Scribd
Upload
39 views

Lecture1 Introduction

The document provides an introduction and overview for a summer school in economics and finance in 2023. It outlines key details such as the instructor, dates, times, and resources. It then discusses why Python is a popular language for economics, and presents a classic dynamic programming problem involving consumption and capital accumulation. It notes that solving this problem involves checking for a unique solution, discretizing the state space, representing the value function on a computer, and using techniques like value function iteration to find the optimal policies.

Uploaded by

sfreire3
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
39 views

Lecture1 Introduction

The document provides an introduction and overview for a summer school in economics and finance in 2023. It outlines key details such as the instructor, dates, times, and resources. It then discusses why Python is a popular language for economics, and presents a classic dynamic programming problem involving consumption and capital accumulation. It notes that solving this problem involves checking for a unique solution, discretizing the state space, representing the value function on a computer, and using techniques like value function iteration to find the optimal policies.

Uploaded by

sfreire3
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

Introduction and Overview

Summer School in Economics and Finance 2023

César Salinas

Indiana University

Spring 2023
Introduction

▶ Instructor: César Salinas

▶ Email: [email protected]

▶ Lectures: from January 3 to January 6, 2023

▶ Lecture time: 7pm-10pm

▶ Zoom link, syllabus, lecture notes and code for tutorials:


Posted at Canvas.
Introduction

▶ I assume prior background in dynamic programming and Python

▶ Additional resources to learn more about Python:


– https://python-programming.quantecon.org/intro.html
– http://scipy-lectures.org/intro/language/python_language.html
– https://jakevdp.github.io/PythonDataScienceHandbook/
Why Python?

Source: From Aruoba and Fernández-Villaverde (2018).


Why Python?

▶ Python is free and open source.

▶ Python is a general-purpose language used and supported extensively by companies and


government agencies.

▶ Python is particularly popular within the scientific community.

▶ Python is also very beginner-friendly and is found to be suitable for students learning
programming and recommended to introduce computational methods.

▶ More about Python: https://python-programming.quantecon.org/about_py.html


A Very Typical Problem

A classic problem we will be focusing throughout the semester:

V (z, k) = max′ u(c) + βEV (z ′ , k ′ )


c,k
′
s.t. c + k = F (z, k) + (1 − δ)k

Many questions here:


1. Does a solution exist?
2. Is there a unique solution or not?
3. What are the relevant z, k?
4. How can V be represented on a computer?
5. How can we compute Ez ?
6. How can we find the optimal c and k ′ ?
The Bellman equation is a functional equation because it can be written as V = T (V ) or

V = TV .

Two useful theorems. 2 crucially theorems in Chapter 3 of Stokey and Lucas: (i) the Blackwell’s
Theorem (BT), and (ii) the Contraction Mapping Theorem (CMT).
▶ To understand the BT you need to be familiarized with metric spaces, norms and contraction
mappings.

▶ The BT gives sufficient conditions for an operator T to be a contraction mapping (it lets
us to invoke the CMT).
▶ The latter theorem establishes that if an operator T is a contraction mapping, then
1. it has a unique fixed point, i.e. there exists a unique function V such that TV = V .
2. T n V0 → V as n → ∞.
▶ The beauty of CMT is powerful. It not only tell us the existence and uniqueness of V ∗ but
it also shows us how to find it!
▶ CMT suggests:
1. Set n = 0. Choose an initial guess V0 ∈ S
2. Find Vn+1 by solving
Vn+1 = TVn = max′ u(c) + βEV (z ′ , k ′ )
c,k

3. Set n = n + 1 and repeat step 2 until |Vn+1 − Vn | < ε

▶ We still need to address some other questions


Discretization. The easiest, flexible and most robust way to solve this and similar problems.
X
V (z, k) = max′ u(c) + β F (z ′ |z)V (z ′ , k ′ )
c,k
z ′ ∈Z
′
s.t. c + k = F (z, k) + (1 − δ)k,
for all k ∈ K, z ∈ Z.

With all this in mind the answers to the previous questions are
1. Does a solution exist? Yes
2. Is there a unique solution or not? Yes
3. What are the relevant z, k? K × Z
4. How can V be represented on a computer? Matrix
5. How can we compute Ez ? Markov Chain
6. How can we find the optimal c and k ′ ? Checking all z, k
A Simple Example

Let’s consider the Neoclassical Growth model with: log utility, Cobb-Douglas production, full
depreciation and non-stochastic

V (k) = max′ log(c) + βV (k ′ )



c,k

s.t. c + k ′ = Ak α

or more compactly
α ′

Vn+1 (k) = TVn = max
′
log(Ak − k) + βVn (k )
k

We want to solve for the value function V (k) and the policy function k ′ = g (k)
Value Function Iteration

1. Make a guess on V0 (k) = 0 for all k, which gives us


′
α
⇒ k ′ = 0 ⇒ V1 (k) = log A + α log k

V1 = max ′
log(Ak − k )
k

2. V2 becomes
log(Ak α − k ′ ) + β(log A + α log k ′ ) ,

V2 = max
′ k
αβAk α
and the FOC implies k′ = 1+αβ

3. Substitute k’ to obtain V2 and keep iterating until convergence. We can verify that in the
limit T → ∞
k ′ = αβk α
Projection Methods

▶ From the previous slide, we can notice that the value function takes the form of V (k) =
a + b log k where a and b are unkown coefficients.
Then, all we need to solve for is

log(Ak α − k ′ ) + β(log A + α log k ′ )



a + b log k = max ′
k

▶ This can be solved with pencil and paper and we should get

log A + (1 − αβ) log(1 − αβ) + αβ log αβ


a=
(1 − β)(1 − αβ)
α
b=
1 − αβ
Projection Methods

▶ This method can also be applied to solve for the policy function directly.
When FOCs are sufficient, then we can find the solution to the above problem using the
FOC:
1 αβA(g (k))α−1
− = 0,
Ak − g (k) A((g (k))α − g (g (k)))
α

where we want to soplve for the functional equation g (k)

▶ Guessing g (k) = sAk α and substituting in the equation above we should be able to find

s = αβ

▶ The functional form will not be as simple and trivial as this one, but we will cover methods
to approximate functions with pretty flexible basis functions later
Content

1. Discretizing the State and Choice Space


1.1 Tauchen
1.2 Rouwenhorst
1.3 Brute Force Grid Search

2. Function Approximation
2.1 Endogenous Grid Method

3. Root Finding
3.1 Value Function Iteration with first order conditions

4. Optimization Methods
4.1 Value Function Iteration with optimization methods

You might also like