Skip to content
All library documents

Unsupervised Learning for Asset Allocation

Article Quant Q&A · Author: develarist
Read the original

Summary

The document asks whether portfolio optimization can avoid using empirical estimates such as historical returns and volatility. Its brief response points to unsupervised learning as a way to investigate relationships among assets, including correlation-like structure, and cites research associated with Marcos López de Prado.

This is only a pointer, not a worked method. It gives no model specification, portfolio construction procedure, empirical evidence, or comparison with supervised allocation approaches. The suggestion therefore introduces a possible research direction rather than showing that unsupervised learning eliminates the need for data: discovering relationships among assets still requires observations or some other representation of those assets. Readers would need to consult the cited work and evaluate its assumptions and results before applying the idea.

Key ideas

  • Unsupervised learning is suggested as a tool for exploring relationships among assets for allocation.
  • The response highlights correlation-like structure as one kind of relationship to investigate.
  • The document provides a citation but no implementation details or performance evidence.
  • Unsupervised methods do not, by themselves, establish that portfolio construction can avoid data.

Tags

Machine learningPortfolio constructionStatistics
Full text
# Unsupervised learning for portfolio construction


# Unsupervised learning for portfolio construction












Are there techniques or models in finance that (unlike supervised learning where input data such as returns and volatility is estimated making the asset allocation data-driven) allow for portfolio optimization without using empirical data for an input?

## Answer by Vitomir (score 1)

https://quant.stackexchange.com/a/54940

I quote a very famous paper from De Prado:

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2708678

In general, unsupervised learning is a very powerful tool in asset allocation especially with regards to exploring deep unsupervised relations among different assets (i.e. sort of correlations).

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.