Historical Performance

The Fama-French three-factor series begin in July 1926, which makes them one of the longest continuous return records in finance. That length is the dataset's great advantage and the source of its most common misuse. A century of monthly observations sounds like overwhelming evidence, but factor cycles run for years, so the number of genuinely independent episodes is small. This article describes what the record actually supports.

The shape of the record

The dataset opens three years before the 1929 crash and therefore covers the Great Depression, the Second World War, the post-war expansion, the inflation of the 1970s, the 1987 crash, the dot-com boom and bust, the 2008 financial crisis, and the 2020 pandemic shock. Few economic environments of the modern era are missing from it.

Across that span the market factor has been the dominant contributor to returns, and the additional factors have been smaller, noisier, and far less consistent than summary statistics suggest. The averages are real. What the averages conceal is how the returns arrived, which is in long uneven stretches rather than steadily.

The lesson that matters most: premiums disappear for years

The single most useful thing in the historical record is the length of the bad periods. Each of the classic factors has gone through multi-year and in some cases multi-decade stretches of no reward or outright loss. The value factor's underperformance beginning around the global financial crisis is the most recent and most discussed example, but it is not unique in kind.

This has a direct practical implication. An investor who allocates to a factor on the strength of its long-run average and then abandons it after three or five poor years is behaving in a way the historical record predicts will be costly. The holding period implied by this data is longer than most investors have in mind when they buy, and the mismatch, rather than the factor itself, is what most often produces a bad outcome.

Regime dependence and crisis behaviour

Factor behaviour varies with the environment. Relationships that appear stable in calm markets can change during stress, and correlations between factors tend to move at exactly the moments when diversification is most needed. This is why the dashboard's regime analysis and drawdown views are more informative than a single long-run average: they show conditional behaviour rather than an unconditional summary.

It is also why applying an average estimated across a century to the next five years is a weak forecast. The economy that generated the 1930s data and the one that generated the 2020s data differ in industry composition, in the proportion of corporate value held as intangible assets, in trading costs, and in who participates in markets at all.

What the early data can and cannot support

Data quality is not uniform across the sample. The earliest decades cover a market with fewer listed companies, far less liquidity, sparser and less consistent accounting disclosure, and trading costs that would be considered prohibitive today. Modern federal securities disclosure requirements did not exist before the mid-1930s.

This does not make the early record useless, and it is the only evidence available on how factors behaved through a genuine depression. It does mean that a premium measured largely in the 1930s and 1940s should not be assumed to be attainable now, and it is a reason to check whether a historical result survives when the sample is restricted to the more recent, better-documented period.

The honest summary

A century of data establishes that these patterns existed and were large enough to matter. It does not establish that they will persist, it does not support precise estimates of future premiums, and it does not support timing decisions based on current readings. What it supports well is calibration: a realistic sense of how volatile these series are, how long they can disappoint, and how far from average a current reading really is. That is genuinely valuable, and it is less than what the data is usually asked to deliver.

All series shown here come from the Kenneth R. French Data Library. Past performance does not indicate future results, and nothing on this page is investment advice.