**Teacher:**Florian Oswald, florian.oswald@sciencespo.fr**Class Times:**Fridays 10:15-12:15 starting 29 Jan 2021**Class Location:**Zoom**Slack**: There will be a slack channel for all communication

This is a course for PhD students at the Department of Economics at Sciences Po in Computational Economics.

In this course you will learn about some commonly used methods in Computational Economics. These methods are being used in all fields of Economics. The course has a clear focus on applying what you learn. We will cover the theoretical concepts that underlie each topic, but you should expect a fair amount of *hands on* action required on your behalf. In the words of the great Che-Lin Su:

Doing Computation is the only way to learn Computation. Doing Computation is the only way to learn Computation. Doing Computation is the only way to learn Computation.

True to that motto, there will be homeworks for you to try out what you learned in class. There will also be a term paper.

You need a laptop.

You should be familiar with the material from

*Introduction to Programming*taught by Clement Mazet in M1. Check out the materials hereYou must sign up for a free account at github.com. Choose a reasonable user name and upload a profile picture.

**Before**you come the first class, please do this:Download the latest stable

`julia`

release for your OS.Download the

`VSCode Editor`

Check out Clement Mazet's materials. You must know this level.

We will be using Julia for this course.

Clement in his course will introduce you to things like the Unix Shell and the verion control system Git. Both of those are very useful - for this course, and for the rest of your life as a scientist.

What is Version Control? watch this 5 minute video. and go back to Clement's stuff if unclear.

This year your term project will be to replicate a paper published in an economics journal. Ideally this would be related to your field of interest. The requirements for choice of paper to replicate are:

Published version and replication kit is available online.

The paper to replicate must not use julia.

You must use julia for your replication.

Ideally your choice will involve at least some level of computational interest (i.e. more than an IV regression)

However, you can replicate a paper with an IV regression, but you have to go all the way to get the exact same results as in the paper. I.e. if the author typed the stata command

`ivreg2 lw s expr tenure rns smsa _I* (iq=med kww age), cluster(year)`

you will have to write (or find) julia code which will match all output from this, including standard errors. I do not recommend to do this.

You need to set up a public github repository where you will build a documentation website of your implementation. You'll learn how to do this in the course.

I encourage you to let the world know about your replication effort via social media and/or email to the authors directly. This is independent of whether you were able or not to replicate the results. Replication is not about finding errors in other peoples' work. If you are able to replicate some result in julia, this may be very interesting for others.

Your grade will be 60% homeworks, 40% term project.

There are some excellent references for computational methods out there. This course will use material from

**Fackler and Miranda**(2002), Applied Computational Economics and Finance, MIT Press**Kenneth Judd**(1998), Numerical Methods in Economics, MIT Press**Nocedal, Jorge, and Stephen J. Wright**(2006): Numerical Optimization, Springer-Verlag**Kochenderfer and Wheeler**(2019), Algorithms for Optimization, MIT Press**A Gentle Introduction to Effective Computing in Quantitative Research**- What Every Research Assistant Should Know, Harry J. Paarsch and Konstantin Golyaev

Programming languages and why

`julia`

Talk through homework requirements

Talk through term project requirements

Show where material is and do first set of slides.

`julia`

setup and Getting Started

Setup environment

Tools and Editors

Examples

Types

Essentials

Speed

Data and Statistical Packages

Integration and Function Approximation

Numerical Integration

Monte-Carlo integration

Gaussian Quadrature

Multidimensional Quadrature

Quadrature with correlated shocks

Function Approximation

Polynomial Interpolation

Basis functions and Coefficients

Regression as Approximation

Colocation Methods

Multidimensional Approximation

The Smolyak Grid

Optimisation 1

Intro

Conditions for Optima

Derivatives and Gradients

Numerical Differentiation

JuliaOpt

Optimisation 2

Bracketing

Local Descent

First/Second Order and Direct Methods

Constraints

JuMP.jl

Numerical Dynamic Programming

Review of DP theory

Different Solution methods for different cases

Discretization

Parametric approximation methods basically Function Approximation

The Endogenous Grid Method

Finite time vs inifinite horizon models

Solving the Growth Model in 7 Different ways

Constrained Optimisation Applications as MPECs

What is an MPEC?

How can we cast constrained problems as MPECs?

Applications:

MPEC on John Rust's Bus Engine Replacement

The Berry-Levinsohn-Pakes (BLP) paper as constrainted optimization problems

HPC

`julia`

Brief intro to parallel computing concepts

Parallel computing with julia

GPU computing with julia

Rust Bus Model and Dynamic Discrete Choice

Intro to Machine Learning with julia

The julia ML stack

julia ML applications

We will try to honour Science Po's anti-plagiarism policy:

Plagiarism occurs when a student submits work that does not allow one to distinguish the student's own thoughts from those of other authors: it can be characterised by the absence of citation of a group of consecutive words (five or more), by reformulation or translation, or by copying directly." (article on intellectual honesty)

Reuse and building upon ideas or code are major parts of modern software development. As an economist writing code, you will (hopefully) never write anything from scratch. This class is structured such that all solutions are public. You are encouraged to learn from the work of your peers. As I said above, I won't hunt down people who are simply copying-and-pasting solutions, because without challenging themselves, they are simply wasting their time and money taking this class.

Please respect the terms of use and/or license of any code you find, and if you reimplement or duplicate an algorithm or code from elsewhere, credit the original source with an inline comment.

The copyright notice to be included in any copies and other derivative work of this material is:

`Copyright 2021 Florian Oswald, Sciences Po Paris, florian.oswald@gmail.com`

Thank you.

This is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License