Evaluating Trading Signals for Collaborative Portfolio Exchange

Summary

The document raises the problem of exchanging trading signals between collaborators who want to preserve some confidentiality. A signal should not be judged by its standalone predictive strength alone: its value depends on how it overlaps with the recipient’s existing signals and whether it improves the portfolio’s overall risk and return characteristics.

The proposed starting point is to share historical signal results without disclosing the signal itself, then assess whether the other party’s signal adds value within a mutually agreed portfolio framework. The question asks whether Pearson correlation is an adequate way to describe alpha and which portfolio theory to use, but it provides no answers, experiments, or recommended framework. As a result, it is most useful as a problem framing note: signal evaluation needs measures of predictive performance and portfolio contribution, while the suitable disclosure and risk model remain unresolved.

Key ideas

• A signal's usefulness depends on its interaction with the recipient's existing signals.
• Standalone predictive strength does not show whether a signal improves portfolio risk.
• Historical results can be shared while withholding the signal construction itself.
• Correlation is posed as a possible measure, but the document does not establish that it adequately captures alpha.
• The appropriate portfolio framework and disclosure terms remain open questions.

Tags

# Mathematical techniques for Trading signals


# Mathematical techniques for Trading signals












I'm trying to come up with a reasonable and mostly mathematical way to trade signals between two people with interests in collaboration but still wary and skeptical. The idea being that you start with trading the signals with weakest alpha.

However, it's not as simple as that, as the signal might not be appropriate given your own current collection of signals because of overlap and portfolio risk. Ie, you want to get a signal that will enhance not just your own predictive power, but also improve your over all portfolio risk.

Some questions:

- What would be the best way to measure and present the alpha in the signal you wish to trade? Would pearsons correlation be reasonable?

- What would be the best way to represent what you're looking for? Currently I'm thinking sharing the results of my signals on historical data (without revealing the signals themselves), and if the other side of the trade has a signal that can improve them within the context of agreed upon portfolio theory, than we're good to go.

- Any recommendations on the portfolio theory to use here?

Note this would be a private transaction, not a public one, so leaking a little bit of information is OK.

Thoughts?

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.