Comment

The paper by Kozlowski, Veldkamp, and Venkateswaran argues that economic agents rationally revised their estimates of tail risk following the Great Recession and that this revision explains, at least in part, the persistent decline of interest rates on safe and liquid assets such as US Treasury securities. In a previous paper (Kozlowski, Veldkamp, and Venkateswaran 2015), the authors argued that the same belief revision can explain the slow recovery of investment and output. One important contribution of this work ismethodological: they propose a tractable approach to embedding learning dynamics in fairly standard quantitative models. Substantively, the overall argument is quite plausible, and I believe the remaining issues are really quantitative: How much did people’s beliefs about tail risk change after the Great Recession? And how sensitive are interest rates (in this paper) or economic activity (in the previous paper) to perceived tail risk? In this discussion, I will address the first question briefly, before turning to the second, and then dissect themechanisms throughwhich interest rates depend on tail risk in the paper. In Kozlowski and colleagues’ model, the risk-free asset combines two qualities: it is safe, and it is excellent collateral. Conceptually, one can separate these two characteristics, even though they are joint in the model and, to some extent, in the data. This allows us to distinguish twomechanisms throughwhich higher tail risk increases the value of the risk-free asset. First, agents’ willingness to pay for safe assets increases with tail risk. I will call this the “safety channel.” This is a standard precautionary savings effect, a wellknown piece of canonical asset-pricing theory. Second, agents’ willingness to pay for assets that are good collateral increases with tail risk, in large part because the tail risk reduces investment and thus the supply