Skip to content
All library documents

Testing Asset-Pricing Models Across CFNAI Business-Cycle Periods

Article Quant Q&A · Author: Dave Fran
Read the original

Summary

The document considers how business-cycle conditions may affect the performance of asset-pricing models applied to industry portfolios. It proposes dividing monthly excess returns into recession and non-recession subsamples, with recession dates identified using the Chicago Fed National Activity Index (CFNAI).

The central concern is that recession episodes vary in length, leaving some subsamples with few monthly observations and potentially weak inference in separate regressions. The author asks about statistical procedures that could improve the power of comparisons across these periods. No method, regression result, or empirical conclusion is supplied. The question highlights a design tradeoff: regime-specific estimates can be difficult to compare when the periods are short or uneven, and any inference must account for the limited sample and the way recession intervals are defined.

Key ideas

  • The proposed analysis tests asset-pricing models on industry portfolios across business-cycle conditions.
  • CFNAI recession dates are used to define return subsamples.
  • Unequal recession lengths can leave too few monthly observations for some regressions.
  • The document raises the need for stronger inference but does not specify or evaluate a statistical procedure.

Tags

EquitiesStatisticsFactor investing
Full text
# Using CFNAI index for identifying sample periods


# Using CFNAI index for identifying sample periods












I'm doing my Thesis on Asset pricing models and I would like to find out the effects of business cycles on the performance of asset pricing models for industry portfolios.

My initial idea was to isolate sub-samples of excess industry returns based on the recession periods identified by the CFNAI (Chicago Fed National Activity Index). However, each recession period has different lengths. If I only use monthly excess return data, there would be very low number of samples for certain periods. So is there any suggestions on statistical procedures that would allow me to give a more powerful inference based on the sub-sample regression results?

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.