Can analysts predict rallies better than crashes?

•We estimate tail dependence between analyst recommendations and security returns.•We use semiparametric copula model fitted to I/B/E/S consensus data.•Analysts' predictive ability is found to be highly asymmetric between tails.•Analysts' found able to identify substantially undervalued but not overvalued stocks.

We use the copula approach to study the structure of dependence between sell-side analysts' consensus recommendations and subsequent security returns, with a focus on asymmetric tail dependence. We match monthly vintages of I/B/E/S recommendations for the period January-December 2011 with excess security returns during six months following recommendation issue. Using a mixed Gaussian-symmetrized Joe-Clayton copula model we find evidence to suggest that analysts can identify stocks that will substantially outperform, but not underperform relative to the market, and that their predictive ability is conditional on recommendation changes.