XXVI Edition

14-15-16 December 2017"

Explorations in the use of artificial intelligence techniques and short-term econometric forecasting in the €-$ market

Tivegna Massimo, University of Greenwich
Pelusi Danilo, University of Teramo

The paper uses a short-term GARCH multiequation model, estimated between 1999 and 2007 in order to issue Long/Short trading signals for €-$ day-trading, based on its appreciation/depreciation forecasts. Optimal stopping, ie. Stop-Loss (SL)nd Take Profit (TP) are determined by two Artificial Intelligence (AI) techniques: a data-mining version of a Genetic Algorithm (GA) and a neuro-fuzzy combination of a Fuzzy Controller and a Neural Network. Optimality here consists of getting the highest trading profit consistent with the smallest number of trading Drawdowns (DD) and the smallest amount of losses, originating from them. The two AI methods are used to reach this goal. Both AI protocols are trained for 750 trading days, between 2008 and 2010. They are then used in Testing or Trading mode for 782 days, between 2011 and 2013. The combination of econometric forecasting and AI produces a Profits-DD trade-off locus of the expected positive slope: the higher the profits, the higher the DD. The results indicate a far superior performance, for €-$ day-trading, of our AI-optimized rules with respect of a Buy&Hold strategy . The same holds true also for cumulative profits obtained with the use of a set of broad consensus TP and SL values among traders. As expected, Profits are lower in the Training Set than in the Trading Set for both methods; DD are slightly lower in the Trading Set than in the Training Set, as hoped for. But in ratio terms, the two techniques yield substantially comparable results. In a broad conclusion:(a) the combination of Econometrics and AI is a winning strategy; (b) this result is confirmed by the similar results of our two AI protocols.

Area: Exchange rates

Keywords: Foreign Exchange Trading Rules, €-$, news, Neuro-Fuzzy Techniques, Genetic Algorithms, GARCH Econometric Models.

Paper file

University Network