Combining Signals with Different Forecast Horizons

Summary

The document asks how to combine a momentum signal whose strongest predictive power is at a three-month horizon with a valuation signal that works best over one year. It proposes scaling the valuation signal by dividing it by four when using both in a three-month portfolio, then asks whether this creates statistical problems for portfolio optimization.

No answer, method, or empirical evidence is provided. The question highlights a real modeling issue: scaling a signal by a horizon ratio does not by itself establish that its expected return, uncertainty, or covariance with other signals is appropriately adjusted. The document leaves unresolved how to align forecasts to a common holding period and how to estimate the inputs used in optimization.

Key ideas

• The document asks how to combine momentum and valuation signals with different predictive horizons.
• It proposes dividing the one-year valuation signal by four for a three-month portfolio.
• Simple horizon scaling may not align expected returns, risk, or signal covariance.
• The document provides no answer or evidence about the statistical effects.

Tags

# Trading Signals with Different Lags


# Trading Signals with Different Lags












I have a momentum signal that gives best predictive power at 3 months and a valuation signal that gives best predictive power at one year. If I combine for a 3 month horizon by "interpolating" the valuation signal i.e. dividing the valuation signal by 4, what if any statistical problems will this generate for the portfolio optimization?

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.