Optimal Incentives for Learning

Project: Research

View graph of relations


Many full-time stock traders and stock investors make use of several complicated technical models to help them in making decisions. These technical models are intertwined with uncertainty; the profitability of different technical models is usually not known by any party. The stock trader can increase his knowledge by using a different model every period but this involves an experimentation cost. How should the employer design incentives for the stock trader in this setting?The majority of the principal agent literature focuses on asymmetric information among parties, i.e., one party has information that is not known by the other party. For example, the employee knows his skills, but the employer is not sure, or the management knows the true profitability of the firm but the outsiders do not. However, real life organizations also face situations where no party knows the true productivity/profitability of a certain task/method/model. How should the employer optimally motivate the employee to learn a certain task or method? The literature in this setting is almost non-existent. This paper attempts to fill this gap in the literature. The objective of the paper is to contribute to the formal modeling and optimal design of incentives in a repeated learning environment.As part of the research methodology, we develop a formal/theoretical model that consists of an employer (‘principal’), an employee (‘agent’). Every period the agent faces a choice between employing a safe method or risky method. The safe method has a success rate that is known, and employing the safe method is costless. The risky method has a success rate that is not known to either party. For example, if the risky method is high type, it has a high success rate. If the method were known to be of high type, the agent would always use the risky method. The agent by using the risky method can increase his knowledge about the type of the method but this experimentation is costly, i.e., the agent has to invest in new knowledge, need to allocate more of his time etc. On the other hand, the principal (‘she’) does not have the time to monitor which method the agent chooses each period. She observes only the outcome of the task (i.e. the realized gain or profit), and hence can design formal contracts depending on the outcome.An optimal wage contract for the principal must achieve two goals in this repeated learning environment: first, it should provide optimal incentives for the agent to experiment, and second, it should minimize principal’s wage bill. The objective, and the major contribution of the research, is to explicitly characterize optimal wage contracts when the principal can commit to stationary (time independent) and non-stationary (time independent) wage contracts. Our model incorporates both hidden action and hidden information. Such models are notoriously challenging when the uninformed party (here, the principal) makes the offers.


Project number7008122
Grant typeSRG
Effective start/end date1/05/1125/03/14