© CFA Institute. For candidate use only. Not for distribution.

Portfolio Risk Measures: Applications of the Normal Distribution 169

portfolio for which E(RP) − RL is largest relative to standard deviation minimizes P(RP
< RL). Therefore, if returns are normally distributed, the safety-first optimal portfolio
maximizes the safety-first ratio (SFRatio), as follows:
SFRatio = [E(RP) − RL]/σP. (9)
The quantity E(RP) − RL is the distance from the mean return to the shortfall level.
Dividing this distance by σP gives the distance in units of standard deviation. When
choosing among portfolios using Roy’s criterion (assuming normality), follow these
two steps:
1. Calculate each portfolio’s SFRatio.
2. Choose the portfolio with the highest SFRatio.
For a portfolio with a given safety-first ratio, the probability that its return will
be less than RL is Normal(–SFRatio), and the safety-first optimal portfolio has the
lowest such probability. For example, suppose an investor’s threshold return, RL, is 2
percent. He is presented with two portfolios. Portfolio 1 has an expected return of 12
percent, with a standard deviation of 15 percent. Portfolio 2 has an expected return
of 14 percent, with a standard deviation of 16 percent. The SFRatios, using Equation
9, are 0.667 = (12 − 2)/15 and 0.75 = (14 − 2)/16 for Portfolios 1 and 2, respectively.
For the superior Portfolio 2, the probability that portfolio return will be less than 2
percent is N(−0.75) = 1 − N(0.75) = 1 − 0.7734 = 0.227, or about 23 percent, assuming
that portfolio returns are normally distributed.
You may have noticed the similarity of the SFRatio to the Sharpe ratio. If we sub-
stitute the risk-free rate, RF, for the critical level RL, the SFRatio becomes the Sharpe
ratio. The safety-first approach provides a new perspective on the Sharpe ratio: When
we evaluate portfolios using the Sharpe ratio, the portfolio with the highest Sharpe
ratio is the one that minimizes the probability that portfolio return will be less than
the risk-free rate (given a normality assumption).

EXAMPLE 3

The Safety-First Optimal Portfolio for a Client

You are researching asset allocations for a client in Canada with a CAD800,000
portfolio. Although her investment objective is long-term growth, at the end
of a year, she may want to liquidate CAD30,000 of the portfolio to fund edu-
cational expenses. If that need arises, she would like to be able to take out the
CAD30,000 without invading the initial capital of CAD800,000. Exhibit 11 shows
three alternative allocations.
​

Exhibit 11: Mean and Standard Deviation for Three Allocations (in
Percent)
​

Allocation A B C

Expected annual return 25 11 14

Standard deviation of return 27 8 20
​

Address these questions (assume normality for Questions 2 and 3):

 © CFA Institute. For candidate use only. Not for distribution.
170 Learning Module 5 Portfolio Mathematics

1. Given the client’s desire not to invade the CAD800,000 principal, what is the
shortfall level, RL? Use this shortfall level to answer question 2.
Solution:
Because CAD30,000/CAD800,000 is 3.75 percent, for any return less
than 3.75 percent the client will need to invade principal if she takes out
CAD30,000. So, RL = 3.75%.

2. According to the safety-first criterion, which of the three allocations is the

best?

(Hint, to decide which of the three allocations is safety-first optimal, select

the alternative with the highest ratio [E(RP) − RL]/σP.)
A. 0.787037 = (25 − 3.75)/27
B. 0.90625 = (11 − 3.75)/8
C. 0.5125 = (14 − 3.75)/20
Solution:
B is correct. Allocation B, with the largest ratio (0.90625), is the best alterna-
tive according to the safety-first criterion.

3. What is the probability that the return on the safety-first optimal portfolio
will be less than the shortfall level?
Solution:
To answer this question, note that P(RB < 3.75) = Normal(−0.90625). We can
round 0.90625 to 0.91 for use with tables of the standard normal CDF. First,
we calculate Normal(−0.91) = 1 − Normal(0.91) = 1 − 0.8186 = 0.1814, or
about 18.1 percent. Using a spreadsheet function for the standard normal
CDF on −0.90625 without rounding, we get 0.182402, or about 18.2 percent.
The safety-first optimal portfolio has a roughly 18 percent chance of not
meeting a 3.75 percent return threshold. This can be seen in the following
graphic, in which Allocation B has the smallest area under the distribution
curve to the left of 3.75 percent.
0.10
Target Return = 3.75%
0.08
B
0.06

0.04 C

0.02 A

0
–80.3 –56.3 –32.3 –8.3 15.8 39.8 63.8 87.8
Percent
Several points are worth noting. First, if the inputs were slightly different,
we could get a different ranking. For example, if the mean return on B
were 10 percent rather than 11 percent, Allocation A would be superior to
B. Second, if meeting the 3.75 percent return threshold were a necessity
rather than a wish, CAD830,000 in one year could be modeled as a liability.
Fixed-income strategies, such as cash flow matching, could be used to offset
or immunize the CAD830,000 quasi-liability.
 © CFA Institute. For candidate use only. Not for distribution.
Portfolio Risk Measures: Applications of the Normal Distribution 171

In many investment contexts besides Roy’s safety-first criterion, we use the normal
distribution to estimate a probability. Another arena in which the normal distribution
plays an important role is financial risk management. Financial institutions, such as
investment banks, security dealers, and commercial banks, have formal systems to
measure and control financial risk at various levels, from trading positions to the overall
risk for the firm. Two mainstays in managing financial risk are value at risk (VaR) and
stress testing/scenario analysis. Stress testing and scenario analysis refer to a set
of techniques for estimating losses in extremely unfavorable combinations of events
or scenarios. Value at risk (VaR) is a money measure of the minimum value of losses
expected over a specified time period (e.g., a day, a quarter, or a year) at a given level
of probability (often 0.05 or 0.01). Suppose we specify a one-day time horizon and a
level of probability of 0.05, which would be called a 95 percent one-day VaR. If this
VaR equaled EUR5 million for a portfolio, there would be a 0.05 probability that the
portfolio would lose EUR5 million or more in a single day (assuming our assumptions
were correct). One of the basic approaches to estimating VaR, the variance–covariance
or analytical method, assumes that returns follow a normal distribution.

QUESTION SET

A client has a portfolio of common stocks and fixed-income instru-

ments with a current value of GBP1,350,000. She intends to liquidate
GBP50,000 from the portfolio at the end of the year to purchase a partnership
share in a business. Furthermore, the client would like to be able to withdraw the
GBP50,000 without reducing the initial capital of GBP1,350,000. The following
table shows four alternative asset allocations.

Mean and Standard Deviation for Four Allocations (in Percent)

​

A B C D

Expected annual return 16 12 10 9

Standard deviation of return 24 17 12 11
​

1. Address the following questions (assume normality for Parts B and C):

A. Given the client’s desire not to invade the GBP1,350,000 princi-

pal, what is the shortfall level, RL? Use this shortfall level to answer
Question 2.
B. According to the safety-first criterion, which of the allocations is the
best?
C. What is the probability that the return on the safety-first optimal port-
folio will be less than the shortfall level, RL?
Solution:

A. Because GBP50,000/GBP1,350,000 is 3.7 percent, for any return less

than 3.7 percent the client will need to invade principal if she takes out
GBP50,000. So RL = 3.7 percent.
B. To decide which of the allocations is safety-first optimal, select the
alternative with the highest ratio [E(RP) − RL]/σP:
Allocation 1 0.5125 = (16 – 3.7)/24.
Allocation 2 0.488235 = (12 – 3.7)/17.
 © CFA Institute. For candidate use only. Not for distribution.
172 Learning Module 5 Portfolio Mathematics

Allocation 3 0.525 = (10 – 3.7)/12.

Allocation 4 0.481818 = (9 – 3.7)/11.
Allocation C, with the largest ratio (0.525), is the best alternative
according to the safety-first criterion.
C. To answer this question, note that P(RC < 3.7) = N(0.037 – 0.10)/0.12)
= Normal(−0.525). By using Excel’s NORM.S.DIST() function, we get
NORM.S.DIST((0.037 – 0.10)/0.12) = 29.98%, or about 30 percent. The
safety-first optimal portfolio has a roughly 30 percent chance of not
meeting a 3.7 percent return threshold.

2. A client holding a GBP2,000,000 portfolio wants to withdraw GBP90,000

in one year without invading the principal. According to Roy’s safety-first
criterion, which of the following portfolio allocations is optimal?
​

Allocation A Allocation B Allocation C

Expected annual return 6.5% 7.5% 8.5%

Standard deviation of returns 8.35% 10.21% 14.34%
​

A. Allocation A
B. Allocation B
C. Allocation C
Solution:
B is correct. Allocation B has the highest safety-first ratio. The threshold
return level, RL, for the portfolio is GBP90,000/GBP2,000,000 = 4.5 percent;
thus, any return less than RL = 4.5% will invade the portfolio principal. To
compute the allocation that is safety-first optimal, select the alternative with
the highest ratio:

_ [​ E​(​RP
​  ​​ − ​RL​  )​​ ]​ ​
​​  ​σ​ P ​
​​ .​

6.5 − 4.5
​Allocation A = ​ _
8.35 ​  = 0.240.​

7.5 − 4.5
​Allocation B = ​ _
10.21 ​  = 0.294.​

8.5 − 4.5
​Allocation C = ​ _
14.34 ​  = 0.279.​
 © CFA Institute. For candidate use only. Not for distribution.
References 173

REFERENCES
Roy, A. D. 1952. “Safety First and the Holding of Assets.” Econometrica20 (3): 431–49.
10.2307/1907413
 © CFA Institute. For candidate use only. Not for distribution.
174 Learning Module 5 Portfolio Mathematics

PRACTICE PROBLEMS
1. An analyst produces the following joint probability function for a foreign index
(FI) and a domestic index (DI).

RDI = 30% RDI = 25% RDI = 15%

RFI = 25% 0.25

RFI = 15% 0.50
RFI = 10% 0.25

The covariance of returns on the foreign index and the returns on the domestic
index is closest to:
A. 26.39.

B. 26.56.

C. 28.12.
 © CFA Institute. For candidate use only. Not for distribution.
Solutions 175

SOLUTIONS
1. B is correct. The covariance is 26.56, calculated as follows. First, expected returns
are
E(RFI) = (0.25 × 25) + (0.50 × 15) + (0.25 × 10)

= 6.25 + 7.50 + 2.50 = 16.25 and

E(RDI) = (0.25 × 30) + (0.50 × 25) + (0.25 × 15)

= 7.50 + 12.50 + 3.75 = 23.75.

Covariance is

Cov(RFI,RDI) = ∑
​​ ​ ​​ ​∑ ​ ​​  P​(​RFI,i ​  ​​)​​(​RFI,i
​  ​​, ​RDI,j ​  ​​)​​(​RDI,j
​  ​​ − E ​RFI ​  ​​)​​
​  ​​ − E ​RDI
i j

= 0.25[(25 – 16.25)(30 – 23.75)] + 0.50[(15 – 16.25)(25 – 23.75)] +

0.25[(10 – 16.25)(15 – 23.75)]

= 13.67 + (–0.78) + 13.67 = 26.56.

© CFA Institute. For candidate use only. Not for distribution.
 © CFA Institute. For candidate use only. Not for distribution.

LEARNING MODULE

6
Simulation Methods
by Kobor Adam, PhD, CFA.
Adam Kobor, PhD, CFA, at New York University Investment Office (USA)

LEARNING OUTCOMES
Mastery The candidate should be able to:

explain the relationship between normal and lognormal distributions

and why the lognormal distribution is used to model asset prices
when using continuously compounded asset returns
describe Monte Carlo simulation and explain how it can be used in
investment applications
describe the use of bootstrap resampling in conducting a simulation
based on observed data in investment applications

INTRODUCTION
The understanding and application of probability distributions is a critical component
1
of forecasting financial variables and asset prices. This learning module provides a
foundation for understanding important concepts related to probability distributions.
Regarding the application of probability distributions, this learning module explains
how to construct and interpret a Monte Carlo simulation analysis. Bootstrapping,
with some similarities to Monte Carlo simulations, is also demonstrated to illustrate
the use and application of this statistical sampling approach.

LEARNING MODULE OVERVIEW

■ The lognormal distribution is widely used for modeling the

probability distribution of financial asset prices because the
distribution is bounded from below by 0 as asset prices and usually
describes accurately the statistical distribution properties of financial
assets prices. Lognormal distribution is typically skewed to the right.
■ Continuously compounded returns play a role in many asset pricing
models, as well as in risk management.
 © CFA Institute. For candidate use only. Not for distribution.
178 Learning Module 6 Simulation Methods

■ Monte Carlo simulation is widely used to estimate risk and return in

investment applications. Specifically, it is commonly used to value
securities with complex features, such as embedded options, where no
analytic pricing formula is available
■ A Monte Carlo simulation generates a large number of random sam-
ples from a specified probability distribution or a series of distribu-
tions to obtain the likelihood of a range of results.
■ Bootstrapping mimics the process of performing random sampling
from a population to construct the sampling distribution by treating
the randomly drawn sample as if it were the population.
■ Because a random sample offers a good representation of the popu-
lation, bootstrapping can simulate sampling from the population by
sampling from the observed sample.

2 LOGNORMAL DISTRIBUTION AND CONTINUOUS

COMPOUNDING

explain the relationship between normal and lognormal distributions

and why the lognormal distribution is used to model asset prices
when using continuously compounded asset returns

The Lognormal Distribution

Closely related to the normal distribution, the lognormal distribution is widely used for
modeling the probability distribution of share and other asset prices. For example, the
lognormal distribution appears in the Black–Scholes–Merton option pricing model.
The Black–Scholes–Merton model assumes that the price of the asset underlying the
option is lognormally distributed.
A random variable Y follows a lognormal distribution if its natural logarithm, ln
Y, is normally distributed. The reverse is also true: If the natural logarithm of random
variable Y, ln Y, is normally distributed, then Y follows a lognormal distribution.
If you think of the term lognormal as “the log is normal,” you will have no trouble
remembering this relationship.
The two most noteworthy observations about the lognormal distribution are that it
is bounded below by 0 and it is skewed to the right (it has a long right tail). Note these
two properties in the graphs of the probability density functions (pdfs) of two lognor-
mal distributions in Exhibit 1. Asset prices are bounded from below by 0. In practice,
the lognormal distribution has been found to be a usefully accurate description of the
distribution of prices for many financial assets. However, the normal distribution is
often a good approximation for returns. For this reason, both distributions are very
important for finance professionals.
 © CFA Institute. For candidate use only. Not for distribution.
Lognormal Distribution and Continuous Compounding 179

Exhibit 1: Two Lognormal Distributions

0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

Like the normal distribution, the lognormal distribution is completely described by

two parameters. Unlike many other distributions, a lognormal distribution is defined
in terms of the parameters of a different distribution. The two parameters of a lognor-
mal distribution are the mean and standard deviation (or variance) of its associated
normal distribution: the mean and variance of ln Y, given that Y is lognormal. So,
we must keep track of two sets of means and standard deviations (or variances): (1)
the mean and standard deviation (or variance) of the associated normal distribution
(these are the parameters) and (2) the mean and standard deviation (or variance) of
the lognormal variable itself.
To illustrate this relationship, we simulated 1,000 scenarios of yearly asset returns,
assuming that returns are normally distributed with 7 percent mean and 12 percent
standard deviation. For each scenario i, we converted the simulated continuously
compounded returns (ri) to future asset prices with the formula Price(1 year later)i =
USD1 × exp(ri), where exp is the exponential function and assuming that the asset’s
price is USD1 today. In Exhibit 2, Panel A shows the distribution of the simulated
returns together with the fitted normal pdf, whereas Panel B shows the distribution
of the corresponding future asset prices together with the fitted lognormal pdf. Again,
note that the lognormal distribution of future asset prices is bounded below by 0 and
has a long right tail.
 © CFA Institute. For candidate use only. Not for distribution.
180 Learning Module 6 Simulation Methods

Exhibit 2: Simulated Returns (Normal PDF) and Asset Prices (Lognormal

PDF)
A. Normal PDF

120

100

80

60

40

20

0
–28.76 –18.58 –8.41 1.76 11.93 22.1 32.27 42.45
Yearly Returns (%)
Histogram Normal Fit

B. Lognormal PDF

120

100

80

60

40

20

0
0.75 0.86 0.98 1.09 1.21 1.32 1.44 1.55
Asset Values 1 Year Later (USD 1 Initial Investment)
Histogram Lognormal Fit

The expressions for the mean and variance of the lognormal variable are challenging.
Suppose a normal random variable X has expected value μ and variance σ2. Define Y
= exp(X). Remember that the operation indicated by exp(X) or eX (where e ≈ 2.7183)
is the opposite operation from taking logs. Because ln Y = ln [exp(X)] = X is normal
(we assume X is normal), Y is lognormal. What is the expected value of Y = exp(X)?
A guess might be that the expected value of Y is exp(μ). The expected value is actually
exp(μ + 0.50σ2), which is larger than exp(μ) by a factor of exp(0.50 σ2) > 1. To get some
insight into this concept, think of what happens if we increase σ2. The distribution
spreads out; it can spread upward, but it cannot spread downward past 0. As a result,
the center of its distribution is pushed to the right: The distribution’s mean increases.
The expressions for the mean and variance of a lognormal variable are summarized
below, where μ and σ2 are the mean and variance of the associated normal distribution
(refer to these expressions as needed, rather than memorizing them):
■ Mean (μL) of a lognormal random variable = exp(μ + 0.50σ2).
 © CFA Institute. For candidate use only. Not for distribution.
Lognormal Distribution and Continuous Compounding 181

■ Variance (σL2) of a lognormal random variable = exp(2μ + σ2) × [exp(σ2)

− 1].

Continuously Compounded Rates of Return

We now explore the relationship between the distribution of stock return and stock
price. In this section, we show that if a stock’s continuously compounded return is
normally distributed, then future stock price is necessarily lognormally distributed.
Furthermore, we show that stock price may be well described by the lognormal
distribution even when continuously compounded returns do not follow a normal
distribution. These results provide the theoretical foundation for using the lognormal
distribution to model asset prices.
Showing that the stock price at some future time T, PT, equals the current stock
price, P0, multiplied by e raised to power r0,T, the continuously compounded return
from 0 to T:
PT = P0exp(r0,T).
We showed in an earlier lesson that r0,T, the continuously compounded return to
time T, is the sum of the one-period continuously compounded returns, as follows:
r0,T = rT−1,T + rT−2,T−1 + . . . + r0,1. (1)
If these shorter-period returns are normally distributed, then r0,T is normally distrib-
uted (given certain assumptions) or approximately normally distributed (not making
those assumptions). As PT is proportional to the log of a normal random variable,
PT is lognormal.
A key assumption in many investment applications is that returns are inde-
pendently and identically distributed (i.i.d.). Independence captures the proposition
that investors cannot predict future returns using past returns. Identical distribution
captures the assumption of stationarity, a property implying that the mean and vari-
ance of return do not change from period to period.
Assume that the one-period continuously compounded returns (such as r0,1) are
i.i.d. random variables with mean μ and variance σ2 (but making no normality or other
distributional assumption). Then,
E(r0,T) = E(rT−1,T) + E(rT−2,T−1) + . . . + E(r0,1) = μT, (2)
(we add up μ for a total of T times), and
σ2(r0,T) = σ2T (3)
(as a consequence of the independence assumption).
The variance of the T holding period continuously compounded return is T
multiplied by _ the variance of the one-period continuously compounded return; also,
σ(r0,T) = σ​  ​√T .​​ If the one-period continuously compounded returns on the right-hand
side of Equation 1 are normally distributed, then the T holding period continuously
compounded return, r0,T, is also normally distributed with mean μT and variance
σ2T. This is because a linear combination of normal random variables is also a normal
random variable.
Even if the one-period continuously compounded returns are not normal, their sum,
r0,T, is approximately normal according to the central limit theorem. Now compare P T
= P0exp(r0,T) to Y = exp(X), where X is normal and Y is lognormal (as we discussed
previously). Clearly, we can model future stock price PT as a lognormal random variable
because r0,T should be at least approximately normal. This assumption of normally
distributed returns is the basis in theory for the lognormal distribution as a model
for the distribution of prices of shares and other financial assets.
 © CFA Institute. For candidate use only. Not for distribution.
182 Learning Module 6 Simulation Methods

Continuously compounded returns play a role in many asset pricing models, as

well as in risk management. Volatility measures the standard deviation of the con-
tinuously compounded returns on the underlying asset; by convention, it is stated
as an annualized measure. In practice, we often estimate volatility using a historical
series of continuously compounded daily returns. We gather a set of daily holding
period returns, convert them into continuously compounded daily returns and then
compute the standard deviation of the continuously compounded daily returns and
annualize that number using Equation 3.
Annualizing is typically done based on 250 days in a year, the approximate number
of business days that financial markets are typically open for trading. Thus,_if daily vol-
atility were 0.01, we would state volatility (on an annual basis) as ​0.01 ​√250 ​  = 0.1581​.
Example 1 illustrates the estimation of volatility for the shares of Astra International.

EXAMPLE 1

Volatility of Share Price

Suppose you are researching Astra International (Indonesia Stock Exchange:
ASII) and are interested in Astra’s price action in a week in which international
economic news had significantly affected the Indonesian stock market. You
decide to use volatility as a measure of the variability of Astra shares during
that week. Exhibit 3 shows closing prices during that week.
​

Exhibit 3: Astra International Daily Closing Prices

​

Closing Price
Day (Indonesian rupiah, IDR)

Monday 6,950
Tuesday 7,000
Wednesday 6,850
Thursday 6,600
Friday 6,350
​

Use the data provided to do the following:

 © CFA Institute. For candidate use only. Not for distribution.
Lognormal Distribution and Continuous Compounding 183

1. Estimate the volatility of Astra shares. (Annualize volatility on the basis of

250 trading days in a year.)
Solution:
First, calculate the continuously compounded daily returns; then, find their
standard deviation in the usual way. In calculating sample variance, to get
sample standard deviation, the divisor is sample size minus 1.

ln(7,000/6,950) = 0.007168.

ln(6,850/7,000) = −0.021661.

ln(6,600/6,850) = −0.037179.

ln(6,350/6,600) = −0.038615.

Sum = −0.090287.

Mean = −0.022572.

Variance = 0.000452.

Standard deviation = 0.021261.

The standard deviation of continuously compounded _ daily returns is
σ ​ ​√T ​ ​​. In this example, σ​​  ˆ​​ is the
σ ​​(​​ ​r​ 0,T​​​)​​  = ​ ˆ
0.021261. Equation 3 states that ​​ ​ ˆ
sample standard deviation of one-period continuously compounded returns.
Thus, σ​​  ˆ​​refers to 0.021261. We want to annualize, so the horizon T corre-
sponds to one year. Because σ​​ ˆ​​is in days, we set T equal to the number of
trading days in a year (250).
Therefore, we find that annualized volatility _ for Astra stock that week was
33.6 percent, calculated as ​0.021261 ​√250 ​  = 0.336165​.

2. Calculate an estimate of the expected continuously compounded annual

return for Astra.
Solution:
Note that the sample mean, −0.022572 (from the Solution to 1), is a sample
estimate of the mean, μ, of the continuously compounded one-period or
daily returns. The sample mean can be translated into an estimate of the
expected continuously compounded annual return using Equation 2, ​​ ​ μˆ​
T = − 0.022572​ (​​250​)​​ ​​(using 250 to be consistent with the calculation of
volatility).

3. Discuss why it may not be prudent to use the sample mean daily return to
estimate the expected continuously compounded annual return for Astra.
Solution:
Four daily return observations are far too few to estimate expected returns.
Further, the variability in the daily returns overwhelms any information
about expected return in a series this short.

4. Identify the probability distribution for Astra share prices if continuously

compounded daily returns follow the normal distribution.
Solution:
Astra share prices should follow the lognormal distribution if the con-
tinuously compounded daily returns on Astra shares follow the normal
distribution.
 © CFA Institute. For candidate use only. Not for distribution.
184 Learning Module 6 Simulation Methods

We have shown that the distribution of stock price is lognormal, given certain
assumptions. Earlier we gave bullet-point expressions for the mean and variance of
a lognormal random variable. In the context of a stock price, the μ ​​ ˆ​ and  ​​ ˆ
σ ​​​  2​​ in these
expressions would refer to the mean and variance of the T horizon, not the one-period,
continuously compounded returns compatible with the horizon of PT.

3 MONTE CARLO SIMULATION

describe Monte Carlo simulation and explain how it can be used in

investment applications

After gaining an understanding of probability distributions used to characterize asset

prices and asset returns, we explore a technique called Monte Carlo simulation in
which probability distributions play an integral role. A characteristic of Monte Carlo
simulation is the generation of a very large number of random samples from a specified
probability distribution or distributions to obtain the likelihood of a range of results.
Monte Carlo simulation is widely used to estimate risk and return in investment
applications. In this setting, we simulate the portfolio’s profit and loss performance for
a specified time horizon, either on an asset-by-asset basis or an aggregate, portfolio
basis. Repeated trials within the simulation, each trial involving a draw of random
observations from a probability distribution, produce a simulated frequency distri-
bution of portfolio returns from which performance and risk measures are derived.
Another important use of Monte Carlo simulation in investments is as a tool for
valuing complex securities for which no analytic pricing formula is available. For
other securities, such as mortgage-backed securities with complex embedded options,
Monte Carlo simulation is also an important modeling resource. Because we control
the assumptions when we carry out a simulation, we can run a model for valuing such
securities through a Monte Carlo simulation to examine the model’s sensitivity to a
change in key assumptions.
To understand the technique of Monte Carlo simulation, we present the process as
a series of steps; these can be viewed as providing an overview rather than a detailed
recipe for implementing a Monte Carlo simulation in its many varied applications.
To illustrate the steps, we use Monte Carlo simulation to value an option, contin-
gent claim, whose value is based on some other underlying security. For this option,
no analytic pricing formula is available. For our purposes, the value of this contingent
claim (an Asian option), equals the difference between the underlying stock price at
that maturity and the average stock price during the life of the contingent claim or
USD 0, whichever is greater. For instance, if the final underlying stock price is USD 34
and the average value over the life of the claim is USD 31, the value of the contingent
claim at its maturity is USD 3 (the greater of USD 34 – USD 31 = USD 3 and USD 0).
Assume that the maturity of the claim is one year from today; we will simulate stock
prices in monthly steps over the next 12 months and will generate 1,000 scenarios to
value this claim. The payoff diagram of this contingent claim security is depicted in
Panel A of Exhibit 4, a histogram of simulated average and final stock prices is shown
in Panel B, and a histogram of simulated payoffs of the contingent claim is presented
in Panel C.
The payoff diagram (Panel A) is a snapshot of the contingent claim at maturity. If
the stock’s final price is less than or equal to its average over the life of the contingent
claim, then the payoff would be zero. However, if the final price exceeds the average
price, the payoff is equal to this difference. Panel B shows histograms of the simulated
 © CFA Institute. For candidate use only. Not for distribution.
Monte Carlo Simulation 185

final and average stock prices. Note that the simulated final price distribution is wider
than the simulated average price distribution. Also, note that the contingent claim’s
value depends on the difference between the final and average stock prices, which
cannot be directly inferred from these histograms.

Exhibit 4: Payoff Diagram, Histogram of Simulated Average, and Final Stock

Prices, and Histogram of Simulated Payoffs for Contingent Claim
A. Contingent Claim Payoff Diagram
Payoff (USD)
35
30
25
20
15
10
5
0
–30 –20 –10 0 10 20 30
Final Price Minus Average Price (USD)

B. Histogram of Simulated Average and Final Stock Prices

Number of Trials
250

200

150

100

50

0
16.0 20.7 25.5 30.2 34.9 39.7 44.4
Stock Price (USD)
Average Stock Price Final Stock Price

C. Histogram of Simulated Contingent Claim Payoffs

Number of Trials
700
600
500
400
300
200
100
0
0 1.7 3.5 5.2 6.9 8.7 10.4
Contingent Claim Payoff (USD)
 © CFA Institute. For candidate use only. Not for distribution.
186 Learning Module 6 Simulation Methods

Finally, Panel C shows the histogram of the contingent claim’s simulated payoffs. In
654 of 1,000 total trials, the final stock price was less than or equal to the average price,
so in 65.4 percent of the trials the contingent claim paid off zero. In the remaining
34.6 percent of the trials, however, the claim paid the positive difference between the
final and average prices, with the maximum payoff being USD 11.
The process flowchart in Exhibit 5 shows the steps for implementing the Monte
Carlo simulation for valuing this contingent claim. Steps 1 through 3 of the process
describe specifying the simulation; Steps 4 through 6 describe running the simulation.

Exhibit 5: Steps in Implementing the Monte Carlo Simulation

Step 1: Specify the quantity of interest (i.e.,
value of the contingent claim)

Specify the Step 2: Specify a time grid; K sub-periods

with t increment for the full time horizon
Monte Carlo Simulation
simulation
Step 1: Specify the model with
Step 3: Specify the method for generating risk factors
the data used in the simulation
Step 2: Specify probability
Step 4: Use the simulated values to distributions for key
produce stock prices used to value risk factors
Run the contingent claim Step 3: Draw random numbers
simulation from distributions for each
over the Step 5: Calculate the average stock price key risk factor over the each
specified and the value of the contingent of the K sub-periods
number of claim payoff
trials Step 4: Use model to convert
Step 6: Repeat steps 4&5 over the required
random numbers to stock
number trials. Then calculate summary
prices
value (i.e., average of the present values
of the contingent claim payoff)

The mechanics of implementing the Monte Carlo simulation for valuing the contingent
claim using the six-step process are described as follows:
1. Specify the quantity of interest in terms of underlying variables. The quan-
tity of interest is the contingent claim value, and the underlying variable
is the stock price. Then, specify the starting value(s) of the underlying
variable(s).
We use CiT to represent the value of the claim at maturity, T. The subscript
i in CiT indicates that CiT is a value resulting from the ith simulation trial,
each simulation trial involving a drawing of random values (an iteration of
Step 4).
2. Specify a time grid. Take the horizon in terms of calendar time and split it
into a number of subperiods—say, K in total. Calendar time divided by the
number of subperiods, K, is the time increment, Δt. In our example, calen-
dar time is one year and K is 12, so Δt equals one month.
3. Specify the method for generating the data used in the simulation. This step
will require that distributional assumptions be made for the key risk factors
that drive the underlying variables. For example, stock price is the under-
lying variable for the contingent claim, so we need a model for stock price
movement, effectively a period return. We choose the following model for
changes in stock price, where Zk stands for the standard normal random
variable:

ΔStock price = (μ × Prior stock price × Δt) + (σ × Prior stock price × Zk).