We demonstrate that common modeling assumptions underlying micro-unit productivity indices induce biases in the evolution and decomposition of standard aggregate productivity measures. After controlling for such biases, we decompose aggregate productivity based on groups of economically significant firm types. We show that large incumbent firms that both export and import determine the evolution of aggregate productivity for the Belgian manufacturing sector. Over time, the increase in average productivity outweighs the decline in the covariance between market shares and productivity of this group. The former result stems from stronger learning-by-doing effects for granular firms. The latter suggests an increase in resource misallocation due to market distortions. This pattern intensifies after the 2008 financial crisis. All other firm types, if anything, contribute negatively to aggregate productivity and productivity growth.