1. Constructive Data Mining: Modeling Consumers' Expenditure in Venezuela
2. Campos and Ericsson illustrate these model selection procedures by empirically modeling Venezuelan consumers' expenditure. Campos and Ericsson (1999) first distinguish between four pejorative senses of data mining, which are repeated testing, data interdependence, corroboration, and over-parameterization. Data mining, in each of its pejorative senses, is empirically detectable. Campos and Ericsson show how Hoover and Perez's (1999a)* modified general-to-specific modeling strategy can counter each of these senses of data mining in practice. Campos and Ericsson then use PcGets to model Venezuelan consumers' expenditure over 1970-1985. In the selected model, income, liquidity, and inflation determine expenditure in an economically sensible fashion; and that model is robust and has constant, well-determined parameter estimates. Even with relatively few observations, high information content in the data helps counter claims of pejorative data mining. Campos and Ericsson identify two limitations to algorithmically based data mining: the initial general model, and data transformations;Ericsson Campos;systematically examine the concept of data mining in econometric modeling. Their discussion provides support for the automated model selection procedures developed by Hoover and Perez (1999a)* and improved upon by Hendry and Krolzig,1999
3. Using this method, the modeler simplifies an initially general model that adequately characterizes the empirical evidence within his or her theoretical framework. The papers reprinted in Campos, Ericsson, and Hendry (2005) articulate many reasons for adopting a general-to-specific approach. In particular, general-to-specific modeling implements the theory of reduction in an empirical context; and it has excellent model selection abilities, as documented in empirical practice and in Monte Carlo studies of automated general-to-specific modeling algorithms. Studies of those algorithms have also clarified and helped resolve key issues in general-to-specific modeling, including path simplification and path dependence, the interpretation of mis-specification tests, the roles of recursive estimation, the value of model selection procedures, and the distinction between the almost inevitable costs of inference and the relatively low costs of search;Conclusions This paper focuses on general-to-specific modeling-a central method for selecting useful empirical models
4. Professor Hendry's Econometric Methodology;C L Gilbert;Oxford Bulletin of Economics and Statistics,1986