1. If you haven’t done so yet, we encourage you to read these excellent pieces in the AQR Cliff’s Perspective series: “Liquid Alt Ragnarök?” (September 7, 2018) and “But What About October?” (November 23, 2018). Cliff Asness’s discussion on investor psychology during painful times, and on the evidence and intuition of liquid alternative investments, resonates with us. We choose not to spend time on these aspects of ARP investing in this article. We recognize these aspects as being important (they are crucial), but our goal is try to gain a better understanding of recent performance across the ARP universe by studying what we view as the core strategies that compose it. Our aim is to contribute to the overarching ARP discussion.
2. The HFR hedge fund classification system also includes fund of funds, risk parity, and blockchain. We exclude these classifications from our study because they are niche categories and, in our view, inadequately represent the ARP space.
3. From 1992 to 1999, Capital Decimation Partners, LP (i.e., an investment strategy that shorts out-of-the-money S&P 500 put options on monthly expiration dates for maturities less than or equal to three months with strikes approximately 7% out of the money) boasted an annual Sharpe ratio that approached 2.0 and an average annual return exceeding 40% (Lo, 2001).
4. As discussed by Mitchell and Pulvino (2001), at the announcement of a merger between two companies, the share prices of the companies’ securities begin to trade at a premium in anticipation of successful completion of the transaction. While many idiosyncratic reasons can lead to a failed merger, share price movements are a common factor. Because a merger is determined at share prices that are current at the time of the bid, large deviations in share prices (or heightened market-price volatility) generally cause one of the companies to pull out of the transaction. The merger spread is well modeled by the simultaneous writing of a call and a put option struck at the merger price. Naturally, these arbitrage strategies tend to thrive when market volatility is stable or falling.
5. The seminal article “Time Series Momentum” by Moskowitz, Ooi, and Pedersen (2012) provides an explanation of time series momentum.
6. Cross-sectional strategies employ the same dollar, beta, or risk long as they do short. They are named cross-sectional because they generally seek to profit from the relative movement between investments of a similar asset class within a given time period, such as a month. An example of this would be to have an equal, long exposure to three commodities and offsetting equal short exposure to three different commodities. If each position had a magnitude of 50%, the dollar net position would be zero (150% − 150% = 0%), but the gross position would be the sum of absolute value of the positions, 600% (6 x 50%).
7. We recognize that including one of the Research Affiliates strategies could be regarded as self-interested. That isn’t our intent. The leading impetus behind this article is to share our findings on our exploration of the underlying drivers of return within the ARP arena. We share strategy-specific results, as it relates to this framework, for those readers interested in understanding our intuition and our thinking behind our product design. Our hope is that these ancillary points do not distract, but rather inform.
8. SARP currently excludes exposure to the equity neutral category and the accompanying micro (security-level) equity factors it exploits. By harvesting only macro (index-level) risk premia, SARP can employ less leverage in seeking to achieve its desired return profile, and implementation costs can be kept significantly lower. Further, the exclusion of equity neutral and micro equity factors leads to improved transparency in portfolio positioning and return attribution as well as reduced due diligence and ongoing monitoring costs.
9. Our 60/40 proxy is the Global 60/40 Index, which is composed of 60% MSCI World Net Index and 40% Barclays Global Aggregate Index.
10. For more information, we refer you to Arnott et al. (2016) and Arnott, Beck, and Kalesnik (2016a,b).
11. We are using the Fundamentally Reweighted definition of value, but similar results hold for most definitions of value, across not only the US market, but also developed and emerging markets.
12. The forward-looking information ratio of the US RAFI Multi-Factor™ strategy, equivalent to a Sharpe ratio in a long–short setting, is 0.71 as of September 30, 2018.
13. Cboe S&P 500 PutWrite Index, Cboe S&P 500 95-110 Collar Index, Cboe S&P 500 2% OTM BuyWrite, Cboe S&P 500 30-Delta BuyWrite Index, Cboe S&P 500 Iron Butterfly Index, Cboe S&P 500 Zero-Cost Put Spread Collar, Cboe S&P 500 Covered Combo Index, Cboe S&P 500 Iron Condor Index, and Cboe S&P 500 5% Put Protection Index.
14. Aked (2016) provides a discussion of the significant implicit transaction costs in equity investing. The issues multiply significantly in less-liquid securities or strategies employed by ARP managers.
15. The data used in this analysis are from February 1989 through June 2018 with instruments added as the data became available. In aggregate, our analysis incorporates 53 different instruments and just over 17,000 data points.
16. We use a LOESS smoothing function with a smoothing parameter of 0.75. The curve is robust to differential smoothing parameters.
17. These strategies are represented by an aggregation of the respective carry, value, and momentum series from the HFR SRP family of indices. For dates before the inception of the HFR SRP series, we calculate our own series of macro strategies. We calculate the performance difference between a long portfolio, consisting of the one-third of the market with the best signal (e.g., value, carry, or momentum), and a short portfolio, consisting of the one-third of the market with the worst signal. The underlying markets include bonds, currencies, equities, and commodities. No adjustment is made for transaction costs, missed trades, cost of leverage, cost of borrowing stock for the short portfolio, or fees.
18. Arnott, Harvey, and Markowitz (2018) offer a seven-point protocol for assessing how statistical tools are applied in backtests.
19. The definitions of the series can be found at https://www.hedgefundresearch.com/family-indices/hfr-bank-systematic-risk-premia and https://lab.credit-suisse.com.
20. To represent the value, carry, and momentum strategies, we calculate the performance difference of a long portfolio, consisting of the one-third of the market with the best signal (e.g., value, carry, or momentum), relative to a short portfolio consisting of the one-third of the market with the worst signal. The underlying markets include bonds, currencies, equities, and commodities. No adjustment is made for transaction costs, missed trades, cost of leverage, cost of borrowing stock for the short portfolio, or fees.
21. A perfectly symmetric return distribution has a skew of zero. Negatively skewed distributions have a longer left tail, which suggests that extreme negative outliers are more likely (and vice versa).
22. Kappa has been calculated as the ratio of the difference between the monthly mean and the monthly median wealth and one-half of the monthly variance. The wealth is the month-end value of an investment divided by the month-beginning value. For mathematicians, it is expressed as where the hat accent follows the normal substitution-of-population-for-sample procedure. We use the difference in the mean and the median rather than a direct measure of the skew, scaled by the expected skew of a log-normal distribution, because of the impact heavy tails have on the skew measure. Given that the majority of asset market returns are better explained by a degree of freedom five t-distribution than a normal distribution, the confidence interval of a kappa using a skew measure is far wider than presented in the table “Skew and Kappa of Distributions.”
23. We calculate the confidence interval from the empirical distribution of a 1,000-sample bootstrap of the kappa statistic. We have 143 monthly data points and round the confidence intervals to the nearest integer. To gain confidence intervals closer to unit one would likely require centuries of data rather than decades.