Back to History

Fields Changed

Registration

Field Before After
Trial Title Respecting or Debiasing Investor Preferences? A Field Experiment on the Effectiveness of Portfolio Recommendations in Enhancing Investor Welfare Robo‑Advising: How Investors Respond to Preference‑Based vs. Debiasing Recommendations
Trial Status in_development on_going
Abstract The household finance literature has found that investors often deviate from optimal investment behavior and suffer wealth losses due to factors such as incomplete information, lack of financial knowledge, and behavioral biases. Investment advisors perform multiple functions including information provision, investor education, and asset allocation recommendations, serving as crucial means to assist investors. Currently, there are two popular logics for investment advisor’s asset allocation recommendations: catering to investor preferences and educating investor. Exploring which asset allocation logic and design are more comprehensive, trustworthy, and beneficial to investors’ welfare holds significant implications for the upgrade of advisory services, development of financial markets, and enhancement of social welfare. The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. The control group will be recommended a uniform non-personalized asset allocation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The household finance literature has found that investors often deviate from optimal investment behavior and suffer wealth losses due to factors such as incomplete information, lack of financial knowledge, and behavioral biases. Investment advisors perform multiple functions including information provision, investor education, and asset allocation recommendations, serving as crucial means to assist investors. Currently, there are two popular logics for investment advisor’s asset allocation recommendations: catering to investor preferences and educating investor. Exploring which asset allocation logic and design are more comprehensive, trustworthy, and beneficial to investors’ welfare holds significant implications for the upgrade of advisory services, development of financial markets, and enhancement of social welfare. The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. All users will first report their current asset allocation and complete a series of questionnaires on basic information and behavioral preferences. The control group will not receive any recommendation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. Finally, all users will report the asset allocation they consider to be most appropriate.
Trial Start Date December 01, 2023 August 01, 2025
Trial End Date June 01, 2024 August 01, 2026
Last Published December 15, 2023 03:28 PM February 24, 2026 10:27 PM
Intervention (Public) The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. The control group will be recommended a uniform non-personalized asset allocation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. All users will first report their current asset allocation and complete a series of questionnaires on basic information and behavioral preferences. The control group will not receive any recommendation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. Finally, all users will report the asset allocation they consider to be most appropriate.
Intervention Start Date December 01, 2023 November 01, 2025
Intervention End Date May 01, 2024 May 01, 2026
Primary Outcomes (End Points) Browsing behavior, Asset reallocation behavior Asset allocation decision
Experimental Design (Public) The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. The control group will be recommended a uniform non-personalized asset allocation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The experiment will randomly divide users of a certain bank’s APP into four groups: control group, preference catering group, loss aversion education group, and mental accounting education group. All users will first report their current asset allocation and complete a series of questionnaires on basic information and behavioral preferences. The control group will not receive any recommendation. The preference catering group will be recommended an asset allocation that is algorithmically calculated to be optimal based on the investor’s behavioral parameters. The loss aversion education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the loss aversion parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. The mental accounting education group will be recommended an asset allocation that is algorithmically calculated to be optimal but with the modification of the mental accounting parameter to be fully rational while keeping the investor’s other behavioral parameters unchanged. Finally, all users will report the asset allocation they consider to be most appropriate.
Planned Number of Clusters about 2000000 individuals About 200,000 individuals exposed to the experiment (about 8,000 actually entering the experiment)
Planned Number of Observations about 2000000 individuals About 200,000 individuals exposed to the experiment (about 8,000 actually entering the experiment)
Sample size (or number of clusters) by treatment arms 500000 individuals control group, 500000 individuals preference catering group, 500000 individuals loss aversion education group, 500000 individuals mental accounting education group 50,000 individuals exposed (about 2,000 actually entering the experiment) in control group; 50,000 individuals exposed (about 2,000 actually entering the experiment) in preference catering group; 50,000 individuals exposed (about 2,000 actually entering the experiment) in loss aversion education group; 50,000 individuals exposed (about 2,000 actually entering the experiment) in mental accounting education group
Secondary Outcomes (End Points) Browsing behavior
Back to top

Fields Removed

Other Primary Investigators

Field Value
Affiliation Peking University
Back to top