This paper investigates the empirical evidence of long-run risk and its implications for the equity premium puzzle. We find that the long-run risk model is generally weakly identified and that standard inferences tend to underestimate the uncertainty of long-run risk. We extend the LM-type test of Ma and Nelson (2010) that remains valid under weak identification to the bivariate VARMA-GARCH model of consumption and dividend growth. The results cast doubt on the validity of long-run risk as an explanation for the equity premium puzzle. We also evaluate the approach of Bansal, Kiku, and Yaron (2007a), which extracts long-run risk by regressing consumption growth and its volatility on predictive variables. The results using the Bonferroni Q-test of Campbell and Yogo (2006) suggest that consumption and dividend growth are generally unpredictable by the price-dividend ratio and risk-free rate. This casts doubt on the validity of the BKY approach.