Aktueller Artikel auf Research Affiliates (Oktober 2018)
https://www.researchaffiliates.com/en_us/publications/articles/686-ignored-risks-of-factor-investing.htmlGrob gesagt spricht der Artikel genau die Punkte an, die in den backtest-geschönten Factsheets gerne verschleiert werden, meist mithilfe selektierter Zeiträume für historische Backtest-Renditen und Risikokennzahlen.
Natürlich wird das im Disclaimer zwar erwähnt, aber unter
"past performance is not indicative of future returns, the excess return is not guaranteed" kann sich der Durchschnittsanleger dann doch nicht so recht etwas vorstellen und konkrete Underperformance-Zeiträume zu Einzel- und Multifaktoren sind nicht immer leicht zu finden.
Hier ein paar (aus meiner Sicht) interessante Aussagen im Artikel:
Quote
Finally, and perhaps most dangerous of all, factor investing presents the risk of data mining. Only factors that show good backtest results are published, let alone used. The resulting upward bias in return estimates is known as selection bias. A factor may look good because it is good or because the historical record is randomly good. Because disentangling the two can be a difficult proposition, the historical results should be expected to exceed future efficacy. In some cases, historical data can even create an illusion of value added by a strategy which in reality has no structural efficacy (Harvey, Liu, and Zhu, 2016). Similarly, the backtest results that attract investors and their capital were mostly earned during a span when little money was being committed to these factor-tilt strategies. An influx of capital can easily arbitrage away the efficacy of a factor, and in some cases this may already have happened for certain factors (McLean and Pontiff, 2016).
Quote
Each of these factors has an Achilles’ heel, giving rise to serious concerns about its future efficacy. For value, it’s the two 11-year bear markets—the most recent from August 2007 to today—that the factor has experienced, which begs the question: Is the value insight today permanently impaired or still indicative of truly cheap stocks? We believe the latter, but cannot prove it. Momentum’s long-term track record would be truly brilliant in the absence of trading costs, but turnover is huge, and trading costs can devour the alpha. Standard momentum, defined as trailing year excluding the latest month, has lost more since 1999 in its crashes than it has earned in its bull markets. Nevertheless, in defense of both value and momentum, these two effects have existed across multiple asset classes (Asness, Moskowitz, and Pedersen, 2013) and, in the case of momentum, for over two centuries (Geczy and Samonov, 2016). Are these two effects now gone, forever, in all their forms?
The size factor, despite being one of the first factors identified, offers very weak empirical evidence of a long-run premium. The performance of the investment factor, which is considered part of quality, is very sensitive to its definition (Cooper, Gulen, and Ion, 2017), raising potential concerns around data mining. Low beta faces the existential question: Should a low-risk portfolio perform on par with, or even better than, a high-risk portfolio? The leverage-constraint explanation for the low-beta effect only works if these constraints remain important. Moreover, Arnott, Beck, and Kalesnik (2016a,b) suggest a portion of the alpha earned by the low-beta factor so far this century is attributable to its rise in “relative valuation”: low-beta stocks used to trade at a deep discount, but now trade at a substantial premium.
Und hier die wohl wichtigste Aussage speziell zum Multi Factor Investing:
Quote
Individual factors are likely to experience lengthy and severe drawdowns, and diversification across factors cannot be expected to eliminate all the risks of factor investing—even though frequently cited low historical correlations, especially derived from backtests, can be very impressive.
The reality is that correlations between factors are not constant over time and multiple factors may be exposed to the same underlying risk drivers.
Thus, investors with exposure to multiple factors may still experience severe drawdowns and decade-long periods of underperformance.
Anmerkung: Die Erstellung der verwendeten simulierten Daten werden im Artikel ausführlich erklärt.
