Who do investors trust more, expert advisors or algorithms? In her March 2019 paper entitled “Algorithmic Decision-Making: The Death of Second Opinions?”, Nizan Packin employs a survey conducted on Amazon Mechanical Turk to assess automation bias when making significant investment decisions. Each of four groups of respondents received one of the following four questions (response scale 1 to 5):
- “You decide to invest 15% of your savings in the stock market. You find a reputable stockbroker, who makes investment recommendations. How confident are you that you got the best recommendation possible for your investment?”
- “You decide to invest 60% of your savings in the stock market. You find a reputable stockbroker, who makes investment recommendations. How confident are you that you got the best recommendation possible for your investment?”
- “You decide to invest 15% of your savings in the stock market. You find a reputable online automated investment advisor, who makes investment recommendations. How confident are you that you got the best recommendation possible for your investment?”
- “You decide to invest 60% of your savings in the stock market. You find a reputable online automated investment advisor, who makes investment recommendations. How confident are you that you got the best recommendation possible for your investment?”
A followup question asked about level of comfort trusting again the same expert (human or algorithmic) after learning that the initial recommendation resulted in a significant loss. Analyses included controls for respondent age, gender, socioeconomic status, having some college education, race and political ideology (liberal/conservative). Based on 800 total responses to specified survey questions, she finds that:
- Respondents significantly prefer recommendations from algorithms over those of human experts. Moreover:
- They still prefer algorithms after poor initial performance.
- Results are consistent across demographics.
- Deference to an algorithm fosters avoidance of second opinions, at potential expense of human (or alternative algorithmic) creativity, innovation and critical thinking.
- Nudging individuals to seek second opinions and requiring audits of algorithms may be beneficial.
In summary, survey results indicate that individual investors are much more comfortable with advice from algorithmic models than from reputable stock brokers.
Cautions regarding findings:
- Only a modest part of the paper consists of original survey research. Most of the content is a review of findings from other papers.
- Amazon Mechanical Turk survey respondents may not be representative of all investors, and survey responses may not be representative of actual investor behaviors.
- It is not obvious that intervention of specially aware public policy experts to suppress automation bias would be preferable to eventual self-correction by individuals.
For other perspectives, see “Machine Learning Factor?”, “Chess, Jeopardy, Poker, Go and… Investing?” and “Models vs. Experts”.
