In recent decades one witnesses a substantial increase in the interest of econometricians for nonlinear models and methods. This is due to: (1) advances in the processing power of personal computers, (2) increased and successful research on algorithms for fast numerical optimization methods, and (3) the availability of large data sets. One of the nonlinear models that received much attention from applied researchers is a neural network, also known as a neural net. The basic idea behind a neural net is the tremendous data-processing capability of the human brain. Human brains consist of an enormous number of cells, labeled neurons. These neurons are connected and signals are transmitted from one cell to another through the connections. These connections are, however, not all equally strong. When a signal is transmitted through a strong connection it arrives more strongly in the receiving neuron. One may argue that there is a particular weight associated with each connection, which varies with the strength of the connection. Neurons may also receive signals from outside the brains. There are then transformed within the brains and returned to the outside world. The whole structure of signal processing between many (unobserved) cells can be described by a particular mathematical model that is therefore known as an artificial neural network model. For convenience we delete, henceforth, the qualification artificial. A more detailed description of the analogy between the mathematical neural network models and the working of the human brain is given by, e.g., Simpson (1990).