Mixture type approaches to modelling binary choice
(B Brown and Z Lu)
In modelling binary choice, researchers conventionally assume a linear combination of all regressors to index the distribution function of the dependent variable.
A mixture type model is being proposed and it is being allowed that each explanatory variable index a distribution function. This is called an exclusive regressors mixture model, or ERMM. The maximum likelihood estimation method is suggested for the estimation of the model.
It is possible that there exist more than one local maximum for the objective function in the defined parameter region, and the simulated annealing algorithm is used for maximum likelihood estimation. An extension of this algorithm to situations where parameters are subject to functional constraints is investigated.