An expert system can be defined as a "knowledge-based system that emulates expert thought to solve significant problems in a particular domain of expertise".
The main characteristics of expert systems compared to neural networks is that they are rule-based. This means that the expert system contains a predefined set of knowledge which is used for all decisions. The system uses the predefined rules to produce results by using inference rules which are coded into the system. Depending on the kind of input and the rules used, expert systems can either be used as quantitative or qualitative tool.
A generic expert system will consists of two main modules: the knowledge base and the inference engine.
The knowledge base contains knowledge of the system regarding the specific domain or area for which it is designed to solve problems or make recommendations. For example, if the system has to work in the financial domain (domain knowledge), the knowledge base will include the specific rules that the system contains - i.e. decisions concerning shares. The knowledge base is coded into the system according to a specific notation which is usually found in the following forms:
Rules
Predicates
Semantic nets
Frames
Objects
The inference engine processes and combines the facts related to the particular problem, case and question, using the part of the knowledge-base which is relevant. The selection of the appropriate data in the knowledge base is performed according to searching criteria. The way in which inference rules are written and applied to the information in the knowledge base vary greatly from system to system and can follow different paths.
Several other modules are usually present in an expert system (e.g. meta-knowledge).
The most important step for the development of financial tools based on expert systems is the acquisition of the domain specific knowledge, consisting of the methods that would be used by a domain expert for making appropriate decisions. This knowledge will normally consist of heuristics which, unfortunately, are extremely difficult to verbalize and the interview process to identify and collect these heuristics can last for weeks.
Unfortunately, information about such systems is generally limited, since disclosure of successful approaches by the financial operators could lead to the loss of competitive advantage, and large sums of money. As a general point, financial operators today tend to prefer neural-networks for real-time forecasting, while expert systems now tend to be used more in other financial fields, where the outcome of the system must be a clear decision - i.e. validating user's credit-card accesses. Expert systems are used in accounting, auditing, decisions in insurance companies, etc.