Custom Search

Developed from a set of decision rules (Figure 1), the performance can be evaluated. As a result of reviewing the performance, alteration in the formulation (phase 1), the indicator series (phase 2) or the decision rules (phase 3) may be considered. Repeating phases 1 through 4 is the essence of a decision rule research study. The amount of time spent on repeating these steps to explore for an acceptable or an improved set of decision rules must always be monitored by the researcher. The research study will usually be terminated by considerations of cost versus expected value.

There are an infinite number of relationships from which to choose in using this approach. Selecting the relationships to be tested, and the ability to modify the tests (after reviewing the preliminary results), determines how efficiently the study effort is conducted. Experience in utilizing the empirical approach, familiarity with the market to be studied and the ability to spot unexpected quirks in the data greatly help in conducting a study.

The indicator pattern for which we are searching is defined by the decision rules. The decision rules obtained (Figure 1) were the best results found after performing several tests, but they are by no means the best possible results or the end result of testing all possible relationships. The rules in Figure 1 were simply considered to be suitable for demonstrating this method. The standards of historical performance suitable for accepting or rejecting the use of a set of decision rules will vary greatly according to the objectives of the study.

Decision rule controls

Phase 5 involves establishing controls over the use of the rules. These are necessary because no matter how well the rules work on the historical data, they will not always work in the future. Thus, guidelines must be established as to what to do when the rules don't work. For instance, in the case of the gold trading rules, guidelines might be established for protective sales whenever a five percent loss is exceeded. Similarly, if there were two consecutive losing trades, use of the rules might be suspended until the rules were revised to successfully work with both the original data and the new data. Finally, phase 6 involves putting the rules to work.

Conclusions

A major pitfall of financial and economic model building can be failure to test the model with data. While an historically successful model has no guarantee of working in the future, any model which does not work historically (when based on the same information and assumptions) is certainly a questionable model for use in the future. Similarly, if a theoretically based model cannot pass muster on historical data, then something is wrong with the theory or the expression of the underlying assumptions. These pitfalls can be sidestepped by supplementing research with historical testing procedures. By utilizing statistical standards for validation, empirical models which meet or exceed the same standards as the theoretically based models can be developed. Pattern recognition decision rules and other empirical modeling approaches can be used effectively for financial and economic applications.