| Abstract [eng] |
This paper analyzes the order execution problem in futures markets. Order execution is one of the problems researched in algorithmic trading field. The main goal of this paper is to research a possibility of decreasing the trading costs using either artificial neural network or decision tree. This is done by employing artificial neural network using multilayer perceptron structure. While decision tree uses CART method. The strategy is implemented and simulated by using 2.5 years of real world historical futures level 1 data. Several distinctive multilayer perceptron architectures are compared with various number of inputs. The results show that proposed method for order execution can sometimes outperform naive strategy. The cost are reduced about 6-7%. The prediction accuracy of price classification is between 49% and 51%. This research could be beneficial for individual investors and other smaller market participants. |
