Is there a better way to identify attractive and unattractive assets than simply ranking them? In the August 2020 version of their paper entitled “Decoding Systematic Relative Investing: A Pairs Approach”, Christian Goulding, Campbell Harvey and Alex Pickard examine a long-short strategy that periodically reforms a portfolio by evaluating all possible pairs within an asset universe based on:
- High positive signal-future return correlation for each asset on its own in a pair.
- Low (or negative) signal correlation between assets in the pair.
- Low (or negative) signal-future return correlations between one asset and the other in the pair.
They use these three inputs to calculate a (somewhat complex) composite score for each pair. Among pairs with the highest composite scores, the member with the higher (lower) signal goes to the long (short) side of the portfolio. They assess usefulness of the three conditions and the composite score using a momentum signal calculated as average past monthly return over a specified lookback interval minus its inception-to-date mean and divided by its inception-to-date standard deviation. They split their sample roughly in half and use the first half for detection of profitable pair strategies and the second half to measure out-of-sample performance. They further test an explicit tactical allocation strategy using a 12-month momentum lookback interval, a rolling 10-year monthly composite score and a scheme that weights the top four asset pairs according to respective composite scores. As a benchmark, they use a comparable conventional relative momentum strategy that simply ranks assets on momentum signal. Using monthly returns for 13 broad asset-class indexes encompassing equities, bonds, real estate investment trusts (REIT) and commodities (78 possible pairs) as available through May 2020, they find that:
- Across different asset pair strategies, each input and composite score have highly significant out-of-sample return predictive power. The composite score is the strongest predictor.
- The composite score is a stronger predictor of future pair strategy performance than is past pair strategy performance.
- The specific test strategy substantially outperforms the conventional relative momentum benchmark strategy over the last 20 years.
- Average annualized monthly gross return of the pairs (benchmark) strategy is 5.3% (2.0%), with annualized gross Sharpe ratio 0.66 (0.28).
- Maximum drawdown of the pairs (benchmark) strategy is -22% (-27%).
- Gross annual alpha and beta relative to the the U.S. stock market for the pairs (benchmark) strategy are -0.10 and 5.8% (-0.16 and 2.8%).
- The pairs approach has lower turnover than the benchmark.
- Pairs outperformance is consistent across subsamples and robust to momentum lookback interval, composite score calculation rolling window length and number of top pairs included each month.
In summary, evidence indicates that more precise isolation of asset sensitivity to signals via pairs analysis improves hedge strategy performance.
Cautions regarding findings include:
- Findings are gross, not net. Accounting for costs of transforming indexes into funds (which likely varies by asset class), periodic portfolio reformation (which may vary by strategy and asset class) and shorting would reduce reported returns.
- The authors do not address whether they use a time lag for strategy calculations in computing returns.
- The proposed strategy is complex enough to be concerned about model snooping in its construction, such that results overstate expected live performance. Running many models on the same data would subvert the out-of-sample methodology.
- Especially for relatively large asset universes, the approach is beyond the reach of many investors, who would bear fees for delegating to an advisor or fund.
