Custom Search

The very nature of empirical modeling results in certain advantages and disadvantages. If a pattern recognition decision rule simply works consistently, it can be extremely useful, particularly if it either out-performs existing theoretical models, or no theoretical model exists. Empirical models are not subjective; they can be explicitly duplicated and tested by someone else. When tested on the same historical time series data, they will always yield identical results. If the model is tested over a sufficiently long period of time to give statistically significant results, it should acquire a certain degree of confidence from its users.

The same testing standards can be applied to more theoretical models for a direct comparison of the reliability of the models. It is even possible to formulate some hypothetical relationships as pattern recognition decision rules for testing, so that if the test is rejected the relationship can be disproved. If the pattern recognition formula of the hypothetical relationship is accepted, it will not prove the existence of the relationship, but will lend more support to that theory.

Similarly, if a set of pattern recognition decision rules indicate a strategically significant qualitative correspondence between two series, it may suggest the possibility of a causal relationship. Pattern recognition models, however, will not provide anything more than the most superficial understanding of the system being modeled. While it is feasible to find an "acceptable" model with decision rules, it is nearly impossible to find an optimum solution or to know if an optimum solution has been accidentally discovered.

The non-functional nature of decision rule models also provides few clues as to how to improve the performance of the model. Because the decision rule models and their results contain so little insight into the process being modeled, there is little guidance as to how to proceed to improve the model. Typically, improvements are the result of trial and error adjustments rather than systematic, incremental changes. Thus, the inefficiencies inherent in altering and retesting can potentially incur large costs without developing an acceptable model.

Finally, as with all models in finance and economics, past success is no guarantee of future success. With these factors in mind, the researcher can decide whether or not embarking on the development of a pattern recognition decision model is one of value.