Adv. Math. Econ.

8, 383-396 (2006) Advances in

MATHEMATICAL
ECONOMICS
©Springer-Verlag 2006

A prepayment model of mortgage-backed

securities based on unobservable prepayment cost
processes*
Hidetoshi Nakagawa^ and Tomoaki Shouda^
^ Graduate School of Innovation Management, Tokyo Institute of Technology, 2-12-1
Ookayama, Meguro-ku, Tokyo 152-8552 Japan
(e-mail: [email protected])
^ Mitsubishi UFJ Trust Investment Technology Institute Co., Ltd. 2-5-6 Shiba, Minato-/
ku, Tokyo, 105-0014 Japan and
Graduate School of International Corporate Strategy, Hitotsubashi University, 2-1-2
Hitotsubashi, Chiyoda-ku, Tokyo 101-8439 Japan
(e-mail: [email protected])

This is a summary version of the original paper [24]^

Received: April 28, 2005
Revised: September 29, 2005

JEL classification: CI 1, C15, G13

Mathematics Subject Classification (2000): 91B28, 91B70

Abstract. We propose a prepayment model of mortgage based on a structural approach
in order to analyze prepayment risk of mortgage-backed securities (MBS). We introduce
a continuous process named prepayment cost process. Specifically, each mortgager's
prepayment time is defined by the first time when her or his prepayment cost process
falls below zero, but prepayment cost processes are supposed to be unobservable in
the market. We also introduce a risk unique to each loan pool of mortgages, called
a loan pool risk (LPR), and we regard LPR as a systematic risk other than interest rate.
Using the model, we discuss the conditional distribution of prepayment times and a risk-
neutral valuation of pass-through MBS. It is shown that each mortgager's conditional
non-prepayment probability and the posterior distribution of LPR play quite important
roles in our study.
Key words: mortgage-backed securities (MBS), prepayment cost, loan pool risk, risk-
neutral valuation
* This research is partially supported by Grant-in-Aid for Young Scientists (B)
No. 16710108 from the Ministry of Education, Culture, Sports, Science and Tech-
nology.
^ Please refer to the original paper [24] for detailed discussion as well as some illus-
tratitions of numerical experiment on simulation.
384 H. Nakagawa, T. Shouda

1. Introduction
The purpose of this study is to present a prepayment model of mortgage which
rationally explains the heterogeneity of prepayments observed in the actual
MBS market.
There have been many literatures on prepayment risk model for MBS
(Mortgage-backed securities) in both theoretical and practical viewpoints. Pre-
payment risk models are classified into several approaches. Among them, we
adopt the option-based approach that is based on the methods used in mathe-
matical finance.
In the option-based approach, if transaction costs are negligible, the value
of mortgage can be regarded as an interest rate option since it depends upon
the difference between the current mortgage rate and the original contract rate
whether prepayment is advantageous or not.
However, actual prepayment is not so simple since mortgagers might be
charged with various costs and commissions at prepayment. Also, prepay-
ment action is strongly affected by mortgager's mental or environmental causes
which are impossible to quantify in terms of money and which are different
from one mortgager to another.
Additional costs which are required at prepayment is explicitly introduced
to the models in some literature. Dunn and Spatt [8] introduce the model in
which transaction cost is necessary for prepayment of the whole remaining
principal and analyze the effect of such a cost on prepayment action.
Stanton [31] presents the model where the transaction cost is proportional
to the remaining principal and different from mortgager to mortgager. He also
assumes that each mortgager's transaction cost follows Beta distribution and
estimates the parameters from observation of the actual prepayments. He dis-
cusses the valuation of American option and derives the optimal interest rate
for prepayment, but actually he specifies prepayment as exogenous event by in-
troducing hazard rate that is different according to whether the current interest
rate is above or below the optimal level.
In this study, we reconsider what prepayment cost is. We give the hypoth-
esis that the difference of prepayment actions among mortgagers in the loan
pool is caused by the heterogeneity of their prepayment costs.
Nakagawa-Shouda [23] assumes that the prepayment cost is static, more
exacdy, it is modeled by a time-invariant random variable that follows a para-
metric distribution. However, we found that we can predict the infimum of
transaction costs (or the supremum of the optimal refinance rates correspond-
ing to the cost) for the mortgagers remaining in the loan pool from the historical
path of the interest rate in their model. This implies that prepayment action de-
pends upon only whether the current interest rate is over or below the optimal
rate, but it does not seem realistic.
 A prepayment model of MBS on prepayment cost processes 385

The motivation of our study is to construct a prepayment model that can

move randomly in order to explain actual prepayments in the loan pool more
than the previous studies. More specifically, we model each mortgager's pre-
payment cost as a stochastic processes instead of a time-invariant random vari-
able, and specify the mortgager's prepayment time by the first time when her
or his prepayment cost process falls below zero.
We also suppose that the prepayment cost processes are not directly ob-
servable to the investors. In short, our model is one of structural models with
incomplete information.
It goes without saying that the prepayment cost process corresponds to the
firm value process appeared in some structural approach of default risk liter-
ature. Therefore some studies on default risk modeling help us to model the
prepayment risk.
Default risk modeling based on the structural model with incomplete in-
formation is studied by Duffie-Lando [5], Giesecke-Goldberg [11, 12]. Their
studies interest us much, but they consider only a single firm's default. Since
a big issue of analyzing the MBS loan pool is how to formulate the dependence
among the mortgager's prepayment costs, we needed to extend their discussion
to the multiple default case. It is the main issue of our study.
As for multi-firm default risk model, we refer to CoUin-Dufresne et al. [2].
They presented a default risk model where each firm's hazard rate is dependent
upon an unobservable exogenous binomial state variable and analyzed a port-
folio exposed to credit risk in terms of the posterior probability of the variable
through Bayesian updating scheme. This is a direct hint that we introduce a dis-
tinctive risk that cannot be diversified away. We will call it the loan pool risk
(LPR) because the risk is unique to each loan pool. We notice that the true
value of LPR is different from pool to pool.
Now return to the subject. In this paper, we discuss the conditional dis-
tributions of mortgagers' prepayment times given the investors' filtration. As
a result, we find that the non-prepayment probability (or hazard process) condi-
tioned by both the macro information and the value of LPR plays an important
role in the prepayment analyses. Particularly, the distribution of prepayment
times can be represented in terms of the conditional non-prepayment probabil-
ity and the posterior distribution of LPR.
Moreover we price a pass-through MBS based on the risk-neutral valuation.
Since there are sufficiently many mortgagers in the loan pool, it is natural to
assume that any risk accompanied with each individual mortgager is diversified
away well. Therefore it is required that the market satisfies some asymptotic
no-arbitrage condition. So we give some additional assumptions to the risk
neutral probability measure and derive the pricing formula of a pass-through
MBS under the assumptions.
386 H. Nakagawa, T. Shouda

2. Model

In this section we define a couple of essential components for our prepayment

model. One is the mortgager's prepayment cost. The other is what we will call
the loan pool risk, which measures the tendency of mortgagers' prepayment in
the underlying loan pool. We implicitly suppose that the true value of LPR is
different from pool to pool.

2.1 Prepayment cost

Let {ft,J^,{Tt)t>o,P) be a complete filtered probability spaced Also, let

{T^t)t>o be a subfiltration of {J^t)t>o on this probability space.
Letting N the number of mortgagers in the underlying loan pool, which
is supposed sufficiently large, we introduce a set of R-valued continuous pro-
cesses X / , . . . , X^ and we will call XI the z-th mortgager's prepayment cost
process.
We assume that X^{t), i = 1 , . . . , A^, are (J^t)-adapted, but do not assume
that they are (Wt)-measurable for any t. Also, we assume that for any i =
l,...,A/^andany t > 0,

P{{Xi),<t € BilHt) = P{{Xi)s<t e Bi\Hoo)^ (1)

We do not suppose that {Xl)s<t^ • • •, {X^)s<t are (W^)-conditionally in-

dependent, i.e.
N
P{{Xl)s<teB^,...,{X^)s<teBN\Ht) and 'Y{P{{Xi)s<t e Bi\Ht)

may not be equal for any J 5 i , . . . , ^^v G S(C([0, t]; R)). Here fi(C([0, ^]; R))
denotes the Borel cr-field on the space of continuous functions.
Next, define the i-th mortgager's prepayment time, denoting by r% by

r^ := mi{t > 0|X^^ < 0}.

It follows from the assumption on XI that r* is an (^t)-stopping time, but may

not be (Wt)-stopping time.
We denote l{r^<t} by Nl the z-th mortgager's prepayment indicator pro-
cess.
Now let us define a new filtration {Qt)t>Q by

g,:=Hty(j{Nl',s<t,i = l,...,N},

^ All thefiltrationsin this article are supported to be right-continuous and completed.

 A prepayment model of MBS on prepayment cost processes 387

Then, r^ comes to be a totally inaccessible (^t)-stopping time, while

{X^^l^^i ^ may not be (^t)-adapted or not be (^t)-conditionally indepen-
dent.
As for the three filtrations {Ht C Qt C Tt) presented above, we imply the
following meanings.

• J^t'- the complete information up to time t including variables and processes

that are hidden to MBS investors.
• Qt'. the information known to MBS investors up to time t.
• Ht'- the macro information up to time t including interest rate, or the in-
vestors' information, from which each mortgager's prepayment and private
information is excluded.

We will calculate several conditional probabilities and the conditional expecta-

tions given the investors' filtration {Qt)-

Remark 2.1. The key point of our model is that MBS investors cannot see the
exact value of each mortgager's prepayment cost.
This scheme is quite similar to the setting of incomplete information or
partial information introduced by Duffie-Lando [5]. They formulated a default
risk model in the framework of so-called structural approach but showed the
existence of hazard rate, an essential tool in the reduced-form approach, by
using the fact that the default time of a firm cannot be a predictable stopping
time since the firm value process cannot be perfectly observed.
However we note that our model is essentially different from one-
dimensional model since we treat multi-dimensional dependent processes.

2.2 Loan pool risk

Let us recall that the prepayment cost processes in the loan pool may not be
(St)-conditionally independent.
To specify the mutual dependence of prepayment cost processes, we give
the hypothesis that the prepayment action of the whole mortgager in a sam-
ple loan pool is influenced by some conmion factors, the macro information
(e.g. interest rate) and each mortgager's characteristics. Indeed, it is difficult
to specify the common factors in both theoretical and practical aspects. In this
study we suppose that any other factor than the macro information and each
mortgager's characteristics is integrated into a single time-invariant factor as
below.
Let 7 a ^-measurable random variable taking value in D ( c R).
We do not assume that 7 is Soo-measurable while we assume that 7 is in-
dependent of Hoo' The first condition implies that it is impossible for investors
to observe 7 exactly.
388 H. Nakagawa, T. Shouda

Moreover we suppose that {Xl)s<t," ",{X^)s<t are ((Wt), 7)-condi-

tionally independent, that is, for any 5 i , . . . , B^ G S(C([0, t]; R)),

z=l

(2)
In short, if the value of the random variable 7 as well as the macro information
are given, the dynamics of prepayment cost processes in the loan pool become
independent. We call 7 the loan pool risk and write LPR hereafter for short.
Remark 2.2. Implicitly we suppose that the loan pool which we will analyze is
regarded as a sample pool chosen from a lot of loan pools.
For example, consider the case that there are several loan pools that seem
quite indistinguishable from each other. It is no wonder to assume that the
difference of actual prepayments among the pools is caused by that of some
hidden factors attached to each pool. LPR is viewed as a simple representation
of such hidden factor.
Such an exogenous hidden variable is a similar idea of an unobservable
exogenous binomial state variable proposed by Collin-Dufresne et al. [2], al-
though it is applied to a reduced-form model of credit risk. The Bayesian ap-
proach to parameter updating is more natural to the analysis of prepayment
risks in the mortgage loan pool than multi-firm default risks, because MBS has
a definite time when the pool is originated while credit portfolio does not have
such an origination time. Furthermore, default events are respectively rare and
individual firms' characteristics seem relatively dominant to the other factors.
We think that such a hidden variable model is suitable to the MBS analysis
rather than the analysis of credit portfolio.
We will make it apparent that the information of prepayments in the loan
pool gathers into the posterior distribution of LPR in the prepayment analyses
in the following section.

3. Conditional distribution of prepayment times in the loan

pool

In this section, we discuss the conditional distribution of prepayment times in

the loan pool given the investors' information Qt,

P{r^ > 5 i , . . . , r ^ > SN\Gt) 5i,...,SAr > 0 .

Firstly, we postulate the following.

 A prepayment model of MBS on prepayment cost processes 389

Assumption 3.1. For each i = 1,,.. ,N, there exists a function ^*: [0, oo) x
D xVt —> [0,1] such that for any t G [0, oo), ^*(t, y) is Ht-measurable for
y E D and
r ( t , 7 ) = P(r'>i|Wt,7).
We also assume that ^*(t, ?/) is non-increasing, continuous and right-
differentiable in t and that ^*(^, y) is measurable on D a.s. Furthermore, we
suppose that ^ ' ( 0 , y) = l and ^\t, y) > 0, t > 0.
We remark that "^^{t^y) can be computed, at least numerically, if each
mortgager's prepayment process XI is specified.
Also, let

r{t,y):=-\og¥{t,y),t>0 <=> ¥{t,y) = e-'^'^'^yl

We see that T^{t^y) is well-defined and that r^{t,y) is non-decreasing and

continuous in t under Assumption 3.1. It goes without saying that r*(t, y) cor-
responds to the hazard process.
Secondly, suppose that ^^-conditional distribution of 7, P ( 7 G dy\Qt), has
a regular conditional density ^t{y), namely, we have

ipt[y)dy := P ( 7 e dylGt).

From the above notation, the next result follows.

Theorem3.2. Forsi,... ,SN >0,wehave

N n r N
My)dy

Ht Vt{y)dy.

As a first step of the proof, conditioning by the value of LPR 7, we can see

P(rl>Si,...,T^>S;v|at)

LD
P ( r i > 5 1 , . . . , T ^ > SN\gul = y)P{l e dy\gt).

Then we pay attention to computing P(r^ > 5 i , . . . , r ^ > SN\Qtil — V)-

For the remainder of the proof, please see the original paper [24].
^t{v) can be viewed as the density of the posterior distribution of LPR 7,
so if Ht and N\<^i are observations on Ht and {Nl)s<t respectively, it follows
from Bayes formula that
390 H. Nakagawa, T. Shouda

, , L{y\Hu{Ni<t}i=i...N)My)
my) = —r = . (3)
JD
where

L(2/|#t,W<Ji=i,...,iv)

P{Ht)J{[-^—^\^^^Nl + ¥{t,y){l-Ni)]\_. (4)

-1 = 1 ^Ht

The formula (3) and (4) imply that ^t{y) is obtained by the prior density
of LPR, (po{y), the z-th mortgager's non-prepayment probability ^*(t,2/) and
its partial derivatives in t.

4. Pricing of pass-through MBS

We discuss a risk-neutral valuation of pass-through MBS in this section.
Denote by rt a R-valued (Wt)-adapted process that is regarded as the
default-free instantaneous interest rate observed in the market at time t. Then
we define the risk-free discount factor as

i-i:
A(t,5) = expf — / Tudu

We also denote by Z(t, s) the theoretical price of default-free zero coupon

bond at time t with maturity s for time t < s, and by V{t, T) the theoretical
value of MBS that is written on the whole loan pool at time t. We postulate
that the constant repayment per time denoted by c, the outstanding principal
denoted by p{t) at time t and the maturity common to all the mortgagers by T.
All cash flows generated by the underlying loan pool are paid to MBS investors
per share without any commission and cost. This constraint is not essential.
Hereafter we suppose no arbitrage for the MBS market, so there exists
a risk-neutral probability measure Q such that the price of a pass-through MBS
as well as a default-free zero coupon bond can be represented as the conditional
expectation under Q as follows.
Definition 4.1. Q is a risk-neutral probability measure on (1), T)for the MBS
market ifQ is a probability measure equivalent to P such that for 0 < t < s <
r,
Z(t, s) = E^[A(f, s)\nt] = ^^[A(t, 5)1^,], (5)

V{t,T) = E^ I cA{t,s){N -As)ds-\- I p{s)A{t,s)dA, (6)

 A prepayment model of MBS on prepayment cost processes 391

N
where we set At := 2 ^ Nl,
i=l

Let us define

V'{t,T) := E^\ f cA{t,s){l-Ni)ds-h f p{s)A{t,s)dNi . (7)

V^{t, T) stands for the fair value of the cash flow that i-th mortgager will pay
back with possibility of prepayment from time t to T.
Then the condition (6) in the previous definition connotes V(t^T) =
iV
Y.V%T).
i=l 1
Conditioning by the value of LPR 7, we can see F*(^, T) as

V\t, T)= f V\t, T; y)Q{j G dylGt), (8)

JD

V\t, T; y) = E"^ U cA(t, 5)(1 - N^) ds +j p(s)A{t, s) dN^

(9)

Our direct purpose of this section is to give a useful expression of

F*(t, T; y) with the prepayment hazard process r*(^, y). Before discussing it,
we examine the possibility that mortgagers' private risks can be diversified
away and then give some reasonable assumptions upon the risk-neutral mea-
sure Q.
For 0 < til < ^2 < T, let

Cir{ui,U2) ••= cA{s,Ty\l-Ni)ds+ p{s)\{s,T)-UNi.

Cj^{ui, U2) can be regarded as the total cash flow from i-\h mortgager between
time ui and U2 evaluated at time T, where the received cash is supposed to be
invested in the default-free interest rate r^.
It follows from the property of conditional expectation and variance^ and
the ((Wt), 7)-conditionally independent condition (2) that for any ui, U2 such
that ui<U2< T,

^ Let X be a random variable with ElX"^] < 00 and ^ be a cr-algebra. Then we have
Var(X) = V&r{E[X\J^]) + £;[Var(X|^)],
where Var(X|:r) = ElX^lJ"] - {ElXlJ"])^.
392 H. Nakagawa, T. Shouda

r /I ^ I \"
+ £ Varf-^C|,(ui,U2)WT,7J
L \ ^_2 I /_
/ I ^ \

1 ^
+ Ar2X^^[^^'(*^^("i'''2)l^5^'7)]. (10)
2=1

where Var() stands for the variance under the original probability P.
The first term in the last expression means the systematic risk part with
respect to the macro information and LPR, while the second term, which means
unsystematic risk part, converges to zero as N goes to 00 since we can suppose
that there exists a constant C (independent of the number N of mortgagers)
such that for any i = 1 , . . . , AT,

£;[Var(C^K,^2)|WT,7])]<C-
This observation implies that any private risk of cash flow during any term
[1*1,^x2] (C [0, T]) is possible to diversify away in the loan pool consisting of
sufficiently large numbers of mortgagers. The equivalence between P and Q
implies that

/ 1 ^
lim Var U2) W T , 7 Q-a.s.
iv-^00 ^ i=l

In other words, the measure Q inherits from P the property that any individ-
ual risk of cash flow during any term between time t and T asymptotically
vanishes.
So we first give the following assumption upon the probability measure Q
on the basis of the above note.

Assumption 4.2. {Xl)s<t, • • •, {^t^)s<t cire also {{Ht)^^)-conditionally in-

dependent under Q, namely, for any ^ 1 , . . . , B^ ^ S(C([0, t]\ R)),

i=l

Under this assumption, we can directly demonstrate that private risks van-
ish asymptotically under Q as we see in (10). This condition is also a bit strong.
 A prepayment model of MBS on prepayment cost processes 393

but it seems natural that if there exists no special mortgager whose prepayment
action influence other mortgagers' prepayment under the original measure P ,
it also holds under the risk-neutral measure Q.
Since private risks asymptotically vanish as we see above, we should also
consider the possibility of asymptotic arbitrage. Indeed, we notice that an
asymptotic arbitrage-free condition holds as follows.
Let us consider a Hnear combination of Cj^{ui, U2) such that

N N
C^:=S"w'CJr{uuU2) s.t. lim Vi(;^£^[CJ.(t/i,U2)|HT,7] = 0 a.s.
i=\ i=l
(11)
for some bounded sequence {^^}i=i,2,... satisfying that

N N
lim y^lit;*! converges and lim y^(tf;M^ = 0.

We call Cj^ the asymptotic hedged position from now on.

Any asymptotic hedged position fulfills

^lim^ V a r ( ' ^ ^ ^ C ^ ( t ^ i , U 2 ) ) = ^Im^ Var('X^^^J5;[C^(txi,U2)|WT,7])

since unsystematic risks also vanish asymptotically due to the calculation sim-
ilar to (10).
Apparently, the initial fair price of the asymptotic hedged position is
given by E^[K{0,T)C^]. Therefore, if the asymptotic arbitrage-free condi-
tion holds, the following equality should be valid:

r N -|
lim E^ K{id,T)C^ lim E^
L 2 = 1 -I

0. (12)

This means that the asymptotic hedged position should be constituted with zero
initial cost. In order to hold equation (12), we add another assumption on the
measure Q as follows.

Assumption 4.3. For every i = 1 , . . . , iV, t G [0, T], a n J P € S(C([0, t]; R)),

Q{{xi)s<t e B\ntn = y) = P{{xi)s<t e B\ntn = y)- (13)

394 H. Nakagawa, T. Shouda

This assumption means that no diversifiable risk has any risk premium,
even if we indeed try to evaluate the loan from mortgager to mortgager. Al-
though the assumption is restrictive, it seems to make sense as far as we see
a large pool of mortgagers.
Under Assumption 4.3, for any linear combination satisfying the prop-
erty (11), we can see that
r N
lim E^ A(0,T)C^1 = lim E^\A{0,T)yw^E'^[C^{uuU2)\nTn]
J TV—^•oo ^-^
2=1

N
lim E*^ A(0,r)^^^E[C^(ui,U2)|WT,7]
i=l
= 0,

which concludes (12). The second equality follows from the condition (13) and
the dominated convergence argument leads to the third equality.
Now we can show our main theorem for MBS pricing under these assump-
tions^.

Theorem 4.4. Suppose Assumptions 3.1, 4.2 and 4.3 are satisfied.
Then F*(t, T; y) defined by (9) is represented as

V\t,T;y)

= (1 - Ni)E'^ /"A(i,s)e-^^'(^'^)-^*(*'^)>[cds+p(s)dr'(s,t/)] Ht (14)

For the proof of the theorem, we need to show that if 7 = y,

1-Ni)dr{u,y)
Jo

is a (Q, {Ht V a{Ni; s < ^}))-martingale.

After showing the property, we apply the proof of Propositions A.1-A.3 in
Nakagawa-Shouda [23] (or the results in section 5.1 of Bielecki-Rutkowski [1])
to obtain the equality (14).
We would like the readers to see the original paper [24] for the complete
proof.

^ In fact, it is the best to show the MBS pricing formula from the minimum set of
assumptions for asymptotic no-arbitrage condition, but the following theorem does
not result directly from it. Anyway we remark that the following pricing formula of
MBS is consistent with the asymptotic arbitrage-free situation.
 A prepayment model of MBS on prepayment cost processes 395

5. Conclusion
In order to study the prepayment mechanism and the risk-neutral valuation
of MBS from a view of MBS investors, we introduce the prepayment cost
processes that determine each mortgager's prepayment time and the systematic
risk of MBS, named the loan pool risk (LPR), which is motivated by some
literature on default risk modeling.
We assume that the prepayment cost processes are conditionally indepen-
dent when the macro information and the value of LPR are given. Under the as-
sumption, we show that the conditional distribution of prepayment times given
the investors' filtration is represented in terms of each mortgager's conditional
non-prepayment probability (or hazard process) and the posterior density of
LPR.
As for MBS pricing, we examine that any idiosyncratic risk in the loan
pool is asymptotically diversified away and that the MBS market satisfies an
asymptotic no-arbitrage condition. Then we achieve the fair value of MBS in
terms of the posterior distribution of LPR under the risk-neutral measure and
the hazard process under the original probability measure.

References

[I] Bielecki, T.R., Rutkowski, M.: Credit Risk: Modeling, Valuation and Hedging.
Springer Finance , Springer-Verlag 2001
[2] Collin-Dufresne, P., Goldstein, R.S., Helwege, J.: Is credit eventriskpriced? Mod-
eling contagion via the updating of beliefs. Working paper (2003)
[3] Deng, Y., Quigley, J.: Woodhead behavior and the pricing of residential mortgages.
Berkeley Program on Housing and Urban Policy. Working paper Series WOO-004
(2001)
[4] Downing, C, Stanton, R., Wallace, N.: An empirical test of a two-factor mortgage
valuation model: How much do house prices matter? Working paper (2003)
[5] Duffie, D., Lando, D.: Term structure of credit spreads with incomplete accounting
information. Econometrica 69, 633-664 (2001)
[6] Dunn, K.B., McConnell, J.J.: A comparison of alternative models for pricing
GNMA mortgage-backed securities. Journal of Finance 36,471-483 (1981a)
[7] Dunn, K.B., McConnell, J.J.: Valuation of mortgage-backed securities. Journal of
Finance 36, 599-617 (1981b)
[8] Dunn, K.B., Spatt, C.S.: The effect of refinancing costs and market imperfections
on optimal call strategy and the pricing of debt contracts. Working paper: Carnegie-
Mellon University (1986)
[9] Durrett, R.: Probability, Theory and Examples 2nd. ed. Duxbury Press 1995
[10] Frey, R., McNeil, A.J.: Dependent defaults in models of portfolio creditrisk.Jour-
nal of Risk 6, 59-92 (2003)
[II] Giesecke, K., Goldberg, L.R.: Forecasting default in the face of uncertainty. To
appear in Journal of Derivatives (2004)
396 H. Nakagawa, T. Shouda

[12] Giesecke, K., Goldberg, L.R.: The market price of credit risk. Working paper
(2003)
[13] Goncharov, Y.: An intensity-based approach for valuation of mortgage contracts
subject to prepayment risk. Working paper (2002)
[14] Ichijo, H.: Prepayment rate model of mortgage that unites structural approach
and reduced-form approach. Proceedings of the 19th JAFEE Conference, 188-207
(2003) (written in Japanese)
[15] Ichijo, H., Moridaira, S.: Prepayment analyses of mortgage. Proceedings of the
15th JAFEE Conference, 221-239 (2001) (written in Japanese)
[16] Jarrow, R.A., Lando, D., Yu, R: Default risk and diversification: Theory and em-
pirical implications. Mathematical Finance 15,1-26 (2005)
[17] Jeanblanc, M., Rutkowski, M.: Modelling of default risk: Mathematical tools.
Working paper (2000)
[18] Karatzas, I., Shreve, S.E.: Brownian Motion and Stochastic Calculus. Springer-
Verlag, New York 1991
[19] Kariya, T., Kobayashi, M.: Pricing mortgage-backed securities (MBS). Asia-
Pacific Financial Markets 7, 189-204 (2000)
[20] Kariya, T., Pliska, S.R., Ushiyama, R: A 3-factor valuation model for mortgage-
backed securities (MBS). Working paper (2002)
[21] Lyuu, YD.: Financial Engineering and Computation. Cambridge 2002
[22] Mcconnell, J.J., Singh, M.: Rational prepayment and the valuation of collateral-
ized mortgage obligations. Journal of Finance 49, 891-921 (1994)
[23] Nakagawa, N., Shouda, T.: Valuation of mortgage-backed securities based on un-
observable prepayment costs. Advances in Mathematical Economics 6, 123-147
(2004)
[24] Nakagawa, N., Shouda, T.: Analyses of mortgage-backed securities based on un-
observable prepayment cost processes. Submitted (2005)
[25] Nakamura, N.: Valuation of mortgage-backed securities based upon a structural
approach. Asia-Pacific Financial Markets 8, 259-289 (2001)
[26] Nakamura, N., Shibasaki, K.: Valuation of mortgage-backed securities based upon
a hazard rate approach. The 15th JAFEE Conference Proceedings (Japanese) (2001)
[27] Popova, E., Popova, I., George, E.: Bayesian modeling of mortgage prepayment
rates. Working paper (2003)
[28] Protter, E.P.: Stochastic Integration and Differential Equations 2nd ed. Springer
2004
[29] Richard, S.F., Roll. R.: Prepayments on fixed-rate mortgage-backed securities.
Journal of Portfolio Management 15, 73-82 (1989)
[30] Schwartz, E.S., Torous, W.N.: Prepayment and the valuation of mortgage-backed
securities. Journal of Finance 44, 375-392 (1989)
[31 ] Stanton, R.: Rational prepayment and the valuation of mortgage-backed securities.
Review of Financial Studies 8, 677-708 (1995)
[32] Sugimura, T.: A prepayment model for the Japanese mortgage loan market:
Prepayment-type-specific parametric model approach. Asia-Pacific Financial Mar-
kets, 305-335 (2002)