Algorithmic portfolio choice: lessons from panel survey data

Abstract

Automated asset management offerings algorithmically assign risky portfolios to individual investors based on investor characteristics such as age, net income, or self-assessment of risk aversion. Using new German household panel data, we investigate the key household characteristics that drive private asset allocation decisions. This information allows us to assess which set of variables should be included in algorithmic portfolio advice. Using heavily cross-validated classification trees, we find that a combination of household balance sheet variables—describing the ability to take risks (e.g., net wealth)—and household personal characteristics—describing the willingness to take risks (e.g., risk aversion)—best explain the cross-sectional variation in household portfolio choice. Our empirical evidence is in line with models of portfolio choice under decreasing relative risk aversion and fixed investment costs. The results suggest the utility of a more holistic modeling of household characteristics. Including background risks in the form of household leverage not only makes investment sense, but is also the new regulatory reality under MIFID II rules. Robo-advisors are strongly advised to act accordingly.