Journal of Macroeconomics 33 (2011) 644655

Contents lists available at ScienceDirect

Journal of Macroeconomics
journal homepage: www.elsevier.com/locate/jmacro

Formal targets, central bank independence and ination dynamics

in the UK: A Markov-Switching approach
William Miles , Chu-Ping Vijverberg 1
Department of Economics, Wichita State University, 1845 Fairmount, Wichita, KS 67260-0078, USA

a r t i c l e

i n f o

Article history:
Received 20 January 2010
Accepted 19 April 2011
Available online 19 May 2011
JEL classication:
E31
E52
E58
Keywords:
Central banks and their policies
Ination
Monetary policy

a b s t r a c t
We examine ination and uncertainty in the UK with a version of the Markov Switching
model, which allows for changes in the variance as well as in the mean and persistence
of a series. We nd that the UKs attempts at exchange rate pegs in the form of shadowing
the deutschmark and entering the ERM were ineffective, and in the latter case counterproductive in lowering ination uncertainty. The 1981 budget, however, greatly lowered
uncertainty, and the adoption of a formal ination target also had a palpable, negative
impact on ination uncertainty. As a suggestive exercise, we examine ination uncertainty
in the US, and nd that, over 20052008, in the absence of an ination target, uncertainty
rose in the US, while uncertainty remained low in the UK over this period of rising commodity prices and nancial turmoil.
2011 Elsevier Inc. All rights reserved.

1. Introduction
Low ination is a major policy goal for both developed and many developing countries. Sadly, in the post-World War II
era, the ination rate in the UK has at times been much higher than hoped for (see Benati, 2008a, for a detailed analysis of
ination in the UK over the Great Ination and Great Moderation periods). Moderate ination in and of itself may be
neutral for long run income. However, ination has been shown to increase ination uncertainty (Friedman, 1977; Ball,
1992), and uncertainty over ination has been shown theoretically and empirically to lower investment and real output.
Accordingly, the UK has tried a number of policies designed to lower ination and its accompanying uncertainty. The
Bank of England was given operational independence in 1997. Prior to this, there were several attempts to peg the exchange
value of the pound, rst by shadowing the deutschmark, and then by entering the Exchange Rate Mechanism (ERM) as a
possible precursor to Euro adoption. Finally, after leaving the ERM in 1992, the Bank of England adopted a formal ination
target. Ination targets have also since been adopted by most other industrialized nations.
The purpose of this paper is to examine the UK ination dynamics with the Markov Switching (MS) technique and, at the
same time, to investigate the impacts of these various polices on UK ination uncertainty. This differs from other papers that
use regression or GARCH models to evaluate ination mean/shock persistence/ination uncertainty. The MS technique utilized here is based on an error correction model in which the mean, shock persistence and uncertainty in two different regimes can be evaluated. Thus, rather than examining changes in the mean, persistence and uncertainty of UK ination
separately or jointly (i.e., any two of the three), we investigate all three issues simultaneously. Furthermore, due to the
Corresponding author. Tel.: +1 316 978 7085; fax: +1 316 978 3308.
1

E-mail addresses: william.miles@wichita.edu (W. Miles), chuping.vijverberg@wichita.edu (C.-P. Vijverberg).

Tel.: +1 316 978 7093; fax: +1 316 978 3308.

0164-0704/$ - see front matter 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmacro.2011.04.003

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

645

special characteristics of the MS model, we are able to obtain the probability of the economy being in the high ination
uncertainty regime. Given the probability derived from the MS model, a probit model is incorporated to assess the impacts
of UKs various policies on ination uncertainty.
We nd that the exchange rate pegs, e.g., shadowing the deutschmark and the ERM, were ineffective in decreasing uncertainty. In the case of the ERM, it actually increased uncertainty over the future path of prices. On the other hand, the 1981
budget lowered ination uncertainty. Moreover, the adoption of the formal ination target had a decidedly negative impact
on ination uncertainty. The ination target did seem to foster clear communication and transparency, thus palpably
decreasing the uncertainty over the future path of prices. While ination rose above its 2% target, prompting a mandatory
explanatory letter from the Bank of Englands governor Mervyn King to the Exchequer starting in April 2007, it is notable that
in our sample, which runs through May of 2008, uncertainty did not pick up at the end. This indicates that, despite the breach
of the target, the public, at least for a year, invested the Bank of England with much credibility.
We investigate this issue further with a comparative exercise. While noting that the effect of ination targeting in the UK
may not be replicated in other countries, it is of interest to contrast the UKs experience with that of the US, as the US is the
largest industrialized country without a formal target, over the last several years of rising commodity prices and nancial
shocks. We nd that over the 20052008 period, while uncertainty in the UK remained low, in the US, without a formal target, uncertainty rose. While one cannot be too bold in drawing clear causal inference from this comparison, it does bolster
the case that in the UK ination targeting did decrease the publics uncertainty of the future path of prices in the face of economic and nancial disturbances.
This paper proceeds as follows. The next section describes the previous literature on ination, uncertainty and policies
designed to decrease both. The third section describes the MS methodology to be employed. The fourth section describes
our results, while the fth section discusses our results in light of previous ndings. The sixth section concludes.
2. Previous literature
Ination has been a source of concern for policymakers in the UK, as elsewhere, due to the potential real economic costs
that a rising price level extracts. While the direct impact on long run output of low or moderate ination may be negligible,
uncertainty about the future price path may be costly. Friedman (1977) posits that the negative real effects of ination arise
mainly from ination uncertainty. Uncertainty about the future price path lowers the information content of prices, therefore making long term contracting difcult. This could lead to lower investment and long run growth (Cukierman et al.,
1993). Grier and Perry (2000) and Grier et al. (2004) provide empirical evidence that ination uncertainty decreases output
growth.
Policymakers thus desire, all else constant, a low rate of ination and an accompanying low level of ination uncertainty.
Additionally, most central bankers nd it desirable to have a low persistence of ination shocks. If shocks to the price level do
not have a signicant long-lasting effect on the future path of prices, the central bank can maintain low ination in the face
of disturbances. For instance, a spike in commodity prices need not require a severe monetary tightening, if the persistence
of shocks is low. Achieving low persistence may necessitate a high level of credibility for the central bank (Cecchetti and
Debelle, 2006; see also Benati 2008b for a discussion of secular changes in UK ination persistence).
While low ination, persistence and uncertainty are desirable, the existence of nominal rigidities makes obtaining and
sustaining a stable price level potentially costly in terms of short run output and employment. This is especially the case,
if, as in the BarroGordon framework, the public mistrusts the central banks commitment to low ination and thus maintains expectations of higher ination. This leads to higher-than-optimal ination in equilibrium.
A possible way out of the BarroGordon conundrum is a commitment mechanism in the form of a formal target, or quantitative goal (the exchange rate, money supply, or ination). Another possible commitment mechanism is to give the central
bank a high degree of independence, so that its decisions are insulated from political pressure for short run monetary stimulation (Cukierman, 1992). There is a large literature, both theoretical and empirical, on the impact of exchange rate targets
on monetary outcomes. Britain may yet go as far as possible in terms of exchange rate targeting by eliminating its currency
and joining the Euro. Even without doing so, its shadowing of the Deutschmark in the 1980s and entrance to the ERM in
the 1990s were variants of exchange rate targets. The desirability of currency targets is still debated. Some nations, especially
in Eastern Europe, still endeavor to join the Euro, while other emerging markets in Asia and Latin America have had devastating balance of payments crises which some economists attribute to inappropriate currency pegs. To anticipate our results,
we nd that Britains recent exchange rate targets were ineffective in generating desired monetary results.
Another goal is a money supply target, which the UK has also employed in the past. The inability to hit such targets in the
UK and elsewhere has led to a wide decrease in popularity (although the European Central bank still maintains a monetary
target as one of its twin pillars of ination and money supply targets). Finally, the difculties of exchange rate and money
supply targets have paved the way for ination targets (IT). An IT is a formal announcement by the central bank of a targeted
rate or range of ination over a coming period. In the wake of the ERM crisis, the Bank of England adopted a formal ination
target in 1992. The UK thus stands with most industrialized (and many emerging market) nations in having an IT, although a
few countries, most notably the United States, have no IT or any formal goal.2
2
Beginning in February 2009, the US Federal Reserve announced that it will publish long-term forecasts of ination, which some interpret as a de facto
ination target. However, there is still no formal ination or any other target for the US central bank.

646

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

Empirically, there are conicting results concerning the impact of targets. Ghosh et al. (1997) nd that exchange rate
rigidity typically lowers the level of ination, while Bleaney and Francisco (2005) nd that only the hardest pegs (currency
boards and unions) have any palpable impact. Fatas et al. (2007) nd that having any of the three formal targets lowers ination, and that ination targets have the greatest negative effect on price increases. On the other hand, Ball and Sheridan
(2005) nd that IT has no signicant effect on the level or the variability of ination.
As for central bank independence, empirical results support the notion that such monetary policy independence lowers
the level of ination (Cukierman, 1992; Alesina and Summers, 1993). The latter paper also nds that independence of the
monetary authority decreases the variability of ination as well.
The above-cited papers are all based on cross-country regressions, and thus may not be reliable for inference regarding
the impact of a given policy or institutional change in a particular country (see Rodrik, 2005 for a discussion of the pitfalls of
cross-country studies of the impact of policies). Our goal in this paper is to examine ination dynamics in the UK in the postwar years and to determine whether and what policies had an impact. In order to do so, we must formulate an empirical
strategy that addresses several important issues.
First, to examine the impact on the level of ination, we must control for potential changes in the persistence of ination
and vice-versa. Cecchetti and Debelle (2006) point out that studies which examine either changes in the mean or persistence
but which fail to control for potential changes in both are likely to lead to erroneous inference. Additionally, given that the
main source of the output costs of ination are through uncertainty, we seek to model this uncertainty and determine
whether it has been signicantly changed through particular policies.
For the mean and persistence of ination, an ARMA model should sufce, as we can test the intercept and slope coefcients for changes. Modeling uncertainty is not as clear-cut. Traditionally, the unconditional variance of ination has
been employed as a proxy for uncertainty, as in Fischer (1981). However, if the determinants of ination are variable,
but predictable, a high unconditional variance of ination will not be a good proxy for uncertainty (Grier and Grier,
2006).
Since the development of ARCH and GARCH models (Engle, 1982; Bollerslev, 1986) the conditional GARCH variance of
the residuals of an estimated ARMA ination specication have been employed as a proxy for uncertainty. Grier and
Perry (1998) for instance, examine ination in the UK, as well as other G-7 countries with GARCH specications. In
the mean time, MS models were developed to capture the changing variances under different regimes (Hamilton,
1989; Kim, 1993; Kim and Nelson, 1999a,b). Note that the changing variances of a GARCH model depend on the previous
innovation within a given structure. It differs from the MS model where changing variances are due to regime changes
in the variance structure. This paper will incorporate changes in the mean, persistence and variance of ination simultaneously in a MS model. We believe that this MS technique can provide different perspectives and/or robust verications about the UK ination behavior.

3. Methodology
To investigate the ination dynamics, we can start with an AR(p) model,

pt a0 b1 pt1 b2 pt2 b3 pt3    bp ptp ut

where pt and pti are the ination rates at time t and t  i, and then transform it by arranging the terms and get the error
correction format,

Dpt a0 q  1pt1 c1 Dpt1 c2 Dpt2    cp1 Dptp1 ut :

P
where q pi1 bi and q is a measure of persistence and c1 = b2 b3; c2 = b3 for an AR(3) model.
Now to examine the changes in the dynamics of the ination process, including uncertainty, we use a MS3 model that
can simultaneously handle changes in the mean, persistence and variance of ination. The previous model can thus be expressed as

Dpt a0 a1 St /0 /1 St pt1 c1 Dpt1 c2 Dpt2    cp1 Dptp1 ut;St

where St is the state variable that denotes the state of the series at time t and /0 + /1St = q  1. In this specication, we
have two different regimes: regime 0 (i.e., St = 0) and regime 1 (i.e., St = 1). A two-state Markov process has the following
~ and PrSt 1jSt1 1 p
~: The uncertainty in each regime is represented
transition probabilities: PrSt 0jSt1 0 q
2
by the variance: Varut;St h0 when St = 0 and Varut;St h0 h1 2 when St = 1. The parameters a1, /1 and h1 capture
the changes in the mean of ination, the persistence of a shock to ination and the variance during regime 1 relative
3
We have benetted from the Gauss programs provided in the book of Kim and Nelson (1999) regarding the Markov-Switching model. However, all the
Markov-Switching programs used in this paper are coded in S+ and R.

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

647

to regime 0. If h1 is positive (negative), it implies that regime 1 has a higher (lower) ination uncertainty than that of
regime 0.4 In this paper, we will use p = 3.5
~. MLE is implemented to estimate the parame~; q
The parameters of the model are thus a0 ; a1 ; /0 ; /1 ; c1 ; c2 ; h0 ; h1 ; p
ters. The log-likelihood function is the following:

lnL

T
X
t3

where

ln

(
1
1
1
X
X
X

f Dpt ; jSt ; St1 ; St2 ; Ut1  PrSt ; St1 ; St2 j; Ut1

St 0 St10 St20

(
)
1
Dpt  a0 a1 St  /0 /1 St pt1  c1 Dpt1  c2 Dpt2 2
f Dpt jSt ; St1 ; St2 ; Ut1 p exp
2r2st
2pr2st

PrSt j; St1 i; St2 kjUt1 PrSt jjSt1 i; St2 kjUt1  St1 i; St2 kjUt1

and
2

where Ut1 denotes past information, r h0 h1 St , and PrSt jjSt1 i; St2 k; Ut1 PrSt jjSt1 i is the transition probability for i, j, k = 0, 1. Note that once the MLE estimates are obtained, we are able to calculate the probability
of St = 1 or St = 0 for each time period6
2
St

4. Results
Monthly data of the UK long-term indicator of prices of consumer goods and services are used. The data are available from
1947:6 to 2008:5.7 The ination rate is calculated as the percentage change of the price index a year ago.8 The average ination
rate of this data set is 5.796%. The ve number summary of the data (i.e., minimum, rst quartile, median, third quartile, and
maximum) is the following: 0.818, 2.755, 4.193, 7.420, and 26.867.
We use Brock, Dechert, and Scheinkman (BDS) test to evaluate non-linearity. BDS is a popular test because of its power
against a variety of nonlinear time series models. For any two m-dimensional points, where m = 2 and 3, we nd that the null
hypothesis of data being iid is rejected. This suggests that UK ination data are not really a candidate for linear modeling.
One may contend that a two-regime MS model may not be optimal. One way to deal with this contention is to implement
three, four and higher regime MS models and to compare their likelihood values. We did not get too far with this technique
because of the difculties in getting convergence in MLE in higher-regime models. An alternative to MLE is the Bayesian
technique that treats both parameters and the state variables as random variables. Higher regime MS models may be implemented through the Gibbs sampling simulation tool, and one may use the marginal data density criteria to nd the appropriate number of regimes. Table 1 shows the results of marginal data density tests of two different priors.9 A four-regime
model is better than one with two regimes under prior I while two regimes are better than four regimes under prior II. Thus,
it is a choice between two and four regimes. As we examined the posterior means and standard deviations of all the parameters
in the four-regime model for both priors, the means and variances of the original ination model in both regimes three and four
have the ratios of posterior mean/ standard deviation much less than 2. This indicates the imprecise estimates of both posterior
4

In our model, we do impose the restriction that h2 > 0 in our computer program. Thus, St = 1 is the higher volatility regime.
Due to the programming complexity in dealing with various state variable statuses at different time points, it is not trivial to have a model with a long lag.
As the lag length of the model increases, the time-dimension of the state variable increases substantially. For example, if the lag length is l (=p  1 as shown in
the equation above), the time-dimension of the state variable is a 2l+1  l + 1 matrix. Thus, for a lag length of l = 2, the time-dimension of the state variable G is
2
3
0 0 0 0 1 1 1 1
an 8  3 matrix, i.e., Gt 4 0 0 1 1 0 0 1 1 5 where the three columns of G represent the statuses of the state variable at t  2, t  1, and t
0 1 0 1 0 1 0 1
respectively. Then for l = 3, G is a 16  4 matrix; for l = 4, G is a 32  5 matrix
P1 P1
P1
6
G i v e n
w e
c a n
u p d a t e
t h e
p r o b a b i l i t y
b y
St 0
St 10
St 20 f Dpt ; St ; St1 ; St2 jUt1 f Dpt jUt1 ,
f Dpt ; St ; St1 ; St2 jUt1 =f Dpt jUt1 f St ; St1 ; S2 jDpt ; Ut1 f St ; St1 ; St2 jUt . Then integrate out St1 and St2, we can get P(St = 1|Ut ).
7
Data were downloaded from http://www.statistics.gov.uk/statbase on June 2008.
8
The ination rates calculated by month to month change of the price index are very noisy. The MS model has difculties in identifying the regime switch,
i.e., we could not get any converged results from using this noisy ination data. The noise was reduced substantially as we calculated ination rates as the
percentage change of the price index a year ago. This is the main reason for using these monthly ination rates. A further justication is the following. If the
noisy ination rates are decomposed into seasonal, trend and noise components, the trend component of the noisy ination rates is very similar to the data
used in this paper. For details, see the attached appendix for referees. Readers may request the detailed illustration from the authors.
9
It is known that the marginal data density results may be affected by the prior. Different priors were set to run for this test. The main competition is
between 2 and 4 regime models. Table 1 presents two sets of priors. Note that in the Bayesian model, we allow the means and variances to change as the regime
changes; we did not let the persistence vary with the regimes. We follow Albert and Chib (1993) and Kim and Nelson (1999a,b) by assuming Normal priors for
ai ; /; cj where i = 1, 2, . . ., 6 and j = 1, 2. For the variance parameters, r21 has an inverted gamma prior distributions and r2i r2i1 1 hi where i = 2, 3, . . . ,6,
hi > 0 and 1 + hi has an inverted gamma prior distribution. The evaluation of the posterior density evaluated at the posterior mean in Table 1 is calculated
according to Chib (1995). All inferences are based on 20,000 Gibbs simulations, after discarding the initial 4000 Gibbs simulations, (these are discarded to
reduce the impact of initial values. Prior 1: we specify for 6 regimes here and the lower regimes are specied similarly.
(a1, a2, . . ., a6)  N((0, 0.1, 0.2, 0.1, 0.2, .2), 5 I6), /  N((0.5, 1/25), (c1, c2)  N((0, 0), 5I2), r2i  inverted Gamma(1, 1), pii  Beta(0, 1, 0.1) for all i and the
adjusted pij  Beta(0.1, 0.1) for all the relevant i and j. Prior 2: we specify for 6 regimes here and the lower regimes are specied similarly.
(a1, a2, . . ., a6)  N((0, 0.1, 0.2, 0.1, 0.2, .2), 2I6), U  N((0.5, 1/25), (c1, c2) N((0, 0), 2I2), r2i  inverted Gamma(1, 10), pii  Beta(1, 10) for all i and the adjusted
pij  Beta(1, 10) for all the relevant i and j (see Kim and Nelson 1999b)
5

648

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

Table 1
Marginal data density test results.
Number of
regimes

Likelihood (i) prior I/

prior II

Prior (ii) prior I/

prior II

Posterior (iii) prior I/

prior II

Total marginal data density: (i) + (ii)  (iii) prior I/

prior II

S=1
S=2
S=3
S=4
S=5
S=6

683.15/683.44
614.17/614.71
619.28/616.70
627.19/628.66
687.45/659.52
696.39/681.27

2.40/14.00
13.80/55.72
23.03/56.86
21.14/61.50
49.16/140.21
47.21/57.36

13.36/13.33
22.52/19.87
10.40/22.67
6.30/18.19
57.44/35.60
30.94/28.90

698.91/710.77
650.49/690.30
652.71/696.23
642.03/708.35
679.17/835.33
712.66/767.53

Table 2
Parameter estimates of Markov Switching models.a
Parameter estimates

With restricted /0 and h1

Without restrictions

With restricted h1

h0
h1
~
p
~
q

0.088 (0.011)
0.001 (0.990)
0.019 (0.017)
0.005 (0.696)
0.230 (0.000)
0.155 (0.011)
0.346 (0.000)
0.582 (0.000)
0.980 (0.114)
0.988 (0.069)

0.467 (0.000)
0.168 (0.069)
0.019
0.075 (0.000)
0.121 (0.001)
0.107 (0.004)
0.565 (0.000)
0
0.976 (0.021)
0.928 (0.005)

0.014 (0.559)
0.089 (0.351)
0.0001 (0.988)
0.028 (0.058)
0.184 (0.000)
0.139 (0.000)
0.295 (0.000)
0.384 (0.000)
0.975 (0.098)
0.992 (0.057)

0.046 (0.047)
1.930 (0.000)
0.017 (0.001)
0.054 (0.327)
0.237 (0.000)
0.115 (0.000)
0.380 (0.000)
0
0.160 (0.000)
0.992 (0.028)

Nobc
Log-like

719
546.61

719
655.87

719
293.47

719
355.76

/0
/1

c1
c2

US 1948:12008:5
b

Without restrictions

a0
a1

UK 1948:12008:5

Numbers in the parentheses are the p-values.

A simple restriction of h1 = 0 could not yield a valid hessian matrix. Thus, an additional restriction is imposed on /0 .
Nob: number of observations.

means and variances in the original ination model. Furthermore, an extension of the model from two to four regimes introduces 14 additional parameters.10 Based on these results, and in the light of the model parsimony principle, a two-regime model
would not be an inappropriate choice. Finally, the best marginal data density value11 does not clearly select between a two-regime model and a four-regime model while the best likelihood value in Table 1 sticks with the two-regime model consistently
even when the prior changes. Thus, after this examination of the appropriate number of regimes, we will focus on a two-regime
MLE model.
Columns 2 and 3 of Table 2 provide the numerical estimates of the Markov-Switching model of UK ination. We will rst
focus on the model that imposes no restrictions on the parameter values. Based on the results shown on Column 2 of Table 2,
~) while the expected duration of the high
the expected duration of the low uncertainty regime 0 is 83.3 months (i.e., 1=1  q
~). The estimate of a0 implies that the expected average ination rate in reuncertainty regime 1 is 50 months (i.e., 1=1  p
^ 0 ). The estimate of a1 signals a shift in the mean ination as St varies from 0
^ 0 =1  q
^ where q
^ 1 /
gime 0 is 4.631% (=a
^0 /
^ 1 ). Though both a
^0 a
^ 1 =1  q
^1
^ where q
^ 1 /
to 1. The expected average ination rate in regime 1 is 6.357% (=a
^ 1 are not statistically signicant, regime 1 shows a higher average ination rate. Since it is important to know whether
and /
there is indeed a signicant difference in the average ination rate, a Wald test is implemented.12 The observed Wald test
statistic is 0.1113 and the p-value for the null hypothesis of no difference in the average ination rates between two regimes
is 0.739. Thus, the average ination rates are basically the same in both regimes. As for ination shock persistence,
q^ 1 /^ 0 /^ 1 St . For the UK, the estimate of /0 is 0.019, indicating the value of q^ being 0.981, which is a highly persistent
number for regime 0. Since the U.K. estimate of /1 is 0.005, the shock persistence measure in regime 1 is 0.986, which implies
that shock persistence is slightly higher in regime 1. However, the p-value of the estimate of /1 is 0.696. Thus, the persistence of
shocks is approximately the same in both regimes. Finally, for ination uncertainty, the estimate of h1 indicates the change in
standard deviation (or uncertainty) when the regime changes. The UK estimate of h1 is 0.582 with a p-value less than 0.001.
^2 when St 0), the measure of ination uncertainty
Comparing with regime 0 where the variance is 0.120 (i.e., Varut;St h
0
2
^
^
in regime 1 is 0.861 (i.e., Varut;St h0 h1 when St 1). The increase in ination uncertainty from regime 0 to regime 1
is signicant.

10

The transition matrix goes from two by two to four by four. Furthermore, there are other additional parameters.
The marginal data density value is calculated according to Chib (1995) and Kim and Nelson (1999a).
12
a1
^ c
^. Using the delta method
The null hypothesis is H0 : /a0 /a0
. Rearranging the terms, we obtain H0 : a1 /0  a0 /1 0. Let gb a1 /0  a0 /1 and gb
0
0 /1
^2 =Varc
^  v2 1.
(Greene, 2008: pp. 10551056), we are able to get the Wald test statistic W c
11

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

649

The MS two-regime model thus shows that, compared to regime 0, regime 1 has similar mean ination and shock persistence, and much higher ination uncertainty. Note that St = 1 means the economy is in regime 1. To have a better grasp
of the MS results, Fig. 1a shows the actual UK ination rate (the solid line) and the probability that St = 1 (the dashed line) for
the period of 1948:62008:5.13 The left vertical axis denotes the ination rate while the right vertical axis indicates the probability. The probability of the economy in St = 1 is considered as high if the value is greater than 0.75. For the UK, the periods of
high probability of being in St = 1 occur in most of the 1950s, the early 1960s, 1974:11978:6, 1979:71981:5, and 1990:4
1991:11.
It is important to note, and bears repeating that the ndings of statistically insignicant a1 and /1 do not imply that there
has been no secular, structural change in the mean or persistence of UK ination over the post-war years. It simply indicates
that the high uncertainty state is not signicantly associated with a high mean level of ination. There is strong evidence that
there have been signicant breaks in mean UK ination over the last several decades. Indeed, a number of papers have carefully documented secular structural breaks in both the mean and persistence of ination for the UK (see Benati, 2008a,
2008b; Cogley et al., 2003; Groen and Mumtaz, 2008; Rapach and Wohar, 2005).14 Again, our nding does not contradict these
results in any way; instead, our MS model captures cyclical regimes changes, and what we have found is that the high uncertainty regime is not robustly associated with a high level of mean ination.
There are two points to note on our nding that high uncertainty is not signicantly correlated with mean ination. First,
this lack of association is both theoretically plausible and consistent with previous empirical ndings.15 Secondly, the attempt to capture three ination characteristics simultaneously in two different regimes may lead to competition among these
three elements, i.e., a dominant element may drive the regime change and diminish the regime differentials in the other two
elements. Thus, to examine whether volatility dominates the other two in the regime switch, we impose the restriction of h1
being zero in the model, i.e., assuming the same volatility in both regimes. Due to the difculty16 of getting a valid t-value
for the estimate of /0 , we also impose the restriction of /0 0:019 in this case. Our results, shown in the third column of
Table 2, indicate that the estimate of /1 is signicantly different from zero and the p-value of the estimate of a1 is drastically
reduced. In this case, the expected duration of regime 0 is 13.9 months (i.e., 1/(1  0.928)) while the expected duration of
regime 1 is 41.7 months (i.e., 1/(1  0.976)). The expected average ination rate in regime 0 is 24.578% while the expected

13

We lose 12 data points because of converting the price index into ination.
Recently some researchers used BVAR or BSVAR to examine the structural break and/or persistence in ination. Even though the structural break issue is
not the focus of this paper, we did have the following observations. First, as mentioned in Table 1, the model with four regimes is optimal under prior I, even
though it is not optimal under prior II. A model with one state variable (i.e., St) with four regimes is similar to, but more general than, a model with two state
variables (i.e., St and Dt) with each state variable having two regimes. For example, compare the model of S = 4, where St = 1, 2, 3, 4, and the model of S = 2 and
D = 2 where the four regimes are S1D1, S1D2, S2D1 and S2D2A. A standard way to deal with structural breaks is to treat Dt as one that either stays at the current
regime or jumps to the next regime (Chib, 1998; Kim and Nelson, 1999a; Barnett et al., 2010). Thus, the transition matrix of Dt is restricted if one intends to use
Dt to indicate structural breaks. A model with one state variable (i.e., St) with four regimes is an un-restricted case because the model does not impose
restrictions on its transition matrix. We did the case for St and Dt with each state variable having two regimes and the transition matrix of Dt being


q11 1  q11
. When we incorporated Dt, we only controlled for the change in the means for Dt, i.e., if Dt moves to a higher regime, both means in either St = 1
0
1
or St = 2 will switch to a different level. Under prior I in Table 1, the marginal data density value of the S = 2 and D = 2 model is 646.53, which is lower than the
value of 642.03 (the case of S = 4 and D = 1) Thus, the optimal choice in Table 1 under prior I remains valid, i.e., the model of S = 4 is better than the model of
S = 2 and D = 2. Second, even though S = 2 and D = 2 is not the optimal model, we would look at its results to nd the implications for the structural break and
persistence. As we examined the results of the S = 2 and D = 2 model, the only signicant ai (i.e., the ratio of the posterior mean/standard deviation is
substantially greater than 2) is the mean at D = 1 and S = 1; all the other ais are not signicant. The implication is that we could not identify a signicant change
in mean ination when D = 1 is moved to D = 2. Thus, we cannot nd sufcient support of a structural break in ination. Again, the culprit is probably in the
volatility, which takes away the ability to detect the change in mean ination rates. This may also due to the fact that we are dealing with a single equation
model and other researchers use BVAR models, which have multiple equations. Third, the posterior mean of the persistence parameter U is .0077 in S2D2
model while it is 0.0083 in S4D1 model. Thus, even when we control the change in the mean for possible structural breaks, the persistence of the ination
remains high.
15
Friedman (1977) and Ball (1992) posit a positive and causal relationship from the level of ination to uncertainty, as higher ination should theoretically
lead to higher uncertainty. However, Ungar and Zilberfarb (1993) note that the effect of ination on uncertainty is ambiguous; if the cost of forecasting ination
is less than the cost of uncertainty, an increase in ination will lead to better forecasts, and actually reduce uncertainty. Moreover, the causality in the other
direction-from higher uncertainty to ination, is also theoretically ambiguous. Cukierman (1992) presents a model in which higher uncertainty raises the
optimal rate of ination, thus predicting a positive relationship. However, Holland (1995) points out that, if a central bank is aware of the negative effects of
uncertainty, it will seek to counter any increase in uncertainty by lowering the rate of ination. This would suggest a negative causal relationship. Empirically,
when ARCH and GARCH modeling began, Engle (1983) and Bollerslev (1986) employed ARCH and GARCH variances as proxies for uncertainty, and found that,
for the US, there was no robust relationship between the mean of ination and uncertainty (see Grier and Perry, 1998, p. 684 for a discussion). Grier and Perry,
employing GARCH models and testing formally for the impact of ination on uncertainty and vice-versa, nd that ination does increase uncertainty in the UK.
However, the authors nd uncertainty in the UK has a negative effect on ination. Thus it is unsurprising that when employing the MS technique we nd that
the relationship between ination and uncertainty is not signicant. These ndings are even less surprising, to anticipate further results, given that the entry
into the ERM raised uncertainty. This joining of the precursor to the Euro occurred during a time of low ination, but the unpredictable nature of the policy (in
the end, of course, Britain dropped out) raised ination uncertainty.
16
Due to the difculty in deriving the analytical gradient and Hessian of the likelihood function, two optimization routines are used in the likelihood
maximization procedure concurrently: nlminb in S+ and nlm in R. In S+, nlminb allows parameter restrictions but it does not provide a numerical gradient and
Hessian. In R, nlm provides numerical gradient and Hessian but it does not allow parameter restrictions automatically, i.e., we have to program the restrictions.
In the case of setting h1 = 0, the optimal estimate of /0 is near zero in S+. As we program the restriction of /0 being between 0 and 1 by using the logistic
function (ex/(1 + ex)) in R, the Hessian matrix fails to be positive denite (in the minimization of ln L) at the point of supposed convergence, probably due to
overow and/or underow problems and the ensuing numerical inaccuracies. Since our goal here is to check if /1 and a1 become signicant estimates as
volatility is restricted, we also x the number of /0 at 0.019, which is the optimal estimate in the unrestricted case.
14

650

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

25

1.0

20

0.8

15

0.6

0.2

10

0.4

0.0

09/01/1948

09/01/1959

09/01/1970

09/01/1981

09/01/1992

09/01/2003

Solid line: inflation rate (12 month ago); Dashed line: Pr(St=1)
Fig. 1a. UK ination and the probability of being in St = 1

average ination rate in regime 1 is 3.180%. The measures of ination shock persistence in regime 0 and regime 1 are 0.981 and
0.906 respectively. Thus, when the volatility is restricted to be the same in both regimes, regime 0, comparing with regime 1,
has a shorter expected duration with a much higher expected average ination rate and slightly stronger shock persistence.
Though the ordering of the regime is different, the signicant switch in shock persistence and a sharp change in average ination rates between regime 0 and regime 1 are results drastically different from the case where no restrictions are imposed.
Fig. 1b shows the restricted case of the probability of the UK economy in regime 0, which is the regime with a higher average
ination rate, comparing to regime 1. It is noticeable that, as we hold the variance constant in both regimes, the high average
ination periods in the restricted case are very similar to the high ination volatility periods in the un-restricted case (i.e.,
Fig. 1a). This shows that, as we model mean, persistence and volatility together in the un-restricted case, it is volatility that
drives the regime change and diminishes the differentials in mean and persistence in the regime switch.
What then is the choice between the un-restricted and the restricted cases? We implement the likelihood ratio test between these two cases and nd that the p-value of the observed test statistic is 0.000. Thus, the null hypothesis of h1 0 and
/0 0:019 is rejected. We will thus from now on focus on the un-restricted case.
As mentioned, the intention of this paper is to evaluate the impact of IT and other policies on UK ination dynamics. A
special characteristic of the MS model is that we are able to calculate the probability of the economy in St = 1 for each time
period. Based on these probability values, we will assess the impacts of different polices on ination uncertainty.
In order to assess the effect of changes in monetary policy, we must rst identify those major actions by the Bank of England that were most likely to have an impact on the level, persistence, and uncertainty of ination. Like other industrialized
countries, Britains ination rate appeared to be high in the 1970s. While informal (and, starting in 1977, formal) money supply targets were implemented in an attempt to lower ination, the targets were missed. Thus prices unsurprisingly continued to rise.
Margaret Thatchers government came to power in May 1979, with a perceived commitment to lower ination. However,
for nearly 2 years, money supply targets continued to be overshot. These overshoots were allowed because of recessionary
concerns at the time (Cobham, 2002, p. 17). Then, in March 1981, the Thatcher administration passed an annual budget that
adjusted the targets to take account of previous overshoots. This was described as a turning point in UK monetary policy.
While ination proceeded to fall through the mid-1980s, money supply targets were still overshot. Thus in March 1987
the Bank of England, under Chancellor Lawson, abandoned money supply targets and began shadowing the Deutschmark.
Informally adopting an exchange-rate target was thought to give a better framework for monetary policy than the discredited money supply targets.
The third major policy action is the entry into the ERM in October 1990. This now gave the Bank of England a formal exchange rate target. Although England would famously leave the ERM during the crisis of 1992, the ERM was clearly a currency peg and led, for most of its member countries, to the adoption of the Euro. The fourth major change in policy was
the adoption of the ination target in October 1992. As noted, the impact of ination targeting on monetary outcomes
has been the source of much inconclusive research. The last policy change we will investigate was in May of 1997, in which
a new Monetary Policy Committee of the Bank of England was given sole authority to conduct monetary policy and was no
longer beholden to the Exchequer. This amounts to central bank independence, which has been found, in cross-country studies, to affect monetary outcomes (Cukierman, 1992; Alesina and Summers, 1993).

651

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

25

1.0

20

0.8

15

0.6

0.2

10

0.4

0.0

09/01/1948

09/01/1959

09/01/1970

09/01/1981

09/01/1992

09/01/2003

Solid line: inflation rate (12 month ago); Dashed line: Pr(St=0)
Fig. 1b. The restricted case: UK ination and the probability of being in St = 0

We thus implement the following Probit model:

F 1 PS1 s0 s1 D81 s2 D87 s3 D90 s4 D92 s5 D97 et

where F 1 is the inverse of the normal CDF, P S1 is the probability of being in the state of St 1, D81, D87, D90, D92, D97 are the
dummy variables for the 1981 budget, Deutschmark shadowing, ERM, IT and Bank of England independence. D81 equals 1 for
the periods of 1981:32008:5 and D81 is 0 for all the other time periods. Similarly, D87 (D90, D92, D97) equals 1 for the periods
of 1987:32008:5 (1990:102008:5; 1992:102008:5; 1997:52008:5) and D87 (D90, D92, D97) is 0 for all the other time periods. Using PS1 derived from the Markov-Switching model in 1948:92008:5, we obtained the following regression results:

F 1 PS1 0:434  1:949D81  0:210D87 1:734D90  2:183D92  0:088D97

p0:0000:0000:2400:0000:0000:016
The numbers in the parentheses are the p-values calculated from robust standard errors. All parameter estimates, except
that of D87, are signicantly different from zero at the 5% signicance level. Again note that the state of St 1 is the regime of
higher ination uncertainty. Thus, the 1981 budget, IT and Bank of England Independence did have signicantly negative
impacts to the probability of the economy being in the state of high ination uncertainty. Of these three polices that reduce
ination uncertainty in the economy, IT has the largest estimate (i.e., the estimate is 2.183). The estimate of the deutschmark shadowing policy is negative but insignicant. This policy thus has no impact on reducing ination uncertainty. Notably, as previously mentioned, the estimate of ERM membership is positive, implying that the ERM actually increased ination
uncertainty.
Another way to evaluate the impacts of the policies on the ination dynamics is to inspect the changes in probability directly. Table 3 shows the impacts of each policy on the probability of being in regime 1. By looking at the third column of
Table 3, it is obvious that 1981 budget has the greatest and the 1992 IT has the second greatest impact in reducing the probability of being in a high ination uncertainty regime. Bank independence actually does not have much impact and 1990
ERM actually increases the probability of being in a high ination uncertainty regime substantially.

5. Discussion
The nding that exchange rate rigidity has no (or, in the case of the ERM, actually a positive) effect on uncertainty must be
interpreted carefully. It bears repeating that the impact is on uncertainty, rather than mean ination. There is a large literature on the impact of xed exchange rates on the level of ination. Ghosh et al. (1997) for instance nd a palpable impact,
while Bleaney and Francisco (2005) found that, once the endogeneity of the exchange rate regime was controlled for, only
hard pegs (currency boards and currency unions) had a signicant, negative effect on price level changes. Examining four
industrialized nations, Mumtaz and Sunder-Plassman (2010) nd that real exchange rate shocks have less importance for
ination since the 1980s than previously. Again, while there have been papers which have shown a link between the mean

652

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

Table 3
Impacts of policies on the probability of being in regime 1.
z-Value

P(St = 1)

DP(St = 1)

0.434
1.515
1.725
0.009
2.192
2.280

0.668
0.065
0.042
0.504
0.014
0.011

0.603
0.023
+0.462
0.490
0.003

15

1948:61981:2
1981:31987:2
1987:31990:9
1990:101992:9
1992:101997:4
1997:52008:5

1.0

10

0.8

0.6

0.4

0.2

0.0

09/01/1948

09/01/1959

09/01/1970

09/01/1981

09/01/1992

09/01/2003

Solid line: inflation rate (12 month ago); Dashed line: Pr(St=1)
Fig. 2a. US ination and the probability of being in St = 1

of ination and uncertainty, our results only suggest that exchange rate rigidity has been ineffective in lowering uncertainty,
and does not have such implications for average ination.
In contrast to the ineffectiveness of exchange rate rigidity, independence for the Bank of England did lower uncertainty.
This is in keeping with the ndings of Alesina and Summers (1993). These authors found, in a cross-country study, that central bank independence lowered the variability of ination. While this latter metric is a far cruder measure of uncertainty
than that employed in this paper, it does make sense that once monetary policy is insulated from political pressure, price
stability will be given greater weight, lessening the uncertainty the public has about ination.
At the time of independence, the great ination had been long over, so the impact, while negative and signicant, was
fairly small in magnitude. The 1981 budget, in contrast, came on the heels of previous failed attempts to hit money supply
targets and lower ination. Thus it is not surprising that this policy change had an effect that was quite large in magnitude.
The adoption of IT had a strong negative impact on ination uncertainty. Fatas et al. (2007), while not examining uncertainty, found in a cross-sectional study that IT was more effective in lowering mean ination than exchange rate targets. As
for uncertainty, our results suggest that the transparency and communication with the public that IT entails clear up much
uncertainty about the path of ination.
While it is important to keep in mind that the UKs experience with IT does not necessarily mean that other countries will
have similar results, it is still interesting to contrast the UKs performance in terms of ination uncertainty with that of the
US, the largest economy without a formal target. IT has been a controversial issue for the US Federal Reserve chair Ben Bernanke is a well-known advocate of IT adoption for the US. Other members of the Federal Reserve Board, such as Donald Kohn,
are on record opposed to IT adoption. The nancial turmoil that began in July of 2007 has pushed the issue of a formal target
temporarily out of public view, as some fear that IT could lead to excessive tightening and hamper the central banks efforts
to deal with the fallout of the subprime mess. On the other hand, the Fed announced in February 2009 that it would publish
long-run ination and economic forecasts, which may be close to a de facto IT. Moreover, some economists, such as Frederic
Mishkin, remain forceful advocates for de jure IT adoption in the US. The issue will thus be revisited in the future, so we will
contrast uncertainty in the US and UK since the UK adopted IT.

653

15

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

1.0

10

0.8

0.6

0.4

0.2

0.0

09/01/1948

09/01/1959

09/01/1970

09/01/1981

09/01/1992

09/01/2003

Solid line: inflation rate (12 month ago); Dashed line: Pr(St=1)
Fig. 2b. The restricted case: US ination and the probability of being in St = 1

Fig. A1. Classical decomposition of monthly UK ination rates

The US data set during the same time period (i.e., 1947:62008:5) is the monthly consumer price index for all urban consumers in the US The data are obtained from the FREDS website of the St. Louis Federal Reserve Bank. We run the same MS

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

+
++
+
++
+
+
+

15
10
0

inflation rate

20

25

654

++++ : xm12 inflation data

solid line : decomposed xm1 inflation data

+
++
+
+
+
+
++
+
+
+
+ +
+
++ +
+ +
+
+
++ +
+
+ ++ + +
+
++
+ ++
+
+
+
++
+
+ + + ++
+
++
+
+++
+
+++
+
+
+ +
+
+++
++
+
+
+
+
+
+
+
+
+ +
+
++ +++
+++
+ ++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
++ +
+
+
+
+
+ +++
++++
+++
+++ +
+
++
+
+
+
+
+
+
++
+
+
+ +
+ ++
+ ++
+
++ +
+
+
+ ++
++ ++ ++
+
+
+
+
+
++ +
+ +
+
++
+++ + +
+
+++
+
++++++
+++++++
+++ +
+ + +++
+
++ ++
++++ ++
+
+
++
+
+
++++
+
+
+
+
+
+
+
+
+
+
++
+ ++ ++++
++++
+
+++ ++
+ + +++
+
++ ++ +
+++
+
+++++
+
+++++ ++
+
+ ++++++
+
+
+
+++++ ++++
+
+
+
+
+
+
+
+
++ +++
+
+
+
+++
+
+
+
+
+
+++ +
+
++
+
++++ + +++ + ++
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+++++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+++ ++
+ ++++++ +
++
+ +++
+
+ + ++++ +++
+
+ ++++
+ + ++
+
+
+
+
++
+
+
+
+
++
+
+
++
+
+
+ +
++
+++ ++
++
+++
++
+++
+
++
+
++
++++
+ + ++++ + ++ +
+++
+
+++ ++
+
++++
+

06/01/1948

06/01/1959

06/01/1970

06/01/1981

06/01/1992

06/01/2003

time
Fig. A2. Comparison of decomposed trend data and (X(t)  X(t  12))/X(t  12) ination data

model with changing mean, persistence and variance. The US estimates are presented in Table 1. By comparing the estimates
of a1 , /1 , and h1 , we cannot see much differentiation between these two countries. However, as we calculate the probability
of the economy being in St = 1, i.e., the high variance regime, there is a sharp contrast between two countries. As mentioned
earlier, for the UK, the periods of high probability of being in St = 1 occur in most of the 1950s, the early 1960s, 1974:1
1978:6, 1979:71981:5, and 1990:41991:11. It is notable that there has not been a period of high variability in the UK since
the adoption of IT. In contrast, for US, shown in Fig. 2a (the restricted case, shown in Fig. 2b),17 the periods of high probability
of being in St = 1 occur in most of the early 1950s, 1973:81975:12, 1981:71983:7, and 2005:92008:1. Thus the US has suffered from a period of high uncertainty in the wake of recent shocks, while the UK has managed to avoid such uncertainty.
While these results are not a formal test and are only suggestive, they are evidence that ought to be weighed as the US contemplates adopting an ination target.

6. Conclusion
To repeat, it is of course the case that results for the UK may not have direct policy implications for other countries. Still,
the nding that intermediate steps toward currency union were ineffective, if not actually harmful in countering ination
uncertainty is something that eastern European and other euro-zone aspirants should keep in mind as they embark on target
zones and other soft pegs as rst steps on the road to euro adoption.
The strong effect of IT on uncertainty does bolster the case that IT provides a clear, observable target and, in the case of the
UK, the Bank of England has been credible in its commitment to IT. The rst 16 years of IT are of course no guarantee of future success. This may especially be the case since the Bank of England has missed the formal target and had to send explanatory letters to the Chancellor of the Exchequer starting in April of 2007. At the same time, our analysis indicates that, more
than a year after Governor Mervyn King rst admitted missing the target of 2%, uncertainty did not yet signicantly rise in
the UK, while the US with no target, has suffered from rising uncertainty over the path of future monetary policy and
ination.

Appendix A
The UK monthly ination rates calculated by (X(t)  X(t  1))/X(t  1) is shown in the rst panel of Fig. A1. Fig. A1 shows
that the actual monthly ination rates may be decomposed into trend, seasonal and random parts. The method used is classical seasonal decomposition by moving averages. Fig. A2 shows that the decomposed time trend data with a lag of 5 (shown
with solid line) match very closely with the (X(t)  X(t  12))/X(t  12) ination data set (shown with +). However, with
17
Note that when we restrict the volatility to be the same in both regimes for the US (i.e. the model only allows the level and persistence of ination to vary
between regimes) we obtain the results shown in the last column of Table 1. Regime 1 has the higher average ination rate. Fig. 2b shows the probability of US
ination being in regime 1, where the period of 20052008 is obviously in a low average ination regime. The LR test results indicate, however, that the
unrestricted case is the better model.

W. Miles, C.-P. Vijverberg / Journal of Macroeconomics 33 (2011) 644655

655

the decomposed time trend data, we lose the rst 6 and the last 6 observations of the original data set while with the
(X(t)  X(t  12))/X(t  12) calculation, we lose the rst 12 observation of the original data set.
References
Albert, James, Chib, Siddhartha, 1993. Bayes inference via Gibbs sampling of autoregressive time series subject to Markov mean and variance shifts. Journal
of Business and Economic Statistics 11, 115.
Alesina, A., Summers, L., 1993. Central bank independence and macroeconomic performance: some comparative evidence. Journal of Money Credit and
Banking 25, 151162.
Ball, L., 1992. Why does high ination raise ination uncertainty? Journal of Monetary Economics 83, 185191.
Ball, L., Sheridan, N., 2005. Does ination targeting matter? In: Bernanke, Ben, Woodford, Michael (Eds.), The Ination Targeting Debate. University of
Chicago Press, pp. 249276.
Barnett, Alina, Groen, Jan, Mumtaz Haroon, 2010. Time-Varying Ination Expectations and Economic Fluctuations in the United Kingdom: A Structural VAR
Analysis. Working Paper 392, Bank of England.
Benati, L., 2008a. The great moderation in the United Kingdom. Journal of Money, Credit and Banking 39, 121147.
Benati, L., 2008b. Investigating ination persistence across monetary regimes. Quarterly Journal of Economics 123, 10051060.
Bleaney, M., Francisco, M., 2005. Exchange rate regimes and ination: only hard pegs make a difference. Canadian Journal of Economics 38, 14531471.
Bollerslev, T., 1986. Generalized autoregressive conditional heteroscedasticity. Journal of Econometrics 31, 307327.
Cecchetti, S., Debelle, G., 2006. Has the ination process changed? Economic Policy 21, 311352.
Chib, Siddhartha, 1995. Marginal likelihood from the Gibbs output. Journal of the American Statistical Association 90, 13131350.
Chib, Siddhartha, 1998. Estimation and comparison of multiple change-point models. Journal of Econometrics 86, 221241.
Cobham, David, 2002. The Making of Monetary Policy in the UK 19752000. John Wiley and Sons, Sussex, England.
Cogley, T., Morozov, S., Sargent, T., 2003. Bayesian fan charts for U.K. ination: forecasting sources of uncertainty in an evolving monetary system. Journal of
Economic Dynamics and Control 29, 18931925.
Cukierman, A., 1992. Central Bank Strategy: Credibility and Independence. MIT Press, Cambridge.
Cukierman, A., Kalaitzidakis, P., Summers, L., Webb, S., 1993. Central bank independence, growth, investment and real rates. Carnegie-Rochester Conference
Series on Public Policy 39, 95140.
Engle, R., 1982. Autoregressive conditional heteroscedasticity with estimates of the variance of united kingdom ination. Econometrica 50, 9871007.
Fatas, A., Mihov, I., Rose, A., 2007. Quantitative goals for monetary policy. Journal of Money, Credit and Banking 39, 11631176.
Fischer, S., 1981. Relative shocks, relative price variability and ination. Brookings Papers on Economic Activity 2, 381431.
Friedman, M., 1977. Nobel lecture: ination and unemployment. Journal of Political Economy 85, 451472.
Ghosh, A., Gulde, A., Wolf, H., 1997. Exchange Rate Regimes: Choices and Consequences. MIT Press, Cambridge, Massachusetts, USA.
Greene, W., 2008. Econometric Analysis, Sixth ed. Pearson Prentice Hall.
Grier, Robin, Grier, Kevin, 2006. On the real effects of ination and ination uncertainty in Mexico. Journal of Development Economics 80, 478500.
Grier, Kevin., Perry, Mark., 1998. On ination and ination uncertainty in the G-7 countries. Journal of International Money and Finance 17, 671689.
Grier, Kevin., Perry, Mark., 2000. The effects of real and nominal uncertainty on ination and output growth in the USA. Journal of Applied Econometrics 1,
4558.
Grier, Kevin, Henry, Olan, Olekalns, Nilss, Shields, Kalvinder, 2004. The asymmetric effects of uncertainty on ination and output growth. Journal of Applied
Econometrics 19, 551565.
Groen, J., Mumtaz, H., 2008. Investigating the Structural Stability of the Phillips Curve Relationship. Bank of England Working Paper No. 350.
Hamilton, J., 1989. A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica 57, 357384.
Kim, Chang-Jin, 1993. Sources of monetary growth uncertainty and economic activity: the time-varying parameter model with heteroscedastic
disturbances. The Review of Economics and Statistics 75, 483492.
Kim, Chang-Jin, Nelson, Charles, 1999a. Has the US economy become more stable? A Bayesian approach based on a Markov-switching model of the business
cycle. The Review of Economics and Statistics 81, 608616.
Kim, Chang-Jin, Charles, Nelson, 1999b. State Space Models with Regime Switching. MIT Press, Cambridge, USA.
Mumtaz, H., Sunder-Plassmann, L., 2010. Time-Varying Dynamics of the Real Exchange Rate: A Structural VAR Analysis. Bank of England Working Paper 382.
Rapach, D., Wohar, M., 2005. Regime changes in international interest rates: are they a monetary phenomenon? Journal of Money, Credit and Banking 37,
887906.
Rodrik, D., 2005. Why We Learn Nothing from Regressing Economic Growth on Policies. Working Paper, Harvard University.
Ungar, M., Zilberfarb, B., 1993. Ination and its unpredictability-theory and empirical evidence. Journal of Money, Credit and Banking 25, 709720.