What Multi-Factor ETFs & Active Managers Don’t Have in Common

By Joe Smith, CFA, CLS Investments

Investors continue to focus on factor investing as more ETF issuers launch single and multi-factor ETFs that target risk premiums through rules-based indexing. Multi-factor ETFs are particularly popular as advisors evaluate the merits of using them in place of traditional active management.

Despite the promise of combining multiple factors into a single portfolio exposure, many of these indices (and their ETF counterparts) are constructed differently and deliver their promises in varied ways. This can lead to questions during the due diligence process, including:

• How much exposure results from each factor being targeted in the portfolio?

• How consistent is the construction of these portfolios with the typical behavior of an active manager?

To answer these questions, we’ll review the current index construction methods for multi-factor ETFs to identify similarities and differences. We will also derive an alternative construction approach based on Bayesian statistics to capture how active managers incorporate factors into their decision-making. Finally, we will compare hypothetical portfolios based on each construction method to compare and contrast their risk and return properties.

What Multi-Factor Indices Have in Common Today

The impetus behind combining factors into a single portfolio is an easy case to make for investors. Over long periods of time, combined factors can provide opportunities to beat the market with odds that offer a statistical edge in both up and down markets.

Most multi-factor indices or ETFs tend to follow two primary methodologies: factor-mixing or factor-tilting. Let’s review a simple example: how do three stocks (JPMorgan, Apple, and Best Buy) show up in each type of multi-factor index based on their factor characteristics measured on December 31, 2016?

We generally use three measures for value: price-to-earnings, price-to-sales, and price-to-book; 12 minus one month price return for momentum; one-year volatility for minimum volatility; and two measures for quality: return on assets and long-term debt to capital. In addition, we measure the inverse of the log- market capitalization to assess the size as a factor characteristic. Each factor characteristic has been normalized into a Z-score relative to the entire universe of stocks in the MSCI USA Index. We use a cumulative distribution function of a standard normal to then generate single-factor scores for each stock in the index. These factor scores can be thought of as the probability of a given stock exhibiting the factor in question.

Factor-mixing involves taking a number of single-factor indices and weighting them (and their underlying holdings) to determine the final list of securities and their allocations represented in the portfolio. In most cases, this will result in owning a considerable representation of the underlying benchmark due to allocating to multiple factors that tend to be diversifying or negatively correlated with each other (value and momentum, for example). As illustrated below, each stock’s representation of each factor index varies both by how strongly it scores on a factor and its overall market capitalization. Larger companies gain a greater weight even if their factor score is not as strong as other stocks in the parent index.

Factor-tilting takes a more bottoms-up approach by tilting towards various stocks based on their exposure to multiple factors simultaneously. This process involves combining each factor score multiplicatively to see what security holds up best to having representative exposure to each factor. As also illustrated below, this process tends to bias towards securities with higher factor scores across each factor, but still depends on each stock’s market-cap. In our example, Apple has strong factor scores for momentum, minimum volatility, and quality, and they nearly double the market-cap of JPMorgan.

Both methods generally yield similar results. Under either method, eligible securities are measured based on how they stack up against other stocks in the entire investment opportunity set. This means the definition of one factor has no bearing on the definition of another and is assessed independently. Ultimately, stocks more likely to score high on multiple factor definitions tend to have a greater emphasis in the final index.

Although there may be more simplicity in this, it begs the question as to whether a typical active manager would make decisions in the same way. We explore this topic in the next section.

What Multi-Factor ETFs and Active Managers Don’t Have in Common

Another long-made argument in the ETF industry is that multi-factor ETFs put investors closer to investment exposures that compete directly against traditional active managers. Yes, capturing the same risk premiums that active managers incorporate at a fraction of the cost is a net benefit to end investors. And yes, active managers do use some form of factor investors in search of alpha through security selection.

But, do active managers pick stocks that simultaneously have the best value, quality, and momentum characteristics, for example? Absolutely not. In fact, many active managers identify pools of stocks that meet basic characteristics largely influenced by their investment philosophies and processes. They then refine those pools of securities based on fundamental, technical, valuation, and risk information that can lead to identifying the best and worst stocks to hold.

To illustrate this point, we’ll apply this process to determine what stocks should be included in a multi-factor index. We will assume we are value managers who prefer owning value stocks that also exhibit momentum, quality, and minimum-volatility characteristics. As value managers, we first want to find stocks that score reasonably well as value (defined as having a value score higher than 0.50). This one rule, by definition, should automatically eliminate 50% of the original index universe. Within this refined universe, we then determine multi-factor scores that reflect the likelihood we’ve identified value stocks with reasonable momentum, minimum volatility, and quality factor scores.

In the case of JPMorgan, the stock exhibits strong value characteristics and modest exposures to momentum and minimum volatility, but not quality. Best Buy has an even stronger score to value and modest tilts to momentum and quality, but not minimum volatility. Apple’s value score is less than 0.5, so it is not considered for inclusion in the final index (or simply put, the probability of Apple being considered a value stock is less than 50%).

In statistical terms, active managers are Bayesians at heart when it comes to selecting securities for their portfolios. The decision to choose stock A over stock B is conditioned first on a prior set of assumptions about each security (in this case, being a value stock). But, it is updated after additional information enables a more informed decision about which stocks to emphasize and avoid (finding value stocks that also exhibit momentum, minimum volatility, and quality characteristics).

This certainly presents the question as to whether multi-factor ETFs in their current form do capture the “essence” of a typical active manager’s process.

The Curious Case of Multi-Factor Index Methods in Practice

Now we will examine how well factor-mixing and factor-tilting compare to a Bayesian approach. To accomplish this, we will evaluate the annual holdings of the MSCI USA Index and construct each multi-factor index based on the appropriate methodologies highlighted above going back to December 31, 2003. Each index is constructed and rebalanced at the end of each year in December. From there, we will compare each portfolio to the MSCI USA Index and to each other on a risk-adjusted basis. We will also compare a version of the multi-factor Bayesian method where the multi-factor score of each stock is adjusted relative to the median market-cap of stocks in the MSCI USA Index instead of individual market-cap.

The results suggest all three multi-factor indices have reasonable exposures to each factor targeted. Generally, the multi-factor mix and multi-factor tilt methods generate average factor scores higher than those of the parent universe. In most respects, however, going from factor-mixing to-factor tilting yields greater amounts of excess tilt to each factor. The factor-Bayesian methods each respectively yields much higher tilts towards value than the two previously mentioned strategies. That is intended with the construction of these portfolios. What is unique is their factor scores are on par with other with the other strategies mentioned, except quality, which is the laggard.

Regarding size exposure, the effects of weighting methods can be shown by applying each stock’s multi-factor score to either its own market-cap or another standard measure. These can yield different exposures to size as a factor. Should a multi-factor index or ETF have a bias to smaller companies as part of its delivery process or stay in line with the average market-cap of its parent index? Many index managers may argue that going up the cap spectrum helps ensure sufficient liquidity and capacity. However, active managers may argue that weighting securities by the greatest amount of available liquidity often reduces the opportunity since more market participants are efficiently pricing in prior information about the company. Hence, de-linking from a parent index can present opportunities that otherwise would have not been well-represented in the portfolio.

So, after all this work, which multi-factor method holds up best on a risk-adjusted basis? Our results show the most basic assumption about building multi-factor portfolios tends to hold. Combining factors that have been historically rewarded over time can help achieve an edge over the market. That result holds true for each multi-factor index strategy considered. The factor-mix method yields the lowest amount of risk-adjusted performance, followed by the factor-Bayesian method that is weighted based on market-cap. The factor-tilting and factor-Bayesian (based on median-cap) methods show the most promising results with nearly identical risk-adjusted performance.

In terms of assessing active returns, there is a stark difference in excess returns and tracking errors between the factor-tilting method and factor-Bayesian (median-cap) method. Although the factor-tilting method handily beats the benchmark, the amount of excess return per-unit of additional tracking error is far less in comparison to the factor-Bayesian (median-cap) method.

Despite the uptick in absolute volatility, the factor-Bayesian (median-cap) method still generates acceptable active results on an excess return and information ratio basis. We can also see the impact of de-linking from market-cap. The factor-Bayesian (median-cap) method generates an additional 2.51% in annualized excess performance relative to the performance profile of the factor-Bayesian (market-cap) method. Although one could argue that tapping into smaller companies will yield an additional premium, the process of selecting undervalued companies should naturally result in this type of behavior.

Key Takeaway – There is Still Room for Innovation in the Multi-Factor ETF Space

Multi-factor indices require further understanding of both their construction and the intention behind their methodologies. Combining factors in a single investment strategy can be a powerful way to achieve what many investors and advisors want in a fee-compressed world: capturing active-like returns within their beta. Such an endeavor can result in meaningful cost savings and more consistency over time if done right.

In our studies, we demonstrated that current multi-factor construction methods may be sufficient in moving towards that goal, but a Bayesian method of identifying stocks to own based on one set of factors and reconditioning based on additional factors that active managers utilize can be more true to the intentions of multi-factor indices. We also showed the ability to de-link a multi-factor index from each company’s individual market-cap can enable the index to capture additional excess return over time on a risk-adjusted basis by better emphasizing other securities that score well for inclusion in a multi-factor index. Over time, we believe such methods will become more mainstream among index providers and ETF manufacturers alike and can ensure future products look more in common with their active management counterparts.

Joe Smith, CFA, is a Senior Market Strategist at CLS Investments, a participant in the ETF Strategist Channel.

Disclosure Information

This information is prepared for general information only. Information contained herein is derived from sources we believe to be reliable, however, we do not represent that this information is complete or accurate and it should not be relied upon as such. All opinions expressed herein are subject to change without notice. The graphs and charts contained in this work are for informational purposes only. No graph or chart should be regarded as a guide to investing. 1123-CLS-1/19/2018