By Martin Andreasen, Giovanni Caggiano, Efrem Castelnuovo and Giovanni Pellegrino
This paper uses a nonlinear vector autoregression and a non-recursive identification strategy to show that an equal-sized uncertainty shock generates a larger contraction in real activity when growth is low (as in recessions) than when growth is high (as in expansions). An estimated New Keynesian model with recursive preferences and approximated to third order around its risky steady state replicates these state-dependent responses. The key mechanism behind this result is that firms display a stronger upward nominal pricing bias in recessions than in expansions, because recessions imply higher inflation volatility and higher marginal utility of consumption than expansions.
I would have thought that the higher marginal utility of consumption in recessions was sufficient to explain why risk matters more in those periods. It appears there is more to it on the firm side. Firms like to set (nominally rigid) prices a bit higher in the face on uncertainty, and there is more uncertainty in recessions.