Working Paper Series: Towards A Robust Monetary Policy Rule For The Euro Area

Worki ng PaPer S eri eS
no 1210 / J une 2010

ToWardS a
robuST moneTary
Policy rule for
The euro area
by Tobias S. Blattner
and Emil Margaritov
WORKI NG PAPER S ERI ES
NO 1210 / J UNE 2010
In 2010 all ECB
publications
feature a motif
taken from the
500 banknote.
TOWARDS A ROBUST MONETARY
POLICY RULE FOR THE
EURO AREA
1
by Tobias S. Blattner
2
and Emil Margaritov
3
1 The views presented in this paper are those of the authors and do not necessarily reflect those of the European Central Bank.
We would like to thank B. Hofmann and an anonymous referee for their comments and suggestions.
2 Corresponding author: European Central Bank, Kaiserstr.29, 60311 Frankfurt am Main, Germany, phone: +49 69 13448351,
fax: +49 69 13447604, e-mail: [email protected]
3 Goethe University, Department of Money and Macroeconomics, House of Finance,
Grueneburgplatz 1, 60323 Frankfurt am Main, Germany, phone: +49 69 79833819,
e mail: [email protected] frankfurt.de.
This paper can be downloaded without charge from http://www.ecb.europa.eu or from the Social Science
Research Network electronic library at http://ssrn.com/abstract_id=1617252.
NOTE: This Working Paper should not be reported as representing
the views of the European Central Bank (ECB).
The views expressed are those of the authors
and do not necessarily reflect those of the ECB.
- -
European Central Bank, 2010
Address
Kaiserstrasse 29
60311 Frankfurt am Main, Germany
Postal address
Postfach 16 03 19
60066 Frankfurt am Main, Germany
Telephone
+49 69 1344 0
Internet
http://www.ecb.europa.eu
Fax
+49 69 1344 6000
All rights reserved.
Any reproduction, publication and
reprint in the form of a different
publication, whether printed or produced
electronically, in whole or in part, is
permitted only with the explicit written
authorisation of the ECB or the authors.
Information on all of the papers published
in the ECB Working Paper Series can be
found on the ECBs website, http://www.
ecb.europa.eu/pub/scientific/wps/date/
html/index.en.html
ISSN 1725-2806 (online)
3
ECB
Working Paper Series No 1210
June 2010
Abstract 4
1 Introduction 5
2 A new real-time database for the euro area 10
3 The thick-modelling exercise 14
3.1 Selection of rules 17
3.2 Estimation results 20
3.3 Condence intervals for thick modelling 26
4 Principal component analysis 27
4.1 Estimation results 30
5 Conclusion 33
References 35
Annexes 38
CONTENTS
4
ECB
June 2010
Abstract
Estimations of simple monetary policy rules are often very rigid. Standard practice
requires that a decision is made as to which indicators the central bank is assumed to
respond to, ignoring the data-rich environment in which policy-makers typically form
their decisions. However, the choice of the feedback variables in the estimations of
simple rules bears non-trivial implications for the prescriptions borne from these
rules. This paper addresses this issue for the euro area using a new comprehensive
real-time database for the euro area and examines the ECBs past interest-rate setting
behaviour in two complementary ways that are designed to deal with both model and
data uncertainty. In a first step we follow the thick-modelling approach suggested
by Granger and Jeon (2004) and estimate a series of 3,330 policy rules. In a second
step we employ a factor-model approach similar to Bernanke and Boivin (2003) for
the US Fed, but with structurally interpretable factors la Belviso and Milani (2006).
Taken together, we find a strong justification for the need of adopting robust
approaches to describe the historical evolution of euro area monetary policy. We also
find that the ECB is neither purely backward nor forward-looking, but reacts to a
synthesis of the available information on the current and future state of the economy.
JEL classification: C50, E52, E58,
Keywords: Taylor rules; Monetary policy; Real-time data
5
ECB
June 2010
I. Introduction
Since the seminal contribution by Taylor (1993), monetary policy rules have become
a prominent feature in the academic literature. Taylor proposed a rule that depicts by
how much a central bank should change the short-term interest rate to deviations of
inflation and output from their target or potential levels. More generally, policy rules
are understood as a positive or normative description of how a policy instrument, e.g.
the short-term interest rate, the monetary base or the exchange rate, responds to
changes in the macroeconomic environment. Such rules are often embedded as a
closing condition in structural macroeconomic models resulting from the first order
condition that solves the optimization problem relevant to the central bank (see, e.g.
Clarida, Gali and Gertler 1999 for interest rate rules or Chowdhury and Schabert 2008
for monetary base rules) or are presented as single-equation reduced-form estimates in
more empirically orientated work (see e.g. Clarida and Gertler 1996 or Orphanides
2001).
In practice, estimations of such rules are often plagued by a number of caveats and
strong underlying assumptions. To begin with, researchers face the issue of data
uncertainty. It is not at all clear whether the estimation of monetary policy rules using
data that could not possibly be known to policy-makers at the time of the decision are
informative for describing actual interest-rate setting. Data uncertainty can take two
dimensions. The first relates to the accuracy of the data. Data are often preliminary
and subject to sizeable and repetitive revisions over the course of time. In fact, it may
take many years before data can be considered final. Past research suggests that the
presence of noise in real-time estimates of inflation and the output gap must be
accounted for in evaluating policy rules (see Orphanides 2001). Table 1 reveals the
large differences that emerge from the use of real time or ex post data in past
estimations of the ECBs monetary policy rule. The second dimension concerns the
timeliness of data. This issue has largely been neglected in past research. Changes in
the policy instrument are often linked to data that was not yet available to policy-
makers at the time of the decision. For instance, interest-rate setting meetings in the
euro area are usually held in the first week of each month. However, in estimating the
ECBs reaction function it is common practice to link the interest-rate decision to data
released during the course of the full month. Even under the assumption that the rule
6
ECB
June 2010
is correctly specified, reliance on possibly unknown data can prove misleading in
identifying the historical pattern of monetary policy.
Table 1. Estimates of Taylor-type rules for the euro area
4
Study Data Type
Inflation
measure
Activity
measure
Period Inertia
Inflation
weight
Output
weight
Fourans and Vranceanu (2004) Ex post FW (+4) HICP IP
1)
01/99 - 10/03 0.84 2.80 0.19
Gerdesmeier and Roffia (2005) Ex post FW (+12) HICP ave
2)
01/99 - 06/03 0.81 0.64 1.44
Real time FW (+12) SPF ave
2)
01/99 - 06/03 0.98 2.13 1.63
Real time FW (+24) SPF ave
2)
01/99 - 06/03 0.95 1.87 1.70
Sauer and Sturm (2007) Ex post FW (+3) HICP IP (GR) 01/99 - 03/03 0.91 1.02 0.47
Ex post FW (+3) HICP IP (HP) 01/99 - 03/03 0.88 0.86 0.86
Real time FW (+3) HICP IP (GR) 01/99 - 03/03 0.85 0.72 0.28
Real time FW (+3) HICP IP (HP) 01/99 - 03/03 0.92 2.31 2.35
Survey FW (+3) ECON.
3)
ESIN 01/99 - 03/03 0.84 1.88 0.44
Hayo and Hofmann (2006) Ex post FW (+12) HICP IP (LT) 01/99 - 05/04 0.85 1.48 0.60
Fourans and Vranceanu (2007) Ex post FW (+12) HICP IP (HP) 01/99 - 03/06 0.96 4.25 1.28
Ex post FW (+12) HICP IP
1)
01/99 - 03/06 0.98 6.80 1.63
Gorter et al. (2007) Ex post FW (+12) HICP IP (HP) 01/97 - 12/06 0.96 0.04 0.86
Survey FW (+12) CEF
4)
CEF
4)
01/97 - 12/06 0.89 1.67 1.65
1) Deviation of IP growth from average growth rate over the sample
3) Inflation forecasts based on a poll of a group of forecasters conducted by the Economist every month.
4) Consensus Economics Forecast.
2) Average of a linear and quadratic trend measures, the one-sided Christiano-Fitzgerlad filter and the output gaps provided by the OECD
and the European Commission.
Researchers also face the issue of model uncertainty. For instance, the most
conventional policy rules are modelled in terms of inflation and the output gap.
Policy-makers, however, are unlikely to agree to the hypothesis that all relevant
information needed to conduct monetary policy is encapsulated in these two variables.
Because central banks face great uncertainty about the state of the economy, they
typically monitor a large array of economic and financial indicators to assess the risks
to price stability. Simple rules linking changes in the monetary policy instrument to
the evolution of a very narrow set of variables ignore the data-rich environment in
which policy makers typically form their decisions (see e.g. Bernanke and Boivin
2003). While it is the nature of models to simplify reality, and a policy rule is an
admittedly simplistic approach to mimic the behaviour of a central bank, the
4
Prior to the inception of the ECB and shortly after, some studies examined the Bundesbanks reaction
function as a benchmark for the ECBs policy rule (Clarida and Gertler 1996, Faust et al. 2001) or
estimated a hypothetical policy rule on the basis of consolidated data for the euro area countries for the
late 1980s and 90s (Gerlach and Schnabl 2000, Domnech et al. 2002, Gerlach-Kristen 2003 and
Gerdesmeier and Roffia 2004). While of interest itself, papers relying on synthetic euro area data
provide little evidence on the ECBs actual interest-rate setting behaviour.
7
ECB
June 2010
limitation to a confined set of indicators is likely to go hand-in-hand with model
misspecification and model uncertainty.

Further, even the most conventional policy rules, specified in terms of inflation and
the output gap, are not as straightforward to implement as is sometimes argued. Do
policy-makers react to past developments, forecasts or a combination thereof? Which
is the measure of economic slack that policy-makers regard as relevant in their
decisions? As regards the latter, there are various different concepts, such as potential,
trend or flexible-price output, in use by both researchers and econometricians. Figure
1 illustrates that, at times, different measures of the slack in real economic activity are
likely to entail a very different policy response depending on both the concept and
method used. In some circumstances this is not just a matter of magnitude. While one
measure suggests a positive output gap, others might suggest the contrary.
5
As a
consequence, the choice of the inflation or output gap indicator is not without impact
on the estimated coefficients of monetary policy rules. Table 1 illustrates that
estimation results generally vary to a significant extent depending on which indicator
for inflation and real economic activity is taken.
6

Figure 1. Real time measures of real economic activity
(as a percentage of potential activity)
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
output gap (linear trend)
output gap (HP filter)
industrial production gap (HP f ilter)

5
Another prominent example in the literature is the choice for the measure of inflation. Typically, in
empirical applications, central banks are assumed to respond to movements in headline inflation. In
contrast, structural models usually rely on the use of the GDP deflator to characterise the interest-rate
setting behaviour of the monetary authority (see e.g. Smets and Wouters 2003).
6
Table 1 only shows the forward-looking rules that have been estimated by the various authors. As we
will discuss below, forward-looking rules typically outperform backward-looking rules and satisfy a
unique equilibrium.
8
ECB
June 2010
Despite these concerns, standard practice in empirical macroeconomics still requires
that a decision is made as to which is the apparently best or dominant specification
amongst those considered, discarding any information in the alternative specifications.
In the current context, this means that researchers examining the ECBs past interest-
rate setting behaviour have largely focused on headline inflation and industrial
production as a measure of price and economic slack respectively (see Table 1),
ignoring any potentially useful information coming from alternative specifications.
7
There are good reasons to believe that this may not be the best approach. To begin
with, reliance on a single specification is clearly at odds with the ECBs two-pillar
monetary policy strategy, which emphasises the assessment of a large number of
indicators for the conduct of monetary policy (ECB 2004). Moreover, it seems unclear
whether a best indicator exists at all. Various authors have argued that HICP
headline inflation would be the correct indicator to be included in any ECB reaction
function because of its prominent appearance in the ECBs monetary policy strategy.
This assumption is at least doubtful. Would it make sense for the ECB to set policy
solely on the grounds of the inflation measure it is held accountable for? In fact, if the
ECB were indeed only to react to past inflation developments, it would face
difficulties in stabilizing the economy given the significant lags in the transmission of
monetary policy. Hence, as a minimum, policy rules should include forecasts of HICP
inflation for the relevant horizon. But forecasts, by definition, are surrounded by a
large degree of uncertainty. This is why policy-makers consider not only a single
best forecast, but look at a variety of projections from different institutions, both for
inflation and output growth.
8
However, the recommendations borne from these
forecasts for setting monetary policy can vary substantially and policy-makers in
practice have to pool the information to derive the (weighted) mean forecast that will
eventually guide their interest-rate setting behaviour.
9
7
An exception for the euro area is Gerdesmeier and Roffia (2005) who synthesize the information of
five different output gap measures and combined them to one that is used in their estimations. For the
US Fed Kozicki (1999) is a notable exception. He tested the robustness of his results employing a small
range of alternative measures of the output gap.
8
The non-existence of a single best forecast is the most evident in the case of the ECB. Staff
projections, which would apply as the most natural candidate for the best forecast, are not approved
by the ECBs Governing Council, but are only considered as one input among many in the decision-
making process.
9
A widely used reference in this regard is Consensus Economics Forecast who pool every month the
forecasts of the largest private and public organisations to form a mean estimate for the main
9
ECB
June 2010
Taken together, a realistic description of any central bank behaviour requires
recognition not only of the fact that policy-makers observe the economy in real time,
but also that they form their decision in a data-rich environment as stipulated by
Bernanke and Boivin (2003). This paper addresses both of these issues for the euro
area. In a first step we construct a new comprehensive real-time database for the euro
area based only on information available up to one day prior to each interest-rate
setting meeting of the ECBs Governing Council since 1999. The database is
described in detail in Section II. On the basis of this dataset we examine the ECBs
past interest-rate setting behaviour in two complementary ways, each time taking into
account the data rich environment in which it operates.

First, we follow the thick-modelling approach suggested by Granger and Jeon
(2004) who propose to estimate all plausible specifications and to pool the parameter
estimates according to some efficiency criteria. We estimate in total a series of 3,330
policy rules for the euro area, exploiting all plausible combinations of the 90 measures
of economic activity and the 37 indicators of inflation and inflation expectations. This
approach explicitly recognises the large information set at hand to policy makers in
practice and does not require taking a stand on the appropriate indicator for inflation
and output. Second, we employ a factor-model approach la Stock and Watson
(1999, 2002) and estimate a policy reaction function for the ECB under which the
short-term interest rate responds to the state of the economy, as measured by the
estimated factors. In so doing, we augment the approach by Bernanke and Boivin
(2003) with structurally interpretable factors la Belviso and Milani (2006) to
summarise the information on the basis of which the ECB conducts its monetary
policy.

The main findings of the paper can be summarised as follows. First, we show that the
prevailing dogma of real-time data, which has mainly been preoccupied with the
implications of ignoring data revisions, is not enough to account for the informational
rigidities that policy-makers face when forming their decisions. Estimations of policy
rules that rely on information that becomes available after the interest-rate setting
meeting necessarily overstretch the boundaries of the actual information set available

macroeconomic variables. Traditionally, the combination of forecasts often outperforms the
individually best forecast.
10
ECB
June 2010
to policy-makers in real time and consequently lead to biased estimates due to the
endogeneity of the explanatory variables. Second, we show that previous
specifications of the ECBs policy rule are likely to be ill-designed in their emphasis
on a single or a small set of economic indicators to which the central bank is supposed
to react to. In particular, we find that almost 300 estimated rules are able to closely
track the ECBs past interest-rate setting behaviour in a meaningful statistical and
economic way. This finding highlights that the quest for the real or unique policy
rule is motivated on false grounds, but that policy-makers receive a large array of
signals when deciding on the level of the short-term interest rate. The
recommendations borne from this battery of rules is likely to reflect the corridor in
which policy makers gear the course of monetary policy. As we will show, ECB
policy-makers are neither purely backward nor forward-looking, but react to a
synthesis of the available information on the current and future state of the economy.

The remainder of this paper is structured as follows. Section II describes the new real-
time database for the euro area. Section III explains the methodology and the results
of the thick-modelling exercise. In Section IV we apply the Bernanke-Boivin
methodology to conduct a positive analysis of the ECBs past interest-rate setting
behaviour. Section V concludes by reviewing the main results from the paper.

II. A new real-time database for the euro area

A striking feature of the past research being done in the area of simple policy rules
and the construction of real-time data sets for that purpose has been the preoccupation
of economists with the sole role of data revisions.
10
While providing a good overview
of the evolution of data over time, such data sets come nowhere close to answering
one of the most essential questions that needs to be addressed in any attempt to
provide a realistic rendition of the historical pattern of monetary policy: What
precisely do policy makers know at each time of decision making?

Traditional real-time data sets are based on standard data frequencies typically used in
the economic literature (e.g. monthly or quarterly). Such a procedure necessarily

10
The Real-Time Data Set for Macroeconomists of the Philadelphia Fed as documented by
Croushore and Stark (2001) can serve as an example.
11
ECB
June 2010
forces the economic analysis to link monetary policy decisions to data to which
policy-makers may actually have no knowledge at the time of making the decision. To
see this, take Orphanides (2001) as an example. Orphanides, like most other
contributors in the area of policy rules, estimates the Feds reaction function at a
quarterly frequency.
11
In particular, Orphanides regresses the quarterly average of the
federal funds rate on the Greenbook information corresponding to the middle month
of each quarter. That is, for the first meeting in each quarter Orphanides implicitly
assumes knowledge of the Greenbook in the following month, i.e. data that could not
possibly be known to policy-makers at that point in time. At the same time, he does
not account for the additional noise in the data resulting from the availability of a new
Greenbook prior to each meeting. An obvious implication of this practice is the issue
of endogeneity. Given that interest-rate decisions are factually linked to future
developments in inflation or economic activity, the resulting estimates under least
squares are biased.

A step into the direction of realism is to use monthly data instead. Researchers,
however, face a non-trivial trade-off when reverting to estimations based on monthly
data. Either they accept, as noted on an earlier note, that modelling then requires
taking a precise stand on indicators that are in fact available at a monthly frequency,
e.g. industrial production or the unemployment rate, or they reduce the frequency of
their quarterly dataset to monthly by means of standard interpolation techniques. The
latter approach, however, is flawed by introducing artificial dynamics in the resulting
dataset that is unlikely to have ever materialized in reality or been observed by policy-
makers. Moreover, even if some of the indicators are readily available at a monthly
frequency, they need not be available prior to the interest-rate setting meeting.

To avoid these misconceptions that have plagued much of the real-time monetary
policy literature we construct a new database for the euro area that takes fully into
account the true informational limitations faced by the ECBs Governing Council in
its decision-making process over the period from 1999 to 2007.

11
In fact, given the short sample of existence, policy rule estimates for the euro area are generally
obtained using monthly data.
12
ECB
June 2010
Table 2. Releases of euro area real GDP in the first half of 2009
Release Indicator name Vintage available for GovC meeting
8 January 2009 National accounts (GDP) Q3/2008 - 2nd release 15 January 2009
13 February 2009 Flash Estimate EU and euro area GDP Q4/2008 5 March 2009
5 March 2009 National accounts (GDP) Q4/2008 - 1st release 5 March 2009
7 April 2009 National accounts (GDP) Q4/2008 - 2nd release 7 May 2009
15 May 2009 Flash Estimate EU and euro area GDP Q1/2009 4 June 2009
3 June 2009 National accounts (GDP) Q1/2009 - 1st release 4 June 2009
8 July 2009 National accounts (GDP) Q1/2009 - 2nd release 6 August 2009

In doing so, we construct a database that is meant to describe the past, current and
future state of the economy as perceived by policy-makers on the day of the interest-
rate setting decision and without the need to assign a particular month or quarter to
any of the series. Contrary to other existing studies on policy rules, we do not impose
any information on the decision-making body that was not actually at hand at any
given point in time. Consider real GDP for the euro area as an example. Though
official national accounts data for the euro area are published with a quarterly
frequency by Eurostat, policy-makers actually receive new information about the state
of the economy every month (see Table 2). By comparing the exact release date with
the dates of the Governing Council meetings we are able to compile the information
set that was actually available to policy-makers at each meeting. From Table 2 it is
easy to see that in some cases a release in one month can only influence the decision
in the next month. Other releases may already be outdated at the next meeting. For
instance, the flash estimate of 13 February was neither of use for the February
meeting that took place on 5 February (too late) nor was it likely to largely influence
the March meeting because a more recent release became available to policy-makers
on the day of the March decision.
12

Figure 2 illustrates the differences that emerge when comparing our approach to the
current standard practice in the literature. The grey solid line shows the approach
chosen by Gerdesmeier and Roffia (2005) who considered a constant interpolation of
their real-time quarterly growth rates into monthly data. The black solid line shows a
constant interpolation of the first quarter of 2009 vintage of euro area real GDP.
13
As
becomes evident, estimations of policy rules that rely on ex post revised or real-time

12
Of course, it could still influence the March decision if the Council responds to the history of events.
We take care of this possibility in our estimations.
13
For a more meaningful and realistic comparison, we already adjusted both interpolated series to
reflect the lagged value at each point in time so as to account for the reporting lag of euro area national
accounts.
13
ECB
June 2010
interpolated data necessarily overstretch the boundaries of the actual information set
available to policy-makers in real time. Moreover, independent of whether the current
or lagged vintage is taken, interpolation of quarterly data does not take into account
that policy-makers usually receive signals about different vintages each quarter. For
example, in the first quarter of 2009 Eurostat issued information about both the third
and the fourth quarter of 2008 (see Table 2). Standard interpolation techniques will
always fail to reproduce this crucial observation.
Figure 2. Euro area real GDP
(quarter-on-quarter growth rate, seasonally adjusted)
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1999 2000 2001 2002 2003 2004 2005 2006 2007
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
real time
real time (interpolated)
ex post (interpolated)
With two meetings per month until the end of 2001 and one meeting per month
thereafter our real-time data set contains a total of 143 data observations for 127 series
that are included starting with the information set available to decision-makers on 21
January 1999 and including as a last observation the information available to the
Governing Council meeting on 6 December 2007. For the purpose of constructing the
real-time data set use was made of various sources, such as the ECBs Monthly
Bulletin, Eurostat, Consensus Economics Forecast, IMF, OECD, and others. The
choice of our variables was dictated by the ECBs recurrent reference to a particular
indicator either in the form of speeches or in official publications.
14
The list of
variables can be split into two groups of relevance for the later estimations: a price
group and a real economic activity group.
15
The price group contains 37 indicators
14
In this sense we follow the approach pursued by Gerlach (2007).
15
The reader is referred to Annex 1 for a complete list of all variables. Our database also includes 30
indicators related to developments in monetary aggregates and financial markets. However, as we are
interested in estimating the original Taylor rule that responds to inflation and output only, we did not
make use of these indicators. See Gerdesmeier and Roffia (2004) for an exposition of rules with
monetary indicators.
14
ECB
June 2010
and the output group includes 90 variables. Importantly, given the pivotal role played
by expectations in the conduct of a stabilizing monetary policy, all groups are
composed of a broad range of real-time macroeconomic projections from various
institutions for the likely future evolution of key economic indicators, such as real
GDP or consumer prices. Furthermore, we collected both soft and hard indicators,
which is particularly important in our context given the partially significant lead of
soft indicators as compared to hard data.
III. The thick-modelling exercise
We focus our analysis on testing the sensitivity of simple policy rules to changes in
the choice of the measures taken for the set of feedback variables that the central bank
is assumed to respond to.
16
The resulting range of rule prescriptions may be
interpreted as the degree of uncertainty or noise stemming from the different signals
policy-makers receive from the broad range of indicators available before each
interest-rate setting meeting. In doing so, we need to make prior assumptions
regarding the general policy framework underlying our sensitivity analysis. Parallel
changes to the specification of the rule would not allow for disentangling the
differences resulting from the choice of the measure or the design of the policy rule.
Various different specifications have been proposed in the literature so far. Consider
first the very general family of monetary policy rules that nests a variety of the
proposed specifications (see e.g. Clarida et al. 2000):
17
where i
t
is the recommended level for the policy rate, r* the real equilibrium interest
rate, * the inflation target of the monetary authority,
t
a measure of price
developments and (y
t
- y*) a measure of economic slack. The integer parameters h and
k refer to the forecast horizon for inflation and output respectively. This specification
collapses to Taylors original rule when r* =2, h = k = = 0 and
=
y
= 0.5.
16
Our analysis is therefore in the spirit of Granger and Jeon (2004).
17
We follow Clarida, Gal and Gertler (1999,2000) in that we consider the nominal equilibrium interest
rate to consist of the real equilibrium rate r* and the inflation objective of the central bank rather than
last periods inflation rate as is often used.
15
ECB
June 2010
Allowing 0 embeds a gradual, partial adjustment of the policy rate to the original
Taylor rate, reflecting the possibility of interest rate smoothing.
18
Estimation of policy rules in the form of equation (1) is common in the literature.
However, estimation of these rules is not without problems. As pointed out by
Orphanides (2003) a critical aspect of policy rules in the form of equation (1) is the
emphasis they place on a concept of the economys potential level of economic
activity, y*, for calculating the output gap. In theory, as noted above, various different
notions of potential output exist (e.g. flexible-price-output, the steady-state-output, the
NAIRU, the linear and HP-filtered-trend) and it is unclear which of the concepts is the
most appropriate for estimating the cyclical position of the economy. The choice is
not without consequences as demonstrated by McCallum and Nelson (2004). They
show that the use of a mistaken concept of the output gap can imply major welfare
losses within a class of policy rules that rely upon measures of the output gap.
19
Similarly, Orphanides (2003) shows that the key source of the policy failure
associated with the Great Inflation in the United States was the pursuit of activist
policies based on real-time estimates of potential output that were severely overstating
the economys capacity at that time.
Assuming that policy-makers have perfect knowledge in real-time over such latent
variables is therefore highly unrealistic, if not problematic. As a consequence, a
number of authors (McCallum 2001, Orphanides et al. 2000, Orphanides 2003,
Orphanides and Williams 2002, Leitemo and Lnning 2006, Williams 2006) advocate
the use of rules that are based on growth rates as they suffer from fewer measurement
problems.
20
Consider the generalised family of growth rules:
18
The inclusion of policy gradualism can be theoretically justified on several grounds, see e.g.
Woodford (1999) and Sack and Wieland (2000).
19
They compare welfare losses with a rule using the McCallum and Nelson (1999) measure of the
output gap, which resembles the flexible-price-output, and a rule employing a simple linear trend as a
measure of potential output.
20
As Orphanides (2003) points out errors in the measurement of the output gap are highly serially
correlated. As a result, mismeasurements in the level of the output gap are more pronounced than in the
first difference of the output gap.
16
ECB
June 2010
where, contrary to the Taylor-type rules in equation (1), these rules do not any longer
require knowledge of the potential level of output or the natural rates of
unemployment for setting policy. Such growth rules are hence a simple and
effective approach for dealing with the ignorance about the degree of uncertainty
surrounding the estimates of potential output in real time. Avoiding the level of the
output gap in the estimation of policy rules is a robust approach for hedging against
non-trivial measurement problem.
Besides their usefulness in this respect, there are two other paramount factors that
speak in favour of using growth rules over other Taylor-type rules. The first factor
relates to the optimality of growth rules. There is now plenty of evidence that
growth or difference rules may outperform standard Taylor rules in their capacity
to stabilise inflation and output volatility (Orphanides and Williams (2002), Walsh
(2003), Stracca (2006)).
21
Another criterion in favour of growth rules is rather
pragmatic albeit no less important. It is the ECBs notorious dismissal of the output
gap as a reliable indicator for the conduct of monetary policy. This can be seen on
various occasions. For instance, as noted already by Gerlach (2007), the ECB never
refers to output gaps in its official communications or publications, such as its
Monthly Bulletin. The ECB itself is also not secretive about its thoughts on the
usefulness of the output gap. For example, Papademos (2005) said that natural rates or
the output gap do not play a prominent role in the ECBs strategy for determining the
monetary policy stance. President Trichet in November 2004 was even clearer in
stating that it would even be dangerous to derive monetary policy decisions from
such an indicator. In my view, the example of the output gap demonstrates that
theoretical economic models and monetary policy practice are, at times, quite far
apart. Hence, if the objective of the paper lies in identifying a robust policy rule for
the euro area, it would seem cynical to impose a reaction of the ECB to a variable that
we know ex ante it deliberately discards as an input to its decision-making process.
Taken together, there is strong theoretical and empirical support for the use of
growth rules when it comes to examining the ECBs past interest rate setting. We
therefore proceed in estimating policy rules in the form of equation (2) for the ECB
21
Difference rules are usually referred to as rules with a smoothing coefficient of one.
17
ECB
June 2010
over the period from 1999 to 2007. The analysis is carried out using our real-time
database described in Section II. As the database consists of both past and
contemporaneous data as well as of projections for inflation and real economic
activity, we can estimate both backward and forward-looking policy rules. For the
most part, estimation of forward-looking rules relying on ex post revised data has
called for the use of instrumental variable (IV) techniques to counter the problem of
endogeneity (e.g. Clarida, Gal and Gertler 2000). Although in theory GMM
estimation has been shown to produce consistent estimates under some assumptions,
the choice of the instruments is critical for determining the finite sample properties, in
particular the bias, of the IV estimator. One thing to remember is that unlike ordinary
least squares (OLS) estimation under a zero conditional mean assumption, IV
methods are never unbiased when at least one explanatory variable is endogenous
(Wooldridge 2001). Even in large samples IV methods can be ill-behaved if the
instruments are weak.
22
For our analysis, and similar to Orphanides (2001), it is not
necessary to instrument for the various measures of inflation and real economic
activity. The reason is that our real-time database is constructed to reflect only
information actually available to policy-makers at any given point in time, avoiding
the issue of endogeneity plaguing estimations employing ex post revised data. Least
squares estimates will therefore provide consistent and unbiased estimates of the
parameter space under standard assumptions.
III.1 Selection of rules
We begin by estimating a total of 3,330 policy rules for the euro area, combining 37
inflation measures and 90 indicators of real economic activity. Naturally, not all
estimated rules are likely to be meaningful from either a statistical or economic point
of view or both. As mentioned above, we are not claiming that any of these rules
mirrors the real reaction function of the ECB, but that the bulk of rules reflects the
degree of uncertainty and the range of signals policy-makers face when combining all
available indicators. We apply four filters that reduce the number of specifications to
a set of rules that delivers a meaningful approximation of the ECBs past interest-rate
setting behaviour. The first filter is of a purely statistical nature. We require the
22
Another potential problem with applying IV methods is that the standard errors have a tendency to be
large. The magnitude of the errors depends, among others, on the quality of the instruments.
18
ECB
June 2010
estimates of
and
y
to be statistically significant at least at the 10% confidence
level. This first selection reduces the number of rules to 617, i.e. slightly more than
2,700 specifications are rejected by the data solely on the grounds of a poor statistical
description of the underlying data generating process.
In a second step we restrict the estimates of
and
y
to be strictly positive.
23
That is,
we only consider rules that have a minimum of stabilising properties and are in line
with the ECBs primary objective of maintaining price stability. Only if monetary
policy responds to an increase in both inflation and output growth by systematically
raising the policy rate, the central bank will be able to stabilise the economy and
deliver price stability. Note that, at this stage, we do not require single rules to fulfil
the Taylor principle, i.e. the requirement of a more than proportional response of the
nominal interest rate to movements in the inflation rate. While the Taylor principle
does not necessarily have to hold for each and every specification, we would expect it
to hold on average (see Section III.2). The second selection criterion further reduces
the number of considered rules to 591.
The third filter applied to the reduced set of rules relates to some identification issues
in estimating equation (2). In the absence of further assumptions estimation of
equation (2) will only allow identifying the term r* + (1-
) *, but not r* or *
separately. While exact identification of these equilibrium measures is not the focus of
our analysis, and indeed all previous studies on the euro area policy rule literally
neglect this issue, we still require our rules to produce reasonable estimates for both
concepts.
24
We pursue the reverse of the approach proposed by Clarida, Gal and
Gertler (2000) and take the sample average of each inflation indicator as a measure of
the implicit inflation objective to recover an estimate of the real equilibrium interest
rate.
25
For the rule to be accepted, we require the resulting equilibrium interest rate to
23
The only exception is the reaction to the unemployment rate or the forecast thereof. An increase in
the unemployment rate should, on average, trigger a monetary policy easing. We therefore multiplied
all series related to the unemployment rate with minus one to allow interpreting the output coefficient
in a similar manner to the other measures of economic activity.
24
Several avenues have been proposed in the literature to identify either the natural rate of interest or
the inflation objective. For instance, given that the US Fed has not announced an explicit inflation
target, Clarida, Gal and Gertler (2000) took the sample average of the real interest rate as a measure of
the equilibrium real rate, allowing them to recover an estimate of the implicit inflation target *.
25
One way that would permit identifying the natural rate of interest r* in studies for the euro area
would be to assume target inflation * to take a value close to the ECBs definition of price stability of
19
ECB
June 2010
lie within the confidence interval suggested by Benati and Vitale (2007). They show
that natural rate estimates in the euro area have historically been characterised by a
significant extent of uncertainty, to the point that the confidence interval
corresponding to one standard deviation, i.e. the 16th and 84th percentiles of the
simulated distributions, stretches from about -2 to 6 per cent in the period from 1999
to 2006 in the euro area.
26
Applying this range to our estimated rules further reduces
the number of considered specifications to 428 rules.
Figure 3. Standard errors of the economic growth variable
0.0
0.5
1.0
1.5
2.0
2.5
1 51 101 151 201 251 301 351 401
Number of rules
S
t
a
n
d
a
r
d

e
r
r
o
r
Finally, from the remaining rules we discard those specifications that have standard
errors of the estimated coefficients significantly different from the general
distribution. Figure 3 illustrates this trimming procedure based on the distribution of
the standard errors of the economic growth variable. As can be seen from Figure 3 the
standard errors of around 85% of rules ranges between 0 and 0.5. Thereafter, the
standard errors start to grow exponentially and become significantly different from the
general distribution of errors of the entire range of estimated rules. We therefore
discard all specifications with a standard error of the economic growth variable larger
than the threshold value of 0.5. In a similar vein, we ignore rules with a standard error
of the smoothing coefficient larger than 0.04, a standard error of the constant larger
than 0.7 and a standard error of the coefficient of the inflation weight larger than 1.5
close, but below 2%. However, while this approach appears as a reasonable approximation in
specifications featuring HICP headline inflation (the index for which the ECB defined its price stability
objective) or the forecasts thereof, it is of no avail in rules that consider alternative inflation measures,
such as wage growth projections, unit labour costs or any sub-index of the HICP.
26
The sample average of the ex post real interest rate in the euro area, calculated as the difference
between the EONIA and annual HICP inflation, is 1.18.
20
ECB
June 2010
(see Annex 2 for the corresponding charts). This final selection reduces the number of
considered rules to 291. That is, around 9 per cent of the original 3,330 rules deliver a
meaningful statistical and economic description of the ECBs past interest-rate setting
behaviour. In the next section we discuss the estimated properties of these rules.
III.2 Estimation results
Table 3 presents an overview of the least squares estimates based on real-time data.
We report simple averages of the estimated coefficients, standard errors and the R
2
.
That is, we assume that policy-makers assign equal weight to each of the relevant
combinations of inflation and output indicators when forming their policy decision.
27
Given that the whole idea behind thick modelling is to hedge against possible
misspecifications, assigning equal weight to each indicator minimises the risks
associated with model uncertainty.
Table 3. Mean estimates of policy rules for the euro area (1999 to 2007)
(stand. errors in brackets; all coefficients statistically significant at least at the 10% confidence level)

A number of interesting results stand out from the estimation. First, considering all
291 rules, which include both backward and forward-looking specifications, the mean
estimate of
equals 1.25, which is noticeably above unity and is indicative for the
fact that real interest rates in the euro area are likely to have risen in response to an
increase in inflation over the estimation sample. However, Table 3 clearly shows that
the strength of the ECBs response to movements in inflation critically depends on
whether the rule is backward or forward-looking. Policy rules that are specified only
in terms of expectations for both inflation and output bear a much larger response (
= 2.22) to inflation as other rules would predict. Purely backward-looking rules,

which respond to lagged or contemporaneous variables, have, on average, an inflation
27
We experimented with various weighting schemes, such as with the standard errors or the R
2
, to
compute the average, but the general results remain virtually identical.
No of
rules
Share
of total
r*

y
SEE R
2
All rules 291 0.92 -0.53 1.25 0.94 0.2007 0.9562
(0.03) (0.37) (0.46) (0.19)
Forward-looking in inflation and output 48 16.5% 0.91 -1.06 2.22 1.05 0.2002 0.9564
(0.03) (0.39) (0.81) (0.15)
Backward-looking in inflation and output 117 40.2% 0.92 -0.20 0.55 0.82 0.2006 0.9562
(0.03) (0.36) (0.18) (0.2)
Mixed rules 126 43.3% 0.91 -0.64 1.52 1.00 0.2010 0.9560
(0.03) (0.39) (0.58) (0.18)
21
ECB
June 2010
coefficient nowhere near one (
= 0.55). Four reasons are likely to drive the

difference in the response. First, a more activist behaviour by policy-makers in rules
based on projections can be partly explained by the larger degree of inertia present in
most forecasts. For instance, the average standard deviation of the 13 HICP headline
projections considered in our estimations amounts to 0.27, which is less than half the
standard deviation of realised past HICP headline inflation (0.59).
28
Second, a
stronger response to projected inflation underlines the ECBs strong commitment to
anchor medium to longer term inflation expectations. Any deviation of these
indicators from their long-run average is strongly accommodated by the ECB.
29
Third,
a coefficient on the inflation variable larger than one in forward-looking rules is
indicative for the fact that successful stabilisation policy requires a more forward-
looking approach. Because monetary policy operates with a lag, real interest rates
need to rise in response to an increase in projected inflation, not past inflation.
Finally, by its nature, a monetary policy aiming at delivering price stability over the
medium term, which is the ECBs defined objective, should not respond too strongly
to short-term movements in past inflation, especially after the economy is hit by a
large shock. This seems all the more warranted at the current juncture where short-
term inflation developments are characterised by a high degree of volatility and a
medium-term orientated view is indispensable in formulating the response of
monetary policy.
30
The extent of variation in the output coefficient
y
is much less substantial. It ranges,
on average, from 0.82 in the fully backward-looking rules to 1.05 in the purely
forward-looking models, with the mean over all rules being estimated at 0.94. The
policy smoothing coefficient on the lagged interest-rate is even less dependent on the
measure of inflation and economic activity and indicates a generally high degree of
inertia, around 0.9, in the ECBs policy response. An interesting result is the fact that
the equilibrium real interest rate is found to be slightly negative on average over the
estimation sample. Only around 25% of all final rules have a positive rate. Naturally,
the identification of the equilibrium real rate depends, as discussed before, on the
28
The same holds true for real annual GDP growth, for which the average standard deviation of its
forecasts accounts for 0.56, which compares to 0.95 for the latest available release.
29
The rate hike of July 2008 is a forceful demonstration of the ECBs commitment in this respect.
30
In this sense, even a two-year horizon, which is the widest horizon of inflation expectations used in
this study, may be too short at present.
22
ECB
June 2010
assumption of the inflation target in each specification. The results should therefore be
interpreted with caution. Nevertheless, the low or even negative level of the average
real rate confirms, on the one hand, the results by Benati and Vitale (2007), who find
a non-trivial probability for the equilibrium rate to be negative in the euro area over a
similar sample, and, on the other hand, the low interest-rate environment prevailing in
the euro area since the advent of the single monetary policy.
Figure 4. The inflation and output coefficient: minimum, mean and maximum
Inflation coefficient
0
1
2
3
4
5
All rules Forward-
looking rules
Backward-
looking rules
Mixed rules
Output coefficient
0
1
2
3
4
5
All rules Forward-
looking rules
Backward-
looking rules
Mixed rules
Figure 4 shows the distribution of the coefficients across different rules. As becomes
apparent from Figure 4 the range of coefficients can, at times, be relatively large,
underlining the plurality of signals that policy-makers receive from the different
indicators and combinations thereof. In other words, rules that share similar statistical
properties, given the homogeneous selection criterions applied to all rules in terms of
goodness of fit, statistical significance or the size of the standard errors, can result in
very different descriptions of the ECBs past interest rate setting. The large
differences highlight the potentially misleading recommendations that can emerge
from the estimation of policy rules based on a single or a small set of indicators.
At the same time, large ranges are not necessarily the result of a thick-modelling
exercise. For example, the inflation coefficients in backward-looking rules or the
output coefficients in forward-looking rules are clustered along a relatively small
range of values (see Figure 4), suggesting that the choice of the indicators plays a less
dominant role in these rules as compared to others. In a similar vein, the distributions
of the coefficients are in most cases not normal, but skewed towards the lower tail,
implying that the broad mass of coefficients is centered around a smaller range.
23
ECB
June 2010
Figure 5. The inflation and output coefficients: distribution over all 291 rules
Inflation coefficient
0
20
40
60
80
100
120
0.0 -
0.5
0.5 -
1.0
1.0 -
1.5
1.5 -
2.0
2.0 -
2.5
2.5 -
3.0
3.0 -
3.5
3.5 -
4.0
4.0 -
4.5
Interval
N
u
m
b
e
r

o
f

r
u
l
e
s
mean
Output coefficient
0
20
40
60
80
100
120
0.0 -
0.5
0.5 -
1.0
1.0 -
1.5
1.5 -
2.0
2.0 -
2.5
2.5 -
3.0
3.0 -
3.5
3.5 -
4.0
4.0 -
4.5
4.5 -
5.0
Interval
N
u
m
b
e
r

o
f

r
u
l
e
s
mean
Figure 5 illustrates this on the basis of a histogram of the inflation and output
coefficients over all 291 rules. As regards the inflation coefficient, 206 specifications,
or around 71% of all final rules, are estimated with a coefficient between 0 and 1.5,
while less than a third of all rules share the much larger interval between 1.5 and 4.5.
The distribution is even more highly skewed for the estimated output coefficients. In
this case, 93% of all rules have a coefficient of below 1.5. Put another way, although
non-trivial differences exist in the estimated coefficients of both output and inflation
across the entire battery of rules, the estimates are largely confined to a relatively
narrow and homogeneous range.
Figure 6. Fitted values of the estimated policy rules
Panel A
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Jan-
99
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
(
f
i
t
t
e
d
)

E
o
n
i
a
minimum bid rate
minimum
mean over all rules
maximum
Panel B
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Jan-
99
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
(
f
i
t
t
e
d
)

E
o
n
i
a
minimum bid rate
mean over all f orward-looking rules
mean over all backward-looking rules
24
ECB
June 2010
This can also be seen by plotting the corridor of recommendations borne from these
rules. Figure 6 (Panel A) shows the very close adherence of the mean fitted values to
the ECBs actual policy-making. The close fit does not originate from any particular
type of policy rule. Panel B of Figure 6 illustrates that, on average, both backward-
and forward-looking rules provide a similar depiction of the recommended path of
policy rates. More strikingly, the actual policy rate virtually lies at any point in time
within the corridor prescribed by the range of policy rules, the only exception being
the extended period of low interest rates in late 2005.
31
In a way, the close fit is a
validation of the fact that the ECBs policy-makers indeed consider a broad range of
indicators, while their decisions seem to converge to the average of the policy path
prescriptions. That does not mean that the rules recommending very large or very low
interest rates are less suited to describe ECB policy-making. Note that these rules
passed the strict selection criteria presented in Section III.1. These rules merely
highlight the fact that, according to the development of some inflation and output
growth indicators, interest rates should on average be higher or lower than what the
majority of rules would predict. Policy-makers should not ignore these signals, but
consider them as one input among many.

Table 4. Mean estimates of policy rules for the euro area: inflation measures
No of
rules
Share
of total
r*

y
SEE R
2
Forward-looking in inflation
1)
108 37.1% 0.92 -0.67 2.48 1.02 0.2016 0.9558
(0.03) (0.35) (0.94) (0.17)
Projections for the current year 26 8.9% 0.92 -0.59 1.23 0.98 0.2025 0.9555
(0.03) (0.34) (0.58) (0.17)
Projections for the next year 82 28.2% 0.92 -0.69 2.88 1.03 0.2013 0.9559
(0.02) (0.35) (1.05) (0.17)
Backward-looking in inflation
1)
183 62.9% 0.91 -0.45 0.51 0.89 0.2002 0.9564
(0.03) (0.39) (0.17) (0.19)
HICP rules 43 14.8% 0.91 -0.27 0.64 0.98 0.2016 0.9558
(0.03) (0.38) (0.27) (0.18)
IPP rules 105 36.1% 0.92 -0.48 0.31 0.71 0.1998 0.9565
(0.02) (0.41) (0.1) (0.2)
Wage growth rules 35 12.0% 0.90 -0.57 0.98 1.30 0.1994 0.9567
(0.03) (0.34) (0.28) (0.18)
1) Output growth measures can be past, contemporaneous or projections.

Table 4 shows in more detail the properties of the estimated coefficients depending on
the measure of inflation that is taken. An interesting result is the stark difference in

31
With the benefit of hindsight an earlier removal of the policy accommodation in line with the
recommendations of the estimated policy rules might have been warranted.
25
ECB
June 2010
the inflation coefficient in rules with inflation projections for the current and next
year. The ECB seems to react almost twice as strongly to movements in inflation
expectations for the next year as compared to the current year. Two other noteworthy
results stand out. First, rules that include industrial producer prices as a measure of
inflation make up a large part of the finally selected specifications (36.1%).
32
This
measure hardly appears in other studies examining the ECBs interest-rate setting
behaviour, emphasising the importance of the thick-modelling exercise. Second, rules
that are specified in terms of wage growth indicators result in a relatively large
inflation coefficient, in particular when compared to other backward-looking rules.
33

Such behaviour on part of policy-makers can partly be explained by the lagged effects
of price developments in labour markets on headline inflation. Past increases in labour
costs may still exert upward pressure on consumer prices in the future, therefore
requiring a stronger response to these indicators as compared to other inflation
measures.

Table 5. Mean estimates of policy rules for the euro area: output measures
No of
rules
Share
of total
r*

y
SEE R
2
Forward-looking in output
1)
114 39.2% 0.91 -0.96 1.20 1.03 0.1998 0.9566
(0.03) (0.43) (0.44) (0.17)
Real GDP growth 63 21.6% 0.90 -1.15 0.87 1.23 0.1992 0.9569
(0.03) (0.43) (0.38) (0.18)
Private consumption growth 14 4.8% 0.92 -1.30 1.89 1.41 0.2028 0.9553
(0.02) (0.55) (0.63) (0.28)
Investment growth 30 10.3% 0.91 -0.54 1.70 0.49 0.1995 0.9567
(0.02) (0.35) (0.51) (0.08)
Industrial production growth 7 2.4% 0.93 -0.42 0.64 0.74 0.2005 0.9563
(0.02) (0.5) (0.29) (0.2)
Backward-looking in output
1)
177 60.8% 0.92 -0.25 1.28 0.88 0.2013 0.9559
(0.03) (0.34) (0.47) (0.2)
Real GDP growth 20 6.9% 0.90 -0.42 0.82 1.17 0.2008 0.9561
(0.03) (0.35) (0.36) (0.21)
Private consumption growth 59 20.3% 0.92 -0.65 1.79 1.08 0.2030 0.9551
(0.03) (0.39) (0.64) (0.22)
Investment growth 65 22.3% 0.92 0.27 0.99 0.66 0.1990 0.9569
(0.02) (0.23) (0.37) (0.16)
Domestic demand growth 33 11.3% 0.93 -0.48 1.24 0.79 0.2030 0.9551
(0.02) (0.48) (0.46) (0.22)
1) Inflation measures can be past, contemporaneous or projections.

32
We consider four measures of industrial producer prices: headline, capital goods, manufacturing
goods and intermediate goods.
33
The two measures capturing price developments in labour markets are unit labour costs and
compensation per employee.
26
ECB
June 2010
Finally, Table 5 highlights the differences in the estimated coefficients that emerge
with the use of different choices for the output growth measure. ECB policy-makers
seem to react very differently to the type of projected variable. They appear to
respond most actively to the annual growth rate of real GDP and private consumption.
Interestingly, this is true for both backward- and forward-looking rules. In particular,
there is hardly any difference in the ECBs estimated response to forecasts (
y
= 1.23)
or past figures (
y
= 1.17) on real GDP growth. Another finding is of importance as
regards a comparison of the results with previous attempts to describe the ECBs past
interest-rate setting behaviour. While the vast majority of these studies had employed
industrial production as a measure of economic activity, this indicator seems to play a
minor, if not subtle role in growth rules and after applying the selection criterions
outlined in Section III.1. Out of 148 rules that could possibly nest industrial
production as a measure of economic activity, only seven rules were finally selected
as providing an economically and statistically meaningful description of the ECBs
conduct of monetary policy over the estimation sample. None of these rules are
backward-looking in industrial production, which is the dominant assumption in most
previous examinations of the ECBs policy setting.
III.3 Confidence intervals for thick modelling

In his seminal paper Taylor (1993) proposes a monetary policy reaction function that
features no distinct central bank preference over the two objectives of stabilising
inflation and output around target. This can be seen by the equal weights of 0.5 that
Taylor assigns on the monetary authoritys reaction to inflation and output. The rich
spectrum of policy rules that we consider in this paper can be used as a laboratory to
investigate the relevance of the originally specified Taylor rule in the historical
conduct of monetary policy in the euro area.
In doing so, we test whether the confidence intervals from thick modelling include the
weights of 0.5 suggested in the Taylor rule. We follow Granger and Jeon (2004) and
construct the confidence intervals by bagging (bootstrap aggregation). For each of
the 291 estimated rules we stationary bootstrap the residuals over 10000 iterations and
re-estimate the rule to reproduce the main properties of the data. The original rule is
then estimated using the bootstrapped data. For each iteration, 291 different sets of the
27
ECB
June 2010
estimated coefficients are pooled to make the thick model estimates. Table 6
summarises the mean estimates over all previously estimated 291 policy rules as well
as the average estimates over all 10000 bootstrap regressions for all 291 rules together
with the corresponding standard deviations.
Table 6. Taylor rule bootstrapping results
r*

y r*

y
All rules 0.92 -0.53 1.25 0.94 0.92 -0.53 1.25 0.94
(0.03) (0.37) (0.46) (0.19) (0.00) (0.19) (0.44) (0.12)
Thick modelling Bootstrap aggregation
In line with the results reported in Granger and Jeon (2004) our bootstrap procedure
produces smaller bagging-implied standard errors compared to those obtained by
simple averaging over the thin specifications. This is particularly true for the S.D.s of
the smoothing, the real interest rate and the output weight coefficients. The 95%
confidence intervals for the inflation and output coefficients implied by the bootstrap
procedure amount to (0.377, 2.122) and (0.696, 1.180) respectively. In contrast to
Granger and Jeon (2004) who document a non-compliance of the magnitude of the US
monetary policy reaction to inflation with the original Taylor (2003) value we find no
evidence to reject the hypothesis that the ECB has been responding to inflationary
pressures in line with the Taylors original specification. In contract, there is empirical
evidence to suggest that the ECB has been more aggressive in its reaction to real
activity developments than proposed by Taylor (2003).
IV. Principal component analysis
Thick-modelling deals with the data-rich environment in which policy-makers
operate by synthesising the estimation results of many alternative specifications. An
alternative to this approach is to condense the information of large datasets prior to
the estimation by extracting common factors that can explain some of the co-
movement in the data and to use these factors in macroeconomic modelling. Amongst
the earliest research in this field have been the seminal papers by Stock and Watson
(1999) and Forni et al. (2000).
28
ECB
June 2010
In what follows we adopt a Principal Component factor-extracting approach that is
designed to summarize the essential dynamics contained in our real-time data set in a
narrower set of factors. While a fundamental feature of most existing research in this
field has been the practice of extracting factors from the whole set of macroeconomic
series available (e.g. Bernanke and Boivin 2003, Bernanke, Boivin and Eliasz 2005),
the factors extracted in this way do not carry a meaningful structural interpretation. To
avoid this limitation we follow the approach suggested by Belviso and Milani (2006)
who suggest the use of structurally interpretable factors in the econometric analysis.
We assume that the variables in our real-time data set are contained in the 1 M
vector
t
X (M stands for the number of all variables present in the data set) while
t
Y
contains the ECB policy instrument with T t ,...., 2 , 1 = . We assume further that the
dynamics of the variables in
t
X is governed by:
(3)
t t t
F X c + A =
where
t
F is a 1 K vector of unobservable fundamental factors,
t
c is a vector of
errors that satisfy ( ) 0 | =
t t
F E c and ( ) 0 |
, ,
=
t n t m
E c c for all N n m ....., , 1 , = and
n m = and A is a conformable matrix of factor loadings. We split our real-time data
set
t
X into subvectors
1
t
X ,
2
t
X ,.,
I
t
X each containing series relating to the same
macroeconomic concept and such that
i
t
X represents an 1
i
N vector satisfying
=
i
i
N N . The fundamental feature allowing us to obtain factors that have a clearly
distinct structural interpretation is our assumption that each subvector
i
t
X is related to
only a particular subset of the fundamental factors vector
t
F . In particular, we
propose a division of
t
F into
1
t
F ,
2
t
F ,.,
I
t
F with
i
t
F being a 1
i
K vector such that
=
i
i
K K and
i i
N K < for all i. Then the dynamics of
i
t
X is assumed to be captured
by exactly one subfactor
i
t
F as given by:
(4)
t
I
t
t
t
f
I
f
f
I
t
t
t
F
F
F
X
X
X
c +
(
(
(
(
(
(
(
(
(
(
A
A
A
=
(
(
(
(
(
....
... 0 0
... ... .... ...
0 ... 0
0 ... 0
.....
2
1
2
1
2
1
with ( ) 0 ,..., |
1
=
I
t t t
F F E c and ( ) 0 |
, ,
=
t n t m
E c c for all N n m ....., , 1 , = and n m = .
Each subfactor
i
t
F is assumed to be the first principal component of the respective
29
ECB
June 2010
variable subgroup
i
t
X ; that is, the principal component that captures the greatest
fraction of the total variability of the subgroup
i
t
X .
We partition our entire real-time data set into three inflation and three output growth
groups that give rise to six distinct factors used in the subsequent econometric
analysis:
34
- Past inflation
- Current year inflation projections
- Next year inflation projections
- Past output growth
- Current year output growth projections
- Next year output growth projections
Prior to the extraction of the factors, each time-series is demeaned and standardized to
have a unit variance. This initial standardization is performed in order to prevent more
volatile series from overly influencing the obtained factors. However, to obtain a
meaningful interpretation of the estimated coefficients, there is a need to restandardise
each principal component before the estimation. Normally, the component series with
the highest correlation with the principal component is taken for the restandardisation
(see e.g. Kapetanios 2004). However, given that this procedure is to a large extent
arbitrary but, at the same time, linked with non-negligible repercussions on the
estimated coefficients in a policy rule context, we choose to normalise each extracted
factor by the means and standard deviations of all component series that have a
correlation with the respective first principal component of more than 80 per cent.
Given that no logical preference can be made for any of the resulting series, we
proceed in estimating the policy rules with all restandardised factors and average the
estimation results over all series that belong to the same group. In so doing, we hedge
against the arbitrariness in basing the first and second moments of the principal
component on just a single component series.

34
The reader is referred to Annex 1 for a description of the composition of each of the groups.
30
ECB
June 2010
IV.1 Estimation results
This procedure results in multiple estimation combinations similar to the thick-
modelling approach. In total, we estimate 2,088 policy rules on the basis of the
restandardised principal components. We apply the same statistical and economic
filters as outlined in Section III.1, the only exception being the trimming by standard
errors since the dispersion of the latter was very limited. The filtering reduces the
number of meaningful rules to 390. Table 7 presents the mean estimates of the
selected factor-based rules.
Table 7. Mean estimates of factor-based rules for the euro area (1999 to 2007)
r*

y SEE R
2
All rules 0.89 3.13 0.72 0.38 0.2004 0.9563
(0.03) (0.17) (0.23) (0.05)
Purely forward-looking rules 0.88 3.73 1.31 0.52 0.1960 0.9582
(0.03) (0.14) (0.31) (0.06)
Inflation (next year) and output growth (current year) 0.90 3.69 0.88 0.40 0.1983 0.9572
(0.02) (0.15) (0.36) (0.05)
Inflation (next year) and output growth (next year) 0.86 3.77 1.74 0.64 0.1937 0.9592
(0.03) (0.13) (0.27) (0.06)
Purely backward-looking rules 0.90 2.42 0.20 0.14 0.2042 0.9546
(0.04) (0.2) (0.1) (0.02)
Mixed rules 0.90 2.97 0.50 0.36 0.2021 0.9555
(0.04) (0.18) (0.23) (0.06)
Inflation (past) and output growth (current year forecast) 0.86 2.81 0.20 0.39 0.1975 0.9575
(0.04) (0.1) (0.05) (0.04)
Inflation (past) and output growth (next year forecast) 0.91 2.82 0.20 0.54 0.2060 0.9538
(0.04) (0.2) (0.09) (0.11)
Inflation (next year forecast) and output growth (past) 0.92 3.27 1.11 0.14 0.2029 0.9552
(0.02) (0.24) (0.54) (0.03)
Table 3 (replicated). Mean estimates of policy rules for the euro area: thick-
modelling (1999 to 2007)
No of
rules
Share
of total
r*

y SEE R
2
All rules 291 0.92 -0.53 1.25 0.94 0.2007 0.9562
(0.03) (0.37) (0.46) (0.19)
Forward-looking in inflation and output 48 16.5% 0.91 -1.06 2.22 1.05 0.2002 0.9564
(0.03) (0.39) (0.81) (0.15)
Backward-looking in inflation and output 117 40.2% 0.92 -0.20 0.55 0.82 0.2006 0.9562
(0.03) (0.36) (0.18) (0.2)
Mixed rules 126 43.3% 0.91 -0.64 1.52 1.00 0.2010 0.9560
(0.03) (0.39) (0.58) (0.18)
For convenience we replicate Table 3 with the results of the thick-modelling exercise
here again. As becomes apparent from Table 7 only six of the nine possible
combinations of inflation and output growth groups have passed the filters of Section
III.1. In particular, the first principal component of the current year inflation
projections group could never be identified as being statistically significant in
31
ECB
June 2010
estimations with any of the output growth groups. To a certain extent, this result
confirms the findings of the thick-modelling exercise where only nine per cent of all
finally selected rules contained an inflation indicator of this group (cf. Table 4).
Several other similarities and differences with respect to the thick-modelling exercise
become apparent. First, while the inflation coefficient in factor-based rules is on
average below unity, purely forward-looking rules are still characterised by
significantly larger inflation coefficients than other rules. Second, similar observations
hold also true for the coefficient on output growth. It is about half the size in the
factor-based rules as compared to the thick-modelling exercise, but its characteristics
across different specifications do not change. Third, the standard errors of all
estimated coefficients are substantially lower as compared to the results presented in
Section III. This reflects the fact that factor-based rules only react to the large drifts in
the data and not to the peculiarities of single indicators. This result is also consistent
with the forecasting power ascribed to factor models (see Stock and Watson 1999).
Finally, large differences emerge with respect to the estimated real interest rate.
35
In
factor-based rules the natural rate is estimated to be in the neighbourhood of around
three per cent, which is substantially larger than the slightly negative rates that were
identified in the thick-modelling approach. This confirms the large degree of
uncertainty surrounding the proper identification of the real interest rate.

Figure 7. Fitted values of the estimated factor-based policy rules
Panel A
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Jan-
99
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
(
f
i
t
t
e
d
)

E
o
n
i
a
minimum bid rate
maximum
mean over all rules
minimum
Panel B
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Jan-
99
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
(
f
i
t
t
e
d
)

E
o
n
i
a
minimum bid rate
mean over all forward-looking rules
mean over all backward-looking rules

35
We did not change the methodology of identifying the real interest rate. Given that the first principal
components were restandardised prior to the estimations, the same inflation means were taken to
approximate the inflation objective (see section III.2).
32
ECB
June 2010
Figure 7 plots the minimum bid rate together with the fitted values for the EONIA rate
implied by the estimates of the factor-based policy rules reported in Table 7. Similar
to the thick-modelling exercise, the mean factor-based rule closely resembles the
ECBs actual interest-rate decisions. The narrower corridor relative to the thick-
modelling exercise mainly stems from the fact that the disperse information coming
from the various indicators has been condensed to fewer factors prior to the
estimation. The qualitative results, however, are similar. The only period in which
policy-makers have deviated from the recommendations borne from estimated factor-
based rules relates to the exceptionally long phase of low policy rates in 2005.
Contrary to the thick-modelling exercise, some differences emerge with respect to the
prescriptions coming from backward- and forward looking rules respectively. While
the latter seem to resemble very closely the ECBs past policy decisions, in particular
the tightening and loosening cycles between 2000 and 2004, backward-looking factor-
based rules recommend, on average, a somehow smoother and at times lagged interest
rate path relative to actual policy-making.
Figure 8. A comparison of the thick-modelling and the factor-based approach
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Jan-
99
Jan-
00
Jan-
01
Jan-
02
Jan-
03
Jan-
04
Jan-
05
Jan-
06
Jan-
07
(
f
i
t
t
e
d
)

E
o
n
i
a
minimum bid rate
f actor-based rules
thick-modelling
Finally, we want to compare the overall fit of the thick-modelling exercise with the
factor-based approach (see Figure 8). The similarity between the two approaches in
terms of fit is striking. Both methods lead to virtually identical prescriptions for the
policy rate in the euro area. This result underlines the mutual power of the thick-
modelling and factor-based approaches in terms of their ability to substantially
compress both the degree of information coming from the variety of indicators and the
33
ECB
June 2010
uncertainty surrounding the recommendations borne from estimated policy reaction
functions based on a single indicator.
V. Conclusion
The economic literature has witnessed the appearance of a wealth of research directed
towards describing the historical pattern of monetary policy. A standard practice of
the majority of the existing research in this field has been the reliance on a few
economic indicators for modelling the conduct of monetary policy making. Such an
approach involves not only an unjustifiably high degree of subjective judgement
concerning which measures of macroeconomic developments are truly relevant for
monetary policy makers but also grossly and unrealistically ignores the vast range of
economic indicators that are constantly monitored by central banks.
The objective of this paper was to address this issue for the euro area using a new
comprehensive real-time database that takes duly into account the informational
rigidities that policy-makers face. We offer two powerful approaches towards dealing
with the potential misspecification of econometric monetary policy models: thick
modelling and factor extraction. Our thick modelling approach aims at exploring the
robustness of the estimated euro area policy reaction function by considering an
exhaustive set of combinations of measures of real economic activity and inflation.
After applying a strict set of efficiency criteria for selecting the final set of estimated
policy reaction functions, we find that almost 300 estimated rules are able to closely
approximate the ECBs past interest-rate setting behaviour in a meaningful statistical
and economic way. This finding constitutes a stark warning for the possible deep
misspecifications that can result from conditioning the evolution of monetary policy
on only a small set of economic indicators. We also find that the ECB is neither
purely backward nor forward-looking, but reacts to a synthesis of the available
information on the current and future state of the economy.
In our second approach to dealing with the issue of model uncertainty factor
extraction we partition our real-time data set into subgroups of similar
macroeconomic concepts. We extract factors with a precise structural meaning and
link monetary policy decisions to changes in these structurally identifiable factors.
34
ECB
June 2010
The major conclusions reached under our factor-based approach largely mirror the
results reached under our thick modelling exercise. In particular, we show that
bundling the large information set that policy-makers are exposed to prior to the
estimation lead to virtually the same prescriptions for the policy rate as compared to
the thick-modelling exercise.
35
ECB
June 2010
References
Belviso, F. and F. Milani (2006), 'Structural Factor-Augmented VARs (SFAVARs) and the
Effects of Monetary Policy', Topics in Macroeconomics, Berkeley Electronic Press, vol. 6(3).
Benati, L. and G. Vitale (2007), 'Joint estimation of the natural rate of interest, the natural rate
of unemployment, expected inflation, and potential output', ECB Working Paper Series No.
797.
Bernanke, B. S. and J. Boivin (2003), 'Monetary policy in a data-rich environment', Journal of
Monetary Economics, Volume 50, Issue 3, April 2003, Pages 525-546.
Bernanke, B., J. Boivin and P. S. Eliasz (2005), 'Measuring the Effects of Monetary Policy: A
Factor-augmented Vector Autoregressive (FAVAR) Approach', The Quarterly Journal of
Economics, MIT Press, vol. 120(1), pages 387-422, January.
Chowdhury, I. and A. Schabert (2008), 'Federal reserve policy viewed through a money
supply lens', Journal of Monetary Economics, Elsevier, vol. 55(2008), pages 825-834.
Clarida, R. and M. Gertler (1996), 'How the Bundesbank conducts monetary policy', in C. D.
Romer and D. H. Romer, eds., Reducing inflation: Motivation and strategy. Chigago:
University of Chicago, 1997.
Clarida, R., J. Gal and M. Gertler (1999), 'The Science of Monetary Policy: A New
Keynesian Perspective', Journal of Economic Literature Vol. XXXVII (December 1999), pp.
16611707.
Clarida, R., J. Gal and M. Gertler (2000), 'Monetary Policy Rules And Macroeconomic
Stability: Evidence And Some Theory', The Quarterly Journal of Economics, MIT Press, vol.
115(1), pages 147-180, February.
Croushore, D. and T. Stark (2001), 'A real-time data set for macroeconomists', Journal of
Econometrics, Elsevier, vol. 105(1), pages 111-130, November.
Domnech, R., M. Ledob, and D. Taguas (2002), 'Some new results on interest rate rules in
EMU and in the US', Journal of Economics and Business 54 (2002) 431446.
ECB (2004), 'The monetary policy of the ECB', European Central Bank, 2nd edition,
Frankfurt am Main.
Faust, J., J. H. Rogers and J. H. Wright (2001), 'An Empirical Comparison of Bundesbank and
ECB Monetary Policy Rules', FRB International Finance Discussion Paper No. 705.
Forni, M., M. Hallin, M. Lippi and L. Reichlin (2000), 'The Generalized Dynamic-Factor
Model: Identification And Estimation', The Review of Economics and Statistics, MIT Press,
vol. 82(4), pages 540-554, November.
Fourans, A. and R. Vranceanu (2004), 'The ECB interest rate rule under the Duisenberg
presidency ', European Journal of Political Economy, Volume 20, Issue 3, September 2004,
Pages 579-595.
Fourans, A. and R. Vranceanu (2007), 'The ECB monetary policy: Choices and challenges',
Journal of Policy Modeling 29 (2007) 181194.
36
ECB
June 2010
Gerdesmeier, D. and B. Roffia (2004), 'Empirical Estimates of Reaction Functions for the
Euro Area', Schweizerische Zeitschrift fr Volkswirtschaft und Statistik 2004, Vol. 140 (1)
3766.

Gerdesmeier, D. and B. Roffia (2005), 'The relevance of real-time data in estimating reaction
functions for the euro area', North American Journal of Economics and Finance, Volume 16,
Issue 3, December 2005, Pages 293-307.

Gerlach, S. (2007), 'Interest rate setting by the ECB, 1999-2006: words and deeds',
International Journal of Central Banking, 3(3), 1-45.

Gerlach, S. and G. Schnabl (2000), 'The Taylor rule and interest rates in the EMU area',
Economics Letters, Volume 67, Issue 2, May 2000, Pages 165-171.

Gerlach-Kristen, P. (2003), 'Interest rate reaction functions and the Taylor rule in the euro
area', ECB Working Paper Series No. 258, September 2003.

Gorter, J., J. Jacobs and J. de Haan (2007), 'Taylor Rules for the ECB using Expectations
Data', Scand. J. of Economics 110(3), 473488, 2008.

Granger, C.W. and Y. Jeon (2004), 'Thick modeling', Economic Modelling 21 (2004) 323-
343.

Hayo, B. and B. Hofmann (2006), 'Comparing monetary policy reaction functions: ECB
versus Bundesbank', Empirical Economics 31:645662 (2006).

Kapetanios, G. (2004), 'Modelling Core Inflation for the UK Using a New Dynamic Factor
Estimation Method and a Large Disaggregated Price Index Dataset', Economics Letters;
85(1), pp. 63-69, (2004).

Kozicki, S. (1999), 'How Useful Are Taylor Rules for Monetary Policy?', Economic Review,
Federal Reserve Bank of Kansas City.

Leitemo, K. and I. Lnning (2006), 'Simple Monetary Policymaking without the Output Gap',
Journal of Money, Credit and Banking, Vol. 38, No. 6 (Sep., 2006), pp. 1619-1640.

McCallum, B. T. (2001), 'Should monetary policy respond strongly to ouput gaps?', American
Economic Association Papers and Proceedings, 91(2), May, 258-262.

McCallum, B. T. and E. Nelson (1999), 'Performance of operational policy rules in an
estimated semi-classical structural model', in Taylor, eds.: Monetary policy rules, Chicago:
University of Chicago Press.

McCallum, B. T. and E. Nelson (2004), 'Timeless perspective vs discretionary monetary
policy in forward looking models', Federal Reserve Bank of St. Louis Review, 86(2), pp. 43-
56.

Orphanides, A. (2001), 'Monetary policy rules based on real-time data', American Economic
Review, Vol. 91, No. 4, pp. 964-985.

Orphanides, A. (2003), 'The Quest for prosperity without inflation', Journal of Monetary
Economics, 2003, pages 633-663.

37
ECB
June 2010
Orphanides, A. and J.C. Williams (2002), 'Robust monetary policy rules with unknown
natural rates', Brookings Papers on Economic Activity, Economic Studies Program, The
Brookings Institution, vol. 33(2002-2), pages 63-146.
Orphanides, A., R. D. Porter, D. Reifschneider, R. Tetlow and F. Finan (2000), 'Errors in the
measurement of the output gap and the design of monetary policy', Journal of Economics and
Business, Volume 52, Issues 1-2, January-April 2000, Pages 117-141.
Papademos, L. (2005), Interview with the Financial Times and Financial Times Deutschland
on 19 December 2005.
Sack, B. and V. Wieland (2000), 'Interest-rate smoothing and optimal monetary policy: a
review of recent empirical evidence', Journal of Economics and Business 52, pp.205228.
Sauer, S. and J.- E. Sturm (2007), 'Using Taylor Rules to Understand ECB Monetary Policy',
German Economic Review, Vol. 8, No. 3. (August 2007), pp. 375-398.
Smets, F. and R. Wouters (2003), 'An estimated stochastic dynamic general equilibrium
model of the euro area', Journal of European Economic Association, September 2003, Vol 1
(5), pp 1123-1175.
Stock, J.H. and M.W. Watson (1999), 'Forecasting Inflation', Journal of Monetary Economics
44 (2), 293335.
Stock, J.H. and M.W. Watson (2002), 'Macroeconomic forecasting using diffusion indexes',
Journal of Business & Economic Statistics 20 (2), 147162.
Stracca, L. (2006), 'A speed limit monetary policy rule for the euro area', ECB Working Paper
Series No. 600, April 2006.
Taylor, J. (1993), 'Discretion versus policy rules in practice', Carnegie-Rochester Conference
Series on Public Policy, 39, pp.195-214.
Trichet, J.-C. (2004), The ECBs use of statistics and other information for monetary
policy, Keynote address on the occasion of the OECDs World Forum on Key Indicators
Statistics, Knowledge and Policy, Palermo 11 November 2004.
Walsh, C. (2003), 'Speed Limit Policies: The Output Gap and Optimal Monetary Policy ',
American Economic Review, 93(1): 265278.
Williams, J. (2006), 'Robust Estimation and Monetary Policy with Unobserved Structural
Change', Economic Review, Federal Reserve Bank of San Francisco, 2006, pages 1-16.
Woodford, M. (1999), 'Optimal monetary policy inertia', NBER Working Paper No.7621.
Wooldridge, J. M. (2001), 'Econometric Analysis of Cross-Section and Panel Data', The MIT
Press, Cambridge, Massachusetts.
38
ECB
June 2010
Annex 1. List of output and inflation measures used in the estimations
Table A1. 1 List of output growth indicators (in alphabetical order):
Factor
36
Indicator
P Consumer confidence (EC, change w.r.t. last GovC meeting)
P Consumer confidence (EC, change w.r.t. 3 months before)
P Domestic demand (qoq, latest available figure)
P Domestic demand (qoq, latest available figure-1)
P Domestic demand (qoq, average of 4 latest available figures)
P Domestic demand (yoy, latest available figure)
P Domestic demand (yoy, latest available figure-1)
P Domestic demand (yoy, average of 4 latest available figures)
P Economic slack (Domestic demand (yoy) - real GDP (yoy))
P Economic slack (Domestic demand (qoq) - real GDP (qoq))
P Foreign annual GDP growth forecast (US,JP,UK)
P Foreign annual industrial production growth forecast (US,JP,UK)
P Net Exports (qoq, latest available figure)
P Net Exports (qoq, latest available figure-1)
P Net Exports (qoq, average of 4 latest available figures)
P Net Exports (yoy, latest available figure)
P Net Exports (yoy, latest available figure-1)
P Net Exports (yoy, average of 4 latest available figures)
P Gross Fixed Capital Formation (qoq, latest available figure)
P Gross Fixed Capital Formation (qoq, latest available figure-1)
P Gross Fixed Capital Formation (qoq, average of 4 latest available figures)
P Gross Fixed Capital Formation (yoy, latest available figure)
P Gross Fixed Capital Formation (yoy, latest available figure-1)
P Gross Fixed Capital Formation (yoy, average of 4 latest available figures)
FC Gross Fixed Capital Formation forecast current year (Consensus)
FN Gross Fixed Capital Formation forecast next year (Consensus)
P Industrial production (yoy)
P Industrial production confidence (EC, change w.r.t. last GovC meeting)
P Industrial production confidence (EC, change w.r.t. 3 months before)
FC Industrial production forecast (Consensus, yoy rate prevailing one year ahead)
FC Industrial production forecast current year (Consensus, yoy)
FN Industrial production forecast next year (Consensus, yoy)
P Labour productivity (yoy)

36
P: Past output growth; FC: Current year forecast; FN: Next year forecast.
39
ECB
June 2010
P Private consumption expenditure (qoq, latest available figure)
P Private consumption expenditure (qoq, latest available figure-1)
P Private consumption expenditure (qoq, average of 4 latest available figures)
P Private consumption expenditure (yoy, latest available figure)
P Private consumption expenditure (yoy, latest available figure-1)
P Private consumption expenditure (yoy, average of 4 latest available figures)
FC Private consumption forecast current year (Consensus)
FN Private consumption forecast next year (Consensus)
P Real GDP (qoq, latest available figure)
P Real GDP (qoq, latest available figure-1)
P Real GDP (qoq, average of 4 latest available figures)
P Real GDP (yoy, latest available figure)
P Real GDP (yoy, latest available figure-1)
P Real GDP (yoy, average of 4 latest available figures)
FC Real GDP growth forecast (Consensus, yoy rate prevailing one year ahead)
FC Real GDP growth forecast current year (AMECO)
FC Real GDP growth forecast current year (Consensus)
FC Real GDP growth forecast current year (ECB)
FC Real GDP growth forecast current year (IMF)
FC Real GDP growth forecast current year (OECD)
FC Real GDP growth forecast current year (SPF)
FN Real GDP growth forecast next year (AMECO)
FN Real GDP growth forecast next year (Consensus)
FN Real GDP growth forecast next year (ECB)
FN Real GDP growth forecast next year (IMF)
FN Real GDP growth forecast next year (OECD)
FN Real GDP growth forecast next year (SPF)
P Retail sales confidence (EC, change w.r.t. last GovC meeting)
P Retail sales confidence (EC, change w.r.t. 3 months before)
P Retail sales volume (yoy)
P Total employment growth (yoy)
P Unemployment rate (change w.r.t. last GovC meeting)
P Unemployment rate (change w.r.t. 3 months before)
FC Unemployment rate forecast cur. year (Consensus, change w.r.t. last GovC meeting)
FC Unemployment rate forecast cur. year (Consensus, change w.r.t. 3 months before)
FN Unemployment rate forecast next year (Consensus, change w.r.t. last GovC meeting)
FN Unemployment rate forecast next year (Consensus, change w.r.t. 3 months before)
40
ECB
June 2010
Table A1. 2 List of inflation indicators (in alphabetical order):
Factor
37
Indicator
P Break-even inflation rate (French indexed bonds)
P Compensation per employee (yoy)
FC Consumer price forecast (Consensus, yoy rate prevailing one year ahead)
FC Consumer price forecast current year (AMECO)
FC Consumer price forecast current year (Consensus)
FC Consumer price forecast current year (ECB)
FC Consumer price forecast current year (IMF)
FC Consumer price forecast current year (OECD)
FC Consumer price forecast current year (SPF)
FN Consumer price forecast next year (AMECO)
FN Consumer price forecast next year (Consensus)
FN Consumer price forecast next year (ECB)
FN Consumer price forecast next year (IMF)
FN Consumer price forecast next year (OECD)
FN Consumer price forecast next year (SPF)
P HICP (yoy) - consumer goods (durables)
P HICP (yoy) - consumer goods (non-durables)
P HICP (yoy) - consumer goods (total)
P HICP (yoy) - energy prices
P HICP (yoy) - headline
P HICP (yoy) - non-energy industrial goods
P HICP (yoy) - processed food
P HICP (yoy) - services
P HICP (yoy) - unprocessed food
P Industrial producer prices (yoy) - capital goods
P Industrial producer prices (yoy) - intermediate goods
P Industrial producer prices (yoy) - manufacturing goods
P Industrial producer prices (yoy) - total (excl. construction)
P Oil price (1-month ahead delivery in EUR, change w.r.t. last GovC meeting)
P Oil price (1-month ahead delivery in EUR, change w.r.t. 3 months before)
P Oil price (Brent in EUR, change w.r.t. 3 months before)
P Oil price (Brent in EUR, change w.r.t. last GovC meeting)
P Oil price (Brent in USD change w.r.t. 3 months before)
P Oil price (Brent in USD, change w.r.t. last GovC meeting)
P Unit labour costs (yoy)
FC Wage growth forecast current year (Consensus, yoy)
FN Wage growth forecast next year (Consensus, yoy)

37
P: Past inflation; FC: Current year forecast; FN: Next year forecast.
41
ECB
June 2010
Annex 2. Distribution of standard errors of the estimated coefficients
Figure A2.1 Standard errors of the smoothing coefficient
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
1 51 101 151 201 251 301 351 401
Number of rules
S
t
a
n
d
a
r
d

e
r
r
o
r
Figure A2.2 Standard errors of the constant
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1 51 101 151 201 251 301 351 401
Number of rules
S
t
a
n
d
a
r
d

e
r
r
o
r
Figure A2.3 Standard errors of the inflation coefficient
0.0
0.5
1.0
1.5
2.0
2.5
3.0
1 51 101 151 201 251 301 351 401
Number of rules
S
t
a
n
d
a
r
d

e
r
r
o
r
Worki ng PaPer S eri eS
no 1118 / november 2009
DiScretionary
FiScal PolicieS
over the cycle
neW eviDence
baSeD on the eScb
DiSaggregateD aPProach
by Luca Agnello
and Jacopo Cimadomo

Working Paper Series: Towards A Robust Monetary Policy Rule For The Euro Area

Uploaded by

Copyright:

Available Formats

Working Paper Series: Towards A Robust Monetary Policy Rule For The Euro Area

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Working Paper Series: Towards A Robust Monetary Policy Rule For The Euro Area

Uploaded by

Copyright:

Available Formats

Worki ng PaPer S eri eS

no 1210 / J une 2010

= 2.22) to inflation as other rules would predict. Purely backward-looking rules,

= 0.55). Four reasons are likely to drive the

You might also like