Author(s)

Niko Matouschek

Steven Callander

Learning by trial and error is ubiquitous. It is also risky. Success is not guaranteed as breakthroughs are mixed with setbacks and the path of learning is typically far from smooth. How decision makers learn by trial and error and the efficacy of the process, therefore, are inextricably linked to the incentives of the decision makers themselves and, in particular, to their tolerance for risk. In this paper we develop a model of trial and error learning with risk averse agents who learn by observing the choices of earlier agents and the outcomes that are realized. We identify sufficient conditions for the existence of optimal actions. We show that behavior within each period varies in risk and performance and that a knowledge trap develops, such that low performing agents opt to not experiment and thus fail to gain the knowledge necessary to improve performance. We also show that the impact of risk reverberates across periods. We identify a causal loop between performance, risk aversion, and experimentation that leads to divergence in long-run performance across agents. This implies that initial differences matter and have a long run impact on performance. Early leaders are more likely to remain leaders over time despite the ups and downs in learning, a persistence that matches observation in a variety of settings.
Date Published: 2016
Citations: Matouschek, Niko, Steven Callander. 2016. The Risk of Failure: Learning by Trial and Error and Long Term Performance. American Economic Journal: Microeconomics. (1)44-78.