Learning from Mistakes

Abstract

We estimate the cost of re-sampling options considered and rejected in the past when the decision maker is learning by doing or searching for some object. We consider a stationary distribution of costs so that recalling past offers is not optimal. Re-sampling rejected options slows learning down, flattens the learning curve, and reduces utility. One can also fit the standard learning curve well, even if no re-sampling occurs during the learning process.

Keywords: Learning, Macroeconomics, Learning by Doing