Analyst sentiment and earning forecast bias in financial markets

Abstract With the natural language processing technique, we quantitatively measure the sentiment expressed in analyst reports of stock markets. Both the analyst sentiment and the earning forecast biases decrease with the fiscal month. The temporal variation of the earning forecast biases displays an asymmetric behavior, i.e., underestimated initial biases may be corrected by the wisdom of crowds, while overestimated ones could not. The Pearson correlation between the analyst sentiment and the earning forecast errors becomes positive, when approaching the end of the fiscal year. In addition, the analyst sentiment is different from the common investor sentiment.