Combining Trading Signals and Portfolio Optimization

Summary

The document raises a portfolio-construction question: how should raw signals that produce instrument positions be combined into a final portfolio? It considers averaging or weighting signals within each instrument, then applying a portfolio method such as hierarchical risk parity or critical line algorithms across instruments. It also asks whether each signal could instead be treated as a separate portfolio asset.

A further issue is preserving the intended exposure of multi-instrument signals, such as spreads, when optimization might assign different weights to their legs. The document asks whether such a strategy should be represented as a distinct instrument. It supplies no answers, empirical comparisons, or recommended workflow, so it serves as a statement of design questions rather than a method. Readers should not infer that it endorses averaging, a specific optimizer, or treating spreads as standalone assets; those choices remain unresolved in the text.

Key ideas

• The document distinguishes signal-level position aggregation from portfolio allocation across instruments.
• It asks whether signals should be combined by averaging or weighting before portfolio optimization.
• It considers treating individual signals as assets for optimization, but gives no recommendation.
• Multi-instrument signals raise a constraint: changing leg weights can alter the intended strategy exposure.
• The discussion poses implementation questions and provides no empirical evidence or resolved approach.

Tags

# Interaction between raw position signals and portfolio optimisation methodologies


# Interaction between raw position signals and portfolio optimisation methodologies












I'm trying to get my head around how the various aspects of constructing a final position generally interact and wonder whether anyone could expand on my (tentative) understanding currently.

As I see it currently, one may have a series of 'signals' which give an output of a 'position' (for the sake of simplicity, let's say this position is just {-1 .. 1}) on a set of instruments. However from here, I'm not entirely clear on any 'standard' method of aggregating these. Some thoughts being:

- It seems like (from things like Advances in Financial Machine Learning) that one can simply average (potentially with weighting) these into an aggregate final position.

- However I'm not sure where something like HRP/CLA comes into it, would one use this across different instruments after using averaging across signals within an instrument?

- Or would one use your portfolio optimisation across your set of signals treating each one as an individual 'instrument'?

- How does this work if your signal has a position in multiple instruments? Obviously you don't want the different legs to be weighted differently as then you're doing something different to your goal. Does one treat this spread as a distinct 'instrument' from each underlying leg?

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.