Custom Search

Modeling with pattern recognition decision rules is an application of the empirical modeling methodology. Patterns may be usefully defined as any repetitive events in a time series. Patterns may be seasonal, cyclical or other recurring variations in the data field series. Pattern recognition decision rules utilize the identification of patterns in a set of indicator series to predict the occurrence of patterns in the forecast series. Through research, the decision maker knows that the occurrence of a pattern in the indicator series has foreshadowed the occurrence of a pattern in the forecast series with a known degree of historical correlation.

Researchers can identify indicator patterns by analyzing historical time series2,3. Pattern recognition indicators are discovered by observing changes in the data series which have occurred at about the same time as the events which the researcher is trying to predict. The series studies can be anything that the researcher believes to be relevant or useful to identifying the events predicted. By testing the patterns as indicators of events statistically, the researcher can confirm or reject the existence of a useful qualitative correspondence between the indicator patterns and the predicted events. The acceptance criteria are based on the degree of reliability of the indicators.

Pattern recognition decision rules contain some common characteristics and some significant differences from another research approach—technical analysis. Technical analysis assumes that there are patterns in market price action which will recur in the future and thus these patterns can be used for predictive purposes. To this extent, technical analysis is a type of empirical modeling.

When models work, they may reflect substantial understanding of the system being simulated.

But typical technical analysis studies have at least two characteristics which exclude them from the status of modeling. First, most technical analysis rules are not well-defined. Without a clear definition, technical analysis procedures cannot be unambiguously utilized by different people. This is especially true when the medium of analysis used consists of charts instead of numerical systems. Thus, identical technical analysis procedures are subject to different interpretations and applications by different users.

Secondly, most adherents of technical analysis tend to test their systems over a very short period of recent history. With so little testing, the statistical validity of the system can rarely be ascertained. Pattern recognition decision rule modeling emphasizes precise definitions and adequate historical testing as well as pattern recognition.