Consensus forecasts and ine¢ cient information aggregation

Consensus forecasts are ine¢ cient, because the weight placed on aggregate new private information relative to the prior re‡ects the optimal choices of the underlying individual forecasters— whose private signals are noisier individually than in aggregate— and as a result the consensus forecast over-weights the prior (Lim, Kim and Shaw, 2001). This paper assesses the empirical relevance of this result using a cross-country panel of growth forecasts. There is strong evidence that consensus forecasts are in fact ine¢ cient. Perhaps more surprisingly, there is no evidence that this ine¢ ciency is understood and corrected for by forecasters themselves. These results have implications for information aggregation via the price mechanism— o¤ering one explanation for the positive short run serial correlation observed in stock prices and the apparent success of momentum trading strategies, and posing a challenge for the e¢ cient markets hypothesis more generally— and also for the wider literature on information processing (Morris and Shin, 2002; Benabou, 2009).