Labs
Throughout the course we will make heavy use of Python and its SciPy ecosystem. We provide a set of learning labs as Jupyter Notebooks.
Tooling
We showcase the basics of Python programming and point students to useful resources to study further. There are numerous excellent introductory lectures on Python programming in economics available online. Among them is Python programming for economics and finance and we will sample some of their material for our labs.
Linear equations
We explore different solution algorithms to solve linear equations. We look at direct methods building on L-U decomposition as well as iterative methods. We study the impact of ill-conditioned matrices on the performance of algorithms. In the process, we learn some basic ideas behind testing and benchmarking numerical routines.
Nonlinear equations
We explore different solution algorithms to solve nonlinear equations. We start with the bisection method. We then turn to function iteration before exploring Newton’s method for nonlinear equations. Finally, we look at Quasi-Newton methods and benchmark their performance in solving a standard Cournot problem. We briefly discuss some criterion to choose the right algorithm for the problem at hand.
Optimization
We discuss the key attributes of optimization algorithms that determine the choice of a suitable optimization algorithm. We explore the role of noise in the criterion function and ill-conditioning for different groups of optimizers: local vs. global, derivative-based vs. derivative-free. We conclude with some programming exercises for nonlinear least squares problems and implement a simple maximum likelihood estimation.
Integration
We examine different strategies for the numerical integration of functions. We discuss rules based on Newton-Cotes quadrature formulas, Gaussian quadrature, and Monte Carlo methods in the uni-dimensional and multi-dimensional case. We conclude by comparing the performance of each approach under different scenarios.
Approximation
We study the function approximation using polynomials. We combine different strategies for the interpolation nodes and basis functions to study how they interact to determine the approximation’s overall quality. We use this as an opportunity to iteratively develop a function that allows to combine the different ingredients to set up an interpolator. Finally, we extend the ideas to the case of multivariate interpolation.