Research Affiliates Compares Competing Low-Vol Approaches

Research Affiliates Low Volatility Equity Research

The millennium is still young, but investors have already suffered a pair of massive selloffs in the past 15 years: First, 2000’s dot-com meltdown saw the S&P 500 fall by 44%; then less than a decade later, the great recession saw stocks plunge by more than 50%. This historic volatility understandably stoked investors’ interest in low-volatility strategies, but after a six-year bull market for stocks, Research Affiliates’ Feifei Li thinks investors are “sleepwalking again – possibly toward a cliff.”

Ms. Li co-authored a recent white paper with Research Affiliates colleague Engin Kose titled “Black Ice: Low-Volatility Investing in Theory and Practice.” The paper looks at the competing approaches to low-volatility portfolio construction – minimum variance and heuristics – and finds that while the former may seem preferable in theory, the latter has certain practical advantages.

Defining the Differences

The minimum-variance approach uses a “numerical optimizer” to select stock weights in a manner designed to minimize expected portfolio volatility in the future. Heuristic approaches, by contrast, typically use a common risk measure, such as the beta coefficient, to screen out particularly volatile stocks, and then weights are assigned to the remaining stocks according to market cap or an inverse “company-specific risk measure.”

The minimum-variance approach is more statistically robust, at least in theory, but Ms. Li says portfolios built with a minimum-variance approach may suffer from “implementation shortfall.” Heuristic approaches, by contrast, are “significantly superior in terms of transaction costs and valuation features.” They also hold up better under self-imposed constraints.

Risks and Constraints

Regardless of whether low-volatility portfolios are constructed using minimum variance or heuristics, they inherently face several types of concentration risk:

  • Geographic/ regional concentration
  • Sector concentration
  • Individual stock concentration

These concentration risks are inherent since certain regions, sectors, and stocks are themselves inherently more or less prone to volatility. The dangers of being overly concentrated are obvious, and low-volatility investors typically seek to mitigate these risks by imposing constraints – but these constraints can result in low-vol portfolios more closely resembling cap-weighted indexes, defeating their whole purpose.

Empirical Data

Ms. Li and her co-author conducted an empirical study of minimum-variance and heuristically constructed low-volatility portfolios across the following regions:

  • United States (1967-2014)
  • Developed markets (1987-2014)
  • Emerging markets (2002-2014)

The authors then looked at how adding variance constraints impacted portfolio turnover, returns, volatility, tracking error, and more. Some of the key takeaways:

  • Among low-vol portfolios constructed using minimum variance, U.S. returns fell when capacity constraints were added, but developed and emerging market returns improved.
  • Returns and Sharpe ratios typically declined as constraints were added.
  • Baseline returns for minimum variance and heuristic U.S. portfolios were the same, but minimum variance had the higher Sharpe ratio – by contrast, heuristic-portfolio returns were much higher than minimum variance for developed and emerging markets.

Theory vs. Practice

Overall, Ms. Li seems to prefer minimum variance in theory, but heuristics in practice. Adding constraints makes minimum-variance portfolios closely resemble cap-weighted indexes, while constrained portfolios designed heuristically are better able to maintain their low-vol characteristics, according to Ms. Li.

Past performance does not necessarily predict future results.

For more information, download a pdf copy of the white paper.

2 Comments

Add a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.