TIME-VARYING FORWARD BIAS AND THE VOLATILITY OF RISK
PREMIUM: A MONETARY EXPLANATION
Juan Ángel Lafuente (*)
Universidad Jaume I
Unidad Predepartamental de Finanzas y Contabilidad
Campus del Riu Sec, 12080, Castellón
Jesús Ruiz (*)
Instituto Complutense de Análisis Económico (ICAE)
Universidad Complutense
Campus de Somosaguas, 28223 Madrid
ABSTRACT
_________________________________________________________________________________
Forward exchange rate unbiassedness is rejected for international exchange markets. This paper proposes
a stochastic general equilibrium model which generates substantial variability in the magnitude of
predictable excess returns. Simulation exercises suggest that high persistency in the monetary policy
produces greater bias in the estimated slope coefficient in the regression of the change in the logarithm
of the spot exchange rate on the forward premium. Also, our model suggest that the nature of the
transmission between monetary shocks can explain the excess return puzzle. Empirical evidence for the
US-UK exchange rate according to our theoretical results is provided.
_________________________________________________________________________________
RESUMEN
_________________________________________________________________________________
La insesgadez del tipo de cambio forward es rechazada para los mercados cambiarios internacionales.
Este trabajo propone un modelo de equilibrio general dinámico y estocástico que genera variabilidad
suficiente en las magnitudes de los excesos de rendimientos predecibles. Los ejercicios de simulación
realizados sugieren que una alta persistencia de la política monetaria produce un mayor sesgo en el
coeficiente estimado de la pendiente de la regresión entre la primera diferencia del logaritmo del tipo de
cambio spot sobre la prima forward. Además, nuestro modelo sugiere que la naturaleza de la transmisión
entre shocks monetarios puede explicar la paradoja del exceso de rendimiento. Por último,
proporcionamos evidencia empírica de acuerdo con nuestros resultados teóricos para el tipo de cambio
entre EEUU y Reino Unido.
_________________________________________________________________________________
(*) Los autores agradecen al profesor Alfonso Novales sus valiosos comentarios y sugerencias. Jesús Ruiz agradece
también los comentarios y sugerencias proporcionados por los miembros del CREB así como los medios que han puesto a su
disposición durante el tiempo que estuvo invitado como visitante (febrero y marzo de 2002). Agradecemos el apoyo financiero
del Ministerio de Educación a través de las becas PB98-0831 y BEC2000-1388-C04-03.
1
I nt r oduct ion
The puzzle of biasedness on forward exchange rat e refers that the estimated
slope coe¢ cients in the regression of the change in the logarithm of the spot
rate on the forward premium signi…cantly departs from one (see Zhu (2002) ,
Tauchen (2001), Baillie and Bollerslev (2000), Baillie and Ostenberg (2000)
and M cCallum (1994), among many others). Such discrepancy from the
underlying valuein theuncovered int er est rate parity impliesthat the forward
rate is not an unbiased predictor of the future spot rat e, suggesting the
possibility of unexploited pro…t opportunities. Potential explanations of this
excess return puzzle generally are assigned to three kind of categor ies: a) the
most popular is that such pattern arises as a consequence of a time-varying
risk premia (see Fama (1984)); b) a second explanation relies on the nature
of expectations. Under no rational expectations agents do not e¢ ciently use
the available informat ion set, incurring in systematic forecasting errors over
a signi…cant number of time periods ahead (see Froot and Frankel (1989)) ;
and c) the peso problem, that is, market part icipants anticipate by rational
learning process a future discrete shift in policy that is not performed within
the sample period analyzed (see Lewis, (1995) ).
Even though a substantial number of studies have addressed the ability of
general equilibrium models related to t he Lucas (1982) model to explain the
forward premium puzzle (see, for example, Hodrick (1989), M acklem (1991) ,
Canova and Marrinan (1993), Bekaert (1994)), they unsuccessfully explained
the substantial variability that occur in the magnitude of predictable excess
returns. Currently, there is no conclusive theory explaining the behavior of
the bias of test for a risk premium in forward exchange rates, and it is yet
regarded as one of the most impor tant unresolved puzzle in international
…nance.
In this paper we develop a theoretical general equilibrium model to explain short and long-run risk premium in forward markets for foreign exchange that, not only pr ovide additional insights about the potential explaining factor s of the for ward risk premium, but also reproduce the forward
pr emium anomaly under rational expectations. T he model takes as benchmark the Dutton’s model (1993) which is based on the general equilibrium
models of Lucas (1982). Our model extend the …rst one in three ways: a)
we consider two forward (one and two periods) exchange rates as a hedge
instruments for spot exchange rate a more realistic approach to real markets in where di¤er ent time to maturity can be traded. T his enriches the
2
analysis because of it would be possible to identify t he e¤ect of the time
to maturity in the derivative contract on the forward market risk premia,
b) it is considered the possibility that domestic and foreign consumptions
goods will be complementary or substitutes. Therefore, the model allows to
estimate the impact of the nature of consumpt ions goods. If, for example,
dollars are relatively risk, the uncertainty about the future spot exchange
should a¤ect di¤erently on the forward risk pr emia under complement aries
or substitutes consumption goods, and c) the weight of each, domestic and
foreign, consumption good in the utility function is not necessary the same.
Consequently, a broad set of scenarios can be simulated in order to explore
for potential explanatory factors of the risk pr emium.
The solut ion of the model involves to evaluate expectations of nonlinear
expressions. Therefore, numerical solutions are provided. Under the assumption of rational expectations, our solution method allows to solve jointly for
both prices and positions in one and two-periods ahead forward contr acts.
This is an interesting extension relative to the Dutton‘s solution method.
Simulation exercises are carried out with a variety of parameter values,
revealing that, even when the econometric bias behind the regression of the
change in the logarithm of the spot rate on the forward premium is taken
into account, the model can reproduce the bias for forward exchange rate to
pr edict the future evolution of spot rate. The model suggest what are the
key factor generating high volatility for the risk premium is the persistence of
the monetary policy. Under a relative high persistence the estimated slopes
dr amatically decreases below one. M oreover , theoretical result s show that
the time to maturity of forwar d contract is negatively correlated with the
size of the slope coe¢ cient in the regression, that is, the estimated slopes
corresponding to the long time to maturity cont ract are relatively lower.
In accor dance with the theoretical simulations of the model, the paper
reports empirical evidence for the US dollar -British pound, not only on the
relationship between the correlation of the monetary shocks and the size
of estimated slopes, but also on the linkage between the persistence of the
monetary policy and the bias for forward exchange rate.
The r est of the paper is organized as follows: sect ion 2 present the model.
In section 3 simulations of risk pr emium are presented and theoretical r esults
about the bias of forward premium are provided. Section 4 refers empirical
evidence for the US-UK exchange rat e. Finally, section 5 summarizes and
makes concluding remarks.
3
2
T he M odel
There are two countries with its own currency and a single consumer. In each
count ry the representative …rm r eceives an endowment of a single traded
good. T he only tradable …nancial assets are the money forward periods
exchange contracts. Also, there is no cont ingent claims markets, so all possibilities to reduce risk are concerning the forward exchange market, where
two maturity contracts are available.
The two consumers own titles to the …rms in their respective countries.
The timing of the model can be summarized as follows: 1) at the beginning
of each period, both …
rms pay to the respective consumers in its count ry
a dividend equal to all incomes achieved the previous period. Then, the
consumer turns in its dividends for a new money, and t he old money becomes
wor thless. This implies that all money will be spent; 2) after receiving the
money supply, consumers liquidate t heir forward contracts traded in foreign
exchange in the two previous periods, 3) consumers spend their money on the
two goods. Domestic goods must be purchased with its own currency. All
transactions take place at equilibr ium prices. 4) At the end of each period,
consumers make forward contracts to delivery of currency in the next two
periods.
Endowments of goods and money supplies are stochastic, and its natural
logarithm follow an autoregressive process with a Nor mal innovation. Let us
to denote X t and M t for any good endowment or money supply, respectively:
ln X t = ¹ X (1 ¡ ½X ) + ½X ln X t¡ 1 + »X ;t ;
³
»X ;t ~N 0; ¾2X
´
³
ln X t¤ = ¹ X ¤ (1 ¡ ½X ¤ ) + ½X ¤ ln X t¡¤ 1 + »¤X ;t ;
»X ¤ ;t ~N 0; ¾2X ¤
ln M t = ¹ M (1 ¡ ½M ) + ½M ln M t¡ 1 + »M ;t ;
»M ;t ~N 0; ¾2M
ln M t¤ = ¹ M ¤ (1 ¡ ½M ¤ ) + ½M ¤ ln M t¡¤ 1 + »¤M ;t ;
³
³
,
´
´
»M ¤ ;t ~N 0; ¾2M ¤
(1)
,
,
´
(2)
(3)
, (4)
wheret he asterisk denotest he for eign country. Correlations between any four
shocks (½M M ¤ ; ½X X ¤ ; ½M X ; ½M X ¤ ; ½M ¤ X ; ½M ¤ X ¤ ) are initially restricted to
be zero.
2.1
T he Consum er ’s problem
The utility function of the home consumer is a CES function:
4
1
[Á( CD;t ) ² + (1 ¡ Á) (CF ;t ) ² ]( 1¡ ° ) =² ,
(5)
1¡ °
where CD ;t and CF;t are the consumption levels of domestic and foreign goods
at time t , ° is the coe¢ cient of r elative risk aversion, and 1¡1 ² is the elasticity of substitution, and Á is the weight for each consumption good. If ²
approaches to zero consumption goods becomes more substitutes, whereas
complementary arises when ² approaches to one. The parameter Á measures
the weight of each consumption good in the utility function. T he optimization problem for the home consumer is:
Ut =
M ax
E0
" 1
X
1
¯
[Á(CD ;t ) ² + (1 ¡ Á) (CF;t ) ² ](1¡
1¡ °
t= 0
t
#
° ) =²
(6)
f CD;t ; CF;t g
s:t:
PD ;t CD;t + St PF;t CF ;t · Yt ,
S ¡ Ft¡ 1;1
S ¡ Ft¡ 2;2
Yt = M t + Tt¡ 1;1 t
+ Tt¡ 2;2 t
,
Ft¡ 1;1
Ft¡ 2;2
where PD;t and PF;t are the prices of domestic and foreign goods at time t,
Yt is the total income in period t, St is the spot exchange rate, Ft¡ 1;1, Ft¡ 2;2
are the prices for the two maturity forward contracts available, Tt¡ 1;1 and
Tt¡ 2;2 are the respective amount of its currency that the home country sold
forward in the two previous periods. The money supply (M t ) plus the pro…ts
on each forward currency trading in period t equals the t otal home income.
A similar optimization problem can be pointed out for the foreign consumer ,
that is:
M ax
E0
¤
f CD;t
; CF¤ ;t g
s:t:
" 1
X
#
³
´ ² i (1¡ ° )=²
1 h ³ ¤ ´²
¤
¯
Á CD;t + (1 ¡ Á) CF;t
(7)
1
¡
°
t= 0
t
¤
PD;t CD¤ ;t + St PF ;t CF;t
· Yt ¤St ,
Yt¤ = M t¤ + Tt¡¤
1;1
St ¡ Ft¡ 1;1
+ Tt¡¤
Ft¡ 1;1St
5
2;2
St ¡ Ft¡ 2;2
.
Ft¡ 2;2 St
2.1.1
Opt im al good choices.
In any period t the home consumer chooses levels of CD ;t and CF ;t that
maximize Ut subject to the level of total home income. First order conditions
for choice of CD ;t and CF ;t ar e:
[Á(CD;t ) ² + (1 ¡ Á) (CF ;t ) ² ]
[Á(CD;t ) ² + (1 ¡ Á) (CF ;t ) ² ]
1¡ °
²
1¡ °
²
¡ 1
¡1
(CD;t ) ² ¡ 1 ¡ ¸ t PD ;t = 0 ,
(8)
( CF;t ) ² ¡ 1 ¡ ¸ t St PF ;t = 0 ,
(9)
Yt ¡ PD;t CD ;t ¡ PF ;t St CF ;t = 0 .
(10)
From 8 and 9 yields the following relationships:
"
CF;t =
(1 ¡ Á) PD;t
ÁPF;t St
#¾
CD;t ,
(11)
where ¾ = 1¡1 ² is the elasticity of substitution. Using 11 and the budget
constr aint, t he demand function for the domestic good is the following:
CD ;t =
1¡ ¾
PD;t
+
³
Yt PD¡ ;t¾
´
1¡ Á ¾
(St PF;t ) 1¡ ¾
Á
.
(12)
Substituting 12 into equation 11 the next demand function for t he foreign
good arises:
"
CF ;t =
(1 ¡ Á) PD ;t
ÁPF;t St
#¾
1¡ ¾
PD;t
+
³
Yt PD¡ ;t¾
´
1¡ Á ¾
(St PF;t ) 1¡ ¾
Á
(13)
Similar equation to (12) and (13) can be easily found for the foreign country:
"
CF¤ ;t
¤
CD;t
=
=
(1 ¡ Á) PD ;t
ÁPF;t St
PD1¡;t ¾ +
³
#¾
CD¤ ;t
Yt¤ St PD¡ ;t¾
´
1¡ Á ¾
(St PF ;t ) 1¡ ¾
Á
¤
Substituting 15 into 14 we obtain the analytical expression for CF;t
.
6
(14)
(15)
2.1.2
For war d Cont r act ing
As well as the allocation of current r esources between the two goods, the
home consumer choose in per iod t the levels of the one and two periods
forward contracting, that is Tt;1 and Tt;2 . The Euler conditions are:
"
Et ¸ t+ 1¯
Ã
"
Et ¸ t+ 2¯
St+ 1 ¡ Ft;1
Ft;1
t+ 1
Ã
St+ 2 ¡ Ft;2
Ft;2
t+ 2
! #
= 0 ,
(16)
= 0 ,
(17)
! #
where Et denotes the conditional expectation to the information set available
in period t. From 16:
Et [¸ t+ 1St+ 1] = E t [¸ t+ 1Ft;1 ] ,
and taking into account 8 yields:
Ft;1 =
h
i
@Ut+ 1
1
@CF ;t+ 1 PF ;t + 1
h
i
Ut + 1
1
Et @@
CF ;t + 1 PF ;t + 1 St + 1
Et
.
(18)
Similar r earranging from 17 when taking into account 8 leads t o the following
expression for the two-periods forward price:
Ft;2 =
h
i
@Ut+ 2
1
@CF ;t+ 2 PF ;t + 2
h
i
Ut + 2
1
Et @@
CF ;t + 2 PF ;t + 2 St + 2
Et
.
(19)
Analogous expr essions to (18) and (19) can be obtained when the foreign
consumer chooses in period t the levels of the one and two periods forward
¤ and T ¤ :
contracting, that is Tt;1
t;2
·
Ft;1 =
Et
·
Et
Et
·
Et
¸
@Ut¤+ 1
1
@CF¤ ;t + 1 PF ;t + 1 St + 1
·
Ft;2 =
¤
@Ut+
1
1
¤
@CF ;t+ 1 PF ;t + 1
¤
@Ut+
1
2
¤
@CF ;t+ 2 PF ;t + 2
,
(20)
¸
.
(21)
¸
@Ut¤+ 2
1
@CF¤ ;t + 2 PF ;t + 2 St + 2
7
¸
2.2
2.2.1
M arket -Clear ing
Equilibr ium in t he Goods M ar ket .
The world constraints on consumptions of the two traded goods in both
count ries implies that the total endowment of the two goods must be equal
the consumption of each good in the respective countries, that is:
¤
CD;t + CD;t
= X D ;t ,
(22)
¤
CF;t + CF;t
= X F;t .
(23)
Equilibrium prices of the two goods depend on t he home and foreign money
suppliesas well as their total endowment in each countr y. Taking into account
that a) money is worthless after each period and b) each country’s good only
can be purchased with the country’s curr ency, the following cash-in-advance
spending constraints must be hold:
PD ;t X D;t = M t ,
(24)
PF;t X F;t = M t¤ .
(25)
Since goods endowments X D;t and X F;t , and money supplies M t and M t¤ are
exogenous, the two above equations determine pr ices of consumpt ion goods.
The solution of the model requires the evaluation of expectations in equations 18 and 19, in where highly non-linear expressions appear. This avoids
the possibility of an analytical solution. Appendix 1 provides detailed explanation about the solution method to obt ain simulated equilibrium in spot
and forward exchange markets. It allows the joint search of all variables
(prices and positions) concerning the forward market. In equilibrium, the
following relationships between home and foreign derivative positions holds
Tt ¡ l ;l = ¡ Tt¡¤
3
l;l
, l = 1; 2.
(26)
Simulat ion of for war d pr i ces and r isk pr emiums
The equilibrium spot rates can be obtained as follows: using the budget
¤ = Y ¤S
constr aints, PD ;t CD ;t + St PF;t CF;t = Yt and PD ;t CD¤ ;t + St PF;t CF;t
t t
8
and equations (11) and (14) , we can solve analytically the spot exchange as
a function of the exogenous stochastic variables X D;t ; X F ;t ; M t ; M t¤:
Ã
1 ¡ Á X F;t
St =
Á
X D;t
3.1
! "
Mt
.
M t¤
(27)
D e…nit ion of R isk Pr em ium
To avoid the implications of Siegel’s paradox we use the following de…nition
of the r isk premium in the forward mar ket:
r pt;t+ l = f t;l ¡ E t (st+ l ) ,
l = 1; 2.
(28)
where Et (¢) denotes the mathematical expectation conditioned on the set of
all relevant infor mation at time t, st is the logarithm of thedomestic currency
pr ice of foreign cur rency at time t and f t;l is the logarithm of the forward
exchange rate with delivery at time t + l.
3.2
3.2.1
Par am et er scenar ios w her e t he for war d pr em ium
anom aly ar ises
Test ing t he unbiasedness hyp ot hesis
The main objective of the paper is to analyze the parameter set that could
repr oduce the forward premium bias. T he central hypothesis that we analyze
in this paper is the Uncovered Interest Rate Parity (UIP) condition, which
states that:
Et (¢ st+ l ) = f t;l ¡ st = i t ¡ i ¤t;
(29)
where Et denotes the conditional expectation to the informat ion set available
on time t; i t and i ¤t are the interest rates on domestic and foreign deposits,
respectively, and ¢ denotes the …rst di¤er ence operator , that is, ¢ st+ l ´
st+ l ¡ st+ l¡ 1.
To test for unbiasedness hypothesis, the lit er ature has widely focused on
the following regression relating the change in the spot rate to the forwardspot spread:
¢ st+ l = ®l + ¯ l (f t;l ¡ st ) + ut+ l;l ;
9
(30)
The estimat ion of equation (30) tries to test the ability of the forwardspot di¤erential to forecast the direction of change in spot rate. Regardless
the sampling frequency, the UIP condition implies that ®l = 0 and ¯ l = 1.
However, empirical evidence has widely reported on estimated slopes that
turn out to be below than one or even negatives1. This …nding not only
reject the UIP condition, but also is contradictory with either form of the
expectations hypothesis.
The analytical expression for the OLS estimation of ¯ l is:
¯ ol s =
Cov (f t ;l ¡ st ; st+ l ¡ st )
,
V ar (f t;l ¡ st )
(31)
where V ar (¢) refers to variance, and Cov (¢) denotes the covariance. As
pointed out in Engel (1996), if the estimator is consistent , under rational
expectations it follows that:
³
´
p lim ¯ ol s = 1 ¡ ¯ r p
where ¯ r p =
Cov ( E t (s t+ l )¡ st ; f t ;l ¡ E t ( st + l ) ) + V ar ( f t ;l ¡ E t ( st + l ) )
V ar ( f t;l ¡ st )
(32)
. From t his expression
it can be obser ved that low values of ¯ ols can be explained under rational
expectations if Var (f t ;l ¡ E t (st+ l )) is enough large. The risk premium is
widely considered themost likely source of thepuzzle, but taking into account
the regression results reported in the literature the required volatility are far
larger than most researchers would accept. One of the major task in the
lit er ature concerns to explain why the risk premium has such a large variance.
Our model provide some insights about t his issue.
3.2.2
T heor et ical R esult s
In all numerical simulations the discount factor ¯ and t he relative risk aversion ° are constant and equal to 0:99 and 1:50, respectively2. We consider
a variety of scenarios than can be summarized as follows: a) we focus the
analysis on the e¤ects of the monetary policy (we leave further work the
analysis of the e¤ects of real shocks on risk premia in for ward markets for
1A
recent survey can be found in Engel (1996) .
values inside t he int erval [0:90; 0:99] and [1:10; 5:00] for ¯ and ° , lead t o
similar r esult s t o t hose report ed in t he paper.
2 Paramet er
10
foreign exchange). T herefore, only a uncertainty source is considered: monetary shocks. This way we consider either one or two shocks; b) we distinguish
between situations in where there is no persistencein theshocks of both countries from other ones in where only the home country have persistence in the
monetary shock 3. The nature of the int er action between monet ary policies is
also examined. When two shocks are considered we allow for three possibilities: uncorrelated, positive and negatively correlated monetary shocks. The
considered absolute value for t he correlation coe¢ cient between domestic and
foreign shocks is 0.9. To summarize the theoretical results from estimating
equat ion 30 using simulated spot and forward exchange rates with " = 14 ,
Table 1 reports the volatility of the forward premium and Figures 1 to 10
(Appendix 3) depict the estimated slopes as a function of the correlation between monetary shock and the persistence of the monetary policy when only
a monetary shock is considered. Also we provide the con…dence intervals at
the 5% signi…cance level based on the simulated distribut ion of slopes with
one hundred of theor et ical observations. Several interesting questions emerge
from this information set:
1. The estimated slopes are generally lower than one, a consistent …nding
with expression 30. This means that ¯ r p > 0. This …nding has been
documented in many empirical studies (see, for example, Bilson (1981) ,
Fama (1984), Bekaert and Hodrick (1993), Backus et al. (1993) and
Mark et al. (1993)).
2. There is a negative relationship between the estimated slope coe¢ cient
and the time to maturity. In the long-run the forward bias is greater
than in shor t-r un, re‡ecting a higher uncertainty in the futures evolution of spot rates,
3. A relative higher persist ence in the monetar y policy pr oduces lower
estimated value for the slope. This …nding is consistent with those
reported in Baillie and Bollerslev (2000) . Those authors simulate forwar d premiums. according to a highly stylized UIP-FIGARCH model
(Fractionally Integrated GARCH model), showing that a long memory
3
Under no corr elat ion bet ween monet ar y shocks t his sit uat ion can be int erpret ed as
t he home count ry behaves as a leader since it can updat e t he forecast ing of money supply.
The considered aut oregressive paramet er is 0.9.
4 Similar result s ar e found wit h " = 0, which are available fr om t he aut hors upon
request .
11
in the forward premium produces wide disper sion in the slope coe¢ cients. Tauchen (2001) simulates the sampling distribution of t he slope
coe¢ cient in equation (30), showing that such to be the case when
spot rates are generated with a near to non-stationary AR(1) process.
This is not surprising when equation (27) is observed. Under high
persistence in the monetary policy of the domest ic countr y, spot rate
is very autocorrelat ed, and consequently the forward premium should
have high persistence. The negative relationship is clearer when.
4. More interestingly, our model suggest that under a relative high persist ence in t he domestic/ foreign monetary policy the volatility of the
forward premium is greater. From table 1, it can be observed t hat the
volatility under persistence is above …ve times the volatility that corresponds to the case where monetary policy forecast can not be updated
using current information.
5. Also, the transmission of the monetary policy e¤ects between both
count ries appears to be a signi…cant factor to explain departures from
the UIP. Under a relat ive high persistence, the estimated slope show
higher discrepancy with t he unitary value when monetary shocks are
positively correlated. Indeed the maximum median anomaly for all
simulations appears when shocks are positively cor related and the domestic monetary policy is very persistent. This a realist ic scenario for
most of empirical studies that analyses the exchange rate between US
and other country, which generally takes as a benchmark the Fed´ s
monetary policy. Deviations from the UIP condition are negligible regardless the correlation between the monetary shocks only under no
persistence in the monetary policy of both countr ies.
But, what about the ability of t he model to generate bias for forward
exchange rate?. To answer this question Figures 11 to 20 depict the sum of
the asymptotic bias plus the median estimated slopes coe¢ cients and their
corresponding con…dence intervals at the 5% signi…cance level, again using
one hundr ed of theoretical observations. T hose graphs show two relevant
aspects:
1. once we have …ltered the econometric bias a discrepancy with the unitary value remains, r evealing that the theoretical model can generate
a bias for forward exchange rate.
12
2. the relative persistence in the monetary policy appears as the key factors behind the forward unbiasedness. It can be observed that the con…dence intervals are larger enough under a relative high persistence,
suggesting a higher variability in the potential estimated slopes. In
particular, when the two countries apply persistence in the monetary
policy the con…dence intervals are much less informative. M oreover ,
under such scenario the corr elation between monetary shocks is an additional factor that explain the forward bias, revealing a higher median
deviation fr om one when shocks are positively rather than negatively
correlated (see Figures 19 and 20). Also in this case the con…dence
intervals are less informative than under no
In summary, our model suggest t hat the anomaly should appear when
one country act as a leader when monetary policy is implemented and a high
persistence is applied. Such is the case in most of empirical analysis that
concerns the dollar exchange rate. In the next section we provide empirical
evidence about this.
4
Em pi r ical evi dence. T he U S dollar -B r it ish
pound ex chage r at e
In this section we provide empirical evidence focusing not only on the relationship between the t ransmission of monetary shocks and slope coe¢ cients,
but also on the link between the monetar y persistence and the bias for the
US dollar-British puund forward exchange rate. The consider ed time to maturity is one month and the sample period covers from December, 1986 to
November 2001.
The model pr edict a negative relationship between the estimated slopes
and the correlation between monetary shocks. Figure 21 show the XY plot
of the r olling correlat ion between the M 1 cyclical components5 of US and
UK and the rolling slopes using one month time to maturity US-UK forward
exchange rate for the already referred sample period. The window size to
compute the rolling statistics corresponds to …ve years. The US average
rolling persistence in this period was 0.77. Clearly, and according with our
theoretical results, a negative relationship ar ises. To quantitative account for
this statement , we perform the following regression:
5 The
Hodr ick-Prescot t …lt er is used t o det rend t he monet ary aggregat e.
13
¯ t = ±0 + ±1½M M ¤ ;t + ul;t l = 1; 2
(33)
where ¯ t denotes the actual rolling slope and ½M M ¤ ;t is the correlation coe¢ cient between the cyclical M1 components. The …tted line is: ¯^ l;t =
0:88(0:09) ¡ 2:39 (0:21) ½M M ¤ ;t where standard errors are in parentheses. The
R-squared becomes 0.42.
Also, the model suggests a negative relationship between the estimated
slopes coe¢ cients and the persistence of the monetar y policy. Figure 22 depict the rolling persistence of the UK monetary policy and the corresponding
rolling slopes using a …ve years moving window over the above referred sample. It can be observed that in three subsamples a negative relationship
appears, suggesting that additional factors are a¤ecting the forward bias
along the overall sample.
5
Sum mar y and concludi ng r emar k s
In this paper we examine the bias of tests for a risk premium in forward exchange rat es which refers to signi…cant discrepancies with t he unitary value
in the estimat ed slope coe¢ cients from regressions of the change in the logarithm of the spot rate on the forward premium. We perform a theoretical
analysis by extending t he dynamic and stochastic general equilibrium model
with goods endowment proposed in Dutton (1993). Our contribution is the
intr oduction of a two-period forward contract in the derivative market . Also,
a solution method under rational expectations is provided.
Our main objective is to explore the e¤ects of the monetary policy and
their interactions between the domest ic and for eign country on the behavior
of the risk premium in order to explain the inconsistency wit h the UIP condition. Our simulations results suggest that a high persist ence in the domestic
monetary policy produces gr eater volatility in the forward premium, and
consequently the estimated slope coe¢ cients show greater deviations from
one. M or eover, the nature of the t ransmission between monetary shocks is a
potential explaining factor for excess return puzzle. Under persistence, the
estimated slopes dramatically decrease below one when monet ary shocks are
positively correlated. Finally, we …nd that the time to maturity of the derivative contract is positively related with the bias of risk premium in forward
exchange rates. The UIP condition only holds in the absence of persistence
14
when monetary shocks are uncorrelat ed. or negatively correlated. However ,
this is an unlikely scenario for most of developed economies.
The paper provides empir ical evidence for the US dollar-British pound
exchange rate. In accordancewith our theoretical results, a negative relationship between the forward bias and the UK monetary persistence is observed
along three di¤erent subsamples from December, 1986 to November 2001.
Moreover, a negative relationship between the forward bias and the cor relation between monet ary shocks arises during the overall sample, where a high
persistence in the US monetary policy is detected.
While the focus of this paper is the e¤ect of the monetary policy, a similar
analysis can be made taking into account t he presence of both monetary and
real shocks. We leave further work under such scenarios for further research.
R efer ences
[1] Backus, D., Gregory A. and C. Telmer (1993), Accounting for forward
rates in markets for foreign currency, Jour nal of Finance 48, 1887-1908.
[2] Baillie, R.T . and T . Bollerslev (2000), T he forward premium anomaly
is not as bad as you think, Jour nal of International Money and Finance
19, 471-488.
[3] Baillie, R.T . and W. P. Ostenberg, Deviations from daily uncovered
int er est rate parity and t he role of intervention, Jour nal of International
Financial Markets, Institutions and Money 10, 363-379.
[4] Bilson, J. (1981), The ” speculative e¢ ciency” hypothesis, Jour nal of
Business 54, 435-452.
[5] Bekaert, G.(1994), Exchange rate volatility and deviations from unbiasedness in a cash-in-advance model, Jour nal of Inter national Economics 36, 29-52.
[6] Bekaert, G. and R. J. Hodrick (1993), On biases in the measurement
of foreign exchange risk pr emiums, Jour nal of Inter national Money and
Finance 12, 115-138.
[7] Canova F. and J. M arr inan (1993), Pro…ts, risk and uncertainty in foreign exchange markets, Jour nal of Monetary Economics 32, 259-286.
15
[8] Dut ton, J. (1993), Real and monetary shocks and risk premia in forward
market s for foreign exchange, Jour nal of Money Credit and Banking
25( 4), 731-754.
[9] Engle, C. (1996) , The forward discount anomaly and the risk premium:
A survey of recent evidence, Journal of Empir ical Finance 3, 123-192.
[10] Fama, E.F. ( 1984), For ward and spot exchange rates, Journal of Monetar y Economics 14, 319-338.
[11] Froot, K .A. and J.A. Frankel (1989), Forward discount bias: is it an
exchange risk premium, The Quater ly Jour nal of Economics 104, 139161.
[12] Hodr ick, R.J. ( 1989), Risk, uncertainty and exchange rates, Journal of
Monetary Economics 23, 433-459.
[13] Lewis, K .K. (1995). Puzzles in international …nancial market s. In: Gr ossman, G. and K . Rogao¤ ( Eds.). Handbook of International Economics,
vol. 3. North Holland, Amsterdam, pp. 1913-1971.
[14] Lucas, Robert (1982), Interest rates and currency prices in a two-count ry
world, Journal of Monetary Economics 10, 335-59.
[15] Macklem, R.T . (1991), Forward exchange rates in art i…cial economies,
Jour nal of International Money and Finance 10, 365-391.
[16] Mark, N., Wu Y. and W. Hai (1993) , Understanding spot and forward
exchange rate regressions. Ohio State University, Columbus, OH.
[17] McCallum, B.T. (1994), A reconsiderat ion of the uncovered interest rate
parity relationship, Journal of Monetar y Economics 33, 105-391.
[18] Tauchen, G. (2001), The bias of test s for a risk premium in forward
exchange rates, Journal of Empir ical Finance 8, 695-704.
[19] Zhu, Z. (2002), T ime-varying forward bias and the expected excess
return, Jour nal of Inter national Financial Markets, Institutions and
Money 12, 119-137.
16
Appendix 1. Solution Method
This appendix contains the step that we use in the solution method. As
we pointed out in Section 3, the problem concerning the home and foreign
consumer is highly non-linear, not allowing to achieve an analytical solution.
Therefore a numerical approach must be used.
After providing numerical values for the structural parameters involved
in the theoretical economy, that is, f ¯ ; ° ; Á; "; ¾X ; ¾X ¤ ; ¾M ; ¾M ¤ ; ¹ X ; ¹ X ¤ ;
¹ M ; ¹ M ¤ ; ½X ; ½X ¤ ; ½M ; ½M ¤ g, the next stages are:
1. We obtain one hundred realizations for the st ochastic variables X D ;t ;
X F;t ; M t ; M t¤ in each time period t = 1; :::100.
2. One hundred realizations of bot h home and foreign prices of the consumption goods are computed according to equations (24) and (25), in
each time period. Let us to denot e this numer ical set as f (PD;t;i ; PF ;t;i ) ;
i; t = 1; :::100g, where i and t denote the realization and the time period, respectively.
3. Similar numerical set to the previous one for PD and PF is computed for
thespot exchange rate using equation (27), that is, f St;i i ; t = 1; :::100g.
Computation of theforward pricesand derivative positions for the one and
two period ahead traded contract s [Ft;1; Ft;2 ; Tt;1Tt;2]. From equations (11)
and (14), substituting into equations (18) and (19) the following expressions
can be obtained:
"
Et
Ft;1 =
"
Et
=
"¡ 1
CD;t+
1
"¡ 1
CD;t+
1
³
³
ÁCD" ;t+ 1
+ (1 ¡
´
Á)CF" ;t+ 1
´
"
"
ÁCD;t+
1 + (1 ¡ Á)C F;t+ 1
E t [WD ;t+ 1]
,
Et [WD;t+ 1 =St+ 1]
1¡ °
"
1¡ °
"
¡ 1
#
1
PD ;t + 1
¡1
#
1
PD ;t + 1 St + 1
(34)
17
"
Et
Ft;2 =
"
"¡ 1
CD;t+
2
Et
³
"¡ 1
CD;t+
2
³
ÁCD" ;t+ 2
´
Á)CF" ;t+ 2
+ (1 ¡
´
"
"
ÁCD;t+
2 + (1 ¡ Á)C F;t+ 2
1¡ °
"
#
¡ 1
1
PD ;t + 2
1¡ °
"
¡1
#
1
PD ;t + 2 St + 2
E t [WD ;t+ 2]
,
Et [WD;t+ 2 =St+ 2]
=
"
¤ "¡ 1
CF;t+
1
Et
F t;1 =
"
CF¤ ;t+"¡11
Et
³
³
(35)
¤ "
ÁCD;t+
1
¤ "
ÁCD;t+
1
´
+ (1 ¡
+ (1 ¡
1¡ °
"
¤ "
Á)CF;t+
1
¤ "
Á)CF;t+
1
´
#
¡ 1
1
PF ;t+
1¡ °
"
¡ 1
1
#
1
PF ;t + 1 St+
1
Et [WF;t+ 1]
,
E t [WF ;t+ 1=St+ 1 ]
=
"
(36)
³
´
1¡ °
"
¤ "¡ 1
¤ "
¤ "
E t CF;t+
2 ÁC D;t+ 2 + (1 ¡ Á)C F;t+ 2
F t;2 =
"
Et
CF¤ ;t+"¡21
³
¤ "
ÁCD;t+
2
+ (1 ¡
¤ "
Á)CF;t+
2
´
1¡ °
"
¡ 1
#
1
PF ;t+
¡ 1
2
#
1
PF ;t + 2 St+
2
Et [WF;t+ 2]
.
(37)
E t [WF ;t+ 2=St+ 2 ]
We solve jointly Ft ;1, Ft;2, Tt;1 and Tt;2 by searching values that satisfy the
following approximations of the equations (34) to (37):
=
"
P N
"¡ 1
i= 1 CD ;t+ 1;i
Ft;1 =
"
P N
"¡ 1
i = 1 CD ;t+ 1;i
³
³
ÁCD" ;t+ 1;i
ÁCD" ;t+ 1;i
+ (1 ¡
+ (1 ¡
"
Á)CF;t
+ 1;i
Á)CF" ;t+ 1;i
´
´
1¡ °
"
#
¡ 1
1
PD ;t + 1;i
1¡ °
"
¡ 1
#
1
PD ;t + 1;i St+
1;i
P N
=
P N
i = 1 [WD;t+ 1]
i=1
P N
[WD ;t+ 1=St + 1 ]
"
"¡ 1
i = 1 C D;t+ 2;i
Ft;2 =
P N
"
"¡ 1
i = 1 CD ;t+ 2;i
=
³
³
(38)
ÁCD" ;t+ 2
"
ÁCD;t+
2;i
P N
P N
,
i = 1 [WD;t+ 2;i ]
i = 1 [WD ;t+ 2;i =St+ 2;i ]
+ (1 ¡
+ (1 ¡
,
Á)CF" ;t+ 2;i
"
Á)CF;t+
2;i
´
´
1¡ °
"
1¡ °
"
#
¡1
1
PD ;t + 2;i
¡ 1
PD ;t+
#
1
2;i St + 2;i
(39)
18
"
P N
F t;1 =
¤ "¡ 1
i = 1 C F;t+ 1;i
"
P N
¤ "¡ 1
i = 1 CF;t+ 1;i
=
"
¤ "¡ 1
i= 1 CF;t+ 2;i
P N
³
³
P N
¤ "
ÁCD;t+
2;i
i = 1 [WF;t + 2;i ]
i = 1 [WF ;t+ 2;i =St+ 2;i ]
#
¡ 1
1
PF ;t +
1¡ °
"
¡1
PF ;t+
1;i
#
1
1;i St + 1;i
(40)
¤ "
ÁCD;t
+ 2;i
P N
´
1¡ °
"
,
i = 1 [WF ;t+ 1;i =St+ 1;i ]
"
+ (1 ¡
´
Á)CF¤ ;t+" 1
¤ "
¤ "
ÁCD;t+
1;i + (1 ¡ Á)C F;t+ 1;i
i = 1 [WF;t + 1;i ]
P N
¤ "¡ 1
i = 1 CF;t+ 2;i
=
"
ÁCD¤ ;t+
1;i
P N
P N
F t;2 =
³
³
+ (1 ¡
+ (1 ¡
¤ "
Á) CF;t+
2;i
¤ "
Á)CF;t+
2;i
´
´
1¡ °
"
1¡ °
"
#
¡ 1
1
PF ;t + 2;i
¡1
PF ;t+
#
1
2;i St + 2;i
.
(41)
Taking into account that under rat ional expectations Et [Wt+ 1 ] = ª 1 at +
Et¡ 1 [Wt+ 1], where at is a white noise, the expression of the two period forwar d price in t ¡ 1 is:
³
P N
Ft¡
1;2
´
P
N
¡ ª D ;1 WD;t ¡
i= 1 [WD ;t;i ]
³
´ , (42)
= PN
P N
~
[W
=S
]
¡
ª
W
=S
¡
[W
=S
]
D;t+
1;i
t+
1;i
D
;1
D;t
t
D;t;i
t
;i
i= 1
i= 1
i= 1 [WD ;t+ 1;i ]
or equivalently for the foreign consumer:
³
P N
´
P
N
i = 1 [WF;t+ 1;i ] ¡ ª F ;1 WF;t ¡
i= 1 [WF ;t;i ]
³
´ . (43)
Ft¡ 1;2 = P N
P N
~
[W
=S
]
¡
ª
W
=S
¡
[W
=S
]
F ;t+ 1;i
t+ 1;i
F ;1
F;t
t
F ;t;i
t;i
i= 1
i=1
Next, we proceed as follows:
i )We posit initial conditions for the paramet ers { ª
(0) ~ (0)
( 0) ~ ( 0)
D;1ª D ;1ª F ;1ª F ;1} .
ii ) Also, we need an init ial vector. Let us to denote it by f F0;1; F¡ 1;2; T0;1 ;
T¡ 1;1g. Then, one hundred realizations of CD ;1;i ; CF ;1;i ; CD¤ ;1;i ; CF¤;1;i ; Y1;i ; Y1;i¤
in t = 1 trough equations (11) , (14) and the following expressions:
Ã
Y1;i =
M 1;i
¤
Y1;i
=
¤
M 1;i
!
Ã
!
S1;i ¡ F0;1
S1;i ¡ F ¡ 1;2
+ T0;1
+ T¡ 1;2
;
F0;1
F¡ 1;2
Ã
!
Ã
!
S1;i ¡ F0;1
S1;i ¡ F¡ 1;2
¡ T0;1
¡ T¡ 1;2
;
F 0;1 S1;i
F¡ 1;2S1;i
19
CD ;1;i =
CD¤ ;1;i =
¾
PD¡ ;1;i
+
¾
PD¡ ;1;i
+
³
¾
Y1;i PD¡ ;1;i
´
1¡ Á ¾
(S1;i PF ;1;i ) 1¡ ¾
Á
¡ ¾
Y1;i¤ S1;i PF;1;i
³
´¾
1¡ Á
(S1;i PF ;1;i ) 1¡ ¾
Á
¤
iii ) With the pr evious data set, f CD;1;i ; CF ;1;i ; CD;1;i
; CF¤;1;i ; Y1;i ; Y1;i¤ g100
i = 1 , we
iterate using the Gauss-Newton algorithm in the system concerning equations ( 38), (39), (42) and (43). After achieving the …xed point in the space
(F1;1; F0;2; T1;1; T0;2 ) and evaluating in t = 1 with the variables f F1;1; F0;2 ;
T1;1; T0;2 g the corresponding expressions, it is possible to compute values for
¤
CD ;1; CF;1 ; CD¤ ;1; CF;1
; Y1; Y1¤, independently of the realization values.
iv) T he steps ii ) and iii ) are repeated recursively for each time period, allowing to obtain the numerical solutions for the remainder of the sample size,
¤
that is, f CD ;t ; CF;t ; CD;t
; CF¤ ;t ; Yt ; Yt¤ g100
t= 2 : However , this solution depends on
(0) ~ (0)
(0) ~ (0)
the initial condit ion { ª D;1 ª D;1 ª F;1ª F;1} . To …lter this e¤ect, we estimate an
autoregressive pr ocess for t he expr essions of WD ;t ; (WD ;t =St ); WF ;t ; (WF ;t =St )
that can be computed with the simulated series of the previous solution. We
use …ve lags in the AR speci…cat ion, a robust structur e in or der to forecast
theprevious expressions. With the …tted autoregressive processes, estimation
0
(0) ~ (0)
~ (0)
of ª s are recovered to evaluate the discrepancy with (ª (0)
D;1 ; ª D;1 ; ª F;1 ; ª F;1 )
¡ 6
using the euclidean norm. The used convergence criterion is 10 : When
¤ ; C ¤ ; Y ; Y ¤} 100 is the …
the norm is lower, { CD ;t ; CF ;t ; CD;t
nal numerical
t
F;t
t t= 1
solution, whereas the norm is higher we back t o step i) to iterate with the
~ ( 0) ( 0) ~ ( 0)
new initial condition for the vector f ª (0)
D ;1ª D ;1ª F ;1ª F ;1g.
20
Appendix 2. St atist ical Tables.
Table 1. Risk premium volatility
¾2M = 0:005; ½M = ½M ¤ = ½X = ½X ¤ = 0; ¾2M ¤ = ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0017 0.0017
0.0012 0.0012
¾2M = 0:005; ½M = 0:9; ½M ¤ = ½X = ½X ¤ = 0; ¾2M ¤ = ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
2
[V ar (f t;l ¡ E (st+ l ))]
0.0092 0.0092
0.0168 0.0169
2
2
¾M = ¾M ¤ = 0:005; ½M M ¤ = 0; ½M = ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0019 0.0019
0.0023 0.0022
2
2
¾M = ¾M ¤ = 0:005; ½M M ¤ = 0:9; ½M = ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0008 0.0008
0.0006 0.0006
2
2
¾M = ¾M ¤ = 0:005; ½M M ¤ = ¡ 0:9; ½M = ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0020 0.0020
0.0020 0.0020
¾2M = ¾2M ¤ = 0:005; ½M M ¤ = 0; ½M = 0:9; ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
2
[V ar (f t;l ¡ E (st+ l ))]
0.0074 0.0074
0.0153 0.0154
2
2
¾M = ¾M ¤ = 0:005; ½M M ¤ = 0:9; ½M = 0:9; ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0094 0.0094
0.0187 0.0188
2
2
¾M = ¾M ¤ = 0:005; ½M M ¤ = ¡ 0:9; ½M = 0:9; ½M ¤ = ½X = ½X ¤ = 0; ¾2X = ¾2X ¤ = 0
regression with l = 1 regression with l = 2
Á = 0.9 Á = 0.1
Á = 0.9 Á = 0.1
1
[V ar (f t;l ¡ E (st+ l ))] 2 0.0070 0.0069
0.0127 0.0109
Not e: Á measures t he degree of substit ut ability or complement ary.
21
Appendix 3. Figures
Median Estimated slopes
One-period ahead forward premium
One monetary shock
Dashed lines are the bands for the 95% confidence interval
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.9
Figure 1
Median estimated slopes
Two-periods ahead forward premium
One monetary shock
Dashed lines are the bands for the 95% confidence interval
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
persistence in the domestic country
Figure 2
22
0.6
0.7
0.8
0.9
1.4
Median estimated slopes
One-period ahead forward premium
Two monetary shocks. No persistency in the foreign country
Dashed lines are the bands for the 95% confidence interval
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.9
Figure 3
Median estimated slopes
Two-periods ahead forward premium
Two monetary shocks. No persistency in the foreign country
Dashed lines are the bands for the 95% confidence interval
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
Figure 4
23
0.7
0.8
0.9
Median estimated slopes
One-period ahead forward premium
Two monetary shocks. High persist. in the foreign country
Dashed lines are the bands for the 95% confidence interval
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.9
Figure 5
Median estimated slopes
Two-periods ahead forward premium
Two monetary shocks. High persist. in the foreign country
Dashed lines are the bands for the 95% confidence interval
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
Figure 6
24
0.7
0.8
0.9
Median estimated slopes
One-period ahead forward premium
Two monetary shocks. No persistency in both countries
Dashed lines are the bands for the 95% confidence interval
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
0.5
0.7
0.9
Figure 7
Median estimated slopes
Two-periods ahead forward premium
Two monetary shocks. No persistency in both countries
Dashed lines are the bands for the 95% confidence interval
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
Figure 8
25
0.3
0.5
0.7
0.9
Median estimated slopes
One period ahead forward premium
Two monetary shocks. High persist. in the domestic country
Dashed lines are the bands for the 95% confidence interval
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
0.5
0.7
0.9
Figure 9
Median estimated slopes
Two- periods ahead forward premium
Two monetary shocks. High persist. in the domestic country
Dashed lines are the bands for the 95% confidence interval
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
Figure 10
26
0.3
0.5
0.7
0.9
Median estimated slopes + asymptotic bias
One-period ahead forward premium
One monetary shock
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.7
0.8
0.9
Figure 11
Median estimated slopes + asymptotic bias
Two-periods ahead forward premium
One monetary shock
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
Figure 12
27
0.9
Median estimated slopes + asymptotic bias
One-period ahead forward premium
Two monetary shocks. No persistency in the foreign country
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.9
0.7
0.8
0.9
Figure 13
Median estimated slopes + asymptotic bias
Two-periods ahead forward premium
Two monetary shocks. No persistency in the foreign country
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
Figure 14
28
Median estimated slopes + asymptotic bias
One-period ahead forward premium
Two monetary shocks. High persist. in the foreign country
Dashed lines are the bands for the 95% confidence interval
2.5
2.0
1.5
1.0
0.5
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
0.7
0.8
0.9
0.7
0.8
0.9
Figure 15
Median estimated slopes + asymptotic bias
Two-periods ahead forward premium
Two monetary shocks. High persist. in the foreign country
Dashed lines are the bands for the 95% confidence interval
2.5
2.0
1.5
1.0
0.5
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
persistence in the domestic country
Figure 16
29
Median estimated slopes + asymptotic bias
One-period ahead forward premium
Two monetary shocks. No persistency in both countries
Dashed lines are the bands for the 95% confidence interval
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
0.5
0.7
0.9
0.5
0.7
0.9
Figure 17
Median estimated slopes + asymptotic bias
Two-periods ahead forward premium
Two monetary shocks. No persistency in both countries
Dashed lines are the bands for the 95% confidence interval
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
Figure 18
30
Median estimated slopes + asymptotic bias
One period ahead forward premium
Two monetary shocks. High persist. in the domestic country
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
0.5
0.7
0.9
0.5
0.7
0.9
Figure 19
Median estimated slopes + asymptotic bias
Two-periods ahead forward premium
Two monetary shocks. High persist. in the domestic country
Dashed lines are the bands for the 95% confidence interval
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.9
-0.7
-0.5
-0.3
-0.1
0.1
Correlation between monetary shocks
0.3
Figure 20
31
a
t
e
s
l
5 years rolling slopes and rolling correlation
Sample: January 1982 - December 2001
1 month time to maturity forward rate
1.5
p
1.0
0.5
e
0.0
E
-0.5
-1.0
-1.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
negative
correlation
0.30
0.8
o
r
correlation between M1 cyclical components
l
r
Figure 21
1.5
1.0
negative
correlation
negative
correlation
o
Rolling slopes and persistence with 5 years moving window
Sample: 1982, January to December 2001
1 month time to maturity UK-USA forward
0.25
0.20
0.5
0.15
0.10
s
0.0
0.05
-0.5
0.00
-1.0
c
-0.05
-1.5
-0.10
rolling slopes
persist_uk
Figure 22
32