Production, Sales, and the Change in Inventories: An Identity That Doesn't Add Up

We examine two measures of monthly manufacturing production. The first is the index of industrial production; the second is constructed from the accounting identity that output equals sales plus the change in inventories. We show that the means, variances, and serial correlation coefficients of the log growth rates differ substantially between the two series, and the cross-correlations are in most cases less than 0.4. A model of classical measurement error indicates that in 15 of 20 two-digit industries measurement error accounts for over 35 percent of the variation in the monthly growth rates of seasonally adjusted industrial production.