Credit growth, the yield curve

and financial crisis prediction:

Evidence from a machine learning approach

Kristina Bluwstein Sujit Kapadia Özgür Şimşek

Marcus Buckmann European Central Bank University of Bath
Andreas Joseph
Miao Kang
Bank of England

Use of Machine Learning Algorithms

AEA, San Diego, CA, 3 January 2020

Disclaimer: The expressed views are those of the authors and not necessarily those of the Bank of England, the European Central Bank. 1
Motivation: Cost and consequences of economic crises

• Financial crises can have severe social, economic and political consequences
• Policy makers would like to minimise these costs or avoid them altogether
• Policy tools, e.g. macropru, could stabilise system if implemented early enough
• Timely and accurate prediction methods needed
• And, understanding of the underlying economic mechanisms

2
Motivation: Cost and consequences of economic crises

• Financial crises can have severe social, economic and political consequences
• Policy makers would like to minimise these costs or avoid them altogether
• Policy tools, e.g. macropru, could stabilise system if implemented early enough
• Timely and accurate prediction methods needed
• And, understanding of the underlying economic mechanisms
⇒ Our paper addresses these points using machine learning (ML) for financial crisis
prediction 2
Preview of main results

• ML models outperform benchmark logit in out-of-sample prediction and

forecasting evaluations
• Shapley value framework enable well-defined inference (Joseph, 2019)
• Small number of factors explain majority of model output:
• Credit growth and flat/negative slope of the yield curve at low nominal rates
Story: search-for-yield in low-interest rate low-returns environment
• Global factors (also credit growth & slope)
Story: shared narrative in coupled economic/financial system
⇒ Global yield curve slope new indicator with greatest robustness across long
sample

3
Related literature in financial crisis analysis

• General/historic: Minsky (1977); Kindleberger (1978); Bordo et al. (2001);

Laeven and Valencia (2008); Reinhart and Rogoff (2009); Cecchetti et al. (2009)

• Credit: Borio and Lowe (2002); Drehmann et al. (2011); Schularick and Taylor
(2012); Aikman et al. (2013)

• Yield curve (not too extensive): Babeckỳ et al. (2014); Joy et al. (2017);
Vermeulen et al. (2015)

• Global factors: Alessi and Detken (2011); Duca and Peltonen (2013);
Cesa-Bianchi et al. (2018)

• Machine learning: Ward (2017); Alessi and Detken (2018); Beutel et al. (2018)

4
Machine Learning (ML) approach

• Statistical toolbox of non-linear & non-parametric models mostly originating

from computer science with a focus on prediction

• Today supervised learning: Universal approximators minimising an error function

of the form
Ex ||y − fˆ(θ)kp
 

• Models we compare:
• logistic regression (benchmark)
• support vector machines (SVM)
• artificial neural networks
• tree models (decision tree, random forests & “extreme trees”)

• Shapley value and regression framework for statistical inference

5
Pros & Cons of ML relative to econometric approach

Advantages

• Often higher accuracy

• Lower risk of misspecification

• Return richer information set

6
Pros & Cons of ML relative to econometric approach

Advantages Disadvantages

• Often higher accuracy • Higher model complexity (“black box

critique”)
• Lower risk of misspecification
• Less analytical guarantees, e.g. risk of
• Return richer information set
overfitting

• Often larger data requirement

6
Jordà-Schularick-Taylor Macrohistory Database

Observations
1870
• 17 developed countries, annual data between 1870 1880
1890
and 2016 1900
1910
• 92 crisis episodes
1920

• 20+ potential indicators 1930

Year
1940
1950
1960
1970
1980
1990
2000
2010

0 2 4 6 8 10
Number of crises

7
Jordà-Schularick-Taylor Macrohistory Database

Observations Subset of variables we use

• 17 developed countries, annual data between 1870 • Non-financial credit
and 2016 • Rates, yield curve
• 92 crisis episodes • Debt service ratio
• 20+ potential indicators • Current account balance
• Stock Prices
• CPI
• Consumption
• Investment
• Broad money
• Public debt
7
Empirical approach

Baseline approach (extensive robustness checks):

• Target: Predict a crisis one and two years in advance (policy space)
• Transformation: 2-year ratio changes or growth rates (sustainability/stationarity)
• Global variables for credit & slope of the yield curve
• Cleaning: Exclude crisis and post-crisis period (5 years), world wars and
1933–1938
Modelling
• Bootstrapped & averaged models (bagging)
• Out-of-sample evaluation: Nested cross-validation & expanding window
forecasting
8
Out-of-sample performance in the ROC space

1.0
0.8
0.6
Hit rate
0.4
0.2
0.0

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
Linear baseline

1.0
0.8
0.6
Hit rate
0.4
0.2
0.0

Logistic regression

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
+ Decision trees

1.0
0.8
0.6
Hit rate
0.4
0.2

Logistic regression
0.0

Decision tree

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
+ Neural network

1.0
0.8
0.6
Hit rate
0.4
0.2

Logistic regression
Decision tree
0.0

Neural network

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
+ SVM

1.0
0.8
0.6
Hit rate
0.4
0.2

Logistic regression
Decision tree
Neural network
0.0

SVM

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
+ Random forest

1.0
0.8
0.6
Hit rate
0.4
0.2

Logistic regression
Decision tree
Neural network
SVM
0.0

Random forest

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
The winner is: Extremely randomized trees

1.0
0.8
0.6
Hit rate
0.4
0.2

Logistic regression
Decision tree
Neural network
SVM
Random forest
0.0

Extreme trees

0.0 0.2 0.4 0.6 0.8 1.0

False alarm rate

9
Area under the curve (AUC) performance

Extreme trees 0.870

Random forest 0.855
SVM 0.832
Neural net 0.829
Logistic regression 0.822
Decision tree 0.759
100 replications of 5-fold cross-validation. Standard errors not shown but consistently below 0.002.

What’s the meaning of this differences?

⇒ Aiming at a 80% true positive rate, extreme trees reduce the number
of false positives by 41% (32%/367 → 19%/219) compared to the logistic regression.
10
Prediction summary for all countries across time (extreme trees)

United States
Sweden
Portugal
Norway
Netherlands
Japan
Italy
United Kingdom
France
Finland
Spain
Denmark
Germany
Switzerland
Canada
Belgium
Australia
1872
1877
1882
1887
1892
1897
1902
1907
1912
1917
1922
1927
1932
1937
1942
1947
1952
1957
1962
1967
1972
1977
1982
1987
1992
1997
2002
2007
2012
11
Prediction summary for all countries across time (extreme trees)

United States
Sweden
Portugal
Norway
Netherlands
Japan
Italy
United Kingdom
France
Finland
Spain
Denmark
Germany
Switzerland
Canada
Belgium
Australia
1872
1877
1882
1887
1892
1897
1902
1907
1912
1917
1922
1927
1932
1937
1942
1947
1952
1957
1962
1967
1972
1977
1982
1987
1992
1997
2002
2007
2012
11
Prediction summary for all countries across time (extreme trees)

Correct crises
Missed crises

United States
Sweden
Portugal
Norway
Netherlands
Japan
Italy
United Kingdom
France
Finland
Spain
Denmark
Germany
Switzerland
Canada
Belgium
Australia
1872
1877
1882
1887
1892
1897
1902
1907
1912
1917
1922
1927
1932
1937
1942
1947
1952
1957
1962
1967
1972
1977
1982
1987
1992
1997
2002
2007
2012
11
Prediction summary for all countries across time (extreme trees)

Correct crises
Missed crises
False alarms

United States
Sweden
Portugal
Norway
Netherlands
Japan
Italy
United Kingdom
France
Finland
Spain
Denmark
Germany
Switzerland
Canada
Belgium
Australia
1872
1877
1882
1887
1892
1897
1902
1907
1912
1917
1922
1927
1932
1937
1942
1947
1952
1957
1962
1967
1972
1977
1982
1987
1992
1997
2002
2007
2012
11
Prediction summary for all countries across time (extreme trees)

Correct crises
Correct non-crises
Missed crises
False alarms
United States
Sweden
Portugal
Norway
Netherlands
Japan
Italy
United Kingdom
France
Finland
Spain
Denmark
Germany
Switzerland
Canada
Belgium
Australia
1872
1877
1882
1887
1892
1897
1902
1907
1912
1917
1922
1927
1932
1937
1942
1947
1952
1957
1962
1967
1972
1977
1982
1987
1992
1997
2002
2007
2012
11
Shapley values for variable importance

Game Theory Machine Learning

N Players Predictors
fˆ/ŷ Collective payoff Predicted value for one observation
S Coalition Predictors used for prediction
Source Shapley (1953) Strumbelj and Kononenko (2010)
Lundberg and Lee (2017)

Model Shapley decomposition:

|S|!(|N|−|S|−1)! ˆ
φSk = ∪ {k} − fˆ(S)]
P
S⊆N\k |N|! [f (S

ΦS fˆ(xik ) = φ0 + m S

P
k=1 φik
Intuitive example 12
Model explanations using Shapley decompositions: high agreement
0.05 0.10 0.15 0.20 0.25

Extreme trees Key indicators:

Random forest
Mean absolute Shapley values

Logistic regression
SVM
Neural network • Domestic credit (Schularick and
(normalized)

Taylor, 2012; Aikman et al.,

2013)
• Global credit (Alessi and Detken,
2011; Cesa-Bianchi et al., 2018)
0.00

• Domestic slope (Babeckỳ et al.,

e

y
e

on
dit

t
o

bt
it

et

nt
en
CP

ne
p
ed
op

ati

2014; Joy et al., 2017)

de

ark

u
slo

cre

pti

tm

co
mo
r
l cr
l sl

um
ce

blic

ac
es
tic

stic
ba

ba

ad
rvi

ck
ns

Inv
s

Pu

nt
Glo

Glo

me

• Global slope (new finding)

me

se

Bro

Sto
Co

rre
Do

Do

bt

Cu
De

13
Extreme trees model Shapley value decomposition

Domestic slope United States

0.1 0.2 0.3 0.4 0.5 0.6

Domestic credit
Global credit
Global slope
Remaining predictors
Predicted value
Shapley contribution

Model mean
Threshold (80% Hit rate)
-0.1

1875
1880
1885
1890
1895
1900
1905
1910
1915
1920
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
14
Extreme trees model Shapley value decomposition

Domestic slope Italy

0.1 0.2 0.3 0.4 0.5 0.6

Domestic credit
Global credit
Global slope
Remaining predictors
Predicted value
Shapley contribution

Model mean
Threshold (80% Hit rate)
-0.1

1875
1880
1885
1890
1895
1900
1905
1910
1915
1920
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
15
Non-linearity of extreme trees for global credit

Global credit

2
Rdegree = 1 = 0.4
2
Rdegree = 3 = 0.92
0.3

Crises • ML models identify strong

Non-crises non-linearities
Shapley values
0.2

• Importantly, these are not known a

priori
0.1

• Directions of associations match those

0.0

in the linear model

0.00 0.05 0.10

Predictor values
16
A closer look at the slope of the yield curve
0.2 0.4 0.6 0.8 1.0

Nominal short-term rate

+1 SD (8.94) • Flat or inverted yield curve slope
Predicted probability

Mean (5.26) increases predicted crisis probability

- 1 SD (1.58)
substantially

• Low nominal short-term rates give

stronger interaction effect
0.0

⇒ Likely search-for-yield behaviour

-4 -2 0 2 4 6 8
Domestic slope • ML models learn nonlinearity and
Logit slope interaction with high/low nominal interactions ‘endogenously’
short-term rates.

17
Shapley regression for econometric analysis (Joseph, 2019)

ŷ = P[ycrisis |x] = Logit φ0 + β̂ S ΦSML (x)

(1)

The Shapley values ΦML (xk )S are interpreted as model-based transformations of

variable xk .

See also: bankunderground.co.uk/opening-the-machine-learning-black-box

18
(Shapley) regression table for extreme trees

Shapley regression Logistic regression

Name Direction Share α-level p Coeff. α-level p
∗∗∗ ∗∗∗
Global slope − 0.23 0.000 -0.61 0.000
∗∗∗ ∗∗∗
Global credit + 0.18 0.000 0.67 0.000
∗∗∗ ∗∗∗
Domestic slope − 0.11 0.000 -0.58 0.000
∗∗∗ ∗∗∗
Domestic credit + 0.11 0.000 0.43 0.002
∗∗∗
CPI − 0.07 0.002 -0.24 0.160
Debt service ratio + 0.05 0.236 0.16 0.347
∗∗ ∗∗∗
Consumption − 0.05 0.029 -0.42 0.003
∗∗∗ ∗∗
Investment + 0.04 0.005 0.32 0.016
∗∗
other variables public debt, money, stock prices , current account

Table 1: Left: Shapley regression. Direction from logistic regression, p-values against the null
hypothesis of neg. or zero regression coefficient (not shown). Right: Coefficients and p-values of a
logistic regression. Significance levels: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. 19
Wrap-up

Insights Potential policy take-aways

• Machine learning models outperform
benchmark logistic regression in
out-of-sample financial crisis prediction
• Most important model drivers:
Credit growth & yield curve slope
(domestically & globally)
• ML models learn pronounced
nonlinearities and interactions
from the data
• Yield curve connects monetary
policy and financial stability
• System-wide leverage suggests
importance of macroprudential
tools, e.g. CyCB or LTV/I-ratios
• Global factors suggest importance
of international policy
coordination

• Especially: global + domestic and

slope + low nominal interest rates 20
The End: THX - Q & A

21
Robustness checks (I)

Setup Crises Extreme Random Logit SVM Neural Decision

trees forest regression net tree
Baseline 93 0.84 0.83 0.80 0.79 0.79 0.73
Testing transformations
Growth rates only 93 0.78 0.77 0.74 0.71 0.72 0.68

Hamilton filter 87 0.82 0.83 0.79 0.78 0.80 0.75

* 87 0.84 0.83 0.80 0.77 0.78 0.76
Adding variables
Nominal rates 93 0.83 0.82 0.80 0.78 0.77 0.73
Real rates 93 0.82 0.82 0.80 0.78 0.79 0.75
Loans by sector 50 0.85 0.84 0.84 0.77 0.82 0.78
* 50 0.87 0.86 0.84 0.76 0.81 0.79
House prices 81 0.86 0.84 0.80 0.78 0.78 0.76
* 81 0.85 0.84 0.80 0.77 0.79 0.76
Robustness checks (II)

Setup Crises Extreme Random Logit SVM Neural Decision

trees forest regression net tree
Baseline 93 0.84 0.83 0.80 0.79 0.79 0.73
Changing the horizon
1 year 93 0.81 0.81 0.80 0.78 0.78 0.71
* 93 0.85 0.83 0.80 0.78 0.79 0.74
3 years 90 0.83 0.83 0.80 0.78 0.77 0.74
* 90 0.84 0.83 0.80 0.79 0.79 0.73
4 years 88 0.86 0.85 0.79 0.80 0.78 0.76
* 88 0.84 0.83 0.80 0.78 0.79 0.75
5 years 87 0.85 0.84 0.79 0.80 0.77 0.75
* 87 0.84 0.83 0.80 0.78 0.79 0.76
Predict one year before crisis
48 0.85 0.81 0.81 0.79 0.80 0.72
Detour: Shapley values in cooperative game theory

• How much does player A contribute a

collective payoff f obtained by a group
of n? (Shapley, 1953).
• Observe payoff of the group with and
without player A.
X |S|!(|n| − |S| − 1)!
• Contribution depends on the other φA = [f (S∪{A}−f (S)]
|n|!
players in the game. S⊆n\A
(2)
• All possible coalitions S need to be
evaluated. 2|n|−1 coalitions are evaluated.
Computationally complex!
Intuitive example: stealing apples together

• Three siblings (strong [S], tall [T] &

smart [M]) set off to nick some apples
A (pay-off) from the neighbour’s tree
• For each sibling, sum over marginal
contribution to coalitions of one and
two
• So, the Shapley value of the strong
Source: 6oxgangsavenueedinburgh
sibling is then:

1 1 1 1
φS = [A(S)−A(∅)]+ [A(T , S)−A(T )]+ [A(M, S)−A(M)]+ [A(T , M, S)−A(T , M)]
6 6 6 3
(3)
 Mean absolute Shapley values
US (normalized)
slo 0.00 0.05 0.10 0.15 0.20 0.25
p e+
No
ise
Glo
ba
l cr
ed
Do
me it
stic
slo
Do pe
m es
tic
cre
dit

CP
I
Replacing global slope with US slope

Co
ns
um
De pti
bt on
se
rvi
ce
rat
Bro io
ad
mo
ne
y
Pu
blic
Cu de
bt
rr en
ta
cco
un
Sto t
ck
m ark
et
Extreme trees

Inv
Random forest

es
tm
en
t
Logistic regression
 Shapley difference
0.00 0.05 0.10 0.15
Glo
ba
l sl
op
Glo e
ba
l cr
Do ed
me it
stic
Do s l o
me pe
stic
cre
dit
Change of Shapley values over time

De CP
bt
se I
rvi
ce
rati
Co o
ns
u mp
tio
n
Inv
es
tm
en
Pu t
blic
de
Bro bt
a dm
on
Sto ey
ck
ma
Cu rke
rre
nt t
ac
co
un
t
Complete data (1870 - 2016)
 Shapley difference
0.00 0.05 0.10 0.15
Glo
ba
l sl
op
Glo e
ba
l cr
Do ed
me it
stic
Do s l o
me pe
stic
cre
dit
Change of Shapley values over time

De CP
bt
se I
rvi
ce
rati
Co o
ns
u mp
tio
n
Inv
es
tm
en
Pu t
blic
de
Bro bt
a dm
on
Sto ey
ck
ma
Cu rke
rre
nt t
ac
co
Before WW2 (1870 - 1933)

un
t
Complete data (1870 - 2016)
 Shapley difference
0.00 0.05 0.10 0.15
Glo
ba
l sl
op
Glo e
ba
l cr
Do ed
me it
stic
Do s l o
me pe
stic
cre
dit
Change of Shapley values over time

De CP
bt
se I
rvi
ce
rati
Co o
ns
u mp
tio
n
Inv
es
tm
en
Pu t
blic
de
Bro bt
a dm
on
Sto ey
ck
ma
Cu rke
rre
nt t
ac
co
1990s crises (1985 - 1992)

un
t
Complete data (1870 - 2016)
 Shapley difference
0.00 0.05 0.10 0.15
Glo
ba
l sl
op
Glo e
ba
l cr
Do ed
me it
stic
Do s l o
me pe
stic
cre
dit
Change of Shapley values over time

De CP
bt
se I
rvi
ce
rati
Co o
ns
u mp
tio
n
Inv
es
tm
en
Pu t
blic
de
Bro bt
a dm
on
Sto ey
ck
ma
Cu rke
t
Complete data (1870 - 2016)

rre
nt
ac
co
un
t
Global financial crisis (2004 - 2010)
Neural net forecasting casting evaluation

True positive
True negative
False negative
False positive

Belgium
Japan
Portugal
Denmark
Germany
France
Australia
Netherlands
Finland
Switzerland
Canada
Norway
United States
Sweden
Italy
United Kingdom
Spain
1948

1953

1958

1963

1968

1973

1978

1983

1988

1993

1998

2003

2008

2013
More interactions with domestic factors

Interaction Sign Share α-lvl p-values

Domestic slope x Domestic credit - 0.08 0.154
∗
Domestic slope x Debt service ratio - 0.15 0.051
∗
Domestic slope x Investment - 0.11 0.070
∗∗
Domestic slope x Consumption + 0.17 0.043
Domestic slope x CPI + 0.04 0.365
Domestic slope x Stock market + 0.09 0.109
Domestic credit x Debt service ratio + -0.13 0.070
∗∗∗
Domestic credit x Investment + 0.21 0.005
Domestic credit x Consumption - -0.20 0.005
∗∗
Domestic credit x CPI + 0.17 0.012
Domestic credit x Stock market + -0.17 0.009
Extreme trees interaction terms, α-level: ∗ : 10%, ∗∗
: 5%, ∗∗∗
: 1%, n-obs: 1249.
Average slope correlations (15yr sliding window)

1.0
All pairwise correlations
Correlation with US

0.8
0.6
0.4
0.2
0.0
-0.2

1880

1900

1920

1940

1960

1980

2000

2020
Shapley interactions Effects: E.g. slope and credit
0.4

Crises
• Many crisis fall into upper left
0.3

quadrant
0.0 0.1 0.2
Domestic credit

• High domestic credit growth and

flat/negative slope of the global yield
curve well separate crisis built-up and
normal times.
-0.1

• Credit booms might be more

-0.2

dangerous in a low/inverted yield

curve global environment
-1 0 1 2
Global slope
Shapley interactions Effects: E.g. slope and credit
0.4

Crises
• Many crisis fall into upper left
0.3

quadrant
0.0 0.1 0.2
Domestic credit

• High domestic credit growth and

flat/negative slope of the global yield
curve well separate crisis built-up and
normal times.
-0.1

• Credit booms might be more

-0.2

dangerous in a low/inverted yield

curve global environment
-1 0 1 2
Global slope
Shapley interactions Effects: E.g. slope and credit
0.4

Crises 0.086
0.069 • Many crisis fall into upper left
0.3

0.052 quadrant
0.0 0.1 0.2

0.035
Domestic credit

0.017
• High domestic credit growth and
0.000 flat/negative slope of the global yield
-0.017 curve well separate crisis built-up and
-0.035 normal times.
-0.1

-0.052
• Credit booms might be more
-0.2

-0.069
dangerous in a low/inverted yield
-0.086
curve global environment
-1 0 1 2
Global slope
Interaction with global factors important

Interaction Sign Share α-lvl p-values

∗∗∗
Global slope x Global credit - 0.06 0.002
Global slope x Domestic slope + 0.03 0.169
∗∗∗
Global slope x Domestic credit - 0.07 0.004
∗∗∗
Global slope x Investment - 0.04 0.000
∗
Global slope x Consumption + 0.03 0.058
∗∗∗
Global slope x CPI + 0.04 0.003
Global slope x Stock market - 0.03 0.185
∗
Global credit x Domestic credit + 0.03 0.083
∗∗
Global credit x Domestic slope - 0.03 0.027
∗∗
Global credit x Investment + 0.02 0.036
∗∗∗
Global credit x CPI - 0.04 0.001
∗∗∗
Global credit x Consumption - 0.03 0.002
∗∗
Global credit x Stock market + 0.03 0.014
Extreme trees interaction terms, α-level: ∗ : 10%, ∗∗
: 5%, ∗∗∗
: 1%, n-obs: 1249.
References i

Aikman, D., Haldane, A. G., and Nelson, B. D. (2013). Curbing the Credit Cycle. The Economic
Journal, 125(585):1072–1109.
Alessi, L. and Detken, C. (2011). Quasi real time early warning indicators for costly asset price
boom/bust cycles: A role for global liquidity. European Journal of Political Economy,
27(3):520–533.
Alessi, L. and Detken, C. (2018). Identifying excessive credit growth and leverage. Journal of Financial
Stability, 35:215–225.
Babeckỳ, J., Havranek, T., Mateju, J., Rusnák, M., Smidkova, K., and Vasicek, B. (2014). Banking,
debt, and currency crises in developed countries: Stylized facts and early warning indicators. Journal
of Financial Stability, 15:1–17.
Beutel, J., List, S., and von Schweinitz, G. (2018). An evaluation of early warning models for systemic
banking crises: Does machine learning improve predictions?
References ii

Bordo, M., Eichengreen, B., Klingebiel, D., and Martinez-Peria, M. S. (2001). Is the crisis problem
growing more severe? Economic policy, 16(32):52–82.
Borio, C. and Lowe, P. (2002). Asset prices, financial and monetary stability: exploring the nexus. BIS
Working Papers 114, Bank for International Settlements.
Breiman, L. (2001). Random forests. Machine Learning, 45(1):5–32.
Cecchetti, S. G., Kohler, M., and Upper, C. (2009). Financial crises and economic activity. Technical
report, National Bureau of Economic Research.
Cesa-Bianchi, A., Martin, F. E., and Thwaites, G. (2018). Foreign booms, domestic busts: The global
dimension of banking crises. Journal of Financial Intermediation.
Drehmann, M., Borio, C., and Tsatsaronis, K. (2011). Anchoring countercyclical capital buffers: The
role of credit aggregates. International Journal of Central Banking.
Duca, M. L. and Peltonen, T. A. (2013). Assessing systemic risks and predicting systemic events.
Journal of Banking & Finance, 37(7):2183–2195.
References iii

Geurts, P., Ernst, D., and Wehenkel, L. (2006). Extremely randomized trees. Machine Learning,
63(1):3–42.
Joseph, A. (2019). Shapley regressions: A universal framework for statistical inference on machine
learning models. Bank of England Staff Working Paper Series, (784).
Joy, M., Rusnák, M., Šmı́dková, K., and Vašı́ček, B. (2017). Banking and currency crises: Differential
diagnostics for developed countries. International Journal of Finance & Economics, 22(1):44–67.
Kindleberger, C. P. (1978). Manias, Panics and Crashes - A History of Financial Crises. New York:
Basic Books.
Laeven, M. L. and Valencia, F. (2008). Systemic banking crises: a new database. Number 8-224.
International Monetary Fund.
Lundberg, S. M. and Lee, S.-I. (2017). A unified approach to interpreting model predictions. In
Advances in Neural Information Processing Systems, pages 4765–4774.
References iv

Minsky, H. P. (1977). The Financial Instability Hypothesis: An Interpretation of Keynes and an

Alternative to“Standard” Theory. Challenge, 20(1):20–27.
Reinhart, C. M. and Rogoff, K. S. (2009). This Time Is Different: Eight Centuries of Financial Folly.
Princeton University Press. Google-Books-ID: ak5fLB24ircC.
Schularick, M. and Taylor, A. M. (2012). Credit booms gone bust: Monetary policy, leverage cycles,
and financial crises, 1870-2008. American Economic Review, 102(2):1029–61.
Shapley, L. S. (1953). A value for n-person games. Contributions to the Theory of Games,
2(28):307–317.
Strumbelj, E. and Kononenko, I. (2010). An efficient explanation of individual classifications using
game theory. The Journal of Machine Learning Research, 11:1–18.
Verikas, A., Vaiciukynas, E., Gelzinis, A., Parker, J., and Olsson, M. C. (2016). Electromyographic
patterns during golf swing: Activation sequence profiling and prediction of shot effectiveness.
Sensors, 16(4):592.
References v

Vermeulen, R., Hoeberichts, M., Vašı́ček, B., Žigraiová, D., Šmı́dková, K., and de Haan, J. (2015).
Financial stress indices and financial crises. Open Economies Review, 26(3):383–406.
Ward, F. (2017). Spotting the danger zone: Forecasting financial crises with classification tree
ensembles and many predictors. Journal of Applied Econometrics, 32(2):359–378.