We analyze a panel of over 28,400 S&P 500 return forecasts by CFOs to examine whether the extent of CFOs’ miscalibration—providing forecast confidence intervals that are too narrow—decreases over time. We find no improvement with task repetition nor evidence of learning, that is, no improvement in response to past performance. Across CFOs, miscalibration appears to be a persistent personal trait. We find some evidence that the degree of miscalibration is related to birth cohort and stock market familiarity.