Predictive Modeling for

Market Risk in the

Banking Book
4EK614

Applied Operational Research in Business Consulting

With You Today

Tomáš Sobotka Our services

Senior Manager in Market Risk Team
Tomas.Sobotka@cz.ey.com
Market Risk &
Regulatory Team
Branislav Lovás
Senior Consultant in Market Risk Team
Branislav.Lovas@cz.ey.com Quantitative Treasury
Services
Eliška Kompanová
Senior Consultant in Market Risk Team
Eliska.Kompanova@cz.ey.com
ALM Regulatory
Adéla Nguyenová
Consultant in Market Risk Team
Adela.Nguyenova@cz.ey.com

Our projects
Model Development Data Analysis Stress Testing
Model Validation Data Mining Forecasting

Securities Valuations Investment RoboAdvisors Regulatory Reporting

Derivatives Valuations Behavioral Modelling Methodological Review

Page 2 Predictive Modeling for Market Risk in the Banking Book

Housekeeping:
Study materials
Course Structure
1. PowerPoint slides, provided after the course

2. Jupyter notebook *.ipynb files, provided after the course

Day 1: Credit Risk
Day 2: Market Risk
Classroom: Online
Time: 9:15am – 10:45am Prerequisites
11:00am – 12:30pm

1. Basic understanding of statistical and mathematical concepts

(mathematical modelling, time series)

2. Elementary knowledge of programming (Python, R, VBA)

Please see Annex to links with helpful sources.

Course Assessment
Please note that you are not limited to methods presented today (and it is more than welcomed to new approaches)
Non-Maturity Deposit Modelling
1. Case study – you will obtain dataset for modelling balance of non-maturity deposits. Your task will be to analyze the dataset, split it to train set and test
set. With train dataset, you will develop selected model for non-maturity deposit balances. Then you will take your test dataset and see how your model
performs.
2. Outputs – PPT presentation or PDF, summarizing the abovementioned outputs, and scripts in Python, R or VBA (including the spreadsheet) that were
used. Please send these outputs to Tomas.Sobotka@cz.ey.com. Submission deadline: 7 May 2021
3. Output presentation for EY (most important!)– short (10-15 minutes) presentation + Q&A about results of this assessment; the presentation will be
delivered online and will take place during the week starting from 10 May 2021.

Page 3 Predictive Modeling for Market Risk in the Banking Book

Contents

1. Mathematical Crash Course 3 Non-Maturity Deposit modelling

I. Somewhat Heuristic Introduction to Probability and I. Segmentation
Stochastic Processes II. Liquidity modeling of NMD
II. Time Series Modelling III. IR sensitivity models – β and optimal portfolio approach
III. Linear Regression IV. Other issues

2. Behavioral modelling in the Banking Book

I. Introduction to the Banking Book 4 Hands-on Case Study: Instructions for Module
II. Behavioral models Assignments

Page 4 Predictive Modeling for Market Risk in the Banking Book

1 Mathematical Crash
Course
Somewhat Heuristic Introduction to
Probability and Stochastic Processes
Time Series Modelling
Linear Regression

Page 5
Questions 1: Basic Demographics

Go to www.menti.com and use the code 15 68 21 92

Page 6 Predictive Modeling for Market Risk in the Banking Book

Probability Space

• Key object is a so-called probability space, which is a triplet

(Ω, ℱ, 𝑃):
• Ω: Sample space (contains all possible outcomes)
• ℱ: 𝜎 –algebra on Ω, (contains all events we are interested in) i.e.:
Example:
• Ω∈ℱ
• If 𝐴1 , 𝐴2 … ∈ ℱ then ‫ڂ‬+∞
𝑖=1 𝐴𝑖 ∈ ℱ
• If A ∈ ℱ then 𝐴𝐶 = Ω − 𝐴 ∈ ℱ • Rolling dice – what would be Ω?
• What is the smallest ℱ?
• 𝑃 – probability measure, i.e. a set function 𝑃: Ω ↦ [0, 1]: • What if we want to know what
is the probability of casting an
• 𝑃 ∅ = 0; 𝑃 Ω = 1 odd / even number?
• For pairwise disjoint sets Ai ∈ ℱ: 𝑃 ‫ڂ‬+∞ +∞
𝑖=1 𝐴𝑖 = σ𝑖=1 𝑃(𝐴𝑖 )

• The above might sound theoretical, but the triplet is a

cornerstone of any stochastic modelling
• Additionally, we might consider a filtered probability space
(Ω, ℱ, ℱ𝑡 𝑡≥0 , 𝑃) where ℱ𝑡 𝑡≥0 is an increasing set of sub-𝜎-
algebras ⊂ ℱ
Page 7 Predictive Modeling for Market Risk in the Banking Book
Probability – Conditioning & Independence

• Assuming probability space Ω, ℱ, 𝑃 and two events 𝐴, 𝐵 ∈ ℱ

Conditional probability:
𝑃(𝐴 ∩ 𝐵)
Example:
• 𝑃 𝐴𝐵 = 𝑃(𝐵)
≈ “probability of event A occurring if event B occurs”
• Rolling dice
Independence of two events: • What is probability of 𝐴 = {1}
• Two events 𝐴, 𝐵 ∈ ℱ are said to be independent if: conditional on B = {1,3,5}?
• 𝑃 𝐴 ∩ 𝐵 = 𝑃 𝐴 𝑃(𝐵) • What is a conditional
probability of two independent
events?

Page 8 Predictive Modeling for Market Risk in the Banking Book

Questions 2: Roll a Dice

Go to www.menti.com and use the code 15 68 21 92

Page 9 Predictive Modeling for Market Risk in the Banking Book

Random Variables

• Real-valued random variable is defined as 𝑋: Ω ↦ ℝ

• For math nerds: 𝑋 is a measurable function on Ω, ℱ, 𝑃

Technically: Examples:
• We denote,
𝑋 = 𝑥 = 𝜔 ∈ Ω: 𝑋 𝜔 = 𝑥 • Rolling dice
𝑋 < 𝑥 = 𝜔 ∈ Ω: 𝑋 𝜔 < 𝑥 • 2 dice rolled at once, 𝑋 is the
𝑎 < 𝑋 < 𝑏 = 𝜔 ∈ Ω: 𝑎 < 𝑋 𝜔 < 𝑏 sum of casted values on dice
• Identity function
• Analogously, random variables 𝑋1 , 𝑋2 are independent if for all • Coin tossing bet
𝑥1 , 𝑥2 : • Random walk of a drunk
𝑃 𝑋1 = 𝑥1 ; 𝑋2 = 𝑥2 = 𝑃 𝑋1 = 𝑥1 𝑃(𝑋2 = 𝑥2 )
𝜔=H $100

Realization
of 𝜔 ∈ Ω

𝜔=T
Page 10 Predictive Modeling for Market Risk in the Banking Book $0
Useful Functions: CDF & PDF

Cumulative distribution function

• Let 𝑋 be a real-valued (continuous) random variable on Ω, ℱ, 𝑃 , then 𝐹 𝑥 : 𝐹 𝑥 = 𝑃 𝑋 ≤ 𝑥 =

𝑥
‫׬‬−∞ 𝑓𝑋 𝑠 𝑑𝑠, where
• 𝐹 𝑥 is the cumulative distribution function (CDF)
• 𝑓𝑋 (𝑥) is the probability density function (PDF), (probability mass function for discrete random variables)

• The CDF: Discrete random variable Continuous random variable

• ranges from 0 to 1,
• is an increasing function,
• is right-continuous.

Page 11 Predictive Modeling for Market Risk in the Banking Book

Important Example: Normal Distribution

• The normal distribution is bell-shaped, entirely defined by two parameters

• Mean - 𝜇
• Standard deviation - 𝜎
• The normal PDF can be analytically expressed:
1 (𝑥–𝜇)2
–
𝑓 𝑥 = 𝑒 2𝜎2
𝜎 2𝜋

Standard normal distribution

• The standard normal distribution has 𝜇 = 0 and 𝜎 = 1

• Why is this important?

-0,3 -0,2 -0,1 0 0,1 0,2 0,3

Page 12 Predictive Modeling for Market Risk in the Banking Book

Questions 3: Other probability distribution functions

Go to www.menti.com and use the code 15 68 21 92

Page 13 Predictive Modeling for Market Risk in the Banking Book

Stochastic processes - Intro

Stochastic (random) process

• Let 𝜏 be a subset of [0, +∞], then a family of random variables indexed by 𝜏: 𝑋𝑡 𝑡∈𝜏 is a stochastic
process. Two basic observations (for real-values stochastic processes):
• 𝑋𝑡 𝜔 : 𝜏 ↦ ℝ for a fixed 𝜔 is a sample path (or trajectory)
• 𝑋𝑡 𝜔 : Ω ↦ ℝ for a fixed 𝑡 is a random variable

Example – constructing a “Random Walk” model

• A simple random walk process 𝑋𝑛 𝑛∈ℕ0 can be constructed:

1. 𝑋0 = 0
2. Increment 𝑋𝑛+1 − 𝑋𝑛 is independent of all previous increments
3. Increment 𝑋𝑛+1 − 𝑋𝑛 has a coin toss distribution (binomial):
1
𝑃 𝑋𝑛+1 − 𝑋𝑛 = +1 = 𝑃 𝑋𝑛+1 − 𝑋𝑛 = −1 =
2

Page 14 Predictive Modeling for Market Risk in the Banking Book

Stochastic time-series models

• Stochastic time-series is a sequence of data collected at specified time points 𝑡1 , 𝑡2 , 𝑡3 , …

• Trend, seasonality, cyclicity, irregularity

ARIMA(p,q): 𝑋𝑡 − 𝛼𝑋𝑡−1 − ⋯ − 𝛼𝑝 𝑋𝑡−𝑝 = 𝜖𝑡 + 𝜃1 𝜖𝑡−1 + ⋯ + 𝜃𝑞 𝜖𝑡−𝑞

• Auto regressive integrated moving average
• Three main parameters:
• p – number of autoregressive terms (AR) … current value as a linear combination of past values
• d – non-seasonal differences to achieve stationarity (I)
• q – number of lagged forecast errors in the prediction equation (MA) … current error as a linear combination of past errors

Generalized GBM

Page 15 Predictive Modeling for Market Risk in the Banking Book

Linear Regression

• Linear approach to modeling a relationship between one dependent variable (outcome) and multiple independent variables
(inputs)
𝑦 = 𝑥𝑇𝛽

• Linearity
y

• Normality – errors have normal distribution

• Homoscedasticity – same variance

• Independence of observations

• Lack of perfect multicollinearity

• MLE:
𝑌෠ = (𝑋 𝑇 𝑋)−1 𝑋 𝑇 𝑌

Page 16 Predictive Modeling for Market Risk in the Banking Book

Questions 4: Other models

Go to www.menti.com and use the code 15 68 21 92

Page 17 Predictive Modeling for Market Risk in the Banking Book

2 Behavioral modelling in
the Banking Book

Page 18
Key products offered by banks

Assets Liabilities

Retail segment Across all segments

Personal loans
Mortgage (various types) Current accounts
Debt consolidation Savings accounts
Overdrafts Term deposits (single/revolving)
Credit cards Structured deposits (embedded
options)
SME segment Bonds
Commercial loans - secured/unsecured

Corporate segment
Investment loans
Working capital loans

Page 19 Predictive Modeling for Market Risk in the Banking Book

 Role of a Typical Commercial Bank

‘Typical’ Balance Sheet Role of Maturity Transformation

Assets Liabilities

Hedge
Asset

Loan
Loans & Deposits
Advances
Time

Liability
Deposit
Other Borrowing

Hedge
Liquid Assets (Wholesale)

Fixed Assets Capital

Banks play an important role as an intermediary in the financial system. They typically have two main functions:
1. Take on deposits from the public (typically ‘short term’ in nature)
2. Lend out loans / mortgages (typically ‘long term’ in nature)

Page 20 Predictive Modeling for Market Risk in the Banking Book

Banking Book

• Banking Book assets and liabilities are those which are intended to be held on the balance sheet until
maturity at amortized value.

• The banking book can also include those derivatives that are used to hedge exposures arising from the
banking book activity, including interest rate risk.

Page 21 Predictive Modeling for Market Risk in the Banking Book

Behavioral Models

Behavioral
• In order to match assets and liabilities with the same interest finance
rate profile, banks need to know the maturity (re-price) date of
Psychology
those assets and liabilities.
• The following assets and liabilities are often associated with
customer optionality and therefore the maturity is not clear:
Finance / Decision-
Product Type Optionality economics making

Fixed Term Loan Customers may have the option to re-pay the loan earlier
than its maturity/reprice date
Fixed Term Deposit Customers may have the option to withdraw deposits
earlier than the fixed term or to top-up deposits at the
original fixed term rate
Non-Maturity Deposit There is no stated maturity but customers have the
option to withdraw at any time
Pipeline Banks need to take a view on assets/liabilities which are
contractual but have not hit the balance sheet yet.

Page 22 Predictive Modeling for Market Risk in the Banking Book

Behavioral Models

Banks employ a number of techniques to assess the likely behaviour of these balances in order to match assets and
liabilities of the same interest rate profile:

1. Fixed rate products with pre-payment / early withdrawal / top up optionality

Determined
using
statistical
2. Non-maturing deposits modelling of
behaviours
under different
scenarios

Page 23 Predictive Modeling for Market Risk in the Banking Book

Behavioral Models – input data types

• Market
• interbank and governments

• Macro-economic
• inflation, economic growth

• Behavioral
• reaction of customers to the market (withdrawing, saving, …)

• Personal
• Age, education, job, ..

Page 24 Predictive Modeling for Market Risk in the Banking Book

What Risks are in the
Banking Book?
3 Non-Maturity Deposit
modelling

Page 26
Questions 5: Non-maturity deposits

Go to www.menti.com and use the code 15 68 21 92

Page 27 Predictive Modeling for Market Risk in the Banking Book

Non-maturity deposits

• Liabilities in the Banking book

• Depositors can withdraw their funds at any time without any penalty
• Interest rates are unilaterally set up by the Bank
• One of the most stable sources of funding for banks; asset
• Contractual maturity – overnight (in reality significant effective duration)
• Retail savings, money-market accounts a) Liquidity runoff model
• Modelling this early redemption option is very challenging
• NMD balances can be modelled under 3 different assumptions:
• Run-off
Liquidity risk
• Static balance-sheet management
• Assuming inflow of new clients
• Factors affecting NMD:
• Market (interbank and governments)
• Macro-economic (inflation, economic growth) FTP
b) IR sensitivity model
• Behavioral (reaction of customers to the market)
Interest rate
► Main components discussed: risk / ALM
a) Liquidity runoff model / balances prediction
b) IR sensitivity model (replication portfolio)

c) Floor add-on

Page 28 Predictive Modeling for Market Risk in the Banking Book

2.1 Modelling of NMD: Segmentation

► The first step in the deposit characterization process is to divide demand deposits into segments, which is critical
for an accurate analysis.
► The main goal of product segmentation is to find homogeneous enough segments, while ensuring a minimum
threshold of materiality and simplicity.
► The segmentation process tries to reach a balance between three ideals: Homogenity

Materiality Simplicity

► Typical segmentation criteria are:

1) Product type: Completely different products should not be examined together
2) Currency: Products with different currencies should not be pooled together.

► Segmentation process requires deep knowledge on the specifics, management strategy and historical behavior of
the products.

Page 29 Predictive Modeling for Market Risk in the Banking Book

2.2 Liquidity modelling of NMD: Overview

Model goal Modelling approaches

► Assign a runoff profile ► Segmentation based
to deposit products on various factors, e.g.
Overview currency, product time,
► A blend of quantifiable country, type of origination
► The profiles should be
statistical models coupled (on-line / branch)
consistent with the historical
with business adjustments
behaviour of the deposits
according to prevailing ► Two types of statistical
market conditions ► Good segmentation of the models (in practice,
products is essential combination of both is used)
(next slides)

Page 30 Predictive Modeling for Market Risk in the Banking Book

2.2 Liquidity modelling of NMD: Run-off profile

Stochastic time-series models

(ARIMA, generalized GBM)
► Modelling deposit balances evolution:
1. Runoff profile as an confidence interval of the fitted model (excl. new production)
2. Prediction of balances (including new production) expressed as expected value of the balance conditional on
current information – evaluated by:
I. Analytic expression (we know how to express this e.g. for GBM)
II. By Monte Carlo simulation

► Seasonality typically taken into

account
► Statistical testing in place to validate
appropriateness of the model

Page 31 Predictive Modeling for Market Risk in the Banking Book

2.2 Liquidity modelling of NMD: Core volume and vintage analyses

15 000
► Balance volatility model determines what level of balances Step 2: Measure variation from the trend
14 000
will remain with the bank in the long term given a level of
13 000
confidence.
12 000 Step 1: Fit the trend line

Balance
► As an output, balance volatility model divides the balances 11 000
into two portions: 10 000
► Core balances which remain with the bank under almost 9 000
all market conditions 8 000 Step 3: Determine the core portion

► Non-core balances which fluctuate over time due to 7 000

I-06 VII-06 I-07 VII-07 I-08 VII-08 I-09 VII-09 I-10 VII-10
market and idiosyncratic reasons and are typically
Months
characterized as short term.
Core balance Balance development Linear fit
► The analysis of core vs. non-core balances is performed
100 Product run-off profile
through a growth based regression model.

Remaining balance in %
80
► Balance volatility model fits the product balances to trend
models (exponential, linear or logarithmic) and find the 60

variation of actual balances around the trend line. 40

20

-
- 10 20 30 40 50 60
Months

Jan 2006 Feb 2006 Mar 2006 Apr 2006

Page 32 Predictive Modeling for Market Risk in the Banking Book

2.2 Liquidity modelling of NMD: Challenges and Market practices highlights

Challenges
► To distinguish between seasonal outflows and surge outflows (due to environmental
changes)
► Lack of historical balance data to support the granular segmentation analytics
► Difficult to have statistically sound models of type 1. for all products
► One might need to switch to a simpler expert model instead
► Good model governance is key here

Market practices highlights

► Hypothesis testing for confidence interval reliability
► Concentration risk
► Average account balance vs # accounts
► Seasonality & surge balance adjustments

Page 33 Predictive Modeling for Market Risk in the Banking Book

2.3 IR sensitivity models – β and portfolio optimization: Overview

Overview
Model goal ► As previously, a blend of Modelling
quantifiable statistical models and approaches
► To stabilize NII and
expert adjustments is put in place
to ensure correct ► Two main concepts
inputs for FTP ► Adjustments should align for are utilized,
mechanisms forward looking expectations on sometimes they are
rates and for business rationale combined together
► Main output of the model – (i.e. via cointegration
replication portfolio analysis)

Page 34 Predictive Modeling for Market Risk in the Banking Book

2.3 Modelling of NMD: Rate sensitivity analysis

► Characterization of administered rates for deposit products describes the significance of market rate changes on the rates paid by bank to the
customer.
► By modeling the behavior of interest rates paid relative to market rates, a better understanding of repricing behavior and the sensitivity of
interest expenses to market rates can be gained.
► The interest rate paid to indeterminate maturity products’ depositors is typically a combination of one or more market rates, and fixed
component.

5,0% Interest rate development ► In order to determine which of market rates

4,5% explain the changes in the rates paid by the
4,0% bank, a multiple regressions are made.
3,5%
Interest rate

3,0% ► Coefficients of this regression equation tell

2,5% what percentage of the portfolio reprices with
2,0%
the associated market rate.
1,5%

1,0% Δproduct rate = 40% Δ 1M PRIBOR

0,5%

► This repricing behavior indicates that 40% of

the portfolio reprices with the 1M PRIBOR,
Months while the remaining 60% of the portfolio is rate
Product rate 1M PRIBOR 3M PRIBOR insensitive.
6M PRIBOR 12M PRIBOR

Page 35 Predictive Modeling for Market Risk in the Banking Book

 2.3 IR sensitivity models – β and portfolio optimization: Modelling approaches

1. Portfolio optimization 2. β – rate sensitivity models

► Typically connected with liquidity runoff (as ► Provide assumptions on how big change in interest rates one can pass to specific
an upper bound or target) and regulatory client categories
requirements
► Beta (β) represents a deposit elasticity to changes in interest rates
► One tries to find a portfolio of market
► Retail NMD typically exhibit lower β compared to Wholesale deposits and savings
instruments that would have led to optimal
accounts
utility function value over relevant historical
time frame ► Short-term/Long-term β and Linear /Dynamic β choices typically considered
► One might consider more than one criterion
(Sharpe, margin stability, duration, impact
under stress scenario)
Typical β models
Model Description
Change in rates proportional to the
Linear rate model
Typical utility functions benchmark

Criterion Optimisation goal Mean reverting equilibrium

Changes due to short- and long-term effects
model
Volatility of the margin Minimize
Direction dependent rate Different sensitivity to increases and
Correlation between model decreases of the benchmark
Maximize
customer and market rates
Different sensitivities when benchmark rates
Level dependent rate model
Sharpe ratio Maximize are low / high

Page 36 Predictive Modeling for Market Risk in the Banking Book

2.3 IR sensitivity models – β and portfolio optimization: Modelling approaches

Dynamic β Linear β Interest rate development

5,0%
► Reflects the changing sensitivity of ► A constant sensitivity of changes in
4,5%
customer rates depending on the level customer rates to changes in
4,0%
of interest rates as well as the relative interest rates

Interest rate
3,5%
change in interest rate across the life Product rate
3,0%
of the deposit
2,5% 1M PRIBOR
2,0% 3M PRIBOR
► Provides a better convexity measure ► Operationally easier 1,5% 6M PRIBOR
to the deposit profile ► Requires a less frequent update to 1,0%
12M PRIBOR
► Captures the asymmetrical behavior the hedging strategy 0,5%
as well as beta response to changes
in the level of rates

► Challenging operationally ► Could be significantly off from the Short term β Long term β
► Difficult to apply to FTP spot beta at a given time, resulting
in disincentives ► Influenced by tactical ► Influenced by statistically determined
► Can be difficult to implement micro- pricing expectations, line of historical experience (long term), and
level hedging strategy ► Reduces transparency for the
business Judgment, and business judgement.
Business
► Requires frequent beta assumption acute historical experience ► Reflects deposit rate paid in equilibrium
review. Most common review
frequency range from 6-12 months

► Constant recalibration and ► Less frequent back-testing

back-testing

Page 37 Predictive Modeling for Market Risk in the Banking Book

2.3 IR sensitivity models – β and portfolio optimization: Challenges and Market
practices highlights

Challenges
► What historical time-frame to use (e.g. short-term β vs long term
β), also applicable for portfolio optimization
► Manage operational complexity, especially for replicating
portfolio and dynamic β approaches
► Finding a correct segmentation of NMD volumes

Market practices highlights

► Portfolio optimization can be done by local optimizer with initial guess equal to previous
weight, or discrete steps
► Embedded floors taken into account:
a) Combination of β-Level and direction dependent rate models
b) via add-ons (incorporating value of IR floors)
c) or ignored

Page 38 Predictive Modeling for Market Risk in the Banking Book

2.4 Other issues

• Negative IR policy (NIRP) and embedded optionality

• FTP rate
• An all-in FTP rate for a deposit product is the sum of base rate, term liquidity premium, and a contingent liquidity
cost.

• New benchmark rates

Page 39 Predictive Modeling for Market Risk in the Banking Book

 Questions 6: Your balance

Balance 1 Balance 2
1500 1000
800
1000
600
400
500
200
0 0
1 4 7 10 13 16 19 22 25 28 31 1 4 7 10 13 16 19 22 25 28 31

Balance 3 Balance 4
115 1500
110
1000
105
500
100
95 0
1 4 7 10 13 16 19 22 25 28 31 1 4 7 10 13 16 19 22 25 28 31

Go to www.menti.com and use the code 15 68 21 92

Page 40 Predictive Modeling for Market Risk in the Banking Book
Questions 7: Modelling approaches

Go to www.menti.com and use the code 15 68 21 92

Page 41 Predictive Modeling for Market Risk in the Banking Book

4 Hands-on Case Study:
Instructions for Module
Assignments

Page 42
Practical Part – Jupyter Notebook

Freely available as part of Anaconda Distribution at: https://www.anaconda.com/distribution/

We use Python 3.7 version.

Page 43 Predictive Modeling for Market Risk in the Banking Book

Course Assessment

• Please note that you are not limited to methods presented today (and it is more than
welcomed to new approaches)
• Non-Maturity Deposit Modelling
• 1. Case study – you will obtain dataset for modelling balance of non-maturity deposits. Your
task will be to analyze the dataset, split it to train set and test set. With train dataset, you will
develop selected model for non-maturity deposit balances. Then you will take your test dataset
and see how your model performs.
• 2. Outputs – PPT presentation or PDF, summarizing the abovementioned outputs, and scripts
in Python, R or VBA (including the spreadsheet) that were used. Please send these outputs to
Tomas.Sobotka@cz.ey.com. Submission deadline: 7 May 2021
• 3. Output presentation to the Bank Board – short (10-15 minutes) presentation about results of
this assessment; the presentation will be delivered online. Presentations will take place during
week starting from 10 May 2021.

Page 44 Predictive Modeling for Market Risk in the Banking Book

Course Assessment – structure of the presentation

1. Data description and model selection

2. Brief model description

3. How your model performs (make prediction for the fourth year and compare the prediction
with the real values)

4. Summarize your results

Page 45 Predictive Modeling for Market Risk in the Banking Book

Course Assessment – evaluation

• You will present your results to the Bank Board (four of us)

• What will be important for us?

1. The quality of your presentation and deliverables
2. The quality of your model (assumptions, errors, ..)
3. The innovation

Page 46 Predictive Modeling for Market Risk in the Banking Book

5 Annex + Q&A

Page 47
Optimization (1)

• The mathematical optimization (also known as mathematical programming) forms a cornerstone of quantitative
portfolio optimization; it can be described as a process of looking for the extreme of a function (maximum or
minimum).
• Apart from asset allocation, optimization has many uses in quantitative finance – model calibration, financial
derivates pricing or asset-liability management.
• Optimization problem can be summarized by three components:

Optimization problem component Asset allocation application

Portfolio risk as a function of portfolio
Objective function 𝑓(𝑥)
weights, e.g. portfolio volatility 𝑤 𝑇 Σ𝑤
Vector of control variables 𝑥 Vector of portfolio weights 𝑤

Constraints set ℂ Portfolio weights must add up to 100%

Useful optimization property:

and can be formally written as: max 𝑓 𝑥 = min −𝑓 𝑥

min 𝑓(𝑥)
𝑤∈ℂ
𝑠. 𝑡. ℂ: 𝑔𝑖 𝒙 ≤ 0 𝑖 = 1, … , 𝐼 (inequality constraints) (25)
ℎ𝑗 𝒙 = 0 𝑗 = 1, … , 𝐽 (equality constraints)

Page 48 Predictive Modeling for Market Risk in the Banking Book

Optimization (2)

• There are two types of function extremes:

• Local (maxima/minima) – extreme of the function in the neighborhood of Global maximum
feasible solutions.

• Global (maxima/minima) – extreme of the function in the set of all feasible Local maxima
solutions.

• The characteristics of the optimization problem (also referred to as

mathematical programs) components of the determine its type, e.g.
linear objective function with linear constrains form a linear
program, a quadratic objective function with linear constraints leads
to quadratic program. Individual types of optimization problems
have their properties which can be usually exploited – e.g. in case of
convex programs all local extremes are also global ones.

• Some of the optimization problems can be solved analytically with

the help of matrix algebra (e.g. portfolio variance minimization with
equality constraints) but majority of problems require numerical,
iterative algorithms, for example Newton-type methods such as
steepest descent.

Page 49 Predictive Modeling for Market Risk in the Banking Book

EY | Assurance | Tax | Transactions | Advisory

About EY
EY is a global leader in assurance, tax, transaction and advisory
services. The insights and quality services we deliver help build trust and
confidence in the capital markets and in economies the world over. We
develop outstanding leaders who team to deliver on our promises to all
of our stakeholders. In so doing, we play a critical role in building a better
working world for our people, for our clients and for our communities.

EY refers to the global organization and may refer to one or more of the
member firms of Ernst & Young Global Limited, each of which is a separate legal
entity. Ernst & Young Global Limited, a UK company limited by guarantee,
does not provide services to clients. For more information about our
organization, please visit ey.com.

© 2019 Ernst & Young, s.r.o. | Ernst & Young Audit, s.r.o. | E & Y Valuations s.r.o.
All Rights Reserved.

This material has been prepared for general informational purposes only and is not
intended to be relied upon as accounting, tax, or other professional advice. Please refer
to your advisors for specific advice.

ey.com