Rationalizing Firm Forecasts
We partner with a large US payment-processing company to run a 5-year, 10 wave panel survey of incentivized quarterly sales forecasts on over 6,000 firms. We match firm predictions to proprietary revenue data to assess accuracy. We find firms forecast poorly, with issues of inaccuracy, over-optimism, predictable errors and over-precision. To assess the causes of these forecasting issues we run experiments on: (i) data use, (ii) incentives, (iii) forecasting skill, and (iv) contingent thinking. We find greater data use primarily decreases noise and reduces over-precision, while higher incentives moderate over-optimism. Both moderately increase accuracy. The other two treatments have no impact. These results suggest forecasting biases can be reduced but are hard to eliminate. In a simple simulation model, we show these biases change firm responsiveness to changes in taxes and productivity, highlighting their macro importance.