Content uploaded by Adam Banai
Author content
All content in this area was uploaded by Adam Banai on Jan 02, 2017
Content may be subject to copyright.
5
Financial and Economic Review, Vol. 15 Issue 4., December 2016, pp. 5–44.
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
In recent years, the average spread on newly extended housing loans above the
3-month interbank interest rate has been consistently higher compared to spreads
in neighbouring countries. This paper invesgates the reasons behind it by using
econometric tools and simple stascal examinaons. In our two-step approach,
we rst idenfy the determinants of spreads based on Hungarian transacon-level
and bank-level data, and then examine the Hungarian banking system’s sectoral
performance relave to other European countries in the main determinants
idened. Our ndings reveal that the higher spreads currently mainly stem from
the high proporon of products with inial rate xaon of over one year, the
relavely large stock of non-performing loans, and credit losses. High operang
costs in internaonal comparison may also have an impact on the seng of spreads.
According to our esmates, demand-side aributes also contribute to the emergence
of high spreads, as does the low level of compeon in certain regions.
G02, G20, G21
: new loan contracts, housing loan, interest rate spread, spread
Interest rate level of household loans plays a pivotal role in shaping households’
nancial decisions. The interest rate, which is in fact the cost of funding, denes
— along with the loan amount and maturity — the burden that debt servicing
represents for the borrower, and thus a relavely higher interest rate can hinder
a signicant poron of households from accessing credit. Given that the Hungarian
populaon tends to prefer property ownership as opposed to property rental (MNB
2016), the pricing of housing loans is of parcular importance in Hungary.
In recent years, the average spread on newly contracted HUF-denominated housing
loans has signicantly exceeded the spreads seen in other regions of Europe (in
* The views expressed in this paper are those of the author(s) and do not necessarily reect the ocal view
of the Magyar Nemze Bank.
Ákos Aczél is a nancial modeller at the Magyar Nemze Bank. E-mail: aczela@mnb.hu.
Ádám Banai is Head of Department at the Magyar Nemze Bank. E-mail: banaia@mnb.hu.
András Borsos is a PhD student at the Central European University. E-mail: andras.borsos@gmail.com.
Bálint Dancsik is an analyst at the Magyar Nemze Bank. E-mail: dancsikb@mnb.hu.
The manuscript was received on 6 October 2016.
6Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
this study, the spread refers to the dierence between the interest rate on loans
and the 3-month interbank interest rate). Although the dierence between the
average annual percentage rate (APR) on new housing loans and the 3-month
money market interest rate has narrowed materially since 2014, the spread sll
exceeds the regional average by 1.6 percentage points and the euro area average
by 1.8 percentage points (Figure 1).
Seng the interest rate is a complex process that depends both on the instuonal
background of a country and its banking system and the bank’s own aributes
(Figure 2). The interest rates applied must be capable of covering the bank’s costs
associated with lending (Buon et al. 2010).
Funding costs. Financial instuons fund their operaons through other economic
agents, and so the price of the funds they receive plays a role in seng the price at
which they lend credit. The price of funds may dier based on loan type, maturity
and type of interest rate. Deposits are generally the most stable and cheapest form
of funding for loans. In addion, covered bonds, of which mortgage bonds constute
a subcategory, also play a major role in several countries (EMF 2012). Prior to the
onset of the crisis, securisaon was on the rise across Europe (ECB 2009), however
it fell short of the degree observed in the United States. Funding costs can also be
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
Percentage point Percentage point
Hungary
Czech Republic
Romania
Slovakia
Poland
Slovenia
Euro area
2008 Q1
Q2
Q3
Q4
2009 Q1
Q2
Q3
Q4
2010 Q1
Q2
Q3
Q4
2011 Q1
Q2
Q3
Q4
2012 Q1
Q2
Q3
Q4
2013 Q1
Q2
Q3
Q4
2014 Q1
Q2
Q3
Q4
2015 Q1
Q2
Q3
Q4
2016 Q1
Note: APR-based spreads. Newly extended loans.
Source: MNB, national central banks.
7
Idenfying the determinants of housing loan margins in the Hungarian banking system
shaped by various state subsidies. Housing loan support schemes are common, for
instance, somemes in the form of liabilies side interest subsidies. Due to these
factors, it is very likely that the internaonal comparison of spreads calculated based
on interbank rates contain biases.
Interest rate risk. The diverging interest rates on assets and liabilies represents
a risk linked to, but disnct from funding costs. The various countries dier
according to (1) the interest rate characterisc of various transacons and (2) the
other unique characteriscs associated with mortgage lending within the region.
For instance, transacons with rates xed over the longer term are predominant
in Belgium, Germany and France, while products that are repriced within one year
are predominant in Portugal, Poland and Ireland. The stability over me of the
proporon of various interest-bearing products also diers: while this proporon is
relavely stable in certain countries, in others, consumers acvely switch between
oang and xed-rate products depending on which seems more benecial at the
me (Johansson et al. 2011).1 This is relevant because consequently, the spread
may dier between two banks with idencal funding structures because one of
them mainly extended loans that are re-priced every three months, while the other
extended loans with a rate xed for ten years. For the laer, there is a signicant
risk of interest rate levels rising substanally over the ten-year period, which is also
reected in future funding costs. This must be taken into account in the interest
spread of extended loans. Prepayment by customers is also a source of risk, which
compels banks to extend the prepaid amount and interest rate environment
dierently — and typically lower — than the one prevailing at the me of original
loan extension. This may be parcularly problemac in countries where the
administrave costs of switching banks and prepayment are low.2
Operang costs. The upward impact of operang costs on interest spreads has
been demonstrated by many studies using various target and control variables
(Gambacorta 2014, Valverde – Fernández 2007). The impact of operang costs
may be parcularly signicant on household loans, as households are sll primarily
served personally, which requires the maintenance of signicant infrastructure (such
as a branch oce network), and the cost of this is reected in spreads. For this
reason, the eciency at which banks use their infrastructure is relevant, because
a signicant relave price decrease (for instance through digitalisaon) may be
reected in credit spreads.
1 Badarinza et al. (2014) demonstrated that the choice between oang- and xed-rate loans is mainly shaped
by the interest spread prevailing between the two product types at a given point in me, and the spread
expected in the short run. A volale inaonary environment should also be menoned: more volale prices
are generally associated with a lower number of xed-rate loans.
2 According to Hungarian regulaons, the early repayment penalty is capped at 2 per cent of the prepaid
amount. However, the debtor may terminate and prepay his debt at the end of the interest period or the
interest spread period free of charge if the interest rate or the spread are set to increase.
8Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Credit losses. An inherent element of bank operaon is that some debtors will
not be able to service their debt. Banks must oset the losses incurred on these
loans through their interest rate spreads (and specically, the risk spread). So
the larger the expected loss on a porolio, the higher the interest spread that
may be necessary. Expected loss is shaped partly by economic fundamentals
(unemployment, changes in GDP, housing price developments) and partly by the
eciency of the legal instuonal system. It is important to note that expected
credit losses are calculated based on historical data, as a result of which a high
volume of non-performing loans may have a lasng impact on pricing. This means
that despite a far beer quality of currently extended loans, the bank may price
them as riskier based on its experiences derived from historical data. Although the
bank may incorporate forward-looking variables in its pricing model, the samples
oen available to banks contain observaons from the crisis period, and thus these
models may possibly capture a higher average risk level.3
Banks’ legal environment also has an impact on spreads. Mortgage loans are
collateralised products, which means that in the event of late payment by the
debtor, banks can hope to recover their loss by selling the property backing the
loan. The rate of recovery depends not only on changes in property prices, but
also on the strength and eciency of the tools available to nancial instuons
for enforcing their rights on the collateral. If legislaon impedes foreclosure (for
instance through long and costly foreclosure proceedings or other administrave
3 Carlehed and Petrov (2012) oer an in-depth discussion of the aspects of this topic that aect risk models.
Instuonal
environment Costs of banks
Operang
costs Expected return,
cost of capital
Interest rate
+
Other fees
Compeon
Funding costs,
Interest rate risk
Expected
losses
Efficiency,
concentraon
Interest rate
environment,
interest rate fixing
Real economy,
legal environment
Financial
culture
Source: own edit.
9
Idenfying the determinants of housing loan margins in the Hungarian banking system
constraints), banks’ expected losses and thus the spreads they apply will also be
higher. The internaonal literature demonstrated this eect both by examining net
interest income (Demirguc-Kunt – Huizinga 1999) and spreads on new loans (Laeven
– Majnoni 2005). Creditor banks’ opon for changing the interest rate through
the duraon of the contract also has signicance. If a bank is able to unilaterally
amend the interest rate at any point during maturity, it does not have to include
all expected future losses into the price at the me of contracng because it has
the opon of responding exibly. These types of loans were prevalent in Hungary
prior to the onset of the crisis, but signicant steps have been taken in recent years
to even out the balance of power between consumers and nancial instuons.4
In addion to the foregoing, the interest rate must also include a prot margin
allowing the instuon to generate the return expected by shareholders. The size
of the prot margin may depend on market structure, the level of compeon, the
instuon’s market power and the level of informaon held by potenal borrowers.
If compeon is weak and future debtors have poor nancial literacy and low price
elascity, then stronger market parcipants are able to enforce costs and high
prot goals in margins. Besides the impact of compeon, Ho and Saunders (1981)
also menon the risk aversion of management, average transacon size and the
variance of interest rates. However, there are contradicng views on compeon:
Maudos and Fernández de Guevara (2004) found that increasing market power is
associated with decreasing spreads.
In the following secon, we seek to idenfy the determinants of the relavely
higher average spreads on newly extended housing loans in Hungary. To idenfy
these determinants, we used econometric methods applied to several databases
alongside simpler stascal tools.5 Unfortunately, the available databases do not
include any that could provide a direct and certain answer to our queson (“Why
are spreads on new housing loans elevated by internaonal standards?”). We are
only able to use banking system aggregates in internaonal databases, and are thus
unable to control for either creditor or borrower composion. Data available only
at a low frequency and for relavely short periods make it even more dicult to
obtain reliable results.6
4 The legislave amendment on “transparent pricing” eecve from April 2012 is one such measure, which
substanally reduced banks’ leeway to unilaterally amend contracts. In keeping with this trend, the “ethical
banking system” regulaon introduced in 2015 only allows the amendment of lending condions based on
predened indicators approved by the MNB.
5 E.g. the examinaon of the composion eect.
6 Considering the available data, it is no surprise that the target variable of most papers published on the
subject is the net interest income role of prot and loss account, rather than the interest spread of newly
extended loans (see for instance Maudos – de Guavera 2004, Demirguc-Kunt et al. 2003, Saunders –
Schumacher 2000, Valverde – Fernández 2007). Using the prot and loss account as the point of departure
enables the use of bank-level internaonal data, but from the perspecve of this study, this is too broad of
a category that also contains non-relevant informaon.
10 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
As a result, we have opted for the following strategy: we aempt to explain
heterogeneity of Hungarian banks’ pricing behaviour using bank-level and
transaconal-level variables, and then examine the main variables idened within
the Hungarian sample in an internaonal comparison. We believe that a Hungarian
bank sets a higher spread compared to other banks based on a specic own
aribute, and then if the Hungarian banking system diers from the internaonal
average in terms of this variable, it may provide an explanaon for the higher spread
relave to other countries. However, it should be noted that this strategy is only
indicave and oers indirect evidence for the invesgaon of internaonally high
spreads, but does not provide a clear explanaon in methodological terms.
To answer our central queson, we performed esmates for three databases.
We examine the impact of bank credit supply and the contract-level aributes
of extended loans using a linear regression applied to microlevel data available
for 2014–2015 and using a panel model esmated for bank-level data between
2004–2014 for bank aributes. We then analyse the impact of demand aributes
using microlevel data available for 2015 using a mul-nominal regression. We
use various databases and methodologies in an eort to present and invesgate
the broadest range of aspects of the issue. This approach obviously comes at the
price of sacricing an in-depth examinaon of the dierent secons that would
be possible if we dedicated a separate paper to each part. We are aware of this
drawback, but nevertheless believe that this comprehensive approach will yield
the greatest benet in light of the relave underrepresentaon of the topic in the
literature.
The MNB’s public analyses (mainly the Trends in Lending and the Financial Stability
Report) generally present the dierence between the average APR of housing loans
extended during a given month and the 3-month BUBOR. However, the pricing of
housing loans may diverge substanally based on the term of interest rate xaon
by the bank for the reasons addressed in the previous chapter. Interest rates xed
for longer periods of up to 5 to 10 years currently materially exceed the inial
interest rate level of oang rate transacons that is ed to the reference rate
and thus changes relavely quickly. As menoned earlier, the main reason for this
is that economic agents generally expect interest rate hikes at the boom of the
interest rate cycle, so the cost of bank funds with rates xed for a longer period is
higher than the cost of shorter-term or oang rate funds (such as the 3-month
interbank interest rate). Banks may access funds with long-term xed rates either
directly or synthecally by interest rate swaps. In the laer case, the xed leg of
the interest rate swap represents the funding cost for the bank. If the bank nances
a xed-rate loan with oang-rate funds, the higher interest rate risk may warrant
11
Idenfying the determinants of housing loan margins in the Hungarian banking system
a higher spread. Based on the distribuon of new loans by the type of interest rate,
Hungary has a relavely high rao of loans with inial rate xaon of over one year,
especially by regional standards (ESRB 2015:28; EMF 2016).
Loans with inial rate xaon of over one year play a key role in explaining spreads
that are high even by internaonal standards. While the above-BUBOR spreads of
transacons with oang rates within one year already approached the levels of
other regional countries (Figure 3), the spreads of products with inial rate xaon
of over one year above the 3-month money market interest rate far outstripped
regional levels (Figure 4).7
7 In addion to the foregoing, there is methodological bias stemming from the fact that the spreads published
by the MNB are based on the APR, and are thus sensive to average loan contract maturity. The dierence
between the annual percentage rate and the interest rate is also shaped by other costs besides interest
(generally disbursement and loan assessment charges, handling charges), which increase APR expressed as
a percentage to greater extent if the maturity is shorter. The average maturity in Hungary in 2013 was 15
years, the shortest among EU countries. Within the region, Romania and Poland exhibit average maturies
of 25-26 years (ESRB 2015). A maturity of 10 years shorter results in an approximately 0.1 percentage point
increase in the APR characterisc of Hungary. A similar eect prevails when other costs are higher relave to
the loan amount taken out (such as nominally xed fees and lower average loan amounts), however there
is no available internaonal informaon on this.
Percentage point Percentage point
Hungary
Czech Republic
Romania
Slovakia
Poland
Slovenia
Eurozone
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
Jan 2010
Apr 2010
Jul 2010
Oct 2010
Jan 2011
Apr 2011
Jul 2011
Oct 2011
Jan 2012
Apr 2012
Jul 2012
Oct 2012
Jan 2013
Apr 2013
Jul 2013
Oct 2013
Jan 2014
Apr 2014
Jul 2014
Oct 2014
Jan 2015
Apr 2015
Jul 2015
Oct 2015
Jan 2016
Note: spread above the 3-month interbank interest rate, interest rate based. Newly extended loans.
Source: MNB, national central banks.
12 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Percentage point Percentage point
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
Hungary
Czech Republic
Romania
Slovakia
Poland
Slovenia
Eurozone
Jan 2010
Apr 2010
Jul 2010
Oct 2010
Jan 2011
Apr 2011
Jul 2011
Oct 2011
Jan 2012
Apr 2012
Jul 2012
Oct 2012
Jan 2013
Apr 2013
Jul 2013
Oct 2013
Jan 2014
Apr 2014
Jul 2014
Oct 2014
Jan 2015
Apr 2015
Jul 2015
Oct 2015
Jan 2016
Note: Spread above the 3-month interbank interest rate, interest rate based. This scheme does not exist
in Poland. For loans with initial rate fixation of over one year, the 3-month interbank interest rate may
diverge substantially from the actual cost of funding, so the spread presented by us may be partially
shaped by higher funding costs. Newly extended loans.
Source: MNB, national central banks.
0
5
10
15
20
25
30
0
5
10
15
20
25
30
–0.2
0.2
0.6
0.9
1.3
1.7
2.1
2.4
2.8
3.2
3.5
3.9
4.3
5.0
5.4
5.7
6.1
6.5
6.8
7.2
7.6
7.9
8.3
8.7
9.0
9.4
Per cent Per cent
Spread over BUBOR (percentage point)
2015, variable
2014, variable
2015, fixed
2014, fixed
Note: Interest rate based. Exclusive of building societies. Newly extended loans.
Source: MNB.
13
Idenfying the determinants of housing loan margins in the Hungarian banking system
Despite the relave widespread nature of products with inial rate xaon of
over one year, a signicant improvement has been observed in recent years in the
pricing of housing loans, as is also reected in the distribuon of spreads above
the BUBOR: In 2015, the distribuon of both oang rate products and products
with an inial rate xaon of over one year shied towards lower spreads relave
to 2014 (Figure 5).
Since early 2014, the MNB has compiled interest rate and other informaon on
new contracts on a transaconal basis. We therefore have a micro-level database
(with over 60,000 observaons aer cleaning the data8),9 which contains the date
of contract, the contracted amount, the maturity of the contract, the lending rate,
the type of the interest rate, the contracng bank, the loan’s subsidisaon status
and any associated collateral for all new housing loan contracts from 1 January
2014 onwards.
We were also able to associate bank aributes to individual contracts since we
have informaon on the creditor nancial instuon. In light of this, we can on
the one hand examine the impact of loan-level characteriscs on the spread while
controlling for the aributes of the creditor bank, and also analyse the paral eect
of bank aributes on spreads. It is important to stress that although we can control
for various loan contract aributes using variables of loan-level characteriscs, the
database does not include informaon on several important traits (such as income,10
collateral value, payment-to-income rao).
In order to idenfy paral eects, we use linear regression (OLS) where the
dependent variable is the spread above the 3-month BUBOR. During the esmaon
of the rst model, contract level characteriscs are given the main focus among
explanatory variables, and we control for the creditor bank using dummy variables.
In the second model, we use variables describing the bank’s operaon instead of
bank dummies in order to idenfy the paral eect of the laer. Santos (2013)
follows a similar methodology to examine the interest rates on loans to Portuguese
non-nancial corporaons. It is important to note that because the database only
8 When cleaning the data, loans extended by building sociees were also ltered out along with apparent data
errors on account of the special nature of these instuons and the schemes oered by them. In addion,
the model using bank variables does not include loans extended by cooperave credit instuons. This
is because the integraon process which cooperave banks underwent over the past two years makes it
uncertain whether individual instuonal aributes play a role in shaping spreads.
9 The main characteriscs of the database are presented in the Annex.
10 However, it is dicult to judge the customer’s income from this perspecve. The price seng of banks may
dier in terms of whether customer income only plays a role in accepng or rejecng loan applicaons, or
in the determinaon of the specic interest rate as well.
14 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
contains data for 2014–2015, the ndings can primarily be applied to these two
years. Because of the special nature of this period in various regards, we use longer
averages instead of the specic quarterly value for some of the bank variables:
• Net credit losses: banks set aside provisions according to their expected losses.
However, the Selement of household loans decreased the gross value of loans
in 2015 H1, which also lowered the amount of expected loss, as the collateral
backing the loans retained its earlier value. The Selement thus decreased the
net value below the collateral value for a poron of loans, and as a consequence
wring back provisions was economically jusable in some cases. Several
instuons took advantage of this opportunity, but this development temporarily
concealed actual credit risk costs and losses in their prot and loss accounts. We
therefore use the average between 2008 and 2014 in the model.
• Rao of net income from fees and commissions: the transacon fee introduced
in 2013 emerged as an “other expense” for banks, but due to the charge being
passed on to customers, its revenue side shows up among fee and commission
income. Consequently, the rao of net income from fees and commissions
increased arcially relave to interest income. In view of this, we use the average
value for the period between 2008 and 2012.
In light of the above, we esmate the following regression model that also includes
bank dummy variables:
SPREAD
i
=
β
0+
β
1CONTRACT
i
+
β
2BANKdummy
i
+
β
3TIMEdummy
i
+
ε
i (1)
where SPREADi is the spread above the 3-month BUBOR for contract i, i.e. the
dierence between the contractual lending rate and the average 3-month interbank
interest rate for the specic month. CONTRACT is the vector containing contract
aributes, and we include two dummy variables: one for the creditor bank and one
for controlling for me (quarter) of contracng. β0 is constant, β1, β2 and β3 refer
to the vectors of the coecients associated with dierent groups of variables, the
element number of which corresponds to the number of variables constung the
group of variables. The contractual variables used in the model are the following:
• Maturity: the original duraon of maturity as specied in the contract, expressed
in months. The model also includes the square of the variable in order to idenfy
non-linear eects.
• Contracted amount: the contractual loan amount expressed in HUF millions,
logarithmised. Similarly to maturity, we also included the square value.
• Collateral dummy: if there is any collateral (generally real estate) associated with
the contract.
15
Idenfying the determinants of housing loan margins in the Hungarian banking system
• Fixed rate dummy: if the interest period dened in the contract is longer than 12
months, the dummy is 1; otherwise it is 0.
• Amount of state subsidy: esmated value of the interest rate subsidy based on
the rules dened in the state interest subsidy decree eecve in 2014-2015.11
The esmated equaon of the model containing bank variables is:
SPREAD
i
=
β
0
+
β
1
CONTRACT
i
+
β
2
BANK_CHARACTHERISTICi+
β
3TIMEdummyi+
ε
i
SPREADi=
β
0+
β
1CONTRACTi+
β
2BANK_CHARACTHERISTIC
i
+
β
3
TIMEdummy
i
+
ε
i (2)
In the second model, besides the above variables, we also include the following
bank variables (instead of bank dummy variables) (BANK_CHARACTERISTIC vector):
• Proporon of liquid assets: the proporon of liquid assets (cash, selement
accounts, central bank bonds and deposits, government securies) relave to
the balance sheet total. We also include the square of the variable in the model.
• Size of the capital buer: the dierence between the consolidated capital
adequacy rao (also factoring in Pillar II requirements) and the minimal regulatory
requirement. We also include the square of the variable in the model.
• Operang cost to assets: the proporon of operang costs (personnel costs, other
administrave costs, depreciaon) relave to the balance sheet total.
• Loan loss provisioning to assets: the average annual amount of the lending losses
relave to assets between 2008 and 2014.
• Rao of branch oces: the rao of network units of a bank/banking group relave
to the aggregate banking system branch oce network.
• Rao of net income from fees and commissions: the rao of net income from
fees and commissions relave to total of net income from interests, fees and
commissions. The average of values measured between 2008 and 2012.
The rst model using bank dummies gives an indicaon of the impact of contract
aributes on spreads. Based on the results of the model (Table 1, model (1)), the
higher the contract amount and the longer the maturity, the smaller the spread
above the BUBOR. However, this eect only applies unl a certain level, as shown
by the posive sign of the squared variables. The signicance of the loan amount
11 The subsidy is dierenated depending whether the loan’s purpose is to purchase used or new property.
In the laer case, the number of children in the household also inuences the subsidy. However, these
two pieces of informaon are not available, so we assumed that every loan was contracted to purchase
a used home. Based on aggregate stascs, this assumpon will not lead to any signicant errors, in view
of the fact that only a small fracon of newly extended loans were used to purchase new homes in 2014-
2015 (MNB 2016).
16 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
is presumably explained partly by the impact of income as an unobserved variable:
wealthier borrowers, represenng a lower risk tend to purchase larger properes
which calls for higher loan amounts. Economies of scale consideraons may
also have an impact: every loan contract comes with certain xed costs (such as
communicang with the customer, handling payment dicules), which requires
a higher spread on smaller credit amounts. However, above a certain level, potenal
loss rises and this is reected in the spread. For maturity, the negave coecient
may capture the eect of shrinking credit risks through the decreasing payment-to-
income rao. This eect however is oset by growing liquidity risks for loans with
very long maturies, so as the maturity grows longer, a higher spread is warranted.
As suggested by intuion, the collateralized nature of a loan decreases the spread,
while interest rate xaon of over one year increases the spread above the
interbank rate. Based on the esmate, the state subsidy also has a relevant impact.
In the database, we were able to observe the total interest rate received by the
bank, which incorporates state subsidies received as well. We are able to esmate
the approximate size of the subsidy based on the rules of the Home Creaon
Scheme being in eect in 2014–2015, and thus are also able to observe whether
the bank prices subsidised loans dierently depending on the amount of subsidy.
Our ndings show that for 1 percentage point of state subsidy, banks apply interest
rates that are over 0.3 percentage points higher on average, ceteris paribus. The
customer sll fares well, geng the loan at a spread that is 0.6 to 0.7 percentage
point smaller than the market rate in case of a 1 percentage point subsidy, while the
bank “keeps” 30–40 per cent of the subsidy. This nding may also give an indicaon
of the level of compeon.12
Based on the coecients idened above, changes in the general contract
characterisc of newly extended loans over the past two years have pointed towards
a reducon in spreads above the BUBOR. Since 2014, both the average of the
contracted amount and the average maturity have increased, while the proporon
of subsidised loans and the amount of state subsidy have connuously decreased,
due to the characteriscs of the pertaining regulaon,13 falling to minimal levels
by 2015 (from February 2015, the average market interest rate was below the 6
per cent corresponding to the lower threshold of the state subsidy). These three
12 Besides a low level of compeon, it may of course reects the impact of unobserved variables characterising
various bank porolios, that has been le out from the model. For example, if a bank specically targets
risky, lower-income customers with its state subsidised schemes, the higher spreads are indeed warranted.
However, the fact that borrowers are aware of state subsidy opons and may specically seek them out
irrespecve of their income status decreases the probability of this distoron. However, the restricons
of subsidisaon pertaining to property value increase the risk of bias.
13 According to the rules of the Home Purchase interest subsidy, the interest rate payable by the customer
must be no less than 6 per cent, so the subsidy can only lower the interest rate to this threshold. Given
that market interest rates approached and even dipped below this level, the state subsidy lost much of its
relevance compared to earlier, reected in the shrinking rao of subsidised loans.
17
Idenfying the determinants of housing loan margins in the Hungarian banking system
characteriscs have all fostered a reducon in transaconal interest rates, and
thus spreads.
For the model supplemented with bank variables, the signs of the coecients
discussed so far do not change, and they retain similar orders of magnitude (Table
1, models (2)-(7)). For bank variables, the credit losses of recent years and higher
operang costs were generally associated with larger spreads, which is in line with
our preliminary expectaons and the ndings of the internaonal literature. The
rao of net income from fees and commissions within net income of interest,
fees and commissions has a negave coecient, which suggests that banks which
generate income through other channels — for instance by selling other services
alongside loans — may take this into account by decreasing spreads. The rao of
liquid assets relave to total assets had a negave impact on spreads in the two
years under review, which may capture the price-reducing eect of growing credit
supply, while the posive coecient of the capital buer coecient may reect the
impact of higher cost of capital. The laer variable, however, loses its signicance
in the broadest specicaon. For both variables, the square values mostly have an
opposite sign (with the excepon of the capital buer, where the sign is the same
in the broadest specicaon, albeit the value of the coecient is parcularly low),
so these eects also only apply up to a certain level. The rao of branch oces
within the banking system branch oce network has a posive coecient, which
may capture market power: banks with relavely more branch oces may have
nearly exclusive presence on a greater amount of local markets, which they may
then enforce in their pricing. We address this eect in depth in the secon on the
model examining demand paerns (Chapter 5).
18 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
(target variable: spread above the 3-month BUBOR)
(1) (2) (3) (4) (5) (6) (7)
contracted_
amount_ln
–1.655***
(0.0171)
–2.131***
(0.0675)
–2.101***
(0.0680)
–2.030***
(0.0713)
–2.008***
(0.0716)
–1.977***
(0.0724)
–1.993***
(0.0727)
contracted_
amount_ln_sq
0.335***
(0.00512)
0.421***
(0.0188)
0.422***
(0.0191)
0.410***
(0.0196)
0.406***
(0.0197)
0.411***
(0.0198)
0.416***
(0.0198)
maturity_month
–0.00265***
(0.000278)
–0.00608***
(0.000376)
–0.00588***
(0.000371)
–0.00282***
(0.000359)
–0.00271***
(0.000358)
–0.00344***
(0.000355)
–0.00339***
(0.000354)
maturity_month_sq
8.57e-06***
(6.79e-07)
1.71e-05***
(8.92e-07)
1.63e-05***
(8.82e-07)
1.00e-05***
(8.50e-07)
9.81e-06***
(8.49e-07)
1.11e-05***
(8.42e-07)
1.09e-05***
(8.39e-07)
d_collateral
–0.777***
(0.0200)
–1.409***
(0.0358)
–1.498***
(0.0356)
–0.899***
(0.0287)
–0.928***
(0.0286)
–1.007***
(0.0285)
–0.994***
(0.0287)
subsidy
0.436***
(0.00930)
0.423***
(0.0101)
0.372***
(0.0100)
0.414***
(0.00932)
0.381***
(0.00944)
0.327***
(0.0100)
0.345***
(0.00999)
d_fixation
1.085***
(0.0127)
1.343***
(0.0132)
1.341***
(0.0132)
1.483***
(0.0139)
1.504***
(0.0139)
1.515***
(0.0137)
1.485***
(0.0136)
liquid
–0.304***
(0.0121)
–0.286***
(0.0126)
–0.162***
(0.00849)
–0.137***
(0.0101)
–0.124***
(0.0101)
–0.0697***
(0.0109)
liquid_sq
0.00405***
(0.000209)
0.00396***
(0.000221)
0.00157***
(0.000141)
0.00145***
(0.000180)
0.00112***
(0.000180)
0.000379**
(0.000193)
capital buffer
0.137***
(0.00480)
0.142***
(0.00442)
0.147***
(0.00449)
0.0238***
(0.00545)
-0.000334
(0.00561)
capital buffer__sq
–0.00505***
(0.000379)
–0.00597***
(0.000339)
–0.00713***
(0.000357)
0.000494
(0.000374)
0.00205***
(0.000396)
cost to asset
0.678***
(0.0171)
0.550***
(0.0173)
0.675***
(0.0182)
0.636***
(0.0176)
prov_avg
0.327***
(0.0100)
0.281***
(0.00976)
0.370***
(0.0108)
branch
0.0303***
(0.000762)
0.0375***
(0.000862)
comm_fee
–0.0350***
(0.00178)
TIME dummy
YES YES YES YES YES YES YES
BANK dummy
YES
Constant
6.426***
(0.0421)
12.62***
(0.173)
11.68***
(0.177)
7.748***
(0.129)
7.091***
(0.141)
6.877***
(0.141)
6.822***
(0.147)
N
64,904 62,848 62,848 62,814 62,280 62,280 62,280
R2
0.671 0.562 0.572 0.621 0.630 0.638 0.641
Note: Robust standard errors in parentheses.* Refers to a 10 per cent, ** to a 5 per cent, and *** to a 1
per cent significance level. The variables: spread above the 3-month BUBOR expressed in percentage
points (BUBOR_SPREAD), contract amount in HUF million, logarithmised (contracted amount_ln), matu-
rity in months (maturity _month), loan collateral dummy (d_collateral), estimated amount of state sub-
sidy (subsidy), interest rate fixation over one year dummy (d_fixation), liquid assets/balance sheet total
(liquid), consolidated capital buffer based on SREP (capital buffer), operating costs to assets (cost to
asset), average loan loss provisioning between 2008 and 2014 (prov_avg), ratio of branch offices within
the branch office network (branch), net income from fees and commissions within net income of interest,
fees and commissions, 2008–2012 average (comm_fee), TIME dummies and institution dummies (BANK
dummy). Variables ending in _sq refer to squared variables.
Source: own calculations.
19
Idenfying the determinants of housing loan margins in the Hungarian banking system
We also used a panel database for our analysis, compiled from Hungarian banking
system data that includes data on the major banks involved in housing lending in
Hungary between 2004 Q1 and 2014 Q4 (OTP Bank, MKB Bank, Budapest Bank,
FHB Bank, Cetelem Bank, Erste Bank, Raieisen Bank, CIB Bank, Unicredit Bank
and K&H Bank).14 Our approach was to use a regression model expressed for the
dierences of the dependent variable and that of the explanatory variables (3).
We esmated a model using xed eects broadly employed in the literature15, and
a dynamic model also containing the dependent variable’s lag, used more rarely
(e.g. Valverde – Fernández 2007).16
Because the presence of unit root processes could not be ruled out for level me
series and because error terms exhibited autocorrelaon when applying the xed
eect model, we instead chose to use a stac model containing the rst dierenal
of the variables:
Δy
it
=ΔX
it
'
β
+e
it (3)
e
it
=
δ
t
+
ω
it (4)
ω
it
∼I.I.D.
, (5)
Where Δy
it
is the annual change in housing loan margins, ΔX
it
is the annual change in
explanatory variables and δ
t
is the me xed eect . Because our panel is balanced,
the calculaon of dierences did not cause any signicant data loss. In the following
secon, we present the ndings of the model esmates, which proved relavely
robust for several specicaons.
The database allows the examinaon of bank-specic factors shaping banks’ pricing
decisions such as operang and funding costs, economies of scale and bank strategy.
Because similarly to the previous database, the sample only includes Hungarian
14 The housing loans oered by one bank are special in that they are pically unsecured transacons concluded
for “other” loan purposes, and as such, can be regarded as consumpon rather than housing loans. For
this reason, we also made the esmate with the omission of this bank and our results proved to be stable.
15 Based on the tests performed using the xed eect model, we cannot exclude the presence of unit root
processes for certain variables, so we rejected this approach due to potenal spurious regression bias.
16 Including the dependent variable lag may be movated by the raonale that when determining bank
lending spreads, earlier periods may serve as an anchor, and addionally, banks are not really capable of
reacng exibly when pricing loans due to market circumstances. Another advantage of this approach is
its capacity to address the endogeneity stemming from reverse causality, which in our case may emerge
in the variables related to bank porolio structure or in the NPL rao. But because the Blundell-Bond and
Arellano-Bond type methods available for short me series can be applied eecvely mainly with large
cross secon element numbers, in our case, esmang too many instruments created issues. Although we
tried out dierent dynamic models, we encountered troubles with model diagnoscs every me.
20 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
data, there is sll no way to conduct a direct internaonal comparison. However,
the 10-year horizon allows us to control for country-level cyclical macroeconomic
developments. In order to capture macroeconomic developments, we included
annual GDP in the model and also included me xed eects in an alternave
specicaon (Model 2). Among the variables used, the ones capturing credit losses,
such as the rao of non-performing loans, the loan-to-value rao and loan loss
provisioning, can be considered as cyclical as well. We also included indicators
represenng market power for the sake of capturing structural eects: the size of
bank branch networks and bank market share within household lending.
The ndings of the esmated model have limited reliability. The sign of key variables
is generally idencal to the ones dictated by economic theory, but signicance levels
are not stable across the dierent specicaons. Because banks are oen unable to
adapt on a quarterly horizon, we consider the ndings of the model expressed for
annual variables as the most convincing, so the following secon addresses these
in detail (Table 2). Overall, from our ndings indicave conclusions can be drawn
on the factors that shape housing loan spreads in the Hungarian banking system.
21
Idenfying the determinants of housing loan margins in the Hungarian banking system
(target variable: spread above the 3-month BUBOR)
Operating cost 0.994
(0.779)
0.489
(1.086)
Other income/interest revenue –0.00831
(0.00604)
–0.00346
(0.00632)
Liquidity 0.0470***
(0.0174)
0.0508***
(0.0183)
CAR 0.0376**
(0.0165)
0.0995***
(0.0350)
Ratio of fixed-rate loans slope of the yield curve 2.774***
(0.699)
3.469***
(1.105)
External liabilities 0.0206
(0.0267)
0.00873
(0.0283)
GDP (YoY) –0.188***
(0.0633) –
LTV 0.0128*
(0.00706)
0.00851
(0.00775)
NPL 0.120***
(0.0364)
0.0882**
(0.0393)
Provisions 0.389***
(0.124)
0.209**
(0.101)
Proportion of branches 0.141*
(0.0786)
0.124*
(0.0683)
Market share 32.44
(19.84)
35.80
(21.84)
Constant 0.0636
(0.211)
1.427*
(0.849)
Time fixed effect – YES
Number of observations 317 317
R-squared 0.22 0.34
Number of banks 10 10
Note: robust standard errors in parantheses.* Refers to a 10 per cent, ** to a 5 per cent, and *** to a
1 per cent significance level.
Variables: operating cost to balance sheet total, non-interest income/interest income, liquid assets/
balance sheet total, capital adequacy ratio expressed as a percentage, the share of fixed loans multipli-
ed by the slope of the yield curve (5-year government security yield – 3-month BUBOR) taken into
account after 2010, the share of external liabilities within the sum of deposits (households and corpora-
te) and external liabilities, GDP growth expressed in percentage points, loan value to the property
pledged as collateral expressed as a percentage, share of non-performing loans in proportion to house-
hold and corporate loans, loan loss provisioning in the given period in proportion to the balance sheet
total expressed as a percentage, market share within the stock of outstanding household loans. We
included the annual change of each factor into the model. Because it takes different amount of time for
the changes of various factors to become incorporated into spreads, we applied an annual lag for ope-
rating costs and a quarterly lag for the capital adequacy ratio, the non-performing loan ratio and the
provision.
Source: own calculations.
22 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
The individual bank factors capture, among others, the dierence between
banks’ business models. The coecient of the share of operang costs to balance
sheet total is not signicant, so in this model, we are unable to reliably conrm
the intuion that banks compensate higher operang costs with seng higher
prices.17 However, this result can also be distorted by the change in the rao’s
denominator (e.g. as a result of deleveraging aer the crisis). The rao of other
income to interest income is also not signicant at the usual signicance levels.
We featured this variable in the model to be able to control for bank strategies
that place greater emphasis on net income from fees and commissions, allowing
the bank to oer more aracve lending rates. The posive sign of the share of
liquid assets to the balance sheet total and the capital adequacy rao suggest
that banks incur addional costs to maintain excess liquidity and excess capital
which they compensate with higher prices.18 We examined the rao of xed rate
loans on variable interest rate loans for the post-crisis period in interacon with
the slope of the yield curve. Based on our expectaons, at those banks where the
rao of xed interest rate loans is higher, the aggregated spread is sensive to
the slope of the yield curve which captures the higher cost of funding and/or the
interest rate risk. This impact was signicantly idened during the panel esmate
on Hungarian banks. We included the variable of the share of external liabilies
to corporate and household deposits in the regression in order to control for the
dierence in business models among banks which are relying on and those which
are not relying on external funds. This variable is not signicant, that is, the results
of the regression do not suggest that banks would price dierently as a result of
their reliance on external funds.
Among cyclical variables, we go into detail about the impact of both the macro
variables and that of the individual bank variables related to the cyclical posion.
The negave coecient esmated for GDP capturing the economic performance
suggests the pro-cyclical nature of spreads. In the case of an economic contracon,
spreads increase in line with high risks as a sign of decreasing credit supply, which
further aggravates the contracon of the economy, while during a boom period,
banks lend with more moderate spreads, thereby further strengthening growth. The
LTV rao entered in the regression with a posive sign. The higher LTV rao reects
higher risk, since in the case of default the bank may migate or avoid credit loss by
selling the collateral. It should be noted that banks can compete not only in price,
17 What makes the idencaon of the impact of operang costs more dicult is that prior to the crisis
several banks gained market share through agent sales, the cost of which — as opposed to operang
their own branch network — did not appear among their operang costs. Considering that following the
onset of the crisis, agent sales decreased signicantly, this may also be the reason why operang costs sll
appear as a signicant factor in the case of a micro-level database only building on data from 2014-2015.
18 In the case of liquidity, this result contradicts the result of the esmaon conducted on the micro-database.
However, t the laer database covers only a two-year period while the panel database processes the data
of one decade, which means a dierence. On the other hand, the impact of capital adequacy is in line with
the results of the micro-database.
23
Idenfying the determinants of housing loan margins in the Hungarian banking system
but also in lending condions, which may cause endogeneity for the LTV variable,
that is, in this case, the underesmaon of the coecient, especially if we examine
newly issued loans. The non-performing loan rao within the loan porolio of the
private sector (NPL) also correlates with the economic cycle: during the period of
an economic boom the share of the NPL porolio is generally low, while during
recession, this rao increases. The high NPL captures both already wrien-o and
potenally expected lending loss; accordingly, the sign of the variable is posive
in the esmated model. Similarly to the non-performance rao, provisioning also
reects the risks, but this indicator only includes the loss already wrien o by the
bank. The sign of the impairment is also posive in the model.
Because the development of economic growth in other countries was similar to the
Hungarian trend, the cyclical variables probably only explain some of the dierence
between spreads in the region. In our view, some structural reasons are also causing
the high spreads. We aempted to capture these factors by the share of the number
of bank branches and the banks’ market share on the household credit market. The
share of the number of bank branches in comparison to the number of branches
of the banks included in the model not only takes into account the bank’s own
branch network, but also the size of that branch network compared to that of the
competors. This variable is signicant and it is featured in the model with a posive
sign which suggests that banks operang a large branch network are able to use
their dominant posion on the market when dening the spreads on mortgage
loans. In our view, the role of the branch network is indeed relevant because the
majority of the populaon can select only from a limited number of banks located
near their place of residence, which decreases compeon between banks. The
market share variable is not signicant, so this simple control variable does not
conrm our impression that banks strive to use their dominant posion on the
market in their pricing.19
It may be a queson whether the levy on banks increased margins aer it was
introduced. We are unable to analyse this impact on the micro data due to the short
me period available, but we have included it in the panel model as an explanatory
variable. Based on our results, the impact of the bank levy is not apparent among
new loans, which also conrms the ndings of the literature according to which
banks have averted this extra cost by modifying the interest rate of their exisng
loan porolios (Capelle-Blancard – Havrylchyk 2013).
19 We can see in the correlaon matrix included in the annex that there is high correlaon between the
share of branches and the market share variable. For this reason, we decided to apply the model without
this laer variable, and the signicance level and the coecient of the branch-proporon variable did not
change signicantly either.
24 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Along with supply factors, it is important to examine whether the demand side
supports the existence of a compeve market or whether there are any fricons
that could result in less compeon. As part of this invesgaon, we developed
a model which belongs to the family of discrete choice models. This allows us
to examine the factors that inuence consumers in bank selecon. During the
modelling, we relied on the Central Credit Register database which contains detailed
loan analycs for new disbursements, including customer characteriscs, from 2015
onwards. The nal model contains the data of seven major banks, covering more
than two thirds of the mortgage loan market. In the following, we present the
intuion behind the model and the main steps of the esmaon (esmang interest
rates and restricng the choice set) and we summarise the results of the esmaon.
We applied a mulnomial regression model for the analysis, placing consumers’
individual choices into the focus of the invesgaon. Factors inuencing the
decisions of consumers can be classied into three groups. First, the condions of
the selected loan product and the characteriscs of the selected bank play a key
role. Beside the interest rate, we can menon the factors that capture the quality
of bank services and that a past relaonship with a given bank may also be an
important aspect. Second, the customer’s taste also maer as the popularity of
the banks may dier in the various segments of the society. Third, the customer-
specic factors which are not observable by the researcher show up in the error
term of the esmate. Based on the above, following Train’s demonstraon (2002),
the ulity of the customer by choosing a given bank can be wrien as:
Uij =Vij xij ,si
( )
+
ε
ij , (6)
Where U
ij
is the ulity of consumer i if he chooses bank j, x
ij
is the vector containing
variables which are customer- and also bank-specic (e.g. transaconal interest
rate). s
i
is the vector that contains solely customer characteriscs (e.g. age, income)
while ε
ij
is the model’s error term, which follows an i.i.d. extreme value distribuon
by assumpon. The model’s starng point is that customers strive to maximise their
ulity, that is, they opt for the oer promising the highest level of ulity compared
to other oers.
U
ij
>U
ik
,∀j≠k
(7)
20 This chapter provides a brief summary of the study, which was presented on the conference entled
„5th EBA Policy Research Workshop: Compeon in Banking: implicaons for nancial regulaon and
supervision” (Aczél 2016).
21 The above study presents in detail the steps of database cleaning and the descripve stascs of the data
used.
25
Idenfying the determinants of housing loan margins in the Hungarian banking system
Approaching the observable part of the ulity funcon with a linear relaonship,
we have:
V
ij
=x
ij
'
β
+D
j
'
γ
s
i
,
(8)
where β is the parameter vector belonging to the characteriscs of the various
alternaves, Dj is a vector containing binary variables denong individual banks,
γ is the matrix containing the parameters belonging to the customer characteriscs
diering by bank. Using all of the above and assuming that the error term follows
an i.i.d. extreme value distribuon, the likelihood that customer i selects bank j
can be wrien as:
P
ij =exij
'
β
+Dj
'
γ
si
exik
'
β
+Dj
'
γ
si
k
∑
(9)
To esmate the model, we also need theorecal interest rate data that show the
interest rate at which the customer would have received a loan had he chosen
another bank instead of the observed choice. We esmated these theorecal
interest rates using linear regression, so that we created a unique model for every
bank where the dependent variable is the interest rate and the explanatory variables
can be classied into two groups. First, we included in the models the characteriscs
of customers who actually borrowed from the specic bank (age, locaon, income)
and second, we also controlled for the transacon aributes (value of the mortgage,
maturity, loan type). The explanatory powers of the models are high (R
2
around 0.9)
and their standard error is low (around 0.3 percentage points).
22
Despite the good
model stascs, the fact that this esmaon may be biased is an issue. The potenal
bias stems from the fact that the esmaon sample is not random, because banks
may be chosen by customers with strongly diverging characteriscs (self-selecon
bias). However, it is important to stress that this esmaon procedure is similar
to the procedure applied by banks, because banks themselves dene their pricing
models based on relaonships esmated with regard to their own clientele. In our
view, our esmated models feature acceptable accuracy and esmate for a sample
similar to banks’ samples, so these esmates provide a good approximaon of the
theorecal interest rate that banks have oered to prospecve borrowers.
Aer esmang theorecal interest rates, we also examined whether the
assumpon that households can choose from the oerings of all banks is well-
founded. We found that households faced both geographic and nancial constraints,
so it is likely that they can only choose from a narrow range of banks when making
mortgage loan decisions. The geographic constraints are reected in the fact that
22 R 2 (0.32) is low for a single bank interest rate model, however this model also yielded an esmate with
a low error (RMSE 0.33).
26 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
no more than two of the eleven major banks are present in half, and no more than
four are present in three-fourths of Hungary’s districts (Figure 6).23
Figure 7 captures the dierences in banks’ business strategies through the
distribuon of customer income associated with the loans extended in 2015. The
gure clearly shows that the banks marked by black mainly serve low-income
customers, while those marked by red mainly target higher-income customers and
barely lend to lower-income segments, or not do not lend to these segments at all.
The distribuon of loan size or the value of the property to be purchased shows
a similar picture. We used these ndings within the models to restrict the group of
banks that customers may choose from.
23 The distribuon of bank presence as a funcon of the populaon would be an interesng addion. More
than a quarter of the Hungarian populaon lives in a district where there are no more than two banks,
and nearly 40 per cent lives in a district where there are no more than four banks present from among
the eleven major banks. Only half of the populaon has access to at least six major banks in their region.
0
25
50
75
100
0
20
40
60
80
1 2 3 4 5 6 7 8 9 10
Per cent
Number of districts
Number of banks in districts
Number of districts
Cumulave share of districts (right-hand scale)
Source: MNB.
27
Idenfying the determinants of housing loan margins in the Hungarian banking system
We run the nal model in eight specicaons; the results are listed in Table 3. In the
rst specicaon, we neither controlled for choice sets nor included demographic
variables (A1). The ndings of this esmate are not in line with expectaons,
because for example the interest rate coecient is posive, which is dicult to
interpret, as it suggests that consumers like high interest rates. This nding also
suggests that endogeneity distorts esmates, which may be because the impact of
demand and supply is not adequately disnguished in this specicaon.
For the sake of ruling out endogeneity, we implemented three changes in the
model. First, we incorporated demographic variables and bank dummies (A2, A4,
B2, B4), second, we restricted the choice sets (B1-B4), and third, we incorporated
a variable that captures previous relaonship with banks (A3, A4, B3, B4). We
obtained intuive results in each case, and the sign of the interest rate is negave,
which is in line with a negavely sloping demand curve.
In the models that included demographic variables and bank dummies (A2, A4,
B2, B4), the issue of endogeneity was signicantly reduced. The procedure applied
addresses the typical problem of a bank taking advantage of its strong brand and
0
5
10
15
20
25
30
35
0
5
10
15
20
25
30
35 Per cent Per cent
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
450,000
500,000
550,000
600,000
650,000
700,000
750,000
800,000
Income (HUF)
Note: The various lines indicate the banks included in the analysis.
Source: MNB.
28 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
lending at high interest rates in response to strong demand. Another key nding
is that by including demographic variables, clearly outlined taste paerns can
be idened. A good example of such paern is that based on the esmated
coecients, older age groups tend to prefer banks that have an established
presence on the Hungarian market, while younger age groups prefer newer market
entrants.
The models esmated by narrowing the choice set (B1-B4) may yield a more realisc
picture because banks that are not potenal choices for customers are le out of
the calculaons. Thus for instance, in the case of a low-income customer, obtaining
a loan from a bank that exclusively targets an auent clientele and oers low
interest rates is not a realisc opon. If we leave out this bank from the customer’s
potenal opons, it would lead to the false conclusion that although the customer
could borrow at a low interest rate, he instead chose to borrow at a higher rate. This
eect may be present in specicaon A1, where we did not control for the choice
sets. A key nding is that narrowing the opons alone results in the esmaon of
a demand curve with a negave slope (B1).
We also included a variable in the models that shows whether the customer has
borrowed from a specic bank (in the past eight years). This variable is signicant
and posive in every specicaon (A3, A4, B3, B4), which suggests that customers
prefer banks that they are familiar with in their borrowing decisions.
In every model, we included a variable among explanatory variables that shows
the number of branch oces that the bank has in the region where the customer
resides. This variable is also signicant and posive in almost every specicaon,
meaning that an expansive branch network is valued by customers.
Overall, the esmaon results suggest that the Hungarian populaon tends
to choose from a specic and narrow range of banks when making borrowing
decisions. This is partly due to the geographic distribuon of banks’ branch networks
and partly to the taste paerns prevailing within society; banks’ business models
are also relevant. These limitaons and paerns allow banks to price their products
according to oligopolisc compeon. These ndings conrm the outcomes of the
bank panel model invesgang supply eects, i.e. that the distribuon of branches
plays a key role in determining spreads. Finally, these esmates demonstrate that
structural factors play an important role on the Hungarian mortgage market.
29
Idenfying the determinants of housing loan margins in the Hungarian banking system
No taste
(A1)
Taste
(A2)
No taste
(A3)
Taste
(A4)
No taste
(B1)
Taste
(B2)
No taste
(B3)
Taste
(B4)
Interest
0.171*** –1.262*** –0.0176 –1.182*** –0.862*** –1.640*** –1.042*** –1.539***
Number of
branches
0.0221*** 0.000881 0.0136*** 0.00213** 0.0181*** 0.00762*** 0.00843*** 0.00971***
History
3.037*** 2.750*** 2.502*** 2.750***
Bank A
Age
0.00311 0.00563 0.00956 0.0167**
Income
1.142*** 1.213*** 0.401*** 0.430***
Constant
–6.447*** –6.015*** –2.614*** –1.945***
Bank B
Age
–0.0266*** –0.0115** –0.0153 –0.00750
Income
1.326*** 1.332*** 0.428*** 0.425***
Constant
–10.22*** –9.233*** –3.936*** –2.947***
Bank C
Age
–0.0155*** –0.00626** –0.0149*** 0.00206
Income
0.979*** 1.032*** 0.439*** 0.441***
Constant
–3.337*** –2.895*** –1.705*** –1.195***
Bank D
Age
–0.0623*** –0.0511*** –0.0747*** –0.0632***
Income
1.270*** 1.289*** 0.523*** 0.515***
Constant
–5.198*** –4.584*** –0.590* 0.283
Bank E
Age
–0.0165*** –0.00558* –0.0180*** –0.00227
Income
0.480*** 0.540*** 0.0768** 0.146***
Constant
–2.626*** –2.676*** –1.362*** –1.509***
Bank G
Age
0.00544 0.0136*** 0.00448 0.0129*
Income
1.245*** 1.275*** 0.504*** 0.502***
Constant
–7.052*** –6.388*** –2.897*** –1.948***
Note: * Refers to a 10 per cent, ** to a 5 per cent, and *** to a 1 per cent significance level.
Source: own edit.
In previous secons we listed a number of characteriscs that may potenally
explain the high Hungarian spreads. In line with our research strategy, in the
rst step we aempted to explain the heterogeneity of Hungarian banks’ price
seng behaviour by using bank-level and customer-level variables. The next step
is to examine the performance of the Hungarian banking sector compared to
internaonal examples with respect to the signicant variables idened in the
models esmated on the Hungarian sample.
30 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Our main ndings, as presented in previous secons, were the following:
• Through a composion eect, the higher share of contracts with an inial interest
rate xaon of over 1 year may account for the higher level of Hungarian spreads.
The slope of the yield curve may also contribute to the relavely high cost of
xed-interest loans.
• An increase in GDP typically reduces spreads, while recession raises them.
• Credit losses and the higher share of nonperforming loans may inuence the
spreads through higher risk costs, partly as a result of banks’ propensity to build
on historical credit experiences involving past – poor quality – loans.
• Higher operang costs have been coupled with higher spreads in recent years.
• The lower share of prots from fees and commissions may induce relavely higher
spreads.
• Similarly, banks’ capital adequacy (capital requirement) may also exert upward
pressure on spreads.
• There is a posive correlaon between the average loan-to-value rao of the
loans disbursed and the spread imposed.
• Banks represenng a higher share in the branch network of the banking sector
applied, ceteris paribus, higher spreads.
• The lack of a sucient number of market parcipants in certain regions and
debtors’ taste paerns may lead to the emergence of an oligopolisc market.
Unfortunately, owing to the limited availability of data, only some of these items
can be analysed in internaonal comparison. In the following, we focus our research
on items that – in light of the internaonal literature and/or our esmated models
– appear to be especially important, and for which relevant internaonal data are
also available. The laer may pose a problem mainly in relaon to the results of
the demand model; indeed, there is praccally no informaon available at the
internaonal level on debtors’ income status, their taste and on the distribuon of
branches. We will not go into detail about the topic of liquid assets and the loan-
to-value rao because – although we found some evidence that these indicators
and the size of the spreads are posively correlated – internaonal literature does
not provide clear guidance on the impact of such aributes on spreads.
31
Idenfying the determinants of housing loan margins in the Hungarian banking system
As pointed out above, the outstandingly high Hungarian spreads observed at the
end of 2015 and in early 2016 can be primarily aributed to the higher spread on
loans with an inial interest rate xaon of over 1 year. The spread between these
lending rates and the interbank rate is partly determined by the yield curve; indeed,
in the case of a steeper (and upward sloping) yield curve, the creditor bank will
also face increased costs of funds when borrowing funds with a long-term inial
rate xaon and consequently, this premium will be priced into the bank’s lending
rate. If the bank relies on short-term and/or oang rate funds to nance loans
extended with a long-term rate xaon, the interest rate risk thus incurred by the
bank juses an increase in the spread. Based on Eurostat data, the yield curve is
relavely steep in Hungary compared to other EU countries. At the end of 2015,
the spread between the ten-year government bond yield and the three-month
interbank interest rate took the h highest value in Hungary.
RO
HU
SI
UK
IE
SE
ES
NL
IT
DK
DE
CZ
BE
SK
0
10
20
30
40
50
60
70
80
90
100
–1.0
–0.5
0.0
0.5
1.0
1.5
2.0
2.5
Share of contracts with an inial fixaon period of
more than 1 year (2015 Q4, per cent)
Difference of average interest rates of contracts with an inial fixaon
period of less then 1 year and more than 1 year (2015 Q4, percentage points)
Note: In the case of loans with an initial interest rate fixation of over 1 year, the most widespread scheme
– the 1Y–5Y initial rate fixation – was considered.
Source: European Mortgage Federation, national central banks.
32 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
As at end-2015, data reveal that in Hungary, the share of products with a rate
xaon of over 1 year was high even though Hungary recorded one of the highest
interest spreads between xed and variable rate products (Figure 8). It should
be noted that, if the interest rate spread between two product types reects
the expected interest rate path, in theory, choosing between the two products
would not make any dierence for a raonal consumer, provided that his interest
expectaons coincide with market expectaons. Experience, however, shows
that instead of looking at the interest rate path as a whole, consumers are far
more concerned about the interest rate spread prevailing at the me of the loan
disbursement and during the short period that follows (Johansson et al. 2011;
Badarinza et al. 2014; Holmberg et al. 2015). It should also be remembered that,
as noted in the introducon, it is oen the given country’s lending “tradions”
or instuonal background that determine consumers’ decisions as they select
from the product types available. Having said that, since the surge in household
lending at the beginning of the 2000s, it has only been observed in recent years
(2015 Q4)
RO
HU
SI
UK
IE
SE
ES
NL
IT
DK
DE
CZ
BE
SK
0
10
20
Difference of average interest rates of contracts with an inial fixaon
period of less then 1 year and more than 1 year, adjusted by the difference
caused by expected change in interest rates (2015 Q4, percentage points)
30
40
50
60
70
80
90
100
–1.5
–1.0
–0.5
0.0
0.5
1.0
1.5
Share of contracts with an inial fixaon period of
more than 1 year (2015 Q4, per cent)
Note: In the case of loans with an interest period of over 1 year, the most widespread scheme – the
5Y–10Y initial rate fixation – was considered. We deducted the difference between the 5-year IRS and
the short-term interbank interes t rate from the difference between the average interest rate on fixed- ra-
te and variable-rate transactions.
Source: European Mortgage Federation, Datastream, national central banks.
33
Idenfying the determinants of housing loan margins in the Hungarian banking system
that households are more likely to become indebted with xed interest rates, on
a market basis (without any state subsidy).
We also analysed the gure above aer adjusng the interest spread by the
dierenal between the 5-year interest swap relevant to the given currency and
the short-term interbank interest rate. Our goal was to exclude, as far as possible,
the eect of interest rate path expectaons from the premium shown in the gure,
in order to obtain a beer approximaon of the “pure” dierenal concerning the
raonal consumer.
24
Based on the values thus received, in Hungary the premium on
xed-interest loans is higher than would be warranted by the dierence between
funding costs; consequently, we sll cannot consider the increase in the share of
xed-rate loans as being trivial (Figure 9).
We believe that the substanal share of xed-rate products suggests that the
Hungarian populaon is more risk averse than borrowers in other countries; indeed,
Hungarian customers are willing to pay a much higher premium for a xed interest
rate. This, in itself, does not imply that this premium (or at least a part of it) is
not jused; it is an interesng development, however, that Hungarian household
debtors are apparently more likely to pay a considerably larger sum in exchange
for a smaller deviaon in monthly payments. In our opinion, this may also reect
households’ negave experiences with foreign currency loans and the extremely
volale instalment amounts associated with them. Banai and Vágó (2016) also
conrm that foreign currency lending gave rise to precauonary moves among
households: based on data derived from the Austrian central bank’s Euro Money
Survey, the authors provided evidence that the negave experiences associated
with foreign currency lending clearly set back credit demand. It is also conceivable
that the “demand” problems presented in Secon 5 can be perceived more strongly
– possibly because of the limited number of acve market parcipants – in the
market of xed-interest loans. The picture appears somewhat more complex once
we consider that the high rao of xed-interest products has partly resulted from
the acvity of building sociees issuing xed-interest loans. Nevertheless, it is also
true for these instuons that the interest rate they impose exceeds the interest
level of variable-rate products; in other words, the customers of building sociees
will also pay the premium between the xed rate and the variable rate in exchange
for a predictable interest rate.
24 However, this method should be viewed as an approximaon only; indeed, the dierenal between the
average interest rate on actually disbursed xed-rate loans and the interest rate on variable-rate loans can
also be inuenced by composion eects, especially when a parcular product has gained dominance in
the given country. Various sub-markets may be dominated by dierent creditors and borrowers and the
dierent characteriscs of these market parcipants may also be reected in aggregate interest rates.
Consequently, the “pure premium” could not be presented even on the second gure.
34 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Model esmates have demonstrated that the rao of credit losses and non-
performing loans play a prominent role in credit spread developments. In calculang
their rate of return, banks should consider the probability of a borrower’s falling into
delinquency during the term of the loan, and calculate the expected recovery rate
on the collateral in case of the borrower’s delinquency. The calculaon of expected
losses is based on historical data; consequently, a substanal non-performing
porolio may have a long-term impact on price-seng. Based on the distribuon
of NPL raos, Hungarian banks are among the more aected instuons of the
region, which may have contributed to the emergence of higher spreads (Figure 10).
Collateral recovery and the eciency of enforcement proceedings play a key role in
credit loss developments. Hungarian legislaon has signicantly hampered banks in
the acquision and sale of real estate property in recent years. Moreover, the legal
environment protecng debtors movated even performing debtors to suspend
their monthly payments, generang even more credit losses for banks (Dancsik
et al. 2015).
(2014)
0
10
20
30
40
50
60
0
10
20
30
40
50
60
GR
BG
SI
HU
RO
HR
IE
PT
LT
LV
PL
ES
AT
DK
BE
IT
SK
GB
CZ
LU
FR
NL
DE
FI
EE
SE
Per cent Per cent
Note: Columns indicate the 25–75th percentiles of the ratios of individual banking systems, while lines
show the 10–90th percentiles. Countries are ranked in descending order based on the 75th percentile.
Source: SNL Financial.
35
Idenfying the determinants of housing loan margins in the Hungarian banking system
Operang costs have an intuive role in the evoluon of spreads, as banks need to
set a price that allows them to achieve prots. Banks’ lower eciency and higher
costs may also call for higher spreads. We demonstrated this eect successfully in
the model featuring microdata; however, we did not receive signicant results in
the panel model. This may be partly aributed to the costs of agent sales preceding
the outbreak of the crisis, as they were not part of banks’ operang costs.
Based on the internaonal data available, the Hungarian banking sector is among
the less cost-ecient banking systems (Figure 11). Obviously, the magnitude of
operang costs cannot be fully separated from non-performing loans; indeed several
items related to the management of the NPL porolio raise the costs incurred by
banks. Such costs include, for example, the need for personal treatment in the
case of a bad loan, or the connuous safeguarding and potenal upkeep of already
recovered collateral.
(2014)
0
2
4
6
8
10
12
0
2
4
6
8
10
12
SI
LU
LT
HU
BE
PL
DK
IE
ES
FI
DE
HR
IT
NL
GB
RO
AT
BG
FR
SK
GR
EE
CZ
LV
PT
SE
Per cent Per cent
Note: Columns indicate the 25–75th percentiles of the ratios of individual banking systems, while lines
show the 10–90th percentiles. Countries are ranked in descending order based on the 75th percentile.
Data may be biased due to the fact that the value of risk-weighted assets is sensitive to the methodology
applied by the bank (standard or IRB method), but for lack of internationally available data, we are
unable to assess the magnitude of this effect. In certain countries, there were few banks for which infor-
mation was available and the distribution might only reflect the data of a single institution or an ext-
remely limited number of institutions.
Source: SNL Financial.
36 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
According to the ndings of the internaonal literature, banks are more prone
to set lower interest rates if they also collect income from services other than
loan contracts. Although this was conrmed by the esmates we performed on
microdata, it was not a signicant variable according to the results of the panel
model. In Hungary, the rao of net income from fees and commissions is relavely
small compared to other European countries (Figure 12), which may also contribute
to higher spreads.
Our models demonstrated that a higher stock of capital is generally associated
with higher spreads. This eect has been idened in the internaonal literature
as well. Based on the CET1 (Common Equity Tier 1) rao – which is composed of
the best capital elements – Hungarian banks cannot be deemed overcapitalised by
European standards (Figure 13). It hinders data comparability – especially in the case
of CEE countries – that the capital posion of a bank largely depends on the capital
allocaon strategy pursued by the non-resident parent bank, i.e. in which country
the bank holds the buer set aside on top of its consolidated capital requirement.
(2008-2013, mean)
0
5
10
15
20
25
30
35
40
45
50
0
5
10
15
20
25
30
35
40
45
50
LU
FR
IT
UK
PT
LV
LT
PL
FI
DE
SE
AT
SI
HR
BE
ES
EE
CZ
RO
NL
BG
SK
DE
HU
IE
CY
GR
MT
Per cent Per cent
Source: ECB Consolidated Banking Data.
37
Idenfying the determinants of housing loan margins in the Hungarian banking system
It is not only the size of the capital buer that is relevant to a bank’s capital posion,
but also the expected minimum statutory adjustment to its level. In parallel to the
development of macroprudenal strategy, regulatory authories have gained access
to several new discreonary instruments in recent years that exert an impact on
banks’ capital posion (systemic risk buer, countercyclical capital buer, capital
buer applicable to systemically important instuons). In Hungary, the level of the
countercyclical capital buer has remained at zero per cent since its introducon,
but the other two instruments have higher levels. In our opinion, however, these
rules cannot be a signicant factor in the deviaon of Hungarian spreads from the
internaonal average; rst, because they are also used in other countries (ESRB
2016:52) and second, because banks are only required to comply with these two
rules, for the rst me, from 2017, which means that their eect must have been
rather muted during our review period (2014–2015 and 2005–2014).
Hungarian banks apply a higher spread on housing loans than most of their
European counterparts. This paper invesgated the reasons for the high spread
using econometric tools, along with simple stascal examinaons. In the absence
of a reliable, adequately detailed internaonal database that covers a suciently
(at end-2015)
0
5
10
15
20
25
30
35
40
0
5
10
15
20
25
30
35
40
EE
LT
IE
FI
LU
BG
LV
SE
MT
SI
HR
RO
GR
DK
SK
CZ
CY
BE
DE
NL
PL
UK
HU
ES
AT
FR
PT
IT
Per cent Per cent
CET1 rao
Source: ECB Consolidated Banking Data.
38 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
long me horizon, we aempted to idenfy the determinants of the spreads on the
basis of Hungarian bank and transacon-level data. In the last step, we examined
the Hungarian banking system’s sectoral performance relave to other European
regions with respect to the main determinants idened.
Our results showed that the spreads diverging from those of the region are primarily
caused by the higher spreads applied for loans extended with an inial interest rate
xaon of over 1 year, while the spread on loans with short-term variable rates has
already approached the regional average. Although the dierence between the
interest rates on variable and xed-rate loans is relavely high in Hungary (partly
as a result of the steeper yield curve), the share of loans with an interest rate
xaon of over 1 year within newly disbursed loans is over 50 per cent. This means
that borrowers are willing to pay a high premium in exchange for a xed interest
rate, even when adjusted for the higher costs associated with xed-interest funds.
Households’ negave experiences during the period of foreign currency lending
may have been an important contributor to this risk aversion.
The rao of non-performing loans, which is also high by internaonal standards, may
have been another factor in the emergence of high spreads. Banks set their spreads
in consideraon of the credit losses incurred, and higher credit risks are typically
coupled with higher spreads. Through collateral recovery rates, the eciency of
the legal enforcement system may also play a role in the evoluon of the spreads.
According to our esmates, the high share of operang costs may also induce
higher spreads. Banks’ expected rate of return will warrant higher spreads if their
cost-eciency is inadequate. The relavely small impact of other net income items
may also play a role: banks are more prone to set higher interest rates if they do not
collect income from any other services. We could only demonstrate these last two
eects in our esmates performed on microdata. Even in terms of these variables,
the performance of the Hungarian banking sector is worse than the internaonal
average.
Our analysis also suggested that, owing to customers’ limited price exibility and
the geographical distribuon of branches, compeon is inadequate in the eld
of housing loans. Our demand model showed that, on the one hand, customers
face geographical limitaons: only a strictly limited group of banks has presence in
many Hungarian administrave districts and customers tend to choose the easily
accessible banks. On the other hand, banks’ business models also reduce the
number of instuons that are perceived by consumers as potenal opportunies;
indeed, the banks which target auent customers do not make eorts to serve
low-income customers.
Thirdly, certain taste paerns suggest that customers rely on an extremely limited,
preferred group of banks in making their borrowing decisions, and are only willing
39
Idenfying the determinants of housing loan margins in the Hungarian banking system
to compare the oers of these chosen banks. These factors, overall, enable banks
to set their prices in the context of oligopolisc compeon. It is also a sign of weak
compeon that banks do not pass on to customers the full subsidy in the case of
subsidised loans as they – according to our esmates – overprice these loans by
about 30–35 per cent of the subsidy.
Aczél, Á. (2016): Who is interested? Esmaon of demand on the Hungarian mortgage
loan market in a discrete choice framework. 5th EBA Policy Research Workshop, under
publicaon.
Badarinza, C. – Campbell, J.Y. – Ramarodai, T. (2014): What calls to ARMs? Internaonal
evidence on interest rates and the choice of adjustable-rate mortgages. NBER Working
paper, No. 20408, Naonal Bureau of Economic Research.
Banai, Á. – Vágó, N. (2016): Drivers of household credit demand before and during the crisis.
Manuscript. Magyar Nemze Bank.
Buon, R. – Pezzini, S. – Rossiter, N. (2010): Understanding the price of new lending to
households. Bank of England Quarterly Bullen, 2010 Q3, pp. 172–182.
Capelle-Blancard, G. – Havrylchyk, O. (2013): Incidence of bank levy and bank market
power. CEPII Working Paper, No. 2013–21. Centre dʼétudes prospecves et dʼinformaons
internaonales.
Carlehed, M. – Petrov, A. (2012): A methodology for point-in-me – through-the-cycle
probability of default decomposion in risk classicaon systems. Journal of Risk Model
Validaon, Volume 6. No. 3. Fall, pp. 3–25.
Dancsik, B. – Fábián, G. – Fellner, Z. – Horváth, G. – Lang, P. – Nagy, G. – Oláh, Zs. – Winkler,
S. (2015): Comprehensive analysis of the non-performing household mortgage porolio
using micro-level data. MNB Occasional Papers, Special Issue. Magyar Nemze Bank.
Demirguc-Kunt, A. – Huizinga, H. (1999): Determinants of commercial bank interest margins
and protability: Some internaonal evidence. World Bank Economic Review, Vol. 13, pp.
379–408.
Demirguc-Kunt, A. – Laeven, L. – Levine, R. (2003): The impact of bank regulaons,
concentraon and instuons on bank margins. World Bank Policy Research Working
Paper, No. 3030. World Bank.
ECB (2009): Housing nance the Euro Area. Occasional Paper, No. 101. European Central
Bank.
40 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
EMF (2012): Study on mortgage interest rates in the EU. European Mortgage Federaon.
EMF (2016): European Mortgage Federaon Quarterly Review, 2016 Q1, European Mortgage
Federaon.
ESRB (2015): Report on residenal real estate and nancial stability in the EU. European
Systemic Risk Board.
ESRB (2016): A Review of Macroprudenal Policy in the EU in 2015. European Systemic Risk
Board.
Gambacorta, L. (2014): How do banks set interest rates? NBER Working Paper, No. 10295.
Naonal Bureau of Economic Research.
Ho, T.S.Y. – Saunders, A. (1981): The Determinants of Bank Interest Margins: Theory and
Empirical Evidence. The Journal of Financial and Quantave Analysis, Vol 16, No. 4.
Proceedings of the 16th Annual Conference of the Western Finance Associaon, pp.
581–600.
Holmberg, U. – Janzén, H. – Oscarius, L. – Van Santen, P. – Spector, E. (2015): An analysis
of the xaon period for Swedish mortgages. Economic Commentaries, No. 7, pp. 1–19.
Johansson, J. – Lagerwall, B. – Lundvall, H. (2011): Larger share of variable mortgages –
how does this aect the impact of monetary policy? In: Sveriges Riksbank: The Riksbank’s
inquiry into the risks in the Swedish housing market. Sveriges Riksbank, pp. 97–108.
Laeven, L. – Majnoni, G. (2005): Does judicial eciency lower the cost of credit? Journal of
Banking & Finance 29, pp. 1791–1812.
MNB (2016): Housing Market Report. May. Magyar Nemze Bank.
Maudos, J. – De Guevara, F. (2004): Factors explaining the interest margin in the banking
sectors of the European Union. Journal of Banking & Finance, No. 28, pp. 2259–2281.
Santos, C. (2013): Bank interest rates on new loans to non-nancial corporaons – one rst
look at a new set of micro data. In: Financial Stability Report 2013, Bank of Portugal, pp
127–134.
Saunders, A. – Schumacher, L. (2000): The determinants of bank interest rate margins: an
internaonal study. Journal of Internaonal Money and Finance, No. 19, pp. 813–832.
Train, K.E. (2002): Discrete Choice Methods with Simulaon. Cambridge University Press.
Cambridge.
Valverde, S.C. – Fernández, F.R. (2007): The determinants of bank margins in European
Banking. Journal of Banking & Finance, No. 31, pp. 2043–2063.
41
Idenfying the determinants of housing loan margins in the Hungarian banking system
Variable rate or initial fixation of
up to 1 year
33,705 51.93
Interest rate with a fixation of
over 1 year
31,199 48.07
Total 64,904 100
Source: MNB.
Number of contracts Distribution of contracts
Market based 45,854 70.65
Subsidised 19,050 29.35
Total 64,904 100
Source: MNB.
Mean Median
Mean Median
2014 Q1 7.9 7.7 5.7 9.7 5.1 4.9 3.0 6.9
2014 Q2 7.7 7.4 5.4 9.7 5.2 4.8 3.0 7.2
2014 Q3 6.8 6.6 4.7 8.5 4.6 4.4 2.5 6.3
2014 Q4 6.5 6.4 4.6 8.0 4.4 4.3 2.5 5.9
2015 Q1 6.3 6.2 4.6 7.7 4.3 4.2 2.5 5.6
2015 Q2 6.1 5.9 4.1 7.4 4.5 4.3 2.5 5.8
2015 Q3 5.7 5.3 3.6 7.2 4.4 3.9 2.2 5.9
2015 Q4 5.7 5.5 3.7 7.2 4.3 4.2 2.4 5.9
Total 6.4 6.3 4.2 8.5 4.5 4.4 2.5 6.3
Source: MNB.
42 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Mean Median
Mean Median
2014 Q1 5.4 4.3 1.5 10.0 173.2 180.2 72.2 241.0
2014 Q2 5.8 4.5 1.5 10.0 172.1 180.2 72.4 241.0
2014 Q3 6.1 5.0 1.7 11.0 175.6 180.3 72.6 241.1
2014 Q4 6.1 5.0 1.9 11.5 180.2 180.4 72.9 241.7
2015 Q1 6.4 5.0 2.0 11.8 179.9 180.3 72.5 264.1
2015 Q2 6.5 5.0 1.6 12.5 175.3 180.2 71.0 265.2
2015 Q3 7.2 5.8 2.0 13.7 174.3 180.0 72.6 252.7
2015 Q4 7.2 5.7 2.0 13.7 180.1 180.4 72.8 299.3
Total 6.5 5.0 1.8 12.0 176.4 180.3 72.5 241.4
Source: MNB.
BUBOR_spread
contracted
amount_ln
maturity_month
d_collateral
subsidy
d_fixation
liquid
capital buffer
cta
prov_avg
branch
fcomm_fee
BUBOR_spread 1.00
contracted
amount_ln
–0.43 1.00
maturity_month –0.19 0.43 1.00
d_collateral –0.34 0.08 0.12 1.00
subsidy 0.28 –0.06 0.04 0.07 1.00
d_fixation 0.45 –0.11 –0.04 –0.06 0.36 1.00
liquid –0.46 0.19 0.07 0.16 –0.28 –0.29 1.00
capital buffer 0.27 –0.16 –0.03 0.06 0.29 0.10 –0.41 1.00
cta 0.32 –0.12 –0.17 –0.38 –0.14 –0.15 0.03 –0.02 1.00
prov_avg 0.45 –0.19 –0.14 –0.24 0.07 0.09 –0.45 0.22 0.47 1.00
branch 0.14 –0.13 0.08 0.24 0.24 0.18 –0.16 0.38 –0.37 0.00 1.00
fcomm_fee –0.22 0.05 0.08 0.24 0.09 –0.12 0.28 –0.01 –0.27 0.00 0.33 1.00
Source: own calculations.
43
Idenfying the determinants of housing loan margins in the Hungarian banking system
Mean Median
Spread 0.00 0.03 –4.56 4.55
Operating costs –0.01 0.01 –0.38 0.36
Other revenue/
interest revenue
25.16 24.21 13.48 37.89
Liquidity 14.74 13.45 3.43 27.06
CAR 12.06 10.70 8.70 16.45
GDP (YoY) 1.76 2.32 –2.17 4.58
NPL 6.68 3.62 0.83 17.12
LTV 55.13 56.10 27.10 80.95
External liabilities 39.48 38.74 14.34 60.02
Provisions 0.22 0.11 0.56 0.01
Market share 0.09 0.05 0.02 0.19
Proportion of
branches
10.73 8.59 1.45 26.66
Ratio of fixed-
interest loans
steepness of the
yield curve
0.71 0.73 0.19 1.10
Source: MNB.
44 Studies
Ákos Aczél – Ádám Banai – András Borsos – Bálint Dancsik
Spread
Operating costs
Other revenue/
interest revenue
Liquidity
CAR
GDP (YoY)
NPL
LTV
External liabilities
Provisions
Market share
Proportion of branches
Ratio of fixed-interest loans
steepness of the yield curve
Spread 1.00
Operating costs 0.01 1.00
Other revenue/
interest revenue –0.03 0.00 1.00
Liquidity 0.01 0.01 0.59 1.00
CAR –0.01 –0.04 0.15 0.21 1.00
GDP (YoY) 0.02 –0.01 0.33 0.12 0.13 1.00
NPL 0.00 0.01 0.20 0.48 0.27 0.13 1.00
LTV 0.02 –0.05 0.13 0.09 –0.27 –0.24 –0.14 1.00
External liabilities 0.00 0.00 –0.49 –0.54 –0.14 –0.15 0.03 –0.27 1.00
Provisions –0.08 0.02 0.18 0.19 0.06 0.13 –0.27 –0.01 –0.38 1.00
Market share 0.00 0.00 0.38 0.01 0.10 0.00 –0.21 0.26 –0.20 0.16 1.00
Proportion of
branches 0.00 –0.01 0.54 0.13 0.02 0.00 –0.26 0.28 –0.34 0.21 0.93 1.00
Ratio of fixed-
interest loans
steepness of the
yield curve
–0.02 0.01 0.16 0.22 0.48 0.34 0.50 –0.22 –0.01 –0.17 –0.08 –0.11 1.00
Source: own calculations.
