CFA

Simulation Methods

www.proschoolonline.com
The Lognormal Distribution
• The lognormal distribution is used to model the
probability distribution of asset prices, including shares. Two Lognormal Distributions
• It plays a significant role in the Black–Scholes–Merton
option pricing model, where it is assumed that the price
of the underlying asset follows a lognormal distribution.
• A random variable Y follows a lognormal distribution if
its natural logarithm, ln Y, is normally distributed, and
vice versa.
• The lognormal distribution is bounded below by 0 and
exhibits a right-skewed (long right tail) shape.
• It accurately describes the distribution of prices for
many financial assets due to the bounded nature of
asset prices.
• In contrast, the normal distribution is often a good
approximation for returns.
www.proschoolonline.com
The Lognormal Distribution
• The lognormal distribution is characterized by two parameters, similar to the normal distribution.
• However, the lognormal distribution is unique in that its parameters are defined in terms of a different
distribution, specifically, the associated normal distribution.
• The two parameters for the lognormal distribution are the mean and standard deviation (or variance) of its
associated normal distribution.
• This means that there are two sets of means and standard deviations (or variances) to consider:
1. The mean and standard deviation (or variance) of the associated normal distribution (these are the
parameters).
2. The mean and standard deviation (or variance) of the lognormal variable itself.

To illustrate this relationship, let us take an example in the next slide

www.proschoolonline.com
The Lognormal Distribution

Simulated Returns (Normal PDF) and Asset Prices

(Lognormal PDF)
• A simulation was conducted with 1,000 scenarios of
yearly asset returns.
• These returns were assumed to follow a normal
distribution with a mean of 7 percent and a standard
deviation of 12 percent.
• Each scenario i involved converting simulated
continuously compounded returns (ri) into future asset
prices using the formula Price(1 year later)i = USD1 ×
exp(ri).
• Notably, the lognormal distribution for future asset
prices is characterized by being bounded from below
by 0 and having a long right tail

www.proschoolonline.com
The Lognormal Distribution
• To understand the mean and variance of a lognormal variable, start with a normal random variable X
with an expected value (μ) and variance (σ²).
• Define Y = exp(X), which transforms X into a lognormal random variable Y. This transformation involves
the exponential function, where e ≈ 2.7183.
• The expected value of Y, denoted as μL, is not simply exp(μ) as one might guess. It is actually exp(μ +
0.50σ²).
• This result indicates that the mean of the lognormal distribution is greater than exp(μ) by a factor of
exp(0.50σ²), which is greater than 1.
• Increasing the variance (σ²) of X causes the distribution to spread out, pushing its center to the right
and increasing the mean.
• The expressions for the mean (μL) and variance (σL²) of a lognormal random variable are summarized as
follows, where μ and σ² are the mean and variance of the associated normal distribution:
1. Mean (μL) of a lognormal random variable = exp(μ + 0.50σ²).
2. Variance (σL²) of a lognormal random variable = exp(2μ + σ²) × [exp(σ²)− 1].
www.proschoolonline.com
Continuously Compounded Rates of Return
Stock Price and Continuously Compounded Return Relationship:
• The future stock price at time T (PT) can be expressed as the current stock price (P0) multiplied by the
exponential function of the continuously compounded return (r0,T) from time 0 to T:
PT = P0 × exp(r0,T)
• The continuously compounded return over time (r0,T) is determined by summing the one-period
continuously compounded returns (rT−1,T, rT−2,T−1, and so on).
Normal Distribution Assumption:
• When these shorter-period returns are normally distributed (with certain conditions), r0,T tends to be
normally distributed or approximately normal.
• PT follows a lognormal distribution due to its relationship with the logarithm of a normal random
variable.

www.proschoolonline.com
Continuously Compounded Rates of Return
• Many financial models assume that returns are independently and identically distributed (i.i.d.),
meaning that past returns do not predict future returns, and the mean and variance remain consistent
over time.
• Assuming the one-period continuously compounded returns (like r0,1) are i.i.d. with mean μ and
variance σ², the mean of the T-period return (r0,T) is μT, and its variance is σ²T.
• Then, E(r0,T) = E(rT−1,T) + E(rT−2,T−1) + . . . + E(r0,1) = μT, (we add up μ for a total of T times), and
• σ2(r0,T) = σ2T (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.
• Even if the one-period returns are not normally distributed, their sum, r0,T, tends to be approximately
normal as T increases, following the central limit theorem.

www.proschoolonline.com
Continuously Compounded Rates of Return
Lognormal Distribution:
• Comparing PT = P0 * exp(r0,T) to Y = exp(X), where X is normally distributed and Y is lognormal, we can use
the lognormal distribution to model future stock prices, assuming that r0,T is at least approximately
normal.
• This assumption of normally distributed returns provides the theoretical basis for using the lognormal
distribution to model the prices of various financial assets, such as stocks.

www.proschoolonline.com
Volatility
• Volatility is a measure of the standard deviation of continuously compounded returns.
• To estimate volatility, historical daily returns are commonly used.
• These daily returns are converted into continuously compounded returns.
• The standard deviation of these continuously compounded daily returns is computed.
• To express volatility on an annual basis, it's annualized using a standard convention of 250 trading days
per year (approximate).
• For example, if the daily volatility is 0.01, the annualized volatility would be calculated as 0.01 * √250 ≈
0.1581.

www.proschoolonline.com
Example 1
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.

Astra International Daily Closing Prices

Estimate the volatility of Astra shares. (Annualize volatility on the basis of 250 trading days in a year.)

www.proschoolonline.com
Example 1
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.
Using equation: σ(r0,T) = σ 𝑇.
σ(r0,T) = 0.021261 250
σ(r0,T) = 0.336165 or 33.6%

www.proschoolonline.com
Monte Carlo Simulation
• Monte Carlo simulation involves generating a large number of random samples from specified
probability distributions.
• It's used to estimate risk and return in investment scenarios and for valuing complex securities without
analytic pricing formulas.
• One common use is to simulate a portfolio's profit and loss performance for a given time horizon.
• It can also value complex securities with embedded options and assess the sensitivity of models to
changes in key assumptions.

www.proschoolonline.com
Example 2
Illustration with an Asian Option:
• An Asian option is used as an example to illustrate the Monte Carlo simulation process.
• This type of option's value depends on the underlying security's final price compared to its average price
over the option's life.
• There's no analytic pricing formula available for this option.
Simulation Scenario:
• In this scenario, the contingent claim (Asian option) is valued one year from today.
• Stock prices are simulated in monthly steps over the next 12 months, generating 1,000 scenarios to value
the claim.

www.proschoolonline.com
Example 2
A. Contingent Claim Payoff Diagram B. Histogram of Simulated Average and Final Stock Prices

If the final stock price is less than or equal The simulated final price distribution, with a wider
to its average during the claim's life, the spread, contrasts with the simulated average price
payoff is zero. However, if the final price distribution, and the contingent claim's value
exceeds the average price, the payoff equals depends on the difference between these two
this difference. prices, not directly evident in the histograms.
www.proschoolonline.com
Example 2

C. Histogram of Simulated Contingent Claim Payoffs

Out of 1,000 trials, in 654 trials (65.4%), the final stock

price was less than or equal to the average price, resulting
in zero payoff. In the remaining 34.6% of trials, the claim
paid the positive difference between the final and average
prices, with the highest payoff being USD 11.

www.proschoolonline.com
Monte Carlo Simulation
Steps in Implementing the Monte Carlo Simulation

www.proschoolonline.com
Monte Carlo Simulation
Steps in Implementing the Monte Carlo Simulation
Step 1: Specify Variables
• Define the quantity of interest (contingent claim value) in terms of underlying variables (stock price).
• Set initial values for these variables, denoted as CiT for the claim value at maturity in the ith simulation
trial.
Step 2: Time Grid
• Divide the time horizon into subperiods, typically using calendar time divided by the number of
subperiods (Δt).
• In this example, one year is divided into 12 subperiods (monthly), so Δt is one month.

www.proschoolonline.com
Monte Carlo Simulation
Step 3: Data Generation
• Define distributional assumptions for key risk factors driving the underlying variables (e.g., stock price).
• Use a model to generate changes in the underlying variables based on these assumptions.
• In this example, the model for changes in stock price incorporates mean (μ) and standard deviation (σ)
along with standard normal random variables (Zk).
• ΔStock price = (μ × Prior stock price × Δt) + (σ × Prior stock price × Zk).
• The term Zk is the key risk factor in the simulation. Through our choice of μ (mean) and σ (standard
deviation), we control the distribution of the stock price variable.
Step 4: Generate Values
• Use a computer program or spreadsheet function to generate K random values for each risk factor.
• These values, such as Z1, Z2, ..., ZK, are drawn from the specified distributions.

www.proschoolonline.com
Monte Carlo Simulation
Step 5: Calculate Values
• Use the simulated values to compute stock price changes (ΔStock price) based on the model.
• Transform these changes into a sequence of K stock prices over the subperiods, starting from the
initial stock price.
• Calculate the average stock price during the life of the contingent claim (average of K stock prices).
• Compute the value of the contingent claim at maturity (CiT).
• Discount the CiT value to the present using an appropriate interest rate, obtaining Ci0.
• This completes one simulation trial.
Step 6: Repeat and Summarize
• Iteratively repeat Steps 4 and 5 for the required number of trials (I).
• The quantity of interest is often the mean value of Ci0 across all simulation trials, providing the Monte
Carlo estimate of the contingent claim's value.

www.proschoolonline.com
Bootstrapping
• Bootstrap is a popular resampling technique that uses computer simulation for statistical inference
without relying on analytical formulas.
• Bootstrap aims to construct the sampling distribution of population parameters (e.g., mean, variance)
by mimicking random sampling from a population.
• Unlike Monte Carlo simulation, Bootstrap doesn't assume knowledge of the population's true
characteristics; it relies on the observed sample.
• In Bootstrap, the observed sample is treated as a representation of the population, and resampling is
done from this sample as if it were the population.
• Bootstrap and Monte Carlo simulation are both resampling methods, but they differ in their approaches.
• Bootstrap focuses on estimating population parameters from a sample, while Monte Carlo simulation
generates data based on known distributions for broader analysis and modeling purposes.

www.proschoolonline.com
Bootstrapping
Bootstrap Resampling

In bootstrap, we repeatedly draw samples from the original sample, and each resample is of the same
size as the original sample.

www.proschoolonline.com
Bootstrapping
Steps in Implementing Simulation using Bootstrapping

www.proschoolonline.com
Bootstrapping
Steps in Implementing Simulation using Bootstrapping
Step 1: Specification
• Specify the quantity of interest, which is the contingent claim value, and identify the underlying variable
(e.g., stock price).
• Set the initial values for the underlying variable.
• Use CiT to denote the value of the claim at maturity (T), with i indicating different simulation trials.
Step 2: Time Grid
• Define a time grid consistent with the periodicity of sample observations.
• Split the calendar time into subperiods (K total), with Δt representing the time increment (e.g., one
month for a one-year horizon).

www.proschoolonline.com
Bootstrapping
Step 3: Data Generation
• Determine the method for generating data used in the simulation.
• In this case, use observed changes in the underlying variable (e.g., historical stock price returns) as the
empirical distribution.
Step 4: Simulate Stock Prices
• Generate simulated values for the underlying variable (stock prices) using the bootstrap method.
• Utilize a computer program or spreadsheet function to draw K random values of the stock process from
the empirical distribution.
• Convert the changes in stock prices (ΔStock price) from Step 3 into stock price dynamics.
• Calculate a sequence of K stock prices, starting with the initial stock price, over the K subperiods.

www.proschoolonline.com
Bootstrapping
Step 5: Calculate Values
• Compute the average stock price during the contingent claim's life (the sum of K stock prices divided by
K).
• Determine the value of the contingent claim at maturity (CiT).
• Calculate the present value (Ci0) by discounting CiT using an appropriate interest rate as of today.
• Each iteration of this process completes one simulation trial.
Step 6: Repetition and Summary
• Repeat Steps 4 and 5 over the required number of trials (I).
• Iteratively return to Step 4 until all I trials are completed.
• Finally, produce summary values and statistics for the simulation, focusing on the mean value of Ci0 over
all bootstrap runs (I = 1,000 in the example).

www.proschoolonline.com