Should investors view financial experts as individuals who, through years of study and experience, overcome behavioral biases and reliably add value to investment decisions? In his May 2014 essay entitled “Are You Trying Too Hard?”, Wesley Gray summarizes research that compares the decision-making of experts to the performance of mechanical models across many fields. He highlights the relevance to of this research to investment decision-making. Based on the body of research pitting expert judgment against mechanical models, he concludes that:
- Contributions from human experts are critical in development and assessment of decision-making processes, but expert tinkering generally degrades process operation.
- A 2000 review of 136 published experts-versus-models studies across many fields finds that:
- Models beat experts 46% of the time.
- Models equal or beat experts 94% of the time.
- Experts beat models only 6% of the time.
- Specific studies suggest that expert second-guessing of model decisions tend to degrade rather than enhance performance.
- Human experts underperform simple models in decision-making because humans:
- Are inconsist due to biases such as: attachment to irrelevant data (anchoring); irrational response to the mode of data presentation (framing); overemphasis of recent or easily recalled information (availability); and, influences from physical state (such as tiredness).
- Prefer a simplifying (and entertaining) narrative/story to primary evidence.
- Tend both to overvalue additional information and to interpret this information in a way that amplifies prior beliefs (self-attribution bias), thereby feeding overconfidence.
In summary, evidence from research across many fields suggests that investors, while perhaps consulting experts for model development, should humbly relegate routine investment decisions to mechanical models.
Cautions regarding conclusions include:
- Financial modelers generally develop models via testing in simplified/constrained data environments. Human financial experts may operate intuitively based on more realistic (complex, adaptive, wild) expectations.
- It may be difficult to draw a line between model interference and model adjustment.
