It is frequently argued that changes in economic variables are measured more reliably
than their levels. This may happen if measurement errors are autocorrelated. Autocorrelation
should be taken into account in the least squares balancing of national accounts.
Formulae are presented for the error structure which is likely to arise from extrapolation
away from or interpolation between bench-mark observations. Balanced estimates are
presented for the UK national accounts, 1920-38, taking into account the autocorrelation
likely to have been generated in data construction.