The drawdown, measured

The long drawdown is usually described. This lesson measures it, from the monthly factor series that the studies themselves are built on, so that every figure below can be recomputed by anyone who downloads the same file.

What was measured, and how

The series is the monthly return on a portfolio that holds high book-to-market companies and sells low book-to-market companies. Compounding those monthly returns from July 1926 gives a cumulative index. The peak of that index is December 2006. The trough is September 2020. The fall between them is 57.8 percent, and the file's last month, July 2026, still sits below the 2006 level.

That is thirteen years and nine months from peak to trough, and the recovery is incomplete after a further five and a half years.

Three ways to say how bad it was

How unusual is that, in the full record

Take every ten-year window in the series, month by month, from July 1926 onwards. There are 1,082 of them. The factor return was negative in 155, which is 14 percent of all of them.

So a decade of no premium is not a freak event in this data: roughly one window in seven. What made the period after 2006 different was its depth and its length, not the mere fact of a losing decade.

What this does and does not establish

It does not establish that the premium is gone, and nothing in an arithmetic exercise could. It establishes the size of the hole a person had to sit in while finding out, which is a fact about the discipline rather than about the statistics.

A method whose evidence is an average across a century commits its user to periods like this one. The honest version of the claim is not that value investing works. It is that in this record it has worked on average across a century, and that the average contains a fourteen-year stretch nobody could have distinguished from failure while living through it.

The recovery is worth reading carefully too, because it is the part most summaries get wrong in both directions. From the September 2020 trough the factor did rise: the index stands at 70.4 in July 2026 against 42.2 at the bottom. That is a substantial run, and it still leaves the series 29.6 percent below where it stood in December 2006, almost twenty years earlier. Both sentences are true, and a reader who has only heard one of them has been told half of it.

There is a second reason to measure rather than describe. Any statement about this episode depends on the date the speaker stopped counting. Measured to December 2019 the drawdown is bad; measured to September 2020 it is far worse; measured to today it is partly recovered. The same series supports three different stories, and the only defence against being told whichever one suits the teller is knowing which month the number ends on.

Check yourself