Factor Timing: Myths and Reality

If you accept that factor premiums are real, an obvious idea follows: rather than holding a constant exposure, increase it when the factor looks cheap and reduce it when the factor looks expensive. This is factor timing, and it is the most natural inference to draw from a dashboard like this one. It is also, on the weight of the evidence, considerably harder than it looks. This article explains why, because the single most likely misuse of the data on this site is to act on a low reading.

The case for timing

The argument is straightforward. If the value premium exists because value stocks are cheap relative to growth stocks, then the size of that valuation gap should tell you something about the expected return from closing it. When the spread between what value and growth stocks cost is unusually wide, the subsequent payoff to owning value should be unusually large. Researchers have found relationships of this kind in the data, and the logic mirrors how equity valuations are used to forecast long-horizon market returns.

Why it disappoints in practice

Several problems compound, and each one alone is serious.

• There are very few independent observations. Factor cycles run for years. A century of monthly data contains a great many months but only a handful of genuine cycles, so the effective sample for testing a timing rule is far smaller than the row count suggests. Confidence intervals around any timing signal are correspondingly wide.
• Cheap can get much cheaper, for years. A spread in the bottom decile can stay there or go further for long enough to exhaust the patience and capital of anyone acting on it. Being early is operationally identical to being wrong.
• Timing concentrates the risk you were diversifying. The main defensible reason to hold multiple factors is that their disappointments do not coincide. A timing overlay deliberately concentrates into whichever factor has most recently done worst, which undoes that benefit at the moment it is most valuable.
• The signal is measured with error. Value spreads depend on accounting definitions that change meaning over time, most acutely the treatment of intangible assets. Part of what looks like a historically wide spread may reflect a drifting measuring instrument rather than a genuine opportunity.
• Costs scale with turnover. A static exposure trades only to rebalance. A timed one trades on every signal change, and the timing gain has to clear that additional cost before it contributes anything.

What the research broadly concludes

The literature has not converged, but a rough consensus exists among practitioners who have studied it seriously. Valuation-based timing signals contain some information, the information is weak relative to its variability, the gains are easily consumed by transaction costs, and the strategies are difficult to hold through the periods when they are losing. Several prominent researchers who have argued in favour of factor timing have also cautioned that the effect is smaller than enthusiasts assume, and the disagreement is largely about whether a modest, slow-moving tilt is worth the added complexity.

A more defensible use of the same information is as context rather than as a signal. Knowing that a factor is at an historical extreme is genuinely useful for setting expectations, for understanding why a portfolio has behaved as it has, and for deciding whether recent performance is anomalous or ordinary. That is different from trading on it.

If you are going to do it anyway

The practices that separate the more careful attempts from the reckless ones are consistent: keep tilts small relative to the strategic allocation, move slowly enough that turnover stays modest, define the rule in advance rather than reacting to whichever number currently looks extreme, use several definitions of the signal rather than one, and decide beforehand how long you are willing to be wrong. That last commitment is the one most often skipped and the one that most often determines the outcome.

Nothing in this article is investment advice. It describes what the historical evidence supports and, more importantly, what it does not.

Asness, C. (2016). "The Siren Song of Factor Timing." Journal of Portfolio Management, 42(5), 1-6.