Using Macroeconomic Principal Components to Predict Stock Returns

Summary

The document describes a proposed study of whether principal components extracted from macroeconomic variables can help predict stock returns. The researcher plans to regress returns on a small number of components and examine several stock styles, including small-cap, value, and momentum portfolios. The central unresolved modeling choice is the lag at which the components should enter the return regressions.

The text provides no completed method for selecting lags, fitted results, or evidence that the components predict returns. It also indicates that the components may later be used in a separate regression, but the specification is cut off. Accordingly, this is a research question and outline rather than a demonstrated forecasting strategy. Any conclusions would depend on choices not addressed here, including the macroeconomic variables, component construction, lag selection, sample period, and evaluation design. The document offers no discussion of out-of-sample testing or controls for data mining, so it does not establish that the proposed relationships would generalize.

Key ideas

• The proposed analysis extracts principal components from a set of macroeconomic variables.
• It tests return predictability across small-cap, value, and momentum styles.
• Choosing the lag for the components is the researcher’s main stated difficulty.
• The text supplies no results, lag-selection procedure, or evidence of predictive performance.
• The later regression specification is incomplete.

Tags

# Predicting stock returns using principal components of macroeconomic variables


# Predicting stock returns using principal components of macroeconomic variables












I'm trying to detect return predictability by regressing stock returns on the first couple of principal components of a set of macroeconomic variables. I'm doing this for different stock styles such as small, value and momentum stocks. Since I don't know at which lag I should insert these principal components, I'm hoping to find someone who can help me out.

Later on I would use these principal components in another regression that looks like this:

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.