Correlation neglect in financial decision-making

Good decision-making often requires people to perceive and handle a myriad of statistical correlations. Notably, optimal portfolio theory depends upon a sophisticated understanding of the correlation among financial assets. In this paper, we examine people's understanding of correlation using a sequence of portfolio-allocation problems and find it to be strongly imperfect. Our experiment uses pairs of portfolio-choice problems that have the same asset span - identical sets of attainable returns - and differ only in the assets' correlation. While any outcome-based theory of choice makes the same prediction across paired problems, subjects behave very differently across pairs. We find evidence for correlation neglect - treating correlated variables as uncorrelated - as well as for a form of 1/N heuristic - investing half of wealth each of the two available assets.