Welcome to AFM 323:

Quantitative Foundations for

Finance
Week 3
Learning Objectives
 Shape characteristics of probability distributions
 Normal and lognormal distributions

Note: I would like you to understand the underlying intuition behind the Math/Stats. I don’t expect you to work
through the Math, but I would like you to work through the examples.

2
Shape Characteristics of a Probability
Distribution
 We continue our discussion of the four most important shape characteristics of a probability distribution of financial
asset returns, which is a random variable (rv).
 For instance, we want to know:
 Where the distribution of the returns is centered
 how spread out the distribution of the return is about the central value
 if the distribution of the return is symmetric about the center or if the distribution of the return has a long left or right tail
 what is the likelihood of observing extreme values for stock returns representing market crashes?
 Four important shape characteristics of a probability distribution of financial asset returns, which is a random variable
(rv)
 Expected value (mean), which measures the center of mass of a distribution of a random variable
 Variance and standard deviation, which measures the spread about the mean a distribution of a random variable
 Skewness, which measures the symmetry about the mean of a distribution of a random variable
 Kurtosis, which measures the tail thickness of a distribution of a random variable.

3
Expected Value or Mean
 Denote the expected value (or mean or average) of a rv X by a constant which we label as i.e.,
 It measures the center of mass of the pdf of X
 What is the intuition of the expected value of a rv X?
 It gives the average value that the rv X takes for a given historical sample.

 How do you calculate the expected value of a discrete rv X?

 For a discrete random variable, this expected value is defined (in terms of the population mean) as a probability
weighted average of n possible values of X lying in the sample space SX
or
 for all in the sample space SX assuming that we have n states of the world.

 Thus,

4
Expected Value or Mean
 How do you calculate the expected value of a continuous rv X?
 For a continuous random variable X with a pdf f(x), the expected value is defined (also in terms of population mean)
as

 It is obtained by replacing the sum over all x in the sample space SX (in the case of the discrete rv) with an integral
from to
 and a probability distribution function, p(x) (in the case of the discrete rv) with a probability density function, f(x).

5
Expected Value or Mean – Discrete
Example
 We use a discrete distribution for the return on Suncor stock with 5 states

 We can calculate the expected value return as

 Which is (-0.3 x 0.05) + (0 x 0.2) + (0.10 x 0.50) + (0.2 x 0.2) + (0.5 x 0.5) = 0.10
(-0.3 x 0.05) + (0 x 0.2) + (0.10 x 0.50) + (0.2 x 0.2) + (0.5 x 0.5) = 0.10

6
Expected Value or Mean – Continuous
Example
 For a continuous rv X, suppose that X has a standard normal distribution, i.e., suppose that

 Then the expected value of this rv is

 We can show the calculation of this by solving the integral, but we won’t do it in this course.

 Remember that the standard normal distribution is centered at zero.

7
Variance of a Random Variable
 Denote the variance of a rv X as
 Variance measures the average deviation of X from its mean.
 Then we define the variance of a random variable X, as:
 Often it can conveniently be calculated by noting that
 What is the intuition of variance of a rv X?
 i.e., it tells us the extent of uncertainty of movements of X on average (for a given historical sample) away from its mean.
 How to calculate the variance of a rv X?
 For a discrete rv X, Var(X) is calculated (in terms of a population variance) as a probability weighted average of n
possible values of the squared deviations of each realized value x from its mean:

or we can use the above two formulas, but they require intermediate steps to calculate x squared and mu squared
 For a continuous rv X, the variance is calculated again simply by replacing the sum by an integral and a probability
distribution functions by a probability density function:

8
Standard Deviation of a Random Variable
 Since represents an average squared deviation, it is not in the same units as X.
 Define a standard deviation of a random variable X as the square root of the variance of the random variable X

 Since Var (X) is in squared units of X, SD(X) is in the same units as X. As a result, SD(X) is easier to interpret than the
variance.
 SD(X) is often used in finance and financial accounting as a measure of risk or uncertainty (more precisely, volatility)
associated with the rv X, which in this case is our financial asset return.
 Strictly speaking, risk is not the same as uncertainty, but in finance we often assume that they are the same.
 risk is present when future events occur with measurable probability which is what we deal with in finance
 uncertainty is present when the likelihood of future events is indefinite or incalculable.

9
Variance Example – Discrete RV
 Calculate the variance of a discrete rv with 5 states:

 Recall our E(X) or Mean from the previous calculation = 0.10

10
Variance Example – Continuous RV
 Calculating the variance of a continuous rv X, where .

 Here, = 0 and it can be shown that

 And hence,

11
The General Normal Distribution
 First, we write down a notation to say a random variable X is distributed as normal with mean and variance

 Then we write down the probability density function (pdf) of a continuous rv X, denoted as f(x), as

 Centered at Symmetric about

 Note: in some statistics textbooks, the results are presented with a strict equality sign, which is incorrect!
 Note says that, for a general normal random variable, about 95% of the time we expect to see
values within ± 2 standard deviations from its mean.
 Note says that, for a general normal random variable, about 99% of the time we expect to see
values within ± 3 standard deviations from its mean - observations more than 3 standard deviations from the mean are very unlikely.
 The pdf of the general normal distribution is completely determined by values of its mean and variance .

12
Finding Areas under a General Normal
Curve
 Excel

 If cumulative = true: a CDF of the rv X is computed.

 If cumulative = false, a pdf of the rv X
is computed

 A α % quantile of the rv (X) is computed.

13
Application to Returns
 Why is a normal distribution not appropriate for simple returns?
 To see this, consider a simple annual return on an asset
 Assume

 Since asset prices must be nonnegative, Rt must always be greater than or equal -1:
 However, the normal distribution is defined for - ∞ ≤ Rt ≤ ∞ and based on the assumed normal distribution, we can use
Excel to calculate
 This says that there is a 1.8% chance that Rt is smaller than -1.
 The above result implies that there is a 1.8% chance that the asset price at the end of the year will be negative!
 This is simply impossible given the definition of the simple return.
 This is why the normal distribution is not appropriate for simple returns.

14
Application to Returns
 However, consider the cc annual return on an asset, rt = ln(1+Rt)
 Then, = simple return
 Assume
 Unlike the simple return, Rt ,the cc return rt can take on values less than –1.
 In fact, we know that the cc return rt is defined for - ∞ ≤ rt ≤ ∞ -> b/c it is a logarithmic function
 For example, suppose that rt = -2. This implies a simple return of
 So,
 Although the normal distribution allows for values of rt <-1, the implied simple return, i.e. Rt (implied by ert - 1) will
always > -1.
 This is why the normal distribution is appropriate for continuously compounded returns.

15
Lognormal Distributions
 Next we look at a distribution with a longer right tail than the normal distribution.
 This is known as a log-normal distribution.
 If X is normally distributed with a certain mean and variance, then the exponent of X, which we call Y, is log-normally
distributed with certain mean and variance. (Y = eX)
 If
 Then the log-normally distributed random variable Y can be determined from the normally distributed random variable X,
using the transformation Y = eX.
 In this case, we say that Y is log-normally distributed and we write it as
 The pdf of the log-normal distribution for Y can be derived from the normal distribution for X using the change-of-
variables formula from calculus

 The mean, variance of Y are given by

16
Lognormal Distribution Example
 A normal distribution for cc return and a log-normal distribution for gross simple return are given by:
 Let denote the cc monthly return on an asset
 Assume that
 Let Rt = (Pt – Pt-1)/Pt-1 denote the simple monthly return
 The relationship between the two returns are given by
 Since rt is normally distributed, (1 + Rt) is log-normally distributed -> by change of variables
 Note: the distribution of (1 + Rt) is only defined for positive values of (1 + Rt)
 This is appropriate since the smallest value that Rt can take is -1.
 Thus, we have that
 where the mean and variance of the gross simple return are:

17
Graphs

18
Skewness of a RV
 Skewness is a measure of symmetry
 Then define the skewness of X simply as:

 When X is far below its mean, is a big negative number; when X is far above its mean, is a big positive
number.
 Hence, if there are more big values of X below then Skew (X) <0. if more big values of X above then Skew (X)>
0
 If X has a symmetric distribution then Skew (X) = 0 since positive and negative values in the expression for Skew (X)
cancel out.
 If Skew (X) > 0 then the distribution of X has a “long right tail”.
 If Skew (X) < 0 the distribution of X has a “long left tail”.
Skewness of a RV
 Skew(X) = 0 ⇒ If X has a symmetric
distribution about then mode =
median = mean
 Skew(X) > 0 ⇒ Its PDF has a long right tail,
and, in this case, then mode < median <
mean
 Skew(X) < 0 ⇒ Its PDF has a long left tail,
and, in this case, then mode > median >
mean

 Mode is the value x at which its

probability density function has its
maximum value, the peak.
Calculating Skewness
 How to calculate the skewness of X?
 For a discrete rv X, Skew(X) is calculated as the sum of probabilities weighted by the standardized deviation of each
realized values from its mean raised to the power of 3:

 for all x1 …….xn in the sample space SX assuming that we have n states of the world.

 For a continuous rv X, the skewness is calculated by replacing the sum by an integral and the probability distribution
functions by probability density functions:
Skewness – Examples
 For a discrete rv with 5 states,

 Continuous random variable – normal distribution

 This result is expected since the normal distribution is symmetric about its mean value .
Kurtosis of a RV
 Kurtosis as a measure of thickness of the tail
 Then define the kurtosis of X simply as:

 Since kurtosis is based on deviations from the mean raised to the fourth power, large deviations get lots of weight
 Hence, distributions with large kurtosis values are ones where there is the possibility of extreme values
 In contrast, if the kurtosis is small then most of the observations are tightly clustered around the mean and there is very
little probability of observing extreme values
 So, values of X far from the mean get blown up resulting in large values of kurtosis
 There are two extreme cases: fat tails (referring to large values of kurtosis) and thin tails (referring to small kurtosis)
Calculating Kurtosis
 For a discrete rv X, Kurt(X) is calculated as the sum of probabilities weighted by the standardized deviation of each
realized value from its mean raised to the power of 4:

 for all x1 …….xn in the sample space SX assuming that we have n states of the world.

 For a continuous rv X, the kurtosis of X is calculated by replacing the sum by an integral and probability distribution
functions by probability density functions:
Interpreting Kurtosis of a Continuous RV
 Sometimes the kurtosis of a random variable is described relative to the kurtosis of a normal random variable
 This relative value of kurtosis is referred to as excess kurtosis and is defined as Excess kurtosis = Kurt(X) - 3
 If Excess kurtosis (X) = 0 then X has the same tail thickness as those of a normal distribution – mesokurtic

 If Excess kurtosis (X) > 0 then X has fatter tails than those of a normal distribution – leptokurtic.

 If Excess kurtosis (X) < 0 then X has thinner tails than those of a normal distribution – platykurtic.

 Let us superimpose a fail-tailed distribution onto a normal distribution for a comparison purpose:

 Leptokurtic distribution (black) – fat tails,

kurtosis >3
 Mesokurtic distribution (red) – normal
distribution, kurtosis = 3
 Platykurtic distribution (green) – thin tail,
kurtosis < 3.
Linear Functions of a Random Variable
 Let X be a discrete or continuous rv with and .
 Let a and b be some known constants.
 Define a new random variable Y to be a linear function of X: Y = a.X + b
 a and b are known constants
 Then the expected value or mean of Y is given by:
 But the expected value of a sum is a sum of expected values; so we have:
 Next, the expectation of a constant is constant itself; so we have

 And

 So, we see that expectations is a linear operator in the following sense:

Linear Functions of a Random Variable
 Next, consider
 But we know that
 Where: (due to definition of a variance); Var (b) = 0 (since there is no uncertainty w.r.t. a constant)
 (there is no uncertainty between a rv and a constant)
 So, we have

 Adding a constant to X does not affect its variance, and the effect of multiplying X by the constant a simply increases the
variance of X by the square of a.
 A normal random variable has the special property that a linear function of it is also a normal random variable.
 The following proposition establishes the result.
 PROPOSITION. Let . Consider a linear function, Y = aX + b as before. Then
where and
 The proof of the result relies on the change-of-variables formula for determining pdf of a function of a random variable.
 Result may or may not hold for random variables whose distributions are not normal.
Standardizing a Normal RV
 Standardizing a normal random variable with mean and variance .
 Consider a normal rv X with nonzero mean and non-unity variance
 Then we can standardize the rv X, so that the resulting rv Z has 0 mean and unit variance, i.e. Z is a standard normal
N(0,1) rv.
 We center X from its mean and scale by its SD:
 So we have:
 Hence, standardization creates a new random variable, Z, with mean = 0 and variance = 1.
Intuition Behind Standardized Normal RV
 Given, we “demean” the variable by its mean to obtain a variable centered from its mean

 We “scale” by the square root of the variance to obtain

 We denote this “standardized” variable as Z, i.e.,

 A non-standard random variable X with mean and variance can be created from a standard random variable
via a linear transformation:
Example
 Let X ~ N(2, 4).

 We want to find Pr (X>5)

 But we only know the probabilities associated with a standard normal random variable Z ~ N(0, 1).

 We solve as follows:

Pr
Pr(X
(X>>5)
5)==Pr
Pr[[(X-2)/2
(X-2)/2>>(5-2/2)
(5-2/2)]]==Pr
Pr(Z
(Z>>1.5)
1.5)
Pr
Pr(Z
(Z>>1.5)
1.5)==1-
1-Pr
Pr(Z
(Z<<1.5)
1.5)==11––NORMDIST(1.5,0,1,TRUE)
NORMDIST(1.5,0,1,TRUE)->
->TRUE
TRUEfor
forCDF
CDF
==11––0.933
0.933
==0.0668
0.0668==6.68%
6.68%
Value-at-Risk (VaR)
 VaR is a method used to assess financial risk

 VaR summarizes the worst expected loss over a given time horizon that, under normal market conditions, will not be
exceeded within a given confidence interval.

 Example
 a bank issues a statement: the daily VaR of its trading portfolio is $10 billion at the 99 percent confidence level

 This means there is only 1 chance in a 100 (or 1-0.99 = 1%), under normal market conditions for the bank to suffer a

daily loss greater than $10 billion on average. This single number summarizes the bank’s exposure to market risk.

 Based on the bank’s bottom line and capitalization, managers can assess the risk and decide if they should reduce the risk.
VaR Calculations for Simple Returns
 Consider a $10,000 investment in Suncor for 1 month
 Let R denote the monthly simple return on Suncor.
 Assume where =.05 and = 0.10.
 We mentioned before Normal distribution is not great for simple returns, but we use it here for demonstration purposes
 How much can we lose on average over a one-month period under normal market conditions with a specified
probability level α ?
 This single number is known as Value at Risk (VaR).
 Let’s look at the following questions:
1. What is the probability distribution of the end-of- month wealth, W1?
2. What is the probability that the end of month wealth is less than $9,000?
3. What value of Rt produces $9,000?
4. What is the monthly VaR on the $10,000 investment with 5% probability, i.e., how much can we lose, on average, if
Rt < q0.05?
VaR Q(1)
 What is the probability distribution of end-of-month wealth, W 1?
 From
 W1 is related to initial wealth, W0 and so W1 = 10,000 (1+R) since W0 = 10,000
 But R is assumed to be a normally distributed random variable.
 By our previous results, we know that W1 = 10,000 (1+R) being a linear function of normally distributed R, is also
normally distributed.
 W1 is normally distributed with mean which is E[W1] = $10,000(1 +E[R])
 and variance:

 We have E[W1] = $10,000*(1 +.05) = $10,500, and Var[W1] = (W0)2 Var (R) = ($10,000)2 (0.10)2 = 1,000,000
 SD[W1] = $1,000.
 So W1 is normally distributed with mean 10,500 and variance 1,000,000, i.e.
VaR Q(1)
VaR Q(2)
 What is the probability that the end of month wealth is less than $9,000, i.e., Pr (W1 <$9,000) ?

 Using the result that

we use Excel to calculate

Pr
Pr(W
(W11<<$9,000)
$9,000)==NORMDIST(9000,10500,
NORMDIST(9000,10500,1000,
1000,True)
True)==0.067
0.067
VaR Q(3)
 What value of Rt produces $9,000?

 To find R that produces W1 = $9,000, we solve

W1 = 10,000(1+ R) or R = (9.000/10,000)
(9.000/10,000)––11==-0.1
-0.1==-10%
-10%
VaR Q(4)
 What is the monthly VaR on the $10,000 investment with 5% probability, i.e. how much can we lose, on average, if
Rt < q0.05? HERE WE FIND THE VAR
 q0.05 Represents the quantile or return value at which the probability of being BELOW that value is 5%
 So when we ask this question we are asking what is the return such that there is a 5% probability of being below that
return?

 How much can we lose, on average, if R < q0.05 under normal market conditions?

 We use and solve for the 5% quantile?

 Using Excel:
Bringing it all Together
 There is a 5% probability that the return is less than -11.4%
-11.4% on average.

 If R = -11.4%
-11.4% , then the average monthly loss in investment value $1,140
$1,140 (5% VaR) -> -11.4% x 10,000
(initial)
$1,140
$1,140
 Hence, is the 5% VaR over the next month on the $10,000 investment in Suncor. (VaR is always a loss but
is represented as a positive number by convention.)
 We finally are in the position to state: There is a 5% chance that we suffer an average monthly loss greater than $1,140
under normal market conditions on our investment in Suncor.
 Because we said that there is a 5% probability that return is less than -11.4% -> 5% probability that loss is worse than this amount
 This single number summarizes our exposure to Suncor’s market risk.
 In general, the x 100% Value-at-Risk (VaRα) for an initial investment of W0 is computed as

 i.e., represents the dollar loss that could occur on average with probability under normal market conditions.
Bringing it all Together
VaR for Continuously Compounded Returns
 The above calculations illustrate how to calculate VaR using the normal distribution for simple returns.
 As argued previously, the normal distribution is not appropriate for characterizing the distribution of simple returns.
 It may be more appropriate for characterizing the distribution of continuously compounded returns.
 How to calculate VaR for cc returns?
 Recall that r = ln(1+R) is the cc monthly return
 The simple monthly return is
 Assume that
 The distribution of R is log-normal
 Let W0 be the initial investment
VaR for Continuously Compounded Returns
 Steps to compute 100 % VaR using cc returns.

1. Step 1: Compute quantile of a normal distribution for cc r as where is the

quantile of a N(0,1) distribution.
2. Step 2: Convert the quantile for cc r into the quantile for simple R, using the transformation:
3. Step 3: Compute 100 × α% VaR using for simple return:

or

We are doing the same thing conceptually in terms of finding the quantile -> the only difference is because we have a
lognormal distribution now, we need to define a formula for the quantile in a different manner and then convert it into the
quantile for the Simple Return.
VaR for Continuously Compounded Returns - Example
 Compute 5% VaR assuming that

 The 5% cc return quantile is = 0.05

0.05+0.10(-1.645)
+0.10(-1.645)==-0.114
-0.114
Where
Wherez(0.05)
z(0.05)==NORM.INV(0.05,0,1)
NORM.INV(0.05,0,1)==-1.645
-1.645

 So the 5% simple return quantile is

 Then the 5% VaR based on a $10,000 initial investment is VaR 0.05 = $10,000(-0.108) = -$1,080

 i.e., There is a 5% chance that we suffer an average monthly loss greater than $1,080 under normal market conditions
on our investment in Suncor.
Wrap Up & Next Class

43
Now What for Week 4?
Week 4 Focus: Probability Continued
• Joint distributions
• Marginal distributions
• Bivariate distributions… and more

Problem Sets:
• Problem Set 2 – Probability now available on Learn (attempt to complete it to test your understanding)
• Review the Probability Review Part 2 and Probability Review Part 3 excel file on Learn for the sample
calculations

Assignments Due:
• Assignment 1 – Asset Returns, will try to get feedback to you as soon as possible
• Please review the assigned stock information (under the Admin folder on Learn) to see what stock you
have been assigned. Note that you will stick with this stock to complete all assignments in the course

Projects & Presentations:

• None

44