This is the final instalment in a two-part series on the parallels between factor investing and trait psychology. My last post explored the similarities in the development of the two fields of research. These include:
- The use of correlation analysis (pioneered in trait psychology)
- A single, foundational data set
- The identification of a handful of factors that can be used to explain and predict results
- The criticisms levelled against them.
In this post, we’ll explore three of these criticisms in more detail.
In 1968, the Austrian-born American psychologist Walter Mischel published Personality and Assessment. Mischel found that numerous empirical studies failed to support the belief that humans behave consistently across different situations (evidence of stable personality traits).
He found that the correlation between traits and behaviour was in most cases weak, with correlations rarely greater that 0.3. In other words, there was a great deal of within-person variability across different situations.
Mischel proposed an alternative view of behaviour. Instead of personality traits being a “signal” of how a person is likely to behave (with the “situation” being just noise), it’s the situation that’s the “signal” (and the trait’s the “noise”). Behaviour was determined by if then relationships which Mischel called “personality signatures”. These personality signatures were highly dependent on situational cues.
This reminds me of the 2016 paper by Rob Arnott, Noah Beck, Vitali Kalesnik and John West of Research Affiliates (“RAFI”) entitled How Can Smart Beta Go Horribly Wrong?. The authors argue that much of the historical performance of popular factors was due to the situation, not the factors themselves:
Many of the most popular new factors and strategies have succeeded solely because they have become more and more expensive. Is the financial engineering community at risk of encouraging performance chasing, under the rubric of smart beta? If so, then smart beta is, well, not very smart.
The charts on pages five and six of their paper show the relationship between relative factor valuation (the “if” or situational cue) and relative factor performance (the “then” or behaviour) over the subsequent period. Higher factor valuations result in lower factor returns over the subsequent period.
Mischel’s critiques of trait psychology drew a predicably strong response from trait psychologists, who argued that:
- Even correlations of 0.3 may have predictive value. For example, such a correlation allows for a 30 per cent reduction in prediction errors, and raises the odds of correct prediction from 1:1 to almost 2:1.
- Mischel underestimated the degree to which traits predict behaviour. His review focused on studies of behaviour in specific instances, not tendencies (i.e. whether or not behaviours are repeated) over time.
Subsequent research showed that aggregated behaviour (i.e. tendencies over time) was more consistent with trait measures than Mischel believed. Still, his work was important because it lead psychologists to a healthy recognition of the importance of situational factors and their role in within-person variability.
In a similar way, the quantitative investment community haven’t pulled their punches in replying to RAFI’s situational critique of factor investing. For example, two blog posts by Cliff Asness at AQR:
Asness’ counter-argument is that stocks move in and out of factor portfolios all the time. And the only way that valuation could be the primary driver of factor returns would be if the same stocks remained in the portfolio and became more expensive.
So how do I reconcile the two viewpoints? I think factor investing survives the situational criticisms levelled against it. There is enough long-term and out-of-sample evidence to suggest that a handful of factors can be used to explain and possibly predict investment returns.
Just like Mischel’s critique of trait psychology, RAFI’s research is an important reminder that factors don’t exist in a vacuum. Context matters! Factors may work over the very long-term. But the within-factor variability in performance over shorter periods is significant. Investors need to consider the variables that might affect factor performance in the future.
Readers of my last post will remember that trait psychology got its start with the lexical hypothesis. American psychologists Gordon Allport and Henry Odbert went through the Webster’s New International Dictionary and identified 17,953 unique words in the English language that relate to personality and behaviour. Their research was based on the belief that language evolves to describe personality types and their associated behaviour.
You might be wondering are personality traits culturally universal? Will a set of factors identified using the English language as a sample apply to other languages and cultures?
The research is inconclusive. Personality tests such as the Big Five and the Eysenck Personality Questionnaire have been translated into several languages. These tests in other languages typically yield more or less identical underlying factors, although the results are weaker in non-European languages.
That said, there are subtle cultural variations. For example, Big Five factors such as Extraversion and Agreeableness don’t appear in Filipino, Japanese and Korean samples but are replaced by other factors. They are: Love (a blend of Extraversion and high-Agreeableness) and Dominance (Extraversion and low-Agreeableness).
Researchers have responded by investigating personality from an indigenous perspective. Rather than use a translated set of trait descriptors, researchers start from scratch and build a database of trait descriptors in the language of interest.
Similar questions have been posed of factor investing. The Center for Research in Securities Prices (CRSP) performed the first long-term study of equity returns using data from the US stock market. This database and it subsequent updates are the foundation for much of the seminal factor research.
Do the factors identified using US data also exist in other markets? There seems to be a lot of evidence that they do, but once again there are subtle differences between markets. Here are a few examples:
- Momentum in Japan
- Size in Australia
- Value in Australia
Its also worth noting that the majority of factor research follows a similar pattern:
- Identify the factor in the US market.
- Test it over different periods.
- See if the results hold in different markets.
The rest-of-the-world is often treated as just a source of out-of-sample data. Perhaps quants could follow the example of trait psychologists and also perform indigenous research? Of course, there are obvious reasons why most factor research is based on US data. The US is the world’s dominant equity market and there’s more data.
I can’t help feeling that more indigenous research will lead to some interesting discoveries. For example, I would really like to see more research into the behaviour of factor portfolios in heavily concentrated markets such as Australia and the Scandinavian countries.
Do Traits Explain Behaviour?
Some trait psychologists believe that traits are hypothetical “latent variables”. We cannot observe traits directly, but we can hypothesise that they exist because we see their effects (i.e. repeated behaviour). An alternative view is that traits do not cause behaviour (i.e. they are not latent variables), they just describe it.
For example, saying that a person is agreeable (kind, sympathetic, cooperative) could mean that Agreeableness is an innate quality that they possess. Or, it could simply mean that they’ve tended to behave agreeably in the past.
Personality traits are observations and not necessarily a cause or an explanation. In short, the explanatory power of personality traits could be limited.
Similar criticisms have been levelled against factors. There is no consensus on why factors exist. Some researchers believe that they are compensation for different kinds of investment risks. Meanwhile, other academics and practitioners believe that they’re the result of investor behaviour.
Some factors seem to stop working once discovered. This begs the questions: is the factor a latent variable? Or was it simply describing what had happened in the past? Either way, it can be hard to know for sure.
Hopefully investors using or considering factor investing strategies have found his brief tour of trait psychology interesting and helpful. There is often much that can be learned from other disciplines.
My biggest take away is that the link between factors (if they exist), returns and risk is quite fuzzy. Just how fuzzy is up for debate. But I’m reasonably certain that it’s a lot fuzzier than most smart beta and quant shops would have us believe.
 As a group (or factor exposure), small stocks have under-performed large stocks over the long-term in Australia. That said, Australian small cap stocks have been a happy hunting ground for active fund managers.