Measuring Cross-Country Differences in Misallocation

In this paper, we discuss the role that data processing and collection have for the measurement of misallocation. First, we turn to the raw self-reported data for the US, reflecting what can be found in most developing countries. In the raw data, measured misallocation (following Hsieh and Klenow 2009) is substantially higher than for any other country for which we have census data. For instance, if Indian firms had the same dispersion of distortions as measured in the reported US data, TFP in the Indian manufacturing sector would decrease by around 2/3. Second, we follow a different strategy for editing and imputing missing data than what is used by the US Census Bureau, by using a method that seeks to replicate the true variance in the underlying data generating process known as Classification and Regression Trees (CART). This change raises the potential gains from removing misallocation in the United States manufacturing sector by around 10%.