Search

A new architectural approach to artificial neural network (ANN) choice modeling is given in order to modify the standard ANN designs. The related studies indicate that a streamlined feed forward ANN trained with a genetic algorithm can provide managers with superior estimates of brand share.

In this paper, a general model of disequilibrium quantity determination under median regression assumptions is combined with weak assumptions on the price adjustment to yield a consistent maximum score estimator. In contrast to Sapra (1986), the supply and demand parameters are separately identifiable and the estimator employs all of the observed data. The price adjustment assumptions can be motivated by the anticipatory pricing model of Green and Laffont (1981). In contrast to the Green-Laffont model, however, our assumptions only require that long-run Walrasian prices have positive probability. As an illustrative application, maximum score estimates of a model of the US commercial loan market are computed and compared to maximum likelihood estimates. Monte Carlo results are also reported. Keywords: Semiparametric; Maximum score; Disequilibrium; Price adjustment; Anticipatory pricing

Tabu search (TS) proved to be an possible alternative to the problematic backpropagation approach used for neural network optimization. Results from the Monte Carlo study reveal that TS generated solutions that were far more superior than the backpropagation method. Moreover, fewer function evaluations are required to find the optimal values.

Backpropagation (BP), the most commonly used algorithm for optimizing neural networks, shows poor results in out-of-sample forecasting. The global search algorithm simulated annealing gives better results compared to BP but is restricted by its point to point search process. The genetic algorithm, which has a process of moving from one population point to another that enables it to discard possible local solutions, gives the best results among the three algorithms.