Economic Modelling 44 (2015) 319326

Contents lists available at ScienceDirect

Economic Modelling
journal homepage: www.elsevier.com/locate/ecmod

Economic forecasts and sovereign yields

Antnio Afonso a,b,c,, Ana Soa Nunes d
a

ISEG/UTLTechnical University of Lisbon, Department of Economics, Portugal

UECEResearch Unit on Complexity and Economics, Portugal 1
c
European Central Bank, Directorate General Economics, Kaiserstrae 29, D-60311 Frankfurt am Main, Germany
d
ISEG/UTLTechnical University of Lisbon, Portugal
b

a r t i c l e

i n f o

Article history:
Accepted 11 March 2014
Available online 26 April 2014
JEL classication:
C23
E44
H68
Keywords:
Macro forecasts
Fiscal forecasts
Sovereign yields

a b s t r a c t
We assess whether the corrections made to the EC macro and scal forecasts (GDP growth rate, ination, budget
balance, debt ratio, current account) have an impact in sovereign yields. We perform a panel analysis of 15 EU
countries (Austria, Belgium, Germany, Denmark, Spain, Finland, France, United Kingdom, Greece, Ireland, Italy,
Luxembourg, Netherlands, Portugal and Sweden), for the period from 1999:1 until 2012:1, and we also analyse
each country individually, on the basis of a SUR estimation. We nd that corrections in the EC's forecasts impinge
on 10-year sovereign bond yields, particularly corrections in scal variables, being more pronounced in countries
with less favourable economic conditions. The penalization for the yields is higher in corrections for the current
and next years than for previous years.
2014 Elsevier B.V. All rights reserved.

1. Introduction
Since 1998 the European Commission (EC) releases in a regular basis
twice a year, in the spring and in autumn, short-term economic forecasts for the member states of the Economic and Monetary Union
(EMU), candidate countries and other important economies, as the
United States, Japan and the United Kingdom.
Since the forecasts are publicly available, investors may use this
information to decide their investment portfolio, notably their investment in the sovereign bonds. Therefore, the release of these forecasts should, theoretically, have an impact on sovereign spreads.
Indeed, we may argue that rational investors use all the available information; thus, a release of new information will cause a rearrangement in their investment portfolio. However, it is not obvious that
this happens in reality.
Hence, we are interested in assessing what is the impact of releasing economic forecasts on the sovereign yields. If, as expected, the
impact on sovereign yields is signicant, the institutions which
The authors are grateful to an anonymous referee, to Patrcia Martins and to participants
at the UECE Conference on Economic and Financial Adjustments in Europe, Technical
University of Lisbon, 28 June 2013 for very useful comments. The opinions expressed herein
are those of the authors and do reect those of the ECB or the Eurosystem.
Corresponding author.
E-mail addresses: [email protected] (A. Afonso), [email protected]
(A.S. Nunes).
1
UECE is supported by Fundaco para a Cincia e a Tecnologia (Portuguese Foundation
for Science and Technology) through the project PEst-OE/EGE/UI0436/2011.

http://dx.doi.org/10.1016/j.econmod.2014.03.012
0264-9993/ 2014 Elsevier B.V. All rights reserved.

release these forecasts (EC, Organization for Economic and Cooperation Development (OECD), European Central Bank (ECB), and
others) and in particular the governments, want to be aware of the
consequences of forecast accuracy. That is particularly relevant
regarding forecasts for current and next years (the ones with most
obvious possible inuence), but also for past years, as there are
often corrections to past data.
Moreover, there is also an interest for private agents to know the impact of macro and scal forecasts, especially traders, as every anticipation of future movements in bond's prices may bring prot. Therefore,
knowing if and how the bond market reacts to the release of these forecasts is paramount.
The present research will try to provide an answer for this problem,
and it is a contribution to the literature since these linkages have not
been much explored, at least to our knowledge, after reading the
existing related literature. In fact, there are only a few studies for the
USA,2 and some were made 15 or more years ago.3 On the contrary,
there are numerous studies on sovereign spreads' determinants, on
forecasts' accuracy, and on the causes of forecast errors (notably,
Jonung and Martin, 2006; Martins and Mora, 2007; Merola and Prez,
2012; Moulin and Wierts, 2006).

2
3

See Canzoneri et al. (2003).

See Porter-Hudak and Quigley (1994).

320

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

We perform an econometric analysis of the linkages between different

economic forecasts and sovereign yield spreads, using a panel of 15 EU
countries (Austria, Belgium, Germany, Denmark, Spain, Finland, France,
United Kingdom, Greece, Ireland, Italy, Luxembourg, Netherlands,
Portugal and Sweden), covering the period from 1999:1 until 2012:1.
First we do the analysis for the entire panel, and afterwards we study
each country individually, specically on the basis of a SUR analysis. Notice that we use as variables the difference between the forecasts of two
consecutive semesters, and not the forecast itself. This has as purpose to
identify not only the impact of the forecasts' corrections in the yields,
but also the credibility of the previsions.
In a nutshell, we can draw an important conclusion from our
study: corrections in the EC's forecasts do impinge on the 10-year
sovereign bond yield spreads, particularly the corrections in scal
variables (public debt and budget balance), but this impact is different across countries, being more pronounced in countries with less
favourable economic conditions. The penalization for the yields is
higher in corrections for the current and next years than for previous
years. The fact that markets react to this information on macro and
scal forecasts could be consistent with the semi-strong form of
efcient market hypothesis.
This paper is organized as follows. Section two covers the related
literature. Section three explains and discusses the data and the construction of the variables. Section four presents the empirical strategy
and the results. Section ve summarizes the conclusions.
2. Literature review

rate, budget and external imbalances have an impact on OECD countries

sovereign spreads.
Thus, our empirical analysis will consider as determinants of the 10year government bond yields the GDP real growth rate, the public debtto-GDP ratio, the budget balance ratio, the ination rate, given by the
harmonized index of consumer prices (HICP), the real effective exchange rate (more specically, the percentage change to the preceding
year), the current account balance, also as a percentage of GDP (all of
these sourced as EC forecasts), the international risk (represented by
the VIXthe S&P 500 implied stock market volatility index), and monetary policy (represented by the short-term interest rates dened by the
monetary authority). We also control for the existence and strength of
scal rules, including as a variable the Fiscal Rude Index, calculated by
the EC.
There is a theoretic economic relation between all the variables
aforementioned and the 10-year government bond yields. For instance,
with high ination a government tends to unilaterally and partially inate away from its scal indebtedness, and the need for a higher nominal and real long-term bond yield cannot be discarded. Moreover,
expected ination is also seen as an indicator of macroeconomic stability, implying higher sovereign risk. Deviations from past ination can be
assumed from the actual ination rate, or taken as an average of past
observations.
In addition, the current account balance-to-GDP ratio can convey the
existence of a gap between saving and investment and provide expectations of a future depreciation of the domestic currency. Under those circumstances the risk premia demanded by the markets on sovereign
debt may also increase.

2.1. Literature on sovereign spreads' determinants

2.2. Literature on forecast errors
To perform our analysis, we need to know the main determinants of
sovereign bond yield spreads. There is a great amount of literature on
this subject, but there are still some conicting results, as there are
many factors which may inuence sovereign spreads.
However, there are some conclusions that are common to the majority of the studies. The variables which more often appear as signicant
are the level of GDP, GDP per capita or GDP growth rate (Afonso,
2010; Hischer and Nosbusch, 2010), scal performance, through public
debt and budget balance (Afonso, 2010; Afonso et al., 2012; Akitoby and
Stratmann, 2006; Amira, 2004; Baldacci and Kumar, 2010; Dell'Erba and
Sola, 2011; Gruber and Kamin, 2010; Laubach, 2009), current account
balance (Amira, 2004) and monetary policy (Gruber and Kamin, 2010).
The literature also presents several interesting conclusions. For example, the impact of the level of public debt is quantitatively lower
than the one of public decits (Faini, 2006; Laubach, 2009), and worst
scal behaviour lowers the ratings of sovereign debt (Afonso and
Gomes, 2010), which may induce a rise in the yields demanded by market participants. Indeed, government balance and the debt-to-GDP ratios could convey relevant information regarding credit risk or
liquidity risk and help in explaining cross-country nancial risk premia.
Dell'Erba and Sola (2011), using a panel of 17 OECD countries from
1989 to 2009, conclude that a budget decit increase has a greater impact in small peripheral countries or in countries with low nancial integration. Baldacci and Kumar (2010), with data from 31 developed and
in developing countries, between 1980 and 2008, report that higher deficits and levels of public debt lead to a signicant increase in long-term
interest rates, and that the magnitude of such increase depends on the
initial scal, institutional and structural conditions, and on the spillovers
of the global nancial markets.
A study by the EC (2011) nds a negative relationship between the
strength of rules-based scal governance and sovereign spreads, using
the Fiscal Rules Index as a measure of the quality of the scal institutions. Alexopoulou et al. (2009) conclude that the current account and
budget balance, ination, exchange and short-term interest rates,
among other factors, inuence the cost of long-term nance of new
EU countries, while Afonso and Rault (2010) conclude that the ination

Regarding forecast errors, there are two different topics usually

explored: errors in government's forecasts and their causes, and
errors in independent agencies' forecasts and their causes. Both
are important for our work due to the dependency of the EC's forecasts on governments' forecasts, since they are based on the information provided by the country's government.
Concerning governments' forecasts, three main conclusions appear
in the literature:
1) preliminary data releases are biased and non-efcient predictors of
the true values, especially for GDP and public decit, and several corrections occur over the subsequent vintages (Castro et al., 2011;
Frankel, 2011; Jonung and Martin, 2006; Martins and Mora, 2007;
Merola and Prez, 2012; Moulin and Wierts, 2006);
2) the economic cycle is not fully included in the GDP forecast, making
GDP forecast errors an important cause of budget decit errors
(Castro et al., 2011; Frankel, 2011; Jonung and Martin, 2006;
Merola and Prez, 2012; Moulin and Wierts, 2006);
3) being subject to a scal rule, without having strong and independent
supervision, leads to an increase in GDP and budget decit errors,
possibly due to creative accounting (Frankel, 2011; von Hagen and
Wolff, 2006).
Bernoth and Wolff (2008), and von Hagen and Wolff (2006) mention that most European Union's members incur in stock ow adjustments (i.e., the change in their government debt is higher
than the budget decit), which increases the yields demanded by
nancial markets. This increase is higher when the events of creative accounting are reported in the media. On the other hand,
Castro et al. (2011) argue that modications in Eurostat budget
rules also explain a signicant part of forecast errors, and forecasts
may be considered rational after 2 years (i.e., forecast for year t may
be considered correct in year t + 2). This conclusion was the reason
for the use in our study of forecast's corrections till 2 years ago as
regressors.

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

Concerning independent agencies' forecasts, two main conclusions

are possible:

321

4. Empirical strategy and results

4.1. Panel estimation results

1) they seem to be unbiased and efcient, either for the EU and for the
non-EU countries (Melander et al., 2007);
2) however, they appear to be correlated with the electoral cycles,
though less than those from the government, and do not include
all the available information, though they consider more information than governments (Merola and Prez, 2012).
Thus, according to the available empirical evidence, it appears
that independent agencies' forecasts are more reliable than governments', which might notably be linked to the fact that governments
have sometimes to accommodate the political cycle. Melander et al.
(2007) show that the forecasts for GDP, ination, current account
balance and public budgets are the most accurate ones, though not
totally correct. Indeed, the authors report that real growth and budget balance forecasts do not show persistent errors (the study covers
the 19701995 period).
In our analysis we will consider EC's forecasts, as they are part of the
basis of budgetary surveillance in the context of the application of the
Excessive Decits Procedure, and are considered more reliable than the
government's, being a major reference for investors, economists and
managers.

3. Data and variables

As already mentioned, in our study we use a panel of 15 countries:
Austria (AT), Belgium (BE), Germany (DE), Denmark (DK), Spain (ES),
Finland (FI), France (FR), United Kingdom (GB), Greece (GR), Ireland
(IE), Italy (IT), Luxemburg (LU), Netherlands (NL), Portugal (PT) and
Sweden (SE).
The EC's forecasts of budget balance-to-GDP ratio (BAL), public debtto-GDP ratio (DEBT), GDP real growth rate (YR), current account balance
(CA), ination (INF) and real effective exchange rate (REER) were retrieved from the EC's website, as well as the short-term interest rates
(I), the 10-year government bond yields (YIELDS) and the scal rule
index (FRI). The VIX was obtained from Bloomberg's.
The forecasts are released twice a year, typically around March
April (the spring forecast) and OctoberNovember (the autumn
forecast); therefore, our data will be bi-annual. As the rst forecasts were made in the second semester of 1998, our analysis
covers the period from 1999:1 till 2012:1. The short-term interest
rates, the yields and the VIX used relate to the month of the release
of the forecast. We use monthly yields instead of daily ones in
order to try to capture some market anticipation of the forecast's
release.
It is important to understand correctly the meaning of all
variables. We will include forecasts made in year t for year t, year
t + 1, and also for years t 1 and t 2. This choice was based on
Castro et al. (2011), as mentioned above. If forecasts may be considered rational after 2 years, investors will not pay much attention to corrections made after that (except if those corrections
are truly signicant, but it is not a frequent occurrence). Moreover,
as already said, we will use forecasts' corrections as variables, and
not the forecast itself.
Therefore, every semester s we have a forecast for variable X,
for country i and year t, X t i,s . Our variable of interest will then be
Xti,s = Xti,s Xti,s 1, the difference between forecasts made for
year t in two consecutive semesters. We are not interested in knowing if the release of the forecast itself has an impact on the yield, but
whether if the corrections made in the forecasts are signicant
enough to alter the yields. This way, we can evaluate if the EC's
and government forecasts have credibility.

We will start by using a panel data approach, to obtain the aggregate

effect of forecasts' corrections on the sovereign yields. The baseline
specication is
T

YIELDs 0 1:  Xi;s 2  Ii;s 3  VIX i;s 4  FRI i;s

where T = {t, t + 1, t 1, t 2} refers to the year of the forecast, and

X = {BAL, CA, DEBT, INF, REER, YR} is the forecasts' vector, and varies
from regression to regression, depending on the variables we want
to study.
Due to the correlation between DEBT and BAL, we never include
them in the same regression. We excluded REERt + 1 as a regressor because it had too few observations. In addition, we perform the analysis
separately for the years when the forecasts are made for, which means
we have a different table with the eight regressions for forecasts for
years t, t + 1, t 1 and t 2. We do this due to the correlation of the
majority of the variables from one year to another. VIX, FRI and the
short-term interest rate are present in all regressions, since they are
control variables, and some forecast variables are repeated in different
regressions in order to test their impact in more than one way. Nonlinear effects were not accounted for since the focus of the study was
specically to check the effect of the corrections in the EC forecasts. In
order to admit residual heteroscedasticity, we always use the White diagonal covariance matrix.
We use instrumental variables for DEBT and BAL, regarding
forecasts for year t and t + 1, since they are correlated with the
YIELDS. Every year, governments have to make interest payments
to bond owners, an expense that it is accounted for in the budget
balance and, consequently, in public debt. Therefore, the higher
the interest rate demanded by investors in the bond's auction, the
higher will be the budget decit and consequently the stock of future debt. Moreover, forecasts for the scal variables for t and for
t + 1 are also likely to be inuenced by the current 10-year secondary market bond yields.
Additionally, we have performed the WuHausman's endogeneity
test for YR, also for the forecasts for year t and t + 1, to exclude a possible effect of the 10-year sovereign yields on the country's economic
growth. Indeed, higher yields may push public balances to critical
values, forcing governments to adopt somewhat more austere programs, reducing their expenses or increasing their revenues, mostly
through higher taxation. Either way, these are negative stimulus to
the economy, and may have a contractionary effect on real GDP. Finally,
we also perform the Hausman's test, to verify if it is more appropriate to
use xed or random effects. 4
For forecasts concerning the year of their release, we have public
debt and the budget balance corrections as signicant. GDP growth
rate corrections have statistical signicance in two of the seven regressions where they are included, having a positive effect on yields, and
real effective exchange rate in one of the four regressions, having a negative coefcient. The constant term, short-term interest rate and FRI also
have an impact on the yields.
Regarding the forecasts for the next year, scal variables remain statistically signicant. Current account balance corrections appear as signicant in one of the regressions, having a positive but smaller impact
than the scal variables. The constant term, short-term interest rate
and FRI are signicant again.
In the results obtained with forecasts for year t 1, only budget balance is signicant. Public debt no longer has an impact, probably because it is difcult to hide the true value of this ratio, when comparing
4
We report the results for years t and t + 1, the most signicant ones; for more results
see the working paper version.

322

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

Table 1
Estimation results for 10-year yields: forecasts for year t.

IV
Ci,t

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Yes
3.526***
(0.159)

Yes
3.494***
(0.166)

Yes
3.555***
(0.169)

Yes
3.662***
(0.177)

Yes
3.523***
(0.254)
0.398*
(0.210)

Yes
3.460***
(0.262)
0.409*
(0.222)
0.027
(0.037)

Yes
3.580***
(0.267)
0.442*
(0.228)

No
3.639***
(0.174)

BALi,t
CAi,t
DEBTi,t
INFi,t
REERi,t
YRi,t
Ii,t
VIXi,t
FRIi,t
R-square
N
Obs
Endogeneity
Hausman

0.135**
(0.062)
0.299
(0.206)
0.017
(0.027)
0.114
(0.088)
0.298***
(0.038)
0.003
(0.006)
0.020
(0.140)
0.357
15
303
0.396
0.005

0.132**
(0.066)

0.020
(0.028)
0.178
(0.112)
0.328***
(0.039)
0.002
(0.005)
0.084
(0.142)
0.329
15
303
0.493
0.003

0.019
(0.020)
0.077
(0.077)

0.150*
(0.078)
0.243
(0.189)

0.013
(0.026)

0.364
(0.246)

0.166
(0.207)
0.342***
(0.040)
0.011
(0.008)
0.189***
(0.054)
0.185
15
314
0.099
0.600

0.152*
(0.090)
0.314***
(0.039)
0.001
(0.007)
0.172***
(0.054)
0.097
15
349
0.204
0.539

0.194*
(0.108)
0.337***
(0.045)
0.001
(0.006)
0.164
(0.103)
0.057
15
349
0.802
0.763

0.031
(0.034)
0.157
(0.119)
0.352***
(0.049)
0.004
(0.006)
0.105
(0.099)
0.048
15
302
0.879
0.191

0.131
(0.099)
0.310***
(0.046)
0.001
(0.007)
0.129
(0.095)
0.077
15
349
0.794
0.825

0.111
(0.141)
0.041*
(0.022)

0.223***
(0.058)
0.007*
(0.003)
0.053
(0.112)
0.315
15
302
0.070

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. IV indicates if instrumental
variables were used in the regression, N is the number of countries included in the sample, Obs is the number of observations, Endogeneity is the p-value obtained by performing the
WuHausman endogeneity test for YR, and Hausman is the p-value for the Hausman's random effect test.

to budget balance. The constant term, short-term interest rate and VIX
remain signicant, and VIX starts to appear as well.
Finally, in the case of forecasts for year t 2, none of the scal variables is signicant. We nd once more that the constant term, shortterm interest rate and FRI are signicant, as in all the tables above.
Moreover, in this case VIX is also signicant in all regressions, probably
because investors do not pay attention to corrections in forecasts of so
far back; thus, VIX gains signicance.

Overall, we observe that the constant term, BAL, DEBT, I and

FRI are signicant in most of the specications. The scal variables,
BAL and DEBT, are the two forecasts' corrections in which investors focus on. Hence, we may say that investors pay attention to
countries' scal behaviour, demanding higher yields when the
public debt ratio increases and the budget balance decreases,
meaning investors penalize countries which engage in an expansionary scal policy nanced by debt issuance.

Table 2
Estimation results for 10-year yields: forecasts for year t + 1.

IV
Ci,t

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Yes
3.581***
(0.207)

Yes
3.664***
(0.224)

Yes
3.633***
(0.203)

Yes
3.681***
(0.202)

Yes
3.881***
(0.378)
0.456*
(0.259)

Yes
3.843***
(0.405)
0.334*
(0.198)
0.061
(0.038)

Yes
3.795***
(0.314)
0.373*
(0.209)

No
3.629***
(0.166)

BALi,t + 1
CAi,t + 1
DEBTi,t + 1
INFi,t + 1
REERi,t
YRi,t + 1
Ii,t
VIXi,t
FRIi,t
R-square
N
T
Endogeneity
Hausman

0.099*
(0.053)
0.200
(0.164)
0.036
(0.029)
0.048
(0.101)
0.335***
(0.048)
0.011
(0.009)
0.006
(0.192)
0.302
15
303
0.945
0.001

0.085*
(0.052)

0.038
(0.031)
0.020
(0.106)
0.342***
(0.048)
0.011
(0.010)
0.173**
(0.069)
0.197
15
303
0.924
0.103

0.094*
(0.050)
0.190
(0.140)

0.098
(0.092)
0.323***
(0.047)
0.004
(0.008)
0.157
(0.171)
0.234
15
349
0.728
0.089

0.028
(0.018)
0.085*
(0.051)

0.056**
(0.027)

0.207
(0.170)

0.004
(0.097)
0.334***
(0.046)
0.010
(0.010)
0.177***
(0.051)
0.215
15
314
0.295
0.231

0.185
(0.178)
0.328***
(0.051)
0.013
(0.014)
0.225**
(0.089)
0.412
15
349
0.466
0.101

0.047
(0.037)
0.017
(0.179)
0.310***
(0.050)
0.016
(0.013)
0.150
(0.120)
0.186
15
302
0.628
0.169

0.106
(0.168)
0.293***
(0.054)
0.008
(0.010)
0.130
(0.206)
0.131
15
349
0.697
0.028

0.024
(0.063)
0.039*
(0.021)

0.244***
(0.043)
0.006*
(0.003)
0.076
(0.106)
0.319
15
302
0.000

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. IV indicates if instrumental
variables were used in the regression, N is the number of countries included in the sample, Obs is the number of observations, Endogeneity is the p-value obtained by performing the
WuHausman endogeneity test for YR, and Hausman is the p-value for the Hausman's random effect test.

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

323

Table 3
Individual results of estimations of forecasts for year t, regression (2).

AT
BE
DE
DK
ES
FI
FR
GB
GR
IE
IT
NL
PT
SE

Ci,t

DEBTi,t

INFi,t

REERi,t

YRi,t

Ii,t

VIXi,t

R-square

Obs

3.120***
(0.198)
3.312***
(0.208)
2.801***
(0.273)
3.568***
(0.352)
4.013***
(0.298)
2.852***
(0.189)
3.143***
(0.199)
3.287***
(0.242)
9.080***
(1.895)
5.799***
(0.658)
3.995***
(0.277)
2.970***
(0.256)
5.823***
(0.798)
2.671***
(0.210)

0.008
(0.015)
0.047***
(0.016)
0.018
(0.017)
0.036***
(0.013)
0.018
(0.047)
0.119***
(0.016)
0.038*
(0.020)
0.056**
(0.027)
0.256***
(0.041)
0.080***
(0.020)
0.016
(0.032)
0.008
(0.015)
0.035
(0.048)
0.039
(0.034)

0.127*
(0.067)
0.006
(0.046)
0.038
(0.074)
0.343
(0.323)
0.230**
(0.110)
0.052
(0.068)
0.030
(0.064)
0.039
(0.086)
0.436
(0.420)
0.571***
(0.209)
0.208
(0.127)
0.041
(0.059)
0.959***
(0.256)
0.156
(0.136)

0.043
(0.031)
0.080***
(0.023)
0.014
(0.011)
0.102
(0.105)
0.102**
(0.040)
0.052**
(0.025)
0.008
(0.019)
0.066***
(0.012)
0.322
(0.212)
0.069
(0.055)
0.066**
(0.033)
0.013
(0.024)
0.277***
(0.076)
0.013
(0.024)

0.095
(0.058)
0.141**
(0.057)
0.032
(0.040)
0.146
(0.220)
0.194
(0.141)
0.010
(0.035)
0.128*
(0.070)
0.073
(0.097)
1.082**
(0.431)
0.027
(0.111)
0.089
(0.105)
0.004
(0.045)
0.063
(0.225)
0.350***
(0.095)

0.415***
(0.044)
0.302***
(0.049)
0.550***
(0.064)
0.313**
(0.147)
0.117
(0.078)
0.520***
(0.046)
0.413***
(0.046)
0.355***
(0.034)
1.431***
(0.519)
0.526***
(0.178)
0.085
(0.072)
0.474***
(0.059)
0.711***
(0.207)
0.632***
(0.062)

0.000
(0.007)
0.012
(0.007)
0.013
(0.010)
0.003
(0.031)
0.005
(0.011)
0.005
(0.007)
0.001
(0.007)
0.005
(0.009)
0.041
(0.066)
0.022
(0.024)
0.017*
(0.010)
0.004
(0.009)
0.052*
(0.028)
0.000
(0.007)

0.700

24

0.599

25

0.678

25

0.771

13

0.283

25

0.817

25

0.686

25

0.763

25

0.612

25

0.474

25

0.247

25

0.643

25

0.492

25

0.838

20

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. Obs is the number of observations.

Looking at the results, we can conclude that investors pay more

attention to corrections made in forecasts for current and next year
than to corrections made in forecasts for 1 and 2 years back. This
may occur due to investor's condence in EC's forecast accuracy
(in fact, corrections for previous years tend to be smaller), or a
higher investor's preference for values of scal variables for current
and next years. In terms of policy implication, if the more accurate
values are only obtained afterwards, there will only be a penalization
for worst budget balances, and it will be lower than if budget
balances' data were corrected before.
The coefcient for the short-term interest rate is positive. When
a central bank increases these rates, it is engaging in contractionary
monetary policy; thus, one can expect a deceleration in economic
activity, which may worsen budget balances and compromise the
country's ability to pay the debt, thus bringing the yields up. The
coefcient for the FRI is negative. The FRI is calculated based on
the Fiscal Rule Strength Index, which evaluates the quality and visibility of a country's institutional features, essential to the correct
application of the scal rule. The higher their quality and visibility,
the higher is the probability and credibility of following the rule,
thus the lower are the yields demanded. If investors believe that
the government will oblige to the limits imposed, then there is
higher credibility that scal imbalances will be quickly corrected.
The constant term may be interpreted as a risk premium demanded
by investors, related to the probability of default. At the aggregate level,
it is, on average, 3.635, but as we will see ahead it differs quite a lot
across countries, depending on the perceived risk attributed to each
one.
Finally, real GDP growth rate forecast's corrections are also signicant for years t and t 2. It could be expected that this variable
would be as meaningful as the scal variables, as it is a vital indicator
of a country's economic viability and debt sustainability. In spite of its
relevant value as an indicator of the state of the economy, real GDP
growth rates forecasts are the most volatile,5 as they depend on external
5
See, for example, Castro et al. (2011), Merola and Prez (2012) and Martins and Mora
(2007).

and non-controllable factors, among others. Hence, investors may not

always react to small corrections in this variable's forecast, as they are
very frequent, or may actually anticipate some errors (for example,
they may anticipate that forecasts are too optimistic). Another possible
explication will be given ahead, after performing the SUR analysis. Indeed, if corrections in GDP growth rate forecasts have opposite effects
in the countries' yields, then when we estimate for the entire panel
these effects may cancel each other, leading to the statistic insignicance of these corrections. On the one hand, higher growth increases
rm's prots, investment returns and, consequently, stock dividends,
which makes the stock market more protable and attractive, leading
to bond selling, decrease in bond's prices and increase in bond's yields,
in order to attract investors again (a positively sloped yield curve also
tends to reect growth expectations). On the other hand, higher growth
can suggest lower debt and budget balance ratios to GDP, implying a
lower probability of default, which makes the country's sovereign
bonds safer investments and, as a consequence, the yields demanded
are lower.
4.2. Robustness tests
Although there are EC's forecasts until 2012:1, the FRI only has data
until 2010:2. Consequently, in the results shown above, three forecast's
releases were not included (spring and autumn of 2011, and spring of
2012). In order to overcome this problem, we did two robustness
tests, to see if the results obtained were still valid: rst, we added one
observation to the FRI, making the value for this variable in 2011
equal to the one veried in 2010; second, we removed the FRI from
the sample. All econometric details (instrumental variables, random or
xed effects, YR endogeneity and White covariance matrix) remain
valid (results are available on request).
Comparing the results with one extra FRI observation with the
initial baseline specication, we observe that scal variables still
remain the most important variables among the forecasts. However, public debt increases its importance, being signicant for all
years (before it was only signicant for years t and t + 1), and the
magnitude of the budget balance coefcient is lower; it only

324

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

Table 4
Individual results of estimations of forecasts for year t + 1, regression (2).

AT
BE
DE
DK
ES
FI
FR
GB
GR
IE
IT
NL
PT
SE

Ci,t

DEBTi,t + 1

INFi,t + 1

REERi,t

YRi,t + 1

Ii,t

VIXi,t

R-square

Obs

2.971***
(0.207)
3.318***
(0.211)
2.269***
(0.276)
2.658***
(0.265)
4.037***
(0.336)
2.669***
(0.203)
3.029***
(0.185)
3.030***
(0.288)
9.761***
(1.983)
5.162***
(0.707)
3.790***
(0.269)
2.874***
(0.253)
5.044***
(0.811)
2.792***
(0.239)

0.011
(0.012)
0.019
(0.012)
0.024*
(0.013)
0.091**
(0.037)
0.011
(0.021)
0.073***
(0.014)
0.043***
(0.011)
0.024
(0.016)
0.119***
(0.041)
0.098***
(0.027)
0.022
(0.020)
0.016
(0.010)
0.162***
(0.042)
0.055**
(0.027)

0.331***
(0.118)
0.008
(0.088)
0.238**
(0.100)
1.442***
(0.440)
0.222
(0.156)
0.220*
(0.124)
0.260
(0.109)
0.152
(0.147)
0.269
(0.949)
1.562***
(0.474)
1.235***
(0.250)
0.005
(0.015)
2.318***
(0.532)
0.249
(0.173)

0.037
(0.036)
0.056***
(0.020)
0.029**
(0.013)
0.150**
(0.064)
0.095**
(0.037)
0.024
(0.023)
0.037*
(0.019)
0.073***
(0.015)
0.630***
(0.214)
0.020
(0.061)
0.053*
(0.030)
0.006
(0.020)
0.311***
(0.085)
0.005
(0.028)

0.174*
(0.093)
0.103
(0.084)
0.155*
(0.091)
1.600***
(0.467)
0.200
(0.130)
0.299***
(0.089)
0.207***
(0.069)
0.209*
(0.117)
1.037**
(0.499)
0.197
(0.168)
0.191
(0.130)
0.169***
(0.062)
0.671**
(0.323)
0.298**
(0.141)

0.444***
(0.047)
0.309***
(0.051)
0.574***
(0.066)
0.519***
(0.123)
0.054
(0.088)
0.527***
(0.046)
0.436***
(0.044)
0.381***
(0.038)
1.692***
(0.554)
0.748***
(0.211)
0.066
(0.068)
0.499***
(0.059)
0.466**
(0.206)
0.694***
(0.071)

0.006
(0.008)
0.010
(0.008)
0.008
(0.010)
0.025
(0.022)
0.011
(0.013)
0.004
(0.007)
0.000
(0.007)
0.004
(0.010)
0.036
(0.071)
0.032
(0.026)
0.024**
(0.010)
0.000
(0.009)
0.049
(0.029)
0.020**
(0.008)

0.686

24

0.584

25

0.702

25

0.847

13

0.192

25

0.810

25

0.741

25

0.781

25

0.535

25

0.453

25

0.385

25

0.658

25

0.518

25

0.810

20

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. Obs is the number of observations.

appears signicant in forecasts for year t 1, and before it was also

for years t and t + 1. The real GDP growth rate never appears as
signicant, as well as the current account balance. On the contrary,
ination now appears as signicant for years t, t + 1 and t 2,
when it used to be signicant only for t 2, and real effective
exchange rate has a signicant impact, regarding forecasts for
years t and t + 1, like current account balance, in forecasts for
year t + 1.
Hence, adding the year 2011 to our sample allows keeping the main
conclusions, but changes some of the results. This may happen due to
instability and uncertainty of this year (in 2011 Portugal asked for a nancial assistance, implementing the EC/ECB/IMF Economic Adjustment
Programme, Greece asked for a second nancial loan, Italian and
Spanish bonds started to be under pressure), which leads to a bigger suspicion by the investors, not relying so much on public balance and GDP growth rate forecasts, as they tend to undergo
several ex-post corrections.
As stated above, we also tested the same regressions without the FRI
data, which allows for three more time series observations per country.
The results (available on request) go in the same direction than those of
the rst robustness test.
These results seem to conrm the idea that the instability and uncertainty of 2011 and 2012 may alter somehow the results obtained in the
initial panel. The disbelief in government's accounts could have led investors to overlook the budget balance corrections, as they did not see
them as very credible in that context, and they started to give more importance to government debt. In addition, countries began to rely on exportations to grow, as their internal demand was sluggish; thus, the real
effective exchange rate increased its importance as an indicator of the
country's economic evolution. Also, the constant term increased, indicating that investors demanded a higher risk premium, due to higher
risk and uncertainty in the bond market.
4.3. Country estimationSUR
In addition to our panel analysis, we have performed an individual
analysis for the countries. Investors may react differently to corrections

in forecasts, as they give different credibility to each country, once they

have different characteristics.
We have estimated a system of equations, one for each country, to
nd the individual coefcients. For that purpose, we used the Seemingly
Unrelated Regressions (SUR) model. We will use this model in two different specications, due to the correlation between public debt and
budget balance, as mentioned above:
T

YIELDs 0 1  DEBTi;s 2  INFi;s 3  REERi;s 4



T
YRi;s

5  Ii;s 6  VIX i;s

YIELDs 0 1  BALi;s 2  CAi;s 3  REERi;s 4



T
YRi;s

5  Ii;s 6  VIX i;s

where T = {t, t + 1, t 1, t 2} refers to the year of the forecast. From

Eq. (2) we will create a system of fourteen regressions, one for each
country (Luxembourg is excluded, because it has very few observations), and we do the same with Eq. (3). Once again, we separate the regressions through year of forecast, so we will have eight systems,
regarding forecasts for year t, t + 1, t 1 and t 2 for both regressions.
Notice that we remove the FRI as a regressor, although it was signicant
in the panel. We need to do this because the FRI is a constant for Greece,
and almost a constant for Belgium and Netherlands, which causes
collinearity problems.
The results of the estimations for years t and t + 1 for regression (2)
and (3) are reported in Tables 3, 4, 5, and 6. 6 (See Tables 1 and 2.)
Looking at the results, we observe that the coefcients and the signicant variables naturally vary across countries. In addition, while in
the initial result corrections to public debt and budget balance, the
short-term interest rate and the constant term were the variables
which stood out, now corrections in the real effective exchange
rate and real GDP growth rate are also important determinants of the
6

The results of forecasts for years t 1 and t 2 are also available from the authors.

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

325

Table 5
Individual results of estimations of forecasts for year t, for regression (3).

AT
BE
DE
DK
ES
FI
FR
GB
GR
IE
IT
NL
PT
SE

Ci,t

BALi,t

CAi,t

REERi,t

YRi,t

Ii,t

VIXi,t

R-square

Obs

3.108***
(0.202)
3.452***
(0.206)
2.840***
(0.262)
3.282***
(0.310)
4.026***
(0.283)
2.858***
(0.215)
3.214***
(0.181)
3.372***
(0.217)
8.554***
(2.051)
5.595***
(0.742)
4.048***
(0.264)
3.003***
(0.244)
5.779***
(0.939)
2.707***
(0.190)

0.114
(0.076)
0.036
(0.045)
0.171*
(0.091)
0.260
(0.163)
0.213***
(0.055)
0.177**
(0.069)
0.168***
(0.050)
0.145***
(0.054)
0.219
(0.188)
0.095***
(0.022)
0.301**
(0.131)
0.049
(0.060)
0.116
(0.219)
0.149
(0.097)

0.032
(0.035)
0.002
(0.021)
0.054
(0.048)
0.150
(0.116)
0.013
(0.016)
0.046
(0.035)
0.041
(0.031)
0.033
(0.056)
0.456**
(0.222)
0.105
(0.140)
0.187***
(0.069)
0.039
(0.028)
0.167
(0.119)
0.097
(0.066)

0.032
(0.040)
0.077***
(0.025)
0.027
(0.019)
0.114*
(0.079)
0.117***
(0.031)
0.081***
(0.028)
0.057***
(0.019)
0.041***
(0.010)
0.851***
(0.192)
0.079
(0.053)
0.050
(0.036)
0.008
(0.038)
0.321***
(0.091)
0.021
(0.021)

0.034
(0.060)
0.068
(0.047)
0.090
(0.077)
0.006
(0.258)
0.357***
(0.086)
0.020
(0.056)
0.059
(0.053)
0.215**
(0.101)
0.623
(0.500)
0.139*
(0.076)
0.169*
(0.099)
0.033
(0.070)
0.175
(0.245)
0.279***
(0.082)

0.390***
(0.044)
0.290***
(0.049)
0.506***
(0.062)
0.305**
(0.134)
0.071
(0.073)
0.526***
(0.052)
0.392***
(0.042)
0.324***
(0.028)
1.576***
(0.561)
0.440**
(0.194)
0.091
(0.069)
0.469***
(0.057)
0.575**
(0.239)
0.677***
(0.057)

0.002
(0.007)
0.006
(0.007)
0.011
(0.009)
0.020
(0.026)
0.013
(0.010)
0.003
(0.008)
0.005
(0.007)
0.004
(0.008)
0.065
(0.075)
0.027
(0.027)
0.018*
(0.009)
0.007
(0.009)
0.051
(0.033)
0.006
(0.006)

0.707

24

0.574

25

0.703

25

0.759

13

0.336

25

0.746

25

0.729

25

0.783

25

0.529

25

0.350

25

0.319

25

0.672

25

0.387

25

0.861

20

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. Obs is the number of observations.

10-year bond yields. In some countries, current account balance and

ination corrections also have a signicant impact on yields.
As it might be expected, the estimated coefcients in Greece, Ireland
and Portugal tend to be higher than in other countries. We seem to conrm that a country's credibility is an essential factor in determining its
funding costs, due to the risk premium demanded, but also because

countries with lower credibility tend to have yields that are more reactive to forecasts' corrections.
After making the individual analysis, it is visible that the real
effective exchange rate and real GDP growth rate corrections are not
so important in the panel results because they have opposite effects in
some countries.

Table 6
Individual results of estimations of forecasts, year t + 1, regression (3).

AT
BE
DE
DK
ES
FI
FR
GB
GR
IE
IT
NL
PT
SE

Ci,t

BALi,t + 1

CAi,t + 1

REERi,t

YRi,t + 1

Ii,t

VIXi,t

R-square

Obs

2.874***
(0.197)
3.385***
(0.210)
2.694***
(0.265)
3.079***
(0.337)
3.975***
(0.318)
2.756***
(0.220)
3.148***
(0.180)
3.039***
(0.249)
9.165***
(1.980)
5.315***
(0.819)
4.028***
(0.300)
2.942***
(0.239)
5.670***
(0.899)
2.799***
(0.224)

0.138***
(0.049)
0.032
(0.031)
0.063
(0.043)
0.266
(0.220)
0.118***
(0.038)
0.180***
(0.040)
0.078***
(0.029)
0.049
(0.042)
0.024
(0.226)
0.202*
(0.115)
0.033
(0.052)
0.051
(0.034)
0.178
(0.205)
0.242*
(0.135)

0.059*
(0.030)
0.014
(0.018)
0.094**
(0.040)
0.109
(0.113)
0.005
(0.014)
0.023
(0.030)
0.030
(0.027)
0.050
(0.061)
0.791***
(0.206)
0.062
(0.143)
0.064
(0.043)
0.036
(0.028)
0.090
(0.136)
0.060
(0.073)

0.025
(0.039)
0.061***
(0.017)
0.024
(0.020)
0.148*
(0.082)
0.074**
(0.034)
0.074***
(0.025)
0.050***
(0.017)
0.051***
(0.012)
1.134***
(0.219)
0.061
(0.064)
0.051**
(0.025)
0.011
(0.029)
0.155
(0.097)
0.016
(0.026)

0.237**
(0.100)
0.047
(0.062)
0.108
(0.101)
0.527
(0.537)
0.242**
(0.100)
0.224**
(0.089)
0.121*
(0.065)
0.175*
(0.101)
0.297
(0.437)
0.071
(0.138)
0.107
(0.102)
0.130*
(0.072)
0.588**
(0.258)
0.423**
(0.213)

0.411***
(0.044)
0.308***
(0.051)
0.537***
(0.064)
0.325**
(0.148)
0.083
(0.080)
0.515***
(0.050)
0.398***
(0.042)
0.370***
(0.032)
2.132***
(0.551)
0.466**
(0.205)
0.085
(0.075)
0.473***
(0.054)
0.495**
(0.226)
0.690***
(0.065)

0.011
(0.007)
0.006
(0.007)
0.009
(0.009)
0.024
(0.031)
0.012
(0.012)
0.001
(0.008)
0.002
(0.007)
0.003
(0.009)
0.110
(0.073)
0.047
(0.030)
0.018*
(0.011)
0.003
(0.009)
0.026
(0.033)
0.018**
(0.008)

0.750

24

0.587

25

0.705

25

0.696

13

0.233

25

0.753

25

0.735

25

0.778

25

0.549

25

0.276

25

0.202

25

0.683

25

0.404

25

0.816

20

Note: the asterisks *, ** and *** represent signicance at 10, 5 and 1% level, respectively. The values present between parentheses are the standard error. Obs is the number of observations.

326

A. Afonso, A.S. Nunes / Economic Modelling 44 (2015) 319326

5. Conclusion

References

In our study we have assessed the relevance of macro and scal forecast vintages for the explanation of sovereign yield developments in a
panel of 15 EU countries. Our analysis covers the period from 1999:1
till 2012:1.
We show that we can draw an important conclusion: corrections in
the EC's forecasts do impinge on the 10-year sovereign bond yield
spreads, particularly the corrections in scal variables (public debt and
budget balance), but this impact is different across countries, being
more pronounced in countries with less favourable economic conditions.
It seems that whether or not macro and scal forecasts are consistently seen as credible by the markets plays a relevant work. On the
one hand, the credibility that investors give to EC's forecasts is relevant,
and on the other hand the credibility that they give to the country and,
consequently, to governments' forecasts is also paramount.
As we have seen, higher credibility means yields will react less to
changes in forecasts. Hence, in spite of the incentive that governments
have to report less accurate forecasts, as the penalization is higher in
corrections for the current and next years than for previous years, if it
lowers its credibility, it may be worse than revealing the right way the
true results.
A relevant policy implication is that if more accurate values are only
known afterwards, the market penalization for worst budget balances
will be lower than if budget balances' data was corrected ex-ante. This
implies the need for a better perception of the forecast errors by market
participants, which could imply additional scepticism regarding the initial vintage forecasts, and already an increase in the yields at the time of
probably too optimistic 1st year vintage forecasts.
We also saw evidences that the sovereign debt crisis altered the variables to which investors pay attention. After including 2011 and 2012
forecasts, the budget balance lost statistical signicance, public debt became a more relevant determinant, and the real effective exchange rate
started to be signicant as well. Also, the constant term increased, indicating that investors demanded a higher risk premium, due to higher
risk and uncertainty in the bond market.
However, it is important to notice that there are some limitations in
our analysis. In fact, the number of observations is not very large, which
may bias our results, especially when we perform the SUR analysis. In
addition, the period under analysis is very typical, since half of the
years considered encompass the subprime and subsequent sovereign
crisis. As follow up work it would be useful to separate the data during
the sovereign crisis, in order to understand its consequences on investors' reactions. However, that is not possible, due to the yet low number
of forecasts made after the beginning of the crisis, but it stays as a
possible future research development.

Afonso, A., 2010. Long-term government bond yields and economic forecasts: evidence
for the EU. Appl. Econ. Lett. 17 (15), 14371441.
Afonso, A., Gomes, P., 2010. Do scal imbalances deteriorate sovereign debt rating? Revue
conomique 62 (6), 11231134.
Afonso, A., Rault, C., 2010. Long-run determinants of sovereign yields. CESifo Working
Paper 3155.
Afonso, A., Arghyrou, M., Kontonikas, A., 2012. The determinants of sovereign bond yield
spreads in the EMU. Department of Economics, ISEG-UTL, Working Paper 36/2012/
DE/UECE.
Akitoby, B., Stratmann, T., 2006. Fiscal policy and nancial markets. IMF Working Paper
16.
Alexopoulou, I., Bunda, I., Ferrando, A., 2009. Determinants of government bond spreads
in new EU countries. ECB Working Paper 1093.
Amira, K., 2004. Determinants of sovereign Eurobonds yield spreads. J. Bus. Finan. Acc. 31
(56), 795821.
Baldacci, E., Kumar, M., 2010. Fiscal decits, public debt and sovereign debt yields. IMF
Working Paper 184.
Bernoth, K., Wolff, G., 2008. Fool the markets? Creative accounting, scal transparency
and sovereign risk premia. Scott. J. Polit. Econ. 55 (4), 465487.
Canzoneri, M., Cumby, R., Diba, B., 2003. Should the European Central Bank and the Federal Reserve be concerned about scal policy? Proceedings, 333-389, Federal Reserve
Bank of Kansas City.
Castro, F., Prez, J., Vives, M., 2011. Fiscal data revisions in Europe. Banco de Espaa,
Documentos de Trabajo N 1106.
Dell'Erba, S., Sola, S., 2011. Fiscal policy, interest rates and risk premia in open economy.
The Graduate Institute of International and Development Studies.
EC, 2011. Public nances in EMU2011. European Commission.
Faini, R., 2006. Fiscal policy and interest rates in Europe. Econ. Policy 21 (47), 443489.
Frankel, J., 2011. Over-optimism in forecasts by ofcial budget agencies and its implications. Oxf. Rev. Econ. Policy 27 (4), 536562.
Gruber, J., Kamin, S., 2010. Fiscal positions and government bond yields in OECD countries. Board of Governors of the Federal Reserve System, International Finance Discussion Papers.
Hischer, J., Nosbusch, Y., 2010. Determinants of sovereign risk: macroeconomic fundamentals and the pricing of sovereign debt. Eur. Finan. Rev. 14, 235262.
Jonung, L., Martin, L., 2006. Improving scal policy in the EU: the case for independent
forecasts. Econ. Policy 21 (47), 491534.
Laubach, T., 2009. New evidence on the interest rate effects of budget decits and debt. J.
Eur. Econ. Assoc. 7 (4), 128.
Martins, J., Mora, L., 2007. How reliable are the statistics for the Stability and Growth Pact?
Banco de Espaa, Notas Estadsticas N. 4.
Melander, A., Sismanidis, G., Grenouilleau, D., 2007. The track record of the Commission's
forecastsan update. Directorate General for Economic and Financial Affairs,
European Commission.
Merola, R., Prez, J., 2012. Fiscal forecast errors: governments vs independent agencies.
Banco de Espaa Working Paper 1233.
Moulin, L., Wierts, P., 2006. How credible are multiannual budgetary plans in the EU?
Banca di Italia.
Porter-Hudak, S., Quigley, M., 1994. A new approach in analyzing the effect of decit announcements of interest rates. J. Money, Credit, Bank. 26 (4), 894902.
von Hagen, J., Wolff, G., 2006. What do decits tell us about debt? Empirical evidence on
creative accounting with scal rules in the EU. J. Bank. Financ. 30 (12), 32593279.