The term structure of CD rates and monetary policy transmission

Yasuo Nishiyama
*
School of Business, Woodbury University, Burbank, CA 91510, United States
a r t i c l e i n f o
Article history:
Received 17 September 2009
Accepted 16 July 2010
Available online 22 July 2010
JEL classication:
E43
G21
Keywords:
Term structure of interest rates
Certicates of deposit
Income smoothing
Monetary policy transmission
a b s t r a c t
The paper investigates the term structure of CD rates and its relationship with the federal funds rate or
monetary policy. The term structure derived in this paper is governed primarily by the federal funds rate
and secondarily by banks income smoothing behavior. It is consistent with the estimation results and
differs signicantly from the standard term structure of interest rates. The downturn phase of business
cycles appears to be accompanied by more aggressive income smoothing by banks (compared with the
upturn phase) due to their pessimistic expectations of future prots. The compositional shift in banks
liabilities during the downturn phase away from CDs toward transaction deposits may pose a greater
withdrawal risk for banks.
2010 Elsevier B.V. All rights reserved.
1. Introduction
The term structure of interest rates has been extensively stud-
ied (see, for example, Shiller, 1990, and references therein;
Gmez-Valle and Martnez-Rodrguez, 2008; Jiang and Yan,
2009). The empirical literature on this topic typically investigates
Treasury securities (bond yields), and the term structure is com-
monly interpreted from the investors perspective. For example,
the term structure is often expressed by equating the long-term
rate to the average of current and expected future short-term rates
plus an extra yield. The extra yield (or a term premium) is inter-
preted as compensation for the interest rate risk that investors
bear from holding long-term bonds instead of short-term bonds.
In this paper, I investigate the term structure of interest rates on
banks certicates of deposit (CDs), which, to my knowledge, has
never been studied. In addition, I investigate (monetary policy)
interest rate pass-through based on the theory of the term struc-
ture of CD rates developed in this paper. More specically, the pri-
mary objective is to derive clearly interpretable components that
constitute a term premium associated with CDs. These compo-
nents are shown to be consistent with observed data, but they rep-
resent the perspective of the suppliers (banks) instead of the
perspective of buyers (investors). The secondary objective is to
estimate the pass-through from the monetary policy target, or
the federal funds rate, to CD rates. Asymmetric characteristics of
pass-through over business cycles are explained based on the pa-
pers theory and empirical evidence of the term structure of CD
rates.
The secondary objective is related to Hofmann and Mizen
(2004) and other recent studies (for example, Liu et al., 2008) in
that short-term rates vis--vis monetary policy are investigated.
Generally, short-term money market rates (including CD rates)
are considered unimportant for the efcacy of monetary policy,
and their pass-through is presumed to be rather trivial. What mat-
ters for monetary policy is the pass-through from the FF rate to
long-term interest rates (Bernanke, 2004). The study by Hofmann
and Mizen (2004), which is especially relevant to the present re-
search for a reason explained below, carefully investigates the pre-
sumption of trivial/complete pass-through of the average of UK
banks base rates (which moves closely with the ofcial rate) to
short-term retail bank rates (90-day deposit account rates). Addi-
tionally, standard variable-rate mortgages are also investigated.
Using a non-linear error correction model (ECM), they uncover rich
complexities or non-linear adjustments of these retail rates to the
average base rate. Specically, when deviations of retail rates from
the average base rate are widening (or expected to widen), the loss
resulting from not adjusting retail rates becomes greater than the
menu cost (the cost of resetting retail rates), thereby forcing
banks to adjust more quickly. Conversely, when deviations are nar-
rowing, banks have less incentive to adjust, resulting in slower or
no change in the adjustment speed. Some differences aside (i.e.,
the UK data of Hofmann and Mizen, 2004, versus the US data of this
paper), considering Hofmann and Mizens (2004) ndings in this
0378-4266/$ - see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jbankn.2010.07.024
* Tel.: +1 818 394 3301; fax: +1 818 394 3311.
E-mail address: yasuo.nishiyama@woodbury.edu
Journal of Banking & Finance 35 (2011) 8294
Contents lists available at ScienceDirect
Journal of Banking & Finance
j our nal homepage: www. el sevi er . com/ l ocat e/ j bf
paper yields additional insights into adjustment of CD rates to the
federal funds rate (explained in Section 6).
The remainder of the paper is organized as follows. In Section 2,
the data used in this paper are described, and stylized facts (three
observations) about CD rates vis--vis the federal funds rate are
advanced. In Section 3, I derive the optimal CD pricing conditions
from an intertemporal bank model. In Section 4, model calibration
illustrates the consistency of each determinant of the CD rates in
isolation with stylized facts. The term structure of CD rates is also
summarized. In Section 5, empirical support for the term structure
of CD rates is presented. Section 6 provides the pass-through esti-
mation results, which are interpreted based on the term structure
of CD rates. Section 7 concludes.
2. Data and three stylized facts
Except for Fig. 3, all of the data used in this paper come from the
Federal Reserve Bank of St. Louis Economic Data FRED II. The
data for Fig. 3 come from the Federal Reserve Bulletins from June
1996, June 2003, and July 2007 (Federal Reserve Board). For CD
rates, the paper uses secondary market rates or averages of dealer
rates on nationally traded CDs (Federal Reserve Statistical Release
H.15). CDs traded in the secondary market (commonly called
negotiable, jumbo, or wholesale CDs) were originally issued
primarily by large banks and purchased directly by corporate trea-
surers, investment companies, banks and others who then sell in
the secondary market. Alternatively, large banks may issue these
CDs indirectly through dealers and brokers. Their denominations
are $100,000 or more. Because CDs were insured up to $100,000
per investor (until October 3, 2008, when the insured amount
was raised to $250,000), banks and brokers have sometimes pack-
aged CDs in $100,000 increments so that the entire amount is fully
insured (Koch and MacDonald, 2003, pp. 435436). For large
banks prots (denoted by p below), the return on assets (ROA)
of large US banks (or banks with average assets of greater than
$15 billion) is used instead of net income because ROA takes into
account mergers, acquisitions, and/or divestitures, while net in-
come does not.
A: Monthly sample, 1987:2 2007:5. The vertical line indicates February 1994.
0
2
4
6
8
10
12
88 90 92 94 96 98 00 02 04 06
____Federal funds rate _ _ _ _Three-month CD rate __ __Six-month CD rate
B: Monthly sample, 1992:1 2001:6. The vertical line indicates February 1994.
2
3
4
5
6
7
92 93 94 95 96 97 98 99 00
____Federal funds rate _ _ _ _Three-month CD rate __ __Six-month CD rate
C: Monthly sample, 1987:2 2007:5. The horizontal line indicates the sample mean of i
c3,t
i
f,t
= 0.1396.
The sample mean of i
c6,t
i
c3,t
= 0.0944 is not shown. (i
c3,t
is the three-month CD rate; i
f,t
is the federal
funds rate; and i
c6,t
is the six-month CD rate.)
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
88 90 92 94 96 98 00 02 04 06
____Spread = i
c3,t
i
f,t
-----Spread = i
c6,t
i
c3,t
Fig. 1. The federal funds rate and the secondary market three- and six-month CD rates.
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 83
The papers objectives are explained clearly in Fig. 1, which
shows the federal funds rate (the FF rate hereafter), the three-
month certicate of deposit rate (the CD3 rate hereafter), and the
six-month certicate of deposit rate (the CD6 rate hereafter). The
sample contains monthly data for the period of 1987:22007:5 be-
cause part of this paper is related to the pass-through of the FF rate
and the fact that the Fed (Federal Reserve) abandoned M1 targets
in February 1987 and gradually shifted to targeting the FF rate in
the following few years. When all rates are plotted over the entire
sample period (Fig. 1A), it is difcult to characterize any discernible
differences between the rates. Differences become clear, however,
in Fig. 1B, which shows a subsample covering the 1992:12001:6
period. Fig. 1C shows the spread between the CD3 and FF rates (re-
lated to monetary policy pass-through) and the spread between
the CD6 and CD3 rates (related to the term structure of CD rates).
Put simply, the papers objectives are to answer the following
questions: What are the underlying determinants of the CD rates
movements relative to those of the FF rate (Fig. 1B)? How do the
CD rates and the term structure of CD rates adjust to changes in
monetary policy or to changes in the FF rate (Fig. 1C)?
Three observations are made. In Fig. 1B, when a decline/rise in
the FF rate is correctly anticipated, the CD rates appear to de-
cline/rise ahead of the FF rate. In addition, the longer-term CD rate
(the CD6 rate) tends to decline/rise ahead of the shorter-term CD
rate (the CD3 rate). That is, the CD rates exhibit a leftward shift
away from the FF rate (Observation 1). This observation is more
apparent after February 1994, when the Fed started announcing FF
rate target changes instead of keeping them secret as it did prior to
1994:2. In Fig. 1B, the two CD rates tend to be higher than the FF
rate; that is, they tend to exhibit an upward shift away from the
FF rate (Observation 2). In Fig. 1C, the positive spread (or positive
equilibrium errors) between the CD3 and FF rates tends to be cor-
rected abruptly, whereas the negative spread tends to persist, a
hint of possibly asymmetric pass-through (Observation 3). The
papers answers to these observations are as follows. The CD rates
Granger cause the FF rate after February 1994, or the current and
expected future FF rates are the primary determinant of the CD
rates (Observation 1). Risk-averse banks are time impatient and
engage in income smoothing, both of which are secondary deter-
minants of the CD rates (Observation 2). Equilibrium error correc-
tions are asymmetric (Observation 3). Two implications arising
from these answers are that the pure expectations hypothesis does
not hold for the CD rates and the term structure of CD rates differs
signicantly from the standard description of the term structure of
interest rates associated with Treasury securities.
Table 1 shows descriptive statistics that appear to bear out two
observations above. Denoting the CD3, CD6 and FF rates by i
c3,t
, i
c6,t
,
and i
f,t
, respectively, i
c3,t3
i
f,t
is on average smaller than i
c3,t
i
f,t
for the 1994:22007:5 period, but not for the 1987:21994:1 per-
iod. It suggests that, after February 1994, i
c3,t3
follows (or antici-
pates) i
f,t
more closely than i
c3,t
. Also, i
c6,t3
i
c3,t
can be
interpreted similarly. These results are consistent with Observa-
tion 1. The CD rates are on average higher than the FF rate, or
Observation 2.
Finally, two clarications regarding the CD rate data are war-
ranted. First, the paper derives the optimal CD rates from a repre-
sentative banks prot-maximizing model. Then, the theoretically
derived CD rates are used to explain the observed behaviors of
the secondary market CD rates (Observations 1, 2 and 3). In the
practical context, the theoretical CD rates represent the rates on
newly issued CDs or the primary market CD rates. Unfortunately,
time-series data on primary market CD rates of individual banks/
aggregates by maturity are not publicly available, so the paper uses
secondary market CD yield data (available from the Federal Re-
serve). The seeming mismatch, however, does not likely pose any
serious problems for the estimation for the following reason. In or-
der for newly issued CDs to coexist with previously issued CDs
traded in the well-functioning secondary market, the rate on newly
issued CDs (determined by the model in this paper) should be, in
equilibrium, approximately equal to the prevailing secondary mar-
ket yield (data used in this paper). Previously issued CDs traded in
the secondary market, as xed-income securities, are priced just
like previously issued coupon bonds, and their trades are settled
at market value plus promised interest (Koch and MacDonald,
2003). Hence, they reect both promised interest and an expected
price appreciation/depreciation (that correspond to the current
yield and an expected capital gain/loss, respectively, in the case
of coupon bonds). The latter (price appreciation/depreciation)
should counterbalance any differences in promised interest be-
tween previously issued (secondary market) and newly issued
CDs. Indeed, Slovin and Sushka (1979) show that the secondary
market yield on CDs closely follows the optimum CD rate deter-
mined by banks (i.e., the primary market rate on CDs). The terms
CD rate and CD yield are used interchangeably in this paper.
Second, the empirical results in Sections 5.1 and 6 are based on
monthly data, whereas those in Sections 5.2 and 5.3 are based on
quarterly data. Quarterly data on the CD and FF (and other) rates
used in Sections 5.2 and 5.3 are simple averages of their respective
monthly data. While end-of-quarter (daily) rate data could be used
instead of simple averages, potential end-of-quarter volatility of
the FF rate makes it an inappropriate choice. Indeed, closer exam-
ination conrmed this concern (available from the author upon
request).
3. Model
The intertemporal bank model in this paper assumes that
banks loans and CDs are differentiated products, and banks engage
in monopolistic competition. This assumption follows Klein (1971),
Prisman et al. (1986), and others, and in particular Hannan and
Berger (1991) with respect to differentiated bank products. A rep-
resentative bank facing a downward-sloping demand curve in the
loan market and an upward-sloping supply curve in the CD market
exercises some market power in setting its loan and CD rates. Gi-
ven the exclusive focus on CDs in this paper, the assumption is con-
sistent with the well-known reality that banks indeed offer a
variety of different CD rates for the same maturity (which can be
easily veried at www.bankrate.com).
Table 1
Descriptive statistics, monthly sample, 1987:22007:5.
Variable Mean Std. dev. Minimum Maximum
Monthly 1987:21994:1 (number of observations = 84)
i
f,t
a
6.213 2.237 2.920 9.850
i
c3,t
b
6.326 2.192 3.090 10.090
i
c6,t
c
6.413 2.185 3.160 10.400
i
c3,t
i
f,t
0.113 0.219 0.480 0.890
i
c6,t
i
c3,t
0.086 0.117 0.200 0.370
i
c3,t3
i
f,t
0.207 0.566 0.880 1.780
i
c6,t3
i
c3,t
0.176 0.597 1.150 1.430
Monthly 1994:22007:5 (number of observations = 160)
i
f,t
a
4.162 1.753 0.980 6.540
i
c3,t
b
4.315 1.782 1.040 6.730
i
c6,t
c
4.414 1.788 1.020 6.940
i
c3,t
i
f,t
0.153 0.207 0.420 0.930
i
c6,t
i
c3,t
0.098 0.137 0.220 0.490
i
c3,t3
i
f,t
0.115 0.332 0.580 1.390
i
c6,t3
i
c3,t
0.061 0.402 0.890 1.460
All statistics are measured in percentage points.
a
The federal funds rate.
b
The three-month CD rate.
c
The six-month CD rate.
84 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294
The representative banks balance sheet at the end of period t is
shown below. (Although other entries such as money market de-
posit accounts can be added to make the balance sheet more
realistic, they are unnecessary for the objectives of this paper,
and their omission does not affect the results.)
Assets
L
m,tj
m-period loans, booked at t j, with interest paid in
each period (from t j + 1 on) and repaid at t j + m;
m = 1, 2, . . . , M; j = 0, 1, . . . , M 1 (m > j);
i
m,tj
= interest rate (determined by the bank)
FS
t
federal funds sold; i
f,t
= federal funds rate (given
exogenously)
B
t
government securities; i
b,t
= yield (given exogenously)
R
t
rD
t
(total required reserves), where r denotes the
reserve requirement ratio (D
t
below)
Liabilities and equity capital
D
t
non-interest-bearing transaction deposits, given
exogenously
D
cn,tj
n-period CDs (certicates of deposit) issued at t j,
with interest paid in each period (from t j + 1 on),
and mature at t j + n; n = 1, 2, . . . , N;
j = 0, 1, . . . , N 1 (n > j); i
cn,tj
= interest rate
(determined by the bank)
FP
t
federal funds purchased; i
f,t
= federal funds rate (given
exogenously)
K
t
equity capital, given exogenously
The demand for loans (L
m,tj
) depends negatively on the interest
rate (i
m,tj
) and positively on the level of economic activity, with
the latter given exogenously. To simplify, I assume that the bank
holds government securities only to manage liquidity, justifying
B
t
= B (constant). For each CD, I assume a simple constant-elasticity
supply function of the following form (omitting the subscripts):
D = ai
e
where D and i denote the supply of CDs and the interest rate,
respectively; a is a constant (a > 0); and e denotes the constant elas-
ticity (e > 0). This assumption has been criticized on theoretical
grounds by Hutchison (1995). He shows that the supply function
(which is called the demand function in Hutchison, 1995), and
hence the resulting elasticity, is a function of both the own interest
rate and the rate on a competing instrument. The criticism calls for
a more general supply function, for example, D(i, i
s
), where i
s
de-
notes the interest rate on a substitute in investors portfolios, and
@D/@i > 0 and @D/@i
s
< 0. Consequently, the elasticity is in general
a function of both i and i
s
, not a constant. I empirically test this crit-
icism below.
The balance sheet constraint is expressed by:
FS
t
FP
t
= (1 r)D
t

P
N
n=1
P
n1
j=0
D
cn;tj
K
t

P
M
m=1
P
m1
j=0
L
m;tj
B,
where FS
t
FP
t
> 0 (<0) indicates the banks net excess reserves
(net reserves shortage) that are lent (borrowed) in the federal
funds market. The banks prot during period t is expressed by:
p
t
=
X
M
m=1
X
m
j=1
i
m;tj
L
m;tj
i
f ;t
(FS
t
FP
t
) i
b;t
B

X
N
n=1
X
n
j=1
i
cn;tj
D
cn;tj
C
X
M
m=1
X
m1
j=0
L
m;tj
; D
t
;
X
N
n=1
X
n1
j=0
D
cn;tj
!
FC;
where C(, , ) is a noninterest cost function with @C/@L
m,tj
> 0, @C/
@D
t
> 0, and c = @C/@D
cn,tj
> 0; and FC = xed cost. The FF rate is the
source of uncertainty for the bank in the model. Following Hannan
and Berger (1991), I assume that the marginal cost (MC) of CDs (c
above) is constant.
One important issue that needs to be addressed is whether a
bank or a bank manager is a utility maximizer or UMAX (i.e., the
bank is risk-averse), or a prot maximizer or PMAX (i.e., the bank
is risk-neutral). In theory, the Fisher separation theorem demon-
strates that, under the assumption of perfect capital markets, it is
optimal for a banks shareholders that the bank manager makes
nancial decisions that maximize the wealth of shareholders (or
current and future prots) regardless of heterogeneous sharehold-
ers utility functions. That is, the bank (or the bank manager) is a
risk-neutral prot maximizer. Empirically, Edwards (1977) shows
that agency problems exist in the banking industry and that utility
maximization better describes the banks (or bank managers)
behavior. Ratti (1980) and Angelini (2000) show evidence that
banks are risk-averse. Many other researchers assume the maximi-
zation of expected utility of prot (or wealth) for the nancial
intermediary (for example, Sealey, 1980). Following these authors,
this paper species that a representative bank maximizes its ex-
pected utility of prot. However, risk-neutral prot maximization
(PMAX) is simply a special case of risk-averse utility maximization
(UMAX) and is examined and tested together with UMAX below.
Subject to the balance sheet constraint, the bank maximizes dis-
counted expected utility, E
t
P

s=t
b
st
u(p
s
), with respect to timet
loan and deposit rates, where u(p
s
) denotes the concave utility
function, p
s
denotes periods prot, b denotes the subjective
time-preference factor (presumably 0 < b < 1), and u
/
(p
s
) > 0 and
u
//
(p
s
) < 0 are assumed. u(p
s
) = p
s
in the PMAX case. Then, the rele-
vant rst-order conditions are:
n = 1 : i
c1;t
= (1 e
1
)
1
E
t
(M
t1
)
1
(i
f ;t
c); (1)
n _ 2 : i
cn;t
= (1 e
1
)
1
X
n
h=1
E
t
Y
h
k=1
M
tk
! " #
1
(i
f ;t
c)
X
n1
h=1
E
t
Y
h
k=1
M
tk
!
(i
f ;th
c)
" # ( )
; (2)
where M
t+1
denotes the intertemporal marginal rate of substitution
of present (time t) for future (time t + 1) prot, M
t+1
= bu(p
t+1
)/
u
/
(p
t
) (=b in the PMAX case), and similarly, M
t+k
= bu
/
(p
t+k
)/u
/
(p
t+k1
)
(=b in the PMAX case); and e denotes the elasticity, which is as-
sumed to be the same for all maturities. I assume i
f,t
> c.
4. Model calibration
4.1. Theoretical explanation for Observation 1 and Observation 2
The purpose of this section is to illustrate how each factor
embodied in Eqs. (1) and (2) affects the CD rates in isolation, there-
by possibly explaining Observation 1 or Observation 2. For the pur-
pose of a simpler and clearer illustration, e = , c = 0, and perfect
foresight (E
t
(x
t+k
) = x
t+k
, k = 1, 2, . . .) are assumed. For the UMAX
case, I assume the CRRA (constant relative risk aversion) utility,
which is commonly used in the literature (Mehra and Prescott,
1985; Feldstein and Ranguelova, 2001; and others):
u(p
t
) = (p
1c
t
1)=(1 c), where c = coefcient of relative risk
aversion (RRA). Fig. 2 shows model calibration based on an as-
sumed exogenous time-path of the FF rate over 50 periods 4% ini-
tially, then rising to 6%, and then falling to 4%. In Fig. 2, one
period = one month. The three-month and six-month CD rates
are calculated by Eq. (2), where n = 3 and n = 6, respectively.
Fig. 2A assumes b = 1 and c = 0 (PMAX case). It is consistent
with the leftward shift, or Observation 1. This case represents the
pure expectations hypothesis of the term structure of interest rates
(PEH) that is applied to bank CD rates.
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 85
Fig. 2B assumes b = 0.95 (time impatience) and c = 0 (PMAX
case). Consider two-period CDs. The optimal two-period CD rate
under the PMAX case is:
i
c2;t
= (1 e
1
)
1
(b b
2
)
1
[(i
f ;t
c) bE
t
(i
f ;t1
c)[
= (b b
2
)
1
[bi
c1;t
b
2
E
t
(i
c1;t1
)[: (3)
Eq. (3) is equivalent to the following equation (by taking logs and
using the linear approximation):
(1 i
c2;t
)
bb
2
= (1 i
c1;t
)
b
(1 E
t
(i
c1;t1
))
b
2
: (4)
Eq. (4) indicates that the bank prices CDs based on b-time com-
pounding (b < 1), not calendar-time compounding (b = 1 or the
PEH case). b < 1 means that the bank values todays dollar higher
than a dollar in the future under PMXA, or the bank values a unit
of todays felicity higher than a unit of felicity in the future under
UMAX. This time impatient behavior (b < 1) results in higher CD
rates than calendar-time CD rates (of Fig. 2A). The reason is as fol-
lows. Consider two banks, Bank A with b = 1.0 and Bank B with
b = 0.95. Assume that the two banks face similar loan demand and
balance sheet constraints. In order to nance todays loan demand,
Bank B raises more funds through CDs (by offering higher CD rates)
and therefore less funds through borrowing (e.g., federal funds)
than Bank A. Because the interest costs of CDs fall in the future,
while those of borrowing fall today, Bank Bs prot today is rela-
tively higher than Bank As. It is consistent with the notion of time
impatience: Bank B values todays prot higher than its prots in
the future. This upward shift of CD rates in Fig. 2B is consistent with
Observation 2.
Fig. 2C assumes b = 1 and c = 1 (UMAX case). In addition, it as-
sumes that the bank prot grows by 5% each period (for
t = 1, 2, . . . , 25) and declines by 5% each period (for
t = 26, 27, . . . , 50); that is, the bank prot is assumed to be procyc-
lical because the FF rate is known to be procyclical (Stock and Wat-
son, 1999). (Note that M
tk
= bg
c
tk
, where g
t+k
= p
t+k
/p
t+k1
is the
growth rate of prot.) In a tree economy la Lucas (1978), the
concavity of the utility function induces consumption smoothing
(Gollier, 2001, Chapter 17). When the economy is growing, agents
A: Assuming perfect foresight, = 1.0, = 0.0 (risk-neutral), g
t+1
=
t+1
/
t
= 1.0 for all t.
3.6
4.0
4.4
4.8
5.2
5.6
6.0
6.4
5 10 15 20 25 30 35 40 45 50
____Federal funds rate _ _ _ _Three-month CD rate __ __Six-month CD rate
B: Assuming perfect foresight, = 0.95 (time impatience), = 0.0, g
t+1
= 1.0 for all t.
3.5
4.0
4.5
5.0
5.5
6.0
6.5
5 10 15 20 25 30 35 40 45 50
____Federal funds rate _ _ _ _Three-month CD rate __ __Six-month CD rate
C: Assuming perfect foresight, = 1.0, = 1.0 (risk-averse or income smoothing), g
t+1
= 1.05 for t = 1,,
25 and g
t+1
= 0.95 for t = 26,, 50 (procyclical profit growth). The vertical line indicates t = 26 where
profit growth changes.
3.5
4.0
4.5
5.0
5.5
6.0
6.5
5 10 15 20 25 30 35 40 45 50
____Federal funds rate _ _ _ _Three-month CD rate __ __Six-month CD rate
Fig. 2. Model calibration.
86 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294
want to borrow today (as they expect larger revenues in the future)
to smooth consumption over time, resulting in the higher equilib-
rium interest rate (Gollier, 2001, Chapter 17). Analogously, the
concavity of the utility function in the bank model of this paper in-
duces income smoothing. (Income smoothing by banks has been
extensively studied in the accounting literature. See, for example,
Liu and Ryan, 2006, and references therein. Following the account-
ing literature, this paper uses the term income smoothing
instead of prot smoothing, though the latter is a more appropri-
ate term in economics.) When the prot is expected to grow, the
bank raises its CD rates. With higher CD rates (compared with
Fig. 2A), the bank is funding todays loan demand more by CDs
and therefore less by borrowing. The interest costs of these CDs
(borrowing) fall in the future (fall today), thereby reducing
(increasing) future higher prot (todays lower prot) and smooth-
ing its prots over time. The explanation is symmetrical when the
prot is declining. Income smoothing is consistent with Observa-
tion 2.
4.2. Additional factor: Too-big-to-fail (TBTF)
It is widely recognized that large prime-name banks can offer
lower CD rates because of their lower default risk and greater mar-
ketability. This prime-name factor, because it is often associated
with the too-big-to-fail (TBTF) guarantee of these banks liabilities
by Federal regulators, is called TBTF in this paper and can be ex-
plained by the model. Consider the two-period CD rate:
i
c2;t
= (1 e
1
)
1
H
1
t
(i
f ;t
c) E
t
[M
t1
(i
f ;t1
c)[
= (1 e
1
)
1
H
1
t
[(i
f ;t
c) E
t
(M
t1
)E
t
(i
f ;t1
c)[
(1 e
1
)
1
H
1
t
cov
t
(M
t1
; i
f ;t1
); (5)
where H
t
= E
t
(M
t+1
+ M
t+1
M
t+2
); and cov
t
(M
t+1,
i
f,t+1
) = (b/u
/
(p
t
))cov
t
(u
/
(p
t+1
), i
f,t+1
). Note that the FF rate moves procyclically (Stock
and Watson, 1999) i
f
rises when the level of economic activity
rises. A rise in the level of economic activity causes loan demand
to increase, leading to higher p and therefore lower u
/
(p), thereby
suggesting cov
t
(M
t+1,
i
f,t+1
) < 0. That is, the bank pays a lower CD rate
if its prot is positively correlated with monetary policy (than it
does otherwise).
Furthermore, |cov
t
| is likely larger for large banks, implying that
large banks pay a lower CD rate than smaller banks, because the
large banks amount of lending and p, and hence u
/
(p), are likely
more strongly correlated with i
f
(monetary policy). The explana-
tion for this comes from convincing evidence by Kashyap and Stein
(2000) that smaller banks lending (to meet greater loan demand)
is constrained by their holdings of liquid assets [as smaller banks
cannot frictionlessly tap uninsured sources of funds to make up
for a Fed-induced shortfall in insured deposits (Kashyap and Stein,
2000, p. 408)], whereas large banks can raise unsecured funds to
meet greater loan demand. Indeed, the smallest banks borrow
hardly at all in the federal funds market, where credit risk is an is-
sue, because federal funds are unsecured borrowing (Kashyap and
Stein, 2000, p. 410). Fig. 3 shows the effective interest rates of time
deposits (not broken down by maturity) paid by four differently
Effective interest rate (percentage points): Large time deposits (large denomination CDs)
2
3
4
5
6
7
8
9
86 88 90 92 94 96 98 00 02 04 06
_______ 10 largest banks by assets
Banks ranked 11 through 100 by assets
_ _ _ _ _ Banks ranked 101 through 1,000 by assets
__ __ __ Banks not ranked among the 1,000 largest banks
Effective interest rate (percentage points): Other time deposits
2
3
4
5
6
7
8
9
88 90 92 94 96 98 00 02 04 06
_______ 10 largest banks by assets
Banks ranked 11 through 100 by assets
_ _ _ _ _ Banks ranked 101 through 1,000 by assets
__ __ __ Banks not ranked among the 1,000 largest banks
Fig. 3. Effective interest rates on time deposits by asset size.
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 87
sized groups of banks. In general, the 10 largest banks time deposit
rates are lower than those of the other groups. TBTF partially off-
sets the upward shift (Observation 2) arising from time impatience
and income smoothing.
4.3. Summary: The term structure of CD rates
Assuming e = and c = 0, the two-period CD rate, or Eq. (5), can
be expressed as:
where i
PEH
c2;t
= [i
f ;t
E
t
(i
f ;t1
)[=2 = [i
c1;t
E
t
(i
c1;t1
)[=2; and Di
f ;t1
=
i
f ;t1
i
f ;t
. The rst term on the right-hand side (PEH) of Eq. (6)
yields Fig. 2A. The second and third terms on the right-hand side
(which are zero if b = 1 and c = 0) yield Figs. 2B and C. The third
term causes the yield curve of CD rates to steepen, but is quantita-
tively much smaller than the second term, when the FF rate is ris-
ing/falling. The fourth term (which is zero if c = 0) represents TBTF
and is negative. The main conclusions are as follows.
(i) The current and expected future FF rates are the primary
determinant of the CD rates, as they also are of other money
market rates such as Treasury bills (Diebold and Sharpe,
1990).
(ii) Time impatience, income smoothing and TBTF are the sec-
ondary determinants of the CD rates. In Section 5.3, the
two-period CD rate (six-month CD rate) is expressed as the
weighted average of the current and expected future one-
period CD rates (three-month CD rates). The weights are
time-varying (explained below) and constitute part of the
secondary determinants in a different way.
Three implications are important.
(a) Suppose that the one- and two-period CD rates are the
short-term and long-term rates, respectively. Then, a term
premium, which is commonly interpreted as an extra yield
demanded by investors to compensate for holding long-term
instruments (here, two-period CDs) instead of short-term
instruments (here, one-period CDs) has theoretically inter-
pretable components from the suppliers (banks) perspec-
tive, as shown in Eq. (6).
(b) PEH does not hold for bank CD rates if the secondary deter-
minants are operative (they indeed are, as shown below).
(c) The one-period and two-period loan rates are expressed,
respectively, as: i
1;t
= (1 e
1
L
)
1
[(1 e
1
)i
c1;t
E
t
(M
t1
)
1
(c c
L
)[ and i
2;t
= (1 e
1
L
)
1
[(1 e
1
)i
c2;t
H
1
t
(1 E
t
(M
t1
))
(c c
L
)[, where e
L
= loan-rate elasticity (e
L
< 0) and c
L
= mar-
ginal noninterest cost of loans (assumed to be constant).
Clearly, the loan rates are linked to the CD rates of the same
maturity, implying that the term structure of CD rates car-
ries over into loan rates. Therefore, bank rates (loan and
CD rates) are governed by the term structure, which has sub-
stantially different characteristics from those of the standard
term structure of interest rates associated with Treasury
securities (i.e., (ii) above and, particularly, the time-varying
weights explained in Section 5.3). This paper does not
empirically investigate loan rates because no loan rate data
by maturity are publicly available.
In Section 5, the objective is, rst, to estimate b (to conrm time
impatience or b < 1) and c (to conrm income smoothing) and, sec-
ond, to estimate the time-varying weights. Because no data are
available for the conditional covariance, TBTF is subsumed under
the random disturbance term and is not investigated.
5. Empirical evidence
In the remainder of this paper, some common notation (such as
e
t
, a
0
, a
1
, etc.) is used repeatedly, but each set of symbols is specic
to an equation where it appears. Notational confusion can be
avoided by noting the following.
Notation in Sections 5.1 and 6 where monthly data are used (one
period = one month)
i
c3,t
three-month CD rate = three-period CD rate (Eq. (2), n = 3)
i
c6,t
six-month CD rate = six-period CD rate (Eq. (2), n = 6)
Notation in Sections 5.2 and 5.3 where quarterly data are used
(one period = one quarter)
i
c3,t
three-month CD rate = one-period CD rate (Eq. (1))
i
c6,t
six-month CD rate = two-period CD rate (Eq. (2), n = 2)
5.1. Empirical evidence for Observation 1: The three-month and six-
month CD rates
The Granger causality test is performed using the following sec-
ond-order VAR (vector autoregression). The lag length is chosen
based on the Schwarz information criterion (SIC).
X
t
= l A
1
X
t1
A
2
X
t2
g
t
; (7)
where X
/
t
= (x
1;t
; x
2;t
; x
3;t
); x
1;t
= i
f ;t
; x
2;t
= i
c3;t
; x
3;t
= i
c6;t
; l is a
(3 1) vector of constants; A
1
and A
2
are each a (3 3) matrix of
coefcients; and g
t
is a (3 1) vector of errors that are iid N(0, R)
(R is positive denite). If variables in a VAR are stationary series,
then the Granger test is a straightforward F test. However, interest
rates are known to have a unit root (Stock and Watson, 1988; Hall
et al., 1992), which is conrmed using the standard unit-root test. In
this case, the standard test statistic based on the VAR estimated in
levels does not have the usual limiting v
2
distribution. An exception
exists under a certain condition, however, which this paper em-
ploys. Toda and Phillips (1993) develop a limit theory for Wald tests
of Granger causality for general VAR systems with an arbitrary
(6)
88 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294
number of cointegrating vectors (Theorem 1 in Toda and Phillips,
1993, p. 1375), which is used in this paper. (The details are available
from the author upon request.)
The results of the Wald tests (one-tailed tests) are shown in
Table 2. For the pre-1994:2 period, the null hypothesis of no Gran-
ger causality (or no predictability) is not rejected in both cases (the
CD3 and CD6 rates) regardless of the two-variable VAR (tests G and
H) or three-variable VAR (tests A and B). For the post-1994:2 per-
iod, both CD rates predict the FF rate in the bivariate VAR (tests G
and H); however, only the CD6 rate predicts the FF rate in the
three-variable VAR (test B). This result is expected because the
CD6 rate is the weighted average of the current and expected fu-
ture CD3 rates, or changes in the CD3 rate in anticipation of future
changes in the FF rate are embodied in the CD6 rate. (This point
becomes clear in Section 5.3.) Therefore, the CD3 rate loses its
predictive power in the presence of lagged CD6 rates in the FF
equation. The different results for the pre- and post-1994:2 periods
are consistent with the Feds disclosure practice introduced in Feb-
ruary 1994, which resulted in greater transparency of the stance of
monetary policy (Nautz and Schmidt, 2009). Indeed, Poole and
Rasche (2003) and others nd that, since February 1994, policy ac-
tions (i.e., changes in the FF rate) have been well anticipated by
market participants.
5.2. Empirical evidence for Observation 2: The three-month CD rate
The three-month CD rate equation, or Eq. (1), is estimated in
two different ways to cross-examine the plausibility of parameter
values. First, the parameters b and c are estimated using the gen-
eralized method of moments, or GMM (Hansen, 1982). Assuming
the CRRA utility, Eq. (1) can be rewritten as: E
t
[1 b(1 + e
1
)(i
c3,t
/
(i
f,t
c))(p
t
/p
t+1
)
c
] = 0. As explained in Section 3, Hutchison
(1995) derives the elasticity, which is a function of the own rate
(i
c3,t
in this paper) and the rate on a substitute in investors portfo-
lios denoted here by i
s,t
. This is expressed by 1 + e
1
= a
0
+
a
1
i
c3,t
+ a
2
i
s,t
, where a
0
> 0, a
1
< 0, and a
2
> 0 are hypothesized.
The signs of a
1
and a
2
follow Hutchison (1995). If a
1
= a
2
= 0, then
1 + e
1
= a
0
> 0. Incorporating this, the equation is expressed as:
E
t
[1 (ba
0
ba
1
Di
c3;t
ba
2
Di
s;t
)(i
c3;t
=(i
f ;t
c))(p
t
=p
t1
)
c
[ = 0;
(8)
where Di
c3,t
= i
c3,t
i
c3,t1
and Di
s,t
= i
s,t
i
s,t1
. Differencing is nec-
essary because GMM requires stationary series. In order to estimate
four parameters (ba
0
, ba
1
, ba
2
, and c), the following four instru-
ments are used: a constant, Di
c3,t
, Di
s,t
, and i
c3,t
/(i
f,t
c)
.
The two
variables, i
s,t
and i
f,t
, are contained in the timet information set,
and i
c3,t
represents a decision that is based solely on the timet
information set. Because there are four parameters and four instru-
ments (four orthogonality conditions), it is a just-identied case.
Second, Eq. (1) is expressed as a log-linear equation as follows.
Following Hansen and Singleton (1983), Attanasio and Low (2000),
and others, I assume that (p
t+1
/p
t
)
c
is conditionally lognormal:
log E
t
(p
t+1
/p
t
)
c
= E
t
log (p
t+1
/p
t
)
c
+ (1/2)var
t
[log (p
t+1
/p
t
)
c
]. Fur-
thermore, following Attanasio and Low (2000), I assume (1/2)var
t
[log (p
t+1
/p
t
)
c
] ~ k + f
t
, where k is a constant (k > 0) and f
t
denotes
a random component. Then, after taking logs and assuming
rational expectations, log (p
t+1
/p
t
) = E
t
log (p
t+1
/p
t
) + v
t+1
, where
v
t+1
denotes an expectation error, Eq. (1) can be rewritten as:
log (p
t+1
/p
t
) = (1/c)[log (1 + e
1
) + log b + k] + (1/c)log (i
c3,t
/(i
f,t
c)) +
g
t+1
, where g
t+1
= (1/c)f
t
+ v
t+1
. For estimation, this equation is gen-
eralized in two respects. Incorporating the criticism by Hutchison
(1995), log (1 + e
1
) = a
0
+ a
1
log (i
c3,t
/i
c3,t1
) + a
2
log(i
s,t
/i
s,t1
),
where a
0
> 0, a
1
< 0 and a
2
> 0 are expected (as above). The two
variables on the right-hand side are expressed as rst differences
(as above). Also, given that the p
t
variable appears to be a station-
ary process (based on a preliminary unit-root test), the strict rst
differencing on the left-hand side is relaxed. Then, the equation
is expressed as:
log p
t1
= c
0
c3
1
log(i
c3;t
=i
c3;t1
) c
2
log(i
s;t
=i
s;t1
)
c
3
log(i
c3;t
=(i
f ;t
c)) c
4
log p
t
g
t1
; (9)
where c
1
= a
1
/c < 0, c
2
= a
2
/c > 0, c
3
= 1/c > 0, and c
4
> 0 (and c
4
~ 1)
are expected.
For the MC of CDs, $0.006 is used as a benchmark in this paper.
This benchmark estimate comes from Humphrey (1985) and is
based on the 1981 Federal Reserves Functional Cost Analysis re-
ports. That is, banks are assumed to incur noninterest costs of
$0.006 per (incremental) dollar of CDs. Because interest rates are
expressed as percentages in the estimation, c = 0.6 (instead of
c = 0.006) is used. For CDs whose denominations are $100,000 or
more, their MC is likely much lower. For example, suppose that
banks incur the MC of $0.006 when they issue a CD of a small
denomination, such as $10,000, or they incur the total cost of $60
(assuming constant MC). Banks likely incur a total cost similar to
$60 when they issue a CD of a large denomination, such as
$100,000, but the per-dollar cost or MC (which is assumed to be
approximately equal to the average cost) is only one tenth as much
as in the $10,000 case, or $0.0006. For this reason, c = 0.06 is also
used in addition to c = 0.6. For the substitute, three money market
Table 2
Granger causality tests,
a
monthly sample, 1987:22007:5.
Null hypothesis 1987:21994:1 1994:2 2007:5
Test statistic v
2
(2)
Test statistic v
2
(2)
Second-order VAR for three variables (i
f,t
, i
c3,t
, i
c6,t
)
b
Test A H
0
: i
c3,t
does not Granger cause i
f,t
3.30 3.07
Test B H
0
: i
c6,t
does not Granger cause i
f,t
4.59 8.18
**
Test C H
0
: i
f,t
does not Granger cause i
c3,t
14.38
***
7.61
**
Test D H
0
: i
c6,t
does not Granger cause i
c3,t
6.01
**
16.22
***
Test E H
0
: i
f,t
does not Granger cause i
c6,t
8.31
**
5.58
*
Test F H
0
: i
c3,t
does not Granger cause i
c6,t
3.82 1.86
Second-order bivariate VAR
c
Test G H
0
: i
c3,t
does not Granger cause i
f,t
0.31 111.12
***
Test H H
0
: i
c6,t
does not Granger cause i
f,t
1.57 119.83
***
For variable denitions, see Table 1.
a
Toda and Phillips (1993, Theorem 1).
b
The lag length of 2 is selected based on the Schwarz information criterion over the 1987:22007:5 period.
c
The lag length of 2 is used.
*
Signicant at the 10% level (the critical value is 4.61; one-tailed test).
**
Signicant at the 5% level (the critical value is 5.99; one-tailed test).
***
Signicant at the 1% level (the critical value is 9.21; one-tailed test).
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 89
instruments are considered: bankers acceptance, commercial pa-
per and Treasury bills. Finally, the sample period is quarterly
1990:12007:1 in order to exclude the 1987:2, 1987:3, 1987:4,
1989:3 (all negative ROA values), and 1989:4 (near-zero ROA)
Table 3
GMM estimation results of the three-month CD rate for various quarterly sample periods.
Rate on a substitute H
1
indicates the alternative hypothesis
ba
0
Estimate
a
ba
1
Estimate
a
ba
2
Estimate
a
c Estimate
a
H
1
: ba
0
> 0 H
1
: ba
1
< 0 H
1
: ba
2
> 0 H
1
: c > 0
Marginal noninterest cost of CDs: c = 0.6
3-Month bankers acceptance
b
0.834
***
1.300 1.429 1.248
*
(0.030) (1.480) (1.513) (0.870)
3-Month commercial paper
c
0.828
***
0.769 0.899 1.211
*
(0.038) (1.613) (1.676) (0.914)
3-Month Treasury bill
d
0.375
***
0.652
e
0.684 11.184
***
(0.098) (0.268) (0.268) (2.741)
Marginal noninterest cost of CDs: c = 0.06
3-Month bankers acceptance
b
0.949
***
0.703 0.804 0.615
*
(0.014) (0.883) (0.899) (0.405)
3-Month commercial paper
c
0.950
***
0.420 0.510 0.478
*
(0.014) (0.736) (0.760) (0.316)
3-Month Treasury bill
d
0.916
***
0.405 0.618 2.443
f
(0.058) (0.356) (0.412) (2.028)
This table shows the estimation results of Eq. (8):
E
t
[1 (ba
0
ba
1
Di
c3;t
ba
2
Di
s;t
)(i
c3;t
=(i
f ;t
c))(p
t
=p
t1
)
c
[ = 0;
where b(1 + e
1
) = b(a
0
+ a
1
Di
c3,t
+ a
2
Di
s,t
); i
c3,t
is the three-month CD rate; i
s,t
is the rate on a substitute in investors portfolios; i
f,t
is the federal
funds rate; and p
t
is ROA (return on assets). The instruments are: a constant; Di
c3,t
; Di
s,t
; and (i
c3,t
/(i
f,t
c)). Standard errors are in parentheses.
a
H
1
: ba
1
< 0 and H
1
: ba
2
> 0 follow Hutchison (1995). The heteroskedasticity consistent covariance is used (White, 1980). Standard errors are
similar (not shown) when the heteroskedasticity and autocorrelation consistent covariance is used (Newey and West, 1987).
b
The sample period is 1990:12000:2. The series was discontinued in June of 2000.
c
The sample period is 1990:11997:2. The series was discontinued in August of 1997.
d
The sample period is 1990:12007:1.
e
Signicant at the 5% level under the two-tailed test.
f
Borderline insignicance. The p value (one-tailed) is 0.11.
*
Signicant at the 10% level.
***
Signicant at the 1% level.
Table 4
Log-linear equation estimation results of the three-month CD rate for various quarterly sample periods.
Rate on a substitute H
1
indicates the alternative hypothesis
c
1
Estimate
a
c
2
Estimate
a
c
3
Estimate
a
Implied c
4
Estimate
a
H
1
: c
1
< 0 H
1
: c
2
> 0 H
1
: c
3
> 0 c = 1/c
3
H
1
: c
4
> 0
Marginal noninterest cost of CDs: c = 0.6
3-Month bankers acceptance
b
1.556 1.658 0.934
**
1.070 0.879
***
R
2
= 0.81, DW
c
= 1.94 (1.503) (1.534) (0.432) (0.085)
3-Month commercial paper
d
1.230 1.326 1.046
**
0.956 0.842
***
R
2
= 0.79, DW
c
= 1.85 (1.371) (1.380) (0.535) (0.120)
3-Month Treasury bill
e
0.226
f
0.297
f
0.055 18.181 0.927
***
R
2
= 0.86, DW
c
= 1.86 (0.103) (0.115) (0.077) (0.054)
Marginal noninterest cost of CDs: c = 0.06
3-Month bankers acceptance
b
1.408 1.552 1.601
**
0.624 0.910
***
R
2
= 0.80, DW
c
= 1.90 (1.522) (1.554) (0.857) (0.086)
3-Month commercial paper
d
1.199 1.363 2.061
**
0.485 0.904
***
R
2
= 0.78, DW
c
= 1.84 (1.383) (1.396) (1.139) (0.119)
3-Month Treasury bill
e
0.193
g
0.302
f
0.545
*
1.834 0.937
***
R
2
= 0.86, DW
c
= 1.86 (0.104) (0.113) (0.355) (0.049)
This table shows the estimation results of Eq. (9): log p
t1
= c
0
c
1
log(i
c3;t
=i
c3;t1
) c
2
log(is;t =i
s;t1
) c
3
log(i
c3;t
=(i
f ;t
c)) c
4
logpt g
t1
; where log (1 + e
1
) = a
0
+ a
1
log (i
c3,t
/i
c3,t1
) + a
2
log (i
s,t
/i
s,t1
); c
1
= a
1
/c (c = coefcient of relative risk aversion or RRA); c
2
= a
2
/c; c
3
= 1/c; c
4
~ 1 is expected; g
t+1
is the error term. For variable denitions,
see Table 3. Because c
0
can take any value (negative, zero or positive), its estimate is not shown. Standard errors are in parentheses.
a
H
1
: c
1
< 0 and H
1
: c
2
> 0 follow Hutchison (1995). Standard errors are based on the heteroskedasticity consistent covariance (White, 1980). Those based on the heter-
oskedasticity and autocorrelation consistent covariance (Newey and West, 1987) are slightly larger (not shown here).
b
The sample period is 1990:12000:2. See Table 3.
c
DurbinWatson statistic.
d
The sample period is 1990:11997:2. See Table 3.
e
The sample period is 1990:12007:1.
f
Signicant at the 5% level under the two-tailed test.
g
Signicant at the 10% level under the two-tailed test.
*
Signicant at the 10% level.
**
Signicant at the 5% level.
***
Signicant at the 1% level.
90 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294
observations. These ROA values reect (one-time) sizable addi-
tions to loss provisions for loans to developing countries and are
considered as outliers.
Table 3 shows the estimation results of Eq. (8). First, ba
0
is sig-
nicant at the 1% level, both ba
1
and ba
2
are insignicant, and c is
signicant at least at the 10% level (except for the borderline case).
The results imply that the elasticity may be considered as a con-
stant, or (1 + e
1
) = a
0
. If the elasticity is assumed to be fairly large
(presumed in this paper), b is close to the ba
0
estimates. This sug-
gests that large banks are time impatient because b ~ 0.95 < 1
(c = 0.06 case). Second, excluding the Treasury bill case where
c = 0.6, c estimates are roughly between 0.5 and 2.0, consistent
with the values economists commonly agree on (for example, Feld-
stein and Ranguelova, 2001). This result suggests that large banks
engage in income smoothing. Table 4 shows the estimation results
of Eq. (9). It does not show the c
0
= (a
0
+ k + log b)/c estimates be-
cause c
0
can theoretically take any value. The implied c estimates
are almost identical to those in Table 3, supporting the plausibility
of the c estimates (excluding the Treasury bill case where c = 0.6).
Finally, the results of Tables 3 and 4 suggest that PEH does not hold
for bank CD rates.
5.3. Empirical evidence for the time-varying weights: The six-month
CD rate
For the CD6 rate, Eq. (2) (where n = 2 or the two-period CD rate)
can be rewritten as follows (assuming the CRRA utility):
i
c6;t
= w
t
i
c3;t
(1 w
t
)E
t
(i
c3;t1
) H
1
t
cov
t
(M
t1
M
t2
; i
c3;t1
); (10)
w
t
= E
t
(M
t1
)=H
t
= 1=[1 (bE
t
(g
c
t2
)=E
t
(g
c
t1
))[; (11)
where g
t+2
= p
t+2
/p
t
; and g
t+1
= p
t+1
/p
t
. A striking feature of the term
structure of CD rates that differs from the existing literature on the
term structure of interest rates is the time-varying weights, w
t
and
1 w
t
. In the existing literature, w
t
= 1/2, whereas the term struc-
ture of CD rates exhibits time-varying weights over business cycles.
In particular, during periods of economic contraction (when both
banks prots and the FF rate are falling), w
t
declines, placing a
greater weight (1 w
t
) on E
t
(i
c3,t+1
) relative to periods of economic
expansion. This shift in the weights creates a further decline in i
c6,t
because i
c3,t
> E
t
(i
c3,t+1
) and represents income smoothing in addi-
tion to that implied by i
c3,t
and E
t
(i
c3,t+1
). As i
c6,t
falls further, fewer
CDs are demanded and issued, resulting in less interest expenses in
the future and counterbalancing lower future prots associated
with rising loan defaults. Interestingly, this income smoothing is
asymmetrical: a greater weight is placed on i
c3,t
relative to periods
of economic contraction when the FF rate (and therefore also i
c3,t
)
is rising. This shift in the weights causes i
c6,t
to move upward slug-
gishly than otherwise because i
c3,t
< E
t
(i
c3,t+1
) when the FF rate is ris-
ing. That is, additional income smoothing takes place only when the
economy is contracting.
Eqs. (10) and (11) are estimated as follows (steps 15 in
Table 5). First, the time-varying weight is estimated using the
Kalman lter. Then, the resulting estimated one-step-ahead {w
t
}
sequence is used to estimate the linear approximation of Eq. (11)
using the instrumental-variables estimation method. (The details
are available from the author upon request.) The results are shown
in Table 5, where c = 0.5, 1.0 and 2.0 are chosen based on the esti-
mates in Tables 3 and 4. The estimates a
1
> 0 and a
2
< 0 (in step 3
in Table 5) implied by the d
1
and d
2
estimates are consistent with
the theory. Also, the theory (Eq. (11)) implies a
1
~ |a
2
|, which is
evidenced in the c = 0.5 and 1.0 cases. (For the c = 2.0 case, a
1
and |a
2
| differ only slightly.) Assuming a
1
~ |a
2
|, w
t
declines during
periods of economic contraction because E
t
g
c
t2

< E
t
(g
c
t1
), as
explained above, hence being consistent with the theoretical
explanation above. (I assume that Jensens inequalities i.e.,
E(1/x) P1/E(x) roughly cancel out between E
t
g
c
t2

and
E
t
g
c
t1

, thereby E
t
(g
c
t2
)=E
t
g
c
t1

~ E
t
g
c
t1

=E
t
g
c
t2

.)
6. Monetary policy transmission: Interest rate pass-through
As indicated above in connection with Fig. 1C, positive devia-
tions of the CD3 rate from the FF rate (and, similarly, of the CD6
rate from the CD3 rate) appear to be adjusted more rapidly than
Table 5
Instrumental-variables estimation results of the six-month CD rate, quarterly sample, 1990:12007:1.
Assumed c value H
1
indicates the alternative hypothesis
d
0
Estimate
a
d
1
Estimate
a
d
2
Estimate
a
R
2
DurbinWatson statistic
H
1
: d
0
0 H
1
: d
1
> 0 H
1
: d
2
> 0
Estimation of time-varying weight wt : g
c
t2
= d
0
d
1
wt d
2
g
c
t1
f
t2
c = 0.5 0.010 0.038
*
1.000
***
0.54 1.93
(0.180) (0.025) (0.178)
c = 1.0 0.073 0.085
**
1.054
***
0.58 1.93
(0.200) (0.052) (0.198)
c = 2.0 0.267 0.217
**
1.225
***
0.67 2.03
(0.260) (0.124) (0.260)
This table shows the estimation results of Eqs. (10) and (11). The estimation procedure is as follows.
Step 1: The six-month CD rate can be expressed as: i
c6,t
= w
t
i
c3,t
+ (1 w
t
) E
t
(i
c3,t+1
) + e
t
.
Step 2: The i
c6,t
(observation) equation and w
t
(state variable) are estimated using the Kalman lter.
Step 3: The linear approximation of w
t
is expressed as: wt = a
0
a
1
Et g
c
t2

a
2
Et g
c
t1

ut .
Step 4: Assuming rational expectations: g
c
t2
= Et (g
c
t2
) v
t2
and g
c
t1
= Et (g
c
t1
) v
t1
.
Step 5: The w
t
equation is expressed as: g
c
t2
= d
0
d
1
wt d
2
g
c
t1
f
t2
.
The variables are dened as follows: i
c6,t
is the six-month CD rate; i
c3,t
is the three-month CD rate; w
t
is the time-varying weight for which one-step-ahead state-space model
estimates are used; e
t
is the time-varying conditional covariance term, assumed to be a random disturbance; g
t+1
= p
t+1
/p
t
and g
t+2
= p
t+2
/p
t
(where p = return on assets of
banks with assets greater than $15 billion); u
t
is the linear approximation error; v
t+2
and v
t+1
are expectation errors; f
t+2
= v
t+2
+ (a
2
/a
1
)v
t+1
(1/a
1
)u
t
. c denotes the relative
risk aversion coefcient, whose value is assumed (given from Tables 3 and 4). The w
t
equation is estimated by the instrumental-variables method. The instruments for g
c
t1
are: f
c
t1
= (p
t1
=pt )
c
, where p
t+1
and p
t
are those of all banks; and h
c
t1
= (p
t1
=pt )
c
, where p
t+1
and p
t
are those of banks with average assets between $1 billion and $15
billion. For the c = 1.0 case, the instruments are highly correlated with g
c
t1
(R
2
= 0.94 when g
c
t1
is regressed on a constant, f
c
t1
, and h
c
t1
). Also, the Sargan instrument validity
test fails to reject the null hypothesis of independence of the instruments and the error term (the test statistic is 3.20, and the 5% critical value of v
2
(1)
is 3.84). Hence, the
chosen instruments appear to be valid. The theory suggests that a
1
> 0 and a
2
< 0, and therefore d
1
= 1/a
1
> 0 and d
2
= a
2
/a
1
> 0. Standard errors are in parentheses.
a
The heteroskedasticity consistent covariance is used (White, 1980). Standard errors are similar (not shown here) when the heteroskedasticity and autocorrelation
consistent covariance is used (Newey and West, 1987).
*
Signicant at the 10% level.
**
Signicant at the 5% level.
***
Signicant at the 1% level.
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 91
negative deviations when corrections occur (Observation 3). In
order to formally test this observation, the following TAR (threshold
autoregressive) model is estimated for the CD3 and FF rates and
separately for the CD6 and CD3 rates (Enders and Granger,
1998): De
t
= q
1
I
t
e
t1
+ q
2
(1 I
t
)e
t1
+ e
t
where e
t
= deviations from
the long-run equilibrium (=OLS residuals); De
t
= e
t
e
t1
; q
1
= rate
of autoregressive adjustment when e
t1
P0; q
2
= rate of autore-
gressive adjustment when e
t1
< 0; I
t
= 1 if e
t1
P0 and =0 if
e
t1
< 0; and e
t
= white-noise disturbance. The long-run equilibrium
(cointegration relationship) is described by i
c3,t
0.183 0.991i
f,t
=
e
t
= 0 and i
c6,t
0.106 0.997i
c3,t
= e
t
= 0, where OLS estimates are
used.
The results are shown in Panel A of Table 6. The unit-root tests
with the entailed alternative of asymmetric adjustment suggest
that the {e
t
} series is I(0) (a stationary series) in both cases. The null
hypothesis of symmetric adjustment is rejected for the CD3 and FF
rates, while, somewhat surprisingly, it is not rejected for the CD6
and CD3 rates. For the CD3 and FF rates, the estimates,
q
1
= 0.336 and q
2
= 0.160, indicate asymmetric adjustment: fas-
ter (slower) adjustment toward the long-run equilibrium when po-
sitive (negative) deviations exist or e
t1
> 0 (e
t1
< 0). Alternatively,
they indicate that negative deviations tend to persist longer. This
nding is consistent with Observation 3.
The degree of monetary policy pass-through can be inferred
from the cointegration relationship (long-run pass-through) and
the error correction model or ECM estimates (short-run pass-
through), with the latter shown in Panel B of Table 6. First, the
parameter values of the cointegrating vectors of 0.991 and 0.997
(shown in Panel A of Table 6) imply a nearly perfect pass-through
of the policy rate (the FF rate) to the CD rates in the long run.
Second, the speed-of-adjustment parameters (as) of the ECM
for the CD3 and FF rates show interesting short-run dynamics of
pass-through. The alternative hypotheses indicate expected direc-
tions (or expected signs) in which each variable should change in
response to previous-period deviations for the deviations to be
eliminated. a
11
takes the expected sign and is signicant, but the
CD3 rate adjusts in the wrong direction (a
12
> 0) for negative devi-
ations (though its p value is 0.31 and hence insignicant at the 10%
level under the two-tailed test). The latter nding is consistent
with Observation 3: negative deviations (i
c3,t
< i
f,t
+ 0.183 assuming
0.991 ~ 1) persist longer than positive deviations, presumably be-
cause the CD rate does not adjust in the expected direction. This
leaves the FF rate to adjust toward the long-run equilibrium in
this case by falling, noting a
22
> 0, I
t
= 0 and e
t1
< 0. Arguably,
the FF rate is responding to underlying deteriorating economic
conditions in this case, not to the disequilibrium between the FF
rate and the CD3 rate caused by the economic conditions. Exclud-
ing the a
12
case, the monthly speed-of-adjustment estimates
(0.202, 0.174 and 0.355) are comparable to those of other studies.
For example, for the euro area and based on ECM, Kok Srensen
and Werner (2006) obtain a weighted-average speed-of-adjust-
ment of 0.257 for retail time deposits, although their estimates
vary across countries from 0.051 to 0.396 (Tables 3 and 4 in
Kok Srensen and Werner, 2006). Also, for New Zealand and based
on ECM, Liu et al. (2008) show that the speed-of-adjustment esti-
mate for six-month time deposits is 0.336 before March 1999,
but it declines to 0.180 after March 1999 (Table 3 in Liu et al.,
2008).
Third, for the CD6 and CD3 rates, the null hypothesis of
symmetric autoregressive adjustment is not rejected (Panel A of
Table 6
Monetary policy pass-through and the term structure of CD rates, monthly sample, 1987:22007:5.
Panel A: TAR (threshold autoregressive) models
a
De
t
= q
1
I
t
e
t1
+ q
2
(1 I
t
)e
t1
+ e
t
b
where e
t
= deviations from the long-run equilibrium (OLS residuals); De
t
= e
t
e
t1
; q
1
= rate of autoregressive adjustment when e
t-1
P 0; q
2
= rate of autoregressive
adjustment when e
t1
< 0; I
t
= 1 if e
t1
P0 and =0 if e
t1
< 0; e
t
= white-noise disturbance. The long-run equilibrium is described by i
c3,t
0.183 0.991i
f,t
= e
t
= 0 and
i
c6,t
0.106 0.997i
c3,t
= e
t
= 0.
q
1
Estimate q
2
Estimate H
0
: q
1
= q
2
= 0 (a unit root) H
0
: q
1
= q
2
(0) (symmetric adjustment)
i
c3,t
and i
f,t
0.336
***
0.160
***
U statistic = 20.609
***c
F statistic = 4.026
**d
(0.056) (0.066)
i
c6,t
and i
c3,t
0.222
***
0.188
***
U statistic = 17.095
***c
F statistic = 0.252
(0.47) (0.051)
Panel B: Error correction model (ECM) for i
c3,t
and i
f,t
, and ECM for i
c6,t
and i
c3,t
ECM for i
c3,t
and i
f,t
Di
c3,t
= a
11
I
t
e
t1
+ a
12
(1 I
t
)e
t1
+ a
11
Di
c3,t1
+ a
12
Di
f,t1
+ disturbance
e
Di
f,t
= a
21
I
t
e
t1
+ a
22
(1 I
t
)e
t1
+ a
21
Di
c3,t1
+ a
22
Di
f,t1
+ disturbance
e
ECM for i
c6,t
and i
c3,t
Replace i
c3,t
with i
c6,t
, and i
f,t
with i
c3,t
, in the above two equations
(H
1
= alternative hypothesis) H
1
: a
11
< 0 H
1
: a
12
< 0 H
1
: a
11
0 H
1
: a
12
0
H
1
: a
21
> 0 H
1
: a
22
> 0 H
1
: a
21
0 H
1
: a
22
0
ECM for i
c3,t
and i
f,t
Di
c3,t
equation 0.202
***
0.105 0.173
*
0.423
***
(0.087) (0.104) (0.091) (0.097)
Di
f,t
equation 0.174
***
0.355
***
0.138
*
0.317
***
(0.068) (0.082) (0.072) (0.076)
ECM for i
c6,t
and i
c3,t
Di
c6,t
equation 0.164 0.463
f
0.577
***
0.248
(0.163) (0.175) (0.198) (0.202)
Di
c3,t
equation 0.457
***
0.720
***
0.285 0.029
(0.148) (0.158) (0.179) (0.182)
For variable denitions, see Table 1. Standard errors are in parentheses.
a
Enders and Granger (1998).
b
Because the {e
t
} sequence is demeaned, the attractor is zero (a
0
= 0 in Enders and Granger, 1998). The Schwarz information criterion (SIC) selects the lag length of 0 for
i
c3,t
and i
f,t
, and the lag length of 1 for i
c6,t
and i
c3,t
. For the latter, the parameter estimate of the lagged variable is not shown.
c
The critical values are 3.10, 3.82 and 5.53 for the 10%, 5% and 1% signicance levels, respectively (Panel A, sample size = 250, Table 1 in Enders and Granger (1998)).
d
The p value is 0.045.
e
SIC selects the lag length of 1 for all cases.
f
Signicant at the 1% level under the two-tailed test.
*
Signicant at the 10% level.
**
Signicant at the 5% level.
***
Signicant at the 1% level.
92 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294
Table 6). However, because Fig. 1C appears to suggest asymmetric
adjustment similar to the case of the CD3 and FF rates, ECM for the
CD6 and CD3 rates is estimated by incorporating both positive and
negative deviations. The results are shown in Panel B of Table 6.
Interestingly, the CD6 rate adjusts in the wrong direction, similar
to the CD3 rate in ECM for the CD3 and FF rates. a
12
(=0.463) takes
the wrong sign and, under the two-tailed test, is signicant at
the 1% level. That is, the CD6 rate falls for negative deviations
(i
c6,t
< i
c3,t
+ 0.106 assuming 0.997 ~ 1), hence widening the gaps
further. The CD3 rate adjusts toward the long-run equilibrium in
this case.
The two wrong directions above may be explained based on
the term structure of CD rates. Note that, in Fig. 1, negative (posi-
tive) deviations tend to arise when the FF rate is falling (rising)
consistent with income smoothing (Fig. 2C). Note also that risk-
averse banks income smoothing behavior (lowering/raising the
CD rates) arises from their expectations of future (lower/higher)
prots see Fig. 2C and the second and third terms on the right-
hand side of Eq. (6). In the CD3 case (a
12
= 0.105 > 0), loan defaults
are rising in deteriorating economic conditions when negative
deviations exist, thereby inducing banks to expect a decline in fu-
ture prots. Then, risk-averse banks engage in more aggressive in-
come smoothing (as in Fig. 2C), which lowers CD rates even further
than the levels driven by the falling FF rate. Hence, the period of
persistent negative deviations, during which the CD3 rate is lower
than, and falls further away from, the FF rate, seems to imply that
banks hold their pessimistic prot expectations longer during the
downturn phase of business cycles than they do their optimistic
prot expectations during the upturn phase. In the CD6 case
(a
12
= 0.463 > 0), the decline in the CD6 rate when negative devia-
tions exist is consistent with a decrease in the weight w
t
discussed
above. The explanation based on the term structure of CD rates is
identical to the CD3 case. Negative deviations tend to arise when
the FF is falling (Fig. 1) or in an environment of deteriorating eco-
nomic conditions that cause loan defaults to rise. A progressive de-
cline in banks prots results in an increase in E
t
g
c
t2

=E
t
g
c
t1

and
hence an increase in the weight (1 w
t
) on E
t
(i
c3,t+1
). Because
i
c3,t
> E
t
(i
c3,t+1
) when the FF rate is falling, this shift in the weight
causes a further decline in the CD6 rate beyond the level driven
by the falling CD3 rate, that is, further income smoothing.
The results above have two implications. First, the paper pre-
sumes that the interest-rate elasticity of the CD supply (e in
the model) is fairly large but not perfectly elastic. Under this pre-
sumption, expansionary monetary policy or a decline in the FF rate
(the primary determinant of CD rates see (i) in Section 4.3) causes
the CD rates to fall and hence the quantity supplied of CDs to fall
signicantly, thereby potentially weakening the standard bank
lending channel that otherwise predicts an increase in bank lend-
ing comparable to an increase in transaction deposits.
Second, a presumably unintended consequence of expansionary
monetary policy under deteriorating economic conditions is a po-
tential increase in withdrawal risk. Because wholesale CDs, whose
quantity supplied decreases as a result of expansionary monetary
policy, can be sold in the secondary market, they pose a lesser
withdrawal risk for banks. By comparison, transaction deposits,
which increase as a result of expansionary monetary policy, pose
a greater withdrawal risk for banks, particularly when banks are
vulnerable because of rising loan defaults. Hence, this composi-
tional shift in banks liabilities during the downturn phase away
from CDs toward transaction deposits may pose a greater with-
drawal risk for banks.
Finally, the ndings of this paper and those of Hofmann and Mi-
zen (2004), which are explained briey in the introduction, can
now be better compared. As illustrated clearly in Figs. 2A and C
(which are supported empirically in this paper), a dominant fea-
ture of the US data is that positive (negative) deviations of the
CD rates from the FF rate or simply the CD rates above (below)
the FF rate tend to occur under the rising (falling) FF rate. Based
on the estimation results of a non-linear ECM, Hofmann and Mizen
(2004) nd that UK retail bank rates are adjusted faster when their
deviations from the average base rate are widening, thereby caus-
ing the loss of not adjusting to rise above the menu cost (i.e., the
cost of adjusting). According to Hofmann and Mizen (2004), such
circumstances arise when deviations are positive (negative) and
the average base rate is falling (rising). In the context of the US
data, they may more likely arise at policy turning points. For exam-
ple, positive deviations of the CD rates from the FF rate in an envi-
ronment of a rising FF rate (Figs. 2A and C) become wider if the FF
rate abruptly changes its direction (i.e., a policy shift) and banks
failed to anticipate the change. If Hofmann and Mizens (2004)
ndings on UK data were applicable to US data, arguably, they
could be encouraging news for US monetary policy makers be-
cause, for policy shifts to be effective at these critical junctures, a
greater degree of pass-through (i.e., faster adjustments of retail
bank rates) is exactly what policy intention needs.
7. Conclusion
In this paper, I investigate the term structure of CD rates and its
relationship with monetary policy (the FF rate). First, the papers
model and its calibration predict that the CD rates, and hence their
term structure, are governed primarily by the FF rate and second-
arily by banks income smoothing and other behaviors. The predic-
tion is consistent with the estimation results, implying that the
pure expectations hypothesis does not hold for the CD rates and
that the term structure of CD rates differs signicantly from the
standard description of the term structure of interest rates. The lat-
ter implication holds because the longer-term CD rate is an average
of the current and expected future shorter-term CD rates with
time-varying weights. Second, the adjustment toward the long-
run equilibrium between the FF and CD rates is asymmetrical over
the business cycle. In particular, the downturn phase of business
cycles (when the FF is falling) appears to be accompanied by more
aggressive income smoothing by banks compared with the up-
turn phase due to their pessimistic expectations of future prots.
It causes the lower CD rates to persist, resulting in the lower quan-
tity supplied of CDs. The compositional shift in banks liabilities
during the downturn phase away from CDs toward transaction
deposits may pose a greater withdrawal risk for banks.
Acknowledgement
The author would like to thank an anonymous referee for help-
ful comments and suggestions.
References
Angelini, P., 2000. Are banks risk averse? Intraday timing of operations in the
interbank market. Journal of Money, Credit, and Banking 32, 5473.
Attanasio, O.P., Low, H., 2000. Estimating Euler Equations. Technical Working Paper
253, National Bureau of Economic Research.
Bernanke, B.S., 2004. The Logic of Monetary Policy. Remarks before the National
Economists Club, Washington, DC, December 2, 2004.
Diebold, F.X., Sharpe, S.A., 1990. Post-deregulation bank-deposit-rate pricing: The
multivariate dynamics. Journal of Business & Economic Statistics 8, 281291.
Edwards, F.R., 1977. Managerial objectives in regulated industries: Expense-
preference behavior in banking. Journal of Political Economy 85, 147162.
Enders, W., Granger, C.W.J., 1998. Unit-root tests and asymmetric adjustment with
an example using the term structure of interest rates. Journal of Business &
Economic Statistics 16, 304311.
Feldstein, M., Ranguelova, E., 2001. Individual risk in an investment-based social
security system. American Economic Review 91, 11161125.
Gollier, C., 2001. The Economics of Risk and Time. MIT Press, Cambridge, MA.
Gmez-Valle, L., Martnez-Rodrguez, J., 2008. Modelling the term structure of
interest rates: An efcient nonparametric approach. Journal of Banking and
Finance 32, 614623.
Y. Nishiyama/ Journal of Banking & Finance 35 (2011) 8294 93
Hall, A.D., Anderson, H.M., Granger, C.W.J., 1992. A cointegration analysis of
Treasury bill yields. Review of Economics and Statistics 74, 116126.
Hannan, T.H., Berger, A.N., 1991. The rigidity of prices: Evidence from the banking
industry. American Economic Review 81, 938945.
Hansen, L.P., 1982. Large sample properties of generalized method of moments
estimators. Econometrica 50, 10291054.
Hansen, L.P., Singleton, K.J., 1983. Stochastic consumption, risk aversion, and
the temporal behavior of asset returns. Journal of Political Economy 91,
249265.
Hofmann, B., Mizen, P., 2004. Interest rate pass-through and monetary
transmission: Evidence from individual nancial institutions retail rates.
Economica 71, 99123.
Humphrey, D.B., 1985. Costs and scale economies in bank intermediation. In:
Aspinwall, R.C., Eisenbeis, R.A. (Eds.), Handbook for Banking Strategy. John
Wiley & Sons, New York.
Hutchison, D.E., 1995. Retail bank deposit pricing: An intertemporal asset pricing
approach. Journal of Money, Credit, and Banking 27, 217231.
Jiang, G., Yan, S., 2009. Linear-quadratic term structure models Toward the
understanding of jumps in interest rates. Journal of Banking and Finance 33,
473485.
Kashyap, A.K., Stein, J.C., 2000. What do a million observations on banks say
about the transmission of monetary policy? American Economic Review 90,
407428.
Klein, M.A., 1971. A theory of the banking rm. Journal of Money, Credit, and
Banking 3, 205218.
Koch, T.W., MacDonald, S.S., 2003. Bank Management, Fifth ed. Thomson South-
Western, Mason, OH.
Kok Srensen, C., Werner, T., 2006. Bank Interest Rate Pass-through in the Euro
Area: A Cross Country Comparison. Working Paper Series 580, European Central
Bank.
Liu, C.-C., Ryan, S.G., 2006. Income smoothing over the business cycle: Changes in
banks coordinated management of provisions for loan losses and loan charge-
offs from the pre-1990 bust to the 1990s boom. Accounting Review 81, 421
441.
Liu, M.-H., Margaritis, D., Tourani-Rad, A., 2008. Monetary policy transparency and
pass-through of retail interest rates. Journal of Banking and Finance 32, 501511.
Lucas, R.E., 1978. Asset prices in an exchange economy. Econometrica 46, 1429
1445.
Mehra, R., Prescott, E.C., 1985. The equity premium: A puzzle. Journal of Monetary
Economics 15, 145161.
Nautz, D., Schmidt, S., 2009. Monetary policy implementation and the federal funds
rate. Journal of Banking and Finance 33, 12741284.
Newey, W.K., West, K.D., 1987. A simple, positive semi-denite, heteroskedasticity
and autocorrelation consistent covariance matrix. Econometrica 55, 703708.
Poole, W., Rasche, R.H., 2003. The impact of changes in FOMC disclosure practices
on the transparency of monetary policy: Are markets and the FOMC better
synched? Federal Reserve Bank of St. Louis Review, January/February, pp. 19.
Prisman, E., Slovin, M.B., Sushka, M.E., 1986. A general model of the banking rm
under conditions of monopoly, uncertainty, and recourse. Journal of Monetary
Economics 17, 293304.
Ratti, R.A., 1980. Bank attitude toward risk, implicit rates of interest, and the
behavior of an index of risk aversion for commercial banks. Quarterly Journal of
Economics 95, 309331.
Sealey, C.W., 1980. Deposit rate-setting, risk aversion, and the theory of depository
nancial intermediaries. Journal of Finance 35, 11391154.
Shiller, R.J., 1990. The term structure of interest rates. In: Friedman, B.M., Hahn, F.H.
(Eds.), Handbook of Monetary Economics, vol. 1. North-Holland, Amsterdam.
Slovin, M.B., Sushka, M.E., 1979. An economic model of the market for negotiable
certicates of deposit. Journal of Monetary Economics 5, 551568.
Stock, J.H., Watson, M.W., 1988. Testing for common trends. Journal of American
Statistical Association 83, 10971107.
Stock, J.H., Watson, M.W., 1999. Business cycle uctuations in US macroeconomic
time series. In: Taylor, J.B., Woodford, M. (Eds.), Handbook of Macroeconomics,
vol. 1A. North-Holland, Amsterdam.
Toda, H.Y., Phillips, P.C.B., 1993. Vector autoregressions and causality. Econometrica
61, 13671393.
White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a
direct test for heteroskedasticity. Econometrica 48, 817838.
94 Y. Nishiyama / Journal of Banking & Finance 35 (2011) 8294