AQR’s Cliff Asness recently wrote It Ain’t What You Don’t Know That Gets You into Trouble where he briefly summarises two recent AQR research papers. The first paper demonstrated that one of the original factors, the size premium, doesn’t exist. Apparently, the data was flawed and the flaws account for the return premium. Use more accurate data, and it turns out that any extra return is due to additional risk (i.e. beta) or illiquidity in the case of micro-cap stocks, but not size. Asness observes:
We know that it’s likely a sobering thought to many that the Ur‑anomaly, the one that’s been used to practically reorganize the entire money management industry, just isn’t there. But, as the man said, it just ain’t so.
Full credit to Asness and the team at AQR for following the evidence and setting the record straight. But Asness’ post also raises another question (also borrowing from Mark Twain): what else do we know that ain’t so? Or to put it more bluntly, of what value is factor research if we can’t be sure that the results are useful?
An interesting way to think about this question is to compare quantitative factor research with research in other fields that have faced similar issues. For example, trait (personality) psychology. Perhaps I’m biased – I started out as a psychology student – but I see close parallels between the two fields. The similarities include:
- The use of statistical analysis such as correlation and regression
- The foundational research is based on a common database
- A handful of “factors” or “traits” are used to both describe and predict
- The critiques levelled against trait psychology and factor investing
Using Probability and Statistics
British mathematician and biostatistician Karl Pearson pioneered the use of probability theory and statistics in academic research. Pearson developed tools such as correlation and regression analysis, borrowing concepts from his mentor Sir Francis Galton. They used correlation analysis to explore questions of heredity; including whether or not intelligence and personality are inherited.
In other words, the statistical tools used in finance and investment today got their start in fields such as psychology and biology. They were applied to the study of trait psychology long before they were ever used in finance.
Founded on a Database
Galton was also one of the first people to opine that human personality traits are reflected in language. His logic was simple. Traits are supposedly stable and enduring personality attributes. This means that they will be expressed in frequent, repeated behaviours. Naturally, language will evolve to describe these behaviours and the personality attributes associated with them. This is now known in Psychology as the Lexical Hypothesis. Galton describes his hypothesis in Measurement of Character, written in 1884:
I tried to gain an idea of the number of the more conspicuous aspects of the character by counting in an appropriate dictionary the words used to express them… I examined many pages of its index here and there as samples of the whole, and estimated that it contained fully one thousand words expressive of character, each of which has a separate shade of meaning, while each shares a large part of its meaning with some of the rest.
Several researchers used a similar approach, most notably the American psychologists Gordon Allport and Henry Odbert. In 1936, the pair went through the Webster’s New International Dictionary and identified 17,953 unique words in the English language that relate to personality and behaviour. They grouped these words into four categories:
- 4,505 words relating to personality attributes
- 4,541 words describing attitudes, emotions and moods (i.e. temporary states)
- 5,226 words used to evaluate another individual’s personality or character (i.e. worthy, insignificant, etc.)
- 3,682 words that didn’t fit the first three categories.
Allport and Odbert acknowledged that their categorisation was subjective and that the four groups were not independent of each other (i.e. some words could easily fit in more than one group). Their “database” proved to be incredibly influential as it became the basis for much of the trait psychology work that was to follow.
This reminds me of work performed by the Chicago (later Booth) School of Business in 1960, when the Center for Research in Securities Prices (CRSP) was launched. After three-and-a-half years of hard work, the Center published the first long-term (using 35 years data) study of equity returns.
Think about it, prior to 1963, nobody knew what the long-term return of US stocks was!
Over the years, the CRSP database grew to include more data as well as additional information on stocks. Just like Allport and Odbert’s work, the CRSP database became the foundation for most of the seminal work on factors.
Allport and Odbert’s list became the subject of a number of psychology studies, which ultimately culminated in the Big Five Model of Personality Traits, developed by American psychologists Dean Peabody and Lewis Goldberg.
The Big Five or Five Factor Model is one of the most widely used models of personality. It describes an individual’s personality across five dimensions: Extraversion (introversion), Agreeableness, Neuroticism, Openness and Conscientiousness.
In a similar way, finance researchers such as Sanjoy Basu (value), Rolf Banz (size), Eugene Fama and Kenneth French (size, value), Mark Carhartt (momentum) and many others have performed factor analysis to identify a set of factors that explain the performance of a portfolios of stocks.
While academics have identified literally hundreds of factors, most agree that only a handful are robust. There is considerable debate around which factors to include in this select group. The most likely candidates include: Market, Size, Value, Momentum, Profitability and Low Volatility (I guess we can now scratch size off the list). And finance also has its own widely-used five-factor model: The Fama and French five factor model.
OK, we’ve established that trait psychology and factor investing have a lot in common. What are some of the critiques of trait psychology? They include:
- The importance of context and ‘within-person variability’ (‘situationist’ critiques).
- Traits aren’t culturally universal.
- Personality is better described using types (categories) rather than continuous dimensions (traits).
- Traits are purely descriptive and not predictive.
- There are better ways to describe personality (e.g. values, interests, motives, etc.).
Close consideration of these critiques is illuminating as they have parallels in factor investing. For example:
- The performance of factors varies widely over time. There is also considerable debate over the effect of variables such as interest rates and factor valuations on factor performance.
- Not every factor works in all markets (e.g. momentum doesn’t seem to work in Japan).
- There are many ways to categorise and measure factors and the method chosen can have a large impact on results.
- Real-life factor portfolios often underperform back-tested factor portfolios.
- Should we be picking stocks using factors, fundamentals or some other method?
We’ll consider these critiques of trait psychology in my next post.
 AQR Footnote: As the paper stresses, there can still be a role for size in explaining monthly returns (in an R-squared sense not an expected return sense) and, importantly, most other factors seem to work better in small caps (at least gross of costs) meaning it might still be rational to tilt towards small, but you’re doing it to capture more premium from other factors, not the size effect itself.