A theory of structural change that can fit the data

We propose a dynamic theory that is consistent with the long-run structural change of consumption expenditure shares in agriculture, manufacturing and services over more than a century. We first document three robust features in the long-run data for the United States, the United Kingdom, Canada and Australia: (i) a monotonic decrease in agriculture; (ii) a hump shape in manufacturing; and (iii) an accelerated rise of services. Using historical panel data on sectoral prices and nominal expenditures from 1900 to 2014, we then test what demand side theory can quantitatively explain the observed structural change. We find that the standard non-homothetic preference specifications used in the literature - the generalized Stone-Geary and the Price Independent Generalized Linearity (PIGL) specification - struggle to do so. Within our intertemporal framework, we then consider the entire class of preferences that allows for the aggregation of individual Euler equations - the class of intertemporally aggregable (IA) preferences. This general class nests the standard specifications and allows for the identification of preference parameters from aggregate data. Moreover, its expenditure system is flexible enough to capture the non-monotonic relationship between sectoral expenditure shares. Despite the flexibility, the IA preference specification is parsimonious and can be used in a multi-sector general equilibrium model with steady growth and heterogeneity in individual consumption expenditures. In the empirical analysis we show that the standard specifications are rejected against the more flexible parameterizations of IA preferences and we document the importance of flexible income effects.