Forecasting earnings with combination of analyst forecasts

We propose a regression-based method for combining analyst forecasts to improve forecasting efficiency. This significantly reduces the bias in earnings forecasts, and generates that consistently outperform consensus over time across firms of different characteristics. Incorporating firm-level macroeconomic information model further improves performance. Forecasting gains increase with dispersion degree under/overreactions news. Moreover, combination forecast produces larger response coefficients, weakens anomaly post-earnings-announcement drift, provides better expected profitability measure has higher power predict stock returns.