Computationally Efficient Portfolio Optimization Algorithms

Group Members

Nick Falshaw – Systems Engineering – nick.falshaw@nick-falshaw

Peter Fadlovich – Systems Engineering – peter.fadlovich@wustl.edu

Moshe Leonard – Systems Engineering – ethan.leonard@wustl.edu

Overview

We implemented four machine learning algorithms to choose an optimal portfolio from a predetermined list of stocks. The four algorithms used are affinity propagation (AP), agglomerative clustering (AC), variational autoencoder (VAE), and principal component analysis (PCA). AP and AC are clustering methods, and VAE and PCA are volatility optimization algorithms. These algorithms are computationally efficient enough to be run from a personal laptop or even a mobile phone, so a web application was created to run these algorithms and display the optimal portfolio that was chosen. Over the four year period from 2018 through 2021 the average annual returns for each algorithm are as follows: AP: 36.76%, AC: 18.43%, VAE: 21.36%, and PCA: 12.89%. The average annual return for the S&P 500 for this same time period was 18.55%

Link to full report: https://docs.google.com/document/d/1Pxa744t7yGXfqvExy5hF–T-lru3KBGy8G_99yJvWn0/edit?usp=sharing