Usually firms operate with a restricted number of energy sources. In t
he dairy industry, most firms have multi-energy systems allowing them
to shift from one energy source to another at almost no cost. Zero exp
enditures are observed because firms minimize costs and non-negativity
constraints on the demand functions are binding. Estimation of demand
systems when the probability of observing zero expenditures is not ni
l has already received attention from econometricians. However, most s
urveys generally report prices only for the subset of goods actually p
urchased. The econometrician faces a problem of missing price observat
ions when zero expenditures occur. The originality of our approach is
to propose a combined and coherent treatment of both the zero expendit
ures and missing data. Price equations are added to the demand system
and the cost-minimizing mix of energy inputs leads to a simultaneous e
quation/limited dependent variable model. A general framework is provi
ded in which it is possible to formulate parameter restrictions which
guarantee consistency of the cost minimizing model and its relationshi
p to a generalized tobit model with errors in variables. An applicatio
n to a sample drawn from a survey on firms of the French dairy industr
y shows the strength of a model which decomposes the choice of energy
mix in two parts: a qualitative preference and a quantitative decision
. A micro-simulation model gives evidence of the practical use of this
study. Nonetheless some theoretical points remain unsolved and should
motivate further research. (C) 1998 Elsevier Science B.V. All rights
reserved.