Risk aversion and effort under an incentive pay scheme with multiplicative noise: Theory and experimental evidence

The purpose of this paper is to consider the influence of individual risk preferences on the effectiveness of incentive pay schemes, by examining the link between individual effort and risk aversion in situations where outcome uncertainty multiplies with effort. Such “multiplicative noise” situations are common, occurring whenever payment is awarded per success rather than per attempt.

The paper develops a theoretical model which predicts a negative risk aversion-effort link under multiplicative noise without a performance target (PT), and a weaker negative link once the target is introduced. This model is then taken to the data from a lab experiment where participants were randomly assigned to a control group, which received fixed pay, and a treatment group, which received a piece rate awarded with a certain probability, with and without a PT. Risk aversion is measured with a menu of lottery choices offered at the end of the experiment.

Compared to their peers in the control group, the more risk-averse participants in the treatment group put in progressively less effort in the absence of a PT. The introduction of a PT substantially weakens this negative risk aversion-effort link, so that there are no more significant differences in performance between the more and the less risk averse.

The paper’s findings speak to the empirical puzzles of incentive pay schemes backfiring and of the proliferation of PTs. The negative risk aversion-effort link may be one reason behind the failure of incentive schemes to deliver improved performance, whereas the weakening of this link may be one justification for the existence of PTs.

In the multiplicative noise environments, managers should take their workers’ risk preferences into account when designing incentive pay schemes. A PT may be a useful motivational tool for the risk-averse workers who are more likely to under-perform.

The multiplicative noise environment has been largely overlooked by the existing literature, yet it is common in practice. An example is the work of a sales agent who receives a bonus per sales which succeeds with a certain probability after each customer contact. This paper is one of the first to model, and test experimentally, worker performance in this environment.