Behavior Analysis of Futures Trading Agents Using Fuzzy Rule Extraction

In this paper, we propose a fuzzy rule-based method that analyzes trading strategies for futures markets. Our purpose is to linguistically understand the behavior of trader agents that have high ability to make a profit. Fuzzy if-then rules are generated from the trade record of the target trader agent. An agent with an unknown trading strategy is used as a target trader agent for our analysis. In the fuzzy if-then rules in this paper, the time series data of stock prices are used as input. The consequent part of fuzzy if-then rules indicates a trading action such as buy, sell, and noaction. The degree of certainty is also included in the consequent part of fuzzy if-then rules in this paper. The number of generated fuzzy if-then rules by the proposed method depends on the number of the antecedent fuzzy sets for each input value. For each combination of antecedent fuzzy sets (i.e., for each fuzzy if-then rule), the consequent action and the grade of certainty are determined from the trading record of the target trader agent. By visualizing the extracted fuzzy rules we can linguistically understand the trading action of the trader agent. In the computer simulations of this paper, we examine the extracted rules from the trading history of the target trader agent. We also examine the trading performance of the extracted fuzzy if-then rules by comparing with that of the target trader agent