Referee Finance

Home of the Financial Intelligence Lab

  • Increase font size
  • Default font size
  • Decrease font size

Reinforcement Learning and Investor Behavior

E-mail Print PDF

Reinforcement Learning and Investor Behavior 

In this paper, James J. Choi, David Laibson, Brigitte C. Madrian and Andrew Metrick demonstrate how investors tend to chase returns and avoid variance with respect to their idiosyncratic 401(k) returns history. If an investor realizes higher returns from his asset allocation within 401(k), then he may decide to increase his 401(k) contributions on the basis of this rationally learned skill. Consequently, this may induce the investor to hold mutual funds that have recently performed well. The authors prove naïve reinforcement learning heuristic i.e. investors expect that investments in which they personally experienced past success will be successful in the future as well, whether or not such a belief is logically justified. Even though, returns chasing and variance avoidance reduces with age, investors continue to follow the above heuristics in their sixties. 

I.Data description
The authors describe the sources of their data.
II.Empirical methodology
The authors attempt to establish a relationship between changes in an individual’s 401(k) contribution rate and two years of 401(k) returns.
A.Summary statistics
The authors summarize the statistics of their results.
B.Main contribution rate change regressions
The authors confirm that the results are consistent with individual investors who abide by naïve reinforcement learning heuristic.
C.Interactions with age and salary
Returns chasing and variance avoidance diminishes swiftly with old investors, although, they tend to exhibit naïve reinforcement learning behavior for the remaining part of their lives.   
IV.Alternative explanations
A.Learning about investing skill
Investors who experience high 401(k) returns with low variance learn that they have greater skill at 401(k) asset allocation than their coworkers who experience low 401(k) returns with high variance. Therefore, the investors with better performance rationally allocate more to their 401(k). In a way, returns chasing and variance-avoidance is driven by rational learning about one’s own investing skill.
The authors infer the rebalancing high 401(k) returns with high non-401(k) asset returns fails to explain volatility avoidance among investors.
The authors construe that investors who experience high returns and/or low variance in the 401(k) portfolio increase their 401(k) contributions more than workers in the same savings plan who experience low returns and/or high variance. This behavior of return chasing and variance avoidance is not welfare-improving since 401(k) portfolio performance is not always consistent. The authors explain naïve enforcement learning heuristic wherein investors who have personally experienced success in assets expect those assets to be successful in the future. Finally, reinforcement learning stays through the entire life of an investor even when return chasing and variance avoidance lessens with age.  

Choi, James J., Laibson, David I., Madrian, Brigitte C. and Metrick, Andrew, "Reinforcement Learning and Investor Behavior" (September 14, 2007). Available at SSRN:



You can follow us on Twitter  @fintlab and facebook