Approximation des identités comptables dans les modéles spécifiés en logarithme

When a linear econometric model is expressed in logarithms, the presence of identities entails a strong non-linearity for the simulation stage of the complete model. This aspect can be damageable for the simulation algorithms which are implemented on microcomputers. A traditional solution consists in linearizing the identity by a Taylor expansion around the mean of the variables. In this paper we propose an approximation based on an error correction mecanism which takes into account the statistical properties of the series. The quality of the approximation is shown on a three equation econometric model estimated on french quartely data covering the period 1963-1985.