Since the financial crisis in 2008, risk parity (RP) strategies have been gaining massive popularity in practice due to their significant outperformance of traditional weighting techniques such as equal- and value-weighting, among others. In addition, RP strategies result in very attractive portfolio characteristics, making them particularly interesting for academia. RP strategies distribute portfolio risk equally across assets, which results in portfolios with lower volatility and higher risk-adjusted returns. Moreover, RP strategies can be adjusted to achieve certain target portfolio risk levels, match investors’ risk preferences, and even be levered to achieve specific target portfolio returns. As a result, not only several large pension plans adopted them, but also numerous risk parity funds and indices were launched. As of May 2018, more than $175 billion of assets are managed using risk parity strategies according to Bloomberg. However, risk parity faces numerous challenges, most prominently is the lack of an underlying theory that helps explain and/or predict its performance.
In his latest study, Dr. Nabil Alkafri aims to shed more light on the risk-return characteristics of the risk parity portfolio by investigating the following questions:
- Which market anomalies and factors help explain the performance of risk parity?
- Can the appealing characteristics of risk parity be attained whilst improving its general risk-return profile as measured by the Sharpe ratio?
The main analysis is restricted to three U.S. asset classes. To simplify the analysis and ensure tractability as well as replicability of results, the analysis is performed with equities, bonds, and commodities. Equities are represented by the S&P 500 Index, bonds are denoted by the Thomson Reuters U.S. 10-year Government Benchmark Index, and commodities are described by the Bloomberg Commodity Index. The 1-year U.S. Dollar deposit rate is utilized as a proxy for the risk-free asset. The data consists of daily total-return prices from 01.01.1994 to 30.09.2019 and an out-of-sample rolling-window regression approach is employed.
Input parameter required to determine the RP portfolio weights, namely the asset variance-covariance matrix, are decomposed and so-called RP component portfolios are developed. These component portfolios utilize either the asset volatility vector (volatility component portfolio or VOLA) or the correlation matrix (correlation component portfolio or CORR) to achieve the risk parity properties under specific assumptions. The approach is similar to Asness et al. (2019), who decompose the Betting-Against-Beta (BAB) factor of Frazzini and Pedersen (2014) into the Betting-Against-Volatility (BAV) and Betting-Against-Correlation (BAC) factors.
The results indicate that the returns of RP are driven by the BAB factor and is closely related to its components, VOLA and CORR, which are driven by the low-volatility anomaly and the BAC factor. In cross-sectional regressions, VOLA and CORR explain almost all of the variation in returns of the RP portfolio. Furthermore, in time-series rolling regressions of RP on the component portfolios, the rolling regression coefficients of VOLA and CORR are almost perfectly negatively correlated and sum up to one (see figure). This implies that RP can be replicated by a linear combination of VOLA and CORR and that these component portfolios can be utilized to improve the general risk-return profile RP while maintaining its properties to some extent.