Constructing Long-Only Multifactor Strategies: Portfolio Blending vs. Signal Blending
- Khalid Ghayur, CFA, Ronan Heaney, and Stephen Platt, CFA
- Financial Analysts Journal
- A version of this paper can be found here
- Want to read our summaries of academic finance papers? Check out our Academic Research Insight category
What are the research questions?
The heightened interest in factor investing has been accompanied by a corresponding focus on the nuts and bolts of constructing multifactor portfolios. There are essentially two ways to go: In a one-step process, single factor signals are blended into a composite signal and one multifactor portfolio is created from the individual stock composites. Or in a two-step process, single factor portfolios are created from single-factor signals and then blended into one multifactor portfolio. Either way, a multifactor portfolio is the result. The question as to which process is superior has been debated in the academic literature and in practice. The weight of the evidence to-date falls on the side of signal blending. However, the results of this study challenge that conclusion. (AA finds similar results here).
- Do factor portfolios built using “signal blending” dominate portfolios built using a “portfolio blending” approach?
What are the Academic Insights
- YES and NO. It depends on the level of active risk and the amount of factor exposure exhibited by each type of portfolio. Portfolio blending produced higher IRs at low/moderate levels of factor exposure and active risk. Signal blending produced higher IRs at high levels of factor exposure and active risk. Results are presented in the chart below. As average exposure increases, portfolio blending produces higher IRs until the average exposure increases beyond approximately .9. For average exposures beyond .9, signal blending is superior. However, signal blending exhibits higher active risk as average exposure increases until approximately .9, at which point portfolio blending exhibits exponential increases in active risk. Portfolio blending remains superior at levels of active risk of approximately 8.0 or lower, at which point the two processes appear to converge.
Why does it matter?
The authors create (average) exposure-matched portfolios at varying levels of factor exposure using each multifactor construction process. The analysis is completed for one, two, three and four-factor portfolios. The US results were validated using international and emerging market data over the periods 1995/1998 to 2016. The cross-validation and long time series both lend a pretty fair amount of credibility to the results.
The conventional wisdom holds that investors are penalized in terms of efficiency when they adopt a portfolio blending (or signal blending for that matter) approach if other issues such as transparency in construction, transparency in return attribution, implementation costs, factor representation and capacity are overriding or conflicting objectives. Defining these tradeoffs should influence the decision as to whether blending signals or blending portfolios is optimal for the investor interested in smart beta or multifactor strategies.
The authors say it best:
…..the focus of the current debate is misplaced. The existing literature makes an attempt to reach a general conclusion about whether portfolio blending is a superior approach to signal blending or vice versa. We argue that such an assessment is more appropriately made in the context of a given investment process.
The most important chart from the paper
Long-only multifactor strategies may be constructed by combining individual-factor portfolios (portfolio blending) or by combining individual-factor signals into a composite signal to construct the portfolio (signal blending). To compare these two approaches, we present a framework for building exposure-matched portfolios. In empirical tests on global equity markets, we find that, generally, portfolio blending generates higher information ratios for low-tomoderate levels of tracking error. At high levels of tracking error, signal blending delivers better risk-adjusted performance. These results generally hold for various factor combinations, and they have important practical implications for investors considering the implementation of multifactor smart-beta strategies.