Meeting the challenge of cyclical factor premia
Urs Raebsamen, Senior Equity Specialist, Systematic and Index Investments team
Both academic as well as practitioners’ research have identified factor premia—or styles—that are widely believed to add value in the long term. The most commonly accepted are value, momentum, quality, low risk and size (small minus large). That said, style performance and correlation can vary over time and through investment cycles.
For example, growth and momentum factors dominated global equity markets in 2017, while value and low risk investing underperformed. This continued into the beginning of 2018 and even well into February, as inflation scares triggered a short market correction. Only in March, when geopolitical risks started to dominate investors’ agenda, styles started to behave differently. This serves as a powerful reminder that a static bet on one style can lead to suboptimal results.
We believe this is particularly relevant now as risks of a regime shift in markets have increased. The rate of acceleration in global growth has moderated in recent months, and global demand growth rates are less synchronized. If geopolitical risks prevail, central banks’ loose monetary policies gradually unwind, and the economic cycle matures, it is reasonable to assume that equity markets’ will remain volatile.
The solution: Isolate the specific components
In essence, investors have three ways of dealing with the challenges of the cyclicality of factor premia and changing correlations: a) active timing of factor premia, b) blending factor premia efficiently, or c) isolating the specific value-add¬ing components of the factor premia and combination thereof. We presented our views on blending factor premia in the last issue of Panorama. In this edition, we will discuss the isolation of the specific components.
The basic idea of isolating the specific components is to build model portfolios that have true exposure to the respective factor premium they aim to harvest. For instance, two stocks can both appear to be driven by momentum. However, while one stock may be a true momentum stock, the other may just be a high beta stock having done well in upward markets. When constructing a momentum factor premium portfolio, we would want to only consider the true momentum stocks, not the ones that happen to have been correlated with momentum.
Our proprietary investment approach aims to achieve exactly that. Our multifactor models are industry-specific and consist of a series of factor premia-based investment themes—called return drivers: Valuation, Capital Usage, Profitability, Growth and Market Behavior. Through our refinement process, we aim to isolate the specific components of our industry-specific multifactor models. The resulting refined return drivers represent pure insights into a factor premium and have expected cross-correlations and correlations to systematic risks that are close to zero. This should allow for a more efficient use of the risk-budget, lower drawdowns and therefore more consistent alpha contributions.
Exhibit 3, based purely on out-of-sample data—demonstrate that our refined return drivers exhibit the aforementioned desired characteristics such as ex ante correlations to each other that are more stable and much closer to zero compared to the ones of raw return drivers. Moreover, refined return drivers on average achieve higher risk-adjusted returns and lower drawdowns than raw return drivers. Therefore, refined return drivers offer a more efficient alpha source.
Exhibit 4, plotting the refined return drivers of our live global multifactor model, demonstrates that none of the return drivers exhibited significant drawdowns and their combination produced a stable and value-adding alpha signal.
Besides the ability to create more stable and lowly correlated alpha signals, a proprietary approach to constructing factor premia indices is flexible in terms of further enhancements. In particular, non-conventional factors for instance based on ESG-related data, and improved ways to identify value-adding factors and methodologies, such as machine learning, can be applied.