Utilizing artificial neural networks and genetic algorithms to build an algo-trading model for intra-day foreign exchange speculation

Direct Link: https://www.sciencedirect.com/science/article/pii/S0895717713000290

This paper introduces an intra-day price prediction and decision making model for the FX market using artificial neural networks and genetic algorithms taught to identify technical patterns. It is directed toward those speculating on intra-day price movements, particularly HFTs. Three strategies are defined and an optimal neural network structure for the profitable model is provided.

One issue often faced when attempting to create a price-prediction model in the FX market is the non-stationarity of intra-day prices, so some transformation of this price data is necessary to provide a sound prediction. According to the literature reviewed in this paper, typical technical indicators such as a Moving Average provide a solution to the noisiness of the data, and an adjusted moving average calculation to correct for market reversals is used for this study. The data includes EURUSD, GBPUSD and EURGPB tick data, collected over a 70-week period. As determined by a regression, bid/ask rates are highly correlated in therefore only ask rates are used to train the neural networks. The data are then divided into three groups—a training set, a validation set and a testing set—used to observe and evaluate the model’s performance.

Evans et al. use a form of supervised learning, feed-forward ANN, to develop the forecasting model. This type of neural network has a layered structure that is determined by several factors including input/output vectors, the activation function and the training function—the selection for each category is explained in the paper. A genetic algorithm is developed using search techniques imitating genetics and natural selection, and is designed to allow the population of many individual observations to evolve under predetermined rules to maximize fitness of the population. A GA is applied to the ANN to optimize the arrangement of the nodes in the neural network. The resulting trading strategy produced an annualized net profit of 23.3%, which indicates the model’s ability to correctly filter through FX price data and identify repeatable patterns.

Risk Warning: The FXCM Group does not guarantee accuracy and will not accept liability for any loss or damage which arise directly or indirectly from use of or reliance on information contained within the webinars. The FXCM Group may provide general commentary which is not intended as investment advice and must not be construed as such. FX/CFD trading carries a risk of losses in excess of your deposited funds and may not be suitable for all investors. Please ensure that you fully understand the risks involved.