Content area
Full Text
There exists a substantial literature supporting the idea that returns are persistent and thus that past returns can help forecast future returns. The seminal papers of Jegadeesh and Titman (1993, 2001) showed that constructing a long–short equity portfolio based on past returns results in excess returns that cannot be explained by the standard capital asset pricing model. A myriad of papers has followed confirming this result and extending the analysis to other asset classes. Among others, Erb and Harvey (2006) and Miffre and Rallis (2007) detected momentum in commodities. Asness, Moskowitz, and Pedersen (2013) extended the analysis to equity indexes, currencies, sovereign bonds, and commodities and found a consistent momentum premium across all these asset classes.
Interestingly, the financial industry recognized far earlier the forecasting power of past returns and the profitability of buying the winners and shorting the losers. Instead of analyzing the relative performance of instruments in the cross section, however, the industry typically adopted time-series momentum, going long markets with recent positive returns and shorting those with recent negative returns. Hurst, Ooi, and Pedersen (2013) argued that these absolute trend-following strategies explain the returns of managed futures/CTA1 funds, a $360 billion industry in 2018 according to BarclaysHedge.
Practitioners have also shaped the academic discussion around time-series momentum. Moskowitz, Ooi, and Pedersen (2012), all affiliated with the global investment management firm AQR, introduced a “time-series momentum factor” and documented the profitability of such strategies. In 2014, AQR published the results of a backtest stressing the persistence of trend following over more than 100 years (Hurst, Ooi, and Pedersen 2017). In a recent paper, Babu et al. (2020) from AQR studied trend following on alternative assets not covered before and on factors and concluded that “momentum is everywhere”.
If we accept the existence of both absolute and relative momentum, we immediately face the question of how to weight these approaches in a trend-following strategy. Moskowitz, Ooi, and Pedersen (2012) decomposed the drivers of time-series and cross-sectional momentum. They showed that the main driver behind both is the positive autocovariance and that time-series momentum is not fully captured by cross-sectional momentum. In a recent paper, Bird, Gao, and Yeung (2017) compared the performance of alternative implementation of absolute and relative trend following in an...