I first propose a quantile-based robust measure of tailedness. The empirical estimates of the new measure indicate that the assumed thick-tailed distributions in the recent literature have to some extent overestimated the degree of macroeconomic tail fatness due to the ambiguity of kurtosis. Further comparing the assumed thick-tailed distributions in forecasting macroeconomic dynamics multiple-period-ahead, I find clear evidence of the following best-performing specifications: the symmetric exponential power distribution for forecasting quarterly macroeconomic dynamics and the symmetric Student’s t distribution with time-varying volatility for forecasting monthly macroeconomic variables. Finally, the forecasting performance decomposition suggests that modeling tail fatness in macroeconomic disturbances provides significantly better predictive content than the benchmark models.