The Impact of Banking Sector Stability on the

Real Economy ?

Pierre Monnin a,∗ , Terhi Jokipii b

a Swiss National Bank, Financial Market Analysis, Boersentrasse 15, 8022 Zurich,
Switzerland
b Swiss National Bank, Financial Stability, Bundesplatz 1, 3001 Bern, Switzerland

Abstract

This article studies the relationship between the degree of banking sector stability
and the subsequent evolution of real output growth and inflation. Adopting a panel
VAR methodology for a sample of 18 OECD countries, we find a positive link be-
tween banking sector stability and real output growth. This finding is predominantly
driven by periods of instability rather than by very stable periods. In addition, we
show that an unstable banking sector increases uncertainty about future output
growth. No clear link between banking sector stability and inflation seems to exist.
We then argue that the link between banking stability and real output growth can
be used to improve output growth forecasts. Using Fed forecast errors, we show
that banking sector stability (instability) results in a significant underestimation
(overestimation) of GDP growth in the subsequent quarters.

Key words: Banking sector stability, real output growth, output growth forecasts.

JEL classification: E20, E44, G21.

? The authors thank Signe Krogstrup for collecting data on inflation and growth forecasts. We
also thank an anonymous referee and the members of the SNB Financial Stability group for their
useful comments. The opinions expressed herein are those of the authors and do not necessarily
reflect the views of the Swiss National Bank.
∗ Corresponding author. Tel: +41 31 327 06 69. Fax: +41 31 327 07 28.
Email addresses: pierre.monnin@snb.ch (Pierre Monnin), terhi.jokipii@snb.ch (Terhi
Jokipii).

June 2010
1 Introduction

The increased incidence of banking and financial crises over the last quarter
century has triggered an active research agenda, not only on the underlying
causes of crises, but also on their impact on the real economy. Literature in
this field has been mainly focussed in two directions: first, on understanding
links between banking sector characteristics and long term growth and, sec-
ond, on quantifying the costs of banking sector crises in terms of real output
losses. In the first strand of research, Levine (1997, 2001) demonstrates the
link between the openness of the banking sector and economic growth. Several
other authors (Levine, 1997, 1998, Levine, Loayza, and Beck, 2000, King and
Levine, 1993a,b, Demirgüç-Kunt and Maksimovic, 1998, Rajan and Zingales,
1986) have highlighted that the degree of development in the financial sector
acts as an important contributor to economic growth. 1 The second strand
of research reaches a clear conclusion: banking crises have usually coincided
with, or preceded, a substantial economic slowdown (see among others Hog-
art, Reis, and Saporta, 2002, Boyd, Kwak, and Smith, 2005, Serwa, 2007,
Kroszner, Laeven, and Klingebiel, 2007, Dell’Ariccia, Detragiache, and Rajan,
2008). The literature is however, far less clear regarding whether or not the
banking sector is the main trigger of the economic slowdown, as it is difficult to
separate cause and effect in the financial sector real economy nexus (Kamin-
sky and Reinhart, 1999, Demirgüç-Kunt and Detragiache, 1997, 2005, Hilbers,
Otker-Robe, Pazarbasioglu, and Johnsen, 2005). These empirical conclusions
have prompted an increased interest in both policy makers and academics to
assess, from a theoretical point of view, the extent to which macroeconomic
policies and banking system soundness depend on one another (Benink and
Benston, 2005, Gupta, 2005, European Central Bank, 2006, Deutsche Bundes-
bank, 2006, Goodhart, Sunirand, and Tsomocos, 2006).

This paper clearly belongs to the second strand of research. Indeed, we are
not concerned by the effect of financial development on long term growth,
but rather by the (possibly transitory) impact of banking sector instability
on real output growth. We are also not interested in investigating the events
at the roots of banking sector instability per se, and thus we take banking
sector instability as given, without examining its causes. One flaw of the stud-
ies cited above in the second strand of research is that they focus solely on
the loss in output growth during or after banking crises, ignoring the banking
sectors impact during more stable times. 2 The main reason for this is that
many authors have chosen to work with binary dependent variables (crisis
vs. non crisis). However, several drawbacks are associated with adopting this
approach. First, bank crises are relatively rare events. Second, the choice of
1
This literature is reviewed in Levine (2004).
2
One exception to this is Rancière, Tornell, and Westermann (2008) who shows a positive link
between average growth and banking sector instability.

1
threshold for defining a crisis is highly discretionary. Finally, binary variables
impose the unrealistic assumption that a banking sector that is not experi-
encing a crisis is necessarily healthy. The focus on crises vs. non crisis periods
has consequently resulted in a lack of research into the impact of the banking
sector on the real economy in less extreme times. It remains unclear whether
’normal’ reductions in banking sector stability – i.e. a level of instability that
can regularly be observed but that does not translate into a banking crisis –
have a significant impact on growth. 3

In this paper, we contribute to the literature by empirically assessing whether

the degree of banking sector stability has an impact on the real economy in
non extreme times. In particular, we are interested in determining whether a
period of banking sector instability is followed by reduced real output growth
in the subsequent quarters. We measure banking sector stability based on the
banking sector’s probability of default. This measure has the advantage of
being continuous by nature and thus captures a continuum of states, rather
than classification as either crisis or non crisis. This contrasts with the bi-
nary variables approach usually adopted in this context. In contrast to other
distance-to-default measures previously used, our measure allows for a time-
varying volatility for banks assets. Following Aspachs, Goodhart, Tsomocos,
and Zicchino (2007), we explore the impact of banking sector stability on real
output growth and inflation rates in the subsequent quarters with a panel vec-
tor autoregressive model 4 . Our sample consists of quarterly data for 18 OECD
countries over the 1980-2008 period. The novelty of our set-up is that it allows
the variance of output growth to be dependant of the state of the banking sec-
tor. This captures the impact of banking sector instability on output growth
as well as on its uncertainty.

Our results can be summarized as follows: banking sector stability is an impor-

tant driver of future GDP growth. Periods of stability are generally followed
by an increase in real output growth and vice versa, a finding which appears
to be driven predominantly by periods of relative instability rather than by
periods of stability. In addition, we show that banking sector instability is
followed by higher uncertainty about output growth.

The link between banking sector stability and economic activity is of particular
interest to policy makers which base their monetary policy decisions on eco-
nomic forecasts. Research focussing on understanding whether financial stabil-
ity should be considered in the setting of monetary policy has investigated the
3
For example, consider a banking sector which suffers credit losses in a business cycle downturn
but which is still able to function without external help. The stability of such a banking sector has
clearly decreased after credit losses, but since it is still functioning, it is not in a fully fledged crisis
either.
4
Allenspach and Perrez (2008) study interactions between the banking sector and the real
economy for Switzerland via a VAR approach, however it is a single country study that does not
make use of the panel VAR methodology adopted here.

2
informational advantages of the central bank, and in particular whether bank
related information can be used to improve macro forecasts. Peek, Rosengren,
and Tootell (1999, 2003), Romer and Romer (2000) show that incorporat-
ing confidential supervisory information about bank health improves central
bank forecasts of both unemployment and inflation, and that in fact, the Fed-
eral Open Market Committee (FOMC) consider this information when setting
monetary policy. We contribute to this literature by extending our analysis
further and investigating the results uncovered thus far. In particular, our re-
sults indicated an apparent importance of banking sector stability on output
growth. We continue by assessing whether additional information embedded in
our stability index might help to improve output growth forecasts. Using Feds
forecasts data, we show that banking sector stability (instability) results in a
significant underestimation (overestimation) of GDP growth in the subsequent
quarters. This finding indicates that additional information embedded in our
stability measure has the potential to further improve economic forecasts.

The rest of this paper is organized as follows. Section 2 defines our measure
of banking sector stability. Section 3 describes the data used. Section 4 esti-
mates the impact of banking sector stability on output and inflation. Section
5 explores the relationship between banking sector stability and central bank
forecast errors. Section 6 concludes.

2 Stability index for the banking sector

2.1 Banking sector stability: definition and assumptions

We define the instability as the probability of the banking sector becoming

insolvent within the next quarter. Hence, a lower probability corresponds to
greater stability and vice verse. We consider a banking sector to be insolvent
if, at the end of the quarter, the market value of the assets owned by all the
banks of a country is not sufficient to repay its total debt (i.e. if at time t + 1,
the total assets At+1 of the banks are smaller than their total debt Dt+1 ). This
approach makes the implicit assumption that the banking sector of country i
is equal to a single bank, computed as the the aggregate of the balance sheets
of banking institutions in that country. Our measure therefore abstracts from
any comptetiveness considerations.

To compute the default probability of the banking sector i at time t, two

elements are required: the distributions of both the asset value and of the debt
value at the end of the period, conditional on their initial values at time t.
With these two distribution functions, we are able to compute the probability
that, at time t + 1, the asset value falls below the debt value.

3
We assume that the asset value of the banking sector follows a geometric
Brownian motion characterized by:

dAt = µAt dt + σt At dw (1)

where µ is the instantaneous growth rate of assets, σt is their instantaneous

volatility rate and dw is a Wiener process. We also assume that, between t
and t + 1, debt grows at the (continuously compounded) growth rate r, which
implies the following one-period transition function:

Dt+1 = Dt er (2)

2.2 Default probability and distance-to-default for the banking sector

To estimate the default probability of a firm (i.e. of the aggregate banking

sector in our case), Merton (1974) suggests considering the firm’s equity as a
call option on its assets. Merton’s method is based on the idea that if, at debt
maturity, the value of the firm’s assets is smaller than its debt, then the firm
cannot repay its debt and becomes insolvent. In this case, the firm’s equity is
worth zero. If the assets are greater than the debt, then the firm repays its
debt and the market value of the firm is equal to the difference between the
asset value and the debt. Thus Merton’s model states that, at debt maturity
(i.e. t + 1 in our case), the firm’s equity is equal to:

max (At+1 − Dt+1 , 0) (3)

By assuming, as stated in section 2.1 that At is log normally distributed and

that debt is insured, 5 Merton (1974) shows, using the Black-Scholes option
pricing formula, that the initial price Et of the banking sector’s equity is:

Et = At Φ (dt + σt ) − Dt Φ (dt ) (4)

with
ln (At /Dt ) − σt2 /2
dt = (5)
σt
where Φ (·) is the cumulative normal distribution.

Note that Φ (dt ) in equation (4) corresponds to the probability of exercising the
option (i.e. the probability that the assets At+1 are greater than the debt Dt+1 ).
Therefore, the probability of default (i.e. the probability of not exercising the
option) is equal to 1 − Φ (dt ). One can show (see Bichsel and Blum, 2004) that
dt measures how far, in terms of standard deviations, the banking sector is
5
This assumption implies that bank’s bonds will yield the risk free rate even in case of default
for the bank.

4
from its default point (i.e. the point where assets are equal to debt). This is
why dt is often called the distance-to-default.

Concretely, for the rest of our analysis, we will not work with the default
probability directly but we will rather use the distance-to-default dt . Note
that these two measures are equivalent since they are linked by a strictly con-
tinuously decreasing function. A lower distance-to-default always implies a
higher probability of default. The distance-to-default measure has the advan-
tage of being unbounded, which is convenient for empirical estimations. The
default probability on the other hand, is bounded between zero and one.

To compute the distance-to-default with equation (5), we need to know the

asset market value At , its volatility rate σt and the debt Dt . In practice how-
ever, only debt is directly observable. To obtain the asset market value, and its
variance rate we make use of equation (4) which links the unobserved assets
with the observed market capitalization of the banking sector (i.e. its equity
value Et ).

2.3 Estimation of the distance-to-defaults with equity market value

Duan (1994, 2000) shows how to estimate the unobserved evolution of the
asset value and its variance making use of the observed evolution of market
capitalization. In his original paper, Duan (1994, 2000) works with a constant
variance; we extend his method to estimate a time-varying variance.

We start by rewriting equation (1) to get the one-period transition density of

At , which is equal to:

∆ ln At+1 = µ − σt2 /2 + ωt (6)

where ∆ ln At+1 = ln At+1 − ln At and ωt is white noise with variance σt2 . Fur-
thermore, we assume that the variance follows a GARCH(1,1) process char-
acterized by:
σt2 = κ + α²2t−1 + βσt−1
2
(7)
The unknown parameters are θ = {µ, κ, α, β}. Duan (1994, 2000) makes use
of the fact that for any given set of parameters θ, equation (4) is a one-to-one
mapping between Et and At . The parameter set θ can therefore be estimated
by maximizing a log likelihood function defined on Et instead of on At . Duan
(1994, 2000) shows that the log likelihood function takes the form:

5
 TX
−1 ³ ´ T −1 1 TX−1
L (Et , θ) = − ln Φ dˆt + σt − ln (2π) − ln σt2
t=1 2 2 t=1
³ ´2
TX
−1
1 TX
−1 ∆ ln Ât+1 − (µ − σt2 /2)
− ln Ât+1 −
t=1 2 t=1 σt2

where Ât is the unique solution to equation (4) for Et , given that the param-
eters θ and dˆt correspond to dt with Ât instead of At . Note that the three last
terms on the right-hand side of the previous equation would correspond to the
log likelihood functions of a GARCH process if the asset value were directly
observable. The first term on the right-hand side of the equation corrects this
traditional log likelihood function to account for the fact that we must obtain
the asset value At from the observable equity price Et via equation (4). The
maximization of the log likelihood function can be performed with ordinary
maximization techniques.

3 Data

Our dataset consists of debt and market value data between 1980Q1 and
2008Q4 for a sample of 521 banks covering 18 OECD countries. 6 All data are
obtained from Datastream. The debt data are annual and have been trans-
formed into quarterly data by linear interpolation. The quarterly market value
is the minimum of daily market values observed during the quarter. 7

In addition to the data used to calculate the distance-to-default index for each
country, we make use of GDP real growth and inflation between 1980Q1 and
2008Q4. The seasonality component in price movement is removed using the
X12 method in Eviews.

Figure 1 presents the estimated distance-to-defaults for each country. We as-

sume that in each country, banking sector stability is measured by the relative
size of distance-to-default deviations from its own historical mean. In other
words, an unusually unstable banking sector corresponds to a distance-to-
default that is substantially below the country historical mean. The horizontal
lines represent the first and third quartiles of the distance-to-defaults of all

6
We selected countries that joined the OECD before 1975 and that are classified as high-
income countries by the World Bank. These countries are Australia (17 banks), Austria (10 banks),
Belgium (12 banks), Canada (10 banks), Denmark (52 banks), France (47 banks), Germany (32
banks), Greece (21 banks), Italy (50 banks), Ireland (5 banks), Japan (110 banks), Norway (32
banks), Portugal (9 banks), Spain (23 banks), Sweden (7 banks), Switzerland (33 banks), United
Kingdom (14 banks) and United States (37 banks). Finland, Iceland, Luxembourg, Netherlands
and New Zealand have been excluded because we could not get data for at least 5 banks.
7
We additionally experiment by using the average market value observed during the quarter.
However the results are not significantly affected and are therefore not presented here for brevity.

6
countries. These quartiles are used in Section 4.2 to define the thresholds of
stability on which we base our assessment. We define three possible states for
the banking sector: unstable, stable and very stable. We consider a banking
sector to be unstable if its distance-to-default is in the first 20% of all distance-
to-defaults observed internationally over the period 1980 to 2007. Similarly,
we say that the banking sector is very stable if its distance-to-default is in the
last quartile of the distribution. Finally, a banking sector is stable in all others
cases.

Figure 2 shows the number of countries with an unstable banking sector in

each period. We can observe four periods in which at more than third of the
countries simultaneously experience instability in their banking sector: 1987Q4
(Black Monday and the following stock market crisis), 1990Q1 – 1993Q2 (hous-
ing crises in several countries), 1998Q3 (Russian and LTCM crises) and in 2008
(subprime crisis).

4 Impact of banking sector stability real output growth and infla-

tion

In order to assess whether banking sector stability is linked to real output

growth and inflation we adopt a panel autoregression (pVAR) approach with
state-dependent variance. We make use of a pVAR approach predominantly for
the reason that the interactions between financial distress and the real econ-
omy have not been rigorously identified theoretically. Using a VAR therefore
allows us to impose as little a priori theorizing as possible. Moreover, a panel
VAR, when compared to a traditional VAR approach, allows us, by using cross
sectional data, to increase the number of observations and thus the precision
of our estimations. This is particularly useful in our context since episodes of
banking sector instability are relatively rare. The choice of a state-dependent
variance derives from a preliminary analysis showing that the residuals of a
simple pVAR were heteroscedastic. Thus, in addition to studying the impact
of banking sector stability on the levels of GDP growth and inflation, our
framework allows us to explore its impact on both GDP growth and inflation
uncertainty.

The pVAR model has the following specification:

1/2
yit = Ayit−1 + Bxit + µi + Hit εit (8)

where yit is a vector of endogenous variables for country i at time t (i.e. out-
put growth, inflation and banking sector stability), xit is a vector of exogenous
variables (i.e. oil prices), µi is a fixed effect for each country, εit is a vector
of independent error terms normally distributed and Hit is a country-specific

7
 Fig. 1. Estimated distance-to-defaults
Australia Austria Belgium Canada Denmark
32 32 32 32 32

28 28 28 28 28

24 24 24 24 24

20 20 20 20 20

16 16 16 16 16

12 12 12 12 12

8 8 8 8 8

4 4 4 4 4

0 0 0 0 0
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

France Germany Greece Ireland Italy

32 32 32 32 32

28 28 28 28 28

24 24 24 24 24

20 20 20 20 20

16 16 16 16 16

12 12 12 12 12

8 8 8 8 8

4 4 4 4 4

0 0 0 0 0
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

8
Japan Norway Portugal Spain Sweden
32 32 32 32 32

28 28 28 28 28

24 24 24 24 24

20 20 20 20 20

16 16 16 16 16

12 12 12 12 12

8 8 8 8 8

4 4 4 4 4

0 0 0 0 0
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

Switzerland United Kingdom United States

32 32 32

28 28 28

24 24 24

20 20 20

16 16 16

12 12 12

8 8 8

4 4 4

0 0 0
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
 Fig. 2. Number of countries with weak banking sector

14

12

10

0
1980 1985 1990 1995 2000 2005

time-varying covariance matrix, which is function of a set of dependent vari-

ables zt−1 (see below).

To model Hit , we adopt a strategy similar to the conditional constant corre-

lation (CCC) of Bollerslev (1990). The covariance matrix can be decomposed
as:
Hit = Dit Ri Dit (9)
with  
1 φi12 φi13 · · · φi1N
 
 
 φi21 1 φi23 · · · φi2N 

Ri =  . .. .. . . . 
 (10)
 .. . . . .. 
 
φiN 1 φiN 2 φiN 3 · · · 1

where φijk is a constant country-specific correlation coefficient between vari-

ables j and k in yit and
 
σ2 0 0 ··· 0
 it1 
 2 
2
 0 σit2 0 ··· 0 
Dit = 
 .. .. .. . . .. 
 (11)
 .
 . . . . 

2
0 0 0 · · · σitN
2
where σitn is the variance of residuals for country i at time t for the nth
variables of yit . Each variance depends of a set of past variables zit through
the following function:
2
σitn = exp(αin + αn zit−1 ) (12)

9
where αin is a fixed effect specific to each country and αn is a common slope.
Note that zit can contain some or all variables in yit and xit .

The likelihood function of the pVAR given θ is:

T Y
Y I µ ¶
1 1
L= N/2 1/2
exp − ε0it Hit−1 εit (13)
t=1 i=1 (2π) |Hit | 2

where T is the number of periods, I is the number of countries and N is the

number of variables in each vector yit . The log likelihood function is then:
T X
X I
N 1 1
l= − ln(2π) − ln |Hit | − ε0it Hit−1 εit (14)
t=1 i=1 2 2 2

The log likelihood can be estimated with traditional maximization techniques.

We use a multivariate version of the estimation methodology proposed by
Harvey (1976).

The model is estimated using GDP real growth, price level and oil price growth
rate data between 1980Q1 and 2008Q4 collected from Datastream. 8 Note that
we standardize 9 the distance-to-default for each country. With this, we elimi-
nate the difference in distance-to-default level and volatility between countries.
This is equivalent to assuming that, in each country, banking sector stability
is measured by the relative size of distance-to-default deviations from each
country historical mean. In other words, an unusually unstable banking sec-
tor corresponds to a distance-to-default that is unusually below the country
historical mean, in terms of historical standard deviations.

4.1 Linear impact of banking sector stability

4.1.1 Impact on real output growth

Table 1 displays the estimated coefficients of the pVAR model with a lag of
two for the endogenous and the exogenous variables. 10 For clarity, only co-
efficients related to the endogenous and exogenous variables are presented;
fixed effects are omitted. The estimated coefficients show that the relation-
ship between banking sector stability and real output growth is a two way
relationship. Real output growth is important for subsequent stability and
similarly, banking sector stability is important for subsequent output growth.
In particular, our results indicate that higher growth is followed by greater

8
We removed the seasonality component in price movement using the X12 method in Eviews.
9
A standardized variable is equal to its value subtracted from its mean and divided by its
standard deviation.
10
The number of lags has been chosen with the Akaike Information Criterion.

10
 Table 1
Linear impact of banking sector stability
on real output growth
Level Variance
2 2
GDPt DDt log(σGDPt
) log(σDDt
)
GDPt−1 0.2035** 2.2104** -16.2861** -3.7156
GDPt−2 0.1982** 0.1754 -6.8144 11.0961**
DDt−1 0.0011** 0.9671** -0.3239** 0.6499**
DDt−2 -0.0010** -0.0093 0.0060 0.3483**
Oilt 0.0006 -0.0495
Oilt−1 -0.0020** -0.0397
Oilt−2 -0.0011 -0.0526*
*(**) indicates that the coefficient at significant at the 5% (1%) level. Level:
estimated coefficients of equation (8) with GDP growth and banking sector’s
distance-to-default as endogenous variables and oil price yield as exogenous
variable. Variance: estimated coefficients for equation (12) with GDP growth
and banking sector’s distance-to-default as independent variables.

banking sector stability (column 2, row 1) and that banking sector stability
induces growth in the subsequent periods (column 1, rows 3 and 4).

The positive impact of banking sector stability on output growth is observable

from the impulse-response functions of the estimated pVAR (lower left-hand
side panel in Figure 3). Moreover, shocks (both to banking sector stability and
to output growth) are more persistent in their effect on banking sector stability
than on output growth (right-hand side vs. left-hand side panels of Figure 3).
The impulse-response functions also show that the impact of a positive GDP
shock on the improvement of banking sector stability is long lasting.

Table 1 additionally presents the estimated coefficients for the output growth
and banking sector stability variances. The results indicate that both the level
of output growth and banking sector stability have a significant impact on their
own variance. Higher output growth is followed by lower output variance in
the following quarters (column 3, rows 1). In other words, uncertainty about
future output growth is smaller in booms than in recessions. 11 In contrast,
the variance of banking sector stability increases when the banking sectors
distance-to-default is high (column 4, rows 3 and 4). This signifies that un-
certainty about future banking sector stability is higher during periods of
stability. Combined with the positive autocorrelation for the banking sector
described previously, our results imply that an unstable banking sector is more
likely to be followed by a further period of instability, than a stable banking
sector is to be followed by another period of stability.

Table 1 additionally shows that banking sector stability has a negative and
significant impact on output growth variance (column 3, row 3). This means

11
This is in line with the negative link between variance and growth identified by Ramey and
Ramey (1995) and Fatás and Mihov (2003).

11
Fig. 3. Impulse-response function for real output growth and banking
sector stability
Real output after real output shock Bank stability after real output shock
1 0.12
Standard deviation

Standard deviation
0.8
0.1
0.6

0.4
0.08
0.2

0 0.06
0 5 10 15 20 0 5 10 15 20
Quarters Quarters
Real output after bank stability shock Bank stability after bank stability shock
0.08 1
Standard deviation

0.06 Standard deviation

0.8
0.04
0.6
0.02

0 0.4
0 5 10 15 20 0 5 10 15 20
Quarters Quarters

that a stable banking sector is followed by reduced uncertainty about future

output growth. Combined with the positive link between banking sector sta-
bility and output growth, our findings indicate that a stable banking sector
is more likely to be followed by high output growth, than an unstable bank-
ing sector is to be followed by low output growth. Interestingly, we find that
higher uncertainty about the stability of the banking sector generally follows
higher output growth (column 4, row 2). Combined with the positive link
between output growth and banking sector stability, this means that, even-
though banking sectors are on average more stable after a growth in output,
the probability of observing a banking crisis is higher after a period of growth.

Figure 4 documents the behavior of output growth and banking sector stability
variances after shocks. The graph shows the difference between the average
variance with and without shocks. As explained previously, a positive shock
(to either output growth or to banking sector stability) reduces the variance
of output growth (left-hand side panels). A positive shock to banking sector
stability increases its own variance (right-hand side lower panel). Finally, a
shock to output growth increases banking sector stability variance (right-hand
side upper panel).

12
Fig. 4. Impulse-response function for real output growth and banking
sector stability variance
Real output var. after real output shock Bank stability var. after real output shock
0 0.3

−0.05 0.2
Difference

Difference
−0.1 0.1

−0.15 0

−0.2 −0.1
0 5 10 15 20 0 5 10 15 20
Quarters Quarters
Real output var. after bank stability shockBank stability var. after bank stability shock
0 1

−0.05 0.8
Difference

Difference

−0.1 0.6

−0.15 0.4

−0.2 0.2
0 5 10 15 20 0 5 10 15 20
Quarters Quarters

4.1.2 Impact on inflation

Table 2 presents the estimated pVAR with inflation and distance-to-default

as endogenous variables with a lag of 2. 12 It shows that neither inflation
nor banking sector stability has an impact on the other. The results for the
variance equations and for their impulse-response functions (Figure 5) show
that higher inflation increases the uncertainty about next period inflation,
but decreases the variance of distance-to-default after one period. Impulse-
response functions additionally show that the variance of inflation tends to
decrease for a very short period after a positive shock to banking sectors
degree of stability.

4.2 Non linear impact of the banking sector stability

In the previous section we showed that some empirical evidence of a linear

link between banking sector stability and the distribution of output growth
exists. However, some authors have suggested that the main impact of banking
sector stability on the real economy is non linear. In this section, we briefly

12
The number of lag is determined by the Akaike Information Criterion.

13
 Table 2
Linear impact of banking sector stability on infla-
tion
Level Variance
2 2
Inf lationt DDt log(σInf lt ) log(σDDt
)
Inf lationt−1 1.0458** 0.0007 0.0623 0.1427**
Inf lationt−2 -0.1237** -0.0080 0.1143* -0.1910**
DDt−1 -0.0487 0.9765** 0.2286** 0.5770**
DDt−2 0.0330 -0.0254 -0.2172** 0.3480**
Oilt 0.5015** -0.0428
Oilt−1 0.1309* -0.0097
Oilt−2 0.3227** -0.0280
*(**) indicates that the coefficient at significant at the 5% (1%) level. Level: estimated co-
efficients of equation (8) with inflation and banking sector’s distance-to-default as endoge-
nous variables and oil price yield as exogenous variable. Variance: estimated coefficients
for equation (12) with inflation and banking sector’s distance-to-default as independent
variables.

Fig. 5. Impulse-response function for inflation and banking sector sta-

bility variance
Inflation var. after inflation shock Bank stability var. after inflation shock
0.2 0.15

0.15 0.1
Difference

Difference

0.1 0.05

0.05 0

0 −0.05
0 5 10 15 20 0 5 10 15 20
Quarters Quarters
Inflation var. after bank stability shock Bank stability var. after bank stability shock
0.05 0.8

0 0.6
Difference

Difference

−0.05 0.4

−0.1 0.2

−0.15 0
0 5 10 15 20 0 5 10 15 20
Quarters Quarters

investigate this possibility with a simple model.

Concretely, we check whether there is an asymmetric reaction of real output

growth and inflation to banking sector stability. Essentially, we assess whether
real output and inflation distributions are different in periods following differ-
ent state of banking sector stability. As per section 3, we define three states for

14
 Table 3
Non linear impact of banking sector stability on
real output growth
Level Variance
2 2
GDPt DDt log(σGDPt
) log(σDDt
)
GDPt−1 0.1969** 4.0539** -17.0826** -2.2483
GDPt−2 0.2115** 3.2440** -8.3040* 7.0877
D1,t−1 -0.0019** -0.6422** 0.1445 -0.3709**
D1,t−2 0.0016** -0.3758** 0.3846** -1.2010**
D2,t−1 0.0008 1.1228** -0.2373 0.3551**
D2,t−2 -0.0004 0.4656** -0.3045* 0.1747**
Oilt 0.0007 -0.0209
Oilt−1 -0.0019** 0.0067
Oilt−2 -0.0012 0.0214
*(**) indicates that the coefficient at significant at the 5% (1%) level. Level: estimated co-
efficients of equation (8) with inflation and banking sector’s distance-to-default as endoge-
nous variables and oil price yield as exogenous variable. Variance: estimated coefficients
for equation (12) with inflation and banking sector’s distance-to-default as independent
variables.

the banking sector: unstable, stable and very stable. We say that the banking
sector is unstable if its (standardized) distance-to-default is in the first 20

Similarly to the linear estimations, we make use of a panel VAR but we replace
the lagged distance-to-default of the right-hand side of equation (8) by two
dummy variables: D1,t , is equal to one when the banking sector is unstable
and zero otherwise; and D2,t , is equal to one when the banking sector is very
stable and zero otherwise.

Table 3 presents the estimated coefficients of the new pVAR. The results for
the dummy variables are particularly interesting. They indicate that only un-
stable banking sectors have a significant (and negative) impact on real output
growth. Particularly stable banking sectors have no impact on real output
growth. In contrary, both unstable and very stable banking sectors have a
significant impact on real output growth variance.

Table 4 presents the estimated coefficients of the non linear pVAR with infla-
tion. The results are less clear than for the output growth, but it seems that
unstable banking sector increase slightly inflation (column 1, row 4).

The non linear model provides some interesting insights into the results from
the linear model for output growth. They show that the linear links uncovered
are predominantly driven by periods of banking sector instability rather than
by a smooth and continuous link between banking sector stability and the real
economy.

15
 Table 4
Non linear impact of banking sector stability on
inflation
Level Variance
2 2
Inf lationt DDt log(σInf lt ) log(σDDt
)
Inf lationt−1 1.0497** -0.0173 0.0654 0.0386
Inf lationt−2 -0.1259** 0.0043 0.1034 -0.0815
D1,t−1 -0.0757 -0.6516** 0.2017 -0.3533**
D1,t−2 0.0985* -0.3387** -0.0224 -1.0098**
D2,t−1 -0.0485 1.1281** -0.0161 0.2686*
D2,t−2 0.0427 0.4536** 0.0456 0.2236
Oilt 0.5162** -0.0189
Oilt−1 0.1336* 0.0421
Oilt−2 0.3342** 0.0847
*(**) indicates that the coefficient at significant at the 5% (1%) level. Level: estimated co-
efficients of equation (8) with inflation and banking sector’s distance-to-default as endoge-
nous variables and oil price yield as exogenous variable. Variance: estimated coefficients
for equation (12) with inflation and banking sector’s distance-to-default as independent
variables.

4.3 Expanding the set of macroeconomic variables

The model estimated in Section 4.1 is very simplistic since it makes the hy-
pothesis that real output is function of its lags, of banking sector stability and
of oil prices only. In reality, several other economic variables have an impact
on output growth. In this section, we extend the set of exogenous macroeco-
nomic variables included in the pVAR to capture their influence on real output
growth. More precisely, we add consumption growth rate (∆Ct ), short term
interest rates (ST IRt ), money growth rate(∆M 2t ) growth and investments
(∆Kt ) in the sample. 13 We chose to present these results in a separate sec-
tion and not directly in the main section because the variables that we include
are not available for all the countries studied or for the same period. Indeed,
after adding these variables, the sample is reduced to 15 countries instead of
18 14 and the number of observation shrinks from 1757 data points to 1052.

Table 5 shows that real output growth and distance-to-default are still func-
tion of their own lag. As previously, real output growth is also dependent on
banking sector stability. However, and contrary to the results in Section 4.1,
banking sector stability does not depend on real output growth anymore. It is
still a function of the macroeconomic environment as banking sector stability
decrease for higher short term interest rates and higher investments (column

13
We chose variables that are likely to influence transitory business cycles of GDP (see e.g. Stock
and Watson, 1999) and not long term growth rates (like e.g. human capital, political environment,
etc...). The reasons for that is mainly that we are not interested in the long term structural influence
of the banking sector on growth but more in the transitory impact of the fluctuation of its stability.
14
Australia, Greece and Sweden are missing from the initial sample.

16
 Table 5
Linear impact of banking sector stability
on real output growth : extended sample
Level Variance
2 2
GDPt DDt log(σGDPt
) log(σDDt
)
GDPt−1 0.2313** 2.2489 -12.5079* -14.8445**
GDPt−2 0.2032** -0.9067 -9.5655 0.3829
DDt−1 0.0008** 0.8984** -0.2344** 0.6176**
DDt−2 -0.0008* -0.0081 -0.0304 0.3826**
Oilt 0.0011 0.1029*
Oilt−1 -0.0011 -0.0155
Oilt−2 -0.0007 0.0074
∆M 2t−1 0.0136* 0.3182
ST IRt−1 -0.0003** -0.0200**
∆Ct−1 0.0848* 0.9168
∆Kt−1 -0.0000** -0.0000**
*(**) indicates that the coefficient at significant at the 5% (1%) level. Level:
estimated coefficients of equation (8) with GDP growth and banking sector’s
distance-to-default as endogenous variables and oil price yield as exogenous
variable. Variance: estimated coefficients for equation (12) with GDP growth
and banking sector’s distance-to-default as independent variables.

2, rows 9 and 11). Table 5 also shows that real output growth variance is
still a negative function of banking sector stability - i.e. uncertainty about
real output growth increases with banking sector instability. The introduction
of the new variables changes the results for distance-to-default variance: the
uncertainty about banking sector stability decreases with higher real output
growth.

5 Banking sector stability and central bank forecast errors

The link between banking stability and economic activity is of particular inter-
est to policy makers which base their monetary policy decisions on economic
forecasts. Our estimations above highlighted the importance of banking sector
stability on output growth, however, they were unable to identify any signif-
icant relationship between stability and inflation. We therefore extend our
analysis on banking sector stability and GDP growth by assessing whether
additional information embedded in our stability index might help to improve
output growth forecasts. We base our analysis on the United States since fore-
cast data is publicly available on the Fed website back to 1965 15 , up to a five
year delay. For comparability with our previous estimations our dataset that
consists of quarterly data from 1980 to 2001.

Despite econometric models playing an important role in Fed projections for

15
http://www.philadelphiafed.org/econ/forecast/greenbook-data/index.cfm.

17
 Table 6
Correlations: distance to default with fore-
cast errors
DD(C) DD(L1) DD(L2) DD(L3) DD(L4)
current -0.51*** -0.50*** -0.47** -0.43** -0.41**
one -0.43 -0.41 -0.37 -0.34 -0.28
two -0.38** -0.34** -0.30** -0.29** -0.23*
three -0.34* -0.29** -0.26** -0.28** -0.04***
five -0.10 -0.37* -0.39 -0.39 -0.34*
six -0.39 -0.29 -0.31 -0.29 -0.25
seven -0.29 -0.31 -0.32* -0.31 -0.27
eight -0.29 -0.41 -0.44 -0.44 -0.41*
Note: *(**) denotes significance at the 5% (1%) level. Forecast errors are cal-
culated as the difference between actual and predicted GDP growth. On the
vertical axis, current, one... refer to the current forecast, the one-period ahead
forecast etc. DD(C), DD(L1), DD(L2) denote the current stability, stability with
a one period lag, stability with a two period lag.

macroeconomic growth, they serve only as part of an input into the forecast
process. Judgement based intervention, incorporating additional information
not grasped by the models, additionally play a vital role (see Reifschneider,
Stockton, and Wilcox, 1997). It is, however, unclear what information is in-
corporated and whether such judgment based adjustments to forecasts are
efficient, i.e. whether they for example make efficient use of all information
contained in banking sector variables. Since economic activity and interest
rates affect financial sector risks, and in turn, the financial sector affects the
real economy, it is possible that considering the state of the banking sector in
a systematic way would improve forecasts. We therefore investigate whether
Fed growth forecasts are indeed making use of information contained in our
measure of banking sector stability.

We start by computing simple correlations between the current distance to de-

fault and the period ahead forecasts. The correlations are presented in Table 6.
Correlations between the stability indicator and the forecast errors are signif-
icant for current forecasts as well as for two- and three-period ahead forecasts
indicating that some degree of association between these two measures exists.
Simple correlations, however, make no a priori assumptions as to whether one
variable is dependent on the other. We therefore proceed to describe the de-
pendence of forecast errors on the banking sector stability; implicitly assuming
that there is a one-way causal effect from the degree of stability to the forecast
errors, regardless of whether the path of effect is direct or indirect.

Our results are presented in Table 7. We find that banking sector stability
has a positive and significant impact on the two-and three-period ahead Fed
forecasts. Since the forecast error is calculated as the difference between the
actual value and the forecasted rate, the result indicates that banking sector
stability today results in a significant underestimation of GDP growth the

18
 Table 7
Estimation: distance to default and Fed forecast errors
Dependent one-period two-period three-period four-period
variable: ahead forecast ahead forecast ahead forecast ahead forecast

LINEAR ESTIMATIONS
distance to default 0.19 0.18** 0.12** 0.12*

NON LINEAR ESTIMATIONS

D1 -0.10 -0.23** -0.24** -0.26
D2 0.21 0.12 0.18 0.28
Note: *(**) denotes significance at the 5% (1%) level. Forecast errors are calculated as the
difference between actual and predicted GDP growth. D1 and D2 are dummies capturing
periods of instability and stability respectively.

subsequent quarters. Essentially, banking sector stability brings more growth

than expected. This finding indicates that additional information embedded in
our stability index measure has the potential to improve Fed growth forecasts.
We additionally experiment with asymmetric effects in this regard, defining
periods of instability, stability and high stability as per section 3. The results
are presented in the lower panel of Table 7. The results from the non linear
estimation further confirm our previous finding that for the two-period (and
three-period) ahead forecasts, instability drives the results.

6 Conclusion

In this paper, we explore the relationship between banking sector stability

and the subsequent evolution of real output growth and inflation. Capturing
the stability of a banking sector by building an index based on banking sector
distance to default, we adopt a panel VAR methodology to assess both linear
and non linear relationships between our variables of interest. We show that
banking sector stability appears to be an important driver of GDP growth.
Periods of stability are generally followed by an increase in real output growth,
while instability corresponds with subsequent periods of reduced growth. We
also show that a stable banking sector reduces real output growth uncertainty.
In addition, we explore the possibility of an asymmetric impact of banking sec-
tor stability on the real economy. To do this we distinguish between unstable,
stable and very stable periods relative to both historical and international
average. We find that the relationship is asymmetric in nature, in that it is
driven predominantly by periods of instability.

We further extend the analysis to assess whether additional information em-

bedded in our stability index might help to improve central bank growth fore-

19
casts. Focussing on data from the Fed, we show that output forecast errors
are correlated with our stability measure. Our findings indicate that banking
sector stability (instability) results in a significant underestimation (overes-
timation) of GDP growth in the subsequent quarters. This result is in line
with the notion that additional information, embedded in our stability index
measure, has the potential to further improve economic forecasts.

Our findings have several important implications for policy makers. First, we
show that for a sample of 18 countries, banking sector stability appears to be
an important driver of GDP growth in subsequent periods, highlighting the
need for greater attention to be paid to banking sector soundness in the imple-
mentation of economic policy. In addition, we find that additional information
contained in our banking sector measure could help forecasters to reduce fore-
cast errors, highlighting the need for policy makers to consider banking sector
measures in their forecast models.

References

Allenspach, N., and J. Perrez (2008): “The interaction between the Swiss
banking sector and the macroeconomy: results from a VAR-analysis,” Swiss
National Bank, mimeo.
Aspachs, O., C. A. Goodhart, D. P. Tsomocos, and L. Zicchino
(2007): “Towards a measure of financial fragility,” Annals of Finance, 3,
37–74.
Benink, H., and G. Benston (2005): “The future of banking regulation
in developed countries: lessons from and for europe,” Financial Markets,
Institutions and Instruments, 14, 289–328.
Bichsel, R., and J. Blum (2004): “The relationship between risk and cap-
ital in Swiss commercial banks: a panel study,” Applied Financial Eco-
nomics, 14, 1–7.
Bollerslev, T. (1990): “Modeling the coherence in short-run nominal ex-
change rates: A multivariate generalized ARCH model,” Review of Eco-
nomics and Statistics, 72, 498–505.
Boyd, J. H., S. Kwak, and B. Smith (2005): “The real output losses asso-
ciated with modern banking crises,” Journal of Money, Credit and Banking,
37(4), 977–999.
Dell’Ariccia, G., E. Detragiache, and R. Rajan (2008): “The real
effect of banking crises,” Journal of Financial Intermediation, 17, 89–112.
Demirgüç-Kunt, A., and E. Detragiache (1997): “The determinants of
banking crises in developing and developed countries,” IMF Staff Papers,
45, 81–109.
(2005): “Cross-country empirical studies of systemic bank distress :
a survey,” IMF Working Paper, 05/96.

20
Demirgüç-Kunt, A., and V. Maksimovic (1998): “Law, finance, and firm
growth,” Mathematical Finance, 53, 2107–2137.
Deutsche Bundesbank (2006): Financial Stability Review.
Duan, J.-C. (1994): “Maximum likelihood estimation using price data of the
derivative contract,” Mathematical Finance, 4, 155–167.
(2000): “Correction: maximum likelihood estimation using price data
of the derivative contract,” Mathematical Finance, 10, 461–462.
European Central Bank (2006): “Country-level macro stress-testing
practices,” Financial Stability Report, pp. 147–154.
Fatás, A., and I. Mihov (2003): “The Case for restricting fiscal policy
discretion,” Quarterly Journal of Economics, 118, 1419–1447.
Goodhart, C. A., P. Sunirand, and D. P. Tsomocos (2006): “A model
to analyse financial fragility,” Economic Theory, 27.
Gupta, P. (2005): “Aftermath of banking crises: Effects on real and monetary
variables,” Journal of International Money and Finance, 24, 675–691.
Harvey, A. C. (1976): “Estimating regression models with multiplicative
heteroscedasticity,” Econometrica, 44(3), 461–465.
Hilbers, P., I. Otker-Robe, C. Pazarbasioglu, and G. Johnsen
(2005): “Assessing and managing rapid credit growth and the role of su-
pervisory and prudential policies,” IMF Working Paper, 05/151.
Hogart, G., R. Reis, and V. Saporta (2002): “Costs of banking system
instability: some empirical evidence,” Journal of Banking and Finance, 26,
825–855.
Kaminsky, G., and C. Reinhart (1999): “The twin crises: the causes of
banking and balance of payments problems,” American Economic Review,
89, 473–500.
King, R. G., and R. Levine (1993a): “Finance and growth: Schumpeter
might be right,” Journal of Financial Economics, 108, 717–738.
(1993b): “Finance, entrepreneurship, and growth: theory and evi-
dence,” Journal of Monetary Economics, 32, 513–542.
Kroszner, R. S., L. Laeven, and D. Klingebiel (2007): “Banking crises,
financial dependence, and growth,” Journal of Financial Economics, 84,
187–225.
Levine, R. (1997): “Financial development and economic growth: views and
agenda,” Journal of Economic Literature, 35, 688–726.
(1998): “The legal environment, banks, and long -run economic
growth,” Journal of Money, Credit, and Banking, 30, 596–620.
(2001): “International financial liberalization and economic growth,”
Review of International Economics, 9, 688–702.
(2004): “Finance and growth: theory, evidence, and mechanisms,” in
Handbook of Economic Growth, ed. by P. Aghion, and S. Durlauf. Elsevier.
Levine, R., N. Loayza, and T. Beck (2000): “Financial intermediation
and growth: causality and causes,” Journal of Monetary Economics, 35,
31–77.
Merton, R. (1974): “On the pricing of corporate debt: the risk structure of

21
 interest rates,” Journal of Finance, 29, 449–470.
Peek, J., E. S. Rosengren, and G. M. B. Tootell (1999): “Is bank
supervision central to central banking?,” Quarterly Journal of Economics,
114, 629–655.
(2003): “Does the Federal Reserve possess an exploitable informa-
tional advantage?,” Journal of Monetary Economics, 50, 817–839.
Rajan, R. G., and L. Zingales (1986): “Financial dependence and
growth,” American Economic Review, 88, 559–586.
Ramey, G., and V. A. Ramey (1995): “Cross-country evidence on the link
between volatility and growth,” American Economic Review, 85, 1138–1151.
Rancière, R., A. Tornell, and F. Westermann (2008): “Systemic
crises and growth,” Quarterly Journal of Economics, 123, 359–406.
Reifschneider, D. L., D. J. Stockton, and D. W. Wilcox (1997):
“Econometric models and the monetary policy process,” Carnegie-Rochester
Conference Series on Public Policy, 47, 1–37.
Romer, C. D., and D. H. Romer (2000): “Pricing risk-adjusted deposit
insurance: an option based model,” American Economic Review, 90, 429–
457.
Serwa, D. (2007): “Larger crises cost more: impact of banking sector insta-
bility on output growth,” MPRA Paper, 5101.
Stock, J. H., and M. W. Watson (1999): “Business cycle fluctuations in
U.S. macroeconomic time series,” in Handbook of Macroeconomics, ed. by
J. Taylor, and M. Woodford. Elsevier.

22