Numerical Methods Bootcamp

Tuesday Assignment
Miscellaneous stuff

In this assignment you are asked to do three things. First, you will simulate draws from
a weird probability distribution and illustrate the Central Limit Theorem.1 Second you
are asked to solve a nonlinear equation on a grid containing several values of a parameter,
numerically approximate the solution for parameter values not on the grid, and graphically
illustrate the results. Lastly, you will solve a fixed point problem. As usual you are provided
with some incomplete scripts to get you started.

The Central Limit Theorem in action. Consider the probability density function
(pdf)

f (x) = a(1 − x), if x ≥ x̄

f (x) = 0, otherwise.

for x ∈ [0, 1]. Figure 1 illustrates what this pdf looks like. In Figure 1 (and the subsequent
figures) x̄ is set to 0.7 and a = 1/(1/2+1/2x̄2 − x̄). Can you verify that this indeed is a pdf?
Derive the cdf, F (z) = P r(x ≤ z), and its inverse. Figure 2 illustrates what these should
look like. The inverse of a cdf is incredibly useful for the purpose of simulating peculiar
looking distributions.
Calculate the mean, µ, and the variance, σ 2 , of the distribution using the integral
function in Matlab.
Set the random seed to 1979 and make N = 100 draws of a random uniform variable on
[0, 1]. Calculate Sn as
√ −1 (e
P
iF i)
Sn = N( − µ).
N
Repeat 10,000 times and plot a histogram. Figure 3 shows the results you should get, ac-
companied by a fitted normal distribution. According to the CLT, limN →∞ Sn should be
normally distributed with mean µ and variance σ 2 . Isn’t the CLT pretty damn amazing
after all? What happens if you set N = 2?

1
This may seem slightly divorced from the course content. But it is a useful exercise to illustrate how
numerical computations can provide some intuition on theoretical results. Seeing is believing.
2

Probability Density Function

7

5
Density

0
0 0.2 0.4 0.6 0.8 1
Support, x

Figure 1. Probability Density Function (pdf).

Cumulative Distribution Function Cumulative Distribution Function

1.2 1

1 0.95
Cumulative distribution

0.9
0.8
Support, x

0.85
0.6
0.8
0.4
0.75
0.2 0.7

0 0.65
0 0.5 1 0 0.5 1
Support, x Cumulative distribution

Figure 2. Cumulative Density Functions (cdf).

 3

Histogram vs. Probability Density Function

6

4
Density

0
0.5 0.6 0.7 0.8 0.9 1 1.1
Support, x

Figure 3. Histogram with fitted normal.

A nonlinear equation. The nonlinear equation you will solve is given by

f (y) = (xη − y)−1 − δηy η−1 (y η − z)−1 ,

where x and z are treated as parameters. z is set equal to (ηδ)1/(1−η) .

For each value of x on a grid solve for y such that f (y) = 0 using Newton’s method.
Plot your results in the (x, y) plane. Compare your results with those in Figure 4 which
illustrates the mapping y = g(x). That is the function y = g(x) such that,

(xη − g(x))−1 − δηg(x)η−1 (g(x)η − z)−1 = 0 ∀x

What is the relationship between z and the intersection of your solution with the 45o line?

A Functional Equation. For this exercise we will think of z as being an endogenous

variable which relates to y in exactly the same way as y relates to x. That is if y = g(x)
then z = g(y), or, equivalently, that z = g(g(x)). As a consequence, the function g(x)
should satisfy the functional equation

(xη − g(x))−1 − δηg(x)η−1 (g(x)η − g(g(x)))−1 = 0 ∀x (1)

4

0.22
0.21
0.2
Solution, y

0.19
0.18
0.17
0.16 Solution
Initial guess
0.15 450 line
0.14
0.14 0.16 0.18 0.2 0.22 0.24
Grid for x

Figure 4. The solution to a nonlinear equation.

Notice that the only difference between this equation and the previous equation is that z is
now not fixed, but a function of y. And it takes the same for as y as a function of x. Hence,
z is replaced by g(g(x)).
There are several ways of solving this functional equation, but I would like you to use
what is known as time iteration – a concept that will be further clarified tomorrow.
Time iteration in this setting works as follows. Guess for a solution to equation (1)
and call this guess g0 (·). For instance, a stupid, but perfectly operational guess, would be
g0 (x) = x.2 Now, use your knowledge of solving nonlinear equations on a grid for x to solve

(xη − y)−1 − δηy η−1 (y η − g0 (y))−1 = 0 ∀x

Thus, for all x we have a solution for y. Combining these solutions we can use an ap-
proximation method to update our guess y = g1 (x). Thus, in general, for any guess gn (x),
solve

(xη − y)−1 − δηy η−1 (y η − gn (y))−1 = 0

for a grid of x. Combine to create y = gn+1 (x), and repeat until

(xη − gn (x))−1 − δηgn (x)η−1 (gn (x)η − gn (gn (x)))−1 ≈ 0, ∀x.

2This is not a general statement. It is a good idea to choose an initial guess with a bit of care.
 5

If you have done things correctly, your results should correspond to Figure 5

0.22
0.21
0.2
Solution, y

0.19
0.18
0.17
Solution
0.16 Initial guess
450 line
0.15 Exact solution
0.14
0.14 0.16 0.18 0.2 0.22 0.24
Grid for x

Figure 5. The solution to a functional equation.