2. Value and momentum: the two that have survived best

Level 9: Factors

Of the list of characteristics someone has at some point proposed as a factor, the vast majority do not withstand serious examination. Two do, with caveats and with very high bills attached: value and momentum. They have the longest track records, they have been replicated most often outside the sample in which they were discovered, and they have handed out the most punishment to those who bought them without knowing what they were buying.

Value: paying little for what is already there

Value compares the price of a share with some measure of what the company already has or already earns: its book value, its profits, its cash flow, its sales. A share is cheap when you pay little for each euro of that measure, and expensive when you pay a lot. The factor is built by comparing the group of the cheapest with the group of the dearest.

Three clarifications that save a lot of grief:

• Cheap does not mean good. Value does not select good companies. It selects companies whose price is low relative to something that already exists, and many of them are cheap for perfectly sensible reasons.
• It is a relative measure, not an absolute one. Nothing guarantees that today's cheap shares are cheap in historical terms. Only that they are cheap compared with the others.
• The idea predates the data. Graham and Dodd were writing about buying below the value of what already exists back in the 1930s. What the 1970s and 1990s contributed was measuring it across thousands of shares instead of defending it with examples.

Momentum: buying what has already risen

Momentum ranks shares by what they have done over the past three to twelve months and buys those that have done best, selling or avoiding the worst. It usually skips the most recent month, because over very short horizons prices tend to bounce back and that bounce contaminates the signal.

It is an unappealing factor for two reasons. First, because it flatly contradicts the popular advice to buy cheap: here you buy what has already become more expensive. Second, because it has no comforting story behind it; nobody buys momentum out of conviction about the business.

And even so, there it remains. The work that put it on the table, by Narasimhan Jegadeesh and Sheridan Titman in the early 1990s, measured the US market between the mid-1960s and the late 1980s and found that buying the recent winners and selling the losers produced on the order of one percentage point a month. A decade later, the same authors checked that the effect was still there in the 1990s, that is, outside the original sample. It was subsequently documented in other countries, in indices, in currencies, in commodities and in bonds, and studies with very long price series trace it back to the nineteenth century.

Why these two and not others

Four reasons, and none of them is that their past returns are the highest:

• A long track record. Decades of data in the case of value, and more than a century in some reconstructions of momentum.
• Out-of-sample replication. They show up in periods and markets that were not used to discover them, which is the test almost no fashionable factor passes.
• Robustness to definition. Value does not depend on which ratio you choose; momentum does not depend on whether you look at six months or twelve. An effect that only exists with one very specific definition is usually an accident of the sample.
• They get on badly with each other. Their bad runs tend not to coincide. Value buys what has been falling and momentum sells exactly that, so combining them smooths the journey without giving up entirely on either. It is the argument Cliff Asness and his team have pressed most insistently.

The bill: long, demoralizing runs

It is worth doing the arithmetic out loud. Suppose a real factor that pays four points a year on average and whose annual result swings with a typical amplitude of twelve points. That is a third of a point of signal for every point of noise. With those numbers, the probability that any given decade ends negative is around one in six, and to distinguish that factor from pure nothing with statistical confidence you would need on the order of thirty-five or forty years of data. And all of this assuming the factor is real and does not change along the way.

Momentum collects its bill in a different way: not slowly, but all at once. Its worst episodes are concentrated in the violent turns after a fall, when the portfolio is short precisely in whatever bounces most. In 1932 and again in the spring of 2009 US momentum portfolios lost, according to the usual academic series, on the order of three quarters of their value within a matter of months.

Real case: value's thirteen-year crossing

Between 2007 and 2020, expensive shares beat cheap ones in the United States almost without interruption, with a cumulative gap counted in hundreds of percentage points. It was the worst run on record for the value factor in its entire measured history, worse than the one at the end of the 1990s, and it lasted thirteen years: longer than many managers' entire working lives. AQR published a paper in 2020 asking openly whether value was dead, which is not something you write lightly when you have spent decades defending it.

What happened next is the other half of the lesson. From the final quarter of 2020 value turned, and in 2022 the US index of cheap shares beat the expensive one by on the order of twenty percentage points in a single year. Many of the investors who had spent thirteen years paying the toll were no longer inside to collect the turn: the redemptions were concentrated just beforehand. That is the part no track record summarizes, and it is the part that decides the real result.