Neural Network Estimators of Binary Choice Processes: Estimation, Marginal Effects and WTP

Abstract

Estimation of binary choice models typically require that the econometric model satisfy the utility maximization hypothesis. The most widely used models for this purpose are the binary logit and probit models. To satisfy the utility maximization hypothesis the logit and probit models must make a priori assumptions regarding the underlying functional form of a representative utility function. Such a theoretical restriction on a statistical model without considering the underlying probabilistic structure of the observed data can leave the postulated estimable model statistically misspecified. Feed-forward back-propagation artificial neural networks (FFBANN) provide a potentially powerful semi-nonparametric method to avoid misspecifications. This paper shows that a single-hidden layer FFBANN can be interpreted as a logistic regression with a flexible index function. An empirical application is conducted using FFBANNs to model a contingent valuation study and estimate marginal effects and willingness-to-pay. Results are used for comparison with more traditional methods such as the binary logit and probit models.