A new formulation is proposed to solve design problems under model uncertai
nty. The proposed method combines well-known techniques for solving multipe
riod design problems with the concept of joint confidence regions. One way
to incorporate uncertainty into design problems is to discretize the uncert
ain parameters into a number of finite values and solve a multiperiod desig
n problem. Traditionally, the uncertain parameters are discretized to a low
er and upper bound based upon their individual confidence intervals. Howeve
r, a more accurate description of the model parameter uncertainty is availa
ble through the use of joint confidence regions. In this work, we propose c
hoosing the discrete values of uncertain model parameters based upon the pr
incipal components of their joint confidence region. We demonstrate the pro
posed method in several examples. The resulting optimal designs are more ac
curate because they incorporate the actual model uncertainty using joint co
nfidence regions. Moreover, faster convergence to an optimal solution is ob
served using a two-stage algorithm because fewer discretization points may
be needed in the multiperiod problem. (C) 1999 Elsevier Science Ltd. All ri
ghts reserved.