This paper surveys research on Emulative Neural Network (ENN) models as economic forecasters. ENNs are statistical methods that seek to mimic neural processing. They serve as trainable analytical tools that "learn" autonomously. ENNs are ideal for finding nonlinear relationships and predicting seemingly unrecognized and unstructured behavioral phenomena. As computing power rapidly progresses, these models are increasingly desirable for economists who recognize that people act in dynamic ways with rational expectations. Unlike traditional regressions, ENNs work well with incomplete data and do not require normal distribution assumptions. ENNs can eliminate substantial uncertainty in forecasting, but never enough to completely overcome indeterminacy.
JEL: C3, C32, C45, C5, C63, F3, G15.
Keywords: Emulative neural networks; Dynamic interrelations; Nonlinear forecasting; Perceptron learning process; Multi-layer perceptron model; Learning; Observational indeterminability; Indeterminacies
This paper surveys the significance of recent work on emulative neural networks (ENNs) by researchers across many disciplines in the light of issues of indeterminacy. Financial and economic forecasters have witnessed the recent development of a number of new forecasting models. Traditionally, popular forecasting techniques include regression analysis, time-series analysis, moving averages and smoothing methods, and numerous judgmental methods. However, all of these have the same drawback because they require assumptions about the form of population distribution. Regression models, for example, assume that the underlying population is normally distributed.
ENNs are members of a family of statistical techniques, as are flexible nonlinear regression models, discriminant models, data reduction models, and nonlinear dynamic systems (Sarle, 1994; Cheng and Tetterington, 1994). They are trainable analytic tools that attempt to mimic information processing patterns in the brain (Krishnaswamy, Gilbert, Pashley, 2000). Because they do not necessarily require assumptions about population distribution, economists, mathematicians and statisticians are increasingly using ENNs for data analysis. Not only do they not require assumptions about the underlying population but are also powerful forecasting tools that draw on the most recent developments in artificial intelligence research.
As Hardin (2002) observes in his essay, "Indeterminacy and Basic Rationality," statistical methods, such as neural networks, were developed partly as the product of the ordinal revolution in economics and choice theory. As he points out, because our choices have social and interactive contexts, it would be extremely difficult to construct a theoretical model that is capable of tracing all of these potential and actual responses and interactions. Such models are bound to exhibit fundamental indeterminacy. These indeterminacies are the inevitable product of strategic interactions among rational individuals who understand that their actions, or inactions, are going to be followed by reactions--those of the other participants in the strategic game and those of the environment. In such circumstances, one may find that responses are not similar, let alone unique.
This is especially true when we add time to the discussion. A player may react in ways very different to what was presupposed in response to an unexpected reaction by his opponent. Such models are inherently dependent upon, and sensitive to, initial conditions, which may not permit accurate predictions even for very near-future states (Brown and Chua, 1998; Smith, 1998; Stone, 1989; Bau and Shachmurove, 2002). As Domotor and Batitsky (2002) point out, a creeping amplification of error will eventually wipe out all predictive accuracy. Even the solar system, reported to be the oldest paradigm of a regular, predictable dynamical system, is unpredictable on the time scale of millions of years. …