Comparing Portfolio Blending and Signal Blending When Constructing Multifactor Portfolios (Summary)

Over recent years, investors have become increasingly interested in constructing their investment portfolios by considering multiple factors, combinations of rules-based approaches—for example, value and momentum—rather than any one factor on its own. The reason for this is that multifactor strategies tend to lead to higher information ratios (IRs), a measure of relative risk-adjusted returns, than strategies using individual factors.

There are two main ways to construct a long-only multifactor portfolio. The first is a two-step approach called “portfolio blending.” This starts by constructing individual factor portfolios and then combines them into a blended portfolio with the desired weighting of factors. The second is a single-step security-scoring approach called “signal blending.” In this approach, each security in the universe is given a score or ranking (i.e., a signal) for each factor, and these signals are blended into an aggregate value. The composite signals are used to build a single portfolio with the desired traits.

Advocates of signal blending argue that it delivers better risk-adjusted returns—or investment efficiency—because it favors securities that have a balanced positive exposure to all the desired factors over securities that rank highly on one factor but poorly on another. The counterview is that these differences are negligible and portfolio blending is better for meeting other investment goals, such as transparency in performance attribution.

The authors first set out to compare how the two approaches affect risk-adjusted returns. One potential hurdle is that portfolios constructed using the different methods may not have the same exposure to each desired factor—which may account for differences observed in efficiency. To mitigate this issue, they propose a straightforward framework for creating exposure-matched portfolios.

They construct these portfolios at both low and high levels of factor exposures—that is, at low and high levels of tracking error, which is the difference between the performance of the portfolio and its benchmark. When factor exposure is low, the percentage overlap of securities between the two types of portfolios is higher; as it increases, each portfolio has more exclusively held securities. They then use these as the basis for a fair comparison of the signal- and portfolio-blending approaches.

Next, the authors conduct an empirical analysis using equally weighted data from the Russell 1000 Index (a benchmark US equity market index) from January 1979 to June 2016. They construct exposure-matched portfolios by considering two factors—value and momentum—and compare their performance over the period. They extend this analysis to a three-factor strategy (using quality as the third factor) and a four-factor strategy (by adjusting the volatility exposures on the three-factor portfolios). They also conduct the analysis on developed markets, excluding the United States, and emerging markets.

Finally, the authors look at multifactor investment goals other than returns. Key among these is transparency in performance attribution to evaluate the sources of risk and return in an investment strategy.

The authors use their framework for constructing exposure-matched multifactor portfolios to lay out what they expect to find. When factor exposures are low, they expect both the portfolio blending and signal blending to generate similar IRs. When factor exposures are high, the signal blend portfolio should show superior performance.

These expectations are partly—but not entirely—borne out in the empirical analysis. As anticipated, at high levels of factor exposures, signal blending generally delivers higher investment efficiency. For example, the signal-blended two-factor portfolio based on the Russell 1000 data has an IR of 0.61—exceeding the portfolio blend’s IR of 0.54.

Contrary to expectations, the corresponding IRs at low to moderate levels of factor exposures are 51% higher for the portfolio blend than the signal blend (0.89 compared with 0.59), a result which is statistically significant. The superior performance of portfolio blending is also observed for various two-, three-, and four-factor combinations in different geographical segments.

This result can largely be explained by the interaction effects between the factors. The authors find that at low levels of tracking error, securities held only in the portfolio blend—that is, those with offsetting factor exposures—reduce active risk and improve active return. This phenomenon does not hold when tracking error rises because the percentage overlap of securities between the two types of portfolios falls. At higher levels of factor exposure, the benefits of interaction effects in the portfolio blend are superseded by greater stock-specific risk. In these cases, the signal-blending portfolio offers better diversification and information ratios.

Table 3. Portfolio Performance vs. the Russell 1000, January 1979–June 2016

Much of the existing research on multifactor portfolios argues that investors who use portfolio blending rather than signal blending sacrifice returns for other investment objectives, such as transparency and flexibility. The results of this study dispute that such trade-offs exist. The authors note that from an investment strategy perspective, factors are usually best captured at low to moderate levels of return predictability. In practice, investors generally avoid implementing multifactor strategies at high levels of factor exposure—the only case in which signal blending is empirically shown to generate higher information ratios.

Another advantage of the portfolio-blending approach is that it provides a direct measure of active risk and active return for the individual factor portfolios that are combined. Many portfolio managers are wary of investment approaches that do not allow them to conduct a detailed attribution analysis, and signal blending lacks transparency in performance attribution. Other investment objectives that should be considered when contrasting the two approaches include simplicity and implementation flexibility.