trend, the table assumes that the discount
rate equals the trend growth rate. In per
cent. 2 The dating of crises is not the same
across studies. If several years are provided,
the references for the crisis dating used in
the studies are (a) Laeven and Va
in Table A2.1 reports the impact of liquidity
and capital ratios on the likelihood that two
or more banks default, conditional on a very
severe macroeconomic shock. In this
environment, higher capital ratios clearly
have large benefits. Increasing capital
emerging markets. Hoggarth et al (2002)
and Cerra and Saxena (2008) distinguish
between high- and low-income countries.
Using this classification, the IMF (2009)
does not find significant differences. Results
shown in Table A1.1 are based on a
classificat
results in the literature are surprisingly
consistent. Studies in the first two groups,
which compare GDP at the beginning of the
crisis to the trough or to the point when its
growth recovers, find a drop of around 10%
relative to precrisis GDP. Costs ten
allowing for permanent effects Cecchetti et
al Laeven and Valencia Hoggarth et al
Cecchetti et al Haugh et al Boyd et al (M 2)3
Boyd et al (M 1)3 Haldane4 Argentina 1980
14.1 10.8 25.9 44.5 Argentina 1989 12.1
10.7 16.1 16.2 Argentina 1995 6.1 7.1 5.8
5.2
episode) chosen as reference points for the
measurement of costs. There are four types
of approach. The first type focuses on the
period between the GDP peak prior to the
crisis and the subsequent through after the
onset of the crisis (time between A and
that assess the costs of crises over a
specified period, hence implicitly assuming
that effects are only transitory (the third
group in the table), the median cumulative
loss estimate is 19%. Studies that explicitly
allow for permanent effects (the last t
trend would imply higher costs of crises. 41
Some studies fix the length of crises at four
years (eg Laeven and Valencia (2008) or
determine it on the basis of expert
judgment (eg Hoggarth et al (2002) or
allow it to be determined endogenously by
assuming
expenses amount to 2.1%. Personnel
expenses represent close to 43% of total
operating expenses. Net income (or ROA) is
0.6%, implying that the average return on
equity (ROE) is 14.8%. The average historical
tax rate is 33.2%. Calculating the impact of
hig
non-G10 members. Haldane (2010) is the
only study estimating costs for the current
crisis. For world GDP, estimates range from
90% to 350% relative to 2009 output (Table
A1.1). Results are even larger for the United
Kingdom, where the upper estimate
excee
calculation of the cumulative costs would
entail some form of discounting (see
discussion below). Table A1.1 lists different
studies grouped along these two
dimensions together with their estimates of
the costs of the average crisis. The first
three group
the non-bank sector, the estimated costs
will be lower. Similarly, there are a number
of factors that could reduce the net
benefits: The existing literature, which is
the basis for this reports estimates of the
costs of banking crises, may overestimate
th
and Mourougane (2009) or deviations of
actual output (eg Cerra and Saxena (2008),
Turini et al (2010), IMF (2009), Furceri and
Zdzienicka (2010). The former studies find
a permanent drop of 2% after a banking
crisis, while the latter find effects of the
o
approximately the difference between trend
and actual levels of output at point C.37
Generally, trend growth in these studies is
calculated as the historical average growth
over a period that ranges (depending on the
study) between three and 10 years prio
costs relative to pre-crisis GDP, especially if
crises are not too long. Hoggarth et al
(2002) prove mathematically that the
difference is actually underestimated for
crises lasting longer than two years, as the
approach does not recognise the reduction
i
minimum is generally zero. A closer look
indicates that this may be driven by
definitions of what constitutes a systemic
banking crisis. For example, some studies
assume that Canada had a banking crisis in
1983. While two small banks failed, experts
at th
indicate an average probability of a systemic
crisis of 4.1% when the capital ratio is 7%
and there is no change in any of the
liquidity ratios from their pre-crisis levels.
This is at the low end of the historical
average of 45%.Within the framework of
t
Nebraska and West Virginia following the US
banking crisis in 1894. While no banks failed
in West Virginia, Nebraska experienced a
relatively high failure rate. Controlling for
other factors, the author shows that
Nebraska grew on average 1% less per year
finite.41 The second and last group of
studies does allow for the possiblity of
permanent effects. Boyd et al (2005) use
two methods of calculating long-run costs
after a crisis. The first method is more
conservative and uses only actual GDP for
the count
be raised by more in order to achieve a
given net benefit. Shifting of risk into the
non-regulated sector could reduce the
financial stability benefits. The results of
the impact of regulatory requirements on
lending spreads are based on aggregate
balance
disaggregate estimates). An assessment of
the long-term economic impact of the new
regulatory framework 35 Table A1.1 Cost of
a banking crisis relative to pre-crisis GDP1
Results reported in the literature Study
Cumulative losses Mean Min Max Industrial
e
correlated. Correlations are based on
Moodys KMV estimate of the institutions
asset-return correlations. In contrast to the
BoE model and the baseline version of
Tarashev and Zhu, which assume that
shocks to banks assets are normally
distributed, the mode
minimise the negative effects on output. In
the absence of such intervention, the
average costs of banking crises are likely to
be significantly higher. In addition, the
discount rate used to estimate the present
value of the multi-year cost of crises is
control more formally for this problem. Two
studies (Bordo et al (2001), Haugh et al
(2009) compare the output costs of normal
recessions with costs of banking crises and
show that the latter are around 3.5 to 4
times larger. Hoggarth et al (2002) try to
correlated; and (ii) direct balance sheet links
between banks, through which the failure of
one bank can cause the failure of other
institutions. The model is calibrated using
data for the five largest UK banks, with a
systemic crisis defined as the joint
variable and/or other explanatory variables,
as well as a dummy that flags the beginning
of a banking crisis. The dummy variable
allows the simulation of impulse response
functions as shown in Graph A1.1. The last
two groups of studies summarised in Table
NIESR model is as follows: 2 )24.0 Cbr3
08.0_11.034.0()(Prob Rhpg1 LiqA 1
Levfcrises Lev is the ratio of total capital
over total assets, A_Liq is the ratio of cash
and balances with the central bank plus
securities over total assets, Rhpg is real
house