T
C
a
b
a
A
R
R
A
A
J
D
G
K
S
H
A
1
m
d
b
i
i
b
s
h
y
h
e
h
p
i
X
W
s
L
c
o
f
c
h
1
0
Journal of Financial Stability 36 (2018) 39–52
Contents lists available at ScienceDirect
Journal  of  Financial  Stability
journal homepage: www.elsevier.com/locate/jfstabil
he  effect  of  the  financial  crisis  on  default  by  Spanish  households
arlos  Allera,∗, Charles Grantb 
Universidad CEU Cardenal Herrera, C/Carmelitas, 3, 03203 Elche, Spain
Brunel University London, Kingston Lane, Uxbridge, Middlesex, UB8 3PH London, United Kingdom
 r  t  i  c  l  e  i  n  f  o a  b  s  t  r  a  c  t
rticle history: We  analyse  the default  behaviour  of Spanish  households  immediately  before  and  after  the  recent  financial
eceived 12 December 2016 crisis.  Using  several  waves  of  the  Survey  of  Household  Finances  (a tri-annual  survey  of  financial  position
eceived in revised form 2 July 2017 of  Spanish  households),  we  show  that  younger,  poorer  and  less  well  educated  households  are  most
ccepted 15 February 2018 likely  to default.  A  key  contribution  is  to explain  the  change  in  arrears  since  the  onset  of  the  crisis.
vailable online 16 February 2018
Using  information  on credit applications  and  acceptances  we  decompose  the  change  in arrears  among  all
households  into  a contribution  from  four parts:  (i)  changes  in characteristics;  (ii) changes  in applications;
EL classification: (iii)  changes  in  acceptances;  (iv)  changes  in  arrears  among  borrowers.  We  show  the last  is the most
14
21 important  contribution.
©  2018  The  Authors.  Published  by  Elsevier  B.V.  This  is  an  open  access  article  under  the  CC  BY-NC-ND
eywords: license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
panish financial crisis
ousehold-debt
rrears
. Introduction
The aim of this paper is to explore the borrowing and repay-
ent behaviour of Spanish households since 2002 using household
ata. The recent international financial crisis has affected a num-
er of developed economies and Spain constitutes a particularly
nteresting country to study due to the severity of the crisis and
ts aftermath. During the span of this study, Spain experienced a
oom followed by a severe crash ensuing from the sub-prime cri-
is. Unemployment more than doubled from 2007 to 2009, and
ouseholds default rates experienced even larger increases in these
ears. Bernardino and Gutiérrez (2012) and Igan et al. (2014) show
ousehold credit in Spain has mirrored these changes in the macro-
conomy. Crook (2006) shows that credit to the household sector
as expanded more rapidly than in other EU countries in the years
rior to the crisis, as Spanish households have become as heavily
ndebted as households in Northern Europe (see Cecchetti et al.,
 Authors are grateful for helpful comments to participants at XVIII (Alicante) and
IX  (Seville) Applied Economics Meeting, Econometric Research in Finance (ERFIN)
orkshop (Warsaw), First Catalan Economic Society Conference (Barcelona) and
eminar participants at University of Balearic Islands and Middlesex University
ondon. We also thank Iftekhar Hasan (editor) and three anonymous referees for
omments which have greatly improved the paper. All remaining errors are our
wn. Financial support from Generalitat Valenciana grant Prometeo/2017/158 and
rom ‘Obra Social La Caixa’ are gratefully acknowledged.
∗ Corresponding author.
E-mail addresses: carlos.allerarranz@uchceu.es (C. Aller),
harles.grant@brunel.ac.uk (C. Grant).
ttps://doi.org/10.1016/j.jfs.2018.02.006
572-3089/© 2018 The Authors. Published by Elsevier B.V. This is an open access article 
/).2011; or Bover et al., 2016). Default rates on mortgages and con-
sumer loans increased dramatically over this period, and a key aim
of this study is to explore the causes of this increase. It will inves-
tigate whether the increase was  due to changes at the household
level in income and unemployment, and whether it was  the result
of credit being extended to previously excluded households.
This paper will use the Survey of Household Finances (EFF), a
household level dataset available for four different waves in the
period 2002–2011. This survey was  commissioned by the Bank
of Spain to collect detailed information about the financial posi-
tion of Spanish households and hence it provides a rich source to
study the debt holding and repayment behaviour of a representa-
tive sample of Spanish families. Using household data enables us to
understand the differences across households in their responses to
the crisis. This paper will also discuss the changes in the borrowing
and arrears behaviour of Spanish households before and after the
financial crisis.
By building on the approach of Grant and Padula (2016), we  will
use a decomposition exercise to understand the changes in arrears
among Spanish households since 2002 (which rose sharply over the
survey period). Christelis et al. (2013) similarly employ a decom-
position exercise when looking at differences in asset holding in
Europe and the US. Their exercise investigates whether character-
istics or coefficients explain these differences. In our decomposition
exercise, we  will note that between any two years the overall
change in arrears in the population can be split into four parts: (i)
changes in characteristics; (ii) changes in applications; (iii) changes
in acceptances; (iv) changes in arrears among borrowers. We  will
under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.
4 inanc
i
a
i
e
s
b
c
s
t
l
b
b
t
d
w
a
e
i
2
a
F
‘
a
a
i
(
m
p
t
E
i
S
u
b
i
o
D
t
l
K
a
i
G
a
p
e
t
w
(
a
2
o
t
c
q
f
t
d
b
0 C. Aller, C. Grant / Journal of F
nvestigate the relative importance of each in explaining the rise in
rrears since 2002. Thus our exercise investigates whether changes
n characteristics (such as unemployment and income changes) can
xplain the increase in arrears during the Great Recession. And
ince there will be separate regressions for credit applications, for
ank acceptances, and for arrears among borrowers, the effect of
hanges in the coefficients in each of these three regressions can be
eparately explored. Thus we will be able to investigate whether
he rise in arrears was caused by an increase in applications for
oans which was accommodated by lenders, or whether it caused
y a weakening in lending standards.
We proceed by first describing existing literature and some
ackground information about the Spanish credit market in Sec-
ion 2 before proceeding with the main part of the paper. Section 3
escribes the Survey of Household Finances, the household data set
hich will be used for the main analysis in the paper. The results
re reported in Section 4. In Section 5 we propose a decomposition
xercise which we will use to understand the cause of the change
n arrears. Our conclusions are reported in Section 6.
. Literature review
There is already an extensive literature on household arrears
s well as the role of arrears in explaining the Great Recession.
or example, Gerlach-Kristen and Lyons (2015) explore the role of
affordability’ and negative equity in explaining mortgage arrears
mong European households. For Spain, Ampudia et al. (2016)
rgue that unemployment and low wealth are among the most
mportant determinants of default, while Sánchez-Martínez et al.
2016) find that Spanish households whose family head is female,
arried or self-employed are at higher risk of missing mortgage
ayments. More interesting is to explore how changes in charac-
eristics explains the crisis. For example, Foote et al. (2009) and
lul et al. (2010) argue that unemployment and unexpected falls in
ncome have driven the increase in default among US households.
imilarly Blanco and Gimeno (2012) have argued that changes in
nemployment explain the surge in Spanish household default
etween 2007 and 2009.
Not all papers have attributed the increase in arrears to changes
n households characteristics. Much of the US literature has focused
n the expansion of credit prior to the crisis. Mian and Sufi (2009),
emyanyk and Van Hemert (2011) and Mayer et al. (2009) argue
hat the records of US lenders show that there was a deterioration of
ending standards which precipitated the sub-prime crisis, which
eys et al. (2010) attributed to the reduced incentive to screen
pplicants which arose with the securitization of mortgage lend-
ng in the US market. Similarly, both Crook (2006) and Duygan and
rant (2009), in cross-European studies, showed there had been
 large increase in borrowing by Spanish households in the years
rior to the crisis. Using lending records supplied by a Spanish real
state company, Akin et al. (2014) and Díaz-Serrano (2015) showed
here was softening of lending standards in the Spanish market,
hile Maddaloni and Peydró (2011) reach a similar conclusion
especially for mortgage loans) analysing the Bank Lending Survey,
 quarterly survey of Euro area banks on their lending practises for
002–2008.
This review of the literature suggests two different explanations
f the rise in arrears during the crisis. In the first, the rise in arrears is
he concomitant consequence of the deterioration in labour market
onditions; while in the second, the increase in arrears is the conse-
uence of the increase in lending to households hitherto excluded
rom borrowing due to low income and poor credit scores. While
hese are the two most popular explanations of the rise in arrears
uring the Great Recession, two other rarer arguments have also
een suggested in the literature on US households. Dell’Arricia et al.ial Stability 36 (2018) 39–52
(2012), by looking at the pool of mortgage applicants, argue there
was an increase in credit demand which was at least as important as
changes in lending policy in the US. While Guiso et al. (2013) argue
that survey evidence suggests that American households became
more willing to default regardless of their circumstances during the
sub-prime crisis.
2.1. The Spanish credit market
This study covers the period 2002–2011, the years immediately
prior to and following the outbreak of the 2008 financial crisis.
This crisis had severe effects on the Spanish financial sector and
the Spanish economy, with important consequences for the house-
hold sector. Spanish households typically hold 25-year or 30-year
variable rate repayment mortgages (which move with the Euro-
zone interest rate), and this has not changed over the crisis years.
Bernardino and Martín de Vidales (2014) discuss how in the years
preceding the crisis there was  a rapid expansion in credit to the
household sector, which they associated with the liberalisation
of the regulation of savings banks which started in 1990. How-
ever, these organisations were newly regulated following the crisis
when it became apparent that many savings banks were under-
capitalized.
Fig. 1a provides information on the household default rate for
the mortgage (housing) loans and for consumer (non-housing)
loans. Campbell and Cocco (2015), in a US context, show that house-
holds are likely to default on their mortgage when they have little
or no equity in the property (which will be during the first few
years of the mortgage agreement). Schwartz and Torous (1993)
provide evidence that mortgage defaults among US  households
peak within 16 quarters of the initiation of the loan (the household
would have entered arrears considerably earlier). The most impor-
tant constituents of consumer loans among Spanish households are
“unsecured personal loans”, “credit lines” and “credit cards”. The
first is an agreed loan with an agreed repayment plan, typically
over a period of up to three years; the second is an agreed bor-
rowing limit, which is repayable on demand; while credit cards
debt is typically repayable over several months when making min-
imum repayments. Saurina (2009), classifying loans to the Spanish
household sector by their riskiness, shows mortgages are the least
risky loan type (especially those with a loan-to-value ratio below
80 percent), while credit cards and credit lines are the most risky.
The figure shows that the default rate on consumer loans is consid-
erably higher than on mortgages, but that the default rate for both
mortgages (solid line) and consumer loans (dashed line) increased
during 2007 and 2008, which plateaued between 2009 and 2011,
before again increasing in the following years. Fig. 1b shows a sharp
increase in mortgage disclosures rates during the survey period,
while the Instituto Nacional de Estadística report a several-fold
increase in individual insolvency procedures since 2004.
In the rest of the paper we will explore this increase in household
default, and assess how household arrears differs between house-
hold types. We  will also investigate the extent to which the overall
change in arrears that can be attributed to changes in credit appli-
cations and/or credit acceptances (as well as exploring other factors
that may  have contributed to changes in arrears). This will enable
us to assess by how much more the increase in arrears would have
been if there had not been change in the supply and/or the demand
for credit.
3. DataThe data used in this paper is taken from the Survey of House-
hold Finances (EFF) developed by the Bank of Spain. This is a survey
of Spanish households which is collected every three years start-
C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52 41
Fig. 1. (a) Households’ default rate. Note: Both rates are defined as the ratio between the total doubtful loans (Loans in which some instalment has not been paid for a period
of  more than 90 days, and those exposures in which there are reasonable doubts as to total repayment under the terms agreed) and total lending to households at Spanish
e of fore
J
i
s
fi
h
h
t
i
m
s
conomy. Source: Bank of Spain. (b) Number of foreclosures. Note: Total number 
udicial.
ng in 2002 (e.g. we have data for 2002, 2005, 2008 and 2011). The
urvey collects detailed information on the financial position and
nancial decisions adopted by a representative sample of Spanish
ouseholds as well as questions about the composition, income,
ousing and labour market participation of household members. In
he design of the survey, richer households were over-represented
n order to collect information on a variety of assets which are
ainly held by wealthy households.1
1 The participation rate is not particularly high, as is typical among these types of
urveys. The overall participation rate, although decreasing with wealth, was 47.3%closures at courts of first instance in Spain, 2007–2012. Source: CGPJ: Estadística
We restrict attention to households whose head is between
30 and 75 years old and exclude those households with multi-
ple unrelated adults (where we define the household head as the
main earner). After making these selections, there are over 4000
households included in each year of the analysis.
Household-related variables are obtained using several ques-
tions asked to the household respondent. We  utilize variables
containing information on the level of education, marital status,
house ownership, labour market status, age and the number of
(2002), 47.3% (2005), 61.9% (2008) and 50.8% (2011) (see Bover et al., 2014, and
references contained therein).
42 C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52
Table 1
Summary statistics.
2002 2005 2008 2011
mean sd mean sd mean sd mean sd
Income 40.6 0.72 38.3 0.82 38.6 0.95 36.7 0.97
Age  50.9 0.27 50.9 0.29 50.6 0.31 50.9 0.35
Couple (%) 77.0 0.86 73.7 0.97 72.3 1.04 71.9 1.14
No.  children 0.5 0.02 0.5 0.02 0.5 0.02 0.5 0.02
male(%) 78.9 0.83 77.8 0.91 77.1 0.95 74.2 1.08
University (%) 17.3 0.78 19.3 0.89 19.0 0.98 21.6 1.10
Employed (%) 57.8 1.01 58.4 1.07 56.4 1.17 54.7 1.28
Unemployed (%) 5.0 0.47 4.9 0.50 8.0 0.64 10.9 0.79
Retiree (%) 26.4 0.85 26.8 0.92 24.7 0.93 23.6 0.97
Self-employed (%) 10.8 0.66 9.9 0.65 10.8 0.79 10.7 0.86
Homeowner (%) 83.3 0.79 84.5 0.79 83.9 0.91 83.5 1.05
Applicant (%) 54.2 1.02 59.3 1.07 59.5 1.16 60.7 1.23
Borrow (%) 52.7 1.03 57.1 1.08 56.4 1.18 55.8 1.27
Arrears (%) 9.2 0.63 8.3 0.59 9.4 0.73 11.1 0.94
No.  observations 4047 4701 4836 4684
A ouseh
h  for 20
h g (dis
t puta
c
i
c
h
t
w
h
d
c
a
h
a
c
c
a
f
i
h
a
p
t
t
f
o
w
t
T
s
i
d
p
h
g
r (households between 70 and 74). However, the LR test for the
T different age dummies shows that differences between the differ-
a ent age-groups is not statistically significant (see Meng and Rubin,
h
i
i
2
uthors own calculations using 2002, 2005, 2008 and 2011 waves of Survey of H
ouseholds holding any kind of debt (including credit card outstanding balances
olders  and those that applied for a loan or refused to do so because of sure rejectin
he  household delayed the payment of any of its debts in the last 12 months. Five im
hildren living in the house. We  also include variables contain-
ng information on household income (adjusted by the monthly
onsumer price index from Instituto Nacional de Estadística).
The main focus of the paper will be to exploit questions on debt
olding. The first key variable is the dummy  “apply” for whether
he household applied for a loan during the last two years (or
as discouraged from doing so because they believed they would
ave been rejected). The survey also allows us to construct the
ummy  “accept” which takes the value one if the application for
redit was accepted and zero if it was rejected (where discour-
ged households are included with the rejected households). Lastly,
ouseholds report whether they were unable to pay as scheduled
ny of their debt payments in the last year, for which we  again
onstruct a dummy  called “arrears”.
The question on arrears covers a wide-range of different out-
omes, since, at one extreme, the household may  be facing court
ction for the recovery of the debt (or the household may  be filing
or bankruptcy because it is unable to pay), while the other extreme,
t may  have been a few days late on a single payment and otherwise
ave an exemplary repayment record. Moreover, the questions on
pplications and acceptances, unfortunately, covers a different time
eriod than the question on arrears since the question on applica-
ions refers to the last two  years, while arrears are reported during
he last year.
Table 1 provides summary statistics for each wave of the survey
or the variables included in the analysis and for the full sample
f households (including those households who do not borrow)
here these calculations, and the regressions, use all five impu-
ations (see Rubin, 1987, for a discussion of multiple imputation).
he table shows that average annual real household income (mea-
ured in thousands) fell over the survey period. While the fall in
ncome between 2005 and 2011 is consistent with the recession
uring those years, the fall between 2002 and 2005 is more sur-
rising. Nevertheless it can be explained by noting the change in
ousehold structure: there was an increase in the number of sin-
le adult households from 40.9% in 2002 to 44.4% in 2005 and a
eduction in the proportion of couple households over this period.
he table also shows that the proportion of households that have
pplied for a loan over the last two years (the variable “applicant”)
as steadily increased between 2002 and 2011. But despite this
ncrease, the proportion of households currently borrowing at first
ncreased between 2002 and 2005, but then slowly fell between
005 and 2011 (suggesting there was an increase in the proportionolds Finances. Household head age: 30–75 years. “Borrow” is the percentage of
05, 2008 and 2011)in the moment of the interview. “Applicant” comprises debt
courage) in the last two years. “Arrears” is a dummy variable recorded as a 1 when
tions (following Rubin, 1987) and weights are used.
of households that had their loan rejected). The pattern over time
of household arrears displays the opposite pattern; at first it fell,
and then between 2005 and 2011 it steadily increased.
4. Regression results
An important aim of the paper is to investigate how the repay-
ment behaviour of Spanish households has changed since the
onset of the recent financial crisis, and to provide some insight
as to what has caused this change. This requires not only study-
ing arrears, but also households’ application behaviour, and the
lending behaviour of credit institutions. We  will investigate the
determinants of whether the household applied for a loan; whether
their loan application was  accepted; and whether they repaid the
debts on schedule or entered arrears. Fundamentally, we  wish to
understand how arrears changed over time. Consequently, we  per-
form separate logit estimations for each wave of the data which
enables us to understand how the household credit market differs
over time, and particularly how it differs before and after the finan-
cial crisis. As explanatory variables, we  include a set of household
characteristics including: dummies for different age strata of the
household, gender of the head, whether the household is headed
by a couple, dummies for level of income (separated into six roughly
equally sized groups), whether the household head has a university
degree, number of children living in the house, a dummy for house
ownership and three different dummies to indicate whether the
head is self-employed, retired or unemployed (employed house-
holds are the left-out group).2 
4.1. The rate of arrears
Table 2 reports results for arrears over the last year for each
wave of the survey. The regressions in the columns 2–5 include all
households (where, clearly, non-borrowers will not report arrears).
The results for the 2002 wave (the second column in Table 2) show
that each age group is significantly different from the left-out group2 Of course, the effect age, time and year-of-birth are not separately identifiable.
We  have reported the changes by age-group and year and have not attributed the
changes in arrears to year-of-birth cohort effects.
C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52 43
Table  2
Logit estimation results I.
Pr(Arrears) Pr(Arrears/Borrow)
2002 2005 2008 2011 2002 2005 2008 2011
Age 30–34 0.967* 1.861*** 1.588*** 2.158*** −0.609 0.677 0.423 0.470
(0.496) (0.510) (0.468) (0.460) (0.572) (0.559) (0.498) (0.487)
Age  35–39 0.821* 1.796*** 1.739*** 1.741*** −0.435 0.697 0.596 0.363
(0.482) (0.505) (0.445) (0.448) (0.564) (0.557) (0.474) (0.470)
Age  40–44 1.123** 1.842*** 1.685*** 1.964*** −0.148 0.794 0.722 0.587
(0.477) (0.500) (0.437) (0.434) (0.560) (0.552) (0.465) (0.460)
Age 45–49 0.957** 1.796*** 1.537*** 1.822***  −0.178 0.768 0.646 0.619
(0.479) (0.494) (0.429) (0.424) (0.563) (0.542) (0.456) (0.447)
Age  50–54 1.008** 1.730*** 1.456*** 1.736*** −0.187 0.689 0.675 0.602
(0.474) (0.497) (0.430) (0.422) (0.561) (0.544) (0.456) (0.442)
Age  55–59 0.840* 1.244** 1.361*** 1.514*** −0.434 0.259 0.684 0.751*
(0.466) (0.494) (0.423) (0.424) (0.543) (0.539) (0.454) (0.448)
Age  60–64 0.762* 0.605 1.010*** 1.088*** 0.035 −0.221 0.539 0.311
(0.425) (0.497) (0.394) (0.397) (0.512) (0.532) (0.431) (0.425)
Age *** 65–69 0.789 0.775** −0.144 0.469 0.291 0.505 −0.389 −0.040
(0.303) (0.390) (0.415) (0.401) (0.394) (0.424) (0.439) (0.429)
Couple −0.073 0.307 0.195 0.314* −0.389*  −0.007 −0.146 0.064
(0.197) (0.198) (0.175) (0.166) (0.224) (0.215) (0.194) (0.182)
Income 15–25 −0.695*** −0.421** 0.224 −0.153 −1.209***   −0.668*** −0.195 −0.601***
(0.233) (0.184) (0.208) (0.176) (0.291) (0.201) (0.231) (0.200)
Income 25–35 −0.712*** −0.835*** 0.367* −0.374* −1.364*** −1.255*** −0.256 −0.893***
(0.262) (0.240) (0.206) (0.196) (0.278) (0.253) (0.227) (0.215)
Income 35–45 −0.797*** −0.936*** −0.332 −0.701*** −1.440*** −1.321*** −0.917*** −1.303***
(0.279) (0.254) (0.313) (0.247) (0.310) (0.261) (0.331) (0.263)
Income 45–57 −0.841*** −1.320*** −0.411 −1.047*** −1.439***   −1.646*** −1.153*** −1.708***
(0.289) (0.365) (0.288) (0.285) (0.328) (0.370) (0.303) (0.297)
Income >57 −1.690***  −1.520*** −0.716*** −1.041*** −2.433*** −1.881*** −1.397*** −1.649***
(0.310) (0.302) (0.261) (0.236) (0.363) (0.313) (0.275) (0.250)
Homeowner *** *** ** *** *** *** −0.763 −0.690 −0.345 0.161 −1.165 −1.012 −0.771 −0.360**
(0.155) (0.147) (0.156) (0.167) (0.187) (0.161) (0.169) (0.183)
Univ  −0.934*** −0.818*** −0.987*** −1.044*** −0.993*** −0.752*** −0.880*** −0.998***
(0.237) (0.214) (0.202) (0.193) (0.252) (0.218) (0.207) (0.201)
Unemployed 1.048*** 0.489** 0.865*** 0.660*** 1.225*** 0.738*** 1.134*** 0.986***
(0.232) (0.246) (0.189) (0.171) (0.292) (0.285) (0.213) (0.194)
Retiree −0.297 −0.271 0.194 −0.234 −0.138 0.068 0.361 −0.081
(0.371) (0.368) (0.314) (0.304) (0.410) (0.408) (0.330) (0.316)
Self-employed −0.354 −0.189 0.539*** 0.315* −0.242 −0.131 0.584*** 0.327*
(0.240) (0.204) (0.175) (0.171) (0.260) (0.214) (0.186) (0.182)
Male  −0.467** −0.235 −0.381** −0.179 −0.448** −0.095 −0.241 −0.170
(0.185) (0.185) (0.167) (0.160) (0.212) (0.205) (0.187) (0.175)
No.  children 0.442*** 0.310*** 0.324*** 0.227*** 0.292*** 0.236*** 0.263*** 0.205**
(0.091) (0.080) (0.078) (0.077) (0.100) (0.088) (0.084) (0.084)
Constant −1.773*** −2.882*** −3.549*** −3.679*** 1.413** −0.622 −1.036** −0.808*
(0.462) (0.495) (0.438) (0.450) (0.562) (0.549) (0.474) (0.486)
LR  test (age) 0.2714 0.0016 0.0066 0.0002 0.6299 0.3181 0.5462 0.6322
LR  test (income) 0.0004 0.0000 0.0004 0.0010 0.0000 0.0000 0.0000 0.0000
Observations 4047 4701 4836 4685 1724 2318 2296 2181
Standard errors in parentheses. Columns 2–5 estimate a logit regression model of whether the reported arrears during the last year using the whole sample of households.
Columns 6–9 estimate a logit regression model of whether the reported arrears during the last year using the sample of households who have a loan. Reference age: 70–75.
Reference income: less than 15,000 real euros. Likelihood ratio (LR) test files report p-value of the joint significance test of the dummies, following Meng and Rubin (1992).
1
h
i
a
w
i
c
o
l
r the same.
r Columns 6–9 of Table 2 looks at the repayment behaviour of
borrowers.3 The results for the sub-sample of borrowers are mostly
a
s
e
c
* p < 0.1.
** p < 0.05.
*** p < 0.01.
992, for a discussion of the LR test). University educated house-
olds are also more likely to repay, while the number of children
ncreases the incidence of arrears. Households with a male head
re less likely to be in arrears than households with a female head
hile unemployed households are significantly more likely to be
n arrears. However, couple is not significant in the 2002 wave. As
ould be expected, high income households are more likely to repay
n schedule than middle income households, who  in turn are more
ikely to repay than low income households; a joint significance test
ejects that these income-dummies are all equal. Owning a house
educes arrears.
The results are similar in the remaining three waves. The effect of
 university education, of being unemployed and of children remain
ignificant; however, male is only significant in the 2008 wave. The
ffect of being self-employed gets larger over time, and is signifi-
ant in the last two waves of the survey. Similarly, the coefficienton couple has changed sign, and in the last wave, has become sig-
nificant. The effect of the age dummies remains significant in the
later waves. The slope effect of age, however, is now bigger since
the coefficients on the age-dummies are larger. Moreover, unlike in
2002, the LR test which tests whether the coefficients are different
(at the bottom of the table) is now significant in the 2005, 2008 and
2011 waves of the survey. The income coefficients are also signifi-
cant in these waves (except for the lower income groups in the last
two waves); the joint-test again rejects that all the coefficients are3 Grant and Padula (2016) estimate what they describe as “the true propensity
to  repay” which they argue can be bounded between the estimates on the whole
44 C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52
Fig. 2. The rate of arrears by income and age. Note: The two  left-hand figures report the predicted probability of the household entering arrears during the last year using
t y of th
h  appli
h s up b
s
v
c
n
a
r
i
e
t
s
h
a
m
d
o
i
2
a
l
t
l
a
a
m
p
e
r
c
a
he  whole sample, while the two right-hand figures report the predicted probabilit
ouseholds who  borrow. Predicted values obtained from logit estimations (Table 2)
ouse  owner employed, house owner, no children and total household income sum
imilar to the results for the whole sample. They show that uni-
ersity education consistently reduces arrears, but the number of
hildren raises arrears. In the sample of borrowers, age is not sig-
ificant in any wave (the LR test never rejects the null that all
ge-groups default at the same rate); this finding differs from the
egressions using the whole sample. In contrast the effect of income
n the regressions is now stronger, since there is a larger differ-
nce between the lowest income and highest income households in
heir repayment behaviour. Similarly unemployed households are
ignificantly less likely to repay, and the effect is slightly stronger.
The results for age and income are also shown in Fig. 2. It plots
ow arrears changes as income changes (in the top panel) and as
ge changes (in the bottom panel), holding other variables at their
edian value.4 The top-left panel plots the rate of arrears for six
ifferent income groups. It shows that in each survey year, the rate
f arrears was higher for low-income groups and lower for higher
ncome groups. It also shows that arrears tended to be lower in the
002 wave of the survey than in later waves, with the increase in
rrears mostly being greater for lower income groups. The bottom-
eft panel looks at arrears among different age-groups. It shows
hat the rate of arrears is highest among younger households, but
ower among older households. The figure also shows that arrears
t younger ages increased between 2002 and 2005, and increased
gain between 2008 and 2011. This results in the age-profile being
uch steeper at the end of the sample period than at the beginning.
opulation, and the estimates on the population of borrowers, e.g., between the
stimates in columns 2–5 and columns 6–9.
4 The median values are taken by pooling all waves of the survey together. The
esulting representative household is aged 45–49, has no children, is headed by a
ouple and the head is an employed man. Total household income is between 15,000
nd 25,000 euros per year (in 2010 real terms).e household entering arrears during the last year where the sample is restricted to
ed to a reference household: headed by a man, 45–49 years old that lives in couple,
etween 15,000 and 25,000 euros (in 2010 real terms) per year.
The two  right hand panels show arrears among borrowers. The
top panel shows how arrears falls with income. It shows the differ-
ences between low income and high income households are even
sharper than the in the panel on the left which includes all house-
holds. The bottom-right panel of Fig. 2 looks at age, and shows that
there seems to be no clear pattern to the age-profile of arrears. This
is in contrast to the bottom-left panel where there is a clear age-
profile to arrears for 2005–2011. However, for households below
60, the rate of arrears increased steadily between waves at a similar
amount for each age-group (the pattern is less clear-cut for older
households).
4.2. Applications and acceptances
Table 2 showed there are some differences between arrears
among borrowers and arrears in the whole population. This sug-
gests that whether a credit application is made and accepted might
have a role in explaining arrears among the whole population of
Spanish households. Furthermore, changes in arrears over time
may  be partly explained by changes in the pool of borrowers. The
first four columns of Table 3 report the determinants of loan appli-
cants, where the left-hand-side variable is a dummy  for whether
the household reports that they applied for a loan during the last
two years. For the most part, the results show there have been few
changes in the effect of these variables over time. The results show
that coupled households are significantly more likely to apply for
a loan than single households; a university education reduces the
demand for loans; home-owners are more likely to apply; while
having children increases the demand for loans. These results help
to explain the rate of arrears among university graduates: the first
four columns of Table 2 showed graduates are less likely to be in
arrears, while here we  have shown that these households are less
C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52 45
Table  3
Logit estimation results II.
Pr(Application) Pr(Accept/Apply)
2002 2005 2008 2011 2002 2005 2008 2011
Age 30–34 2.339*** 1.959*** 2.073*** 2.659*** 1.376 1.730** 1.058 1.647***    
(0.234) (0.212) (0.222) (0.259) (1.182) (0.830) (0.849) (0.576)
Age  35–39 2.019*** 1.711*** 2.203*** 2.141*** 1.349 1.239 1.385 1.269**
(0.222) (0.202) (0.209) (0.209) (1.168) (0.796) (0.846) (0.523)
Age 40–44 1.693*** 1.379*** 1.598*** 1.828*** 1.671 2.093**   1.691* 1.894***
(0.218) (0.194) (0.191) (0.193) (1.193) (0.883) (0.867) (0.550)
Age *** *** 45–49 1.484 1.397 1.528*** 1.611*** 0.889 1.612** 1.746** 1.579***
(0.208) (0.179) (0.179) (0.172) (1.117) (0.807) (0.854) (0.508)
Age  50–54 1.513*** 1.284*** 1.220*** 1.381*** 1.437 1.381* 1.360* 2.024***
(0.202) (0.179) (0.171) (0.160) (1.147) (0.767) (0.804) (0.547)
Age  55–59 1.500*** 1.068*** 0.957*** 1.021*** 1.812* 2.365*** 1.957** 0.982**
(0.194) (0.166) (0.166) (0.158) (1.090) (0.918) (0.859) (0.480)
Age  60–64 0.914*** 0.699*** 0.648*** 0.706*** 1.634* 1.608** 1.245 1.298***
(0.180) (0.154) (0.146) (0.138) (0.989) (0.664) (0.759) (0.437)
Age  65–69 0.574*** 0.444*** 0.125 0.401*** 0.870* 0.506 0.962** 0.583*
(0.148) (0.129) (0.127) (0.123) (0.470) (0.352) (0.474) (0.336)
Couple 0.361*** 0.426*** 0.438*** 0.184** −0.413 1.084*** 0.667** 0.707***
(0.104) (0.096) (0.090) (0.090) (0.472) (0.349) (0.319) (0.252)
Income 15–25 0.132 0.197* 0.277** 0.095 0.817 0.324 1.083*** 0.919***
(0.146) (0.116) (0.122) (0.123) (0.505) (0.304) (0.300) (0.242)
Income 25–35 * *** *** 0.280 0.407 0.671 0.191 1.200* 1.018*** 1.780*** 1.612***
(0.151) (0.120) (0.122) (0.123) (0.683) (0.393) (0.405) (0.317)
Income 35–45 0.344* 0.376*** 0.425*** 0.200 1.875** 1.424** 1.836*** 2.812*** 
(0.177) (0.139) (0.138) (0.140) (0.934) (0.672) (0.507) (0.693)
Income 45–57 0.229 0.276 0.748*** 0.480*** 0.963 2.022* 2.336*** 2.852***
(0.172) (0.168) (0.145) (0.147) (0.938) (1.089) (0.640) (0.675)
Income >57 0.463*** 0.331*** 0.692*** 0.334*** 2.002*** 3.170*** 3.967*** 3.261***  
(0.150) (0.120) (0.128) (0.127) (0.760) (1.118) (1.061) (0.528)
Homeowner 0.263** 0.222** 0.151 .0218** 1.629***  0.837*** 1.994*** 1.492***
(0.104) (0.098) (0.102) (0.104) (0.340) (0.285) (0.253) (0.206)
Univ −0.197** −0.291*** −0.325*** −0.285*** 0.108 0.206 0.546 0.066
(0.089) (0.081) (0.080) (0.081) (0.507) (0.389) (0.439) (0.288)
Unemployed 0.216 0.113 −0.013 0.040 −0.347 −1.415*** −1.290*** −0.932***      
(0.180) (0.177) (0.140) (0.130) (0.670) (0.414) (0.328) (0.276)
Retiree *** −0.158 −0.491 −0.139 −0.171 −0.853 −0.858 −0.091 −0.351
(0.150) (0.134) (0.134) (0.126) (0.989) (0.684) (0.742) (0.414)
Self-employed −0.169 −0.088 0.021 0.095 −0.281 −0.203 −0.372 −0.471
(0.105) (0.099) (0.099) (0.101) (0.677) (0.529) (0.433) (0.337)
Male  −0.167 −0.071 −0.156* 0.091 0.837* −0.140 −0.275 −0.081
(0.102) (0.095) (0.091) (0.088) (0.453) (0.360) (0.327) (0.260)
No. children 0.328*** 0.297*** 0.250*** 0.284*** 0.119 0.052 0.326*    0.040
(0.059) (0.056) (0.057) (0.060) (0.266) (0.180) (0.175) (0.140)
Constant −2.133 −1.518 −1.741 −1.600 0.274 0.399 −1.045 −1.175
(0.230) (0.198) (0.196) (0.192) (1.064) (0.751) (0.812) (0.499)
LR  test (age) 0.0000 0.0000 0.0000 0.0000 0.4628 0.0658 0.2827 0.0776
LR  test (income) 0.0699 0.0476 0.0004 0.0575 0.1520 0.0007 0.0000 0.0000
Observations 4047 4701 4836 4685 1773 2405 2399 2349
Standard errors in parentheses. Reference age: 70–75. Reference income: less than 15,000 real euros. Likelihood ratio (LR) test files report p-value of the joint significance
test  of the dummies, following Meng and Rubin (1992).
l
a
h
o
a
a
f
g
b
a
s
n
h
h
c
* p < 0.1.
** p < 0.05.
*** p < 0.01.
ikely to want a loan. Similarly, children raise the demand for loans
nd increase the rate of arrears. However, couples do not have
igher arrears even though they have higher credit demand. The
bserved effect of male on applications is consistent with their
rrears in Table 2 since the two waves where males had lower
rrears are also the two waves in which they are less likely to apply
or a loan. The effect of age is strongly significant; younger age-
roups are more likely to apply for credit (the LR test reported at the
ottom of the table shows that the differences across age-groups
re statistically significant). The results for unemployment are not
ignificant. This shows the high rate of arrears we  found earlier is
ot due to their high demand for loans since unemployed house-
olds are no more likely to apply for a loan than other types of
ousehold.
The last four columns in Table 3 look at whether the credit appli-
ation was accepted or rejected by the lender (e.g. whether theapplicant is credit constrained) in each wave of the survey. The
regression runs a Logit regression model where the left-hand side
variable is a dummy  variable for whether the loan is accepted, and
the regression is run using the sample of applicant households. The
estimated coefficients on the age-dummies suggest a hump shape
to credit constraints: middle-aged households are more likely to
be accepted than either the youngest or the oldest households (the
left-out group is the 70–74 group, and the largest coefficient is
always for households in their fifties). This is slightly surprising
since we  found that the oldest households have the lowest rate
of arrears. Couple is positive and significant in all but the first
wave, while the number of children is significant only in 2008,
even though Table 2 suggests children increase arrears. Similarly
university educated are just as likely to be refused credit as other
households despite their lower arrears.
4 inanc
g
h
c
d
t
t
a
a
t
c
c
s
t
a
2
o
W
r
m
m
r
h
t
t
l
g
h
c
i
a
a
t
5
a
i
o
o
p
a
s
w
c
p
t
h
a
m
a
t
s
t
t
p
v
i
c
6 C. Aller, C. Grant / Journal of F
The table suggests that there are some differences across income
roups in the acceptance rate on loans. Middle and high income
ouseholds seem to be more likely to have their application for
redit accepted than low income households (although the LR test
oes not reject the null in the 2002 wave). This is consistent with
he fact our results showed these households are the least likely
o enter arrears. Similarly, except in 2002, unemployed households
re more likely to be refused credit; Table 2 shows these households
re significantly more likely to be in arrears.
Fig. 3 plots the prediction of the level of applications and accep-
ances for a representative household through the four waves
onsidered in this study (taken, as before, at the median). We
onsider two explanatory variables: age and income. The figure
uggests that younger households are more likely to apply for credit
han older households. The figure also suggests that applications
mong younger households (below 60) are noticeably lower in
002 than in later years, and that applications from households
ver 50 peaked in 2005, before falling back in the last two waves.
e found that differences between age-groups in their rejection
ates by lenders was not significant, but the figure suggests there
ay  be a slight increase in rejection at the oldest ages. The figure
ore clearly suggests, however, there was a decline in acceptance
ates in the 2011 wave, particularly concentrated among the oldest
ouseholds. The final picture for the last wave seems paradoxical:
hose households (the oldest households) with the best propensity
o repay in Table 2 are the ones who became more rationed in the
oan market.
When we look at the estimate of applicants for different income
roups, we observe that in 2002 there were significantly fewer
ouseholds applying for loans at every income level, with appli-
ation rates continuing to drift upwards for higher income groups
n 2008 before falling slightly in 2011. The figure also shows that
cceptance rates were higher for higher income groups. Moreover,
cceptance rates fell dramatically for lower income households in
he 2008 and 2011 waves.
. Explaining the change in arrears
The regression results in Table 2 shows which factors affect
rrears in each year. While the results for unemployment and
ncome seem plausible and intuitive, other results, such as the effect
f male households, are surprising. However, the observed pattern
f arrears might be because, for instance, low income or unem-
loyed households do not have credit, and hence can not thus enter
rrears. The results also show that the effect of some of the variables
eems to have changed over the sample period. Earlier, in Table 1,
e found that the rate of arrears increased during the financial
risis, and we also found that there was an increase in the pro-
ortion of households that were credit constrained. We  would like
o understand how changes in borrower and in lender behaviour
as contributed to the change in arrears during the years before
nd after the crisis. Recall that for a household to enter arrears it
ust first apply for a loan; it must then have the loan application
ccepted; and then it must fail to repay the loan when required
o do so. A change in the arrears behaviour of a household with
ome given characteristics (we will denote the set of characteris-
ics as X) could be the consequence of a change in any of these
hree things. It would be useful to disentangle these three different
ossible explanations for the results we observe. Figs. 2 and 3 pro-
ided some indication of the effect of age and income on arrears;
n this section we will explore the role of age, income, and other
haracteristics in explaining arrears in more detail.ial Stability 36 (2018) 39–52
5.1. Design of the decomposition exercise
Let Di denote a dummy  for whether household i repays its debts
on schedule (it takes the value one if the household is in arrears
and zero otherwise). Similarly, let Ai denote a dummy  variable that
takes the value one of the household has applied for a loan (and
zero otherwise), and let Ci be a dummy  that takes the value one if
the household’s application for credit was accepted (and zero oth-
erwise). In each wave t (where we  have separate estimates in 2002,
2005, 2008 and 2011) we  can estimate the probability of arrears of
household i, given characteristics Xi as Prt (Di = 1|Xi). And we can
compare how the arrears behaviour of households differs in dif-
ferent waves. Note, however, we  can only observe a household in
arrears if it borrows, that is if Ai = 1 and Ci = 1, hence we  can make
the following decomposition
Prt(D ti = 1|Xi) = Pr (Di = 1|Ai = 1, Ci = 1, Xi)
(1)
· Prt(Ci = 1|Ai = 1, X ) ·Prti   (Ai = 1|Xi)
where each of the probabilities has already been estimated and
is reported in Tables 2 and 3. Using the coefficients reported in
Tables 2 and 3 and using Eq. (1), we can compute the probability of
arrears for each household in the sample, given their characteris-
tics. We  can then calculate the arrears rate in the whole population
in wave t by taking the weighted average of each individual house-
hold probability of arrears in that wave. Consequently we can
investigate how the arrears rate has changed over time.
Of fundamental interest, however, is to understand what has
caused the change in arrears rates over time. First note that the
arrears rate between two periods can change because the distri-
bution of underlying household characteristics has changed. To
calculate how changes in characteristics affect arrears between
two waves we can take the household characteristics for wave
t + 1 but use the estimated probability function for wave t (where,
slightly abusing notation, t we  write this as Pr (D|Xt+1)). Arrears can
also change because the composition of borrowers has changed.
Using Eq. (1), we  can decompose the overall change in the arrears
rate between two  periods for a household with characteristics X, to
changes in the arrears behaviour of borrowers, to changes in accep-
tance behaviour of lenders, and to changes in application behaviour
of households between those waves. Eq. (1) has written the prob-
ability of arrears t t in wave t as Pr (D|A = 1, C = 1, Xt) · Pr (C|A = 1,
Xt) · Prt(A|Xt). This formulation uses the estimates for application,
acceptance and borrower behaviour (or arrears behaviour of bor-
rowers) for wave t that have already been estimated. A measure
of the effect of changes in application behaviour of households can
be obtained by using the period t + 1 estimated application regres-
sion but with the period t characteristics. That is, to investigate
how changes in applications affects the overall level of arrears we
can calculate Prt(D|A t t+1 = 1, C = 1, Xt) · Pr (C|A = 1, Xt) · Pr (A|Xt) and see
how the estimated arrears rate changes when Prt(A|Xt) is replaced
by t+1 Pr (A|Xt). Similarly, we  can investigate the effect of changes in
lenders’ acceptance behaviour and of the repayment behaviour of
borrowers by, in turn, using the next period estimate for lenders’
acceptance t+1 behaviour Pr (C|A = 1, Xt), and arrears among borrow-
ers Prt+1 (D|A = 1, C = 1, Xt). Results from this thought experiment are
reported in the next section.
5.2. Results of the decomposition exercise
5.2.1. All sample
The top row of Table 4 reports the predicted arrears rate ineach wave of the data (compiled as the weighted sum of each
household’s predicted arrears rate using the logit estimates in
Tables 2 and 3). The table shows that predicted arrears fell from
9.3% to 8.3% between 2002 and 2005, before increasing to 9.5% in
C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52 47
Fig. 3. The rate of applications and acceptances by income and age. Note: The two  left-hand figures report the predicted probability of the household applying for a loan during
the  last two years, while the two right-hand figures report the predicted probability that t
logit  estimations (Table 3) applied to a reference household: headed by a man, 45–49 yea
income  sums up between 15,000 and 25,000 euros (in 2010 real terms) per year.
Table 4
Decomposing the change in default.
2002 2005 2008 2011
Prt(D|Xt) 9.28 8.31 9.45 11.10
Prt(D|Xt+1) 9.48 8.53 9.40 .
Prt(D|Xt) using:
Prt+1(A|Xt) 10.58 7.86 9.95 .
.  . . and Prt+1(C|A = 1, Xt) 10.47 7.37 9.72 .
.  . . and Prt+1   (D|C = 1, A = 1, Xt) 8.13 8.72 10.55 .
Notes: The first row calculates the predicted probability of default D using the
weighted sum of each household’s predicted default (using the logit regressions
for  each wave) over the observations with characteristics X in wave t. In the second
row, the probability of default is calculated by using the probability of default for
wave t but the observations from wave t + 1. The remaining rows report the effect
on default holding the characteristics fixed for wave t but using: (i) the probability
applying for a loan A in wave t + 1; (ii) and additionally the probability of receiving
c
i
2
r
t
c
c
t
9
8
e
r
f two rows of the decomposition exercise suggest that there was an
c
i
c
redit C in wave t + 1; and (iii) and additionally the probability of default given credit
n  wave t + 1.
008 and more sharply to 11.1% in 2011. This pattern closely mir-
ors the raw data in Table 1. The second row of Table 4 investigates
he effect of changing household characteristics in explaining the
hange in arrears over time. It shows that if we had used the 2005
haracteristics but the 2002 logit estimates for applications, accep-
ances and arrears then the arrears rate would have remained at
.5%. The table shows that the predicted rate of arrears in 2005 is
.3%, hence it means changes in household characteristics can not
xplain the change in arrears between these two waves. The top
ow of Table 4 shows that the predicted rate of arrears increased
rom 8.3% in 2005 to 9.5% in 2008. The second row shows that
hanges in characteristics alone increases arrears to 8.5%, mean-
ng, again, changes in characteristics had only a small role in the
hange in arrears between 2005 and 2008. Finally, this decomposi-he loan application is accepted rather than rejected. Predicted values obtained from
rs old that lives in couple, employed, house owner, no children and total household
tion for the change between the last two  waves shows that changes
in characteristics did not increase arrears, but arrears increased to
11.1%.
The results in Table 2 had shown that both income and unem-
ployment are significant when explaining which households enter
arrears. This is consistent with the results reported in Ampudia
et al. (2016). However, the results in Table 4 show that changes
in characteristics (of which income and unemployment are impor-
tant components) had little effect on the change in arrears between
waves. This is true even between the last two waves of the survey
despite the fact that the economic crisis resulted in larger unem-
ployment and lower income per capita in Spain in 2011 compared
to 2008 (the 2008 survey was mostly undertaken just prior to the
onset of the crisis in Spain). This result contradicts Blanco and
Gimeno (2012) who  argued that changes in unemployment explain
the increase in default. During this period, this result suggests that
additional insights can be gained exploring the change in behaviour
rather than the change in characteristics. The next three rows of
Table 4 separate the change in the level of arrears into a change in
the rate of applications; a change in the rate of acceptances; and a
change in the rate of arrears among borrowers where we  will hold
characteristics at their current level.
The table shows that using the 2005 coefficient estimates for
application behaviour of households, but the 2002 characteristics,
would have results in an increase in arrears rates from 9.3% to
10.6% (this is shown by the third row of the results for 2002).
Using the 2005 estimates of both applications and acceptances (the
fourth row) would have reduced this rate very slightly to 10.5%. Theincrease in the proportion of households borrowing, and that the
decline in the arrears rate between 2002 and 2005 was despite
the expansion in borrowing. The last row additionally investigates
48 C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52
Table 5
Decomposing the change in default by age-group.
30–34 years old 35–39 years old
2002 2005 2008 2011 2005 2005 2008 2011
Prt(D|Xt) 12.16 13.40 11.25 20.52 11.37 11.40 15.12 11.86
Prt(D|Xt) using:
Prt+1(A|Xt) 12.48 13.27 12.12 . 13.00 11.65 15.15 .
. . . and Prt+1   (C|A = 1, Xt) 12.27 10.97 13.10 . 13.14 11.19 14.05 .
. . . and Prt+1    (D|C = 1, A = 1, Xt) 11.23 10.37 18.10 . 11.43 14.29 12.58 .
40–44  years old 45–49 years old
2002 2005 2008 2011 2005 2005 2008 2011
Prt(D|Xt) 15.11 12.42 13.37 15.02 10.10 11.41 12.21 14.72
Prt(D|Xt) using:
Prt+1(A|Xt) 16.51 11.90 14.41 . 12.04 10.87 12.57 .
.  . . t+1  and Pr (C|A = 1, Xt) 16.26 11.30 14.23 . 11.93 11.01 11.59 .
.  . . and Prt+1 (D|C = 1, A = 1, Xt) 12.64 12.76 13.87 . 11.43 10.73 13.62 .
50–54  years old 55–59 years old
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 8.66 7.01 7.62 13.55 6.78 3.67 9.21 9.73
Prt(D|Xt) using:
Prt+1(A|Xt) 10.44 5.70 9.01 . 7.92 3.17 9.06 .
.  . t+1 . and Pr (C|A = 1, Xt) 10.42 5.15 9.65 . 7.97 2.94 8.69 .
. . t+1  . and Pr (D|C = 1, A = 1, Xt) 6.25 7.48 12.35 . 3.87 8.43 9.19 .
60–64  years old 65–75 years old
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 6.77 2.09 3.93 4.74 4.82 4.42 3.50 2.55
Prt(D|Xt) using:
Prt+1(A|Xt) 8.41 1.92 4.54 . 5.76 3.81 3.62 .
. . . and Prt+1    (C|A = 1, Xt) 8.09 1.75 4.39 . 5.58 3.72 3.33 .
.  . . t+1 and Pr (D|C = 1, A = 1, Xt) 2.21 3.94 4.74 . 4.81 3.31 2.75 .
Notes: For each sub-sample, the first row calculates the predicted probability of default D using the weighted sum of each household’s predicted default (using the logit
r e rem
b the pr
o
t
2
f
a
f
r
t
b
c
c
2
h
t
a
i
a
b
1 younger households. For example, in 2002, although arrears are
t highest among households aged 40–44 (at 15.1%), the pattern oth-
t
r
r
i
b
egressions for each wave) over the observations with characteristics X in wave t. Th
ut  using: (i) the probability applying for a loan A in wave t + 1; (ii) and additionally 
f  default given credit in wave t + 1.
he effect of changes in the arrears among borrowers (using the
002 characteristics), which shows that the arrears rate would have
allen to 8.1%. The results in this column shows that between 2002
nd 2005 the increase in borrowing was countered by a much larger
all in the arrears rate among borrowers, and this change in bor-
ower behaviour is the major explanation for the fall in arrears in
he whole population.
We found that nearly a quarter of the increase in arrears
etween 2005 and 2008 can be explained by changes in household
haracteristics. However, the second column of Table 4 shows that
hanges in applications and changes in acceptances (e.g. using the
008 logit estimates but the 2005 characteristics) together would
ave reduced the arrears rate from 8.3% to 7.4% as credit condi-
ions slightly tightened. It shows that a great deal of the change in
rrears in the top row of the table can be explained by a rebound
n the arrears among of borrowers (the bottom row) between 2005
nd 2008. However, the full change in arrears requires a change in
oth characteristics and in the repayment behaviour of borrowers.
The top row of Table 4 shows how arrears increased from 9.5% to
1.1% between 2008 and 2011. The surge in arrears between these
wo years can partly be explained by an increase in credit applica-
ions, which would have raised arrears to 9.9%. However, the fourth
ow shows lenders reduced the availability of credit, which slightly
educed the default rate.5 The most substantial contribution to the
5 Blanco and Gimeno (2012) associate, for almost the same period of time, the
ncrease in credit with a fall in default ratios that are related to the sample of
orrowers rather than the whole population as in our case.aining rows report the effect on default holding the characteristics fixed for wave t
obability of receiving credit C in wave t + 1; and (iii) and additionally the probability
increase in default in the top row of the table between 2008 and
2011, however, was  due to changes in the arrears behaviour of bor-
rowers (shown by the bottom row of the 2008 column).6   Overall,
the table suggests that changes in the repayment behaviour of bor-
rowers (rather than changes in the pool of borrowers) are the most
important explanation for the changes in arrears over the period
between 2002 and 2011. Nevertheless, changes in characteristics
are also a necessary part of the explanation.
5.2.2. Age
In Table 5 we investigate the differences in arrears across age-
groups. The table reports results for each of the 5-year age groups
that have been used in the analysis (except that we  have merged
to two oldest age-groups as there are relatively fewer older house-
holds). For each group (for example households between 30 and
34 years old) we have calculated their predicted level of arrears
given their other characteristics. The general pattern shows that,
on the whole, older households have lower rates of arrears thanerwise shows that the rate of arrears fell from 12.2% for households
6 Blanco and Gimeno (2012) argue that unemployment is the main explanation
for  the surge in default rate in these years, (e.g. characteristics are the most impor-
tant explanation); Ampudia et al. (2016) make a similar argument. These authors
argue the debt-to-income ratio plays an important role in explaining the evolution
of  default rate.
C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52 49
Table  6
Decomposing the change in default by income.
<15,000 euros 15,000–25,000 euros
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 15.05 13.08 8.73 15.44 10.56 10.95 11.65 12.62
Prt(D|Xt) using:
Prt+1(A|Xt) 17.93 11.32 10.37 . 12.46 10.30 12.17 .
. . . and Prt+1   (C|A = 1, Xt) 17.28 9.19 10.32 . 12.26 10.13 11.45 .
.  . . and Prt+1 (D|C = 1, A = 1, Xt) 12.80 7.64 13.31 . 11.23 11.43 12.38 .
25,000–35,000 euros 35,000–45,000 euros
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 9.40 7.60 11.87 13.71 8.87 7.19 9.81 8.61
Prt(D|Xt) using:
Prt+1(A|Xt) 10.00 8.14 11.55 . 9.89 6.71 10.51 .
.  . . and Prt+1 (C|A = 1, Xt) 9.92 8.12 11.23 . 9.96 6.60 10.63 .
.  . . and Prt+1 (D|C = 1, A = 1, Xt) 7.90 11.07 13.97 . 7.34 8.68 9.43 .
45,000–57,000 euros >57,000 euros
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 8.48 4.19 7.95 6.53 3.94 3.05 5.20 3.98
Prt(D|Xt) using:
Prt+1(A|Xt) 9.76 4.36 8.56 . 4.23 3.05 5.21 .
.  . . and Prt+1(C|A = 1, Xt) 10.04 4.33 8.44 . 4.25 3.06 5.17 .
.  . t+1 . and Pr (D|C = 1, A = 1, Xt) 4.87 6.97 6.70 . 3.41 5.03 4.35 .
Notes: For each sub-sample, the first row calculates the predicted probability of default D using the weighted sum of each household’s predicted default (using the logit
r e rem
b the pr
o
a
d
a
g
r
3
(
t
a
o
c
t
r
t
a
i
h
a
t
t
m
o
t
m
s
t
a
a
t
f
v
i
a
5
egressions for each wave) over the observations with characteristics X in wave t. Th
ut  using: (i) the probability applying for a loan A in wave t + 1; (ii) and additionally 
f  default given credit in wave t + 1.
ged 30–34 to 4.8% for households aged 65–75 when investigating
ifferences across age groups.
Three of the four age-groups under 50 saw a slight increase in
rrears between 2002 and 2005. In contrast, all household age-
roups over 50 decreased their rates of arrears, even though their
ates were already lower in 2002. For the youngest age-group aged
0–34 years old, although there was an increase in applications
the second row shows applications increased arrears to 12.5%),
here was also a reduction in acceptances (the third row shows
rrears falling to 12.3%), hence a small increase in their incidence
f borrowing. However, they sharply reduced their level of arrears
onditional on getting a loan; the fourth row shows arrears falling
o 11.2%. Moving from the bottom row of the 2002 column to the top
ow of the 2005 column shows the effect of changes in characteris-
ics between these two waves, showing characteristics increased
rrears from 11.2% to 13.4%. Thus the much larger countervail-
ng effect of changes in characteristics meant that overall these
ouseholds increased their level of arrears. The other age-groups
ll saw a slight increase in arrears due to an increase in applica-
ions between 2002 and 2005 (shown by the difference between
he first and the second row in each case) which was mostly accom-
odated by lenders (the changes in acceptances had little effect
n arrears since there is little difference between the second and
hird row for 2002). They also all saw an improvement in the repay-
ent behaviour of borrowers as the bottom row of the 2002 column
hows a reduction in arrears compared to the third row. As a result
he 35–39 and 45–49 age groups saw only modest increases in over-
ll arrears, while the other age-groups saw a reduction in overall
rrears; and changes in characteristics are not an important part of
he story for these other age-groups (going from the bottom row
or 2002 to the top row for 2005 barely changes arrears).
A less clear pattern is observed for the period 2005–2008. The
ery youngest and oldest cohorts experienced a modest decline
n their arrears, while the other age-groups all saw their rate of
rrears increase (where the increase was substantial for the 35–39,
5–59 and 60–64 age-groups). All but the 35–39 age-group sawaining rows report the effect on default holding the characteristics fixed for wave t
obability of receiving credit C in wave t + 1; and (iii) and additionally the probability
a reduction in applications and all groups except the 45–49 age-
group saw a reduction in acceptances. Borrowers became less likely
to repay between 2005 and 2008 for all age-groups except 30–34,
45–49 and 65–75 households, with particularly large effect for
some groups. But changes in characteristics are also important for
younger households.
Finally, we  analyse the period 2008–2011, in which most groups
experienced a surge in their arrears rate. The youngest house-
holds (30–34 years old) shows the largest increment, since arrears
increased from 11.3% to 20.5%; while all factors contribute to that
rise, changes in borrower behaviour (the difference between the
third and fourth row) is the most important. A similar pattern
is apparent for 50–54 years old households: the large increase
in arrears of over 5% can partially be attributed to changes in
characteristics, applications and acceptances, but arrears among
borrowers makes the largest contribution to this increase. The table
shows that 45–49 years old households experience a similar surge
in arrears of 2.5%, which is mostly explained by the worsening
repayment behaviour of borrower households that more than com-
pensates for the more restrictive lending behaviour of banks. For
60–64 years old households, repayment and application behaviour
by households, as opposed to the restrictive granting behaviour by
banks, explain most of the increase in the overall arrears rate. The
results for 65–75 years old households show the opposite results,
as for this age-group the overall level of arrears fell between 2008
and 2011; although applications slightly increased (the second row
shows they increased to 3.6%), it is more than cancelled by the bet-
ter characteristics and repayment behaviour of borrowers as well
as the better granting behaviour by banks to result in the decline
of their overall rate of arrears. The other group that experienced a
decline in arrears is the 35–39 years old age-group; in this case the
fall is explained by their better repayment behaviour (the move-
ment from the third to the fourth row) and by the more restrictive
granting behaviour by banks (the reduction in arrears by moving
from the second to the third row).
50 C. Aller, C. Grant / Journal of Financial Stability 36 (2018) 39–52
Table 7
Decomposing the Change in default by employment status, house ownership and education.
Employed Unemployed
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 9.80 9.03 9.49 12.71 29.11 21.11 22.75 20.99
Prt(D|Xt) using:
Prt+1(A|Xt) 11.12 8.82 9.70 . 33.06 17.86 25.11 .
. . . and Prt+1   (C|A = 1, Xt) 11.21 8.43 9.65 . 30.56 14.08 22.80 .
. . . and Prt+1    (D|C = 1, A = 1, Xt) 8.71 9.45 12.35 . 20.21 18.36 22.04 .
Retired Self-employed
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 4.94 4.42 3.80 3.44 7.97 8.34 12.29 9.67
Prt(D|Xt) using:
Prt+1(A|Xt) 5.76 3.81 4.21 . 9.17 8.27 13.07 .
.  . . and Prt+1(C|A = 1, Xt) 5.61 3.72 3.94 . 9.14 7.70 13.51 .
.  . . and Prt+1(D|C = 1, A = 1, Xt) 4.68 3.87 3.43 . 7.84 12.79 8.92 .
University education Non university education
2002 2005 2008 2011 2002 2005 2008 2011
Prt(D|Xt) 3.75 3.15 2.99 3.21 10.44 9.55 10.97 13.27
Prt(D|Xt) using:
Prt+1(A|Xt) 4.13 3.31 2.84 . 11.94 8.95 11.62 .
.  . . and Prt+1 (C|A = 1, Xt) 4.07 3.15 2.80 . 11.81 8.38 11.34 .
.  . . and Prt+1(D|C = 1, A = 1, Xt) 3.26 3.03 3.40 . 9.15 10.08 12.24 .
Notes: For each sub-sample, the first row calculates the predicted probability of default D using the weighted sum of each household’s predicted default (using the logit
r e rem
b the pr
o
5
t
b
F
(
t
l
s
r
i
2
r
i
2
f
a
a
(
b
s
w
h
f
i
o
i
a
i
t
r
w
t
f
egressions for each wave) over the observations with characteristics X in wave t. Th
ut  using: (i) the probability applying for a loan A in wave t + 1; (ii) and additionally 
f  default given credit in wave t + 1.
.2.3. Income
Table 6 divides the sample into six different income groups. The
op row of each sub-table shows that the level of arrears decreased
etween 2002 and 2005 for all except the 15–25,000 income group.
or all income groups, there was an increase in the application rate
the predicted arrears in the second row increases compared to
he top row). While lenders reduced their acceptance rate for the
owest three income groups(shown by the difference between the
econd and third row), there is a slight increase in the acceptance
ate of for the other income groups. All income groups saw a fall
n the rate at which actual borrowers fell into arrears between
002 and 2005 (the move to the bottom row in each sub-table
educes arrears), which explains most of the fall in arrears reported
n the top row for each group (e.g. the move from the results for
002–2005).
Between 2005 and 2008 the overall rate if arrears fell sharply,
rom 13.1% to 8.7%, for the lowest income group. This fall in arrears
mong the poorest households can be explained by the large fall in
pplications (reducing arrears to 11.3%), in acceptance by lenders
further reducing arrears to 9.2%), and in the repayment of loans
y borrowers (reducing arrears to 7.6%). All other income groups
aw an increase in the rate of arrears between 2005 and 2008,
ith especially large increases among middle income groups as
ouseholds with an income of 25–35,000 euros increased arrears
rom 7.6% to 11.9%, households with income of 35–45,000 euros
ncreased arrears from 7.2% to 9.8%, while households with income
f 45–57,000 euros increased arrears from 4.2% to 8.0%. Changes
n application behaviour of households(shown by the change in
rrears in the second row) do not explain this increase since the
ncrease in the rate of applications is rather small (and applica-
ions actually fell for the 35–45,000 euro group). Similarly, the third
ow shows that there was also a small reduction in the rate at
hich applicants received credit. Arrears among borrowers fell forhe lowest income group. However, the change in overall arrears
or all the other income groups between 2005 and 2008 is mostlyaining rows report the effect on default holding the characteristics fixed for wave t
obability of receiving credit C in wave t + 1; and (iii) and additionally the probability
attributable to the increase in the rate of arrears among borrowers
(shown by the increase in arrears in the bottom row).
Between 2008 and 2011, all except the 25–35,000 euro income
group saw an increase in credit applications (the effect on arrears
is shown in the second row), and almost all groups saw a substan-
tial in the rate at which credit was  granted (shown by the third
row). The rate of arrears among actual borrowers, shown by the
bottom row, increased sharply for the lowest three income groups,
but improved for the three highest income groups. Changes in char-
acteristics (shown by the move from the bottom row of the 2008
column to the top row of the 2011 column) are never a partic-
ularly important factor in explaining the changes between 2008
and 2011, although they play some part in the increase in arrears
for the lowest income households and the reduction in arrears for
the 35–45,000 group. By far the largest part of the explanation of
the changes in overall arrears is due to changes in the repayment
behaviour of borrowers.
5.2.4. Employment and education
The top four panels of Table 7 investigates four different
employment groups: employees, self-employed, unemployed (or
households outside employment) and retirees. The pattern of
arrears for employees is very similar to the pattern for the whole
sample reported in Table 4. The top row of the sub-table shows
that the overall rate of arrears fell between 2002 and 2005, before
increasing in 2008 and then, for employed households, increasing
again in 2011. As in Table 4, the fall in arrears between each wave
is mainly attributable to the changing arrears behaviour of bor-
rowers (shown by the move from the third and the fourth row
of the panel). The unemployed group have much higher rates of
arrears in all the waves, but the top row shows this rate fell sharply
between 2002 and 2005. The bottom row of the panel shows this is
due to a sharp reduction in the rate at which unemployed borrow-
ers entered arrears. Their small increase in arrears between 2005
and 2008 is mainly due to changes in the households’ characteris-
tics (the second row shows there was a reduction in applications
inanc
w
i
c
b
e
s
C
I
b
(
c
a
r
i
2
e
a
a
a
t
f
t
w
r
i
u
t
w
h
t
r
b
t
a
e
s
t
a
s
r
i
t
s
a
e
i
a
a
c
6
b
a
O
h
t
c
B
c
C. Aller, C. Grant / Journal of F
hich would otherwise have reduced arrears). Their small decrease
n overall arrears between 2008 and 2011 is due to a reduction in
redit acceptances (shown in the third row) and in arrears among
orrowers (shown in the fourth row).
Retired households have the lowest rate of arrears among all
mployment categories, and the top row of this panel shows they
teadily reduced their rate of arrears throughout the sample period.
hanges in characteristics played very little part in this decline.
nstead, it can be explained by a decline in the rate of arrears among
orrowers between 2002 and 2005 and between 2008 and 2011
which in both cases outweighed the effect of an increase in appli-
ations). Between 2005 and 2008 there was a small reduction in
pplications which reduced their overall arrears rate.
Except in 2008, self-employed households had lower arrears
ates than employees. However, the pattern of changes over time
s rather different: their arrears increase steadily from 2002 and
005 until 2008, and then fell in 2011. The second row of the self-
mployment panel shows that the increase in arrears between 2002
nd 2005 is largely due to an increase in the rate of applications,
s the fourth row shows self-employed borrowers reduced their
rrears. In contrast, the increase in arrears in 2008 is mainly due to
he very large increase in arrears among borrowers (shown in the
ourth row), as there was a reduction in applications and accep-
ances in 2008. This pattern was reversed in 2011, where there
as an increase in applications and acceptances and a very sharp
eduction in arrears by borrowers leading to an overall reduction
n arrears.
The bottom two panels of Table 7 contrast households with a
niversity education with those who did not go to college. The
op row shows that university educated households are, in all
aves, much less likely to enter arrears. University educated house-
olds experienced a decline in arrears from 2002 to 2008, where
he increasing tendency to apply for loans (shown in the second
ow) is more than compensated by an improvement in repayment
ehaviour among borrowers (shown in the fourth row). In 2011
here was a very small reduction in the rate of applications, and
 small increase in the rate of non-payment by borrowers. How-
ver, changes over time for college educated households are rather
mall.
Non-university educated households see larger changes over
ime: they reduced the rate of arrears from 10.4% to 9.6% in 2005,
nd increased their arrears to 11.0% in 2008 and then, more sub-
tantively to 13.3%, in 2011. The third row of the panel shows that
ate of applications increased in 2005 and again in 2011, but fell
n 2008, while the third row shows there were also small reduc-
ions in the rate of acceptances over time. However, the fourth row
hows that most of the changes in the overall rate of arrears in 2005
nd 2008 was  due to changes in the rate of arrears among borrow-
rs, which fell in 2005, but increased in 2008. In 2011, increases
n applications, in arrears among borrowers, and in characteristics
ll caused similarly large increases in the overall change in arrears
mong non-college educated households which was  only partially
ounter-balanced by a reduction in credit acceptances by lenders.
. Conclusion
This paper contributes to our understanding the arrears
ehaviour of Spanish households before and after the recent crisis,
n issue that has attracted the attention of a number of researchers.
ur paper complements the analysis made in other papers that
ave hitherto used aggregate data or data from lenders’ adminis-
rative records. We  utilise households’ self-reported information
ollected by the Survey of Household Finances provided by the
ank of Spain for the years before and immediately after the recent
risis, and, to the best of our knowledge, this is the first paper toial Stability 36 (2018) 39–52 51
exploit such data. Thus we  can not only look at which households
are more likely to enter arrears, but also how this has changed
over the sample period. Moreover, there is separate information
on applications, acceptances and arrears (however, note the differ-
ent time frame of these questions as applications and acceptances
are over the last two years while arrears are during the last year).
These questions allow us to distinguish between different expla-
nations for the changes in the level of arrears observed for Spanish
households. Our paper conducts a decomposition exercise where it
apportions the rise in default to a part caused by changes in applica-
tions, a part caused by changes in acceptances, and a part caused by
changes in borrower behaviour, as well as a part caused by changes
in household characteristics.
The raw results show that the rate of arrears fell between
2002 and 2005, and then rose in 2008 and again in 2011. Blanco
and Gimeno (2012) and Ampudia et al. (2016) highlighted the
role of unemployment and low wealth in explaining arrears. Our
regression results also show that lower income and unemployed
households are more likely to enter arrears, as are younger house-
holds and households with lower levels of education. However,
these effects are apparent in all four waves included in the analy-
sis, and hence in themselves do not explain the changes in the level
of arrears during and after the crisis: although we also find that
changes in unemployment and income form part of the explana-
tion, they are too small to be able to explain all of the changes that
are observed in the data. Thus the results reported in this paper
do not provide much support for the argument put forward by
Blanco and Gimeno (2012) for Spain (or for Foote et al., 2009, for
US households).
A key contribution of our analysis is the decomposition exer-
cise. The results suggest that the overall rate at which Spanish
households enter arrears fell between 2002 and 2005 and that
this occurred despite the increase in credit applications which was
mostly met  by lenders. Similarly, the increase in arrears between
2005 and 2008 happened even though applications fell between
these two years. Maddaloni and Peydró (2011) and Díaz-Serrano
(2015) both suggested that weakening lending standards were an
explanation for the increase in arrears during the crisis (as did Mian
and Sufi, 2009, and Demyanyk and Van Hemert, 2011, for US house-
holds). However, the evidence presented here does not support
their argument. Dell’Arricia et al. (2012) argue instead that there
was an increase in credit demand among American households
prior to the sub-prime crisis. However, our study shows that appli-
cations fell between 2005 and 2008, the period in which arrears
increased. Rather than the change in arrears being driven by either
increases in credit demand, or a softening in lending standards, this
study shows that changes in the arrears behaviour of households
given credit drove these changes. Between 2008 and 2011 there was
an increase in the incidence of applications among Spanish house-
holds and in arrears among those receiving a loan, which together
explain a substantial proportion of the change in arrears between
these two  years.
These results suggest that changes in the behaviour of actual
borrowers is driving the arrears rate of Spanish households, rather
than changes in the type of households that borrow. Our findings
thus seem to support the claim made by Guiso et al. (2013) for
US households; borrowers became more willing to enter arrears
regardless of their circumstances. Note that since applications and
acceptances fell when arrears increased, we  cannot attribute this
increase to a change in the composition of borrowers with new
borrowers being higher credit risks since our results showed that
fewer households were borrowing. Although the decomposition
exercise can not explain the reason for the increased willingness
to default which Guiso et al. (2013) suggest, we believe it is consis-
tent with a decline in the stigma attached to default over time, and
5 inanc
a
d
e
p
i
h
h
e
h
h
v
a
a
t
l
t
t
l
s
m
b
b
g
f
a
d
M
h
r
h
s
m
t
R
A
A
B
B
Saurina, J., 2009. Loan loss provisions in Spain. A working macroprudential tool.
Revista de Estabilidad Financiera of the Bank of Spain 17, 11–26.
B
B
2 C. Aller, C. Grant / Journal of F
n increase in the sympathy with which the civil courts deal with
ebtor households.
The overall picture of arrears masks substantial variation in the
xperience of arrears among different household types. For exam-
le, the oldest households in the sample did not see an increase
n their rate of arrears, while for some (but not all) of the younger
ouseholds, the increase in arrears was substantial; middle-income
ouseholds increased their rate of arrears although the very rich-
st and poorest households did not; while unemployed households
ave higher rates of arrears, only employed and self-employed
ouseholds increased the rate at which they entered arrears; uni-
ersity educated households had little change in their rate of
rrears, but poorly educated households both had higher arrears,
nd their rates increased substantially through this period.
For most groups, repayment behaviour among borrowers drove
hese changes in arrears, although there is some evidence that
enders reduced credit to low income households. However, all but
he highest income groups increased their application rate during
he crisis, as did middle-aged households. Hence the results high-
ight a puzzle: why did lenders not react by reducing lending more
ubstantively? Moreover, given that the income and unemploy-
ent characteristics are not driving the change in arrears among
orrowers, we cannot explain why borrowers did change their
ehaviour.
Lastly, we make a note of caution. First, we have been investi-
ating arrears, which we earlier noted covers a range of behaviour
rom being a few days late on a single payment, to facing legal
ction for the recovery of debt. It may  well be true that each inci-
ent of arrears has become more serious following the recession.
oreover, throughout the analysis, we only capture whether the
ousehold is borrowing and has repaid (the extensive margin)
ather than the size of the loans (the intensive margin). This paper
ighlights that changes in the composition of borrowers is not
ubstantively explaining changes in arrears, hence future research
ight usefully analyse changes in loan size and its contribution to
he debt crisis for Spanish households during the last few years.
eferences
kin, O., García Montalvo, J., García Villar, J., Peydró, J.L., Raya, J.M., 2014. The real
estate and credit bubble: evidence from Spain. SERIEs 5 (2–3), 223–243, http://
dx.doi.org/10.1007/s13209-014-0115-9.
mpudia, M.,  van Vlokhoven, H., Zochowski, D., 2016. Financial fragility of euro
area households. J. Financ. Stab. 27, 250–262, http://dx.doi.org/10.1016/j.jfs.
2016.02.003.
ernardino, A.C., Gutiérrez, P.J., 2012. El crédito al sector privado en España: del
crecimiento desbordado a la fuerte restricción. REVESCO 109, 13–44, http://dx.
doi.org/10.5209/rev REVE.2012.v109.40651.
ernardino, A.C., Martín de Vidales, I., 2014. Crisis y cambios estructurales en el
sector bancario español: Una comparación con otros sistemas financieros.
Estudios de Economía Aplicada 32 (2), 197–213.lanco, R., Gimeno, R., 2012. Determinants of default ratios in the segment of loans
to  households in Spain. In: Banco de España Working Paper, No 1210., http://
dx.doi.org/10.2139/ssrn.2007059.
over, O., Casado, J.M., Costa, S., Du Caju, P., McCarthy, Y., Sierminska, E.,
Tzamourani, P., Villanueva, E., Zavadil, Y., 2016. The distribution of debt acrossial Stability 36 (2018) 39–52
euro area countries: the role of individual characteristics, institutions and
credit conditions. Int. J. Cent. Bank. 12, 71–128 http://www.ijcb.org/journal/
ijcb16q2a3.htm.
Bover, O., Coronado, E., Velilla, P., 2014. The Spanish Survey of Household Finances
(EFF): description and methods of the 2011 wave. In: Occasional paper no.
1407. Banco de Espa na., http://dx.doi.org/10.2139/ssrn.2514206.
Campbell, J.Y., Cocco, J.F., 2015. A model of mortgage default. J. Finance 70,
1495–1554, http://dx.doi.org/10.1111/jofi.12252.
Cecchetti, S.G., Mohanty, M.S., Zampolli, F., 2011. The real effects of debt. In: BIS
Working Papers, No 352. https://ssrn.com/abstract=1946170.
Christelis, D., Georgarakos, D., Haliassos, M.,  2013. Differences in portfolios across
countries: economic environment versus household characteristics. Rev. Econ.
Stat.  95 (1), 220–236, http://dx.doi.org/10.1162/REST a 00260.
Crook, J., 2006. Household debt demand and supply. In: Bertola, G., Disney, R.,
Grant, C. (Eds.), The Economics of Consumer Credit. MIT Press, pp. 63–92.
Dell’Arricia, G., Igan, D., Laeven, L., 2012. Credit booms and lending standards:
evidence from the subprime mortgage market. J. Money Credit Bank. 44,
367–384, http://dx.doi.org/10.1111/j.1538-4616.2011.00491.x.
Demyanyk, Y., Van Hemert, O., 2011. Understanding the sub-prime mortgage crisis.
Rev. Financ. Stud. 24 (6), 1848–1880, http://dx.doi.org/10.1093/rfs/hhp033.
Díaz-Serrano, L., 2015. Banks, Lies and Bricks: The Determinants of Home Value
Inflation in Spain during the Housing Boom. CREIP Working Paper, 09-2015.
Duygan, B., Grant, C., 2009. Household arrears: what role do institutions play?
Econ. Policy 57, 107–140, http://dx.doi.org/10.1111/j.1468-0327.2009.00215.x.
Elul, R., Souleles, N.S., Chomsisengphet, S., Glennon, D., Hunt, R., 2010. What
‘triggers’ mortgage default? Am.  Econ. Rev. 100 (2), 490–494, http://dx.doi.org/
10.1257/aer.100.2.490.
Foote, C., Gerardi, K., Goette, L., Willen, P., 2009. Reducing foreclosure: no easy
answers. NBER Macroecon. Annu. 24, 89–138.
Gerlach-Kristen, P., Lyons, S., 2015. Mortgage arrears in Europe: the impact of
monetary and macroprudential policies. In: Swiss National Bank Working
Paper.
Grant, C., Padula, M.,  2016. The repayment of unsecured debt by European
households. J. R. Stat. Assoc. (Ser. A) 181, 59–83, http://dx.doi.org/10.1111/rssa.
12248.
Guiso, L., Sapienza, P., Zingales, L., 2013. The determinants of attitudes toward
strategic default on mortgages. J. Finance 68 (4), 1473–1515, http://dx.doi.org/
10.1111/jofi.12044.
Igan, D., Leigh, D., Simon, J., Topalova, P., 2014. Dealing with household debt. In:
Financial Crises: Causes, Consequences, and Policy Responses. International
Monetary Fund.
Keys, B.J., Mukherjee, T., Seru, A., Vig, V., 2010. Did securitization lead to lax
screening? Evidence from subprime loans. Q. J. Econ. 125 (1), 307–362, http://
dx.doi.org/10.1162/qjec.2010.125.1.307.
Maddaloni, A., Peydró, J.L., 2011. Bank risk-taking, securitization, supervision, and
low  interest rates: evidence from the Euro-area and the US lending standards.
Rev.  Financ. Stud. 24 (6), 2121–2165, http://dx.doi.org/10.1093/rfs/hhr015.
Mayer, C., Pence, K., Sherland, S.M., 2009. The rise in mortgage defaults. J. Econ.
Perspect. 23 (1), 27–50 http://www.jstor.org/stable/27648293.
Meng, X.L., Rubin, D.B., 1992. Performing likelihood ratio tests with
multiply-imputed data sets. Biometrika 79 (1), 103–111, http://dx.doi.org/10.
1093/biomet/79.1.103.
Mian, A., Sufi, A., 2009. The consequences of mortgage credit expansion: evidence
from the U.S. mortgage default crisis. Q. J. Econ. 124 (4), 1449–1496, http://dx.
doi.org/10.1162/qjec.2009.124.4.1449.
Rubin, D.B., 1987. Multiple Imputation for Nonresponse in Surveys. John Wiley &
Sons, http://dx.doi.org/10.1002/9780470316696.
Sánchez-Martínez, M.T., Sánchez-Campillo, J., Moreno-Herrero, D., 2016. Mortgage
debt and household vulnerability: evidence from Spain before and during the
global financial crisis. Int. J. Housing Markets Anal. 9 (3), 400–420, http://dx.
doi.org/10.1108/IJHMA-07-2015-0038.Schwartz, E.S., Torous, W.N., 1993. Mortgage prepayment and default decisions: a
Poisson regression approach. Real Estate Econ. 21, 431–449, http://dx.doi.org/
10.1111/1540-6229.00619.