Keep it sophisticatedly simple [chapter]

Arnold Zellner, Arnold Zellner, Hugo A. Keuzenkamp, Michael McAleer
Simplicity, Inference and Modelling  
and prediction, namely s = 1 / 2 gt2, E = mc 2 , PV = RT, maxent, etc. in the physical sciences and the laws of demand and supply, the Fisher equation, no arbitrage conditions, Marshall's competitive industry model, Friedman's and Becker's consumer models, the method of least squares, maximum likelihood techniques, Zayesian analysis, etc. in economics, econometrics and statistics. See Zeitliner (1997) for further discussion of these topics. Further, for many years, I have challenged many
more » ... llenged many audiences to give me an example of a large, complicated model in any field that works well in explaining past data and experience and in predicting new data. As yet, I have not heard of a single one. Certainly, the many large-scale, complicated macroeconometric models of national economies, involving hundreds of nonlinear stochastic difference equations have not been forecasting performance of a sample of complicated macroeconometric models. In general, these studies found their forecasting performance to be unsatisfactory and in a number of instances no better or worse than that of random walk models and other simple, univariate time series models. As many have noted, if a large scale model using many variables and data on them along with much background subject matter information and theory can not perform better in prediction than a simple random walk model, the large, complicated model is probably defective. Rather than complicated, large Stanley steamers that often break down and sometimes explode, we need dependable Model T or A Fords which can be improved with additional work.
doi:10.1017/cbo9780511493164.014 fatcat:22pleymugjew7o53ks35xp3yde