ABSTRACT
Introduction It has been known for a long time that consistent aggregation of microeconomic relationships is only possible under restrictive conditions. Some classic and some newer references in this respect are Theil (1954), Muellbauer (1975), Barnett (1979). Yet, applied macroeconomics is mainly still proceeding under the representative agent assumption. Even when consistent aggregation is possible, for instance when relationships are linear, estimation of these relationships for macro and for micro data tends to yield quite different results. A standard example is a simple Keynesian consumption function, which typically exhibits a far larger marginal propensity to consume in macro time-series data than in micro cross-section data. Various explanations have been advanced for this phenomenon, including the relative in co me and the permanent income hypothesis (Duesenberry, 1967, Friedman, 1957). Pakes (1983) analyses in more general terms the differences between micro and macro relations that arise as a result of explanatory variables that exhibit a group structure. An example of such a situation is provided by the micro model used in this paper: preferences of agents are allowed to depend on the behaviour of others in their social reference group. For individual agents the behaviour of others is exogenous, but at a macro level one observes the behaviour of agents jointly and reference group behaviour is r10 Ion ger exogenous. Thus macro behaviour differs from micro behaviour (see section 12.1 for a more detailed exposition).
