Question 0: Answer as many of the questions on this problem sheet as you can. Write up a short report on your findings, with answers to the questions you have studied. Make sure to include relevant graphs produced and the R code you used.
Remember that this worksheet is just for practice, and not assessed for credit. However, Computational Worksheet 2 is assessed work. I am happy for that report to be submitted in any format, including: HTML or PDF produced by R Markdown, PDF produced by LaTeX, Microsoft Word document, or handwritten report with R printouts for plots.
Question 1. We will be studying the simple random walk starting from \(X_0 = 0\) with up probability \(p = 0.6\) and down probability \(q = 1 - p = 0.4\). Produce a sample of the first \(N = 10\) steps of this simple random walk.
We want to start by building the increments process \((Z_n)\), where each \(Z_n\) is \(+1\) with probability \(p = 0.6\) and \(-1\) with probability \(q = 0.4\). We will use the sample()
function to do this. You can find out how to use sample()
by reading the help file, which you can open by entering ?sample
in the console. What arguments do you need to give to sample()
to get \(10\) values of \(\pm 1\) with the correct probabilities? What, for example, do we get from the following?
Z <- sample(c(1, 7, 4), 20, replace = TRUE, prob = c(0.5, 0.3, 0.2))
Z
What is the purpose of the argument replace = TRUE
? Can you adapt the above code to get the desired increments process?
Once we have the increments process \((Z_n)\), we need to transform it into a random walk \((X_n)\). A useful function here is the cumulative sum function cumsum
. The cumulative sum of a vector \((y_1, y_2, \dots, y_n)\) is the vector of partial sums \[ (y_1, y_1 + y_2, y_1 + y_2 + y_3, \dots, y_1 + y_2 + \cdots + y_n) . \] How can we use this function to form a random walk from the increments?
All together, your code to produce the random walk should look something like this:
p <- 0.6
q <- 1 - p
N <- 10
Z <- # Add your code for increments
Z # Should produce a string of 10 plus-or-minus 1s
X <- # Turn increments into a random walk
X # Should produce something that looks like a random walk
Question 2. Plot a sample of the random walk for \(N = 10\) steps and for \(N = 10000\) steps. Try to make your plots look smart, for example by giving them titles and labelling the axes. Briefly comment on the differences between the plots.
The standard command to plot a vector \(\mathbf{X}\) of length \(N\) against \(1, 2, \dots, N\) is plot(1:N, X)
; or just plot(X)
for short. Here 1:N
means the vector \((1, 2, \dots, N)\).
You will probably want to include the starting point \(X_0 = 0\) on your plot too. How can you do this?
The plot function can take extra optional arguments. You can find out about these by typing ?plot
, which brings up the help for the function plot()
. For example, what does the following do?
plot(1:N, X,
type = "b",
col = "red",
ylim = c(-15, 15),
xlab = "Test",
main = "Hello!"
)
Use ?plot
(or Google, or ask a friend) to find out the possible arguments to type =
, and make sure to pick the most appropriate one. You may find that different arguments are appropriate for the \(N = 10\) and \(N = 10000\) cases.
Remember to comment on the differences between the plots.
Question 3. Make a function in R that produces a sample of a simple random walk for given \(p\) of length \(N\). Test your function to check it works. What are the advantages of making a a function like this?
You’ll want a syntax something like this:
RandomWalk <- function(N, p) {
Z <- # Your code for N increments that are +1 with probability p
# and -1 otherwise
cumsum(Z)
}
If you do this correctly, then, for example,
X <- RandomWalk(20, 0.3)
X
should produce a simple random walk of length \(20\) with \(p = 0.3\). It might be helpful to check by plotting a random walk of your choice.
Question 4. Estimate the expected value of of the simple random walk at \(N = 100\) steps by simulating many random walks and taking an average. How does this compare with the true answer?
Given a vector x
, we know that, for example, x[10]
gives the 10th element of the vector. So, using your RandomWalk()
function, RandomWalk(N, p)[N]
should give the final value of an random walk of length \(N\).
We can use the replicate()
function (read ?replicate
to learn more) to get a large number of samples of the random walk.
N <- 100
trials <- 10000 # Or some other appropriate large number
samples <- replicate(trials, RandomWalk(N, p)[N])
# Now add code to output the average of samples
What does theory we have learned say the answer should be?
Question 5. (Optional) Investigate estimating other properties of the simple random walk through simulation.