Problem set 5

Download source

This problem set explores probability through two settings: repeated dice rolls and American roulette. You will use both Monte Carlo simulation and exact mathematical calculations to understand these models.

Please answer each of the exercises below. For those asking for a mathematical calculation, use LaTeX to show your work.

Important: Make sure that your document renders in less than 5 minutes.

Part I: Rolling two dice

  1. Write a function called double_six that takes a number n as an argument, simulates rolling two fair dice n times, and returns TRUE if at least one of the n rolls is a double-six.

    Hint: generate two vectors of rolls and check whether any pair equals (6, 6).

double_six <- function(n){
  ## Your code here
}
  1. Suppose you roll two fair dice 24 times. If we define success to mean seeing at least one double-six, what is the estimated probability of success? Use a Monte Carlo simulation with B = 1000 trials based on the function double_six from the previous exercise.
B <- 10^3
n <- 24
## Your code here
  1. Redo the previous exercise for several values of n to determine at what number of rolls the probability becomes greater than 50%. Set the seed at 1997.
set.seed(1997)
compute_prob <- function(n, B = 10^3){
  ## Your code here
}

n_vals <- 1:60
## Your code here
  1. These probabilities can be computed exactly instead of relying on Monte Carlo approximations. Since the probability of not getting a double-six on a single roll is 35/36, we have

\[ \Pr(\text{at least one double-six in } n \text{ rolls}) = 1 - \left(\frac{35}{36}\right)^n. \]

Plot the probabilities you obtained using Monte Carlo as points and the exact probabilities with a red line.

exact_prob <- function(n){
  ## Your code here
}

## Your code here
  1. The Monte Carlo points in question 4 will not exactly match the red curve because the simulation uses only 1,000 iterations. Repeat exercise 2 for n = 24, but now try B <- seq(10, 250, 5)^2 iterations. Plot the estimated probability against sqrt(B). At what value of sqrt(B) do the estimates consistently stay within 0.005 of the exact probability? Add horizontal lines at the exact probability plus and minus 0.005. Set the seed to 1998.
set.seed(1998)
B <- seq(10, 250, 5)^2
n <- 24
## Your code here

  1. Repeat the comparison from question 4 (Monte Carlo points versus exact red line), but now use your findings from question 5 to choose an appropriate value of B so that the points practically fall on the red curve.

    Hint: choose a value of B that is large enough for a good plot, but not so large that your document takes too long to render.

n_vals <- 1:60
## Your code here

Part II: American roulette

In American roulette there are 38 slots total: 18 red, 18 black, and 2 green.

  1. If a player bets $1 on red, what is the probability that the casino wins the bet?

\[ \text{Derivation here} \]

  1. If a player bets $1 on red, the casino’s profit from that single bet is represented by the random variable \(X\):
  • \(X = -1\) if the ball lands on red,
  • \(X = +1\) otherwise.

Create a sampling model for X using the sample function.

## Your code here
  1. Now create a random variable \(S\) for the casino’s total profit if n = 1000 people each make a $1 bet on red. Use Monte Carlo simulation with B = 10000 trials to estimate the probability that the casino loses money.
n <- 1000
B <- 10^4
## Your code here
  1. What is the expected value of \(X\)?

\[ \text{Your derivation here} \]

  1. What is the standard error of \(X\)?

\[ \text{Your derivation here} \]

  1. What is the expected value of \(S\)? Does the Monte Carlo simulation confirm this?

\[ \text{Your derivation here} \]

## Your code here
  1. What is the standard error of \(S\)? Does the Monte Carlo simulation confirm this?

\[ \text{Your derivation here} \]

## Your code here
  1. Use data visualization to convince yourself that the distribution of \(S\) is approximately normal. Make a histogram and a QQ-plot of the standardized values of \(S\). The QQ-plot should be close to the identity line.
## Your code here
  1. The normal approximation will not be perfect in the tails. What would improve the approximation more: increasing the number of people playing n, or increasing the number of Monte Carlo iterations B? Explain your reasoning.

Answer here

  1. Use the Central Limit Theorem to approximate the probability that the casino loses money. Does your CLT approximation agree with the Monte Carlo estimate from question 9?

\[ \text{Your derivation here} \]

## Your code here
  1. What is the minimum number of people n who must each bet $1 on red so that the probability the casino loses money is approximately 1% according to the CLT? Check your answer with a Monte Carlo simulation.

\[ \text{Your derivation here} \]

## Your code here