Journal of Banking & Finance 36 (2012) 2323–2343

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf

New measures of monetary policy surprises and jumps in interest rates q

Ángel León a, Szabolcs Sebestyén b,⇑
a
Dpt. Métodos Cuantitativos y Teoría Económica, Universidad de Alicante, San Vicente del Raspeig, 03080 Alicante, Spain
b
ISCTE-IUL Business School, Avenida das Forças Armadas, 1649-026 Lisbon, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: We propose new surprise measures to characterise two important dimensions of monetary policy. Our
Received 15 May 2011 measures outperform the traditional monetary shocks in explaining variation of interest rates in the
Accepted 18 April 2012 event-study framework. We also study the extent to which the ECB caused jumps in euro area interest
Available online 28 April 2012
rates. The new surprises still prevail upon the traditional ones. Jumps play a great role in the variation
of interest rates and the ECB induced several jumps with its decisions, but its predictability has improved
JEL classification: over time. We find that, although the surprise measures become somewhat distorted due to money mar-
C58
ket tensions during the financial turmoil, our model still provides an interesting insight into interest rate
E43
E52
behaviour throughout the crisis.
E58 Ó 2012 Elsevier B.V. All rights reserved.

Keywords:
Monetary policy surprises
Jumps
Interest rates
Principal components

1. Introduction
If William Shakespeare had lived over the last decade or so, he
could have hardly found more interesting characters for his plays
than central banks in that their words are at least as effective as
their deeds. Modern monetary policy is much more than just
changing a specified target rate. Instead, central banking has be-
come rather an art of shaping market expectations across the term
structure of interest rates, which henceforth affects the economic
agents’ investment decisions and the real economy. For an effective
implementation of monetary policy, central banks have given a
bigger weight on communication than ever before, and are using
it as a key tool to manage expectations of future interest rates
and inflation. As Woodford (2003) states: ‘‘Not only do expectations
about policy matter, but. . . very little else matters’’.
q
The paper has benefitted much from discussions with Gabriel Pérez-Quirós and
Roberto Rigobon. We are also very grateful to Francis X. Diebold, Antonio Diez de los
Rios, András Fülöp, Raquel Gaspar and Jun Pan for their valuable comments and
suggestions to a previous version of the paper, titled as ‘‘Jumps in interest rates: To
what extent do news surprises matter?’’, and we thank seminar participants at
workshops, conferences and seminars in Barcelona, Berlin, Münster, Coimbra,
Budapest, Madrid, Lisbon, Lancaster and Oslo for the useful discussions. We
especially thank an anonymous referee for his/her helpful comments. Evidently, all
remaining errors are ours.
⇑ Corresponding author. Tel.: +351 217 903931.
E-mail addresses: aleon@ua.es (Á. León), szabolcs.sebestyen@iscte.pt (S. Sebestyén).
A monetary policy decision consists of two parts: the first is the
decision itself whether or not to change the target rate, and if so, to
what extent. The second part is a statement which is released with
(or following) the decision, and explains the rationale underlying
the decision. Although the first component only affects the policy
rate and very short-term market rates, it is part of the whole deci-
sion whose aim is to steer market expectations on longer horizons,
and this objective is achieved through the statement. Hence, these
two parts should not be separated, since they exert influence
jointly on the markets. To illustrate, for example, take two 50 basis
point increases in the policy rate, one accompanied by a statement
of no changes in the near future, while the other accompanied by a
statement which considers the current move as a start of a tighten-
ing process. The decision component is the same in both cases, but
their message and impact on longer horizon interest rates are evi-
dently completely different.
Effective communication implies better predictability of mone-
tary decisions as the markets are able to price the impact of a deci-
sion even before it actually takes place. As a consequence, only
unexpected monetary decisions affect interest rates after the deci-
sion is released (see Kuttner, 2001). Measuring monetary surprises
and assessing their effects on interest rates provide important
information for policy-makers for several reasons. First, they help
the understanding of the functioning of financial markets in that
policy makers can gauge how investors interpret monetary deci-
sions and revise their expectations, as well as whether decisions
are consistent with market expectations. Second, by identifying

0378-4266/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jbankfin.2012.04.014
2324 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
the asset classes which are the most sensitive to monetary deci-
sions central banks can examine how effectively the monetary
transmission mechanism works.
Following Kuttner (2001) a broad literature has emerged to
quantify the impacts of monetary surprises on asset prices. The
unexpected component of monetary decisions is usually measured
as the daily difference of some very short term interest rate on
monetary meeting days.1 Alternatively, some survey-based mea-
sures have also been constructed. While these surprise measures
have widely been used in the literature, they have at least two draw-
backs. First, Rigobon and Sack (2004, 2008) emphasise the possible
endogeneity between the surprise and the asset price on which its
effect is to be estimated, and the omitted-variables problem, i.e. that
other variables other than the monetary surprise may affect asset
prices in the given time interval. On the other hand, survey-based
measures may suffer from measurement errors. While they propose
estimators which are void of these problems, these are still subject to
the second drawback that a surprise measure, based on either ap-
proaches, cannot reflect all dimensions of monetary decisions.
Gürkaynak et al. (2005) raise the following question: how many
dimensions are necessary to adequately characterise monetary pol-
icy decisions? They let the data determine the surprise through
principal component analysis and find that two factors are suffi-
cient. While Gürkaynak et al. (2005) greatly recognise that to de-
scribe monetary decisions more than one dimension is required,
the factors are constructed in a way that they separate the two com-
ponents of a monetary decision described above. Their first factor
only deals with the current target and the other corresponds to
longer horizon impacts. Whereas the latter is relevant, the former
does not provide much insight into the nature of a monetary deci-
sion. It is not surprising then that they find that the second factor
plays a great role in explaining the variation of longer term rates.
The objective of this paper is twofold. First, we propose new
monetary surprise measures which describe important dimensions
of monetary policy. We rely on the methodology of Gürkaynak
et al. (2005), but our factors supply two dimensions of monetary
decisions which are more useful in the understanding how central
banks shape market expectations across the yield curve than the
previously proposed factors in the literature. Particularly, the level
factor corresponds to decisions which shift the yield curve level,
while the slope factor reflects monetary decisions causing changes
in the slope of the term structure. Both are key variables, since the
level of the yield curve is consistent with the markets’ long-term
inflation expectations and the slope appears to be a good predictor
of the business cycle (see Estrella and Hardouvelis, 1991). More-
over, we discuss methodological problems in Gürkaynak et al.
(2005) and provide improvements.
We find, in the standard event-study framework, that the struc-
tural interpretation of our level and slope factors is adequate, they
considerably outperform the traditional surprises and explain
much more of the daily variation of interest rates on the European
Central Bank (ECB)’s meeting days. The estimation results provide
compelling evidence that the two factors represent important
dimensions of the ECB’s monetary policy.
The second objective is to evaluate the impacts of monetary
surprises on euro area interest rates and assess the ECB’s predict-
ability in a novel way. Central banks prefer to steer expectations
in a smooth way and thus avoid to induce abrupt movements in
interest rates. We build an econometric model which disentangles
diffusive and jump information flows and examine the extent to
which monetary decisions are related to jumps. Both the jump ar-
rival process and the jump distribution evolve over time as a func-
1
Or, if intraday data is available, the variation over a narrow window surrounding
the decision.
tion of monetary surprise variables. Our model produces time
series of jump probabilities, which allow us to analyse whether
the ECB’s monetary decisions were incorporated smoothly into
interest rates of different maturities or instead induced sharp
adjustments. Moreover, we can study how the ECB’s predictability
changed over time since the introduction of the euro in 1999, and
which decisions caused the biggest jumps in interest rates.
We find that our new monetary surprise measures still prevail
upon the traditional surprises in most cases. In addition, our results
show that the ECB caused jumps mostly in the first half of the sample
period, a learning period for both the ECB and the markets, and that
the jump frequency reduced considerably when the ECB switched to
hold meetings less frequently. We also find that the contribution of
jumps to total volatility is substantial, especially in the short term,
suggesting that jumps play a great role in the daily variation of euro
area interest rates. Moreover, monetary surprises associated with
jump innovations are incorporated more quickly into current rates
than those associated with normal innovations. Our model also
appears to be robust to additional variables other than monetary
surprises, particularly macroeconomic news surprises and mone-
tary surprises of other central banks, as the response coefficients
of monetary surprises are essentially unchanged, and the model-
implied jump characteristics change only slightly.
Although we estimate our model over the sample period which
ends on December 29, 2006, we also consider a longer sample
including the recent financial turmoil both to study how our sur-
prise measures are affected by the crisis and to check the perfor-
mance of our monetary-jump model. We find that the monetary
surprise measures involving Euribor rates are somewhat distorted
due to money market tensions in 2008–2009. The impacts of this
distortion are evident from our estimation results. Nonetheless,
the estimates are similar to those of the shorter sample.
On the other hand, our findings warn against a blind use of con-
ventional monetary policy surprises in times of serious money
market tensions. The recent crisis has required the implementation
of unconventional monetary measures and the relevance of target
rate changes has considerably reduced, whereas other – formerly
much less important or completely irrelevant – factors, such as
quantitative and credit easing, have been brought into limelight.
These factors, jointly with the increased and volatile spread be-
tween unsecured and secured rates, must explain to a great extent
the distortions of our surprise measures. This issue is beyond the
scope of our paper, although it is an important direction for our fu-
ture research. Moreover, our sample period mainly consists of
‘‘calm’’ times (so does the history of monetary policy) when mon-
etary policy is pursued in a conventional way, and our paper can
provide important lessons for central banks regarding predictabil-
ity and measurement of the effects of unexpected decisions.
The paper is structured as follows. Section 2 discusses the sta-
tistical properties of the data used. In Section 3 we briefly overview
the literature on monetary surprises, we analyse the possible cave-
ats, and also introduce the principal component methodology. The
section then provides the details of our new monetary surprise
measures and the interpretation and descriptive analysis of the
new factors. Section 4 deals with the econometric framework. It
describes first the event-study setting and presents the estimation
results. The second part of the section introduces our monetary-
jump model in detail and reports some preliminary results. Section
5 presents and discusses the results of the monetary-jump model.
Finally, Section 6 concludes.
2. Descriptive statistics of the interest rate data
The interest rate data analysed here consist of daily observa-
tions in euro area interest rates. Since the euro area is a currency
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
Fig. 1. Daily variation in euro area interest rates. Evolution of daily changes in euro area interest rates in percentage terms over the period February 19, 1999 through June 30,
2010. The vertical grid lines show the end of the sample period used for estimation purposes, December 29, 2006.
area without a single common bond market, it is not obvious
which instruments to choose to represent euro area interest rates.
Hence, consistent with the ECB’s earlier practice,2 our analysis is
based on money market rates for maturities of up to 1 year and on
the fixed side of interest rate swap contracts for maturities of one
to ten years. For a detailed description of the interest rate data used
in this paper, see Section A in the on-line appendix of the paper,
available on the journal’s website.
For our analysis four maturities are chosen: 1 month, 6 months,
2 years and 10 years. This choice aims both to represent different
segments of the term structure and to reflect market participants’
preferences. Our sample covers the period from February 18,
1999 (the first day for which we obtained swap yield data) through
June 30, 2010. However, due to the financial crisis, which started in
2007, we carry out our main analysis for the sample up to December
29, 2006, and use the remaining period to study the behaviour of
our monetary surprise measures during financial distress and
how their impacts on interest rates have changed. Weekends and
holidays are excluded from the data set, providing 2002 useful daily
observations (with the crisis period included, 2889 observations).
For the sample period, the average term structure exhibits an
upward-sloping pattern. For the level of all interest rates, denoted
by rt, the augmented Dickey–Fuller (ADF) test shows that the null
hypothesis of a unit root cannot be rejected at any significance le-
vel. This test, however, rejects the presence of a unit root for all
first-difference series, denoted by Drt. Hence, it is a good approxi-
2
Since 10 July 2007, the ECB releases yield curve estimates calculated from euro
area government bonds on a daily basis.
2325
mation to model the daily changes in interest rates as they are
stationary.
The statistical properties of Drt show that the null hypothesis of
normality can clearly be rejected as the sample skewness and kur-
tosis values are far away from those of the Gaussian distribution.
The kurtosis is considerably higher for short rates. Fig. 1 may pro-
vide a possible explanation for the very high values of kurtosis of
money market rates. The series of 1-month and 6-month interest
rate changes are relatively smooth with extremely sharp jumps
(mainly the 1-month rate), while the 2-year and 10-year swap
yields seem inherently more volatile with jumps of lower size.
Note that the very large jumps in money market rates mainly oc-
curred in the first half of the sample period. Finally, the Ljung-
Box statistics for the squares of Drt show strong autocorrelation
for all series. This suggests modelling the conditional variance,
for instance, via a GARCH-type model.
Fig. 1 provides clear evidence that the financial crisis changed
somewhat the time-series characteristics of the series by exhibit-
ing extreme observations and higher volatility. These patterns
are less pronounced for bond yields, but the impacts of the turmoil
on them are also striking.
3. How to measure monetary policy surprises?
3.1. Survey and market surprises
It is reasonable to suggest that rather than an announcement it-
self, it is the surprise contained in an announcement that moves
interest rates. To define the surprise component for monetary pol-
2326 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
icy decisions, two main approaches have arisen in the literature.
The first studies whether the released decision agrees with the
market participants’ a priori expectations. The difference between
the actual change and the market expectation (the average or the
median of all analysts’ opinions) provides a survey-based measure
of the monetary decision. This approach was proposed by Balduzzi
et al. (2001) for macroeconomic variables, and applied by Anders-
son et al. (2009) to study the ECB’s surprises in an intraday frame-
work, among others.
The second approach is more popular among researchers and
relies on the assumption that, on a monetary meeting day, the
main driver of movements in very short rates is the monetary deci-
sion itself. Thus a natural measure for the unexpected component
of central banks’ decisions is the daily change in a very short-term
interest rate on meeting days. With the availability of higher fre-
quency data, this surprise can be calculated in a more sophisticated
way by considering a narrow window surrounding the monetary
decision. Some examples are Kuttner (2001), Bernanke and Kuttner
(2005) and Chuliá et al. (2010), among others, which focus exclu-
sively on US financial markets, and the monetary surprise is con-
structed either from the fed funds futures rate or from the
Eurodollar rate. Regarding the euro area, Pérez-Quirós and Sicilia
(2002) propose to use Eonia swap rates3 and find that the 2-week
rate predicts well the monetary policy decisions before November
2001, while afterwards the 1-month rate is preferable.4 Bernoth
and von Hagen (2004) use the closest-to-delivery 3-month Euribor
futures rates as a surprise measure.
3.2. Caveats and solutions
Rigobon and Sack (2004) study the shortcomings of these ap-
proaches and emphasise the possible endogeneity between asset
prices and short rates, as well as the possible problem of omitted
variables. They propose an estimator that identifies the response
coefficient through the heteroskedasticity of monetary shocks.
Rigobon and Sack (2008) examine the measurement problems with
survey surprises and propose two econometric approaches to ac-
count for these measurement noises.
An alternative way to deal with the endogeneity and measure-
ment error problems is to let the data determine the surprise. This
can be achieved by the use of principal component analysis (PCA).
This approach was independently proposed by Rigobon and Sack
(2008) (to capture macroeconomic surprises) and Gürkaynak
et al. (2005) (to characterise the dimensions of monetary policy re-
leases),5 although the idea was probably first applied in this context
by Pérez-Quirós and Sicilia (2002). The main advantage of PCA is that
it is void of the problem that the surprise variable does not capture
all the unexpected component of a release. On the other hand, it may
capture some variation which is not associated with the monetary
decision, a possibly relevant problem with lower frequency data.
3.3. The principal component analysis approach
In this paper we choose the PCA approach. Although we are
aware of the fact that the obtained components may accidentally
explain some variation from other sources, we are confident that
this is less severe than the endogeneity and measurement error
problems. Moreover, the possible impacts of omitted variables will
be examined later as a robustness test. Finally, our results show
3
For a description of Eonia and Eonia swaps, see Section A in the on-line appendix.
4
Before November 8, 2001 the ECB held bimonthly meetings, but afterwards it
switched to monthly discussions.
5 7
For an application of the methodology of Gürkaynak et al. (2005) for euro area
data, see Brand et al. (2010).
that the principal components do a better job than the traditional
surprise variables.
Similarly to Gürkaynak et al. (2005), our question is how many
dimensions are necessary to adequately characterise monetary
policy decisions. They write down a latent factor model, and then
apply the matrix rank test of Cragg and Donald (1997) to deter-
mine the optimal number of factors. Given the inference results,
the factors are computed by PCA, and are finally rotated in order
to obtain a structural interpretation. However, this approach suf-
fers from both conceptual and statistical problems. First, factor
analysis (FA) and PCA are conceptually quite distinct techniques.6
Although both aim to reduce the dimensionality of a dataset by
focusing on the covariance matrix of the data, PCA concentrates on
the diagonal elements (variation), whereas FA’s interest is in the
off-diagonal elements (correlation). Moreover, by increasing the
dimensionality of the model the original PCs remain unchanged,
but the new factors can be very different from the old ones. Finally,
while PCs are exact linear combinations of the data, factors are not
due to an error term. The Cragg and Donald (1997) test can be ap-
plied in the FA framework, but not in the PCA approach. To deter-
mine the ‘‘optimal’’ number of PCs several rules have been
proposed in the literature (see Jolliffe (2002, Chapter 6)), of which
most are ad hoc rules, whereas formal rules either require unrealistic
distributional assumptions or are computationally intensive. There-
fore, using inference applicable to FA for PCA is pointless and may
lead to an inconsistent choice regarding the number of PCs as the
‘‘optimal’’ number of factors can only be arbitrarily correct.
Two other problems in Gürkaynak et al. (2005) arise with rota-
tion. First, when rotating PCs, the total variation explained by the
first, say, m PCs remains unchanged; however, the distribution of
explanatory power among the rotated components does change
and becomes more even than before rotation. Hence, with rotation
we lose information regarding the nature of any really dominant
components. Second, under the usual PCA normalisation scheme,
the loading vectors are orthogonal and the obtained PC scores
are uncorrelated. Nonetheless, after rotation, the loadings remain
orthogonal, but the component scores are no longer uncorrelated,
see Jolliffe (1995). While Gürkaynak et al. (2005) impose a restric-
tion to ensure the orthogonality of the new rotated factors, it fails
to do so, resulting in correlated new PCs which distorts regression
results. In Section B in the on-line appendix we show this formally
and we also quantify the possible correlation between rotated
factors.
3.3.1. Data for PCA and analysis of PC factors
Before carrying out PCA, an important issue is to select the vari-
ables for the data matrix, i.e. the ones that are likely to provide the
‘‘purest’’ monetary policy surprise. In this paper, in line with previ-
ous studies cited above, we use Eonia swap and 3-month Euribor
futures rates.7 The market rates we choose to extract the ECB’s mon-
etary surprises from are the 2-week and 1-month Eonia swap rates
and the first five deliveries of 3-month Euribor futures rates. The first
delivery contains the contracts that expire first in the quarterly se-
quence (closest-to-delivery contracts with average horizon of
1.5 months), the second contains contracts that have settlements
in the second delivery quarter from now on (with average horizon
of 4.5 months), and so on. This way we obtain seven time series of
market expectations with horizons of over 1 year. The rows of the in-
put data matrix for PCA then contain daily changes in Eonia swap
and Euribor futures rates on ECB meeting days (normalised to have
unit variance), and the columns correspond to the different maturi-
ties (totally seven columns).
6
For a thorough comparison of the two procedures, see Jolliffe (2002, Chapter 7).
See Section A in the on-line appendix for a description of these contracts and for
some details regarding their markets.
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2327

Table 1 merely change the target rate. This is achieved through central
Results of principal component analysis. bank statements, which may give hints to market participants at
Shorter sample Full sample the future stance of monetary policy. Hence, the target factor per
S1 S2 S1 S2 se does not reflect an important dimension of monetary decisions,
since it only captures surprises over a very short horizon. It is not
Panel A: on monetary meeting days
Eonia swaps
surprising then that the authors find that the path factor has a
2-Week 0.2391 0.6380 0.1918 0.6776 much greater impact on long-term yields.
1-Month 0.2784 0.6002 0.2301 0.6488 Our surprise variables focus on how central banks can shape
Euribor futures expectations across the term structure. The first PC (level factor) ex-
1st Delivery 0.4077 0.1086 0.4298 0.0002 plains variations in the level of the yield curve on monetary meet-
2nd Delivery 0.4320 0.1020 0.4496 0.1221 ing days, and indicates a roughly parallel shift of the term structure
3rd Delivery 0.4299 0.1958 0.4467 0.1975
when an unexpected decision occurs. Unlike the Fed, the ECB not
4th Delivery 0.4161 0.2644 0.4343 0.2429
5th Delivery 0.3931 0.3198 0.3678 0.0835 only announces its decision, but also explains it at a press confer-
ence on the day of the meeting. Whereas on non-meeting days the
Explained variation (%) 72.12 21.05 64.99 23.51
Cumulative explained 72.12 93.17 64.99 88.50 loadings on very short rates are much smaller than on Euribor fu-
variation (%) tures (see Panel B of Table 1), on meeting days this difference is
substantially smaller (see Panel A), leading to more even weights.
Panel B: on non-meeting days
In addition, about 72% of the variation of short rates on monetary
Eonia swaps
2-Week 0.1374 0.6988 0.1643 0.6900
meeting days is explained by the level factor, while on non-meet-
1-Month 0.1929 0.6616 0.2123 0.6593 ing days only 58%. These findings suggest that the level factor can
Euribor futures
be related to and represents an important dimension of monetary
1st Delivery 0.3535 0.0269 0.3268 0.0218 policy.
2nd Delivery 0.4634 0.0700 0.4880 0.1012 Table 2 describes important events on the meeting days with
3rd Delivery 0.4679 0.1295 0.4919 0.1537 the largest surprises across the two dimensions we consider. A
4th Delivery 0.4628 0.1508 0.4778 0.1721
large value in the level factor may be realised in different situa-
5th Delivery 0.4138 0.1699 0.3357 0.1587
tions. It may reflect surprise decisions accompanied by clear state-
Explained variation (%) 57.97 21.63 51.31 22.44
ments that steer market expectations and shift the yield curve (e.g.
Cumulative explained 57.97 79.60 51.31 73.74
variation (%) April 8, 1999 and June 8, 2000). On the other hand, it may corre-
spond to a timing surprise, when the expected change in the policy
The table presents the loadings obtained by principal component analysis, applied
rate does not occur (or of a lesser amount), thus all rates adjust,
on Eonia swap and Euribor futures rates. The analysis is carried out both on ECB
Governing Council meeting days and on other days, and changes in interest rates on resulting in a shift in the term structure (see April 11, 2001 and
all days are standardised to have unit variance. Shorter sample refers to the period May 10, 2001). Finally, a large level factor can also be observed
February 19, 1999 through December 29, 2006, while the full sample ends on June when the target rate remains unchanged, but a clear statement is
30, 2010. released regarding policy moves in the future (see July 15, 1999).
The second PC (slope factor) conveys important information
regarding movements in the slope of the money market yield
The obtained loadings and explained variations for the first two curve, another key indicator for central banks. The loadings of this
PCs are given in Panel A of Table 1.8 The table contains the derived factor in Table 1 are somewhat different on meeting and non-
loadings both for the sample period analysed here and for the full meeting days. Whereas Eonia swap rates have similar large
sample including the crisis period. Further PCs explain little varia- weights in both cases, the loadings on Euribor futures rates are
tion in interest rate changes, it is difficult to find a structural inter- considerably larger (in absolute value) on ECB meeting days. This
pretation for them (except for the third PC which can be seen as result emphasises the relevance of Euribor futures in measuring
curvature), and an econometric analysis have found them insignifi- monetary policy surprises, and the relatively large weights on
cant in explaining interest rate movements. longer maturity futures suggest that on monetary meeting days
The pattern in the loadings is very similar to what is commonly the slope of the money market term structure can change not only
observed in the yield curve literature, that is, the first PC can be due to changes in very short rates, but also because of movements
interpreted as level and the second as slope. Of course, in our case in longer term rates. This also supports the view that monetary
only rates up to at most 15 months are considered and only on policy can effectively shape expectations across the yield curve.
monetary meeting days, thus the interpretation of PCs changes The slope factor explains around 21% of the variation of money
slightly. A change in the first PC induces a roughly parallel shift market rates both on monetary meeting and on non-meeting days,
in the money market yield curve. A movement in the second PC re- indicating that this factor has a qualitative rather than a quantita-
sults in a change in slope of the money market yield curve with tive meaning. After all, the striking large difference between the
large weights on the very short rates, suggesting that the slope cumulative explained variations on meeting (over 90%) and on
changes mostly due to movements in very short rates, which are non-meeting days (below 80%) reflect the relevance of the chosen
directly affected by monetary decisions. rates in conveying market expectations of policy changes.
Gürkaynak et al. (2005) define two factors through rotation: Large slope factor realisations occur, for instance, when market
unexpected changes in the current fed funds futures rate are driven participants expect that short-term interest rates are likely to rise
exclusively by the target factor, while the path factor captures all more in the future than in the course of the next few months (see
other aspects of the Fed’s decisions that move rates in the future July 15, 1999 in Table 2). Alternatively, it may also indicate a tim-
and that are unrelated with the change of the current fed funds ing surprise when an expected key rate change does not occur (or
rate. However, central banks, instead of controlling only very short of a lesser amount), but is very likely to occur in the next meeting.
rates, aim to shape market expectations across the yield curve. This In such a case short rates adjust, but longer term yields remain vir-
means that each monetary decision has a longer-term focus than to tually unchanged (see November 7, 2002 in Table 2).
Although a great part of the variation on monetary meeting
days comes from the level factor, it is striking in Table 2 that all
8
Note that the PCs are also normalised to have unit variance. but one (April 11, 2001) large level factor realisations are also asso-
2328 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

Table 2
The largest level and slope factors and monetary events.

Date Level Slope Observations

April 8, 1999 3.60 4.28 Unexpected 50 bp cut and a clear statement that no further changes will occur; the yield curve regained a positive
slope
July 15, 1999 2.02 3.26 Market participants expected a rise in 2000 rather than in the course of the next few months, thus rates with more
than 3 months of maturity rose and the yield curve steepened
October 7, 1999 1.22 3.27 The expected increase did not occur (it took place at the next meeting), the whole yield curve adjusted, but especially
short rates
December 15, 1999 0.00 2.40 Large upward movement in the 2-week rate due to millenium problem
June 8, 2000 4.10 1.70 Bigger increase than expected (50 bp) that curbed expectations of future increases in the near term
April 11, 2001 2.79 0.07 Expected cut that only occurred at the next meeting, and tight liquidity conditions caused by underbidding in the
main refinancing operations
May 10, 2001 3.39 2.61 Smaller cut than expected (25 bp), short rates decreased more than long rates
November 7, 2002 0.25 2.30 Expected cut that only occurred at the next meeting, so only short rates adjusted
March 6, 2003 0.32 2.11 25 bp expected cut, but very short rates increased due to underbidding in the main refinancing operations, while long
rates remained broadly unchanged
February 7, 2008 2.09 1.23 Large drop in Euribor futures rates due to uncertain money market condition; this period was characterised by large
volatility on unsecured markets
June 5, 2008 3.76 1.91 Sharp increase in Euribor futures rates, reflecting (i) expectations of future money market tensions, and (ii) increase in
market expectations regarding future ECB interest rates
July 3, 2008 1.69 2.24 25 bp hike after which Euribor futures rates dropped, indicating an anticipation of some narrowing of the spread
between unsecured and secured rates
October 2, 2008 2.03 1.18 Downward adjustment, especially in Euribor futures rates, after sharp increases due to Lehman bankruptcy, AIG bail-
out and other crisis events
November 6, 2008 2.02 1.46 50 bp cut, expanded collateral framework at the ECB, and government supports to insolvent banks
December 4, 2008 0.21 4.71 75 bp cut, resulted in large drops in very short rates, while rates at longer horizons did not change considerably
January 15, 2009 0.19 2.56 50 bp cut, resulted in large drops in very short rates, while rates at longer horizons did not change considerably
April 2, 2009 1.90 2.12 25 bp cut, resulted in drops in very short rates and in increases in Euribor futures rates, steepening the yield curve
June 4, 2009 2.22 0.71 No key rate change, but uncertainty regarding liquidity conditions, resulted in hikes in rates

The table presents the largest level and slope monetary policy surprises of the ECB. Only days on which at least one of the surprises exceeded 2 are reported. The level and
slope surprises are obtained by principal component analysis, applied on Eonia swap and Euribor futures rates. The table separates surprises over the shorter sample period
from February 19, 1999 through December 29, 2006, and those over the full sample which ends on June 30, 2010.

ciated with large slope factor observations. A graphical analysis
supports this view (see Fig. 1 in the on-line appendix) for our sam-
ple period, meaning that, though the ECB’s unexpected decisions
resulted primarily in shift of the yield curve level, big surprises also
changed the slope of the term structure.
If we consider the full sample period, the loadings on Euribor
futures rates for the level factor become larger, while those on Eo-
nia swaps decrease. This reflects the fact that during the turmoil
the spread between unsecured (such as Euribor) and secured (such
as Eonia swaps) rates was high and volatile, and shifts in the yield
curve level on monetary meeting days were primarily driven by
changes in Euribor futures rates. The big monetary surprises, oc-
curred during the crisis and shown in the lower part of Table 2,
support this view in that the biggest level changes were caused
by sharp movements in the Euribor futures rates. Moreover, the
loadings of the level factor on non-meeting days also exhibit the
same pattern, suggesting that the high volatility of Euribor futures
does not only characterise monetary meeting days, but was a gen-
eral phenomenon in that period. The explained variation by the le-
vel factor is considerably smaller in the full sample, which
indicates that the rates that earlier conveyed relatively precise
information regarding policy expectations were driven by factors
other than monetary policy during the financial turmoil.
Turning to the slope factor, the difference between loadings in
the short and full samples is more pronounced. Those on very short
rates increase substantially, which is likely to be the consequence of
the ECB’s heavy cuts, inducing large movements mostly in very short
rates. On the other hand, the loading on the nearest-to-delivery
Euribor futures rate is essentially zero, while that on the fifth deliv-
ery is much smaller (in absolute value) than in the shorter sample.9
9
Deriving the PCs without the fifth delivery leads to higher explained variation for
the slope factor, but the loading on the nearest-to-delivery Euribor futures rates
remains very small (0.03). The level factor loadings do not change significantly,
although short rates gain somewhat larger loadings.
Similarly to our previous conclusions, large level surprises are
likely to be accompanied by large slope surprises (see Fig. 1 in
the on-line appendix). All these findings suggest that the ECB’s
monetary policy surprises are well characterised by the level and
slope factors before the financial turmoil, while the crisis period
causes some distortion in the surprise measures.
3.3.2. Some descriptive results
The PCA methodology ensures that the two factors are orthog-
onal, facilitating the econometric analysis. Both factors are corre-
lated with the monetary policy surprise measures widely used in
the literature. The correlations of the level and slope factors with
the survey measure, calculated from Reuters polls and used by
Andersson et al. (2009), are 0.52 and 0.52, respectively. The cor-
relations with the most widely used market measures, computed
as daily changes in the 1-month Eonia swap rate and in the clos-
est-to-delivery 3-month Euribor futures rates, are 0.63 and
0.73, and 0.92 and 0.13, respectively. Hence, the traditionally
used surprise measures reflect both important dimensions of mon-
etary policy, but we can only estimate their joint effects.
As regards the relation between our two new factors and inter-
est rate changes, Fig. 2 shows scatter plots of the level and slope
factors against daily differences of the four interest rates we con-
sider here. The level factor exhibits a strong positive correlation
with all rates, especially with the 6-month rate, with a correlation
coefficient of 0.9 (see Panel B). The slope factor correlates nega-
tively with money market rates, while has a positive relation with
swap yields. The strongest correlation coefficients are observed at
the two extremes of the yield curve, as expected, while on the 6-
month horizon the correlation is quite weak. Considering the full
sample, the correlations are substantially weaker, except the 2-
year rate, which is more strongly correlated with both the level
and slope factors over the crisis period. This finding will be studied
later in more detail.
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2329

Fig. 2. The level and slope factors against interest rate changes on monetary meeting days. Scatter plots of level and slope factors against daily changes in the 1-month, 6-
month, 2-year and 10-year rates on ECB Governing Council meeting days. Filled circles correspond to observations of the pre-crisis period February 19, 1999 through
December 29, 2006, while empty circles correspond to observations of the crisis period January 1, 2007 through June 30, 2010. qpc denotes linear correlation in the pre-crisis
period, while ql stands for linear correlation in the crisis period.
2330 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
The summary statistics for the two factors are not reported here
to save space, but are presented in Table 1 in the on-line appendix.
The following conclusions can be drawn. The average surprise is
generally small and statistically indistinguishable from zero. Aver-
age absolute surprises are over half a standard deviation for each
factor, indicating that the ECB’s unexpected decisions and commu-
nication induced substantial changes in the level and slope of the
euro area money market yield curve. This is further supported by
the big ranges of the variables which reach six standard deviations.
Finally, there is no pronounced pattern in terms of asymmetry for
the level factor, while for the slope factor more negative values can
be observed, pointing to more flattening than steepening surprises.
It is striking in Table 2, considering only the upper half which
refers to the shorter sample, that almost all big surprises occurred
in the first half of the sample, in particular before November 8,
2001, when the ECB switched to monthly discussions from fort-
nightly meetings (see footnote 4). For this reason Table 1 in the
on-line appendix also shows the descriptive statistics for our sur-
prise measures for three subsamples: before November 8, 2001
(learning period, since both the ECB was a new monetary institu-
tion and the analysed markets were recently developed in this per-
iod), between November 8, 2001 and December 29, 2006 (quiet
period with less frequent meetings and few key rate changes),
and after December 29, 2006 (crisis period).
Both the mean absolute surprise and the range of surprises are
the largest in the learning period, indicating that, on average, the
ECB surprised the markets with its decisions more over its first
years of functioning than even during the crisis. However, the sur-
prises in the learning and in the crisis period are of different nat-
ure. The former reflect some lack of predictability of the ECB’s
decisions (especially their timing) due to the too frequent meet-
ings10, whereas the latter realised during a period of financial dis-
tress when monetary policy has not worked in the conventional
way, thus surprises also reflect non-monetary and unconventional
events.
4. Econometric analysis
In this paper we apply two approaches to study the extent to
which the level and slope factors affect euro area interest rates.
The first is the so-called ‘‘event-study’’ approach,11 which consists
of regressing returns (or in case of interest rates, changes) on the sur-
prise component of monetary policy actions on monetary meeting
days. The second is the time-series approach in which, instead of
considering only monetary meeting days, we build an econometric
model for the time series of asset returns, and control for the
time-series characteristics of the data such as conditional heteroske-
dasticity. An advantage of this approach is that, in addition to exam-
ine the impacts of monetary shocks on the conditional mean of
returns, their effects on conditional volatility can also be studied.12
To assess the performance of our new surprise variables, we
compare the results with those obtained by using the traditional
monetary surprise measures considered in the literature. The first
is a survey-based measure, which is calculated on meeting days as
the difference between the ECB’s actual decision and the mean of
analysts’ expectations, collected 1 week before the Governing
10
The reason is that with fortnightly meetings the timings of the ECB’s decisions
were hard to anticipate, leading to bigger surprises. The issue of improved
predictability of the ECB’s decisions after November 8, 2001 is studied by ECB
(2002). Also note that the better anticipation was also thanks to that, during the
period June 5, 2003 through December 1, 2005, the ECB did not change its policy rate,
and all these no-change decisions were perfectly anticipated by analysts according to
Reuters’ polls.
11
See Kuttner (2001) and Bernanke and Kuttner (2005), among others.
12
See, Bomfim (2003) and Ehrmann and Fratzscher (2003) as examples for this
approach, among others.
Council meetings by Reuters, divided by the standard deviation
of the forecast error. We also examine the two market-based mea-
sures cited above, in particular, daily differences in 1-month Eonia
swap and 3-month closest-to-delivery Euribor futures rates. These
surprise components are also standardised so that we can directly
compare their impacts.
4.1. The event-study approach
The basic equation of the event-study regression is
Drt ¼ a þ bSt þ et ; ð1Þ
where St stands for the different monetary surprise measures, and t
runs across Governing Council meeting days. The main focus is on
the response coefficient b and on the explanatory power of mone-
tary surprises on interest rate changes.
The estimated coefficients and R2s of Eq. (1) are presented in Pa-
nel A of Table 3. There are several important findings to report.
First, in line with our a priori expectations, considering longer
maturity surprise measures results in effects on longer horizons.
The survey measure, which corresponds to the current decision,
and the Eonia measure, which reflects changes in investors’ expec-
tations on the 1-month horizon, affect significantly euro area inter-
est rates up to 2 years of maturity. Both their impact and
explanatory power decrease with maturity and on the 10-year
horizon they become insignificant. However, on the very short
term these two surprise measures prevail over all individual vari-
ables. Second, the Euribor surprise remains significant across the
whole term structure, and exerts the strongest influence on the
6-month rate, the maturity closest to its horizon.
Regarding our new surprise measures, they convincingly out-
perform all other traditional monetary surprise variables, support-
ing our hypothesis that these two dimensions of monetary policy
decisions are very important. Although individually the level factor
is dominated by the traditional surprises on the 1-month horizon,
and only the Euribor-based measure outperforms it slightly for the
6-month rate, it has by far the largest explanatory power for bond
yields. It is striking that it explains 58% and 24% of the daily vari-
ations of the 2-year and 10-year rate, respectively. The slope factor
also seems very relevant, especially at the very short and at the
long end of the yield curve. On the 6-month horizon it is individu-
ally insignificant, but it is likely due to the omission-of-relevant-
variables problem, which is supported by its high significance
when it is jointly estimated with the level factor (see column
(6)). The magnitude and sign of the response coefficients of the
two factors are broadly consistent with our expectations and with
the visual inspections of Fig. 2, indicating that the structural inter-
pretation of the factors as level and slope is adequate.
When estimating the regression jointly with the level and slope
factors, the explanatory power increases substantially and the esti-
mates become more precise. The two factors jointly outperform all
traditional surprise measures at any horizon, even on the very
short term. Moreover, they explain 85%, 84%, 67% and 36% of the
daily variations of the four interest rates, as the maturity increases.
For the full sample we estimate a slightly different model to
examine how the effects of monetary surprises have been changed
by the financial distress. The estimated equation is given by
Drt ¼ a þ ac Dt þ bSt þ bc Dt St þ et ; ð2Þ
where Dt denotes a dummy variable that takes on the value one
after January 1, 2007, and zero otherwise. We allow this dummy
to affect both the intercept and the slope of the regression. Hence,
the impact of the monetary surprise variable prior to the crisis is
captured by b, whereas for the full sample its coefficient is b + bc.
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2331

Table 3
Main results of event-study regressions.

1-Month 6-Month
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Panel A
Survey 0.0375 – – – – – 0.0278 – – – – –
(7.73) (5.36)
Eonia – 0.0445 – – – – – 0.0301 – – – –
(12.08) (5.61)
Euribor – – 0.0336 – – – – – 0.0391 – – –
(5.41) (14.18)
Level – – – 0.0318 – 0.0318 – – – 0.0381 – 0.0381
(5.00) (14.64) (11.80) (15.47)
Slope – – – – 0.0313 0.0313 – – – – 0.0086 0.0087
(4.25) (10.10) (1.12) (4.39)
R2 0.5999 0.8340 0.4818 0.4306 0.4167 0.8481 0.4267 0.4930 0.8437 0.8028 0.0409 0.8441
2-Year 10-Year
Survey 0.0145 – – – – – 0.0031 – – – – –
(2.67) (0.71)
Eonia – 0.0135 – – – – – 0.0023 – – – –
(2.02) (0.47)
Euribor – – 0.0346 – – – – – 0.0187 – – –
(8.45) (4.71)
Level – – – 0.0398 – 0.0398 – – – 0.0230 – 0.0229
(10.71) (15.29) (5.09) (6.10)
Slope – – – – 0.0155 0.0155 – – – – 0.0159 0.0159
(2.07) (6.72) (2.52) (4.37)
2
R 0.0774 0.0659 0.4400 0.5837 0.0883 0.6715 0.0043 0.0024 0.1602 0.2415 0.1160 0.3571

Panel B
1-Month 6-Month
Survey 0.0429 – – – – – 0.0318 – – – – –
(7.70) (5.34)
Dt  Survey 0.0400 – – – – – 0.0305 – – – – –
(7.10) (5.10)
Eonia – 0.0422 – – – – – 0.0285 – – – –
(12.03) (5.59)
Dt  Eonia – 0.0091 – – – – – 0.0058 – – – –
(1.21) (0.81)
Euribor – – 0.0345 – – – – – 0.0401 – – –
(5.39) (14.13)
Dt  Euribor – – 0.0326 – – – – – 0.0208 – – –
(3.82) (2.09)
Level – – – 0.0324 – 0.0260 – – – 0.0393 – 0.0376
(4.91) (10.11) (11.72) (14.64)
Dt  Level – – – 0.0235 – 0.0050 – – – 0.0137 – 0.0047
(2.71) (0.83) (1.63) (0.50)
Slope – – – – 0.0390 0.0338 – – – – 0.0163 0.0087
(6.06) (10.02) (2.08) (3.93)
Dt  Slope – – – – 0.0223 0.0084 – – – – 0.0141 0.0072
(2.78) (1.51) (1.15) (1.61)
R2 0.5224 0.8092 0.4199 0.3803 0.5483 0.8223 0.3376 0.4381 0.7252 0.7400 0.1060 0.8064
2-Year 10-Year
Survey 0.0166 – – – – – 0.0035 – – – – –
(2.65) (0.70)
Dt  Survey 0.0079 – – – – – 0.0032 – – – – –
(0.71) (0.32)
Eonia – 0.0128 – – – – – 0.0022 – – – –
(2.01) (0.47)
Dt  Eonia – 0.0283 – – – – – 0.0155 – – – –
(1.61) (0.67)
Euribor – – 0.0355 – – – – – 0.0192 – – –
(8.42) (4.69)
Dt  Euribor – – 0.0359 – – – – – 0.0098 – – –
(4.37) (0.91)
Level – – – 0.0411 – 0.0442 – – – 0.0237 – 0.0268
(10.86) (15.29) (5.11) (6.86)
Dt  Level – – – 0.0358 – 0.0243 – – – 0.0067 – 0.0019
(5.88) (5.07) (0.68) (0.17)
Slope – – – – 0.0074 0.0163 – – – – 0.0110 0.0164
(0.90) (6.69) (1.61) (4.13)

(continued on next page)

2332 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

Table 3 (continued)

1-Month 6-Month
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Dt  Slope – – – – 0.0387 0.0013 – – – – 0.0110 0.0048
(2.30) (0.30) (0.79) (0.30)
R2 0.0539 0.0501 0.5862 0.7169 0.1869 0.7787 0.0094 0.0122 0.1828 0.2413 0.0867 0.3167

The table presents the main results of the event-study regressions for individual euro area interest rates. The model for the results in Panel A is Drt = a + bSt + et, while the
model in Panel B is Drt = a + alDt + bSt + blDtSt + et, where St denotes the different monetary surprise measures and Dt denotes a dummy variable that takes on the value one
after January 1, 2007, and zero otherwise. The monetary surprises considered here are as follows. First, the survey measure is defined as the difference between the ECB’s
actual decision and the mean of analysts’ expectations, collected 1 week before the Governing Council meetings by Reuters, divided by the standard deviation of these
differences. Second, the daily difference on monetary meeting days of 1-month Eonia swap rates (Eonia) and 3-month Euribor futures rates (Euribor). Third, the level and
slope factors, derived by principal component analysis where the data matrix contains 2-week and 1-month Eonia swap rates and the first five deliveries of 3-month Euribor
futures rates. The table contains the estimated b coefficients and R2s which are based on least squares regressions of observations only on ECB Governing Council meeting
days: 127 in the period February 19, 1999 through December 29, 2006 and 169 in the period February 19, 1999 through June 10, 2010. Heteroskedasticity-consistent t-values
are in parentheses.

The main results of this regression are presented in Panel B of Table
3.13
The most important finding to report is that for money market
rates b^c has the opposite sign to b,
^ and it is significant considering
the Euribor, level and slope surprises for the 1-month rate and con-
sidering the Euribor surprise in the equation of the 6-month rate.
Hence, the effect of monetary surprises over the full sample is
weaker than over the shorter sample, and the R2s of the regressions
are also lower.14 It is tempting to interpret these results as an empir-
ical evidence that the effectiveness of the interest rate channel of the
monetary transmission mechanism has reduced due to the financial
turmoil. This is partly true because, although market interest rates
followed the ECB’s cuts and this reduction also passed through to
bank lending rates, it occurred with a significant lag (see ECB,
2010). However, the lower explanatory power can rather be ex-
plained by the fact that the spread between unsecured and secured
rates in euro area money markets increased substantially and be-
came volatile from the onset of the crisis, which reflects higher credit
and liquidity risks in the unsecured Euribor market rather than
uncertainty regarding the ECB’s monetary decisions. This is sup-
ported by our estimation results, namely that b ^c is significant consid-
ering surprise variables which rely heavily on Euribor rates such as
the Euribor measure itself, as well as the level and slope factors. Also
recall from Panel A of Table 1 that the variation explained by the le-
vel factor is substantially lower over the full sample on ECB meeting
days than over the shorter sample.
Contrary to the results presented in Panel A, the parameter esti-
mates of the joint model differ considerably from the individual
estimates of the level and slope factors in some cases, in that the
estimated coefficients change and the interaction terms with the
crisis dummy lose their significance in most cases when the two
factors are estimated jointly. The insignificance of the interaction
terms reflects both the correlation between the regressors15 and
the fact that the principal components capture variation in several
interest rates among which the secured Eonia swap rates were much
less affected by money market tensions as the unsecured Euribor
13
Note that market expectations from Reuters polls are not available for the period
September through December 2008, thus zero surprises are assumed. Evidently, given
that part of the ECB’s heavy cuts took place in that period, the survey measure
becomes somewhat distorted. Surprises for the missing months are likely to be large,
resulting in a larger standard deviation of the forecast error, which implies smaller
surprises. Hence, by assuming zeroes for those months we overestimate survey-based
monetary surprises.
14
The only exception where the R2 increases is the slope factor, likely reflecting the
impacts of the ECB’s large cuts in the fall of 2008, which led to a steepening of the
yield curve. This is in line with the finding in Panel A of Table 1 where the explained
variation of the slope factor appears to be higher over the full sample than over the
shorter sample.
15
Although the level and slope factors are orthogonal, they are no longer
uncorrelated with the interaction terms, and this correlation induces higher variances
for the coefficient estimates.
rates, leading to a less biased measure of monetary policy surprise.16
For bond swap yields b ^c shows a positive sign considering all
surprises but the Eonia measure, and it is highly significant in
the equation of the 2-year rate considering the Euribor, the level
and slope surprises. In addition, the coefficient of the level factor
remains significant even in the joint model. These can be explained
by the high correlation between the 2-year swap yields and the
monetary surprises measured by either the Euribor (the correlation
is 0.72 for the full sample and 0.85 for the crisis period) or the level
factor (0.76 and 0.92) on ECB meeting days. This strong relation
indicates that during the financial turmoil movements in the 2-
year rate on meeting days were mostly driven by factors underly-
ing the changes in Euribor. In fact, the 2-year swap rate followed
quite closely the Euribor futures rates over the full sample period,
which is not surprising given that the reference rate for longer
term swaps is the Euribor.
Overall, (i) big changes on monetary meeting days during the
crisis period are to a great extent related to non-monetary events,
particularly, to money market tensions; (ii) the sharp policy rate
cuts in the crisis period induced large changes in the slope of the
term structure, supplying more explanatory power to the slope
surprise.
4.2. The monetary-jump model
In this subsection we build a model for individual interest rates.
Our aim is not only to capture the time-series characteristics of the
series and estimate the effects of monetary policy surprises, but
also to measure these impacts in a novel way. News incorporates
into asset prices either smoothly or abruptly, inducing sharp move-
ments in the latter case. It is a well known empirical phenomenon
that asset returns are subject to infrequent jumps, and Johannes
(2004) shows that jumps play a dominant role in interest rate
dynamics. According to the general view, jumps occur when new
and unexpected information arrives to the markets, which makes
investors revise their expectations and this adjustment occurs in
a blunt way. This new information is generally some public news
such as macroeconomic announcements or monetary policy
decisions.
The distinction between normal and abrupt information flows is
important, since central banks try to pursue their monetary policy
and shape market expectations in a smooth way. Large surprises
cause uncertainty in the markets and reduce central banks’ pre-
dictability. The bigger transparency and predictability of central
banks over the past decades have considerably improved the effi-
ciency of policy implementation. By studying the extent to which
16
Note that the lower sensitivity of the Eonia swap rates to the financial distress is
also reflected by the insignificant interaction term in column (2).
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
monetary policy is responsible for jumps in interest rates we can
analyse the ECB’s predictability and effectiveness over the sample
period. Moreover, unlike the event-study approach and previous
models, which provide a single coefficient for the impacts of mon-
etary surprises on interest rates, our model supplies a characterisa-
tion of the jump distribution over time that allows the examination
of each individual monetary policy decision. We characterise the
jump distribution over time and let the different surprise variables
affect the interest rate dynamics in a time-varying manner.
The behaviour of daily changes in euro area interest rates is
modelled as
Dr t ¼ lt þ e1;t þ e2;t ; ð3Þ
where
X
p
lt ¼ a0 þ ai Drti ; ð4Þ
i¼1
e1;t ¼ rt zt ; ð5Þ
X
nt
e2;t ¼ J t;k  lJ kt : ð6Þ
k¼1
The ai coefficients stand for possible autoregressive terms.17 In Eq.
(5), e1,t indicates a zero-mean normal innovation, representing diffu-
sive information flow, where zt is an i.i.d. standard normal variable
and rt is the conditional volatility of normal innovations, to be spec-
ified below. Eq. (6) defines the jump innovation term, representing
the impact of abrupt information arrival. Moreover, Jt is the jump
size which is assumed to be normally distributed with constant
mean and variance denoted by lJ and r2J , respectively.18 Note that
the term lJkt serves for adjusting the jump innovation to have con-
ditionally zero mean. We assume that zt and Jt are independent. Fi-
nally, nt refers to a Poisson process with time-varying conditional
intensity parameter kt for the number of jumps (nt 2 {0, 1, 2, . . .}),
occurring in the interval (t  1, t], and nt is also assumed to be inde-
pendent of zt. The density of nt conditional on the information set
I t1 19 is
expðkt Þkjt
Prðnt ¼ jjI t1 Þ ¼ j ¼ 0; 1; 2; . . . ð7Þ
j!
The jump intensity can also be written as kt ¼ Eðnt jI t1 Þ, thus it
can be interpreted as our ex ante assessment of the expected num-
ber of jumps over the interval (t  1,t]. It would also be desirable to
quantify the change in our conditional forecast as the information
set is updated, i.e., the ex post assessment of the expected number
of jumps. Using Bayes’ rule,20 we have
f ðDr t jnt ¼ j; I t1 Þ  Prðnt ¼ jjI t1 Þ
Prðnt ¼ jjI t Þ ¼ j ¼ 0; 1; 2; . . .
f ðDr t jI t1 Þ
ð8Þ
The first term in the numerator is the density of daily interest rate
changes conditional to I t1 and j jumps, and will be given later in
Eq. (17). The second term is in (7) and the denominator is the value
of the likelihood function at time t and given in Eq. (18). The filter in
Eq. (8) is very useful for inference purposes, since the probability
that at least one jump occurred on a day is simply
17
We determine the optimal number of lags both by studying the autocorrelation
structure of the series and by using an information criterion, particularly, the Schwarz
Information Criterion (SIC). Both point to one lag in the 1-month series, two lags in
the 6-month series, and no lags in the daily differences of swap yields.
18
Later we relax the assumption of a constant jump size mean.
19
Note that the information set may contain events that occur between time t  1
and time t. Specifically, it is the amount of information available right before the
closing time of date t.
20
See also Maheu and McCurdy (2004).
2333
Prðnt P 1jI t Þ ¼ 1  Prðnt ¼ 0jI t Þ, which can be directly calculated
from Eq. (8).
Now the ex post expected number of jumps can be straightfor-
wardly calculated as
X
1
Eðnt jI t Þ ¼ jPrðnt ¼ jjI t Þ: ð9Þ
j¼0
This expression will be useful in the following section.
4.2.1. Modelling the conditional variance
The first two conditional moments of Drt are
EðDr t jI t1 Þ ¼ lt ; ð10Þ
 
VarðDr t jI t1 Þ ¼ r2t þ kt r2J þ l2J : ð11Þ
 
The second term in (11), kt r2J þ l2J , is the jump contribution to
the conditional variance.
Suppose first that r2t is assumed to follow a GARCH (1, 1) pro-
cess, i.e.
r2t ¼ x0 þ x1 e2t1 þ x2 r2t1 ; ð12Þ
where et1 = e1,t1 + e2,t1 denotes the total innovation observed at
time t  1. Hence, r2t , besides the impacts of past diffusive innova-
tions, also includes the effects of past jump innovations to interest
rate changes.
Therefore, instead of assuming that jumps affect the conditional
variance only through the time-varying jump intensity, it is rea-
sonable to think that previously realised jumps also have some im-
pact on the GARCH component of the conditional volatility. This
effect is realised via the squared past innovations, e2t1 , and is cap-
tured by the parameter x1 in Eq. (12). As Maheu and McCurdy
(2004) argue, this feedback can be important, since realised inno-
vations may induce different trading strategies in the markets. Evi-
dently, these activities generate further volatility clustering,
besides clustering of jump arrivals.
The problem is that it is difficult to decompose the total feed-
back impact (et) into normal (e1,t) and jump (e2,t) components. Ma-
heu and McCurdy (2004) propose the use of a proxy for the jump
contribution.21 We thus calculate the ex-post expected number of
jumps by (9) and allow it to affect the feedback of past innovations
on future volatility. Hence, we rewrite Eq. (12) as
r2t ¼ x0 þ gðxÞe2t1 þ x2 r2t1 ð13Þ
with22
gðxÞ ¼ exp½x1 þ x1;J Eðnt1 jI t1 Þ: ð14Þ
Because the function g(x) needs to be positive for a well-spec-
ified GARCH process, it is defined in terms of an exponential func-
tion. Likelihood tests, not reported here, show that for money
market rates jump innovations have a different effect on condi-
tional volatility than normal innovations. Hence, for short rates
we use Eq. (14), whereas for long-term rates the simple GARCH
(1, 1) specification seems sufficient to model time-varying diffusive
volatility. These results reflect the conclusions from the visual
inspection of Fig. 1 and from our preliminary estimation results,
described later in Section 4.3, that jumps play a great role in short
rates’ volatility.
21
See also Beber and Brandt (2010) for an application.
22
Maheu and McCurdy (2004) and Beber and Brandt (2010) also include terms in
(14) which allow for asymmetric effects of positive and negative innovations. We also
examined that specification, but found it irrelevant for all interest rates.
2334 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

4.2.2. Modelling the jump intensity and jump size mean 4.2.3. Estimation of the model
The dynamics of jumps can be very simple, e.g. a Poisson pro- The hypotheses underlying Eq. (3) imply that the distribution of
cess with constant arrival rate, see Jorion (1988). Alternatively, Drt, conditional to I t1 and to j jumps, is normal,
the jump arrival process may evolve endogenously over time
f ðDr t jnt ¼ j; I t1 Þ
according to a, say, ARMA process, such as the autoregressive jump 2 3
intensity (ARJI) model of Maheu and McCurdy (2004). They assume 1 ðDr t  lt þ lJ kt  jlJ Þ2
¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4   5:
a constant jump size mean. In addition, both the jump intensity   exp  ð17Þ
and the jump size mean can depend on macroeconomic announce- 2p r2t þ jr2J 2 r2t þ jr2J
ments, as in Beber and Brandt (2010). They include announcement
dummies in the specification of kt, and lJ,t is modelled as a function Integrating out nt yields the conditional density function of Drt in
of news surprises. The combination of the ARJI model and macro terms of observable variables,
variables is examined in Rangel (2011).
X
1
We model the jump intensity as a function of monetary policy f ðDr t jI t1 Þ ¼ f ðDr t jnt ¼ j; I t1 Þ  Prðnt ¼ jjI t1 Þ: ð18Þ
surprises. We do not introduce lagged values of kt in the specifica- j¼0
tion to avoid that large past jumps drive our results. We have seen
in Table 3 that unexpected monetary decisions can explain a large Intuitively, this is a mixture-normal model where the mixture coef-
proportion of interest rate variation on Governing Council meeting ficients are Poisson probabilities driven by monetary surprises. That
days. Now we analyse whether these daily changes reflect jumps or is, every day the jump intensity is governed by monetary surprises,
smooth adjustments. Our intuition is that the mere occurrence of a which in turn determines the probabilities of i jumps occurring on
monetary meeting should not increase the probability of a jump, that day (i = 0, 1, 2, . . .). Notice that the likelihood in (18) involves an
but sharp changes are more likely the larger surprise the ECB infinite summation over the possible number of jumps. For feasible
causes. Thus a simple dummy variable is unable to capture accu- estimation we truncate the summation at 20. Given our model
rately the dynamics of the jump arrival process; instead, jump parameter estimates, the probability of more than ten jumps is
intensity depends on the surprise itself. Moreover, our prior intui- essentially zero. In order to obtain estimates for the unknown
PT
tion is that it is not the sign of a surprise that matters in inducing parameters, the log-likelihood function, given by t¼1 ln f ðDr t j

jumps, but rather its magnitude.23 Therefore, the intensity of the I t1 Þ, has to be maximised.
jump arrival process is given by
4.3. Preliminary empirical results
kt ¼ k0 þ k1 jSt j: ð15Þ
Some simpler models, though not reported here, have been esti-
Since this model nests the constant jump intensity specification
mated first for the shorter sample ending December 29, 2006.24
(with the constraint k1 = 0), standard likelihood ratio tests can be
These models are: (i) a constant Gaussian volatility model; (ii) a
carried out to test whether our monetary-jump model is superior
GARCH (1,1) model; (iii) a jump model with constant Gaussian vol-
to the constant k model.
atility and constant intensity parameter; and (iv) a GARCH-jump
Concerning the jump size mean, the question is now whether, in
model with constant jump intensity. Note that formal comparison
addition to increasing the probability of a jump, monetary sur-
of the models is not possible because they are not nested due to
prises can also affect the size of jumps. From Fig. 2 it can be seen
the presence of nuisance parameter(s) in some specifications (k in
that very large surprises tend to be associated with large daily
the models with jumps, and the GARCH parameters in the models
changes in money market rates, but for long-term swap yields
with GARCH errors). Instead, we use SIC to compare models,
the picture is not so clear. Moreover, the figure only plots observa-
although it is well known that it is not a formal hypothesis test.
tions on meeting days, while it may be the case that the largest dai-
Nonetheless, substantial differences in SIC values may be indicative
ly changes occur on non-meeting days. It turns out that 8 out of the
for model choice.
ten largest movements in the 1-month rate occurred on monetary
We may conclude the following results. First, the Gaussian
meeting days (although one, September 17, 2001, was unsched-
jump model outperforms the pure Gaussian one. This superiority
uled). For the 6-month rate this number is 5, and for the 2-year
of the specifications with jumps also holds when adding GARCH
and 10-year swap yields are 2 and zero, respectively. In addition,
structure to the model. The large differences in SIC values are strik-
the two cases for the 2-year rate are only the 9th and 10th largest
ing when comparing the models with and without jumps. Second,
daily changes. Hence, it seems adequate to let the jump size mean
the GARCH structure is significant and it leads to a noteworthy de-
depend on monetary surprises for money market rates, while for
crease in SIC values. GARCH models for money market rates turn
swap yields we retain a constant lJ. Econometric analysis supports
out to be non-stationary, i.e. x1 + x2 > 1 in Eq. (12), but the intro-
our conjecture. Considering the full sample period the pattern is
duction of jumps eliminates this feature. A similar behaviour is re-
similar, although for the 10-year rate there appear to be two large
ported by Benito et al. (2007) in modelling the overnight Eonia
observations on monetary meeting days, both in the crisis period.
rate. This indicates that jumps account for a considerable compo-
Formally, similarly to Beber and Brandt (2010), we allow the
nent of interest rate volatility. It also supports our finding that pre-
jump size mean to be time-varying for the 1-month and 6-month
viously realised jumps play a substantial role in the daily
rates, and model it as a function of monetary surprise variables.
variability of money market rates.
Moreover, now it is reasonable to think that not only does the size
Some important results of the GARCH-jump model with con-
of the surprise drive the mean of the jump size, but also its direc-
stant jump intensity are presented in Table 4. The estimated jump
tion. We also allow for asymmetric reactions, i.e., that negative sur-
intensities are significantly different from zero up to 2 years of
prises have different impacts than positive surprises:
maturity and it is insignificant for the 10-year rate, suggesting that
lJ;t ¼ b0 þ b1 St þ bþ1 Sþt ; ð16Þ jumps do matter in the daily variation of most euro area interest
rates, whereas for the 10-year rate a simple GARCH (1, 1) specifica-
where Sþ  tion seems satisfactory. However, we will include monetary sur-
t  maxðSt ; 0Þ and St  minðSt ; 0Þ. For swap rates, lJ,t = lJ.
prises in the conditional mean equation of the 10-year rate,
23
We also estimated the model with St in the jump intensity specification instead of
24
its absolute value, but it yields considerably worse fit. The estimation results for these models are available upon request.
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2335

Table 4
GARCH-jump model with constant jump intensity.

1-Month 6-Month 2-Year 10-Year

k 0.1297 0.1453 0.0851 0.1651
(11.92) (7.58) (2.03) (0.91)
lJ 0.0002 0.0021 0.0195 0.0143
(0.54) (1.50) (1.80) (1.13)
rJ 0.0492 0.0389 0.0662 0.0415
(16.63) (13.35) (5.55) (3.16)
Log-L 7120.66 5649.51 3592.80 3618.97
Average jump contribution to total volatility 0.8624 0.6467 0.2473 0.2103
# of jumps 183 142 41 56
# of jumps on ECB meeting days over all jumps 0.1530 0.2465 0.1707 0.0714
# of jumps on ECB meeting days over all meeting days 0.2205 0.2756 0.0551 0.0315

The table contains the key results for the GARCH-jump model with constant jump intensity parameter for the period February 19, 1999 through December 29, 2006. The
P P t  
model is defined as Drt = lt + e1,t + e2,t, where lt ¼ a0 þ pi¼1 ai Dr ti , e1,t = rtzt, e2;t ¼ nk¼1 Jt;k  lJ k, zt  N ð0; 1Þ, J t  N lJ ; r2J , and nt is a Poisson process with constant
intensity k. rt follows a GARCH (1, 1) process, r2t ¼ x0 þ x1 e2t1 þ x2 r2t1 for the 2-year and 10-year rates, while for the 1-month and 6-month rates, rt is defined as
r2t ¼ x0 þ gðxÞe2t1 þ x2 r2t1 , where gðxÞ ¼ exp½x1 þ x1;J Eðnt1 jI t1 Þ. It is assumed that zt, Jt and nt are independent. We use one lag in the 1-month series, two lags in the
6-month series, and no lags in the daily differences of swap yields. The # of jumps is the number of days on which the ex-post probability of at least one jump, Prðnt P 1jI t Þ, is
bigger than 0.5. The average jump contribution to total volatility is the average ratio of kðr2J þ l2J Þ to the total conditional variance, VarðDr t jI t1 Þ. The # of jumps on ECB
meeting days over all jumps is the ratio of the number of jumps on ECB meeting days to the total number of jumps, and the # of jumps on ECB meeting days over all meeting
days is the ratio of the number of jumps on ECB meeting days to the total number of Governing Council meeting days. t-Values are in parentheses.

since, though they appear not to induce jumps, they seem to affect
the level of the daily changes of this rate, as seen in Table 3.
Although the significant k’s are quite similar in magnitude (all
around 10%), this can be explained by the fact that we set k to be
constant. Other jump statistics exhibit considerably different pic-
tures for rates of different maturities.
The contribution of jumps to total volatility decreases with
maturity, but even on the 2-year horizon is around 25%. This find-
ing is in line with that of Beber and Brandt (2010) for US bond re-
turns, although our results for short rates indicate a much higher
contribution of jumps to the total conditional variance. It is also
clear from the table that much more jumps occurred in money
market rates than in swap yields.25 This is consistent with the visual
inspection of Fig. 1, although it may also reflect the fact that, due to
their larger inherent volatility, jumps in long-term rates are more
difficult to identify. This is also supported by the fact that the x2
parameter in the GARCH specification is substantially smaller for
money market rates, indicating that for long-term yields many small
jumps are identified by the model as volatility.
Regarding jumps on Governing Council meeting days, about one
in every six jumps occurred on meeting days in the 1-month and 2-
year rates, while one in every four in the 6-month rate. The high
proportion for the 2-year rate seems contradictory with our earlier
findings and intuition, but note that the 17% means 7 jumps in
absolute terms, whereas the 15% proportion for the 1-month rate
means 28 jumps, that is, four times more. Looking at the ratio of
the number of jumps on meeting days to the total number of meet-
ing days makes the picture clearer. On around a quarter of ECB
meeting days occurred at least one jump in money market rates,
while in the 2-year yield the proportion is only 1/20.
For the full sample, the jump arrival process is specified as
kt = k0 + kcDt, where Dt is the crisis dummy taking on the value
one after January 1, 2007, and zero otherwise, as in (2). Hence,
the jump intensity is equal to k0 over the shorter sample, while it
is k0 + kc over the crisis period. The results, not reported here, show
that the estimated k0’s are similar to the ones presented in Table 4,
and that ^ kc is significantly negative for money market rates and is
positive but insignificant for swap yields. This suggests that the
jump intensity decreased during the crisis, which can be explained
25
We define a jump occurrence as when the ex post probability of at least one jump
is greater than 0.5.
by the fact that, due to the increased uncertainty in the financial
markets, the diffusive component of volatility increased substan-
tially, blurring the identification of jumps. We will study this issue
in the next section in more detail.
The 10-year rate is particular in that ^
k0 turns out to be very high
(0.43) and significant for the full sample. Moreover, the constant
jump intensity model identifies 257 jumps for this rate, which sug-
gests that around 9% of the days were characterised by at least one
jump and around 22 jumps occurred a year. This seems unrealistic,
and looking at Panel D of Fig. 1 we may conjecture that the model
identifies many diffusive innovations as jumps.26
5. Analysis of results
This section reports and discusses the results of the monetary-
jump model. All results examined here refer to the model esti-
mated for the shorter sample, that is, which ends on December
29, 2006. Findings of the model for the full sample will be dis-
cussed in Section 5.5.
5.1. General fit
The key estimation results are given in Panels A and B of Table
5. Comparing the log-likelihood values in Table 5 to those of the
constant jump intensity specification in Table 4, a substantial in-
crease can be observed, especially for money market rates. The
big differences between the log-likelihood values are striking and
show that the monetary-jump model is superior to the constant
k set-up, and all formal likelihood ratio statistics turn out to be
highly significant. In addition, for the short rates the time-varying
jump size mean specification is also statistically superior to the
constant lJ model. All these results suggest that the ECB’s mone-
tary actions appear to contribute significantly both to the jump ar-
rival process and to the jump distribution of euro area interest
rates.
26
We have also checked whether monetary surprises could help this identification
and provide a more realistic picture of the jump arrival process, but only the level
surprise appeared to be significant, and the model implied almost 400 jumps and that
on 70% of monetary meeting days at least one jump realised. This is evidently
unrealistic and reflects the jump identification problem discussed above. Therefore, it
seems that a simple GARCH (1, 1) specification is adequate for the 10-year rate for the
full sample too, with surprise variables in the conditional mean.
2336 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

Table 5
The impacts of monetary policy surprises on individual interest rates in the monetary-jump model.

1-Month 6-Month
Survey Eonia Euribor Level Slope Joint Survey Eonia Euribor Level Slope Joint
Panel A
x0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
(6.01) (6.31) (5.87) (6.06) (6.15) (6.37) (1.73) (1.89) (1.69) (1.53) (1.88) (1.77)
x1 0.3939 0.4718 0.3744 0.3740 0.4119 0.3609 1.6407 1.8694 1.6898 1.6225 1.9115 1.7440
(4.05) (4.64) (3.59) (3.83) (3.99) (3.82) (7.95) (10.14) (8.69) (7.97) (10.18) (9.26)
x1,J 1.7029 1.7348 1.7871 1.8147 1.6985 1.8568 1.4960 1.3872 1.6229 1.6057 1.1886 1.5829
(11.29) (10.86) (12.33) (12.72) (10.47) (12.20) (6.76) (6.02) (7.48) (7.01) (5.42) (6.98)
x2 0.3791 0.4005 0.3878 0.3909 0.3886 0.3703 0.8120 0.8395 0.8220 0.8103 0.8369 0.8217
(12.69) (13.27) (11.16) (12.63) (13.25) (13.11) (28.84) (38.50) (31.36) (27.82) (40.06) (33.33)
k0 0.1271 0.1214 0.1211 0.1184 0.1212 0.1187 0.1596 0.1289 0.1313 0.1337 0.1195 0.1227
(11.46) (10.98) (10.74) (10.96) (10.90) (11.11) (7.18) (7.09) (6.86) (6.66) (6.67) (6.61)
k1 1.4897 0.9379 0.8569 0.6139 0.4810 0.5293 1.0064 1.2068 1.1519 1.2175 0.9285 0.7772
(9.45) (8.18) (4.20) (4.97) (4.27) (3.34) (7.12) (4.55) (8.45) (6.93) (4.57) (5.35)
k2 – – – – – 0.3681 – – – – – 0.2218
(2.95) (2.32)
b0 0.0004 0.0002 0.0002 0.0004 0.0005 0.0000 0.0015 0.0018 0.0024 0.0020 0.0008 0.0021
(0.89) (0.48) (0.55) (1.00) (1.24) (0.05) (1.24) (1.20) (1.66) (1.51) (0.54) (1.39)
b
1 0.0222 0.0361 0.0018 0.0007 0.0009 0.0059 0.0182 0.0197 0.0255 0.0235 0.0005 0.0285
(2.54) (4.88) (0.82) (0.69) (1.81) (3.98) (2.84) (3.15) (5.56) (6.41) (0.45) (6.10)
bþ
1
0.0067 0.0170 0.0011 0.0011 0.0016 0.0142 0.0187 0.0108 0.0148 0.0225 0.0189 0.0264
(8.30) (4.86) (1.38) (2.02) (2.13) (4.28) (5.86) (3.06) (5.29) (5.29) (4.18) (4.71)
b
2 – – – – – 0.0012 – – – – – 0.0036
(1.60) (1.28)
bþ
2
– – – – – 0.0169 – – – – – 0.0161
(4.36) (2.80)
rJ 0.0422 0.0425 0.0443 0.0456 0.0451 0.0444 0.0321 0.0356 0.0341 0.0322 0.0360 0.0344
(16.78) (16.56) (16.15) (16.54) (15.82) (16.29) (13.38) (13.19) (12.85) (12.32) (12.55) (12.95)
Log-L 7184.05 7205.50 7164.75 7156.30 7152.89 7166.98 5700.31 5713.85 5749.70 5750.77 5704.31 5759.88
2-Year 10-Year
Panel B
a0 0.0008 0.0008 0.0008 0.0008 0.0009 0.0006 0.0005 0.0005 0.0006 0.0006 0.0005 0.0006
(0.93) (0.84) (0.91) (0.91) (0.97) (0.71) (0.60) (0.59) (0.68) (0.73) (0.58) (0.73)
a1 – – – – – – 0.0038 0.0046 0.0170 0.0211 0.0159 0.0213
(0.93) (1.08) (4.51) (5.08) (2.98) (6.55)
a2 – – – – – – – – – – – 0.0161
(4.28)
x0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
(1.77) (1.79) (1.82) (1.81) (1.80) (1.77) (2.15) (2.15) (2.15) (2.15) (2.12) (2.09)
x1 0.0304 0.0304 0.0308 0.0308 0.0303 0.0307 0.0434 0.0434 0.0453 0.0453 0.0436 0.0451
(4.05) (4.01) (4.03) (4.01) (3.97) (4.05) (4.21) (4.23) (4.20) (4.12) (4.08) (4.01)
x2 0.9507 0.9512 0.9489 0.9485 0.9501 0.9490 0.9463 0.9463 0.9437 0.9436 0.9459 0.9437
(72.94) (72.67) (69.75) (68.84) (70.46) (70.70) (75.03) (75.50) (71.26) (69.68) (72.65) (67.89)
k0 0.0989 0.0858 0.0914 0.0965 0.0874 0.1009 – – – – – –
(1.89) (1.68) (1.83) (1.81) (1.89) (1.93)
k1 0.4271 0.3859 0.5843 0.5901 0.4392 0.8899 – – – – – –
(1.50) (1.51) (2.22) (2.32) (2.02) (2.21)
k2 – – – – – 0.3397 – – – – – –
(1.23)
b0 0.0164 0.0162 0.0167 0.0163 0.0185 0.0130 – – – – – –
(1.67) (1.51) (1.62) (1.58) (1.79) (1.31)
rJ 0.0607 0.0628 0.0601 0.0588 0.0617 0.0586 – – – – – –
(5.28) (4.76) (5.44) (5.45) (5.56) (5.70)
Log-L 3598.28 3599.48 3606.34 3607.48 3601.42 3608.27 3600.24 3600.42 3611.24 3617.24 3607.76 3625.50

The table contains the results for the monetary-jump model for the period February 19, 1999 through December 29, 2006. The basic model is described in the notes of Table 4,
and the following changes are applied here. For the 1-month, 6-month and 2-year rates the jump intensity is time varying and modelled as kt = k0 + k1jStj, where St stands for
   
   
the different monetary policy surprise measures. In the Joint case the specification of kt changes to kt ¼ k0 þ k1 Stlev el  þ k2 Sslope
t . For the 1-month and 6-month rates, the
 þ þ þ 
specification for the jump size mean is given by lJ;t ¼ b0 þ b 1 St þ b1 St , where St  maxðSt ; 0Þ and St  minðSt ; 0Þ, and in the Joint case lJ,t becomes

lJ;t ¼ b0 þ b1 Slet v el þ bþ1 Stlev elþ þ b2 Sslope

t þ b þ slopeþ
S
2 t . For the 10-year rate a standard GARCH (1, 1) model is estimated with conditional mean lt = a0 + a1St, and in the Joint
case lt ¼ a0 þ a1 Stlev el þ a1 Sslope t . The conditional mean parameters of the models for the 1-month and 6-month rates are not reported to save space. The considered monetary
surprises are the same as in Table 3. t-Values are in parentheses.

For the 10-year rate no formal test is available due to the nui-
sance parameter problem, but it is notable that for the surprise
measures which are significant in the conditional mean equation,
the log-likelihood is almost as high as that of the constant jump
intensity model with more parameters, and for the joint model it
is considerably higher. Looking at the SIC values (not reported here)
which penalise for extra parameters, the findings are similar. This
provides further evidence that, although the series of 10-year yields
cannot be characterised by jumps, monetary policy surprises do af-
fect its level, but they incorporate smoothly rather than abruptly.
5.2. The impacts of monetary policy surprises
The log-likelihoods in Table 5 exhibit a similar pattern to the R2s
in Table 3 in terms of the relative importance of the monetary sur-
prise measures. That is, for the 1-month rate the survey and Eonia
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
measures matter, while on longer horizons the Euribor surprise as
well as the level and slope factors jointly dominate. For all but the
1-month rate the joint estimation of the level and slope factors
yields the highest log-likelihood value, and this increase is highly
significant comparing to the individual estimates with either the
level or the slope factor, except for the 2-year rate. The case of
the 2-year rate is illustrative and helps us better understand the
nature of the level and slope factors, thus we discuss it briefly.
Even though the two factors are orthogonal, they are jointly dif-
ferent from zero only on meeting days, thus the identification of
their individual impacts is not straightforward. In the event-study
setting the combination of linear regression and orthogonal regres-
sors makes this identification very easy. However, in the mone-
tary-jump model we study how likely they induce jumps. Hence,
for an accurate identification we need a sufficient number of large
interest rate realisations on meeting days, as well as a sufficient
number of meeting days when one of the factors is relatively large
while the other is relatively small. However, as we discussed ear-
lier (see Fig. 1 in the on-line appendix), large level surprises are
mostly accompanied by large slope surprises and vice versa. Fur-
thermore, large interest rate realisations mainly occur on days
when the level factor is large.27 As a consequence, the identification
of the slope factor becomes more difficult, and individually it is likely
to capture much of the variation caused by the level factor. This does
not mean, however, that our new factors are wrong measures of
monetary surprises. They still reflect important dimensions of mon-
etary decisions, as shown in Table 3, it is an empirical question how
to capture their impacts on interest rates.
The estimation results of the joint model for the 2-year rate (see
Panel B of Table 5) reflect adequately this identification problem.
The negative coefficient in the jump intensity equation (k2 =
0.34) implies negative jump arrival probabilities for the handful
of days when the slope factor is large but the level factor is very
small.28 The same applies to the 6-month rate, k2 turns negative
and insignificant in the joint model, but including the surprises in
the jump size mean equation helps the identification and tackles
the problem. Even so the estimates of the slope factor in the joint mod-
el for all rates are considerably smaller than individually. For the 2-
year rate this is not sufficient due to the constant jump size mean.
All monetary surprises significantly make a jump more likely in
money market rates, and the sum k0 + k1 is almost always greater
than 0.7, indicating a high probability of at least one jump ex ante
when a one standard deviation surprise occurs.29 This does not
mean, however, that the jump probability ex post is also high, since
it also depends on the change in the given interest rate on that day.
Figs. 3–5 depict variables (only on meeting days) that provide an in-
sight into the characteristics of jump arrivals and their impact on the
conditional volatility of interest rate changes. Moreover, Table 6 re-
ports summary statistics for jump and volatility characteristics im-
plied by the monetary-jump model. For each rate the model with
the highest log-likelihood value is considered, that is, the model with
the Eonia surprise for the 1-month rate, the joint model for the 6-
month rate, and the model with the level surprise for the 2-year rate.
Plots and statistics with the other surprises are similar.
The following findings can be reported. First, the ECB’s unex-
pected decisions induced numerous jumps in euro area interest
rates. In particular, on one-fifth of meeting days the ex post jump
27
Unsurprisingly, since the level factor is the first principal component of daily
interest rate variation on meeting days, and although different rates are used to
compute the PCs than the ones used for estimation, they are very closely related. Thus
the level factor is likely to reflect a great part of large movements in other interest
rates too on meeting days.
28
We estimated the model without parameter restrictions. However, with non-
negativity constraints on the jump intensity parameters k2 converges to zero.
29
When kt is bigger than 0.7 (more precisely, bigger than ln(0.5) = 0.693), the ex
ante probability of at least one jump (see Eq. (7)) is higher than 0.5.
2337
probability exceeded 0.5 for the shortest maturity rate, while for
the 6-month and 2-year rates the proportions are 27% and 21%,
respectively. Evidently, not only on meeting days did jumps occur,
but, as Table 6 shows, both the average ex post probability and
thus the average ex post expected number of jumps are substan-
tially higher on monetary meeting days than on other days. This
can partly be explained by the fact that the jump arrival process
is modelled as a function of monetary surprises, but most of the
implied jumps occurred on non-meeting days, indicating that our
model in general does a good job in identifying jumps.
The results point to some lack of predictability of the ECB’s
monetary policy. Visual inspection of Panels C and D of Figs. 3–5
suggests that great part of this unpredictability originates from
the first half of the sample period. As mentioned earlier (see Foot-
note 4), at its meeting in November 8, 2001 the ECB switched from
bimonthly monetary discussions to monthly meetings, which has
improved its predictability, since the (mainly timing) surprises
due to the too frequent meetings reduced considerably (see Table
1 in the on-line appendix). Despite of this fact, the results of Table
6 show that the average ex post jump probabilities and the average
ex post expected numbers of jumps are not substantially higher
before November 8, 2001 than afterwards, except for the 1-month
rate. Hence, on average, the proportion of days characterised by
jumps in the second subsample is roughly the same as in the first
subsample.
Second, a bigger monetary surprise can induce more than one
jump on the corresponding meeting day. From Panel D of Figs. 3–
5 it is remarkable that 3–4 jumps realised on some meeting days
with large surprises, for example, April 8, 1999 for the money mar-
ket rates or June 8, 2000 for longer horizons. The highest number of
jumps are observed primarily on the days reported in Table 2. This
suggests that after some very large unexpected target changes of
the ECB it took some time for investors to find a consensus and
sharp movements took place in the markets. Hence, our model is
not only able to assess whether or not a given monetary decision
induced abrupt movements in interest rates, but can also quantify
the effect of surprise in several ways such as via the ex post ex-
pected number of jumps.
Third, jumps provide an important contribution to the condi-
tional volatility of interest rate changes, specially on meeting days
(see Table 6) and this contribution decreases with maturity. The
average jump-induced volatility is substantially higher on Govern-
ing Council meeting days, and this difference is bigger the longer
the maturity of interest rates. For example, for the 2-year rate
the jump contribution to the total conditional volatility is more
than twice on meeting days than on other days. A striking feature
is that jumps appear to have added more to volatility in the second
subsample than in the first where most of the big jumps occurred.
The explanation is that diffusive volatility reduced substantially in
the second subsample. In particular, the GARCH volatility of the 1-
month rate changes was 4.5 times bigger before November 8, 2001
than afterwards, while for the 6-month and 2-year rates the ratios
are 2.1 and 1.2. Whether or not this strong moderation of interest
rate volatility happened thanks to the ECB’s decision on discussing
monetary issues more rarely cannot be certainly claimed from our
results (and it is beyond the scope of our paper), but the fact that
the ECB surprised the markets by its unexpected decision consider-
ably less after the change of its meeting practices is likely to have
played an important role.
Turning to the jump size mean parameters, the results in Panel
A of Table 5 indicate that almost all monetary surprise measures
affect significantly the mean of the jump size distribution of money
market rates. Moreover, the fact that both positive and negative
surprises have significant response coefficients provides evidence
that not only the magnitude but also the sign of the surprises mat-
ters. The signs of the coefficients are quite stably negative across
2338 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

Fig. 3. Time series of results of the 1-month rate. Time series derived from the monetary-jump model for the 1-month rate on ECB Governing Council meeting days. Solid
lines correspond to the series implied from the model estimated for the shorter sample period February 19, 1999 through December 29, 2006, while dotted lines correspond
to the series implied from the model estimated for the full sample which ends on June 30, 2010.

surprises, except for the slope factor for which they are positive in
the joint models and negative for positive surprises in the individ-
ual estimations, likely caused by the identification problem dis-
cussed above. Negative coefficients imply a positive jump size
mean after negative surprises and a negative jump size mean after
positive surprises. This seems puzzling, but it can be explained by
two reasons. First, lJ appears with a negative sign in Eq. (6), requir-
ing a positive value for a negative jump in Drt. Second, the sign of
the conditional skewness30 is determined by the sign of lJ. Thus, a
negative jump size mean implies a negative conditional skewness,
and, as a consequence, the mass of the distribution is concentrated
on the positive region, making a large positive change more likely.
The coefficients of the slope factor are positive, which results in
negative jump size mean after a negative surprise and positive lJ
after a positive surprise. This is also broadly in line with intuition,
since a negative slope surprise means a flattening of the yield curve
which, in most cases, occurs due to an increase in short rates. Like-
wise, a positive slope surprise reflects a steepening yield curve, pri-
marily caused by dropping money market rates.
30
For the formula of the conditional skewness, see Eq. (20) in Maheu and McCurdy
(2004).
Although there is clear evidence on asymmetric responses after
positive and negative surprises, it is difficult to link directly this
finding to some specific feature of the ECB’s monetary policy over
the sample period. The higher coefficients of negative surprises (in
absolute value) can be explained by the empirical fact that the big-
gest changes on meeting days in the 1-month and 6-month rates
were mostly negative caused by large negative surprises. Similarly,
on the days of these large drops the slope factor exhibited large po-
sitive realisations, leading to higher response coefficients for posi-
tive slope surprises.
Regarding the model results for the 10-year rate, the joint mod-
el outperforms all other specifications. The coefficients of the level
and slope factors are very similar to those reported for the event-
study regression in Table 3, which shows that the results are robust
even if we consider the whole time series of the 10-year rate and
control for conditional heteroskedasticity.
5.3. Conditional volatility
Recall that after a normal innovation, g(x) = exp(x1), while
after a jump innovation we have g(x) = exp(x1 + x1,J) in Eq. (14).
These values are not presented here, but are plotted in Fig. 2 in
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2339

Fig. 4. Time series of results of the 6-month rate. Time series derived from the monetary-jump model for the 6-month rate on ECB Governing Council meeting days. Solid
lines correspond to the series implied from the model estimated for the shorter sample period February 19, 1999 through December 29, 2006, while dotted lines correspond
to the series implied from the model estimated for the full sample which ends on June 30, 2010.

the on-line appendix for the different monetary policy surprise
measures for the two money market rates.
The following conclusion can be drawn. First, jumps result in
much smaller feedback coefficients than do normal innovations.
This implies that monetary surprises associated with jump innova-
tions are incorporated more quickly into current rates. This finding
is in line with Maheu and McCurdy (2004), and is consistent with
the stylised fact that large changes in rates often affect less persis-
tently expected volatility than smaller changes. Second, the feed-
back coefficients after both normal and jump innovations are
very similar across surprise measures for both rates, suggesting
that regardless of how we define the unexpected component of
monetary decisions, if at least one jump realises previously it has
a quantitatively similar impact on current volatility.
5.4. Robustness test
As we argued in Section 3.2, although the principal components
approach ensures that the derived factors capture all the unex-
pected component of a monetary decision, PCs may also capture
some variation which is not associated with the release. With daily
data this is an important issue, since during a day several other
important pieces of news may affect interest rates. Among them,
perhaps the most relevant are macroeconomic announcements.
For example, the initial jobless claims release is a weekly
announcement in the US and it is always released on a Thursday
at 14:30 CET, only 45 min after the ECB announces its decision
once a month and always on a Thursday, and exactly at the time
when the president’s press conference starts. Hence, if this
announcement contains a big surprise it may also affect interest
rates, and thus its impact may be captured by the PCs.
Another important release which is concurrent many times
with the ECB’s decisions is the monetary meetings of the Bank of
England.31 Of the 127(169) meetings of the ECB over the shorter
(full) sample period, 48(79) occurred concurrently with Bank of Eng-
land meetings, although in more than half of the cases both central
banks left their policy rates unchanged. Bredin et al. (2010) find that
UK monetary policy surprises affect German/euro area bond returns,
even accounting for contemporaneous ECB surprises.
31
We would like to thank an anonymous referee for drawing our attention to this
release.
2340 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343

Fig. 5. Time series of results of the 2-year rate. Time series derived from the monetary-jump model for the 2-year rate on ECB Governing Council meeting days. Solid lines
correspond to the series implied from the model estimated for the shorter sample period February 19, 1999 through December 29, 2006, while dotted lines correspond to the
series implied from the model estimated for the full sample which ends on June 30, 2010.

To check the robustness of our results, we collected the most
important macroeconomic data releases both from the euro area
and from the US (including the Fed’s monetary surprises). In addi-
tion to the aggregated euro area variables, numbers from the three
biggest euro area economies (Germany, France and Italy) and from
the UK were also gathered. In total, 45 macro announcements are
considered here.32 We define a surprise component for each vari-
able, proposed by Balduzzi et al. (2001), as
Akt  Ekt
Skt ¼ ; ð19Þ
rk
where Akt and Ekt are the actual and expected values of variable k,
respectively, and rk denotes the standard deviation of Akt  Ekt . The
median expectation values taken from surveys conducted by
Bloomberg are used as expectations for the macro variables. Overall,
if on day t an announcement occurs in variable k, Skt takes on the va-
lue given in Eq. (19), and zero otherwise.
32
The complete list of releases considered can be found in Section C in the on-line
appendix.
Regarding the monetary releases, the Fed’s monetary surprises
are defined, following Kuttner (2001), as the daily difference in
the current month federal funds futures rates on meeting days
and zero otherwise. The UK monetary surprise is defined similarly
and is captured by the daily change in the 3-month sterling futures
rates on policy meeting days and zero otherwise (see Bredin et al.,
2010). Moreover, both are standardised to have unit variance and
to be comparable to the other surprise variables.
Not only do we consider other variables which are released on a
Thursday, but our aim is to examine how the model parameters
and the implied jump characteristics change when additional vari-
ables are added to the model. Since there are a large number of
macro surprises, we estimate our jump model with each macro
surprise one by one, including them into the jump intensity spec-
ification. Then the insignificant variables are dropped, and finally
all significant variables are included in the model jointly with the
monetary surprise that yielded the highest likelihood before.
The results (not reported to save space) show that our model is
robust to the inclusion of additional variables. The parameter esti-
mates of monetary surprises change only slightly, and the other
model parameters are also unaffected by the macro surprises. Fur-
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2341

Table 6
Jump characteristics for daily changes in euro area interest rates.

1-Month 6-Month 2-Year

All Meeting days Non-meeting days All Meeting days Non-meeting days All Meeting days Non-meeting days
kt ¼ Eðnt jUt1 Þ
Full sample 0.1530 0.6195 0.1214 0.1693 0.8583 0.1227 0.1242 0.5335 0.0965
Before 11/08/01 0.1888 0.8301 0.1214 0.1984 0.9182 0.1227 0.1411 0.5662 0.0965
After 11/08/01 0.1340 0.3916 0.1214 0.1538 0.7892 0.1227 0.1152 0.4980 0.0965
Pr (nt P 1jUt)
Full sample 0.1254 0.2886 0.1156 0.1316 0.3696 0.1154 0.1089 0.3517 0.0924
Before 11/08/01 0.1777 0.3210 0.1626 0.1674 0.3701 0.1460 0.1208 0.3569 0.0961
After 11/08/01 0.0976 0.2119 0.0920 0.1126 0.3690 0.1000 0.1025 0.3461 0.0906
Eðnt jUt Þ
Full sample 0.1439 0.4180 0.1253 0.1492 0.5295 0.1234 0.1242 0.5209 0.0973
Before 11/08/01 0.2162 0.5444 0.1817 0.1970 0.5594 0.1588 0.1449 0.5534 0.1021
After 11/08/01 0.1054 0.2814 0.0968 0.1238 0.4972 0.1055 0.1131 0.4856 0.0949
Jump contribution to total volatility
Full sample 0.8472 0.9163 0.8426 0.5957 0.8731 0.5769 0.2535 0.5492 0.2335
Before 11/08/01 0.7297 0.8839 0.7135 0.5258 0.8456 0.4922 0.2439 0.5318 0.2136
After 11/08/01 0.9097 0.9513 0.9077 0.6328 0.9030 0.6196 0.2587 0.5681 0.2435
# of jumps
Full sample 199 26 173 141 34 107 67 27 40
Before 11/08/01 96 15 81 68 19 49 28 13 15
After 11/08/01 103 11 92 73 15 58 39 14 25

The table reports the main jump and volatility characteristics of our monetary-jump model. The details of the model are described in Table 5.

thermore, the jumps statistics in Table 6 change only moderately,
leaving the main conclusions of the model unchanged. Only for
the 2-year rate, where the impact of macro factors is more rele-
vant, can we observe some increase in the implied jump probabil-
ities, but the qualitative results do not change.
5.5. Including the crisis period
To study the impacts of monetary policy surprises over the full
sample, the specification of the jump arrival process is changed to
kt ¼ k0 þ kc Dt þ k1 jSt j þ kc1 Dt jSt j, thus we allow the jump intensity
to take different values both on meeting and on non-meeting days.
On the other hand, the jump size mean lJ is kept constant even for
money market rates to avoid overfitting of the model.
The results, not reported here but are available upon request,
exhibit a similar pattern to that found in the event-study regres-
sion framework, see Section 4.1. The following findings are worth
reporting. First, after including monetary surprise variables, ^ kc still
remains highly significantly negative for both money market rates
and considering all surprise measures. This means that over the
crisis period jumps were less likely to occur in money market rates
than in the pre-crisis period, regardless whether meeting or non-
meeting days are considered. As in the case of the constant inten-
sity model discussed in Section 4.3, the possible explanation for
this finding is that money market tensions, caused by the financial
distress, increased substantially the volatility of interest rates,
making the disentanglement of jump and diffusion innovations
more difficult. In particular, after the fall of Lehman Brothers in
mid-September 2008 volatility in interest rate markets shot up to
unprecedented levels, and thus the contribution of jumps to total
volatility fell to very low levels (less than 10%) and remained low
for months, see Figs. 3 and 4, where dotted lines represent the im-
plied time series for the full sample on ECB meeting days.
Second, ^kc1 turns out to be significant in the model of money
market rates when considering the Euribor and level surprises, al-
beit only barely for the 6-month rate when the level factor is used,
and it has negative sign. This implies that the ECB’s monetary pol-
icy surprises, when measured by the Euribor or level factor, were
less likely to induce jumps in euro area money market rates over
the period of financial turmoil. This finding partly reflects the
above explanation, namely that the increased interest rate (in par-
ticular, Euribor) volatility blurred the identification of jumps. The
marginal significance of ^ kc1 in the model of the 6-month rate with
the level factor suggests that this must be the driving factor under-
lying the lower jump arrival rate for this rate. Moreover, when we
estimate the joint model, ^ kc1 becomes insignificant, while ^ kc re-
mains highly significant, providing further evidence that jumps
were as likely from the onset of the turmoil as in the pre-crisis per-
iod on ECB meeting days, but the intensified money market ten-
sions make the identification of jumps more difficult. This result
again shows that our new monetary surprise measures help the
better understanding of the impacts of monetary policy on interest
rates than traditional surprises.
From Figs. 3 and 4 it is striking that the estimation of the model
over the full sample results in little changes in the implied mea-
sures compared to the estimation results over the shorter sample.
The differences are somewhat more pronounced for the 6-month
rate because in that case the model is estimated by using the level
and slope surprises, both of which affected by the high variability
of Euribor rates. The graphs also show that during the crisis period
several jumps are identified by the model and that the expected
numbers of jumps do not exceed those observed in the learning
period.
As regards jump statistics, the ECB’s heavy cuts in the fall of
2008 were followed by sharp changes in euro area interest rates,
particularly at shorter maturities, indicating that the interest rate
channel of monetary transmission remained operative.33 While
for the 1-month rate 20% of jumps realised on monetary meeting
days before the financial crisis, from the onset of the turmoil this ra-
tio increased to 23%. For the 6-month rate the surge of jumps on
meeting days is even more pronounced, from 27% to 49%, likely indi-
cating the higher variability of unsecured Euribor rates due to the in-
creased credit and liquidity risks. Therefore, jumps in money
markets rates over the crisis period mostly reflect non-monetary
events rather than uncertainty regarding the ECB’s decisions on its
policy rate. Note however that, in addition to money market ten-
sions, the ECB’s unconventional policy responses to the turmoil also
33
Although, as already mentioned in Section 4.1, the pass-through to bank lending
rates occurred with a considerable lag.
2342 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343
led to higher uncertainty in the markets. Whereas for the 1-month
rate 32% of meeting days were characterised by at least one jump
in the pre-crisis period, the ratio drops to 26% in the crisis period.
On the other hand, for the 6-month rate it increases from 35% to
40%, providing evidence that longer maturities were more affected
by money market tensions and by the ECB’s unconventional
measures.
Turning to the 2-year rate34, ^kc is positive albeit insignificant in
all cases, and so is ^
kc1 . Of course, it does not mean that jumps were
unlikely to occur during the crisis period, but rather that the jump
arrival process is not significantly different in the two subsamples.
Despite of this fact, Panels C and D of Fig. 5 indicate that from the
onset of the financial turmoil several jumps occurred and that the
expected number of jumps exhibits much higher values in some
cases than in the pre-crisis period. Expressed in numbers, before
2007 43% of jumps realised on monetary meeting days, while after
the crisis emerged half of jumps occurred on ECB meeting days. In
addition, whereas only 18% of meeting days were characterised by
at least one jump in the pre-crisis period, the ratio rose to 38% in
the turmoil period. This does not mean, however, that daily changes
in 2-year swap rates became much more sensitive to the ECB’s mon-
etary policy actions, but rather reflects the co-movement of the 2-
year rate with Euribor futures rates which exhibited extreme volatil-
ity levels due to money market tensions.
The conclusions for the 10-year rate remain unchanged, since it
was much less affected by the financial turmoil than rates of short-
er maturities. The joint model outperforms all other specifications,
and although we allow for distinct response coefficients to mone-
tary surprises for the two subsamples in the conditional mean,
both the crisis dummy and the interaction terms turn out to be
insignificant. Obviously, the crisis resulted in a substantial increase
in volatility, but daily changes remained smooth rather than
abrupt.
6. Conclusions
This paper’s aim is twofold. First, we propose new measures of
monetary policy surprises which represent two important dimen-
sions of the ECB’s monetary policy. The level factor captures shifts
in the level of the yield curve, while the slope factor reflects
changes in the slope of the term structure. Both provide relevant
information to policy-makers regarding the monetary transmission
mechanism. A preliminary statistical analysis shows that the inter-
pretation of factors is adequate and that they supply a new way to
examine the ECB’s decisions.
Standard event-study regressions show that our new factors
compellingly outperform traditional monetary surprise measures.
The results provide further overwhelming evidence that the level
and slope factors represent important dimensions of monetary pol-
icy decisions and their structural interpretations is meaningful and
accurate.
Our second objective is to assess the impacts of monetary sur-
prises on euro area interest rates in a novel way. Since central
banks pursue their monetary policy by shaping market expecta-
tions smoothly across the yield curve, a useful way to evaluate
their predictability is to estimate the extent to which they induce
jumps in interest rates with their decision. Hence, we build an
econometric model which allows both the jump arrival process
and the jump distribution to be time-varying and dependent on
monetary policy surprises.
The model outperforms standard time-series specifications and
our results testify the good performance and high relevance of our
34
Note that we find the same identification problem as above, thus the model with
the level factor is taken for analysis.
new measures. We find that, especially in the first half of the sam-
ple, the ECB caused several jumps in interest rates with its unex-
pected decisions or the unexpected timings of its decision, but
after switching to less frequent monetary meetings it has substan-
tially improved its predictability. Our results are proved to be ro-
bust when additional variables are included in the jump arrival
process.
Our model has also been estimated for the recent period of
financial turmoil. The findings show that the model is flexible en-
ough to capture the changed characteristics of the time series due
to the financial distress, and can provide important insights into
how interest rates of different maturities behaved during the dif-
ferent phases of the crisis and how they reacted to the ECB’s policy
responses.
Acknowledgement
Ángel León acknowledges financial support provided by the
Spanish Ministry of Science and Innovation through the Grant
ECO 2011-29751.
Appendix A. Supplementary material
Supplementary data associated with this article can be found,
in the online version, at http://dx.doi.org/10.1016/j.jbankfin.
2012.04.014.
References
Andersson, M., Overby, L.J., Sebestyén, S., 2009. Which news move the euro area
bond market? German Economic Review 10 (1), 1–31.
Balduzzi, P., Elton, E.J., Green, T.C., 2001. Economic news and bond prices: evidence
from the US Treasury markets. Journal of Financial and Quantitative Analysis
36, 523–543.
Beber, A., Brandt, M.W., 2010. When it cannot get better or worse: the asymmetric
impact of good and bad news on bond returns in expansions and recessions.
Review of Finance 14, 119–155.
Benito, F., León, A., Nave, J., 2007. Modeling the euro overnight rate. Journal of
Empirical Finance 14, 756–782.
Bernanke, B., Kuttner, K., 2005. What explains the stock market’s reaction to Federal
Reserve policy? Journal of Finance 60, 1221–1257.
Bernoth, K., von Hagen, J., 2004. The Euribor futures market: efficiency and the
impact of ECB policy announcements. International Finance 7 (1), 1–24.
Bomfim, A.N., 2003. Pre-announcement effects, news, and volatility: monetary
policy and the stock market. Journal of Banking and Finance 27 (1), 133–151.
Brand, C., Buncic, D., Turunen, J., 2010. The impact of ECB monetary policy decisions
and communication on the yield curve. Journal of the European Economic
Association 8 (6), 1266–1298.
Bredin, D., Hyde, S., O’Reilly, G., 2010. Monetary policy surprises and international
bond markets. Journal of International Money and Finance 29, 988–1002.
Chuliá, H., Martens, M., van Dijk, D., 2010. Asymmetric effects of federal funds target
rate changes on S&P100 stock returns, volatilities and correlations. Journal of
Banking and Finance 34, 834–839.
Cragg, J.G., Donald, S.G., 1997. Inferring the rank of a matrix. Journal of
Econometrics 76, 223–250.
ECB, 2002. Experiences with the impact on money market volatility of the
Governing Council’s decision to switch to monthly assessments of the
monetary policy stance. European Central Bank Annual Report 2002, Box 1,
pp. 12–13.
ECB, 2010. Monetary policy transmission in the euro area, a decade after the
introduction of the euro. European Central Bank Monthly Bulletin (May) 85–98.
Ehrmann, M., Fratzscher, M., 2003. Monetary policy announcements and money
markets: a transatlantic perspective. International Finance 6 (3), 309–328.
Estrella, A., Hardouvelis, G.A., 1991. The term structure as a predictor of real
economic activity. Journal of Finance 46, 555–576.
Gürkaynak, R., Sack, B., Swanson, E., 2005. Do actions speak louder than words? The
response of asset prices to monetary policy actions and statements.
International Journal of Central Banking 1, 55–93.
Johannes, M., 2004. The statistical and economic role of jumps in continuous-time
interest rate models. Journal of Finance 59, 227–260.
Jolliffe, I.T., 1995. Rotation of principal components: choice of normalization
constraints. Journal of Applied Statistics 22 (1), 29–35.
Jolliffe, I.T., 2002. Principal Component Analysis, second ed. Springer.
Jorion, P., 1988. On jump processes in the foreign exchange and stock markets.
Review of Financial Studies 1, 427–445.
Kuttner, K., 2001. Monetary policy surprises and interest rates: evidence from the
Fed funds futures markets. Journal of Monetary Economics 47, 523–544.
 Á. León, S. Sebestyén / Journal of Banking & Finance 36 (2012) 2323–2343 2343

Maheu, J.M., McCurdy, T.H., 2004. News arrival, jump dynamics, and volatility
components for individual stock returns. Journal of Finance 49, 755–793.
Pérez-Quirós, G., Sicilia, J., 2002. Is the European Central Bank (and the United States
Federal Reserve) Predictable? ECB Working Paper No. 192.
Rangel, J.G., 2011. Macroeconomic news, announcements, and stock market jump
intensity dynamics. Journal of Banking and Finance 35 (5), 1263–1276.
Rigobon, R., Sack, B., 2004. The impact of monetary policy on asset prices. Journal of
Monetary Economics 51, 1553–1575.
Rigobon, R., Sack, B., 2008. Noisy Macroeconomic Announcements, Monetary Policy,
and Asset Prices. NBER, pp. 335–370 (Chapter 8).
Woodford, M., 2003. Interest and Prices: Foundations of a Theory of Monetary
Policy. Princeton University Press.