One year ago, I wrote about the U.S. Value factor and what I found by extending its history back in time before 1926.
In summary, I wrote that Value’s drawdown in March 2020 was normal and likely close to its bottom. Without the insights from the extended history, Value had appeared ‘dead’ given it had crossed the previous all-time maximum drawdown.
As far as factor timing goes, I was close -- about three months too early -- as it reached its bottom around August 2020. Since last May and through the end of March 2021, however, Value is up about 21%. Once again, thanks, long-run history!
In this blog, I explore the U.K. Value factor, by showing its previously unseen history extended back to 1800, and illustrating the similarity between its recent drawdown and bounce back as compared to the U.S. value factor.
To extend the U.K. Value factor back, I used Dividend Yield as a proxy. While not a perfect substitute, it’s close enough. Since 1975, the U.K. Dividend Yield correlated 58% with Book-to-Market, making it a not-so-awful-proxy for Value. In fact, over the same timeframe, the cumulative performance for the U.K. Dividend Yield was better than Book-to-Market.
The chart below, based on data from Professor’s French website, compares the two factors from 1975 onwards.
From the amazing history available at the Global Financial Data (GFD), which includes a stock-level database of prices, total returns, and dividend yields for ~3,610 U.K. common stocks trading between 1800 and 1975, I can extend the U.K. Dividend Yield back to 1800.
GFD’s data actually goes even further back, but I wanted to stay consistent with my other extension projects which all go back to 1800.
When comparing the recent performance histories, the long-run pattern of the U.K Value factor is similar to the U.S Value factor. Both in the U.K. and in the U.S.
value factor ‘works’ in the previously untested history,
periodically ‘crashes’
has experienced a noticeable cumulative return slowdown over the past two to three decades
Despite all the existing evidence, it is fascinating to see fresh evidence that a simple version of the Value factor has consistently helped to identify outperforming companies for 200+ years.
This outperformance, however, comes with occasional and violent crash risk. The long-run history reveals an early period when Value crashed more than -50% (1840’s) which makes the early 1980’s drawdown appear more normal. We also see a handful of other -20% and -30% losses throughout the deep history, which, again, makes more recent history crashes appear in line with the trends rather than anomalous.
However, as with the U.S. analysis, we see an alarming trend of recently flattening cumulative performance and increased drawdown frequency. There are many possible reasons for the structural slowdown, but I’m speculating: more factor crowding, the shift from tangible to intangible value, missing share-buybacks, or cleaner recent data. Whatever the reason, the U.K. Dividend Yield does not appear to be a reliable long-term structural source of ‘alpha’ going forward. However, as a risk factor, it’s likely to continue its mean reversion, similar to how an underperforming sector or industry eventually transforms and experiences a period of mean reversion.
So how does the above analysis connect to what we do at Two Centuries?
We view long-run history as an unappreciated source of insights. We try to test all our strategies against the longest timeframe possible . Sometimes that means going back a century or two. Other times we can only add an extra decade. For example, with alternative datasets, doubling the history is a significant feat. Long-run data is helpful in showing us:
If something worked as we expected in the untested data that reduces the odds of it being a product of datamining on the overly tested recent data.
A richer distribution of outcomes including painful crashes that might not show up as often in shorter history, because of their rareness, meaning investors disregard them as one off events, rather than seeing them as a repeatable part of the distribution. It takes much longer history to start seeing crashes as more normal outcomes.
If a factor has a recent (20-30 year) slowdown like Value, a longer history shows whether it is ‘normal’ or ‘alarming’ given the additional context. In the case of Value, we think the recent structural slow down in the average return to value (not the drawdown, for which we do expect some mean reversion) is actually alarming. As a result, we prefer to focus our R&D on more innovative sources of a company’s value, such as intangible assets rather than current divided yields.