The problem of endowing large applied general equilibrium models with
numerical values for parameters is formidable. For example, a complete
set of own- and cross-price elasticities of demands for the MONASH mo
del of the Australian economy involves in excess of 50 000 items. Invo
king the minimal assumptions that demand is generated by utility maxim
ization reduces the load to about 26 000 - obviously still a number mu
ch too large for unrestrained econometric estimation. To obtain demand
systems estimates for a dozen or so generic commodities at a top leve
l of aggregation (categories like 'food', 'clothing and footwear', etc
.), typically Johansen's lead has been followed, and directly additive
preferences imposed upon the underlying utility function. With the mo
ve beyond one-step linearized solutions of applied general equilibrium
models, the functional form of the demand system adopted becomes an i
ssue. The most celebrated of the additive-preference demand systems, t
he linear expenditure system (LES), has one drawback for empirical wor
k; namely, the constancy of marginal budget shares (MBSs)- a liability
shared with the Rotterdam system. To get around this, Theil and Cleme
nts used Holbrook Working's Engel specification in conjunction with ad
ditive preferences; unfortunately both Working's formulation and Deato
n and Muellbauer's AIDS have the problem that, under large changes in
real incomes, budget shares can stray outside the [0, 1] interval. It
was such behaviour that led Cooper and McLaren to devise systems with
better regularity properties. These systems, however, are not globally
compatible with any additive preference system. In this paper we spec
ify, and estimate, at the six-commodity level, an implicitly directly
additive-preference demand system which allows MBSs to vary as a funct
ion of total real expenditure and which is globally regular throughout
that part of the the price-expenditure space in which the consumer is
at least affluent enough to meet subsistence requirements.