Central Bank Communication and Tail Risk in Inflation Expectations

Abstract This paper integrates extreme value theory, rational inattention, and social learning into a framework for inflation expectations under heavy-tailed signals. The framework delivers a tail-biased worst-case forecast, entropy-dispersion tail dependence that rises in the tail index and falls in attention capacity, a globally attracting social learning steady state, and an optimal heavy-tailed signal. In quarterly United States data from 2000 to 2025, headline inflation has a heavy upper tail, Federal Reserve statement entropy explains roughly 30 % of the variation in household survey forecast errors, and the slope of forecast bias on tail-loss covariance flattens after the 2012 inflation target adoption. The structural levers are communication entropy and the inflation tail index.