Noise in Expectations: Evidence from Analyst Forecasts

This paper quantifies the amount of noise and bias in analysts’ forecast corporate earnings at various horizons. We first show analyst forecasts outperform statistical short-horizons, but underperform longer next decompose relative accuracy these into three components: (i) noise, (ii) (iii) information advantage over forecasts. find is constant across forecasting horizons, while both are increase linearly. then most existing models lack a mechanism to account for facts. To generate such mechanism, we consider parsimonious variant model Patton Timmermann (2010) with noisy cognitive default it quantitatively fits data. The intuition underlying this that forecasters rely on their biased defaults more as rational less accurate. also matches two non-targeted empirical relationships: disagreement increases horizon an increasing function volatility.