We examine the asymmetric impact of shocks to macroeconomic expectations and their underlying dispersion on equity risk premia across different market regimes. First, we rely on a two-state logit mixture vector autoregressive model and use Consensus Economics survey data on GDP growth, inflation, and short-term interest rates to approximate macroeconomic expectations and the underlying disagreement in the United States for the period 1989M10-2022M09. We demonstrate that unexpected changes of survey forecasts and their dispersion significantly affect cyclical factor returns in a dynamic setting and that the state of the economy matters for the magnitude, persistence, and occasionally also for the sign of the effect. Second, by extending the dynamic asset pricing model of Adrian et al. (2015), we show that GDP forecasts and their dispersion are priced in the cross section and drive the size and value premium, whereas inflation expectations serve as robust predictors for the price of risk. We also document that the survey expectationsaugmented specification reduces pricing and premium errors when compared to a common benchmark of return predictors.