We document that the expectations of households, firms, and professional forecasters
in standard surveys simultaneously extrapolate from recent events and underreact to new
information. Existing models of expectation formation, whether behavioral or rational,
cannot account for these observations. We develop a rational theory of extrapolation
based on limited attention, which is consistent with this evidence. In particular, we show
that limited, asymmetric attention to procyclical variables can explain the co-existence
of extrapolation and underreactions. We illustrate these mechanisms in a microfounded
macroeconomic model, which generates expectations consistent with the survey data,
and show that asymmetric attention increases business cycle fluctuations.