Cheng et al. (2015) de-
pict partial search equilibria, while we drive a more general equilibrium encom-
passing search, bargaining and financing.
7
Note that TN is a random variable. In our model, the seller will optimally choose
N and TN , and we discuss this optimization in Section 3.2.
Throughout the paper, we use “she” when referring to the seller and “he” when
referring to the buyer.
4 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12

number of available offers at waiting time on market TN , where E, and (2) the present value of mortgage repayment to the buyer.

TN = Nj=1 t j . We denote V max = max{V1 , V2 , . . . , VNζ }, and the total Comparing Eqs. (11) to (1) reveals that buyer’s gain from trade un-
gain from trade is the highest when a transaction occurs between der a cash transaction is a special case of Eq. (11), in which P = E.
the seller and the buyer with the highest valuation of the property, Characterizing the buyer’s gain more generally, Eq. (11) is critical in
i.e., showing how financing alters the price formation process. Specifi-
cally, with financing, the transaction price is no longer bounded by
max = V max − R (4)
the buyer’s and the seller’s valuations of the property.
There will be no trade if there is no gain from the trade, i.e.,
max = V max − R < 0 or Vmax < R. The necessary condition for the Theorem 1. With mortgage financing, the total gain from trade be-
trade to occur is the existence of non-negative gain from the trade, comes,
Vmax ≥ R. In addition, suppose that the offer price for the buyer  
who has the highest valuation of the property is P.9 Since both r
 max
=V max
− R + (P − E ) 1 − (12)
the buyer’s gain and the seller’s gain should be non-negative, we i
have V max − P ≥ 0 and P − R ≥ 0. As a result, P must satisfy,
Financing creates new ground for trading. In contrast to conventional
V max ≥ P ≥ R (5)
wisdom, a buyer does not have to value a property more than its seller
Therefore, with a cash transaction, the price is bounded by for a mutually beneficial trade to occur, and transaction price is not
seller’s and buyer’s valuations of the property. In particular, it is bounded and it can be above the buyer’s valuation.
above the seller’s valuation but below the buyer’s valuation.
In reality, cash purchases are rare, and mortgages are often used Proof. Comparing Eqs. (12) to (4), we observe that the introduc-
to finance acquisitions of real estate. To introduce financing in the tion of mortgage financing into the model transforms the total gain
model, suppose that the buyer makes an equity investment, E, and from trade from (V max − R ) to V max − R + (1 − 
 )(P − E ). This
r
i
finances the rest of the purchase price by borrowing (P − E ) at an transformation suggests: first, mortgage financing alters the neces-
interest rate i under a T-period conventional fixed-rate mortgage sary condition under which a mutually beneficial trade can occur.
(FRM).10 With cash purchase, there will be no transaction if Vmax < R. How-
With the T-period FRM, the periodic mortgage payment, PMT, ever, that is no longer the case when mortgage financing is used.
must satisfy, Even with Vmax < R , we can still have max > 0 if (1 −   )(P −
r
i

T E ) > R − V max . In other words, favorable mortgage financing makes
P MT
P−E = (6) the impossibility of trading under cash transaction possible. Sec-
τ =1 (1 + i )τ ond, with financing, transaction price, P, is no longer bounded by
Hence, the seller’s and the buyer’s valuations of the property. To see this,
 we rewrite Eq. (11) as

T
1
P MT = (P − E ) (7)  
τ =1 (1 + i )τ  B,max
=V max
− P + (P − E ) 1 −
r
(13)
i
Suppose that the buyer has a discount rate of r for his future
cash flows, the present value of his mortgage payments can be cal-
Eq. (13) implies that when (1 − 
 )(P − E ) is strictly greater
r
culated as follows, i
 than zero, the transaction price P can be higher than the buyer’s

T
P MT value of the property, Vmax , because it is still possible for the buyer
PV B = (P − E ) (8)
τ =1
(1 + r )τ to have a positive gain from the transaction as long as Eq. (13) is
strictly positive. QED 
Substituting Eq. (7) into Eq. (8) yields,
   Theorem 1suggests that when mortgage financing is used,

T
P MT 
T
P MT
PV B = (P − E ) (9) transaction price reflects not only a property’s collateral value but
τ =1
(1 + r )τ τ =1 (1 + i )τ also how financing is structured.

To simplify notation, we adopt two definitions: r ≡

T 
1
and i ≡ Tτ =1 (1+1i )τ . Therefore, Eq. (9) can be
τ =1 (1+r )τ 3.2. Seller’s optimal waiting time and price formation
rewritten as
r A defining feature of real estate transaction process is the se-
PV B = (P − E )( ) (10)
i quential but random arrival of potential buyers. Both the time of
offer arrival and offer prices are stochastic. An optimization prob-
With financing, the seller’s gain from trade remains to be S =
lem confronted by the seller is when to stop search and begin price
P − R.11 However, the buyer’s gain from trade becomes
  negotiation. On the one hand, the seller has an incentive to wait
r for more buyers in order to increase the chance of receiving higher
 B,max
=V max
− E − (P − E ) (11)
i offer prices. On the other hand, waiting is costly. Beyond a certain
point, the marginal cost of waiting may outweigh its benefit, and
Eq. (11) indicates that the buyer’s gain is equal to his valua- waiting further is no longer justified. In other words, the seller first
tion of the property, Vmax , minus two costs: (1) equity investment, needs to determine when is the optimal time to stop the search
and start a negotiation.
9
In the remainder of the paper, when referring to the buyer we mean the buyer Following Arnold (1999), we assume that buyer j’s valuation,
who has the highest valuation of the property and is expected to outbid the others. Vj , is distributed over support [P , P ] with mean μ and variance
10
By conventional FRM, we mean a fully amortizing fixed-rate mortgage with a
σ 2 , where P and P are respectively the lower and upper bounds.
constant periodic payment. For simplicity, we consider a conventional FRM first.
Financing through other types of mortgages will be discussed later in the paper. Suppose that f(V) and F(V) are respectively the probability density
11
We assume the source of the buyer’s financing is a third-party financial institu- function and cumulative distribution function of Vj . Given that Vmax
tion, not the seller. is the highest valuation among Nζ available buyers, the density
 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12 5

function of Vmax can be expressed as follows:12 Eqs. (17) and (18) indicate the seller’s and the buyer’s gain
from trade depend on two things: (1) the size of total gain from
gV max (V ) = Nζ F (V )(Nζ −1) f (V ) (14) trade, max ; (2) the seller’s and the buyer’s bargaining powers
We thus have (θSBgn , θBBgn ). In particular, both the seller and the buyer receive a
greater payoff if the total gain from trade is greater. In addition,
V
E[V max ] = V Nζ F (V )(Nζ −1 ) f (V )dV (15) seller’s (buyer’s) gain decreases with buyer’s (seller’s) bargaining
V power and increases with her (his) own bargaining power. Intu-
itively, the seller (the buyer) receives a larger share if she (he) is
We now evaluate the seller’s marginal benefit of waiting. We
more patient or if the buyer (the seller) is less patient.
model the bargaining between the seller and the buyer by using
With Eq. (15), we can rewrite Eq. (12) as follows:
the non-cooperative bargaining model of Rubinstein (1982). If the
 
total gain from trade, max in Eq. (12), is greater than zero, the r
seller and the buyer will engage in a bargaining game described in  max
= E[V max
] − R + (P − E ) 1 − (19)
i
Rubinstein (1982) to divide the gain. The seller first makes an of-
fer by proposing a split (x, max − x ). If the buyer accepts the pro- From Eqs. (19) and (17), the seller’s expected gain from trade is,
posal, the game ends and a deal is reached. The payoffs received by
the seller and the buyer are x and (max − x ), respectively. If the
 
1 − θBBgn r
proposal is rejected, the buyer makes a counter-offer by proposing S = E[V max ] − R + (P − E ) 1 − (20)
1 − θSBgn θBBgn i
a new split (x , max − x ). If the seller accepts it, the game ends,
and the payoff to the seller and the buyer are now equal to θS x
Bgn
Note that seller’s gain is also equal to (P − R ). Therefore, the
and θB (max − x ), where (θS < 1 ) and (θB < 1 ), respectively,
Bgn Bgn Bgn
transaction price, P, must satisfy,
represent the seller’s and the buyer’s bargaining power. There are
 
many factors affecting θS and θB , including seller’s and buyer’s
Bgn Bgn
1 − θBBgn r
search cost and their anxiety of waiting. Alternatively, if the seller P−R= E[V max
] − R + (P − E ) 1 − (21)
1 − θSBgn θBBgn i
rejects the offer, she makes a counter-offer. Thereafter, the seller
and the buyer take turns to make proposals. Solving for P from Eq. (21), we can obtain a closed-form so-
We follow the technique developed by Shaked and Sut- lution of the equilibrium transaction price when bargaining and
ton (1984) to determine the perfect equilibrium of this bargain- mortgage financing are considered simultaneously:
ing game. We define M as the supremum of the payoff that the
1−θBBgn   1−θBBgn
seller can obtain in any perfect equilibrium of the game. Suppose R− 1− 
i E +
r
E[V max ] − R
1−θSBgn θBBgn 1−θSBgn θBBgn
the buyer rejects the seller’s first offer, the seller’s payoff becomes P= (22)
1−θBBgn
θSBgn M. Now consider a proposal made by the buyer after rejection. 1− 1− 
i
r
1 −θ θ
Bgn Bgn
Any split proposed by the buyer which gives the seller more than S B

θSBgn M will be accepted by the seller, so there is no perfect equilib- Therefore, the marginal benefit of waiting for the Nth buyer’s
rium in which the seller receives more than θS
Bgn arrival can be estimated by,
M. It follows that
the buyer will get at least max − θS
Bgn
M in any perfect equilibrium 1−θBBgn ∂ E[V max ]
of the subgame beginning from that point. In fact, max − θS M is
Bgn ∂P 1−θSBgn θBBgn ∂N
= (23)
the infimum of the payoff received by the buyer in this subgame. ∂ N 1 − 1 − θB Bgn
1− r
1−θ Bgn θ Bgn i
Now consider a proposal made by the seller in the first offer. S B

Any split proposed by the seller which gives the buyer anything Substituting Eq. (15) into Eq. (23), we have,
less than θB (max − θS M ) will not be accepted by the buyer.
Bgn Bgn
 
Hence the seller will obtain at most max − θB (max − θS M ).
Bgn Bgn
1−θBBgn
∂ V Nζ F (V )(Nζ −1 ) f (V )dV
V
In fact, as before, this is the supremum of what the seller will re- ∂P θ
1− SBgn BBgn θ V
=  
ceive in the first offer. ∂N 1 − 1− BBgn θ
1−  r ∂N
θ
1− SBgn BBgn θ i
But the game of the seller’s second offer is identical to the
game of the seller’s first offer, apart from shrinkage of all payoffs (24)
due to θS and θB . Hence, it follows that the supremum of the
Bgn Bgn
As we know, the total waiting time for the Nth buyer’s arrival is
seller’s payoff must equal M. Therefore, M should satisfy, 
TN = Nj=1 t j . For the distribution of tj , Bond et al. (2007) examine
M = max − θBBgn max − θSBgn M (16) the U.K. data and find that the buyer’s stochastic arrival tj follows
the Poisson process with rate λ. Therefore,
(1−θBBgn )
Solving for M, we have M = max . Hence, the split of
(1−θSBgn θBBgn ) 
N
N
the gain between the seller and the buyers are as follows, E[TN ] = E[t j ] = (25)
λ
(a) The seller’s gain from trade is: j=1

(1 − θBBgn ) Following Haurin (1988), suppose that the seller’s holding cost
S = M = max (17) of waiting for the Nth buyer’s arrival is g(TN ) = cTN , where c is the
(1 − θSBgn θBBgn ) holding cost per unit of time. The expected holding cost of waiting
(b) The buyer’s gain, which is simply the difference between the for the Nth buyer’s arrival can thus be estimated as follows,
total gain and the seller’s gain, is, cN
E[g(TN )] = (26)
θ (
Bgn
θ
1 − SBgn ) max λ
B = max − S = B
 (18)
(1 − θ θ
Bgn Bgn
) Since the optimal N∗ must meet the condition of “marginal ben-
S B
efit of waiting for the Nth buyer’s arrival = marginal cost of wait-
ing for the Nth buyer’s arrival”, N∗ should satisfy the following con-
12
See (Ross, 2002, p. 275). dition:
6 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12

 
1−θBBgn Four observations can be made on Eq. (33). First, consistent
∂ V Nζ F (V )(Nζ −1 ) f (V )dV
V
θ
1− SBgn BBgnθ V c with standard bargaining theory, we show that P is strictly in-
  =
θ
1− BBgn ∂N λ creasing in the bargaining power of the seller (i.e., ∂Bgn
P
> 0)
1− 1− 
i
r
∂θS
θ θ
1− SBgn BBgn
and strictly decreasing in the bargaining power of the buyer (i.e.,
(27) ∂ P < 0). Second, the transaction price increases with the un-
Bgn
∂θB
Read (1988) and Lin and Vandell (2007) assume that buyer’s
derlying value of the property (i.e., ∂∂ RP > 0). Third, the transac-
offer price is uniformly distributed. If this is the case, we can show
tion price increases with the buyer’s discount rate for his future
that,13
mortgage payments (i.e., ∂∂Pr > 0). Finally, holding other things be-
V
ζ NV + V ing equal, transaction price can be far away from the underlying
V Nζ F (V )(Nζ −1 ) f (V )dV = (28) value of the property (E[Vmax ] or R) when buyer’s discount rate r
V ζN + 1
is relatively bigger and his down-payment (E) is small.
Therefore,
V
∂ V V Nζ F (V )(Nζ −1 ) f (V )dV ζ V −V 3.3. The bias of price-based loan-to-value (LTV) ratio
= (29)
∂N (ζ N + 1 ) 2
With the transaction price under the seller’s optimal N∗ in Eq.
√
V −V
Since = σ2
, we have V − V = 2 3σ . Therefore, we can
12
(31), we next derive the bias of price-based LTV ratio. Suppose that
rewrite Eq. (29) as follows: both the seller and the buyer agree on the underlying value of the
V √ property, i.e., E (V max |N ∗ ) = R.14 In other words, both the seller and
∂ V V Nζ F (V )(Nζ −1 ) f (V )dV 2 3σ the buyer think the property is worth E (V max |N ∗ ) or R, and this
= (30)
∂N (ζ N + 1 )2 should be the property’s collateral value. Therefore, Eq. (33) be-
comes,
From Eqs. (27) and (30), we can solve the optimal N∗ and the
1−θBBgn r
seller’s expected optimal waiting time on market in Theorem 2. R−
1−θSBgn θBBgn
(1 − i E )
P= (34)
Theorem 2. The optimal number of potential buyers, (N∗ ), which the 1−θBBgn r
1−
1 −θ θ
Bgn Bgn (1 − i )
seller should wait for, can be expressed as follows, S B
⎛ ⎞
 √  1−θ Bgn  Because E (V max |N ∗ ) or R is unobservable in the market, only
 2 3 λζ σ
1 ⎜ B
1−θSBgn θBBgn ⎟ transaction price is revealed after bargaining between the seller
∗
N = ⎜ − 1⎟ (31)
ζ ⎝ 1−θBBgn r
⎠ and the buyer. From Eq. (34), we can impute the underlying value
1−
1 −θ θ
Bgn Bgn 1 − i c of the property (E (V max |N ∗ ) or R) from the observable transaction
S B
price P as follows,
where ζ is the percentage of buyers who are still available. λ is the
1 − θBBgn r
buyers’ arrival rate. c is the holding cost per unit of time. σ is the R=P− (1 − )(P − E ) (35)
volatility of offer prices. θS and θB are respectively the seller’s and
Bgn Bgn
1−θ S B
θ
Bgn Bgn i
the buyer’s bargaining powers. In addition, from Eqs. (25) and (31),
Because R (i.e., the underlying value of the property) is unob-
the seller’s expected optimal waiting time on market is
⎛ ⎞ servable, the transaction price is commonly used as a proxy of col-
 √  1−θ Bgn  lateral value to control for borrower’s credit risk. Given the fact
 2 3 λζ σ
1 ⎜  ⎟
B
1−θSBgn θBBgn that the second term in the right-hand side of Eq. (35) is always
E[TN∗ ] = ⎜  − 1⎟ (32)
λζ ⎝ 1−θBBgn r
⎠ positive, the finding in Eq. (35) suggests that transaction price is
1− 1 − i c likely to overestimate collateral value (R) and result in a LTV ratio
1 −θ θ
Bgn Bgn
S B
that is too low. As a result, high default risks could hide behind a
Theorem 2 suggests that the expected optimal waiting time on seemingly reasonable LTV ratio. We next examine how the price-
market is affected by many factors such as seller’s holding cost based LTV ratio underestimates true financial leverage and hence
(c), market conditions (λ), the tightness of the market (ζ ), and the underestimates the borrower’s default risk.
dispersion of buyers’ offer prices. More importantly, E[TN∗ ] is also Since the underlying value of the property is E (V max |N ∗ ) or R,
determined by both bargaining and financing. First, E[TN∗ ] is in- the true LTV ratio should be calculated as,
creasing in seller’s bargaining power and decreasing in the buyer’s P−E
bargaining power. This is consistent with our common intuition. LT V true = (36)
R
Stronger seller bargaining power allows her to get a bigger share of
the pie. As a result, she is incentivized to search longer and more However, the underlying value of the property cannot be ob-
extensively to make the pie larger. Second, Eq. (32) also reveals served in the market, in practice, the transaction price is used in-
that E[TN∗ ] is strictly decreasing in the mortgage rate, i. This sug- stead to calculate the LTV ratio, i.e.,
gests that increase in credit supply (i.e., lower mortgage rates), ce- P−E
teris paribus, makes sellers search more extensively and wait longer LT V price−based = (37)
P
for potential buyers.
Under the seller’s optimal N∗ , we can express the resulted equi- The difference between Eqs. (36) and (37) is the bias of the LTV
librium transaction price as follows: ratio caused by this common practice, namely,
  1 1

1−θBBgn 1−θBBgn
R− 1− 
i E +
r
E[V max | N ∗ ] − R LT V bias = LT V price−based − LT V true = (P − E ) − (38)
1 −θ θ
Bgn Bgn
1 −θ θ
Bgn Bgn
P R
P∗ = S B S B
(33)
1− BBgn θ
1− 1− 
i
r
θ
1− SBgn BBgn θ
14
Our main results are unchanged if seller and buyer do not agree on the under-
lying value of the properties: either buyer’s valuation > seller’s valuation or buyer’s
13
See the proof in Appendix I. valuation < seller’s valuation.
 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12 7

When mortgage financing is used, the buyer is willing to Proof. Both mortgages have the same loan amount of P − E, we
pay more for the property. As a result, P > R.15 In other words, thus have
LTVbias < 0, and the LT V price−based is likely to underestimate the true 
T 
T
P MTt P MTtalt
loan-to-value ratio and thus fails to capture the credit risk of the =P−E = (45)
loan. From Eq. (35), we can rewrite Eq. (38) as follows,
(1 + i )
t=1
t (1 + i )t t=1

From Eq. (44), we can simplify Eq. (45) as follows;

LT V bias = LT V price−based − LT V true
⎛ ⎞ P MT jalt
P MT j P MTk P MTkalt
1 1 + = + (46)
= ( P − E )⎝ − ⎠ (39) (1 + i ) j (1 + i ) k (1 + i ) j ( 1 + i )k
P θ
1− BBgn r
P−
θ θ
1− SBgn BBgn
(1 − i )(P − E ) Since the buyer chooses to take a loan only when r > i, thus
1+i
1+r < 1. Given that P MT jalt < P MT and P MTkalt > P MT , and with Eq.
Note that P − E is the loan amount, which can be rewritten as (46), we have
LTVprice-based P. Substituting it into Eq. (39), we can obtain a closed-
 k − j  k− j
form solution of the LTV bias in the following theorem. P MT j P MTk 1+i P MT jalt P MTkalt 1+i
+ > +
Theorem 3. The LTV bias of Fixed-RM mortgages can be expressed as ( 1 + i ) j ( 1 + i )k 1+r (1 + i ) j ( 1 + i )k 1+r
follows, (47)
  Multiplying both sides of Eq. (47) by ( 1+ r)
1+i k− j
yields,
1
LT V bias = LT V price-based 1 − P MT jalt
1−θBBgn r P MT j P MTk P MTkalt
1−
1−θSBgn θBBgn
(1 − i )LT V price-based + > + (48)
(1 + r ) j (1 + r ) k (1 + r ) j ( 1 + r )k
(40) Thus we have,

T
P MTt 
T
P MTtalt
3.4. Alternative lending products > (49)
(1 + r )
t=1
t (1 + r )t
t=1
Now we consider a more general type of mortgage in which The total gain from trade is the sum of buyer’s gain and seller’s
the buyer borrows (P − E) at an interest rate i with varying mort- gain, which can be expressed as follows:
gage payments. Suppose that the term of the mortgage is T, and
the buyer’s mortgage payments from period 1 to T are denoted as

T
P MTt
max = [E[V max |N∗ ] − E − ] + (P − R ) (50)
P MT1 , P MT2 , . . . , P MTt , . . . , P MTT −1 , P MTT , then P MTt (t = 1, 2, . . . , T ) (1 + r )t
t=1
must satisfy,
If the buyer takes the alternative mortgage, Eq. (50) becomes,

T  
P MTt 
T
P MTtalt
P−E = (41)
(1 + i )t max,alt = E[V max |N ∗ ] − E − + (P − R ) (51)
t=1 (1 + r )t
t=1
If the buyer has an alternative mortgage with everything else The first term in the bracket in Eqs. (50) and (51) is the buyer’s
being equal but with different payments in the periods j and k as gain from trade. From Eqs. (49)–(51), we can conclude that the
follows, buyer’s gain increases due to delaying mortgage payments, which
P MTtalt = P MTt , ∀t ∈ {1, 2, . . . , j − 1, j + 1, . . . , k − 1, k + 1, . . . , T }, results in an increase in the total gain from trade,

P MT jalt = P MT j and P MTkalt = P MT (k > j ) (42) max,alt > max . (52)

Therefore, delaying mortgage payments increase both the
Suppose that the alternative mortgage pays less at period j with
buyer’s gain and the total gain from trade, which results in higher
delaying payments, i.e., P MT jalt < P MT j . Since the loan amount for
transaction prices.16 QED 
the two mortgages is the same, we thus have,
During the 20 04–20 06 housing boom, the U.S. home prices had

T
P MTtalt 
T
P MTt a double-digit increase annually, which resulted in affordability is-
= (43)
(1 + i )t
t=1
(1 + i )t
t=1
sues in many areas across the country. Because interest-only (IO)
mortgages allowed borrowers to pay only interest portion, it be-
We can readily obtain, came very popular during the housing boom. Compared with FRM
loans, IO mortgages are a good example of delaying mortgage pay-
P MTtalt = P MTt ∀t ∈ {1, 2, . . . , j − 1, j + 1, . . . , k − 1, k + 1, . . . , T }, ments, and consequently, they will lead to not only higher trans-
P MT jalt < P MT j and P MTkalt > P MT (k > j ) (44) action price but also a larger LTV bias.

Theorem 4. Both the buyer’s gain and the total gain from trade Theorem 5. The LTV bias of IO mortgages can be calculated as fol-
are higher under the mortgages with delaying payments. As a result, lows:
the more aggressive lending products such as interest-only loans and LT V bias = LT V price-based
mortgages that allow negative amortization (i.e., two perfect examples  
1
of the mortgages with delaying payments) will lead to higher trans- × 1− N
1−θBBgn
action prices. 1−
1−θSBgn θBBgn
(1 − τ =1 (1+r )τ − (1+r )N )LT V
i 1 price-based

(53)
15
This result is consistent with the findings of many empirical studies that
properties with mortgage financing always sell higher prices than those with
cash transactions. For example, Lusht and Hansz (1994) find a 16.5% discount for
16
cash-only transactions in Lehigh County, Pennsylvania. More recently, Aroul and As payments are pushed back, prices are more inflated and cash flows become
Hansz (2011) and Jauregui et al. (2017) find 13.5% and 9% discount for cash trans- more concentrated on the far end. As a result, duration of the loan increases, and
actions in Clovis, California and Franklin County, Ohio, respectively. this subjects the lender to greater default risk.
8 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12

Table 1
Simulated LTVbias : base case – fully amortizing FRM.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
30-year fully amortizing FRM (LTV = 80%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −2.00% 0.00%
4.00% −3.92% −1.95% 0.00%
4.50% −5.76% −3.81% −1.89% 0.00%
5.00% −7.51% −5.60% −3.71% −1.84% 0.00%
5.50% −9.18% −7.30% −5.44% −3.60% −1.78% 0.00%
6.00% −10.78% −8.93% −7.10% −5.28% −3.49% −1.73% 0.00%
6.50% −12.29% −10.48% −8.68% −6.89% −5.13% −3.39% −1.68% 0.00%
7.00% −13.73% −11.96% −10.19% −8.43% −6.70% −4.98% −3.29% −1.63% 0.00%
1−θBBgn
15−year fully amortizing FRM (LTV = 80%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −1.10% 0.00%
4.00% −2.18% −1.09% 0.00%
4.50% −3.24% −2.15% −1.07% 0.00%
5.00% −4.27% −3.19% −2.12% −1.06% 0.00%
5.50% −5.28% −4.21% −3.15% −2.09% −1.04% 0.00%
6.00% −6.27% −5.21% −4.16% −3.11% −2.06% −1.03% 0.00%
6.50% −7.23% −6.18% −5.14% −4.10% −3.06% −2.04% −1.01% 0.00%
7.00% −8.17% −7.13% −6.10% −5.07% −4.04% −3.02% −2.01% −1.00% 0.00%

Proof. With IO mortgage, the present value of its payments in prediction of our model, when r > i, LTVbias is strictly negative.
Eq. (9) can be rewritten as follows: This finding is also in line with empirical findings that properties

N
i 1
! with mortgage financing sell at higher prices than those paid with
PV B = (P − E ) + (54) cash. Our simulation enables an examination of the magnitude of
t=1
(1 + r )t (1 + r )N LTVbias . For example, when {r = 3.5%, i = 3%}, LTVbias is −2.00 per-
centage points. To put this in perspective, at a LT V price−based of 80%,
We can express r
i in Eq. (10) as follows: LTVtrue , which is calculated based on collateral value, is 82.00%.
LTVbias increases as the gap between r and i grows wider. When
r N
i 1
{r = 7%, i = 3%}, LTVbias becomes −13.73 percentage points, which
= + (55)
i t=1 (1 + r )t (1 + r )N suggests a LTVtrue of 93.73%. In other words, a mortgage originated
We can obtain the LTV bias of IO mortgages by using Eq. at an 80% LTV is, in fact, just slightly “above water” at origination.
(40) with replacing  Furthermore, we also see that the same gap between r and i, say a
 in Eq. (55). QED 
r
i one percentage point difference between r and i, leads to a greater
LTVbias when mortgage rates are low. For example, LTVbias is -3.92
4. Simulations
percentage points when {r = 4%, i = 3%}. In comparison, it is −3.29
percentage points, when {r = 7%, i = 6%}. We ran a similar simula-
To understand the economic significance of the LTVbias and to
tion with a 15-year FRM and present results in the bottom panel of
quantify the influence of financing and bargaining on LTVbias , we
Table 1. First, all patterns we previously identified using a 30-year
conduct a series of simulations.
FRM continue to exhibit under a 15-year FRM. More importantly,
it is clear that a shorter term reduces LTVbias for all {i, r} combi-
4.1. Base case
nations. This observation is consistent with our model’s prediction
that the ability to postpone repayments leads to a greater LTVbias .
We begin simulating a base case with the assumptions that
We further simulate LTVbias with interest-only mortgages. To es-
(1) LT V price-based = 80% and (2) the buyer and the seller possessing
Bgn
1−θB
tablish comparability, we use the same set of parameters previ-
equal bargaining power, which implies Bgn Bgn = 0.5. We choose ously used in our simulations of fully-amortizing FRMs. Simulated
1−θS θB
those parameter values to simulate a typical mortgage under a nor- values of LTVbias are shown in Table 2. It is evident that shifting
mal market condition, in which the buyer and the seller make a from a conservative amortization schedule to a more aggressive
“50–50” split of the gain from trade. We let buyer’s discount rate one substantially increases LTVbias . For example, with {r = 3.5%, i =
r and mortgage rate i vary between 3% and 7% (with a 0.5% per- 3%}, LTVbias is −3.08 percentage points for an IO mortgage (previ-
centage point increment). This generates 45 different {i, r} combi- ously −2.00 percentage points under a fully-amortizing FRM). Ex-
nations. For each {i, r} combinations, we simulate LTVbias with a amining other {i, r} combinations, we observe LTVbias is greater
30-year and a 15-year fully amortizing FRM. Simulated values of across the board under an IO mortgage. Again, this set of results
LTVbias are presented in Table 1. is consistent with our model’s prediction that postponement of re-
Several important observations can be made from Table 1. First, payments leads to a greater LTVbias .
when r = i, LTVbias is unbiased. In this case, financing adds no gain
to the trade, and a buyer is indifferent between obtaining mort- 4.2. Financial leverage
gage financing and a cash purchase. As a result, the property trans-
acts at its collateral value. However, when r is strictly greater than In addition to the base case, we further simulate LTVbias with
i, financing enhances the total gain from trade, and transaction high financial leverage under both fully-amortizing FRMs and IO
price exceeds the property’s collateral value. Consequently, price- mortgages. Specifically, we increase LTVprice-based from 80% in our
based LTV ratio understates financial leverage. Consistent with the base case scenarios to 90% and then to 95%. Simulated values
 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12 9

Table 2
Simulated LTVbias : base case – IO mortgage.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
30-year IO mortgage (LTV = 80%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −3.08% 0.00%
4.00% −6.00% −2.89% 0.00%
4.50% −8.76% −5.63% −2.72% 0.00%
5.00% −11.34% −8.22% −5.30% −2.56% 0.00%
5.50% −13.76% −10.64% −7.73% −4.99% −2.42% 0.00%
6.00% −16.01% −12.91% −10.01% −7.28% −4.71% −2.29% 0.00%
6.50% −18.11% −15.04% −12.15% −9.43% −6.87% −4.45% −2.17% 0.00%
7.00% −20.05% −17.01% −14.15% −11.46% −8.91% −6.50% −4.22% −2.06% 0.00%
1−θBBgn
15-year IO mortgage (LTV = 80%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −1.91% 0.00%
4.00% −3.78% −1.84% 0.00%
4.50% −5.59% −3.64% −1.78% 0.00%
5.00% −7.37% −5.40% −3.52% −1.72% 0.00%
5.50% −9.09% −7.11% −5.21% −3.40% −1.67% 0.00%
6.00% −10.75% −8.77% −6.86% −5.04% −3.29% −1.61% 0.00%
6.50% −12.37% −10.37% −8.46% −6.63% −4.87% −3.18% −1.56% 0.00%
7.00% −13.93% −11.93% −10.01% −8.17% −6.41% −4.71% −3.08% −1.51% 0.00%

Table 3
Simulated LTVbias : high leverage – fully amortizing FRM.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
30-year fully amortizing FRM (LTV = 90%, = 0.5)
1−θSBgn θBBgn

3.00% 0.00%
3.50% −2.54% 0.00%
4.00% −5.00% −2.47% 0.00%
4.50% −7.36% −4.86% −2.40% 0.00%
5.00% −9.62% −7.15% −4.72% −2.33% 0.00%
5.50% −11.79% −9.35% −6.94% −4.58% −2.26% 0.00%
6.00% −13.87% −11.46% −9.08% −6.74% −4.45% −2.20% 0.00%
6.50% −15.86% −13.48% −11.14% −8.82% −6.55% −4.32% −2.13% 0.00%
7.00% −17.76% −15.42% −13.10% −10.82% −8.57% −6.36% −4.19% −2.07% 0.00%
1−θBBgn
30-year fully amortizing FRM (LTV = 95%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −2.84% 0.00%
4.00% −5.59% −2.76% 0.00%
4.50% −8.23% −5.43% −2.68% 0.00%
5.00% −10.78% −8.00% −5.27% −2.60% 0.00%
5.50% −13.24% −10.48% −7.77% −5.12% −2.53% 0.00%
6.00% −15.59% −12.86% −10.18% −7.54% −4.97% −2.45% 0.00%
6.50% −17.85% −15.15% −12.49% −9.88% −7.32% −4.82% −2.38% 0.00%
7.00% −20.01% −17.35% −14.72% −12.13% −9.59% −7.11% −4.68% −2.31% 0.00%

of LTVbias under a 30-year fully-amortizing FRM are presented in
Table 3. It is clear that higher financial leverage drastically exacer-
bates LTVbias . For example, with {r = 3.5%, i = 3%}, LTVbias increases
from −2.00 to −2.54 percentage points when LTVprice-based goes
from 80% to 90%, and it further goes up to −2.84 percentage points
with a LTVprice-based of 95%. When high financial leverage is accom-
panied by a larger difference between r and i, the bias of the price-
based LTV ratio is quite substantial. At a of 95%, {r = 7%, i = 3%}
produces a LTVbias of −20.01 percentage points and a LTVtrue of
115.01% (i.e., 95% + 20.01% = 115.01%). In other words, the loan
amount of a mortgage originated at a 95% LTV is in fact 15.01%
greater than the collateral value of the property. Table 4 presents
simulation results of a 30-year IO mortgage with high leverage. It
is clear that, ceteris paribus, highly levered IO loans are riskier than
a fully amortizing FRM with the same level of leverage. LTVbias is
substantially higher for all {i, r} combinations.
4.3. Bargaining power
We now explore the influence of bargaining power on LTVbias .
So far, we assumed the buyer and the seller possess equal bar-
gaining powers. However, in a “hot” market, increased demand for
housing may put the seller in an advantageous bargaining position
and allow her to get a larger share of the pie. To simulate such a
market environment, we first increase the seller’s share of the total
Bgn
1−θB
gain from trade, Bgn Bgn , from 0.5 to 0.8. Results are reported in
1−θS θB
the top panel of Table 5.
Observing from the top panel of Table 5, enhanced seller’s bar-
gaining power exacerbates LTVbias . For example, with {r = 3.5%, i =
3%}, LTVbias goes from −2.00 percentage points in the base-case
scenario to −3.26 percentage points. As shown in the bottom panel
of Table 5, LT V price-based understates default risk even more when
10 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12

Table 4
Simulated LTVbias : high leverage − IO mortgage.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
30-year IO mortgage (LTV = 90%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −3.92% 0.00%
4.00% −7.67% −3.68% 0.00%
4.50% −11.24% −7.19% −3.46% 0.00%
5.00% −14.62% −10.53% −6.76% −3.26% 0.00%
5.50% −17.80% −13.70% −9.90% −6.36% −3.07% 0.00%
6.00% −20.79% −16.68% −12.87% −9.32% −6.00% −2.91% 0.00%
6.50% −23.59% −19.49% −15.67% −12.12% −8.79% −5.68% −2.75% 0.00%
7.00% −26.20% −22.12% −18.32% −14.76% −11.43% −8.31% −5.38% −2.61% 0.00%
1−θBBgn
30-year IO mortgage (LTV = 95%, = 0.5)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −4.38% 0.00%
4.00% −8.59% −4.11% 0.00%
4.50% −12.61% −8.05% −3.86% 0.00%
5.00% −16.43% −11.81% −7.56% −3.64% 0.00%
5.50% −20.05% −15.39% −11.09% −7.12% −3.43% 0.00%
6.00% −23.46% −18.78% −14.45% −10.44% −6.72% −3.24% 0.00%
6.50% −26.67% −21.98% −17.63% −13.60% −9.85% −6.35% −3.07% 0.00%
7.00% −29.67% −24.99% −20.64% −16.60% −12.83% −9.31% −6.01% −2.91% 0.00%

Table 5
Simulated LTVbias : high seller bargaining power – fully amortizing FRM.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
30-year fully amortizing FRM (LTV = 80%, = 0.8)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −3.26% 0.00%
4.00% −6.47% −3.16% 0.00%
4.50% −9.63% −6.28% −3.07% 0.00%
5.00% −12.74% −9.35% −6.10% −2.98% 0.00%
5.50% −15.78% −12.36% −9.07% −5.92% −2.89% 0.00%
6.00% −18.76% −15.31% −11.99% −8.80% −5.74% −2.81% 0.00%
6.50% −21.67% −18.20% −14.85% −11.63% −8.54% −5.57% −2.72% 0.00%
7.00% −24.50% −21.01% −17.65% −14.40% −11.28% −8.28% −5.40% −2.64% 0.00%
θ
1− BBgn
30-year fully amortizing FRM (LTV = 80%, = 1)
θ θ
1− SBgn BBgn
3.00% 0.00%
3.50% −4.11% 0.00%
4.00% −8.25% −3.99% 0.00%
4.50% −12.41% −8.01% −3.88% 0.00%
5.00% −16.58% −12.04% −7.77% −3.76% 0.00%
5.50% −20.75% −16.07% −11.67% −7.54% −3.65% 0.00%
6.00% −24.91% −20.10% −15.58% −11.32% −7.31% −3.54% 0.00%
6.50% −29.05% −24.12% −19.47% −15.09% −10.97% −7.08% −3.43% 0.00%
7.00% −33.16% −28.11% −23.35% −18.85% −14.62% −10.63% −6.87% −3.33% 0.00%

1−θB
the seller possesses all the bargaining power (e.g. Bgn Bgn
1−θS
LT V price-based becomes −4.11 percentage points with {r = 3.5%, i =
3%}. Table 6 presents simulation results of a 15-year IO mort-
gage with seller possessing more bargaining power than the buyer.
LTVbias is substantial when aggressive lending products are used
in a “hot” market. In the case of {r = 7%, i = 3%} and the seller
Bgn
1−θB
possessing all bargaining power (e.g. Bgn Bgn = 1). The LTVbias is
1−θS θB
−33.75 percentage point under a 15-year IO mortgage. This sug-
gests a LTVtrue of 113.75% (i.e., 80%+33.75%=113.75%).
5. Conclusion
Real estate prices are often established through bargaining and
financed with mortgage debt. We formally study the effect of bar-
gaining and financing on prices. In this paper, we explore the price
Bgn
formation of a mortgage-financed property in a bilateral bargain-
= 1).
θB ing game. We find that when the price of a mortgage-financed
property is formed through bargaining, the resulted transaction
price reflects not only the value of the property, it also captures
the value created through financing. As a result, mortgage-financed
properties can trade at prices that are well above their collateral
values, and the loan-to-value (LTV) ratio contains a bias that un-
derstates credit risk. This bias is exacerbated when a mortgage is
originated with a longer term, at a higher LTV ratio, or when the
seller possesses stronger bargaining power relative to the buyer.
Furthermore, this bias is larger under aggressive lending prod-
ucts, such as interest-only loans and mortgages that allow negative
amortization. Our simulation results suggest that many mortgages
originated at the peak of the housing bubble are likely “under
water” at origination. In particular, the loan amount of a 30-year
mortgage at a 95% LTV can be 15.01% greater than the collateral
value of the property, suggesting the mortgage is already deeply
 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12 11

Table 6
Simulated LTVbias : high seller bargaining power – IO mortgage.

r 3.00% 3.50% 4.00% 4.50% 5.00% 5.50% 6.00% 6.50% 7.00

1−θBBgn
15-year IO mortgage (LTV = 80%, = 0.8)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −3.10% 0.00%
4.00% −6.22% −2.99% 0.00%
4.50% −9.34% −6.00% −2.89% 0.00%
5.00% −12.47% −9.00% −5.79% −2.79% 0.00%
5.50% −15.60% −12.01% −8.68% −5.59% −2.70% 0.00%
6.00% −18.72% −15.01% −11.58% −8.38% −5.40% −2.61% 0.00%
6.50% −21.82% −18.00% −14.46% −11.16% −8.09% −5.22% −2.53% 0.00%
7.00% −24.90% −20.97% −17.32% −13.93% −10.77% −7.82% −5.05% −2.45% 0.00%
1−θBBgn
15-year IO mortgage (LTV = 80%, = 1)
1−θSBgn θBBgn
3.00% 0.00%
3.50% −3.91% 0.00%
4.00% −7.92% −3.78% 0.00%
4.50% −12.03% −7.64% −3.64% 0.00%
5.00% −16.22% −11.58% −7.37% −3.52% 0.00%
5.50% −20.50% −15.60% −11.16% −7.11% −3.40% 0.00%
6.00% −24.85% −19.69% −15.01% −10.75% −6.86% −3.29% 0.00%
6.50% −29.27% −23.84% −18.93% −14.46% −10.37% −6.63% −3.18% 0.00%
7.00% −33.75% −28.05% −22.90% −18.21% −13.93% −10.01% −6.41% −3.08% 0.00%

“under water” at origination. The bias of the LTV ratio also exists Given that Vmax is the highest valuation among all available
in many other collateralized debts (e.g. commercial mortgages, car buyers, the density function of Vmax is
loans). gV max (V ) = Nζ F (V )(Nζ −1 ) f (V ) (58)
Our findings cast doubt on the widely adopted practice of using
transaction price to estimate collateral value and call into ques- Therefore, we have,
tions underwriting and risk control practices of mortgage indus-
 (Nζ −1)  
V
V −V 1 ζ NV + V
tries and other collateralized debts. For example, in the United E[V max ] = V Nζ dV = (59)
V V −V V −V ζN + 1
States, an 80% LTV is the cutoff for determining whether or not
private mortgage insurance is required on a conventional mort- References
gage. This 80% cut-off has been consistently applied by Fannie Mae
and Freddie Mac for many years. Our findings suggest that this Abreu, D., Gul, F., 20 0 0. Bargaining and reputation. Econometrica 68 (1), 85–117.
Arnold, M., 1999. Search, bargaining and optimal asking prices. Real Estate Econ. 27
lack of variation on the cutoff is likely to be suboptimal. Since
(3), 453–481.
20 0 0, both U.S. mortgage market and housing market have un- Aroul, R., Hansz, A., 2011. The role of dual-pane windows and improvement age in
dergone some dramatic changes. For example, mortgage rates de- explaining residential property values. J. Sustain. Real Estate 3, 142–161.
Baliga, S., Serrano, R., 1995. Multilateral bargaining with imperfect information. J.
clined continually and repeatedly hit historical lows. In addition,
Econ. Theory 67 (2), 578–589.
U.S. housing market experienced a full boom-and-bust cycle. Our Binmore, K., 1985. Bargaining and coalitions. In: Roth, A.E. (Ed.), Game-theoretic
analysis suggests that with changing market conditions, it is un- Models of Bargaining. Cambridge University Press.
likely an 80% LTV represents the same level of credit risk at differ- Binmore, K., Rubinstein, A., Wolinsky, A., 1986. The Nash bargaining solution in eco-
nomic modelling. RAND J. Econ. 176–188.
ent times. An optimal credit risk control strategy should involve Bond, S., Hwang, S., Lin, Z., Vandell, K., 2007. Marketing period risk in a portfolio
having time-varying and market-specific underwriting standards. context: theory and empirical estimates from the UK commercial real estate
As we show in our analysis, when mortgage rates are low, LTVbias is market? J. Real Estate Finance Econ. 34 (4), 183–191.
Brueggeman, W., Zerbst, R., 1979. Discount points and housing prices: a reply.. J.
greater and, as a result, credit risk increases. Therefore, originators Finance 34 (4), 1055–1060.
should either charge a larger risk premium (higher mortgage rate) Cheng, P., Lin, Z., Liu, Y., Seiler, M.J., 2015. The benefit of search in housing markets.
or apply tighter underwriting standards (e.g. require higher credit J. Real Estate Res. 37 (4), 597–621.
Collins, A., Harrison, D., Seiler, M., 2015. Mortgage modification and the decision to
scores and/or larger down payments). Furthermore, transactions in strategically default: a game theoretic approach.. J. Real Estate Res. 37, 439–470.
hot housing markets, in which sellers possess stronger bargaining Colwell, P., Guntermann, K., Sirmans, C., 1979. Discount points and housing prices:
power relative to buyers, are more susceptible to the bias we have comment.. J. Finance 34 (4), 1049–1054.
Conklin, J., Coulson, N.E., Diop, M., Le, T., 2018. Competition and appraisal inflation.
identified. Loans originated in those markets should be subject to
Working Paper.
stricter underwriting standards or charged a larger risk premium. Corominas-Bosch, M., 2004. Bargaining in a network of buyers and sellers. J. Econ.
Theory 115 (1), 35–77.
Cowan, A., Cowan, C., 2004. Default correlation: an empirical investigation of a sub-
Appendix I prime lender.. J. Bank. Finance 28 (4), 753–771.
Fuchs, W., Skrzypacz, A., 2010. Bargaining with arrival of new traders. Am. Econ.
Suppose that buyers’ valuations are uniformly distributed over Rev. 100 (3), 802–836.
Hansz, J., Hayunga, D., 2016. Revisiting cash financing in residential transaction
support [P , P ], hence their probability density function is prices. Real Estate Finance 33 (1), 33–46.
" Haurin, D., 1988. The duration of marketing time of residential housing.. AREUEA J.
1/(V − V ) if V j ∈ [V , V ] 16, 396–410.
f (V j ) = (56) He, X., Lin, Z., Liu, Y., Seiler, M.J., 2018. Search benefit in housing markets: an in-
0 otherwise
verted u-shaped price and TOM relation.. Real Estate Econ. Forthcoming.
Jauregui, A., Tidwell, A., Hite, D., 2017. Sample selection approaches to estimating
The cumulative distribution function of buyers’ valuation is house price cash differentials. J. Real Estate Finance Econ. 54, 117–137.
# Krishna, V., Serrano, R., 1996. Multilateral bargaining. Rev. Econ. Stud. 63 (1), 61–80.
1 if V j > V Lin, Z., Vandell, K., 2007. Illiquidity and pricing biases in the real estate market. Real
F (V j ) = (V j − V )/(V − V ) if V j ∈ [V , V ] (57) Estate Econ. 35, 291–330.
0 otherwise
12 X. Bian et al. / Journal of Banking and Finance 92 (2018) 1–12
Lusht, K., Hansz, A., 1994. Some further evidence on the price of mortgage contin-
gency clause. J. Real Estate Res. 9, 213–217.
Myerson, R., Satterthwaite, M., 1983. Efficient mechanisms for bilateral trading. J.
Eco.n Theory 29 (2), 265–281.
Federal Reserve Bank of New York, 2017. Quarterly report on household debt
and credit (q2 2017). Available at https://www.newyorkfed.org/microeconomics/
hhdc.html (accessed on Oct. 16 2017).
Pavlov, A.D., Wachter, S.M., 2011. Subprime lending and real estate prices. Real Es-
tate Econ. 38, 1–17.
Min, Q., Yang, X., 2009. Loss given default of high loan-to-value residential mort-
gages. J. Bank. Finance 33 (5), 788–799.
Read, C., 1988. Price strategies for idiosyncratic goods—the case of housing. Real
Estate Econ. 16 (4), 379–395.
Rosen, K., 1982. Creative financing and house prices: a study of capitalization ef-
fects.. Working Paper.
Ross, S., 2002. A First Course in Probability., sixth ed Prentice Hall.
Rubinstein, A., 1982. Perfect equilibrium in a bargaining model. Econometrica 50,
97–109.
Rubinstein, A., Wolinskey, A., 1990. Decentralized trading, strategic behavior and the
Walrasian outcome. Rev. Econ. Stud. 57, 63–78.
Samuelson, W., 1984. Bargaining under asymmetric information. Econometrica 52,
995–1005.
Seiler, M., 2015a. Do as i say, not as i do: the role of advice versus actions in the
decision to strategically default. J. Real Estate Res. 37, 191–215.
Seiler, M., 2015b. The role of informational uncertainty in the decision to strategi-
cally default. J. Hous. Econ. 27, 49–59.
Seiler, M., 2018. Asymmetric dominance and its impact on mortgage default defi-
ciency collection efforts. Real Estate Econ. Forthcoming.
Shaked, A., Sutton, J., 1984. Involuntary unemployment as a perfect equilibrium in
a bargaining model. Econometrica 52 (6), 1351–1364.
Sirmans, G., Smith, S., Sirmans, C., 1983. Assumption financing and selling price of
single-family homes. J. Financ. Quant. Anal. 18 (3), 307–317.
Zerbst, R., Brueggeman, W., 1977. FHA And VA mortgage discount points and hous-
ing prices.. J. Finance 32 (5), 1766–1773.