Making Financial Trading by Recurrent Reinforcement Learning
KES '07 Knowledge-Based Intelligent Information and Engineering Systems and the XVII Italian Workshop on Neural Networks on Proceedings of the 11th International Conference
| Hi-index | 0.00 |
For building an automatic trading system one needs: a significant variable for characterizing the financial asset behaviours; a suitable algorithm for finding out the information hidden in such a variable; and a proper Trading Strategy for transforming these information in operative indications. Starting from recent results proposed in literature, we have conjectured that the Technical Analysis approach could reasonably extract the information present in prices and volumes. Like tool able to find out the relation existing between the Technical Analysis inputs and an output we properly defined, we use the Group Method of Data Handling, a soft-computing approach which gives back a polynomial approximation of the unknown relationship between the inputs and the output. The automatic Trading Strategy we implement is able both to work in real-time and to return operative signals. The system we create in such a way not only performs pattern recognition, but also generates its own patterns. The results obtained during an intraday operating simulation on the US T-Bond futures is satisfactory, particularly from the point of view of the trend direction detection, and from the net profit standpoint.