With greater certainty in anthropogenic influence on observed changes in cl
imate there is increasing pressure for agreements to control emissions of g
reenhouse gases (Houghton et al., 1996). While it is difficult to assess th
e appropriate level of mitigation, it has been argued that flexibility in m
eeting emission targets offers significant economic savings. Such flexibili
ty can be exercised in terms of timing of mitigation (i.e. delay) or geogra
phic location of the intervention (e.g. permit trading and Joint-Implementa
tion). Much of this insight is based on standard models of technical change
in energy supply and demand. However, standard model formulations rarely c
onsider: (i) a link between the pattern of technical change and policy inte
rventions; (ii) economies of learning; and (iii) technical progress in disc
overy and recovery of oil and gas. While there is evidence to support the i
mportance of these factors in historic patterns of technical progress, the
data necessary to calibrate internally consistent economic models of these
phenomena have not been available. In this paper simple representations of
endogenous and induced technical change have been used to explore the sensi
tivity of mitigation cost estimates to how technical change is represented
in energy economics models. The scenarios involve control of CO2 emissions
to limit its concentration to no more than 550 ppm(v), starting in the year
2000, and delayed to 2025. This sensitivity analysis has revealed four rob
ust insights: (i) If endogenous technical change is assumed, expected busin
ess as usual emissions are higher than otherwise estimated - nevertheless,
while 25% greater CO2 control is required for meeting the CO2 concentration
target, the cost of mitigation is 40% lower; (ii) If technical progress in
oil and gas discovery and recovery is assumed, energy use and CO2 emission
s increase by 75% and 65%, respectively above the standard estimates; (iii)
If the economies of learning exhibited in various manufacturing sectors ar
e repeated in development of non-fossil technologies and abatement of CO2 e
missions, the costs of abatement can be 50% lower than those assessed using
standard models; and (iv) In this sequential learning framework, delay in
abatement towards a 550 ppm(v) CO2 concentration target leads to expected n
et economic loss in seven of nine model structures studied. Only when the m
odel structure permits new oil and gas discoveries while keeping other feat
ures of standard models does delay offer economic gain with greater than 60
% confidence. (C) 1998 Elsevier Science B.V. All rights reserved.