Thread by @ReformedTrader

1/ Constructing Long-Only Multi-Factor Strategies: Portfolio Blending versus Signal Blending (Ghayur, Heaney, Platt)

"For low-to-moderate factor exposures, portfolio blending works better than signal blending due to interaction effects between factors."

papers.ssrn.com/sol3/papers.cf…

2/ "We describe a basic framework for constructing average exposure-matched portfolios using the signal blending and portfolio blending approaches.

"Securities are overweighting and underweighting based on a percentile threshold (e.g., top 50% based on factor z-scores)."

3/ For *low* factor exposure portfolios, stocks are included if z-scores are in the top half.

"Given the large overlap and similarities in active weight, we expect the portfolios' performances to be similar. However, the portfolio blend generates a higher IR (0.89 vs. 0.59)."

4/ "The securities held exclusively in the portfolio blend (Area 3, Panel A, Fig. 3) have a higher average active return than for those held only in the signal blend (Area 2).

"The portfolio blend securities also reduce active risk, unlike securities for the signal blend."

5/ "The finding that securities held only in the portfolio blend generate a positive average active return can be largely explained by interaction effects between value and momentum.

"These contextual relationships are also present to varying degrees in other factor pairings."

6/ "As a next step, we create portfolios that target a much higher level of factor exposures.

"The signal blend now achieves these high exposures with better diversification, as represented by a higher count percentage of universe securities and lower active share.

7/ "The signal blend generates 13% higher IR, as shown in Table 8, with the improvement driven by lower active risk.

"In this case, the interaction effects are overwhelmed by the high concentration and stock-specific risk embedded in the portfolio blend."

8/ "At low-to-moderate [high] levels of factor exposure (starting with the top 50% signal blend portfolio), the portfolio [signal] blend is superior.

"Except for the momentum-volatility combination, similar patterns are found for various two-factor combinations."

9/ Quality = gross profits / total assets
Volatility = 1/σ of trailing 12-month daily total returns

"Due to value’s neg. correlation with both momentum and quality, we observe a lower exposure to value. As previously, we match exposures across the 3 factors in both approaches."

10/ "Figure 5 plots the average factor exposure against information ratio for the three- and four-factor portfolios.

"Overall, we see a pattern that is quite consistent with the pattern observed with the value-momentum combination."

11/ "The results in the U.S. are generally validated when analyzed in other developed and global emerging markets."

12/ "Portfolio blending is more transparent: the building-block framework facilitates cause/effect performance attribution.

"For example, momentum has higher active risk than value (9.32% vs. 6.85%), suggesting that diversification could be improved by balancing active risk."

13/ "The way a manager defines, constructs, & implements factor strategies can have a meaningful impact on correlation structure and efficiency of factor capture.

"It is our view that choosing an approach is more appropriately made in the context of a given investment process."

14/ Related research

Long-Only Style Investing: Don't Just Mix, Integrate

https://twitter.com/ReformedTrader/status/1127035098849660928