On this page
A tilt that adds one-fifth of the value factor’s return to the market, run at roughly 2% tracking error against the cap-weighted index, has earned 0.5 points a year since 1985. Almost all of that edge arrived before 2007. Measured against the market, the tilt’s relative return peaked in December 2006 and, as of July 2026, has not made a new high since.
What a 2% tracking error actually buys
Tracking error is the leash length on a tilt, not its payoff. A 0.20 HML weight runs at roughly 2% annualized tracking error against the cap-weighted index here, which means a household holding it should expect its return to land within about 2 points of the plain index in a typical year, and further away in an unusual one. The leash says nothing about which direction the miss runs. Since 1985 it has run positive more often than not, by 0.5 points a year on the full sample, but the path was not smooth: 39% of the rolling 10-year windows in that sample show the tilt behind the market, not ahead of it.
Nineteen years without a new high
Most of the 0.5-point annual edge was earned in a single stretch: the value rebound after the 2000 to 2002 dot-com crash, when growth stocks fell hard and cheap stocks did not. By December 2006, the tilt’s cumulative return relative to the market had reached its high point for the sample. It has not gotten there again. The rolling 10-year edge that stayed near 2 points a year through the mid-2000s turned negative by the mid-2010s and bottomed around minus 1.4 points a year in the window ending September 2020, before recovering toward flat by 2026. A 2021 paper by AQR researchers Israel, Laursen and Richardson took the question directly, “Is (Systematic) Value Investing Dead?”, and concluded no, arguing the drought reflected a valuation gap between cheap and expensive stocks that widened rather than a factor that stopped working. The data here does not resolve which reading is correct; it only shows how long “not yet” can run.
Why a long-only tilt keeps only part of the premium
Nothing above is the value factor’s full return. HML itself is a long-short number: long the cheap half of the market, short the expensive half. A household’s actual tilt cannot short anything; it can only underweight the expensive names it already holds down toward zero. Clarke, de Silva and Thorley’s 2002 transfer coefficient work put a typical long-only manager’s correlation to their own unconstrained signal at roughly 0.6 to 0.8, meaning a long-only implementation should be expected, before a single year of results comes in, to capture only part of what the factor itself earned. The 0.5-point edge in this card is already the long-only, constrained number, not the textbook one.
The premium is real over a career-length sample and was invisible for the back half of it. Someone who cannot sit through a decade of underperformance without pulling the position should not expect to collect the 0.5 points; someone who can hold it through a stretch like 2007 to 2020 is the one being paid for that patience, not for having found a better stock-picking model.