Journal Papers

A Study of Low-Volatility Portfolio Construction Methods

By Jason Hsu Tzee Chow Feifei Li Li-Lan Kuo

SEPTEMBER 2014 Read Time: 30 min

Executive Summary

Although low-volatility investing underperforms cap-weighted indexing in strongly uptrending markets, it generally provides markedly superior risk-adjusted returns in the long run. Setting aside its possible usefulness in market timing, the pertinent questions for long-term investors are whether to include a low-volatility portfolio in the strategic asset allocation and, if so, whether to opt for a minimum-variance approach or another set of portfolio construction rules. This practitioner-oriented article contributes to the extensive literature by comparing and contrasting optimized and heuristically constructed low-volatility strategies on the basis of their simulated long-term performance; sector, country, and stock concentrations; and implementation features such as turnover, liquidity, and capacity. Investors and consultants will find this research informative when deciding between minimum-variance and other rules-based approaches in order to narrow the selection of a low-volatility fund.

Using the most extensive historical data available from CRSP and Datastream, the authors build low-volatility portfolios of large companies in the United States (1967–2012), global developed markets (1987–2012), and emerging markets (2002–2012). They determine that all the portfolios examined, when compared to a cap-weighted index, have a lower exposure to the market risk factor and added exposures to the value, BAB, and duration factors; thus, low-volatility investing can be a risk-diversifying strategy when blended with standard equity portfolios. But as a result of the construction methodology employed, the portfolios are meaningfully unalike in their industry and country biases as well as their turnover and liquidity characteristics.

The most popular version of the low-volatility strategies is the minimum-variance (MV) portfolio, which takes as input a covariance matrix for the stocks in the prescribed universe. The authors employ two shrinkage methods and two factor models to estimate covariance matrices for the 1,000 largest companies on the basis of trailing five years of monthly returns. They then compute four long-only MV portfolios for each region by means of a quadratic programming solver that they constrain with 5% position limits to control single-stock concentrations.

Compared with the cap-weighted index, the US MV portfolios have about 25% less volatility and a return advantage of 134–182 basis points (bps). The global developed MV portfolios reflect a reduction of more than 30% in volatility vis-à-vis the corresponding cap-weighted index. One of them underperforms the cap-weighted index by 8 bps, while the other three outperform by up to 143 bps. The emerging markets MV portfolios have about 50% less volatility than the comparable cap-weighted index, with a return advantage of 97–409 bps. Without exception, the MV portfolios in all regions register statistically strong improvements in their Sharpe ratios. No decisive quantitative evidence favors any one of the covariance estimation methods used in their construction.

To test alternate approaches to low-volatility investing, the authors examine two distinct weighting heuristics, driven, respectively, by the constituent stocks’ CAPM betas and total volatilities. They select the 200 lowest-beta (lowest-volatility) stocks from the 1,000 largest companies in each region and weight them in the portfolios by their inverses (i.e., 1/beta or 1/volatility, as appropriate). They additionally construct equal-weighted portfolios containing the lowest-beta (lowest-volatility) stocks.

The heuristically constructed portfolios outperform the cap-weighted indices by 184–249 bps in the United States, 278–302 bps in the global developed markets, and 655–887 bps in the emerging markets. In general, the heuristic approaches also prove reasonably effective in reducing volatility; in emerging markets, however, the average volatility reduction they achieve is 35%, considerably less than the 50% reduction observed in the MV methodology. In all regions, the Sharpe ratios calculated for the heuristically constructed portfolios are close to or higher than those of the average MV portfolio, but the authors remark that no theoretical grounds exist for concluding either approach would consistently produce better risk-adjusted returns. As with the MV portfolios constructed on the basis of different covariance estimation methods, none of the various heuristic strategies is decisively superior to the others.

In order to quantify the sources of return, the authors conduct attribution analyses using both the Fama–French–Carhart four-factor model and an augmented model that additionally includes BAB and duration. They observe that replacing a beta-one equity portfolio with a low-volatility portfolio reduces risk without decreasing the overall equity allocation: All the low-volatility portfolios’ market betas are significantly below unity (about 0.7 for the US strategies and lower for the global developed and emerging markets). The value factor generally makes a positive contribution outside the emerging markets, where a moderately anti-value bias comes into view: In the emerging markets, the value factor is insignificant in the four-factor model and economically negative when BAB is included. In all regions, the duration factor reveals positive exposure to interest rate risk; investors seeking income and safety may see stocks with high dividend yields and low volatility as an attractive alternative to fixed-income securities in a low-rate environment. The relationship between low-volatility stocks and small stocks is inconsistent across regions.

Overall, the authors conclude from the attribution analyses that the factor exposures for the MV and heuristically assembled portfolios are statistically similar within each region. Investing in more than one low-volatility portfolio construction methodology would not meaningfully increase factor diversification. They further report that, with the exception of the simulated emerging markets portfolios, there is no evidence of skill-related alpha, and they caution investors against including unexplained alpha in estimates of future low-volatility performance.

The authors also examine by region the MV, inverse-volatility, and inverse-beta portfolios’ sector, country, and stock concentrations in comparison with those of the cap-weighted indices. The portfolios’ inverse Herfindahl scores indicate the MV portfolios generally have greater concentrations than the portfolios whose holdings are weighted by 1/volatility and 1/beta; the sole exception is the US MV portfolio’s modestly higher effective N for sectors.

Low-volatility portfolios’ tendency toward extreme concentrations in a particular industry or country is potentially concerning, and constraining the portfolio construction process to mitigate this tendency can increase volatility and reduce returns. The authors present charts depicting, over time, the industry allocations of naïve US cap-weighted, MV, and heuristically constructed low-volatility portfolios and the country allocations resulting from those methodologies in global developed and emerging markets portfolios.

Finally, due to an aggressive securities selection process, low-volatility portfolios inevitably have higher turnover than cap-weighted indices. Nonetheless, the differences in turnover resulting from portfolio construction methodologies are significant. Across regions, the MV portfolios have average one-way turnover between 44.9% (US) and 47.35% (global developed markets), whereas the inverse-volatility portfolios’ average turnover ranges from 18.92% (US) to 28.32% (emerging markets). The inverse-beta portfolios fall between the two other approaches to low-volatility investing. The turnover statistics along with liquidity indicators such as bid–ask spreads and average daily volumes reveal that, in comparison with cap-weighted funds, low-volatility strategies can be much more costly to trade and more difficult to implement at scale. The authors argue in closing that careful portfolio engineering is necessary to improve low-volatility portfolios’ liquidity and investment capacity as well as to ensure adequate country and sector diversification.

Summarized by Philip Lawton, CFA

Featured Tags

Learn More About the Authors

Senior Vice President, Head of Smart Beta

Li-Lan Kuo

Senior Researcher, Smart Beta
Senior Advisor

Feifei Li

Partner, Head of Investment Strategy