MomentumSecurities that have outperformed over the last 1–12 months continue to outperform over the next 1–3 months because investors under-react to information and herd into trends.
Jegadeesh & Titman (1993); Moskowitz, Ooi & Pedersen (2012)
Short-term reversalOver days to a couple of weeks, sharp moves overshoot and partially reverse as liquidity providers are compensated for absorbing order imbalances.
Lehmann (1990); Jegadeesh (1990); Nagel (2012)
Volatility compression → expansionVolatility clusters: after a period of unusually narrow ranges, a break of the range on volume starts a move that continues as the new information is absorbed.
Mandelbrot (1963); Engle (1982) ARCH; Donchian channel practice
Post-news driftPrices under-react to salient news (earnings, guidance, ratings, contracts); the initial move is followed by drift in the same direction over the following days as attention spreads.
Bernard & Thomas (1989); Tetlock (2007); Hirshleifer, Lim & Teoh (2009)
Sector rotationRelative strength between sectors persists for weeks because capital rotates slowly with the business cycle and with macro surprises; buying the leading sectors captures the rotation.
Moskowitz & Grinblatt (1999); Conover, Jensen, Johnson & Mercer (2008)
Statistical arbitrage (pairs)Two securities exposed to the same economic drivers share a long-run relative price; temporary divergence of the spread mean-reverts, so buy the lagging leg (and short the leading leg where allowed).
Gatev, Goetzmann & Rouwenhorst (2006); Avellaneda & Lee (2010)
Variance risk premiumImplied volatility is on average higher than the volatility that is later realised, because investors pay for insurance. Selling premium when it is rich and buying it when it is cheap harvests the premium.
Carr & Wu (2009); Bakshi & Kapadia (2003)
Intraday VWAP reversionWithin a session, prices stretched away from the volume-weighted average revert toward it as institutional execution algorithms benchmark to VWAP — except on trend days, which must be avoided.
Market microstructure; Madhavan (2002) on VWAP execution
Learned conditional edgeNo single hand-written rule captures when the market is predictable. An online learner over the full feature vector (trend, reversal, volatility, volume, relative strength, news, regime) finds conditional combinations with a persistent edge and keeps re-fitting as the market changes.
López de Prado (2018) meta-labelling; online convex optimisation (Zinkevich, 2003)
Efficient market, trend- and volatility-managed (null hypothesis & passive core)Prices already reflect available information, so after costs nothing beats holding the market — except holding less of it when volatility is high and stepping aside when price is below its long moving average: volatility scaling and a 10-month trend filter have historically cut drawdowns without giving up return. This is the default allocation every other theory must beat.
Fama (1970); Faber (2007) 'A Quantitative Approach to Tactical Asset Allocation'; Moreira & Muir (2017)