III.

Probability Modeling

Term 2, 2024/2025 DSA211: Lecture 3 1

 Chi-square distribution

[Link]

Term 2, 2024/2025 DSA211: Lecture 3 2

Term 2, 2024/2025 DSA211: Lecture 3 3
 t-Distribution

[Link]

Term 2, 2024/2025 DSA211: Lecture 3 4

Term 2, 2024/2025 DSA211: Lecture 3 5
 Why is it t-distribution?

Term 2, 2024/2025 DSA211: Lecture 3 6

 Poisson modelling
• Over a period of three years (=1096 days) in London,
records showed that a total of 903 deaths occurred among
males eight-five years of age and older.
• The data are given as follows:
Number of Deaths Number of Days Proportion
0 484 0.442
1 391 0.357
2 164 0.150
3 45 0.041
4 11 0.010
5 1 0.001
6+ 0 0.000

We may use the Poisson distribution to model the number of

death per day.
Term 2, 2024/2025 DSA211: Lecture 3 7
 Number of Number of Proportion Poisson
Deaths Days Model
0 484 0.442 0.4387
1 391 0.357 0.3615
2 164 0.150 0.1489
3 45 0.041 0.0409
4 11 0.010 0.0084
5 1 0.001 0.0014
6+ 0 0.000 0.0002
1096 1.000 1.000

• Estimate the Poisson parameter µ

• Calculate the probabilities based on
estimated parameter.

Term 2, 2024/2025 DSA211: Lecture 3 8

 Chi-square goodness-of-fit test

Number of Number of Proportion Poisson Expected Expected

Deaths Days Model Frequencies Frequencies
0 484 0.442 0.4387 480.834 480.834
1 391 0.357 0.3615 396.159 396.159
2 164 0.150 0.1489 163.198 163.198
3 45 0.041 0.0409 44.820 44.820
4 11 0.010 0.0084 9.232 10.989
5 1 0.001 0.0014 1.521
6+ 0 0.000 0.0002 0.236
1096 1.000 1.000
Term 2, 2024/2025 DSA211: Lecture 3 9
Term 2, 2024/2025 DSA211: Lecture 3 10
R-code
library(fitdistrplus) Call the package

fpois <- fitdist(death, distri="pois") Poisson: pois

Negative Binomial: nbinom
Geometric: geom
fpois <- fitdist(death, "pois")
result1 <- gofstat(fpois, Determine the number of classes
chisqbreaks=c(0:3), discrete=TRUE, such that each theoretic class
count is at least 5
fitnames=c("Poisson"))

Term 2, 2024/2025 DSA211: Lecture 3 11

Term 2, 2024/2025 DSA211: Lecture 3 12
Term 2, 2024/2025 DSA211: Lecture 3 13
 Interval between eruptions (in
months)

14
12
Frequency

10
8
6
4
2
0
20 40 60 80 100 120 140
Interval in months

Term 2, 2024/2025 DSA211: Lecture 3 14

Term 2, 2024/2025 DSA211: Lecture 3 15
Term 2, 2024/2025 DSA211: Lecture 3 16
Term 2, 2024/2025 DSA211: Lecture 3 17
Term 2, 2024/2025 DSA211: Lecture 3 18
 Daily Return Rates of SingTel ([Link]) shares from 1 August 2017 to 29 July 2019
(501 data points)

R1=(3.3-3.3)/3.3
R2=(3.3-3.363)/3.363

Term 2, 2024/2025 DSA211: Lecture 3 19

 Import Dataset .csv
• Put all data files in one particular folder in your computer
• Use the following way to choose that folder as Working
Directory
• Then, use [Link] to input the data file

Term 2, 2024/2025 DSA211: Lecture 3 20

Term 2, 2024/2025 DSA211: Lecture 3 21
 Introduction of Value at Risk (VaR)
VaR is an attempt to provide a single number that
summarizes the total risk in a portfolio.

It was pioneered by J.P. Morgan in 1990 and has

become widely used by corporate management
and fund managers as well as by financial
institutions.

It is also the measure regulators have chosen to set

capital requirements for market risk, credit risk and
operational risk.

Term 2, 2024/2025 DSA211: Lecture 3 22

 Definition of VaR
• Value at Risk (VaR) is defined as the worst loss expected from holding a
portfolio over a given period of time, given a confidence level 1 -- α. (It is
equivalent to the percentile of the distribution)

• For example, VaR(1 day, 99%)=$10 million.

• It means that the realized daily losses will, on average, be higher than $10
million in only one day every 100 trading days.

• Or, we can make a statement such as: The maximum loss over one day is
about $10 million at the 99% confidence level.

• It states how likely the actual losses will exceed the VaR.

• It is designed to measure risk over a short period of time.

• VaR is therefore a function of two parameters: the time horizon and the
confidence level.
Term 2, 2024/2025 DSA211: Lecture 3 23
 VaR and regulatory capital

• Regulators set the capital level of banks in

Basel III (or insurance companies in Solvency
II) based on VaR.

• For example, the market-risk capital is 3 times

the 10-day 99% VaR.

Term 2, 2024/2025 DSA211: Lecture 3 24

• Based on the definition of VaR,
VaR = expected profit/loss minus worst case
loss at the 1-α confidence level

• Based on the definition of absolute VaR,

• Absolute VaR (VaR') = - worst case loss at the
1-α confidence level

Term 2, 2024/2025 DSA211: Lecture 3 25

 Equation of VaR
• Based on the definition of VaR,
VaR = E(Profit) minus worst case loss at the 1-α confidence level
• Based on the definition of absolute VaR,
• Absolute VaR (VaR') = - worst case loss at the 1-α confidence level

Term 2, 2024/2025 DSA211: Lecture 3 26

 Historical VaR for an investment
• Historical VaR which is a non-parametric approach, is also called Nonparametric VaR.

• For example, observe 500 daily return rates (501 observations) of SingTel stock from
1 August 2017 to 29 July 2019.

• Rank the daily return rates from the smallest to the largest. Then, use the 1%
percentage return rate (about the fifth smallest rate) to estimate the VaR(1, 0.99) of
an investment of 4,000 shares of DBS on 29 July 2019 with price $3.30 per share.

• Ans: Abs VaR(1, 0.99)=291.64 and VaR(1, 0.99)=291.36

Term 2, 2024/2025 DSA211: Lecture 3 27

Parametric VaR
Assume that the returns have an analytic density function, f(R).

Historical data are used to estimate the unknown parameters of the assumed
distribution.

A portfolio can be assumed to have a normal returns distribution if the portfolio

is fairly well diversified and the returns are sufficiently independent from each
other.

If R is normally distributed with mean µ and standard deviation 𝜎 in the time

horizon H.

Term 2, 2024/2025 DSA211: Lecture 3 28

 SingTel Ans: VaR(1, 0.99)=267.60

Term 2, 2024/2025 DSA211: Lecture 3 29

 Some other models [not tested]
Model 1: Generalized Autoregressive
Heteroscedasticity GARCH(1,1)

Term 2, 2024/2025 DSA211: Lecture 3 30

 Model 2: Extreme Value Theory

Term 2, 2024/2025 DSA211: Lecture 3 31

 Model 3: Option Pricing Model

Term 2, 2024/2025 DSA211: Lecture 3 32

 Model 4: n-year Endowment Insurance Model

Term 2, 2024/2025 DSA211: Lecture 3 33

 Model 5: Pricing Model of Home Reversion
Products for HDB’s Flat Owners

Term 2, 2024/2025 DSA211: Lecture 3 34