How big was the value factor really?

Before I dive into today’s post, I have to make clear that I am not an anti-value investor. I started my career as a value investor, and I still think that the value factor is real and has historically been large. In fact, I think value and momentum are the two most reliable factors investors should try to exploit to gain superior returns. However, the world is changing and over the years, I have come to believe that the value factor has declined in size over time and probably is much smaller today than it was 30 or 50 years ago, something that I have looked at here.

And now, Mathias Hasler from Boston College has presented some findings that make me wonder if the value factor was historically as large as we thought to begin with. He went back to the original publication by Fama and French that systematically described and popularised the value factor as the difference in performance between the 30% stocks with the highest book-to-market ratio vs the 30% stocks with the lowest book-to-market ratio. 

Hasler noted that in order to compute their results Fama and French had to make several seemingly innocuous choices on how to calculate the book-to-market ratio and how to sort the companies into deciles. He then changed these choices slightly to create 96 alternative ways to calculate the original value factor and see how these alternatives stacked up against the results from Fama and French:

In terms of reported book value, Fama and French used the book value as reported at the end of a company’s fiscal year, subject to the restriction that the fiscal year ended at least 6 months earlier to make sure the data was available to the public in real time. As an alternative, Hasler chose the book value of a company reported at the last interim or final results as long as that result was at least six months prior. So, in the case of Fama and French, the book value used could be six months old sometimes and up to 17 months old if the company’s fiscal year coincidentally ended in the month after they formed their portfolios. In the case of Hasler’s alternative specification, the book value typically was between 6 and 9 months old because he admitted quarterly results as well.

In terms of market value, Fama and French used the market value of stocks at the end of the last fiscal year (same as book value), while Hasler alternatively used the most recent market value of the company or the market value that is one month old.

Fama and French excluded all companies with a negative book value from their study, while Hasler alternatively allowed companies with negative book value to be included.

Fama and French included financial companies but in later studies on the profitability factor excluded them because financial companies (especially banks) have much higher financial leverage than other companies due to the nature of their business and thus have a systematically distorted book-to-market ratio. Hasler alternatively excluded financial companies from the calculations.

Finally, Fama and French used the top and bottom 30% of companies in each year to form their portfolios, while Hasler allowed for this threshold to vary to 20% or 40%.

The different combinations of all these original and alternative ways to calculate the original value factor creates 96 different results. The chart below shows the monthly excess return of value stocks vs. growth for the period originally investigated by Fama and French from 1963 to 1991. The point marked ‘HML’ shows the original result from Fama and French, while the point market AHML is the average value premium of the 96 different specifications. 

The value premium in the original specification compared to other specifications

Source: Hasler (2021)

The difference between the original value premium and the average of the alternative specifications is 0.09% per month or about 1.1% per year. This difference is both statistically and economically significant and indicates that by pure chance, Fama and French may have picked a specification that accidentally ended up giving one of the largest value premiums they could have found. To be clear here, I don’t accuse Fama and French of any wrongdoing or p-hacking. When we do research we all make tons of these decisions and while we are careful to make sensible decisions and check for potential distortions, no researcher can ever check all possibilities or may simply fail to consider them.

That the historical value premium may have been reported as too high can also be seen by Hasler’s out-of-sample tests. If he examines the value premium for the time period following the original publication by Fama and French, he finds that the average value premium is statistically indistinguishable from zero. This indicates, that the original publication from Fama and French may have been upward biased by chance and that the decline of the value factor we have seen over the last 30 years was simply a return towards more normal premiums that were observable all along had we measured it in alternative ways.