Data-Driven Optimal Portfolio Selection Project
This project aims to evaluate, under simple constraints, an optimal portfolio choice for both retirement and non-retirement contexts. Investing choices are limited to U.S. equities, U.S. bonds, and U.S. short-duration treasuries, with the assumption that these broad classes alone (likely purchased as common index funds) can construct a reasonable portfolio for individual investors with arbitrary time horizons. The goal is to provide a solid data-driven reference point for reasonable asset-class allocation as a function of age or horizon, without diving into the other complexities of an individual investor’s particular situation.
I have separate pages diving into the dataset, statistical framing, optimization strategy, and a (slightly interactive) page of results.
3 assets U.S. stocks, bonds, and T-bills
1927-2025 historical real returns
Retirement and Non-Retirement investment contexts
Explore the project
Project Goals
A basic overview of the project framing, and what financial questions I wanted to answer.
Dataset and Simulation Strategy
Details covering the historical dataset and how I sampled it to rank portfolio choices.
Optimization Algorithms
A mathematical description of the algorithm I designed to reliably arrive at an optimal glide path.
Why Trust My Results?
The ML techniques and validation strategy I employed to produce robust, trustworthy results.
Core Results
What interesting results did this project produce?