2.8 Simulation and Bootstrapping

2.8 Simulation and Bootstrapping

Question 1

A portfolio manager has asked each of four analysts to use Monte Carlo simulation to price a path-dependent derivative contract on a stock. The derivative expires in nine months and the risk-free rate is $4\%$ per year compounded continuously. The analysts generate a total of $20,000$ paths using a geometric Brownian motion model, record the payoff for each path, and present the results in the table shown below.

Analyst	Number of Paths	Average Derivative Payoff per Path (USD)
1	$2,000$	$43$
2	$4,000$	$44$
3	$10,000$	$46$
4	$4,000$	$45$

What is the estimated price of the derivative?

A. USD $43.33$
B. USD $43.77$
C. USD $44.21$
D. USD $45.10$

Answer: B
Following the risk neutral valuation methodology, the price of the derivative is obtained by calculating the weighted average nine month payoff and then discounting this figure by the risk free rate.

Average payoff calculation: $(2000\times 43+4000\times 44+10000\times 46+4000\times 45)/20000=45.1$

Discounted payoff calculation: $45.10\times e^{-0.04\times (9/12)}=43.77$

---

Question 2

Which of the following statements about Monte Carlo simulation is incorrect?

A. Correlations among variables can be incorporated into a Monte Carlo simulation.
B. Monte Carlo simulations can handle time-varying volatility.
C. Monte Carlo methods can be used to estimate value-at-risk (VaR) but cannot be used to price options.
D. For estimating VaR, Monte Carlo methods generally require more computing power than historical simulations.

Answer: C
Monte Carlo simulations cannot price options with early exercise accurately. All of the other statements are correct.

Correlation can be incorporated using the method of Cholesky decomposition, Monte Carlo simulations can be designed to handle time varying volatility, and Monte Carlo simulations are computationally more intensive than historic simulations.

---

Question 3

Consider a stock that pays no dividends, has a vol. of $30\%$ per annum, and provide an expected return of $15\%$ per annum with continuous compounding. The stock price movements follow GBM. Consider a time interval of $1$ week and the initial stock price is $100$, then the stock price increase has a normal distribution with:

A. Mean = $0.268\%$, standard deviation = $4.03\%$
B. Mean = $0.278\%$, standard deviation = $4.13\%$
C. Mean = $0.288\%$, standard deviation = $4.16\%$
D. Mean = $0.288\%$, standard deviation = $4.27\%$

Answer: C
Mean: $\mu /n=15\%/52=0.288\%$

SD: $\sigma /\sqrt{n}=30\%/\sqrt{52}=4.16\%$

---

Question 4

Consider a stock that pays no dividends, has a volatility of $25\%$ per annum and an expected return of $13\%$ per annum. Suppose that the current share price of the stock, $S_0$, is USD $30$. You decide to model the stock price behavior using a discrete-time version of geometric Brownian motion and to simulate paths of the stock price using Monte Carlo simulation. Let $\Delta t$ denote the time interval used and let $S_t$ denote the stock price at time interval $t$. So, according to your model,

$S_{t+1}=S_t\times (1+0.13\Delta t+0.25\sqrt{\Delta t}\times ϵ)$ ,where $ϵ$ is a standard normal variable.

To implement this simulation, you generate a path of the stock price by starting at $t=0$, generating a sample for e updating the stock price according to the model, incrementing $t$ by $1$, and repeating this process until the end of the horizon is reached. Which of the following strategies for generating a sample for $\Delta$ will implement this simulation properly?

A. Generate a sample for $ϵ$ by using the inverse of the standard normal cumulative distribution of a sample value drawn from a uniform distribution between $0$ and $1$.
B. Generate a sample for $ϵ$ by sampling from a normal distribution with mean $0.13$ and standard deviation $0.25$.
C. Generate a sample for $ϵ$ by using the inverse of the standard normal cumulative distribution of a sample value drawn from a uniform distribution between $0$ and $1$. Use Cholesky decomposition to correlate this sample with the sample from the previous time interval.
D. Generate a sample for $ϵ$ by sampling from a normal distribution with mean $0.13$ and standard deviation $0.25$. Use Cholesky decomposition to correlate this sample with the sample from the previous time interval.

Answer: A
Monte Carlo Simulation assumes independence across time so there is no need to correlate samples from time period to time period, eliminating C and D. Choice A describes a valid method for generating a sample from a standard normal distribution.

---

Question 5

Monte Carlo simulation is suitable for pricing options in which of the following cases?

I. An Asian option on a stock market index (payoff based on average stock price).
II. A look-back put option on XYZ stock (payoff based on maximum or minimum stock price).
III. An American call option on ABC stock (possible early exercise).
IV. A cash-or-nothing call option (i.e., binary option) on SCU stock (payoff is fixed amount or nothing).

A. I and IV
B. I, II, and IV
C. Il and II
D. IIl and IV

Answer: B
Monte Carlo simulation is suitable for pricing options in each case except when early exercise of the option is possible. This means that the Monte Carlo approach could not accurately price the American call option. Monte Carlo simulation is very useful for options with price-dependent paths (such as Asian options and look-back options) and can also handle options with complex payoff, such as binary options.

---

Question 6

A risk manager has been requested to provide some indication of accuracy of a Monte Carlo simulation. Using $1,000$ replications of a normally distributed variable $S$, the relative error in the one-day $99\%$ VaR is $5\%$. Under these conditions:

A. Using $1,000$ replications of a long option position on $S$ should create a larger relative error.
B. Using $10,000$ replications should create a larger relative error.
C. Using another set of $1,000$ replications will create an exact measure of $5.0\%$ for relative error.
D. Using $1,000$ replications of a short option position on $S$ should create a larger relative error.

Answer: D
Short option positions have long left tails, which makes it more difficult to estimate a left-tailed quantile precisely. Accuracy with independent draws increases with the square root of $K$. Thus increasing the number of replications should shrink the standard error, so answer B is incorrect.

---

Question 7

Suppose you simulate the price path of stock HHF using a geometric Brownian motion model with drift $\mu =0$, volatility $\sigma =0.14$, and time step $\Delta t=0.01$. Let $S_t$ be the price of the stock at time $t$. If $S_0=100$, and the first two simulated (randomly selected) standard normal variables are ${ϵ}_1=0.263$ and ${ϵ}_2=-0.475$, what is the simulated stock price after the second step?

A. $96.79$
B. $99.79$
C. $99.97$
D. $99.70$

Answer: D
The process for the stock prices has mean of zero and volatility of Hence the first step is $\sigma \sqrt{\Delta t}=0.14\sqrt{0.01}=0.014$, $S_1=S_0(1+0.014\times 0.263)=100.37$

The second step is $S_2=S_1(1+0.014\times -0.475)=99.70$

---

Question 8

A quantitative risk analyst is comparing the computational efficiency of different estimators generated using Monte Carlo simulation. Relevant information is summarized in the following table:

	Estimator A	Estimator B	Estimator C	Estimator D
Standard deviation	$0.30$	$0.40$	$0.25$	$0.35$
Time for generating one scenario (seconds)	$35$	$25$	$40$	$30$
Scenarios	$20$	$40$	$30$	$50$
Total time for generating scenarios (seconds)	$400$	$1,000$	$1,200$	$1,500$

Which of the estirnators is most computationally efficient? (Important)

A. Estimator A
B. Estimator B
C. Estimator C
D. Estimator D

---

Question 9

You are running a Monte Carlo simulation to price a portfolio of options. When generating random numbers for use in the simulation:

A. The stratified sampling method eliminates extreme observations.
B. A truly random number generator would avoid clustered observations.
C. A congruential pseudorandom number generator creates sequences converging to a constant value.
D. The Latin hypercube sampling method ensures that all strata are sufficiently well-represented.

---

Question 10

Which of the following statements about simulation is invalid?

A. The historical simulation approach is a nonparametric method that makes no specific assumption about the distribution of asset returns.
B. When simulating asset returns using Monte Carlo simulation, a sufficient number of trials must be used to ensure simulated returns are risk neutral.
C. Bootstrapping is an effective simulation approach that naturally incorporates correlations between asset returns and non-normality of asset returns, but does not generally capture autocorrelation of asset returns.
D. Monte Carlo simulation can be a valuable method for pricing derivatives and examining asset return scenarios.

Answer: B
Risk neutrality has nothing to do with sample size.

---

Question 11

PE2020Q30 /PE2021Q30
A modeling team at a risk management consulting firm is debating whether it is appropriate to use the bootstrap technique to analyze a particular sample of data. Which of the following represents a situation where the bootstrap technique will be ineffective?

A. The data follow an asymmetric distribution.
B. The data are independent and identically distributed.
C. The data contain outliers.
D. The data are normally distributed.

Answer: C
Learning Objective: Describe situations where the bootstrapping method is ineffective.

C is correct. There are at least two situations where the bootstrap will not work well: if there are outliers in the data, and if the data are dependent on one another.

A is incorrect. One benefit of the bootstrap is the fact that they do not have distributional requirements - the bootstrap does not care what distribution the data come from.

B is incorrect. The very opposite is true, data must be independent in order for the bootstrap to be effective.

D is incorrect. The same explanation for A applies here.

---

Question 12

PE2020Q29 / PE2021Q29 / PE2022Q29
A data analyst at a large bank is evaluating the valuation of a unique stock option with few known properties. The analyst is considering using simulation to model the option’s potential value. The analyst assesses whether to use Monte Carlo simulation or bootstrapping to conduct the analysis. Which of the following statements about bootstrapping is correct?

A. Data used for bootstrapping must follow a standard normal distribution.
B. Data used for bootstrapping must be resampled with replacement.
C. Data used for bootstrapping must come from a variable with known properties.
D. Data used for bootstrapping must be resampled such that all possible outcomes in a probability space are present.

Answer: B
Learning Objective: Describe the bootstrapping method and its advantage over Monte Carlo simulation.

B is correct. In bootstrapping, data are resampled with replacement in order to empirically estimate the sampling distribution.

A is incorrect. One advantage of bootstrapping over Monte Carlo simulation is that the data do not have to follow any distribution.

C is incorrect. Same explanation as A.

D is incorrect. This would be ideal but not always possible.

---

Question 13

An analyst is conducting a Monte Carlo simulation to estimate the expected value of a random variable. The analyst wants to reduce the standard error of the simulated expectation. Which of the following correctly describes a method for reducing the standard error?

A. Increasing the expected value of the simulation
B. Increasing the number of replications
C. Increasing the variance of the distribution
D. Increasing the confidence level of the simulation

Answer: B
Learning Objective: Describe ways to reduce Monte Carlo sampling error.

B is correct. One way to increase the accuracy of estimation from Monte Carlo simulation is to increase the number of replications.

A is incorrect. Increasing the expected value does not affect the accuracy of the simulation.

C is incorrect. Increasing the variance does not improve accuracy of the simulation; the opposite effect occurs.

D is incorrect. Increasing the confidence level only provides a wider range of values but
does not improve accuracy.

---