IRRBB Under Low Rate Environment

DEGREE PROJECT IN MATHEMATICS,
SECOND CYCLE, 30 CREDITS

STOCKHOLM, SWEDEN 2020
IRRBB in a Low Interest Rate

Environment
SIMON BERG
VICTOR ELFSTRÖM
KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ENGINEERING SCIENCES
IRRBB in a Low Interest Rate
Environment
SIMON BERG
VICTOR ELFSTRÖM
Degree Projects in Financial Mathematics (30 ECTS credits)

Master's Programme in Applied and Computational (120 credits)
Mathematics KTH Royal Institute of Technology year 2020
Supervisors at zeb Nordics: Markus Ahlgren
Supervisor at KTH: Boualem Djehiche
Examiner at KTH: Boualem Djehiche
TRITA-SCI-GRU 2020:058
MAT-E 2020:021
Royal Institute of Technology

School of Engineering Sciences
KTH SCI
SE-100 44 Stockholm, Sweden
URL: www.kth.se/sci
ii
Abstract
Financial institutions are exposed to several different types of risk. One of the risks
that can have a significant impact is the interest rate risk in the bank book (IRRBB).
In 2018, the European Banking Authority (EBA) released a regulation on IRRBB
to ensure that institutions make adequate risk calculations. This article proposes
an IRRBB model that follows EBA’s regulations. Among other things, this frame-
work contains a deterministic stress test of the risk-free yield curve, in addition to
this, two different types of stochastic stress tests of the yield curve were made. The
results show that the deterministic stress tests give the highest risk, but that the out-
comes are considered less likely to occur compared to the outcomes generated by
the stochastic models. It is also demonstrated that EBA’s proposal for a stress model
could be better adapted to the low interest rate environment that we experience now.
Furthermore, a discussion is held on the need for a more standardized framework
to clarify, both for the institutions themselves and the supervisory authorities, the
risks that institutes are exposed to.
iii
IRRBB i en Lågräntemiljö
Sammanfattning
Finansiella institutioner är exponerade mot flera olika typer av risker. En av de ris-
ker som kan ha en stor påverkan är ränterisk i bankboken (IRRBB). 2018 släppte
European Banking Authority (EBA) ett regelverk gällande IRRBB som ska se till
att institutioner gör tillräckliga riskberäkningar. Detta papper föreslår en IRRBB
modell som följer EBAs regelverk. Detta regelverk innehåller bland annat ett deter-
ministiskt stresstest av den riskfria avkastningskurvan, utöver detta så gjordes två
olika typer av stokastiska stresstest av avkastningskurvan. Resultatet visar att de
deterministiska stresstesten ger högst riskutslag men att utfallen anses vara mindre
sannolika att inträffa jämfört med utfallen som de stokastiska modellera generera-
de. Det påvisas även att EBAs förslag på stressmodell skulle kunna anpassas bättre
mot den lågräntemiljö som vi för tillfället befinner oss i. Vidare förs en diskus-
sion gällande ett behov av ett mer standardiserat ramverk för att tydliggöra, både
för institutioner själva och samt övervakande myndigheter, vilka risker institutioner
utsätts för.
iv
Acknowledgements
We would express our gratitude to everyone at zeb, who has welcomed us and
contributed their expertise. In particular, we want to thank Markus Ahlgren, our
supervisor at zeb, for his advice and support. We would also like to thank Boualem
Djehiche, our supervisor at KTH, for his feedback and guidance. Last but not least,
we want to thank our families for their support. We are very grateful that you are
always there for us.
Contents
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Previous Research . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Theoretical Background 7
2.1 IRRBB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Gap Risk . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Basis Risk . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Behavioral Risk . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The Legal Environment of IRRBB . . . . . . . . . . . . . . . . . 10
2.3 Measures of IRRBB . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Balance Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Mathematical Framework 14
3.1 Nelson-Siegel-Svensson Parameter Estimation . . . . . . . . . . . 14
3.2 Principal Component Analysis . . . . . . . . . . . . . . . . . . . 16
3.3 Cholesky Factorization . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Forward Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.5 Bachelier Pricing Model . . . . . . . . . . . . . . . . . . . . . . 19
4 Model 21
4.1 Mapping of the Cash Flow Data . . . . . . . . . . . . . . . . . . 21
4.2 Yield Curve Construction . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Interest Rate Models . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1 Interest Rate Restrictions . . . . . . . . . . . . . . . . . . 24
4.3.2 EBA Interest Rate Model . . . . . . . . . . . . . . . . . . 24
v
CONTENTS vi
4.3.3 PCA Interest Rate Model . . . . . . . . . . . . . . . . . . 26

4.3.4 Cholesky Interest Rate Model . . . . . . . . . . . . . . . 27
4.4 Refinancing Risk . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.5 Gap Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5.1 EVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5.2 NII . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.6 Basis Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.7 Option Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 Results and Discussion 35

5.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Generated Yield Curves . . . . . . . . . . . . . . . . . . . . . . . 37
5.3 Gap Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.4 Basis Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.5 Behavioral Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.6 Discussion About the IRRBB Model . . . . . . . . . . . . . . . . 51
6 Conclusions 54
6.1 Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Bibliography 57
List of Figures
5.1 Cash flows in SEK, EUR and DKK converted to SEK. Note the
scale difference for the different currencies. . . . . . . . . . . . . 36
5.2 Original yield curves for SEK, EUR and DKK. . . . . . . . . . . 36
5.3 Yield curves in SEK, EUR and DKK generated by the EBA model. 37
5.4 Describes the generated yield curves of the Cholesky and PCA
models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.5 Describes the maximum/minimum yield generated for each bucket
by the PCA and Cholesky model. . . . . . . . . . . . . . . . . . . 41
5.6 Describes the gap risk measure for both the PCA and Cholesky
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
vii
List of Tables
4.1 Describes the different time buckets and the time period each bucket
represent. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Describes the individual shock parameters for some currencies. . . 25
4.3 Describes scenario multipliers for each EBA scenario. . . . . . . . 29
5.1 Gap risk measures given in million SEK for each of the six EBA
scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Describes gap risk measures given in million SEK for the three
models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.3 Describes the basis risk in each currency given in million SEK. . . 48
5.4 Behavioral risk measures given in million SEK for each of the six
EBA scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.5 Describes behavioral risk measures given in million SEK for the
three models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
viii
Chapter 1
Introduction
In this chapter, a shorter background about interest rate risk in the banking book
and the paper itself will be presented. Furthermore, this chapter will present the
paper’s aim, research question, and outline.
1.1 Background
When talking about investments, people mainly want to analyze the expected earn-
ings and the risk associated with an investment. When talking about investment
strategies, it is common to use the words risk-averse, risk-neutral, and risk-seeking
to explain how much risk an investor is willing to take in order to get a higher
expected return on the investment. Financial institutions have different investment
strategies, and their risk exposure is varying. However, regardless of an institution’s
investment strategy, they need to do risk assessments regularly [1]. This in order
to ensure that their business still can operate, even in case of unfavorable events.
Financial institutions are exposed to many different kinds of risks. This paper will
focus on the interest rate risk, which is one type of risk that can have a significant
impact on companies. Interest rate risk is a subject that has been important for
banks and institutions for decades [2]. One of the more common interest rate risks
is the so-called Interest Rate Risk In The Banking Book (IRRBB).
“IRRBB refers to the current or prospective risk to the bank’s capital and earn-
ings arising from adverse movements in interest rates that affect the bank’s banking
book positions. When interest rates change, the present value and timing of future
cash flows change. This, in turn, changes the underlying value of a bank’s assets,
liabilities, and off-balance sheet items and hence its economic value. Changes in
1
CHAPTER 1. INTRODUCTION 2
interest rates also affect a bank’s earnings by altering interest-rate-sensitive in-

come and expenses, affecting its net interest income (NII). Excessive IRRBB can
pose a significant threat to a bank’s current capital base and/or future earnings if
not managed appropriately” [3].
The quote above is the Basel Committee on Banking Supervision (BCBS) descrip-
tion of IRRBB. The importance of IRRBB comes from the possibility of wrongly
estimate an institution’s exposure to this risk. Inaccurate estimates of IRRBB can
lead banks to reserve an amount of internal capital that does not reflect the bank’s
actual riskiness. Both underestimation and overestimation potentially have negative
consequences. Underestimation might jeopardize banks’ stability, whereas overes-
timation might charge banks higher opportunity costs and, eventually, reduce their
credit supply to the economy [2].
The banking sector plays a vital role in society, and therefore the role of regu-
lators becomes essential in this sector. The purpose, briefly explained, with the
regulations is to promote stability and efficiency in the financial system as well as
to ensure effective consumer protection [4]. Different institutions are controlled by
different supervisory authorities and are also acting under various regulations. The
Swedish financial supervisory authority, which decides the rules that the Swedish
banking sector needs to act under, is called Finansinspektionen (FI) [5]. FI recently
decided that the Swedish banking sector is required to follow the European Banking
Authority’s (EBA) new guidelines on the management of interest rate risk arising
from non-trading book activities, i.e., the IRRBB. EBA’s guidelines on IRRBB are
based on the Basel Committee on Banking Supervision updated standards on IR-
RBB, which was published in 2016.
“Institutions should manage and mitigate risks arising from their IRRBB exposures
that affect both their earnings and economic value “ [1].
Risks arising from adverse interest rate movements can be analyzed in several ways,
where Economic Value (EV) and Net Interest Income (NII) are commonly used ap-
proaches. However, historically there has been no clear opinion about which per-
spective an institution should mainly focus on when determining its risk exposure
and hence its capital requirement. Both of the measures have advantages and dis-
advantages, and neither has yet to prevail as the standard [6]. Swedish institutions
have typically focused on EV, since FI’s old regulations mainly took this measure
into account.
The new guidelines from EBA, which came into force in summer 2019, require the
banking sector to manage their interest rate risks that arise from both the changes
in economic value and earnings [1]. More precisely, the EBA framework requires
financial institutions to have at least one NII measure, and one EVE measures that
combined should calculate the three risks Gap, Basis, and Option. By which meth-
ods these measures should be calculated is not defined by the EBA. This creates
great freedom for a risk manager to choose its own methods. However, do this cre-
ate uncertainties in the reported calculations from the financial institutions? Which
methods are, in fact, best adopted to calculate these risk measures under a low-
interest environment? Furthermore, EBA has a few suggestions on how to calculate
some features in an IRRBB model. One of these is how to stress the current yield
curve, where the suggested method is highly standardized. How do these standard-
ized stresses affect the IRRBB calculations? Is there any other method during the
current circumstances that is better to use. These questions are some that this paper
set out to answer.
1.2 Previous Research

Many previous studies have dealt with the topic of IRRBB. Several different ways of
generating interest rate scenarios and different methods of calculating the changes
in earnings and economic value have been proposed.
For instance, Fiori and Iannotti [7] evaluated risk exposure through a principal
component Value-at-Risk method based on Monte Carlo simulations. They cre-
ated different interest rates scenarios by doing a principal component analysis on
historical yield curves. By shocking the principal components randomly and using
a Monte Carlo simulation, they got theoretical yield curves scenarios. Kreinin et al.
[8] used a similar approach, and they demonstrate that dimensionality reduction can
result in substantial computational savings reduction. Combining PCA and Monte
Carlo simulation is a useful way to revalue a portfolio fully under many different
scenarios [9].
Abdymomunov and Gerlach [10] looked into six different methods based on in-
dustry practices, academic studies, and banking supervisory requirements. One of
these methods was based upon a PCA approach. Both the studies of Fiori and Ian-
notti [7] and Abdymomunov and Gerlach [10] came to the conclusion that stochas-
tic approaches can be used in favor of the deterministic scenarios.
Cerrone et al. [2] created their own model based upon a Cholesky factorization
of historical yields covariance matrix. They argue in their paper that the model
is superior to other models due to that it is more consistent with actual riskiness
than, for example, the standardized shocks presented in BCBS [11]. Furthermore,
they argue that stochastic simulation techniques overcome the limits of the current
deterministic scenarios since they allow estimating a bank’s equity sensitivity in a
wider set of adverse scenarios and capturing interest rates dynamics over time [2].
Fiori and Iannotti [7], Cerrone et al. [2] and Abdymomunov and Gerlach [10] eval-
uated their models towards a Value-at-Risk approach where they only measured
the change of economic value of equity, i.e. EV EGap , which is not a sufficient
model according to EBA [1]. However, at the time of their research, the new EBA
framework was not yet presented. Therefore they compare their model to earlier
frameworks such as BCBS [11] and BCBS [3]. These frameworks were less strict
where, for example, in BCBS [11], the only required deterministic stresses were the
parallel up and parallel down. In comparison, EBA [1] has four additional stresses
in its framework, see Section 4.3.2. Furthermore, the old frameworks had more
differing features that make these studies a bit out of date. All three of the earlier
mentioned paper restricted the yield curves according to BCBS [11] and, BCBS
[3], where no negative interest rates nor negative forward rates were allowed. Neg-
ative interest rates are not only allowed in the new EBA framework, but it is also
encouraged to exist. Furthermore, these articles are mainly focused on how interest
rate scenarios could be generated, not on how the interest rate generators affect a
complete IRRBB model.
As mentioned earlier, there exist many studies that investigate how to measure IR-
RBB. For instance, Barbi [12] takes a closer look at the skewness behavior of the
basis risk and how to hedge this risk in an appropriate way. Both Memmel [13] and
Hibiki and Iwakuma [14] papers evaluate how the behavior of gap risk earnings and
Economic value of equity measures relates to each other. Both evaluate these risk
measures overtime, where Hibiki and Iwakuma [14] simulate data with the help
of a Nelson-Siegel-Svensson, see Equation 3.1, and an AR(1) approach. While
Memmel [13] looks at actual yield data over the period January 1980 to Decem-
ber 2010. Memmel [13] concludes that changes in banks’ economic value and net
interest income are highly correlated. Moreover, banks’ portfolio composition has
a huge impact on the ratio of changes in net interest income relative to changes in
Economic value [13]. However, to the best of our knowledge, there exist no articles
that suggest a complete IRRBB model that captures all parts of the risk agreeing
to EBA [1].
To conclude, as far as we know, this is the first paper that proposes a complete
IRRBB model. In other words, a model that follows the EBA-framework and cap-
tures gap, basis, and behavioral risk. Furthermore, this paper analyzes how differ-
ent interest rate scenarios affect the EVE and NII for a given banking book. It also
provides an insight into how modeling of IRRBB can look like and how different
assumptions can be made.
1.3 Purpose
The new framework is creating new challenges for risk managers who need to adapt
to these new guidelines. Especially under the historically low interest rate environ-
ment that the financial world finds themself in. This paper aims to test this new
framework by creating a method that follows the EBA standards to calculate a fi-
nancial institution’s IRRBB. The model will be stressed with different techniques to
generate yield curve scenarios. This in order to test how an EBA model behaves in
a low interest environment and to investigate which method is better suited to stress
this model under the current climate. Furthermore, a discussion will be included on
which mathematical choices a risk manager has when creating their IRRBB model
and which advantages some calculations have compared to others. This will show
the significant variation each institutions’ IRRBB model will have in both resulting
risk and behavior. This paper will contribute to the current research in the field of
IRRBB by creating a better understanding of the effects of a low interest environ-
ment on the calculations of IRRBB.
In order to perform this research, some limitations of the scope were needed. This
paper will only focus on the actual calculations in an IRRBB model. This article
will leave the more managerial questions unaddressed, for example, who should be
in charge of the IRRBB management and how the information system connected to
IRRBB should work.
1.4 Research Question

To fulfill the purpose of the article, the main question that this paper set out to
answer is formulated as:
• How to manage IRRBB according to the new EBA-regulations in a low-

interest environment?
In order to answer this research question, an IRRBB model that fulfills the new EBA
standards will be created. This model, combined with the following sub-research
questions, will be used to enrich the answer to the main question:
• How to build a model that fulfills the EBA standards?

• How do the individual choices of a risk manager affect the calculated risk?
• How to stress test this model?
• What are the pros and cons of the different stress test?
• What effects do a low-interest environment have on the risk model?
• What weaknesses do the new EBA-regulations have?
1.5 Outline of the Thesis

The layout of the report is formatted in a way to first give the reader an introduction
and the necessary information about the topic. Then comes a more mathematical
and theoretical base that has laid the foundation for the report’s calculation. Fi-
nally, the report’s results are presented with associated analysis, discussion, and
conclusions. In particular, chapter 2 will address the background on IRRBB, ex-
plaining the three sub-categories which constitute the IRRBB, and the standard
measures that are used to determine the risk. Furthermore, a summary of the signif-
icant changes between previous frameworks and the current framework will be pre-
sented. In Chapter 3, the mathematical framework is provided. It will address the
essential mathematical methods and concepts considered in this paper, like Prin-
cipal Component Analysis, Cholesky decomposition, and Bachelier option pricing
model. Chapter 4 explains how the mathematical framework was used to construct
the models that, in turn, generated the result. The result, together with analysis
and discussion, are presented in chapter 5. Finally, conclusions and suggestions for
further research are presented in Chapter 6.
Chapter 2
Theoretical Background
This chapter will provide a more thorough background description of the topic as
a whole as well as a description of concepts that are important for understanding
the necessary calculations.
2.1 IRRBB
Interest rate risk in the banking book may, for practitioners, be a self-explanatory
phrase. It refers to the risk which arises from adverse movements in interest rates
that affect the banking book positions [15]. Where banking book positions refer to
all interest-rate sensitive instruments from non-trading activities, i.e., assets, liabili-
ties, and off-balance-sheet items in the non-trading book, excluding assets deducted
from CET1 capital [1]. To understand the difference between the banking book and
the trading book, one must understand the difference in the assets held by these ac-
counts. The clear distinction between these accounting books is that the banking
book instruments are, in general, intended to be held to maturity. In contrast, the
assets in the trading book are, in general, held on a shorter horizon. For example,
trading book assets might be bought or sold to facilitate trading activities for clients
or to profit from spreads between the bid and ask prices.
Interest rate changes affect both a financial institution’s economic value and its
earning capability. It is these risks that constitute interest rate risk in the bank
book. An institution’s IRRBB may sound simple to determine, but a banking book
exists of multiple positions in a wide range of asset classes, which makes the calcu-
lations more complicated. Therefore, to investigate this risk, one must understand
the drivers of it. There are different names and definitions of these drivers, but since
this paper aims to propose a model that fulfills the EBA-standards, its definitions
7
CHAPTER 2. THEORETICAL BACKGROUND 8
will mainly be used in this paper. EBA [1] divides the drivers into the three major
groups Gap risk, Basis risk, and Option risk. To avoid confusion between option
risk and the automated option risk, we will, in this paper, refer to option risk as be-
havioral risk, which is a commonly adopted name of this risk. These three drivers
depend directly on the level and shape of the yield curve, which forces financial
institutions to do significant stress tests on the yield curve. The stress testing is es-
sential to get information about how different economic environments would affect
the IRRBB for an institution. Therefore, the effectiveness of a stress test depends
directly on the yield curve scenarios that have been used in the stress test [10].
2.1.1 Gap Risk

The first major driver of risk defined by EBA [1] is called gap risk and is defined as :
"Risk resulting from the term structure of interest rate sensitive instruments that
arises from differences in the timing of their rate changes, covering changes to the
term structure of interest rates occurring consistently across the yield curve (par-
allel risk) or differentially by period (non-parallel risk)." [1]
This driver of risk is essentially one measure of two different risks that are closely
related to each other, namely yield curve risk and repricing risk [11].
Repricing risk is the primary and most discussed form of interest rate risk, and
it arises mainly from timing differences in the maturity and repricing of financial
institutions assets, liabilities, and OBS position. While such repricing mismatches
are fundamental to the business of banking, they can expose a bank’s income and
underlying economic value to unanticipated fluctuations as interest rates vary. For
instance, a bank that funded a long-term fixed-rate loan with a short-term deposit
could face a decline in both the future income arising from the position and its
underlying value if interest rates increase. These declines occur because the cash
flows from the loan are fixed over its lifetime, while the interest paid on the funding
is varying [11].
Repricing mismatches can also expose a bank to risks in case of changes in the
slope and shape of the yield curve (Yield Curve Risk). This risk occurs when
unanticipated shifts of the yield curve have adverse effects on a financial institu-
tion’s income or underlying economic value. Consider a portfolio with only two
instruments consisting of a long position in 10-year government bonds and a short
position in 5-year government notes. The economic value of this portfolio could
decline sharply if the yield curve steepens (the longer-horizon yield drastically in-
creases while medium-horizon yield stays the same). The reason for this is that the
value of the long position would, in this scenario, drastically decrease while the
short position would essentially remain the same.
2.1.2 Basis Risk

The second major driver of IRRBB is commonly called basis risk. This risk refers
to the impact of an imperfect correlation between indices underlying asset and li-
ability rates, with otherwise similar repricing frequency. For example, 6-month
PLR-linked loans can be funded by 6-month LIBOR-linked deposits, but there is
no guarantee that a 1% change in LIBOR will imply a 1% change in PLR. If the
spread between index rates suddenly changes, it might lead to a decrease in earn-
ings and net worth [16]. In other words, basis risk captures the possible loss from
the imperfect correlation between different interest rates, which assets and liabili-
ties are priced upon [12]. It is important to capture this risk since it is a common
practice to ignore which yield curve instruments are based upon when doing gap
risk calculations. Thus, one can see the basis calculations as a complement to the
gap risk calculations.
2.1.3 Behavioral Risk

The third major driver of IRRBB is called behavioral risk, which EBA call option
risk. There is no consensus about a definition of behavioral risk [17]. However,
in this paper, behavioral risk will be parted into two different categories in accor-
dance with EBA [1]. The first part of the risk arises from options (embedded and
explicit), where the institution or its customer can alter the level and timing of their
cash flows. Hence, this risk arises from interest-rate sensitive instruments where
the holder will almost certainly exercise the option if it is in their financial interest
to do so (embedded or explicit automatic options) [1]. This part of the behavioral
risk will, in this paper, be referred to as option risk, which is not to be confused
with EBA’s definition. They use this term when they are talking about what this
paper refers to as behavioral risk, i.e., both risk that occurs from financial options
and customer behavior.
The second part is commonly called prepayment risk, withdrawal risk, refinanc-
ing risk, or optionality risk. This risk occurs when banking books contain numer-
ous implicit options such as prepayment options on mortgages, borrowing options,
early withdrawal options, interest rate options, etc. As these options may be exer-
cised in response to market interest rate changes, they induce (non-linear) interest
rate risk [18]. In other words, this risk occurs because different interest rate swings
affect customer behavior. For example, people tend to pay back more on their fixed
interest rate loans during periods of falling interest rates and less during periods of
rising interest rates [3]. One of the main problems of estimating this risk is that one
needs to predict the behavior of the customers, which are extremely hard in some
businesses, like in retail products [18]. Consider the case when a customer has a
fixed-rate mortgage loan at 5%. This customer has strong incentives to refinance
his loan when the interest rate drops, and there is, for example, a similar fixed-rate
loan at 3%. This phenomenon creates uncertainties in future cash flow for institu-
tions, which may affect the NII or/and EVE significantly. In this paper, this part of
the behavioral risk will be referred to as refinancing risk.
2.2 The Legal Environment of IRRBB

The Basel Committee on Banking Supervision is the primary global standard-setter
for the prudential regulation of banks and provides a forum for regular cooperation
on banking supervisory matters. Its 45 members comprise central banks and bank
supervisors from 28 jurisdictions [4]. The committee publishes standards for bank-
ing supervision, but each country has its authority who sets the regulations. It is
up to each authority to take action to implement these regulations through their na-
tional systems. In Sweden, it is the Swedish Financial Supervisory Authority (FI)
who is the government agency responsible for financial regulation. FI is, in turn,
governed by EU laws. However, it is of great importance to keep these regulations
updated so that they hold for the current environment.
In 2016, BCBS [3] issued new standards for the interest rate risk in the banking
book. These new standards are a revised version of the committees "Principles for
the Management and Supervision of Interest Rate Risk" from 2004. Much has hap-
pened since 2004. Several countries are experiencing a low-interest rate climate,
which they have not experienced before. However, BCBS’s revised version con-
tains a lot of changes and updates. This section will mainly focus on the changes
that have an impact on the modeling and calculation of the IRRBB, and less on the
management/governance changes.
To understand the new standards and how it affects financial institutions, it is a

good start to get a view of how the standards from 2004 looked like. First of all, the
old standards contained principles on how to manage the interest rate risk (IRR),
i.e., the interest rate risk for both the banking book and the trading book. The new
standards are only focusing on the interest rate risk in the banking book. Moreover,
the previous regulations contained only two stress scenarios, one positive and one
negative parallel shift of the yield curve. It was considered important to add several
possible scenarios to capture potential risks. Hence, the new framework contains
six stress scenarios. However, BCBS decided to update the standards and came up
with two options for the regulatory treatments of IRRBB. Pillar 1, which is a stan-
dardized approach that calculates the Minimum Capital Requirements, and pilar 2,
a less standardized approach which instead contains principals to guideline banks
to measuring risk and assessing capital adequacy internally. BCBS concluded that
Pilar 2 was more appropriate to capture the heterogeneous nature of IRRBB. This
was also the proposal the industry preferred since the feasibility of Pilar 1 was
questionable. This since all banks have different products, plans, strategy, etc. and
therefore, there is no model which fit all banks.
The European Banking Authority (EBA) is one of the EU’s three joint financial reg-
ulators. On 19 July 2018, EBA published the new guidelines on the Management of
IRRBB. This update was the first step towards the implementation of BSBC’s new
standards [1]. In Sweden, FI decided to follow these standards, and this has had a
quite significant impact on Swedish financial institutions since the guidelines from
EBA differ a bit from the old regulations provided by FI. The old standards from
FI required institutions to calculate the changes in EVE under different stress tests,
which mostly capture the gap risk [5]. Furthermore, institutions were supposed to
fill out a questionnaire about their exposure to basis and options risk, i.e., they did
not have to do any calculations about it.
The updated standards require institutions to use at least one economic value mea-
sure and one earnings-based measure. Such requirements did not exist before.
Usually, only one of those measures was used. Nevertheless, it is still not com-
pletely clear how these calculations should be applied. Institutions still have a lot
of freedom in how the IRRBB should be measured, as long as they stay inside the
boundaries of the framework.
2.3 Measures of IRRBB

"The Committee acknowledges the importance of managing IRRBB through both
economic value and earnings-based measures. If a bank solely minimises its eco-
nomic value risk by matching the repricing of its assets with liabilities beyond the
short term, it could run the risk of earnings volatility." [3]
There is no single method for measuring IRRBB. Still, it is common to divide the
used methods into two different types of measures, namely earning-based measures
and economic-value measures. Interest rate changes can affect both a bank’s earn-
ings and economic value, therefore this categorization. The two categories differ
from each other and provide different perspectives on IRRBB.
Earnings measures look at changes in expected future profitability within a given

time horizon resulting from interest rate movements. Changes in the interest rate
will affect the size of cash flows, and it is the sensibility against these changes
that earnings measures look at. The time horizon for earnings-based measures is
usually the nearest 1-5 years, i.e., it captures the short term risks [1]. However,
the most common earnings-based measure is ∆N II, which basically compares
cash flows under different interest rate scenarios. How sensitive institutions are to
∆N II depends on their specific assets and liabilities, where fixed or floating rates
and repricing dates have a significant impact.
In contrast to earnings measures, economic value measures look at the net present
value of the interest rate sensitive instruments over their remaining lifetime, i.e., un-
til all positions have expired. One of the most common economic value measures
is the Economic Value of Equity. It is calculated by taking the present value of all
expected asset cash flows and subtract it with the present value of the expected cash
flows on liabilities, where equity is excluded from the cash flows. To calculate risk
with this measure, it is a common practice to compare the present scenario to all
other created scenarios and look at the change in EVE, i.e., ∆EV E
"For measuring and monitoring of IRRBB, institutions should use at least one
earnings-based measure and at least one economic value measurement method
that, in combination, capture all components of IRRBB." [1]
It is important for institutions to use both types of measures since small changes
in earnings measures dose not imply small changes in the economic value. Banks
can reduce their risk measured by earnings measures at the expense of EV. There-
fore, according to the new guidelines from EBA, institutions need to capture all of
the components of IRRBB (gap, basis, and behavioral risk) by using at least one
earnings-based measure and at least one economic value measurement method.
"In addition to the existing and prospective exposure to IRRBB, when implementing
the guidelines, institutions should also consider their general level of sophistication
and internal approaches to risk management to make sure that their approaches,
processes and systems for the management of IRRBB are coherent with their gen-
eral approach to risk management and their specific approaches, processes and
systems implemented for the purpose of the management of other risks." [1]
EBA [1] has made a sophistication matrix containing four different categories.
The different categories reflect different sizes, structures, the nature scope, and
complexity of activities of institutions. Category 1 is the one with the most so-
phisticated institutions, and category 4 is the one for less sophisticated institutions.
The guidelines for measuring IRRBB differ a bit between the categories. Supervi-
sors have different expectations on metric and modeling depending on the level of
sophistication an institution is placed in. This paper will mainly be based on the
guidelines for institutions in categories 3 and 4.
2.4 Balance Sheet

When calculating the IRRBB, an institution’s current balance sheet is used. But
since the calculations are done over a long time horizon, one must make assump-
tions about how the balance sheet will change over time. Basically, it is three dif-
ferent ways to do this; assume a run-off balance sheet, a dynamic balance sheet, or
a constant balance sheet. A run-off balance sheet is a balance sheet where existing
positions are not replaced when matured. A dynamic balance sheet is a balance
sheet where future business expectations, adjusted for the relevant scenario, are in-
cluded. A constant balance sheet is a balance sheet where maturing or repricing
cash flows are replaced with new cash flows that have identical features with regard
to the amount, repricing period, and spread components. Different calculations as-
sume different types of balance sheets. Hence, in Chapter 4, the assumed portfolio
type will be mentioned before each associated risk calculation is defined.
Chapter 3
Mathematical Framework
This chapter presents the most important and more complex mathematical concepts
used in this article. Here, the theories are presented in a comprehensive way, and
the next chapter describes how these theories have been applied in this study.
3.1 Nelson-Siegel-Svensson Parameter Estima-

tion
Central banks and other market participants widely use the Nelson-Siegel-Svensson
(NSS) equation as a model for the term structure of interest rates [19]. The NSS
equation models the yield curve at a point in time as follows [20]:
−t −t
−t

1 − e λ1 1 − e λ1
Ŷ (t) = β1 + β2 + β3 − e λ1
t/λ1 t/λ1
−t
(3.1)
−t

1− e λ2
+β4 − e λ2
t/λ2
Where Ŷ (t) denotes the zero rate for maturity t. To calibrate the model we need
to estimate six parameters: β1 ,β2 ,β3 , β4 and λ1 , λ2 . For N observed yields with
different maturities t1 , . . . , tN , we have N equations. These parameters can be
estimated by minimizing the difference between the model rates Ŷ , and observed
rates Y M where the superscript stands for ‘market’ [20]. An optimization prob-
lem can be stated and solved by a least squared method minimizing the quadratic
difference between the approximated yields and market yields by calibrating the β
and λ values in accordance to:
14
CHAPTER 3. MATHEMATICAL FRAMEWORK 15
X
min( (Ŷ − Y M )2 ) (3.2)
λ,β
To avoid potential challenges in numerical optimization, Diebold and Li [21] in-

troduces a method of fixing the λ1 and λ2 parameters and then estimate the rest
of the parameters using a least squares algorithm. Based upon the earlier research
of Diebold and Li [21], Muvingim and Kwinjo [22] tested which values of λ1 and
λ2 that would generate the best-fitted yield curve. Their research showed that a λ1
value of 0.714 and λ2 value of 10 gives the most appropriate curves. With these
values set, the optimization problem can be solved with the ordinary least square
estimation:
YM
t = X t · β̂ + t (3.3)
Where Y M t expresses the interest rates at maturity date t, t is the random error
term, X t is the vector of prediction variables that explain the dependent variable
yield. The vector X t can be estimated given maturity t and λ. Using the equation
for Nelson-Siegel-Svensson model, see Equation 3.1, the following vector for X t
can be obtained:
 −t1 −t1 −t1 

1 − e λ1 1 − e λ1 −t1 1 − e λ2 −t1
1 − e λ1 − e λ2 
 t1 /λ1 t1 /λ1 t1 /λ2 
−t2 −t2 −t2
 
−t2 −t2 
 
 1 − e λ1 1 − e λ1 1 − e λ2
1 − e λ1 −e λ 2 
Xt = 
 t2 /λ1 t2 /λ1 t2 /λ2 
 (3.4)
 .. .. .. .. 
. . . . 
−tN −tN −tN
 
 
 1 − e λ1 1 − e λ1 −t 1 − e λ2 −tN 
1 − e λ1 −e 2λ
tN /λ1 tN /λ1 tN /λ2
Where β̂ is a column vector consisting of the four βi constants:

 
β1
β2 
β̂ =  
β3  (3.5)
β4
The optimal parameters of β̂ can then be estimate by the equation:
β̂ EST = (X T X)−1 X T Y M (3.6)

Finally the fitted yield value Ŷ (t) for maturity t can be obtained by inserting the
estimated values of βi and λi into Equation 3.1:
3.2 Principal Component Analysis

Principal component analysis (PCA) is a method that can be used to reduce the di-
mension of a data set and still retain most of the information. This is done by a linear
orthogonal transformation of a d-dimensional data set onto a k-dimensional linear
space (where k < d), such that the variance of the projected data is maximized.
The dimension of the transformed data are orthogonal, i.e., they are independent
and are mutually uncorrelated [23]. This technique can briefly be described in the
following steps;
i. Normalize and center the data to have mean zero (Let us call this data for X)
ii. Determine the corresponding covariance (or correlation) matrix of the data X
and extract eigenvectors and eigenvalues from the covariance matrix or correlation
matrix. Where Singular Vector Decomposition also can be used.
iii. Construct a projection matrix W consisting of the k selected eigenvectors that

correspond to the k largest eigenvalues.
iv. Use the projection matrix W to transformed the observed d-dimensional data
to the new k-dimensional feature subspace. A deeper theoretical explanation of the
steps is presented below.
Consider a data set with d features X 1 , X 2 , ..., X d . Each of the dimensions found
by PCA is a linear combination of the d features.The first principal component of a
set of features X 1 , X 2 , ..., X d is the normalized linear combination of the features
that has the largest variance. The normalized linear combination of the features can
be expressed as:
Z 1 = φ11 X 1 + φ21 X 2 + ... + φd1 X d (3.7)
Pp
Moreover, normalized can be expressed as j=1 φ2j1 = 1. The elements φ11 , ..., φd1
can be seen as the direction of the first principal component, which, together, make
up the principal component direction vector φ1 = φ11 , φ21 , ..., φTd1 [23].
Now, consider the n · d data set X, where X has been normalized and centered to
have mean zero. To calculate the first principal component, the linear combination
of the sample feature values is considered:
zi1 = φ11 xi1 + φ21 xi2 + ... + φp1 xip (3.8)
Pp
for i = 1,...,n that has largest sample variance, subject to the constraint j=1 φ2j1 =
1. This means, that the first principal component vectors solves the following opti-
mization problem
  2 
n p p
1 X X X
max  (φj1 xij )  subject to φ2j1 = 1

  (3.9)
{φ11 ,...,φp1 } n i=1 j=1 j=1
We refer to Z1 as the first principal component with realized values z11 , ..., zn1 .
The optimization problem can be solved by different methods, e.g., by finding the
eigenvectors and eigenvalues from the covariance matrix or correlation matrix of
X. The obtained eigenvectors and eigenvalues can be seen as the core of a PCA.
The eigenvectors are the principal components, which define the directions of the
new feature space. The eigenvalues represents the magnitude of the principal com-
ponents [23].
The second principal component is the linear combination of X 1 , X 2 , ..., X d that

has maximal variance among all linear combinations that are uncorrelated with Z1
[23]. The second principal component scores z12 , z22 , ..., zn2 take the form:
zi2 = φ12 xi1 + φ22 xi2 + ... + φd2 xid (3.10)
Where φ2 is the second principal component direction vector with elements φ12 , φ22 , ..., φd2 .
It turns out that constraining Z2 to be uncorrelated with Z1 is equivalent to con-
straining the direction φ2 to be orthogonal to the direction φ1 , and so on [23].
Principal components Z can be expressed as
Z = XΦ (3.11)
Where X is the normalized observed data with zero mean and Φ is the matrix
with the eigenvectors (principal components) of the sample covariance matrix of
X, i.e., the columns in the matrix φ are the principal component direction vectors
φ1 , φ2 , ..., φk [23]. We refer to the matrix φ as the projection matrix.
The inverted PCA representation can be used to reproduce the correlation struc-
ture of the original risk factors
X = ZW −1 (3.12)
Where W −1 is the inverse of the projection matrix W .
3.3 Cholesky Factorization

Every positive definite matrix ARn×n can be factored as
A = LLT (3.13)
where L is a lower triangular matrix with real and positive diagonal entries, and LT
denotes the conjugate transpose of L [24]. The elements of L can be determined
for k = 1, 2...n and for i = 1, 2...k − 1 in accordance with Datta [25] as:
v
u
u k−1
X
Lk,k = tAk,k − L2k,j
j=1
(3.14)
i−1
1 X
Lk,i = (Ak,i − Li,j Lk,j )
Lii j=1
The Cholesky decomposition is a commonly used mathematical tool in a Monte

Carlo simulation for systems with multiple correlated variables. The covariance
matrix is decomposed to give the lower-triangular L. Applying L to a vector of
uncorrelated samples u produces a sample vector Lu with the covariance proper-
ties of the system being modeled [24].
3.4 Forward Rate

The forward rate is a projection of future interest rates between the period t to T at
time s, where the continuously compounded forward rate are given by:
ln(p(s, T ) − ln(p(s, t))

F (s; t, T ) = − (3.15)
T −t
Where p(t, T ) is the price of a non-defaultable zero-coupon bond with maturity at

time T. This is a financial contract that, with certainty, pays its owner 1 at time T
and are given by the formula:
p(s, T ) = e−r(s,T )·(T −s) (3.16)
Where r(s, T ) is the interest rate between s and T at time s, if s = 0 we approxi-

mate in this paper that the (continuously compounded) interest rate is equal to the
yield.
3.5 Bachelier Pricing Model

In 1990, the French mathematician Louis Bachelier presented the normal model,
which is now one of many methods that can be used to price interest rate options
[26]. For simplicity, we will refer to the model as the Bachelier model. Unlike the
Black-76 Model, which probably is the most widely used option pricing method,
Bachelier’s model allows negative interest rates. However, it is known that caps
and floors are the interest rate options that are most likely to occur in the banking
book [3]. Caps and Floors are a series of caplets and floors. Hence, we start to
calculate them. Bachelier’s pricing formula of caplets follows:
√
Caplet = F (s; t, T ) − K · N (d) + σ · T − t · n(d) (3.17)
Where k is the strike rate, T is the option maturity date, N is the cumulative normal
distribution function, n is the normal density function, and d is given by:
F (s; t, T ) − K
d= √ (3.18)
σ· T
As mention before, caplets are used to price caps, which are instruments containing
a series of caplets. Hence, the pricing formula for a cap is given by:
N
X
Cap = capleti · τti−1,ti · DF (ti ) (3.19)
i
Each caplet is weighted by the year-fraction τti−1,ti and the discount factor DF (Ti ).
Bachelier’s pricing formula of floorlets follows:
√
F loorlet = F (s; t, T ) − K · (−N (d)) + σ · T − t · n(d) (3.20)
Just like a cap, a floor can be priced as a discounted sum of floorlets:

N
X
F loor = F loorleti · τti−1,ti · DF (ti ) (3.21)
i
It can be seen that the valuation of caps and floors are quite similar. A cap contains
a series of caplets. Each caplet is a potential cash flow and can be seen as a call
option on a floating rate index level. Whereas, a floor contains a series of floorlets.
A floorlet can be seen as a put option on a floating rate index level [26].
Chapter 4
Model
In this chapter, we introduce notations and calculation that has been used to build
the models in this study. Several methods have been used in order to determine
different risks in different scenarios. The primary model is a model that follows
EBA’s guidelines on how to measure IRRBB. We have referred to this model as
the IRRBB model. In addition to the IRRBB model, three different models that
generate interest rates scenarios have been built. We will, in this paper, refer to
these models as the EBA model, Cholesky model, and PCA model.
4.1 Mapping of the Cash Flow Data

One essential part of the IRRBB calculation is cash flows. The timing, size, and of
course, the type of cash flows will have significant effects on the IRRBB. Which
cash flows should be used, and how should they be modeled? This section will
answer how these questions were handled in this paper.
All future notional repricing cash flows arising from interest rate sensitive assets, li-
abilities, and off-balance sheet items were aggregated and mapped onto 19 different
time buckets (indexed numerically by k). A notional repricing cash flow includes
any repayment of principal, any repricing of principal and any interest payment on
a tranche of principal that has not yet been repaid or repriced [3]. Cash Flows under
interest rate shock scenario i and currency c is denoted as CFi,c (k). Each bucket
is represented by a time period, and cash flows were mapped into the different time
buckets according to their repricing date. All cash flows in each time bucket were
assumed to take place at the midpoint of the buckets time interval. The 19 time
buckets, with corresponding time interval and midpoint, can be seen in Table 4.1.
The time bucket approach is commonly used and suggested by EBA [1]. An alter-
21
CHAPTER 4. MODEL 22
native approach is to measure each cash flow to its exact time, which would give a
more precise result but is hard in practice.
Table 4.1: Describes the different time buckets and the time period each bucket
represent.
Time bucket intervals
(M: months; Y: years)
CF CF
Short- O/N< t 1M< t 3M< tC F 6M< tC F 9M< tC F 1Y< t C F 1.5Y< tC F
Overnight
term 1M 3M 6M 9M 1Y 1.5Y 2Y
(0.0028Y)
rates (0.0417Y) (0.1667Y) (0.375Y) (0.625Y) (0.875Y) (1.25Y) (1.75Y)
2Y< tC F 3Y< t C F 4Y < t C F 5Y< t C F
Medium-term 6Y< t C F
3Y 4Y 5Y 6Y
rates 7Y (6.5Y)
(2.5Y) (3.5Y) (4.5Y) (5.5Y)
Long
7Y< t C F 8Y < t C F 9Y < t C F 10Y<t C F 15Y< tC F
term t C F > 20Y(25Y)
8Y (7.5Y) 9Y (8.5Y) 10Y(9.5Y) 15Y(12.5Y) 20Y(17.5Y)
rates
The type of a particular position decides how its cash flow should be slotted.
For fixed-rate positions, which generate cash flows that are certain till the point
of contractual maturity, should all coupon cash flows, and periodic or final princi-
pal be mapped into the time bucket closest to the contractual maturity. Examples
of such positions are fixed-rate loans without embedded prepayment options, term
deposits without redemption risk, and other amortizing products such as mortgage
loans.
Floating rate instruments are managed in a slightly different way since they gener-
ate cash flows that are not predictable past the next repricing date other than that the
present value would be reset to par. These instruments are assumed to reprice fully
at the first reset date. Hence, the entire principal amount is slotted into the bucket
in which that date falls, with no additional slotting of notional repricing cash flows
to latter time buckets.
4.2 Yield Curve Construction

Before any consideration of IRRBB could be made, an interest rate curve, more
commonly known as yield curve, needed to be built. A yield curve is a plot of
yields of bonds having equal credit quality but different maturity dates. This will
be the foundation of the entire model, especially for the gap risk calculations, since
these calculations assume that for each currency, all assets and liabilities are priced
upon the same yield curve.
In order to generate a yield curve, bond prices or equivalent instruments need to

be extracted from the market. In this paper, the yield curves for each currency
are based upon the market price of interest rate swaps, which is recommended by
CHAPTER 4. MODEL 23
BCBS [27]. However, one is not done there. Remember that the EBA framework
suggests an IRRBB model that is based upon a time bucket approach. Since the
maturity dates of the market’s bonds do not precisely match the time bucket’s mid-
points, a bootstrapping method was needed to generate an approximation of the
yield for each bucket.
In the model proposed in this paper, the Nielson-Siegel-Svensson (NSS) method

was used. NSS is a commonly used method to generate yield curves, and it can, for
instance, also be found in Hibiki and Iwakuma [14]. Remember from Section 3.1
that the NSS equation models the yield curve at a point in time as follows [20]:
−t −t
−t

1 − e λ1 1 − e λ1
Ŷ (t) = β1 + β2 + β3 − e λ1
t/λ1 t/λ1
−t
(4.1)
−t

1− e λ2
+β4 − e λ2
t/λ2
Where Ŷ (t) denotes the estimated zero rate for maturity t. The parameters β1 ,β2 ,β3 ,
β4 and λ1 , λ2 were calibrated with a least square method see Section 3.1. Once
these parameters have been estimated, one can use this equation to generate yields
for any maturity. In this study, yields for each specific scenario were created for
each time bucket, see Table 4.1, based upon each simulated market data point. Re-
member that in this paper, yields with a specific maturity approximate the interest
rate for instruments with the same maturity.
4.3 Interest Rate Models

Hypothetical interest rate scenarios can be generated with many different methods.
In this paper, the different methods that are presented can be divided into two differ-
ent main approaches, namely deterministic scenarios and stochastic scenarios. In
the deterministic approach, key scenarios are hypothetically assumed and created
by predefined stresses. In the stochastic approach, a large number of hypothetical
scenarios are randomly generated [10]. In this section, the deterministic method
that EBA [1] proposes will be introduced first. This paper has also implemented
two different stochastic approaches, which will also be presented in this section.
CHAPTER 4. MODEL 24
4.3.1 Interest Rate Restrictions

Before presenting the different models for generating the yield curve, it is of utmost
importance to be aware of the limitations EBA has on acceptable yield curves.
Earlier frameworks, such as BCBS [11], suggested that no negative interest rates
should be allowed. Since then, negative interest rates have become more prevalent
in many countries. Hence, the EBA has given a lower allowed limit which can be
described by the function:

−1%¸
 for t ∈ [0, 1]
YF loor (t) = −1%¸ + 0.05%¸ · (t − 1) for t ∈ (1, 20] (4.2)

0%¸ for t ∈ [21, 25]

In the yield curve generation models presented in this paper, this lower limit was
handle in two ways. For the deterministic EBA method, yields that fall below the
interest rate boundary were replaced with the value of the floor. For example, if the
1-year yield in one scenario was equal to -2 %, this value was replaced and set to
−1%. For the stochastic models yield curves that did not fulfill the requirement, set
by EBA [1], these curves were rejected and replaced by a new unrelated stochastic
simulation.
4.3.2 EBA Interest Rate Model

To evaluate how interest rate sensitive a financial institution is, EBA [1] suggests
that predefined stress tests should stress the yield curve for each currency c. The
six suggested interest rate shocks suggested by the EBA framework are:
(i) Parallel shock up

(ii) Parallel shock down
(iii) Short rates shock up
(iv) Short rates shock down
(v) Steepener shock (short rates down and long rates up)
(vi) Flattener shock (short rates up and long rates down)
In order to provide these stressed yield curves, EBA suggests a simple model where
a constant R̄x,c are generated from a predefined method see EBA [1] for calcula-
tions, where x is the type of stress, and c is for which currency the stress is applied
on, see Table 4.2 for currency depended constants.
CHAPTER 4. MODEL 25
Table 4.2: Describes the individual shock parameters for some currencies.
Currency SEK USD GBP DKK EUR JPY

R̄parallel,c 200 200 250 200 200 100
R̄short,c 300 300 300 250 250 100
R̄long,c 150 150 150 150 100 100
Parallel shock for currency c:
A constant parallel shock up or down across all time buckets a certain amount of
basis points based upon in which currency is to be stressed [1].
∆Rparallel,c (tk ) = ±R̄parallel,c (4.3)
Short rate shock for currency c:
Shock up or down that is greatest at the shortest tenor and diminishes toward zero
at the tenor of the longest point on the term structure [1].
−tk
∆Rshort,c (tk ) = ±R̄short,c · e 4 (4.4)
Rotational steepener shock for currency c:
Involving rotations to the term structure of the interest rates, whereby both the long
(up) and short (down) rates are shocked and the shift in interest rates at each tenor
midpoint is obtained by applying the following formula [1]:
−tk −tk
∆Rsteepener,c (tk ) = −0.65 · |R̄short,c · e 4 | + 0.8 · |R̄long,c · 1 − e 4 | (4.5)
Rotation flattener shock for currency c:
Involving rotations to the term structure of the interest rates, whereby both the long
(down) and short (up) rates are shocked and the shift in interest rates at each tenor
midpoint is obtained by applying the following formula [1]:
−tk −tk
∆Rf lattener,c (tk ) = 0.8 · |R̄short,c · e 4 | − 0.6 · |R̄long,c · 1 − e 4 | (4.6)
Where the resulting yield for scenario i, and currency c is described as

CHAPTER 4. MODEL 26
yi,c (tk ) = y0,c (tk ) + Ri,c (tk ) (4.7)
4.3.3 PCA Interest Rate Model

One common yield scenario-generating method is the Principal Component Analy-
sis (PCA) [10], which is described in Section 3.2. Several studies show that the first
three principal components factors can describe most variations in the yield curve.
These factors represent the shift, tilt and twist of the yield curve [10] [7]. This pa-
per is also using the first three principal components. Furthermore, the model can
be described in the following way;
i) Semiannual historical data of interest rate changes were collected. The PCA
was implied on the data set, and the first three principal components were extracted.
ii) The PCA based Monte Carlo simulation was performed by drawing indepen-
dent random shocks from each principal component distribution. We know from
Section 3.2 that the yield curve can be expressed by inverting the PCA representa-
tion, i.e.:
X = Z · W −1 (4.8)
Here, the coefficients W −1 give the sensitivity of interest rate changes along the
yield curve with respect to each Z.
iii) By shocking the right hand side of the equation, new theoretical yield curve
changes was obtained. The shock vector U that represents the simulated change in
each scenario can be expressed as:
√ p p p
U = W −1 Λη = η1 λ1 W1 + η2 λ2 W2 + ... + ηk λk Wk (4.9)
Where Λ is the diagonal matrix of the corresponding eigenvalues, and η = η1 , η2 , ..., ηk

are the vectors of independent shocks. Usually, scenarios based on PCs are sim-
ulated through vectors of standard normal shocks η j ∼ N (0, Ik ), which also has
been used in this case [7].
iv) Finally, the simulated change for each scenario was applied upon the current
yield curve to generate the i:th scenario yi , i.e.:
yi = y0 + U (4.10)
CHAPTER 4. MODEL 27
This procedure was repeated K times until 1000 generated scenarios that fulfills
the EBA standards of a yield curve were created.
4.3.4 Cholesky Interest Rate Model

Cerrone et al. [2] used a Monte-Carlo approach in their research in order to gener-
ate simulated scenarios. They argue that their model was superior to other methods
suggested by other papers. However, in their simulation, they used some constrains
on the yield curve simulation that is no longer up to date. For instance, they did not
allow negative yields nor negative forward rates. In newer standards like EBA [1],
these constraints should not be applied. Furthermore, Cerrone et al. [2] only tested
their generated scenarios on an EVE gap risk model. To check if their conclusions
are still in effect, we tested their method on a more up-to-date IRRBB model with
current standards. The model can be explained in seven numbers of steps:
i) Estimate the means µi of the semiannual changes in the key rates and their
variance-covariance matrix Ω.
ii) Generate a random number ui (i = 1, . . . 14) ranging from zero to one at each
node of our key rates term structure.
iii) Convert each ui into a value zi (i = 1, . . . 14) distributed according to a stan-

dard normal. In symbols:
zi = F −1 (ui ) (4.11)
Where F −1 is the inverse of the distribution function of a standard normal.
iv) Use the algorithm of Cholesky 3.3 in order to decompose the matrix Ω in two
matrices L and LT such that:
Ω = L · LT (4.12)
Remember from section 3.3 that L is a lower-triangular matrix containing the

variance-covariance matrix relationship between the key rates.
v) Calculate the vector x, whose elements are the simulated semiannual changes
in the key rates, through the following formula:
CHAPTER 4. MODEL 28
x=L·z+µ (4.13)
Where z is the vector of the values calculated in step iii) and µ is the vector of the
14 means of the distributions of the key rates annual changes calculated in step i).
x represents a simulated change of the yield curve.
vi) The simulate change vector x is then applied to the current yield curve to get
the simulated scenario i
yi = y0 + x (4.14)
vii) The steps ii)-vi) were then repeated until we reached a number of 1000 scenar-
ios [2].
4.4 Refinancing Risk

After both stressed and non-stressed yield curves had been generated, the second
part of behavioral risk was modeled, namely refinancing risk. This since the gap
risk is highly dependent on each CFi,c (k) size, which will differ for each yield and
currency for the instruments that are behavioral sensitive.
Several factors are important determinants of financial institutions estimations of

how different interest rate scenarios will affect the average prepayment and with-
drawal speed. A financial institution must assess the expected average prepayment
and withdrawal speed under each scenario [27]. It is also essential for a risk man-
ager to have a deep understanding of the balance sheet in order to identify which
products are sensitive to refinancing risk. This since it is far from every product
that the customer has the embedded option to prepay or withdrawal capital.
To calculate how the different cash flows change during different stress tests, a
function for the behavioral sensitive instruments based on BCBS [27] original idea
with smaller alterations was created.
p
First, the conditional prepayment rate, CP Ri,c , was calculated with the formula
p p
CP Ri,c = min(1, γi · CP R0,c ) (4.15)
p
Where CP R0,c denotes the average historical prepayment rate for each prepayment
CHAPTER 4. MODEL 29
sensitive instrument, and γi denotes the multiplier applied for scenario i. To sim-
plify the model, assumptions for all prepayments sensitive instruments were made.
p
The original CP R0,c was set, in accordance with BCBS [28] recommendations, to
0.01. For the EBA yield curves scenarios, the γi were defined in accordance with
BCBS [28] and can bee seen in Table 4.3.
Table 4.3: Describes scenario multipliers for each EBA scenario.
Scenario number (i) Interest rate shock scenarios γi (scenario multiplier)

1 Parallel up 0.8
2 Parallel down 1.2
3 Short rate up 0.8
4 Short rate down 1.2
5 Steepener 0.8
6 Flattener 1.2
For the Monte-Carlo simulations, a similar approach was used to include con-
sistency between the models as far as possible. It was assumed that γi was equal to
0.8 if the short term rates were increasing and that γi was equal to 1.2 if the short
term rates were decreasing.
When the CPRs for each scenario had been calculated, the new scenario-depended
cash flows were calculated. It was assumed that all prepayments were made in each
time bucket and went into effect instantly. In accordance with the formula:
p s p
CFi,c (k) = CFi,c (k) · CP Ri,c (4.16)
s
CFi,c denotes the originally scheduled interest rate and principal payment. These
new scenario sensitive cash flows were used in the gap risk calculation, see section
4.5. Hence, this option risk is not calculated in an isolated fashion as the other risks
are.
4.5 Gap Risk

Gap risk is the most comprehensive risk and has, in this study, been calculated in
terms of EVE and NII. Both of these risk measures will be presented separately
below.
CHAPTER 4. MODEL 30
4.5.1 EVE
Under each scenario i, all notional repricing cash flows are slotted into the respec-
tive time bucket k (1, 2, . . . , 19). Remember that for each scenario, the cash flows
differ due to refinancing risk. Within a given time bucket k, all positive and nega-
tive notional repricing cash flows are netted to form a single cash flow. Hence, the
offsetting part of each cash flow within the time bucket k was removed from the
calculation. By following this process across all the time buckets, a set of netted
p
notional repricing cash flows denoted as CFi,c (k) was obtained [1]. Netted cash
flows in each time bucket k were weighted by a continuously compounded discount
factor computed as:
DFi,c (tk ) = e−ri,c (tk )·tk (4.17)
Which leads to that for each scenario i and currency c the estimated EV E is de-
scribed as:
19
X p
EV Ei,c = CFi,c (k) · DFi,c (tk ) (4.18)
k=1
Finally, the changes in EVE in currency c under the interest rate scenario i, denoted
GAP
as ∆EV Ei,c , is obtained by subtracting the original EV E, denoted as EV E0,c ,
from the EV E for scenario i.
N
X
∆EV EiGAP = EV Ei,c − EV E0,c (4.19)
c=1
4.5.2 NII
The risk of loss in earnings in this model will be measured with the use of net in-
terest income. More precisely explained, the changes in NII depending on shifts
in the yield curve. Even though earlier framework such as BCBS [3] only required
risk managers to measure their NII risk for parallel shifts of the yield curve, EBA
[1] has implemented stricter rules for this measure. EBA [1] requires risk managers
to evaluate the NII risk fully, i.e., evaluate the change of NII for all six stress tests.
In this section, a method to measure this risk in accordance with the demands of
EBA [1] is presented.
First and foremost, a more precise explanation of NII and which instruments are
affected by this measure is necessary. NII risk is measured in the relative change
CHAPTER 4. MODEL 31
of NII in accordance with 4.22. Since NII gap is a measure of the difference in
interest earnings, the net worth of asset and liabilities are not of interest for these
calculations but instead the interest payments of these instruments. This implies
that fixed interest rate instruments are not of interest since these interest incomes
will offset each other when the interest rate changes. Hence, the only relevant part
of fixed interest rate instruments is the change of prepayment rate, since this change
the amount of interest payment received.
However, instruments affected by a floating interest rate are of high importance in

these calculations. Floating interest rate instruments do not change continuously.
Instead, the interest rate for these changes occurs periodically at the instruments’
specific repricing date. The frequency of how often an instrument’s repricing date
occurs will, in this paper, be referred to as repricing frequency, which typically
takes place every month, every two months, etc. The effect of this is that a floating
interest rate instrument can be handled as a fixed interest rate instrument between
each of the repricing dates for that instrument. Furthermore, for these calculations,
it is important to remember that NII is calculated with a so-called constant balance
sheet, which means that maturing assets are instantly replaced with an instrument
with identical features as the instrument matures. Another important aspect to re-
member for these calculations is that the EBA approach requires risk managers to
slot institutions’ notional cash flows into time buckets in accordance with each cash
flows repricing date. This means that the interest rate income during the period t
to t + R, where R denotes the repricing frequency, will be assumed to act under
the period tk to Tj . Where tk is the midpoint of the bucket closest to t and Tj is
the midpoint of the bucket closest to t + R.
The first step to calculate the NII gap risk is to aggregate all notional cash flows
with the same repricing bucket, repricing frequency, and currency.
The second step is to evaluate the interest earnings of the notional cash flow with
repricing frequency R under scenario i, between tk and Tj , this is done with the
formula:
N IIi,c,k = CFi,c,k, · F (0; tk , Tj ) (4.20)
Where i refers to stress scenario, c currency, k is the index of the bucket closes to t,
and j is the index of time bucket closest to t + R. The effective interest rate under
the period tk to Tj is given by the forward rate described in section 3.4:
After a cash flow of interest payment between the period tk to Tj had been made,
CHAPTER 4. MODEL 32
the notional cash flow was added to the aggregated notional cash flow with the same
repricing frequency in the time bucket of Tj . This procedure was then repeated for
all time buckets and repricing frequencies until the NII for the next three years had
been calculated.
To calculate the total interest earnings for scenario i, the NII for all repricing fre-
quencies was summed with the formula:
C X
X 19
N IIi,c = N IIi,c,k (4.21)
c=1 k=1
The final step to receive the resulting NII exposure was to calculate the difference
between the interest earnings in scenario i and the original scenario.
GAP
∆N IIi,c = N IIi,c − N II0,c (4.22)
4.6 Basis Risk

The next necessary step to measure a financial institution’s IRRBB is to calculate
the basis risk in the banking book. Remember that basis risk refers to the impact
of an imperfect correlation between indices underlying asset and liability rates. in
other words, the possible impact of relative changes in interest rates on interest rate
sensitive instruments that have similar tenors but are priced using different interest
rate indices [1]. In this paper, an approach suggested by BCBS [28] was used to
determine the basis risk. In this method, basis risk is computed by first aggregating
the net notional repricing cash flows for a specific time period and for each refer-
ence rate. In this paper, the used time period is three years. After that, the net cash
flows exposed to basis risk between each pair of reference rates are computed as
the minimum of the associated aggregated net notional repricing cash flows. This
procedure is repeated for all combinations of pairs of reference rates in currency c.
This can be expressed by the following equation:
RR X
X Y
∗ ∗
∆N IIcrb = ((min(max(CF0,c,rrX
; 0); |min(CF0,c,rrY
; 0)|)
Y =2 X=1
(4.23)
∗ ∗ C
+min(|min(CF0,c,rrX
; 0)|; max(CF0,c,rrY
; 0))) · HrrX ,rrY
)
∗
Where CF0,c,rr is the aggregated notional cash flows during the chosen time pe-
riod for the specific reference rate rr. RR is the number of reference rates which
C
the bank is exposed to in currency C. Hrrx,rry is the historical basis risk shock
CHAPTER 4. MODEL 33
between two reference rates and should be calculated by taking the 99th percentile
of the three-month moving average of the rate differentials between the pair of the
three-month reference rates by data from the latest ten year period. If no such data
is available for the institution, 50 basis points can be used instead. The latter was
used in this research.
4.7 Option Risk

As mention in section 2.1.3, the behavioral risk can be divided into two parts; refi-
nancing risk and the risk that occurs from embedded or explicit automatic options,
namely option risk. This subsection will handle the latter. Remember that this risk
arises when an institution or its customer can alter the level and timing of their
cash flows through different financial derivatives. There are several interest rate
sensitive instruments obtaining such features, where caps and floors are the most
important ones that are likely to occur in the banking book [3]. This paper will
mainly handle these two.
There are many different methods for calculating the automatic interest rate op-
tions risk, and EBA leaves it up to the institutions to choose which method to use.
This paper will use an approach suggested by BCBS [3], which is applied to sold
automatic interest rate options and bought automatic options. The method can be
described in the following steps:
o
The first step was to calculate the value change, denoted ∆F V AOi,c . This was
done by taking the value of the sold automatic options under a stress scenario, mi-
nus the value of the sold option at the valuation date. In other words, calculate the
value of the sold automatic option o in currency c , for each interest rate scenario
i. Subtract it with the value of the sold option, given the yield curve in currency c
at the valuation date. In addition to the interest rate scenarios i, a relative increase
in the implicit volatility of 25% is applied for each scenario.
The same procedure was repeated for the bought automatic interest rate option q,
q
i.e., the change in value, ∆F V AOi,c , of the option between interest rate shock sce-
nario i and the current yield curve, combined with a relative increase in the implicit
volatility of 25% was calculated.
The last step was to calculate the total measure for automatic interest rate option
risk under interest rate shock scenario i in currency c. This was done by the fol-
lowing equation:
CHAPTER 4. MODEL 34
nc
X
o q
KAOi,c = = ∆F V AOi,c − ∆F V AOi,c (4.24)
o
Where nc and mc is the number of sold respectively bought options in currency c.
The given approach requires a method to determine the value of the interest rate
options. Many different methods can be used, but it is important to keep in mind
that EBA’s framework state that negative interest rates should not be neglected. In
this paper, the Bachelier model, which was presented in section 3.5, was used. One
of the advantages of this model is that it allows negative interest rates.
Chapter 5
Results and Discussion
In this chapter, the result of this paper will be presented with the aid of graphs
and tables. Differences and similarities between the models will be identified, and
underlying factors for this will be addressed. The results will be analyzed and
discussed on an ongoing basis. Furthermore, a discussion about the IRRBB model
will finish the chapter.
5.1 Data
In Figure 5.1, the cash flows generated from the fictional banking book are shown.
The fictional banking book is a made-up banking book, which is based on studies
on how typical banking books look like. Also, real banking books have been used
as inspiration to ensure that the made-up banking book is realistic. The fictional
banking book generates cash flows in three different currencies, Swedish krona
(SEK), Euro (EUR), and Danish krone (DKK). All cash flows have been converted
to SEK to make the analysis suitable. It is assumed that the cash flows only change
due to varying prepayment speeds under different scenarios, beyond that, it stays the
same. Remember that it is common for banks to borrow short and lend long, which
generates a slight overweight of liabilities in the short term buckets and assets in
the long time buckets. Moreover, the made-up banking book has more liabilities
with floating rate than assets with floating rate. If positions have fixed or floating
rate has an extra big impact on ∆N II.
35
CHAPTER 5. RESULTS AND DISCUSSION 36
Figure 5.1: Cash flows in SEK, EUR and DKK converted to SEK. Note the scale
difference for the different currencies.
Figure 5.2 shows the original yield curves for SEK, EUR, and DKK. The curves
are based on data retrieved on March 1, 2020, from Thomas Reuter Eikon. All
historical yield curve data used in the model building are taken from Thomas Reuter
Eikon.
Figure 5.2: Original yield curves for SEK, EUR and DKK.
5.2 Generated Yield Curves

Figure 5.3 shows the original yield curve, together with EBA’s six interest rates
scenarios for each of the currencies SEK, EUR, and DKK. One advantage while
analyzing the EBA-scenarios is that it only constitutes of six different scenarios.
Thanks to this, all of the scenarios can be examined, and it gives risk managers a
clear picture of how various interest rate changes would affect their banking book.
Before looking into the different scenarios, it is essential to remember that the EBA
framework suggests a lower interest rate boundary that starts on −1% and increases
by five basis points per year until it reaches 0%, see Section 4.3.1. This interest rate
floor had a significant effect on some of the yield curve scenarios. It can be seen
in Figure 5.3 that it was only two scenarios that were not affected by the constrain,
namely parallel up and short rate up. For the remaining four scenarios, there is a
varying amount of change that occurs due to this lower bound, which, in the end,
had an impact on the calculated risk.
Figure 5.3: Yield curves in SEK, EUR and DKK generated by the EBA model.
The parallel down scenario was supposed to be a parallel shift of the yield curve
with the size of 200 basis points. However, due to the low-interest environment,
the parallel down scenario was equal to the lower bound for all currencies. Which
resulted in that the parallel down scenario was the same for all of the different cur-
rencies, even though the original yield curves did not have the same shape and rates.
In other words, this scenario was not adapted for the specific currencies which it
was supposed to be. Furthermore, it is interesting to have a look at the discounting
factor, which for this scenario is greater than 1 for all maturities 20 years forward.
This goes against the concept known as "The time value of money" which says that
money available at the present time is worth more than the same amount in the fu-
ture [29].
Furthermore, due to the lower bound combined with the low original yield, the
short rate down and the parallel down scenarios are very similar to each other.
This since both of them are on the lower boundary for most of the maturities. Also,
the steeper shock rate is identical to these two scenarios for, approximately, the first
1.25 years. If an institution is calculating the ∆N II for the first year, the result for
these three scenarios would be the same. In contrast, earlier frameworks did not
allow negative interest rates, which would result in even more similar scenarios.
However, since many countries nowadays are having interest rates close to or even
below zero, the new framework allows negative rates.
Since EBA’s method only generates six different scenarios, it is, from a risk man-
ager’s perspective, risky to have several scenarios that are almost the same. This
since only a few neuances of the risk can be identified and there could be some
curves that provides significant risk that are not properly identified.
It is not only the fact that several of the scenario curves are more similar to each
other than what they would have been without the floor rate, but they are also los-
ing their supposed shape. For instance, for the flattener and steepener stresses. The
original yield curves for all the currencies are quite flat, which makes the flattener
and steepener scenario questionable. The purpose of these two scenarios is to ro-
tate the yield curve. As one can understand of their respective name, the flattener
case is supposed to generate a yield curve that is flatter than the original yield by
increasing the short rates and decreasing the long rates and vice versa for the steep-
ener case. The problem is that this scenario does not make the investigated yield
curves flatter. Instead, it makes them inverted, i.e., downward sloping. In this case,
the flattener starts with a positive rate that later becomes negative until it reaches
the lower limit, where this limit forces the interest rate to increase again. The re-
sulting shape of the stressed curve can be said to be unreasonable or at least very
rare. The same reasoning applies to the steepener scenario, but in this case, the
original idea is to create a scenario that is steeper than the original curve. Due to
the lower limit, the interest rate during the first years is only increasing with five
basis points per year. Which is less or almost equal to the original curve and for
later term buckets have a significant slope upwards. This can be deemed, at least
in theory, as an unrealistic shape of the curve.
Positive shocks do not have the same issue as the negative shocks since there is
no upper limit. One can see that the shape of the parallel up and short rate up
curves differ a lot, while the parallel down and short rate down are quite similar.
However, as mentioned earlier, there is a clear similarity between the short rate up
case and the flattener case. Furthermore, in general, the shocks on the short interest
rates are relatively large, especially for the positive shocks. For the short rate up
scenario, the stress generates short rates that are above the maximum short rate that
has occurred during the past ten years. On the other hand, if we go further back
than ten years, short rates have been way above the stressed level. For instance, in
September 1992, Sweden faced a 500 % rate. However, at present, EBA’s shocks
on the short rate can be seen as unrealistic or at least highly conservative for the
analyzed currencies.
Another disadvantage with the EBA-scenarios is that it does not generate any yield
curve that is of the more common shape, i.e, increasing and concave [21]. This
is the most prevalent shape, and it can be deemed reasonable to have at least one
scenario that gives this shape of the curve.
(a) Cholesky generated SEK yields (b) PCA generated SEK yields
(c) Cholesky generated EUR yields (d) PCA generated EUR yields
(e) Cholesky generated DKK yields (f) PCA generated DKK yields
Figure 5.4: Describes the generated yield curves of the Cholesky and PCA models
Figure 5.4 illustrates the simulated yield curve scenarios produced by the PCA
and Cholesky model. If examined carefully, one can see that the models produced
yield curves with different sizes and shapes, as suggested by the EBA [1]. Even
though there is a clear shape shift, the general shapes of the yield curves have a
closer resemblance with the original yield in the original scenario compared to the
scenarios the EBA model produced. Furthermore, in comparison, one can see that
there is a smaller variance on the shorter term buckets in the stochastic approaches
than in the EBA model. This depends on that the EBA approach has a stiffer way
of producing its yield curve with more predefined stresses on all buckets. In a
low-interest environment, this creates greater relative changes to the yield curves,
especially in the shorter term buckets. The PCA and Cholesky models show small
variance in the shorter buckets. As mentioned before, there has historically been
times where the short rate has been significantly higher than now. However, from a
risk manager’s perspective, it is reasonable to assume that on a short time horizon,
smaller changes in the short term rate are more reasonable. If one compares the
yields for the long term buckets in the EBA and Stochastic models, there is more of
a similarity between these. All three models create an upper yield of around 2% for
all currencies in the last bucket. However, there is a discrepancy between the PCA
and Cholesky models in the lower bounds of the long term buckets. One can see
that the Cholesky model created scenarios that are closely aligned, with the lower
limit suggested by the EBA, see Section 4.3.1, while the PCA model creates no
curves that are on the lower boundary in the long term buckets. For example, one
can see in Figure 5.5 that the lowest yield the PCA model generates in the 25-year
bucket is around 0.5 % but that the lower bound allows for the yield to go down to
zero percent. Even though by looking at the yield curves can give the perspective of
a big difference between the Cholesky and PCA model, it has a small impact on the
approximated IRRBB since it is unusual for financial institutions to have projected
cash flows so far into the future.
(a) Maximum/minimum yield for the (b) Maximum/minimum yield for the PCA
Cholesky model model
Figure 5.5: Describes the maximum/minimum yield generated for each bucket by
the PCA and Cholesky model.
By analyzing Figure 5.5 one can see that there is more discrepancy between the
PCA and Cholesky model then only in the lower bound of the long term buckets.
There is a significantly wider path that the Cholesky curves use with both higher
and lower rates for both short and long term buckets. It is reasonable to assume
from this that the Cholesky model captures a wider range of possible yield curves
and is less likely to underestimate potential scenarios. Further analysis of the PCA
graph in Figure 5.5 shows that the lower and upper bound of the yield curves have,
in a more visible way, inherit the original shape of the yield curve compared to the
Cholesky model.
Furthermore, it is also worth mentioning that the interest rate constrain is not imple-
mented in the same way for the EBA model as for the other two models. Remember
that, for the EBA model, in cases where the stressed rate went under the floor rate,
the stress rate was automatically set to be equal to the floor rate. For the Cholesky
and PCA model, whole curves were rejected if they went below the constrain. This
resulted in approximately 800 scenarios were rejected for the PCA- model, and 220
for the Cholesky model. This results, especially for the PCA model, in an overrep-
resentation of positive changes in yield curve scenarios. If the same approach had
been used for the EBA model, only two scenarios would not been rejected. It is
hard to say which approach is the best. The method used in EBA gives rise to un-
realistic yield curve shapes while the method used for the other two deficiencies in
the generation of different yield curve shapes, especially the PCA model. One can
conclude that the floor rate constrain makes it challenging to generate the perfect
setup of scenarios. We are not saying that the floor rate in isolation is the origin of
the problem, rather the relationship between the interest rate shocks and the lower
limit. They are not adapted to each other and therefore give unrealistic curves. To
conclude, none or at least only a few, of the EBA-scenarios is seen to be as realistic
scenarios.
5.3 Gap Risk

The timing of cash flows is essential when calculating interest rate risk and espe-
cially gap risk, this since it affects both the EVE and NII. In the EVE case, the
discounting factor has a central role since it operates over the whole timeline, and
therefore, cash flows that occur later in the future are affected more than nearby
cash flows. Hence, institutions EVE can simply change just by changing the timing
of the cash flows. For NII, the cash flows are not discounted, but the timing is still
an essential factor. In this paper, the NII is based on the cash flows in the nearest
three years, thus the size and type of cash flows within this time-period matters.
This implies that institutions NII also can be changed by changing the timing of
the cash flows. It is not only the timing that affects these measures, but the yield
curves also have a big impact. Hence, the chosen method plays an important part.
The effects on the gap risk that occurs from different interest rate scenarios will be
presented in this section. When analyzing the different scenarios, it is important to
remember that the result is highly dependent on the banking book. Timing, size,
and type of cash flows are different for every banking book. Hence, the same inter-
est scenarios will not give the same result for different banking books. As mention
before, this paper takes the bank’s balance sheet as a given, and study which inter-
est rate environments would generate the largest risk. The analyzed banking book
characterizes of slightly more negative cash flows in the first two years and more
positive cash flows after that, see Figure 5.1.
The EBA model only generates six scenarios, as mentioned earlier, this has both
advantages and disadvantages, which will be further analyzed later in this section.
One of these advantages is that this allows each scenario to be analyzed one by one,
i.e., no black box calculations are needed. This is useful since some generated yield
curves can be seen to have a low probability of occurring. In Table 5.1 the resulting
gap risk for the EBA scenarios in both EVE and NII can be seen. Remember that
in our model, there is a built-in prepayment/withdrawal risk in the EVE gap risk
parameters. For the exact sum of the prepayment for each scenario, see Table 5.4.
Table 5.1: Gap risk measures given in million SEK for each of the six EBA scenar-
ios.
Gaprisk Parallel up Parallel down Short rate up Short rate down Flattener Steepener
EVE: -172 38 -38 26 30 -92

NII: 34 -17 33 -17 21 -13
For the parallel down scenario, the interest rate is on the interest rate floor for all
the currencies. This also implies that the interest rate is equal to the lower bound,
see Section 4.2, during the whole period. The negative rate also indicates a dis-
counting factor that is greater than one, meaning that the discounted cash flows
are greater than their nominal value. Since the discounting factor is dependent on
the time as well as the interest rate, the discounting effect is more significant on
cash flows with later maturities. Positive cash flows usually over represent these
late cash flows, which is one of the main reasons for the positive ∆EV E in this
scenario.
For the parallel up scenario, the generated yield curves were almost equal to 2%
over the whole period. Remember that the discounting effect has a more signifi-
cant impact as further away the cash flow occurs, which, in our case, means that
positive cash flows are discounted by a higher factor than the negative cash flows.
This results in the lowest ∆EV E.
The short down and the parallel down scenario are very similar to each other since
both of them are on the lower boundary for most of the maturities. The short rate
down scenario generates a slightly lower ∆EV E than parallel down since this yield
curve is a bit over the floor after six years. The cash flows after this time are there-
fore not getting as much weight as the ones in the parallel down case.
As mentioned in chapter 5.2, the shape of the flattener shock results in a s-like
curve, going from a positive interest rate to a negative interest rate. This implies
that the early cash flows (mainly negative) are discounted by a factor less than one
and the later cash flow (mainly positive) with a factor greater than 1. Which results
in the second-highest ∆EV E. The steepener shock works oppositely, and results
in the second smallest ∆EV E. Furthermore, it can be seen that the short rate up
and the flattener case are giving completely different result even though the shape
of the curves are similar. In particular, the flattener case is almost like a parallel
downshift of the short rate up curve. This means that the parallel up scenario has a
smaller discounting factor than the flattener case at all maturities, but the difference
between them is growing as the time increase. If you compare the scenarios, the
difference between the discounted assets is greater than the difference between the
discounted liabilities, which has a big impact on the final result.
If we have a look at the ∆N II measure, it can be seen that this risk has a lower
amplitude and less volatile behavior compare to the economic value risk. This is
because the NII calculations have a much shorter time-horizon and, in most cases,
the yield curve over this period has a lower amplitude and is less volatile. Further-
more, the analyzed banking book has more floating assets than floating liabilities
in the earlier time buckets. Hence, the scenarios with positive shocks on the short-
term rates resulted in more positive NII:s and vice versa. Furthermore, this creates
a problem when measuring NII risk. This since EBA suggests a lower cap on in-
terest rates in their directives, and multiple curves within three years are on this
lower boundary. This results in that multiple curves are practically the same, for
example, parallel down and short rate down are both on the lower boundary dur-
ing the whole period. Furthermore, this also limits the model’s maximum possible
risk in this measure since no lower interests are possible. Therefore, there is no real
simulation of this measure but rather just a statement of the worst-case scenario for
this particular banking book.
Table 5.2: Describes gap risk measures given in million SEK for the three models.
Gap Risk: EBA PCA Cholesky
EVE:
M ax 38 4 21
M in -172 -105 -110
F −1 (0.05) - -68 -68
F −1 (0.95) - -8 4
ST DV - 16 22
NII:
M ax 34 17 19
M in -17 -10 -12
F −1 (0.05) - -5 -7
F −1 (0.95) - 11 12
ST DV - 5 6
In Table 5.2, it can be seen that the EBA model generates the biggest value
of ∆EV EM ax and the smallest ∆EV EM in , while the PCA model generates the
smallest ∆EV EM ax and smallest ∆EV Emin . In other words, the spread be-
tween max and min is highest for the EBA model and lowest for the PCA model.
∆EV EM in , for the Cholesky model, is quite close to ∆EV EM in for the PCA
model, while its ∆EV EM ax, is precisely between the corresponding measures
generated from the EBA and PCA model.
By analyzing the standard deviation from Table 5.2 in combination with Figure
5.6, one can conclude that the Cholesky model creates a broader span of risk mea-
sures compared to the PCA model. The Cholesky model shows less of a clustering
behavior and, at some level, a more evident linear behavior between NII and EVE.
From a risk manager’s perspective, it is alluding to assume that this would create
a wider scope of scenarios and, therefore, capture the risk in a better way. This
conclusion may, in some scenarios, be valid, but in this case, we can clearly see
by looking at the F −1 (0.05), in Table 5.2, that this is not the case. The different
models show a similar amount of risk both in the earlier mentioned max measure
and quantiles, at least on the downside of the measures. The upside of the models
shows a lesser resemblance and values that are much closer to zero. This may, for
a different banking book, give rise to more notable variation between the risk mod-
els.
(a) Cholesky gap risk measures (b) PCA gap risk measures
Figure 5.6: Describes the gap risk measure for both the PCA and Cholesky model.
From Figure 5.6, it is easy to conclude that the Cholesky and PCA model gen-
erates more negative than positive ∆EV E since several scenarios with a general
increase in interest rates are obtained. Since the inventive bank borrows short and
lends long, a general increase in interest rates returns a decrease in EVE and vice
versa. Since decreasing interest rate induces increasing EVE, the lower interest rate
limit can be seen as a limitation on the rise in EVE. In addition, many curves were
rejected since they reached the lower limit. Which, in the end, gives a majority
of scenarios with increasing interest rates. This becomes a trade-off problem in a
low-interest environment. Where the alternative solution would be to replace "too
low yields" with the lowest accepted yield, similar to how this paper solves this
problem for the EBA model, instead of the approach of rejecting yield curves. In
a high volume scenario generating approach, this would create too many similari-
ties and duplicates of lower bound curves. This could lead to that the purpose of
the models is lost, and the calculations of risk become questionable. Generating
many different forms of the yield curve is one of the main strengths of the stochastic
models compared to the EBA model. It is, therefore, a disadvantage that so many
curves are rejected due to the lower boundary. On the other hand, in this study, it
is not the potential worst-case scenarios that are dismissed, since decreasing rates
implies, in our case, an increase in EVE.
For all models, the absolute value of ∆EV EM in is considerably greater than ∆EV EM ax ,
this since positive changes are weighted by a factor of 50% when aggregating
∆EV E for the different currencies. This penalty factor can have a big impact
on institutions’ capital requirements. For instance, let us say that a bank has many
negative cash flows in SEK and many positive cash flows in EUR. Suppose that
∆EV ESEK = −100 and ∆EV EEU R = 100. When aggregating the changes
in EVE, the positive ∆EV E is weighted by a "penalty factor" of 50%, that is,
∆EV ET otal = 0.5 · 100 − 100 = −50. Now, let us say that the bank only has
assets and liabilities in SEK and that ∆EV ESEK = −30. This means that the
bank’s interest rate risk measured in economic value is considered to be bigger in
the first scenario, even though the changes in EVE between SEK and EUR offset
each other to 0 (without the penalty factor) and the ∆EV E in the second sce-
nario is negative. Of course, exchange exposure may increase the risk, but this
risk should be captured in other ways. This paper finds it unreasonable to have the
same factor between all currencies, and that a penalty factor of 0.5 is unfair. In-
stead, a correlation between currencies could be used to get a more fair aggregation
All of the models have their pros and cons, and it is hard to determine which one
is the best. It has been mentioned before that the EBA model does not contribute
with any reasonable yield curves, which is in line with what Letizia [15] presents.
He means that the parallel shift of the yield curve is unrealistic and objectively
incapable to deliver a meaningful appraisal of IRRBB. It is also this scenario that
generates the highest risk in our analysis. But does it mean that the EBA model ends
up with a worse risk calculation? No, that does not have to be the case. For this
banking book, the parallel up scenario implicates the highest risk for the economic
value measure. The reason why this stress rate is considered not to be realistic is
primarily due to the high increase in the short rates. Let us say that the stress on the
short rate was not that big, the ∆EV E, in this case, would be even smaller. In other
words, the ∆EV E, corresponding to the greatest risk, gives a reasonable result.
On the other hand, it is not sure that the parallel up scenario is the worst case for all
financial institutions. Several of the other EBA-scenarios would overestimate the
IRRBB, and in these cases, one of the other two models would possibly give more
accurate results.
Another reflection that can be made is that the calculated risk would be completely
different if the banking book contained larger positions in EUR instead of SEK. In
a case like that, the difference between the calculated risk generated from the EBA
model and the other two models would be even bigger. This because the Cholesky
and PCA model generates curves with significantly lower yields. The EBA stress
scenarios constitute interest rates that are higher than what the maximum yields
have been in the latest years. But, in the past ten years, the yields have been over
2.5 %, which is the highest yield generated from any of the models. The yields
have been even higher if we go further back than ten years. On the other hand,
the annual interest rate changes during the past ten years have not been close to
the same size as the EBA-scenarios. In general, historical interest rate changes are
more homogeneous than historical interest rate levels. The overall level fluctuates a
lot over time, and so the interest rate levels. Hence it is more reasonable to analyze
historical changes and use this as input data to models [30]. Based on this, it can
be argued that the EBA model are more conservative risk tolerance and are more
likely to overestimate the risk. In contrast, the Cholesky and PCA model generates
scenarios that are more likely to occur based on the behavior for the past years.
However, it is impossible to know to what degree historical data will explain the
future, and which data gives the best approximations. We assume that the infor-
mation in the samples is representative of future values and that other probability
beliefs are not relevant.
5.4 Basis Risk

The second driver of risk that this paper modeled is the so-called basis risk. Re-
member that the method used is a non-yield sensitive method that measures the risk
of an imperfect correlation between different underlying in each currency. Since
the method is non-yield sensitive, there are no measures that are unique for the dif-
ferent yield generator methods nor for each specific scenario. Instead, there is a
constant measure that should be added together with the more varying risks, i.e.,
gap and option risk. Remember that this measure could be seen as a complement
to the gap risk calculation, where one assumes that all assets within one currency
are based upon the same underlying.
Table 5.3: Describes the basis risk in each currency given in million SEK.
Basis Risk SEK EUR DKK
NII:
Total 20 0 10
In Table 5.3, we can see that we have the highest basis risk in SEK, which is rea-
sonable since we have the most floating assets and liabilities in this currency. The
reason why the basis risk in EUR is equal to zero is due to that there is only one
type of instrument used in this currency, both for assets and liabilities. These two
also have the same underlying. Furthermore, it is essential to understand that this
measure is supposed to be seen as the worst-case scenario within each currency.
Hence, it is more than probable that the actual risk is far less than the presented
number. A risk manager needs to analyze the institution’s own portfolio in an at-
tempt to identify how severe the basis risk actually is. For example, if an institution
has a high number of instruments with varying underlying assets, it is unreason-
able to assume that all the different underlying should differ from each other with
at least 50 basis points. A solution to this might be for the risk manager to reduce
C
the stress parameter Hrr X ,rrY
, see Section 4.6, and by that reduce an unrealistic
basis risk to a more probable value.
5.5 Behavioral Risk

Table 5.4 shows the option and refinancing risk, which together constitute the be-
havioral risk. Both of these two risks go under behavioral risk, but they are calcu-
lated differently and do not act in the same way. The refinancing risk is integrated
into the gap risk, hence, its contribution to the total risk calculation has already
been included. Anyway, this section breaks out the refinancing risk from the gap
risk in order to give a more unobstructed view of its impact and behavior.
Table 5.4: Behavioral risk measures given in million SEK for each of the six EBA
scenarios.
Option Risk Parallel up Parallel down Short rate up Short rate down Flattener Steepener
Option: 28 -12 1 -0.5 -14 22

Prepayment -8 12 -12 13 15 -13
In Table 5.4, one can see that the ∆EV E for the refinancing is negative in the
scenarios parallel up, short rate up, and steepener and positive for the other three
scenarios. In other words, a general increase in short rates causes a decrease in
∆EV E, while a general decrease in short rates generates an increase in ∆EV E.
This is in line with the theory saying that refinancing tends to increase during pe-
riods of falling interest rates and vice versa [3].
Unlike the refinancing risk, the option risk is calculated separately from other risks.
This risk is dependent on an institution’s interest rate options. In this case, the
made-up bank was using two interest rate caps to hedge against increases in long-
term interest rates. In Table 5.4, it can be seen that the biggest increase in value is
the parallel up scenario, and the biggest decrease occurs in the flattener case. Re-
member that this risk is not looking at the actual cash flows these options generates.
Instead, it is the value of the derivative we are interested in. In our case, we can see
that the value increases when the rates go up and decrease when the interest rates
go down.
Table 5.5: Describes behavioral risk measures given in million SEK for the three
models.
Gap Risk: EBA PCA Cholesky
Option:
M ax 28 31 26
M in -15 -17 -17
F −1 (0.05) - -0.5 -7
F −1 (0.95) - 15 16
ST DV - 5 7
Prepayment:
M ax 15 15 17
M in -13 -17 -17
F −1 (0.05) - -16 -16
F −1 (0.95) - 14 15
ST DV - 10 11
In Table 5.5 it can be seen that the result is almost the same for the different
methods. Remember that the refinancing depends on a scenario multiplier that is set
to different values during different interest rate scenarios. This scenario multiplier
should be set to values that typically indicate the refinancing behavior under the
specific scenarios. The calculation of this multiplier is complex, and a lot of data
is needed. However, in this paper, a simplified approach, recommended by BCBS
[28] for institutions that lacks the data needed, has been used. Where the scenario
multiplier only can take two different values, in which the value of the multiplier
depends on the interest rate scenario. Even tough it is important to notice that this is
a simplified model compared to other more sophisticated options, this is still seen as
a valid approach according to the EBA-framework. On account of this, the result is
almost the same for the different methods. In other words, the refinancing increase
with the same factor for all the scenarios when the short rate is decreasing, and it
decreases with the same factor for all the scenarios when the short rate increases.
This is the reason why ∆EV EM ax and ∆EV Emin is close to each other for the
different models. The reason why they are not exactly the same is because they
have various discount factors.
5.6 Discussion About the IRRBB Model

The IRRBB model that is suggested in this paper fulfills the restrictions given by
the new EBA framework. This does not mean that our model is the only possible
solution, nor that there exists a model with the same traits. The choices of methods
to create an IRRBB model are close to infinite. This results in, depending on the
designer of the model, that the resulting risk would also differ. In this section, we
will discuss some of the decisions that were made when the model was created,
what other alternatives there is, and why we chose the methods that we did.
The EBA framework contains some information about how EVE and NII should be
measured, but the information given is far from a standardized method. How these
measures should be used to calculate the various sub-risks is not specified, nor is it
indicated when it is appropriate to use EVE or NII. This means that it can be diffi-
cult for supervisors or governing agencies (FI in Sweden) to assess the numbers. It
is almost like comparing apples and pears. Comparing the same risk that has been
measured with different methods is not appropriate. The lack of a standard model
and clear guidelines makes it difficult to understand the real interest rate risk in-
stitutions actually are facing. This paper means that standardized methods/models
should be developed to give a continuity of calculations and thus give a more ac-
curate result that becomes easier to evaluate. This standardized model should also
state clearly how ∆EV E and ∆N II should be aggregated in an appropriate way.
This to make sure that all the risks are captured, but also that double accounting
does not occur.
The measure of basis risk is one widely discussed subject in the world of inter-
est rate risk. This due to the trade-off between capturing the full risk of non-perfect
correlated yield curves and not "double count" the same risk multiple times. We
have chosen only to calculate the basis risk within each currency. We see it as
the basis risk between each currency in, for example, the EBA scenario model is
captured by the gap risk measure. This due to that for each currency, the stresses
for each yield curve are of the same type but of different amplitude. For instance,
Rshort,SEK is 300 basis points and Rshort,DKK is 250 basis points. Hence, they
are closely correlated but not perfectly correlated. The argument that the gap risk
measure itself captures much of the true basis risk is, for instance, argued by the
JBA [31] opposing document on BCBS [28] guidance on how IRRBB can be calcu-
lated. Furthermore, the non-yield sensitive measure we have chosen for this model
to calculate basis risk, see Equation 4.23, also got criticism by the Japanese Bankers
Association (JBA) as a too crude instrument that, in most cases, would overestimate
the basis risk measure. However, we have chosen this model due to its simplicity
and transparency, i.e., no black box calculations. Furthermore, from a risk man-
ager’s perspective, the choice of an overestimating risk model is seen as a better
choice than an underestimating model. Another reason why we do not see the
overestimation of basis risk as a problem is due to that, in our case study, the am-
plitude of the basis risk itself is not so big that a slight overestimation of the risk
has a significant impact on the result. However, we find that there are deficiencies
in EBA [1] on how this risk should be managed most appropriately.
How commercial margins should be handled when calculating IRRBB is another

decision a risk manager must take. In the EBA [1] framework, there is a paragraph
that states:
"If commercial margins and other spread components are excluded from economic
value measures, institutions should (i) use a transparent methodology for identify-
ing the risk-free rate at inception of each instrument; and (ii) use a methodology
that is applied consistently across all interest rate sensitive instruments and all
business units." [1]
This gives a risk manager the option not to incorporate the marginal interest rate
in their IRRBB model. Since our model is based upon a fictive bank, we chose not
to incorporate this feature in our model. This due to the lack of accurate historical
data on how the margin change due to yield-curves behavior, which may lead to
high uncertainty in the model. This uncertainty was deemed to have a big impact
on the resulting risk measures and was therefore left out of the calculations. This
choice led to that one can see our marginal interest as a constant. Banks choose
their margin spread to protect themselves somewhat from risk and ensure that the
organization makes a profit. Hence, this simplification of the model may result in
that risks are somewhat overestimated and that if a marginal interest rate were in-
corporated, the risk measures would change to slightly smaller numbers.
As mentioned before, the option risk is not looking at the actual cash flows the hold
options generate. Instead, we are interested in the value of the derivative. Hence,
the valuation method of the derivative plays a significant role. There exist several
different methods that can be used to price options, and, in most cases, they will
not give the same result. The Black 76 model, which has been the most common
model to use when pricing interest rate options, do not accept negative interest rates
nor negative forward rates. Since EBA [1] clearly states that these phenomenons
should be included as plausible scenarios, the Black-76 is not an appropriate model
to use when measuring option risk for some of the scenarios. Thus, Bachelier’s op-
tion pricing model has been used in this paper for all scenarios. It is important to
be consistent with the choice of pricing model since different models generate dif-
ferent results. However, when the forward rate is positive, the prices obtained from
Black-76 and Bachelier- model coincide well [26]. Hence, this paper assumes that
the Bachelier model is an appropriate model to use. Furthermore, some assump-
tions were made when the Bachelier model was integrated onto the IRRBB model.
It was assumed that each caplet/floorlet in a specific cap/floor option belonged to
different time buckets. In other words, the time to expiry for each caplet was the
same as the time interval for the time bucket the caplet belonged to.
In general, the measurement of behavioral risk is the most unpredictable risk since
it is supposed to incorporate human choices in both parts, refinancing and option
[18]. For the refinancing part of this risk, a simple predefined, yet EBA approved
method was chosen. This to, once more, avoid black box calculations and for lack
of accurate historical data on our fictive customers. This is a realistic scenario for
many of the "younger" financial institutions since yield curves, in general, keeps an
approximate similar behavior over longer time periods. For example, the current
low interest-environment has remained similar for the last five years. However, for
banks with more extended behavioral data on their customers, there are other cal-
culation options available with more complex calculations, which produces more
varying and accurate results. For example Bissiri and Cogo [17] presents in their
paper a model that incorporates nonmarket factors, which may contribute signifi-
cantly to the variance of future cash flows.
To conclude, as mentioned earlier, there exists a significant amount of modeling

options each risk manager may take into account, where each choice affects the
level of measured risk. This paper finds this as one of the weaknesses in a frame-
work such as EBA [1] since IRRBB reports to governing agencies may vary sig-
nificantly depending not on the actual risk, but due to the methodological choices
the institution have made.
Chapter 6
Conclusions
This report proposes an example of an IRRBB model that fulfills the EBA require-
ments, see Chapter 4. Furthermore, the report shows some of the limitations that the
new EBA framework possesses. We conclude that in a low-interest environment,
the shape of most of the curves generated by the EBA model becomes unrealistic,
and only a few scenarios are probable. Due to the lower interest rate floor and the
amplitude of the stress constants, several of the generated yield curves are highly
similar. This results in only a few nuances of the actual risk financial institutions
are facing.
For the banking book that has been analyzed in this study, the EBA model cre-
ates more protective risk measures. For a risk-averse risk manager, these calcu-
lations may be deemed prudent enough. However, with a different banking book,
this might not be the case due to the limited amount of credible and diverse yield
curves. One of the major new aspects of the new framework was to create more
scenarios for a risk manager to analyze. The old framework only considered the
parallel up and the parallel down scenario. In a low-interest environment, we argue
that this objective is somewhat lost. To improve this, more flexible stress constants
may be needed where the calculations take the actual level of the yield curve into
consideration and adopt after that. The suggested stochastic models that this paper
proposes as alternative yield curve generating models have, in contrast, this consid-
eration prebuilt in their calculations. The disadvantage for this is that the amplitude
of rates, especially the short rate, does not have as big variation as the EBA model.
This problem may be addressed with an increase of the stress variable, which in
the Cholesky model is a standard uniform variable, see Section 4.3.4, and for the
PCA model is a standard normal variable, see Section 4.3.3.
54
CHAPTER 6. CONCLUSIONS 55
The yield curves produced by both the PCA model and the Cholesky model gener-
ates, in general, a lower amount of risk. Since these simulations are more adaptive
to the recent interest rate climate and, is commonly accepted methods to generate
probable yield curves, these estimates are seen as more probable and more use-
ful in a low-interest environment. The weakness of these is, as mentioned earlier,
the amplitude of the changes in these yield curves is in comparison lower. We ar-
gue that this is a more probable and accurate result. Still, it may create a problem
if some significant unsuspected macro-environment shifts do occur over the short
term. Since an IRRBB model is supposed to be used in a quarterly fashion but
also is to protect a financial institution from liquidity problems due to interest rate
changes, it becomes a trade-off between risk and accuracy.
Further disadvantages the Cholesky and PCA model faces in a low-interest environ-
ment is mainly due to the lower interest barrier suggested by the EBA framework.
Because of the lower barrier, a large number of yield curves are rejected, which
results in an overrepresentation of positive scenarios. Which in this case study, re-
sults in an overrepresentation of negative EVE scenarios.
We can conclude that the EBA-framework gives the institutions a lot of freedom in
their IRRBB calculations. Different methods can be used to calculate the same risk
but give different results. Moreover, it is possible to see that the new requirement
that requires institutions to include both earnings and economic value measures
affects their risk calculation. We see that in our case, ∆EV E and ∆N II have
a degree of negative correlation, which means different scenarios causes differ-
ent risks. This increases the risk overview and makes it harder for institutions to
minimize one risk on behalf of another risk. On the other hand, it is unclear how
the calculated ∆EV E and ∆N II should be aggregated in an appropriate way.
Furthermore, there is no consistency in how financial institutions should calculate
their risk. Thus, IRRBB reports for financial institutions may vary from institution
to institution, although they should not. This may create problems for governing
agencies which task is to control financial institutions and ensure that institutions
are properly protected to interest-rate shifts.
6.1 Further Research

In this section, we present examples of further research that we consider to be rel-
evant. These examples are based on unaddressed questions by this paper and pre-
vious research, which we deem important within the field of interest rate risk.
• According to EBA [1], positive ∆EV E for currency c should be multiplied

CHAPTER 6. CONCLUSIONS 56
by a factor of 0.5. This paper finds this factor too crude and too small. Con-
tinued research could investigate an appropriate way to aggregate the calcu-
lated risk between currencies. One way to do this may be to look at the cor-
relation between currencies yield curves to determine a more reality-based
factor between the IRRBB in different currencies.
• EBA [1] states that credit risk should be investigated separately but should
be included in the IRRBB reporting. This is something that has been beyond
the scope of this paper but would be relevant for further research. Imple-
mentation of credit risk and how to incorporate this measure to an IRRBB
model, such as the one this paper proposes, could give a deeper insight into
the IRRBB credit risk relation and how large these risks are for a specific
institution.
• This paper has looked at some methods for how risk scenarios can be gen-
erated and how the different risks can be calculated. This could be further
investigated by looking at additional methods that could be used when cal-
culating the IRRBB. This may provide a deeper knowledge of how a more
standardized model could look like.
Bibliography
[1] EBA. “Guidelines on the management of interest rate risk arising from non-
trading book activities”. In: (2018). url: https : / / eba . europa .
eu/sites/default/documents/files/documents/10180/
2282655/169993e1-ad7a-4d78-8a27-1975b4860da0/Guidelines%
20on % 20the % 20management % 20of % 20interest % 20rate %
20risk%20arising%20from%20non-trading%20activities%
20%28EBA-GL-2018-02%29.pdf?retry=1&fbclid=IwAR33WQtd7g-
B7acKq_EpSuydOttG0DCqEOZvQn7o5svqepmiMdjp7ou_vJU.
[2] Rosaria Cerrone et al. “Does prudential regulation contribute to effective
measurement and management of interest rate risk? Evidence from Italian
banks”. eng. In: Journal of Financial Stability 30 (2017), pp. 126–138. issn:
1572-3089.
[3] BCBS. “Standards: Interest rate risk in the banking book”. In: (2016). url:
http://www.bis.org/bcbs/publ/d368.pdf.
[4] BCBS. The Basel Committee - overview. url: https : / / www . bis .
org/bcbs/ (visited on 04/14/2020).
[5] Finansinspektionen. “FI:s metoder för bedömning av enskilda risktyper inom
pelare 2”. In: (2018). url: https://www.fi.se/contentassets/
85eefa6358994289b932615a72b99c07 / andrad _ p2 _ 2018 -
05-31.pdf.
[6] Piergiorgio Alessandri and Mathias Drehmann. “An economic capital model
integrating credit and interest rate risk in the banking book”. eng. In: Journal
of Banking and Finance 34.4 (2010), pp. 730–742. issn: 0378-4266.
[7] Roberta Fiori and Simonetta Iannotti. “Scenario Based Principal Compo-
nent Value-at-Risk: An Application to Italian Banks’ Interest Rate Risk Ex-
posure”. eng. In: SSRN Electronic Journal (2006). issn: 1556-5068.
57
BIBLIOGRAPHY 58
[8] Alexander Kreinin et al. “Principal Component Analysis in Quasi Monte

Carlo Simulation”. eng. In: Algo Research Quarterly 1.2 (1998), pp. 21–29.
issn: 0377-2217.
[9] William H. Press et al. Numerical Recipes. The Art of Scientific Computing.
eng. 3rd ed. Cambridge University Press, 2007. isbn: 0-521-88068-8.
[10] Azamat Abdymomunov and Jeffrey Gerlach. “Stress testing interest rate risk
exposure”. eng. In: Journal of Banking and Finance 49 (2014), pp. 287–301.
issn: 0378-4266.
[11] BCBS. “Principles for the Management and Supervision of Interest Rate
Risk”. In: (2004). url: https://www.bis.org/publ/bcbs108.
pdf?fbclid=IwAR1-bDIWau2NUPHezAsCkbnThTwEd5gj70j1vGvoMWxHg_
MsqOrRmgxYw8.
[12] Massimiliano Barbi. “Skewness, basis risk, and optimal futures demand”.
eng. In: International review of economics finance (2018). issn: 10590560.
[13] Christoph Memmel. “Banks’ interest rate risk: the net interest income per-
spective versus the market value perspective”. In: Quantitative Finance 14.6
(2014), pp. 1059–1068. url: https://doi.org/10.1080/14697688.
2011.630326.
[14] Norio Hibiki and Norio Iwakuma. “A Model for Managing Interest Rate
Risk in the Banking Book: A comparison between a modified earnings-
based approach and economic value approach”. In: (2015). url: http :
/ / lab . ae . keio . ac . jp / ~hibiki _ lab / pdf _ k - ris / IH _
APIEMS_2015.pdf.
[15] Aldo Letizia. “A Stochastic Approach to the Measurement of Interest Rate
Risk in the Banking Book”. In: SSRN (2017). url: https : / / ssrn .
com/abstract=3059537.
[16] Asish Saha et al. “Networth exposure to interest rate risk: An empirical anal-
ysis of Indian commercial banks”. eng. In: European Journal of Operational
Research 193.2 (2009), pp. 581–590. issn: 0377-2217.
[17] Matteo Bissiri and Riccardo Cogo. “BEHAVIORAL VALUE ADJUSTMENTS”.
eng. In: International Journal of Theoretical and Applied Finance (2017).
[18] Antoine Frachot. “A Note on Behavioral Models for Managing Optionality
in Banking Books”. eng. In: Groupe de Recherche Operationnelle, Credit
Lyonnais (2001).
[19] Manfred Gilli, Stefan Große, and Enrico Schumann. “Calibrating the Nel-
son–Siegel–Svensson model”. eng. In: Computational Optimization Meth-
ods in Statistics, Econometrics and Finance (2010).
BIBLIOGRAPHY 59
[20] Enrico Schumann Manfred Gilli Stefan Groze. “Calibrating the Nelson–Siegel–Svensson
model”. eng. In: Comisef Working Papers Series (2010).
[21] Francis X Diebold and Canlin Li. “Forecasting the term structure of govern-
ment bond yields”. eng. In: Journal of Econometrics 130.2 (2006), pp. 337–
364. issn: 0304-4076.
[22] Jacob Muvingim and Takudzwa Kwinjo. “Estimation of Term Structures us-
ing Nelson-Siegel and Nelson-Siegel-Svensson: A Case of a Zimbabwean
Bank”. In: Journal of Applied Finance & Banking 4.6 (2014), pp. 1–9. url:
https://ideas.repec.org/a/spt/apfiba/v4y2014i6f4_
6_9.html.
[23] Gareth James et al. Machine Learning and Data Science: An Introduction
to Statistical Learning Methods with R. eng. 7th ed. 2017. Springer Texts in
Statistics, 2013. isbn: 978-1-4614-7137-0.
[24] Nicholas J. Higham. “Cholesky factorization”. eng. In: 1.2 (2009), pp. 251–
254. issn: 1939-5108.
[25] Biswa Datta. Numerical Methods for Linear Control Systems. eng. 1st ed..
2004. Chap. 3. isbn: 9780080537887.
[26] Walter Schachermayer and Josef Teichmann. “How close are the option pric-
ing formulas of Bachelier and Black-Merton-Scholes?” In: (2007).
[27] BCBS. “SRP31 Interest rate risk in the banking book”. In: (2019). url:
https://www.bis.org/basel_ framework/chapter/SRP/
31.htm?tldate=20221231&inforce=20191215.
[28] BCBS. “Consultative Document:Interest rate risk in the banking book”. In:
(2015). url: https://www.bis.org/bcbs/publ/d319.htm.
[29] Morten Helbæk, Ragnar Løvaas, and Jon Olav Mjølhus. Time value of money.
eng. Routledge, 2013, pp. 71–79. isbn: 9780415625968.
[30] Henrik Hult et al. Risk and Portfolio Analysis : Principles and Methods. eng.
New York: Springer, 2012. isbn: 9781493900312.
[31] JBA. “Comments on the Consultative Document: Interest Rate Risk in the
Banking Book issued by the Basel Committee on Banking Supervision”.
eng. In: (2015).
TRITA SCI-GRU 2020:058
www.kth.se

IRRBB Under Low Rate Environment

Uploaded by

Copyright:

Available Formats

IRRBB Under Low Rate Environment

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IRRBB Under Low Rate Environment

Uploaded by

Copyright:

Available Formats

DEGREE PROJECT IN MATHEMATICS,

SECOND CYCLE, 30 CREDITS

IRRBB in a Low Interest Rate

KTH ROYAL INSTITUTE OF TECHNOLOGY

Degree Projects in Financial Mathematics (30 ECTS credits)

Royal Institute of Technology

4.3.3 PCA Interest Rate Model . . . . . . . . . . . . . . . . . . 26

5 Results and Discussion 35

interest rates also affect a bank’s earnings by altering interest-rate-sensitive in-

1.2 Previous Research

1.4 Research Question

• How to manage IRRBB according to the new EBA-regulations in a low-

• How to build a model that fulfills the EBA standards?

• How to stress test this model?

• What effects do a low-interest environment have on the risk model?

• What weaknesses do the new EBA-regulations have?

1.5 Outline of the Thesis

2.1.1 Gap Risk

2.1.2 Basis Risk

2.1.3 Behavioral Risk

2.2 The Legal Environment of IRRBB

To understand the new standards and how it affects financial institutions, it is a

2.3 Measures of IRRBB

Earnings measures look at changes in expected future profitability within a given

2.4 Balance Sheet

3.1 Nelson-Siegel-Svensson Parameter Estima-

To avoid potential challenges in numerical optimization, Diebold and Li [21] in-

 −t1 −t1 −t1 

Where β̂ is a column vector consisting of the four βi constants:

The optimal parameters of β̂ can then be estimate by the equation:

β̂ EST = (X T X)−1 X T Y M (3.6)

3.2 Principal Component Analysis

iii. Construct a projection matrix W consisting of the k selected eigenvectors that

Z 1 = φ11 X 1 + φ21 X 2 + ... + φd1 X d (3.7)

zi1 = φ11 xi1 + φ21 xi2 + ... + φp1 xip (3.8)

The second principal component is the linear combination of X 1 , X 2 , ..., X d that

zi2 = φ12 xi1 + φ22 xi2 + ... + φd2 xid (3.10)

φ1 , φ2 , ..., φk [23]. We refer to the matrix φ as the projection matrix.

Where W −1 is the inverse of the projection matrix W .

3.3 Cholesky Factorization

The Cholesky decomposition is a commonly used mathematical tool in a Monte

3.4 Forward Rate

ln(p(s, T ) − ln(p(s, t))

Where p(t, T ) is the price of a non-defaultable zero-coupon bond with maturity at

p(s, T ) = e−r(s,T )·(T −s) (3.16)

Where r(s, T ) is the interest rate between s and T at time s, if s = 0 we approxi-

3.5 Bachelier Pricing Model

Just like a cap, a floor can be priced as a discounted sum of floorlets:

4.1 Mapping of the Cash Flow Data

4.2 Yield Curve Construction

In order to generate a yield curve, bond prices or equivalent instruments need to

In the model proposed in this paper, the Nielson-Siegel-Svensson (NSS) method

4.3 Interest Rate Models

4.3.1 Interest Rate Restrictions

4.3.2 EBA Interest Rate Model

(i) Parallel shock up

Currency SEK USD GBP DKK EUR JPY

Parallel shock for currency c:

∆Rparallel,c (tk ) = ±R̄parallel,c (4.3)