Risk Measures Value at Risk - one asset

Risk Measures - Value at Risk (VaR)

Value at Risk (VaR) for an investment in one asset.

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 33 / 140
 Risk Measures Value at Risk - one asset

Asset return

Definition 3.1 (Asset return)

The return of an asset for a given period is the percentage change in
wealth that results from holding this asset.
The return is given by:

Pt+1 − Pt + Dt+1
Rt+1 =
Pt
where Pt is the price at time t and Dt are the dividends.

The return of one asset is in general a random variable because, we

know the price at which we trade the asset today, Pt , but we know neither
the price, Pt+1 , nor the dividend, Dt+1 that will be payed tomorrow.
Therefore, there is uncertainty about the asset return because Pt+1 and
Dt+1 are both random variables.
Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 34 / 140
 Risk Measures Value at Risk - one asset

VaR - Aplications

• The minimal required capital (MRC) for regulatory purposes is

calculated with Article 364 of the Regulation (EU) No 575/2013 of
the European Parliament and of the Council of 26 June 2013 on
prudential requirements for credit institutions and investment firms:
60
( )
10 mc X 10
M RC = max V aRt (99%), V aRt−i (99%) +
60
i=1
60
( )
m s
X
+ max sV aRt10 (99%), 10
V aRt−i (99%)
60
i=1

• Management tool to define internal limits in the asset management.

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 35 / 140
 Risk Measures Value at Risk - one asset

Value at Risk - VaR

Consider an asset with uncertain future financial returns denoted by R,
and let f (R) be the densitity function of the future returns.

Z R∗
∗
α = P (R ≤ R ) = f (R)dR
−∞

With probability (1 − α) the return is not less then R∗ .

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 36 / 140
 Risk Measures Value at Risk - one asset

Value at Risk - VaR

Let α be the significance level, 1 − α the confidence level and W the
invested wealth.

VaR is defined using R∗ ,

Z R∗
α = P (R ≤ R∗ ) = f (R)dR
−∞

which is the quantile α of the distribution.

Definition 3.2
VaRd (1 − α) = −W R∗ , that is, with 100(1 − α)% of confidence the loss
will not exceed −W R∗ during the period d.

Interpretation: VaR is the maximum loss, which should not be exceeded

during a specified period of time with a given probability level.
(VaR10 (1 − 0.05) = 1000).
Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 37 / 140
 Risk Measures Value at Risk - one asset

Normal VaR

If f (R) is the density of the normal distribution with mean µ and standard
deviation σ, that is, R ∼ N (µ, σ 2 ) then,

1 (R−µ)2
f (R) = √ e− 2σ2 .
2πσ

Let us compute R∗ ,
Z R∗ Z R∗
(R−µ)2
1
α= f (R) dR = e− 2σ2 dR =
√
−∞ −∞ 2πσ
R∗ −µ  ∗ 
R −µ
Z
σ 1 − x2
= √ e 2 dx = Φ
−∞ 2π σ
where Φ is the cumulative distribution of the normal standard (0, 1).

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 38 / 140
 Risk Measures Value at Risk - one asset

Normal VaR

R∗ − µ
 
α=Φ
σ

Since Φ is increasing we can compute the inverse and then,

R∗ − µ
= Φ−1 (α) ⇔
σ

R∗ = µ + σΦ−1 (α)

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 39 / 140
 Risk Measures Value at Risk - one asset

Normal VaR - Example

An investor wants to invest 1 million euros on the PSI 20 asset. What is

the maximum loss over a one-week horizon?
The mean of the returns over one-week is 0 and the standard deviation is
2.5%. What is the normal VaR with a 99% confidence level?

Φ−1 (1 − 0.99) = −2.33.

VaR= −Φ−1 (1 − 0.99)W σ = 2.33 ∗ 1 million ∗0.025 = 58.158 euros.

Remarks:
• The VaR increases with the wealth invested.
• The VaR also increases with the standard deviation, that is, with the
volatility of the asset.

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 40 / 140
 Risk Measures Value at Risk - one asset

Normal VaR - Unknown parameters

When the returns follow a normal distribution but the parameters µ and
σ 2 are unknown, we can use a sample to estimate them,

T T
1X 2 1 X
µ̂ = Rt and σ̂ = (Rt − µ̂)2
T T −1
t=1 t=1

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 41 / 140
 Risk Measures Value at Risk - one asset

Historical VaR - Empirical distribution function

To compute the historical VaR we use the empirical distribution of a data

set.

Given a sample of a random variable, x1 , x2 , ..., xn . The empirical

distribution function associated to this sample is the distribution function
whose probability mass function assigns the same weight to all the
elements of the sample, that is,

#{ i : xi ≤ x}
F̂ (x) = .
n

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 42 / 140
 Risk Measures Value at Risk - one asset

Empirical distribution function - Example

Sample=c(1,1,1,1,1,1,1,2,2,3,3,3,3,3,3,3,4,5,7,7,7,7,7,7).
(24 observations)

hist(amostra, probability = TRUE) - Draws the probability mass function

graph.

plot(ecdf(x)) - Draws the empirical distribution function graph.

7 2 7
P (X ≤ 3) = F̂ (3) = p(1) + p(2) + p(3) = 24 + 24 + 24 = 0.6(6).
Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 43 / 140
 Risk Measures Value at Risk - one asset

Historical VaR - Empirical distribution function

R∗ = R[int(αn)]
when the sample is sorted in ascending order.

If αn is an integer number:

αn
P̂ R ≤ R[αn] = F̂ R[αn] = = α.
n

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 44 / 140
 Risk Measures Value at Risk - one asset

Temporal aggregation
Consider an asset with price Pt at time t.
The simple return between times t and t + 1 is defined as:
Pt+1 − Pt Pt+1
Rt,t+1 = = − 1.
Pt Pt

The log return is defined as

 
Pt+1
rt,t+1 = ln .
Pt

When
Pt+1
∼ 1 then Rt,t+1 ∼ rt,t+1
Pt
(Prove using the Taylor formula).
Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 45 / 140
 Risk Measures Value at Risk - one asset

Temporal aggregation of log returns

 
Pt+k
rt,t+k = ln
Pt
   
Pt+k Pt+1 Pt+2 Pt+k
rt,t+k = ln = ln ··· =
Pt Pt Pt+1 Pt+k−1
     
Pt+1 Pt+2 Pt+k
= ln + ln + · · · ln ,
Pt Pt+1 Pt+k−1
that is,
rt,t+k = rt,t+1 + rt+1,t+2 + · · · + rt+k−1,t+k .

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 46 / 140
 Risk Measures Value at Risk - one asset

Temporal aggregation of log returns

If we assume that the returns are i.i.d with expected value µday and
2 then the expected value for k days is given by,
variance σday

µk days = E(rt,t+k ) = E [(rt,t+1 ) + E(rt+1,t+2 ) + · · · + E(rt+k−1,t+k )] =

= E(rt,t+1 ) + E(rt,t+1 ) + · · · + E(rt,t+1 ) = kµday

and the standard deviation for k days is given by,

q q
σk days = V (rt,t+k ) = V (rt,t+1 ) + V (rt+1,t+2 ) + · · · + V (rt+k−1,t+k ) =
p √
= V (rt ) + V (rt ) + · · · + V (rt ) = kσday .

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 47 / 140
 Risk Measures Value at Risk - one asset

Normal and Historical VaR

Exercise 3.1
Choose a stock from the Yahoo Finance, download the daily prices for one
year. Compute the daily log returns. Consider an investment of 10 000
euros in that stock.

(a) Compute the daily normal VaR.

(b) Compute the one week normal VaR.
(c) Compute the daily historical VaR.
(d) Compute the one week historical VaR.

Marta Faias (NOVA FCT - DM) Teoria do Investimento Ano letivo 2024/2025 48 / 140