Forecasting and trading the EUR/USD exchange rate with Gene Expression and Psi Sigma Neural Networks

https://doi.org/10.1016/j.eswa.2012.02.022Get rights and content

Abstract

The motivation for this paper is to investigate the use of two promising classes of artificial intelligence models, the Psi Sigma Neural Network (PSI) and the Gene Expression algorithm (GEP), when applied to the task of forecasting and trading the EUR/USD exchange rate. This is done by benchmarking their results with a Multi-Layer Perceptron (MLP), a Recurrent Neural Network (RNN), a genetic programming algorithm (GP), an autoregressive moving average model (ARMA) plus a naïve strategy. We also examine if the introduction of a time-varying leverage strategy can improve the trading performance of our models.

Highlights

► We investigate the use of Psi Sigma Neural Network and the Gene Expression. ► We benchmark their results with five different linear and non-linear models. ► We introduce a time-varying leverage strategy.

Introduction

Neural networks and genetic programming algorithms are emergent technologies with an increasing number of real-world applications including finance (Lisboa and Vellido, 2000, Chen, 2002). However their numerous limitations and contradictory empirical evidence around their forecasting power are often creating scepticism about their use among practitioners. This scepticism is further fuelled by the fact that the selection of each algorithm inputs is based more on trial and error and the practitioner’s market knowledge rather than on some formal statistical procedure.

The motivation for this paper is to investigate the financial forecasting performance of two promising classes of artificial intelligence models, the Psi Sigma (PSI) Neural Network and the Gene Expression algorithm (GEP), which try to overcome some of these limitations and have provided some promising empirical evidence in pattern recognition in other fields of science. This is done by benchmarking their trading performance over the EUR/USD daily fixing by the ECB with a Multi-Layer Perceptron (MLP), a Recurrent Neural Network (RNN), a genetic programming algorithm (GP), an autoregressive moving average model (ARMA) and a naïve trading strategy. Moreover, we explore if the application of a time-varying leverage strategy based on Risk Metrics volatility forecasts can increase the trading efficiency of our models.

The main reason behind our decision to use the EUR/USD daily fixing by the ECB is that it is possible to leave orders with a bank and trade on that basis. It is therefore a tradable quantity which makes our trading simulation more realistic.

As it turns out, both the Psi Sigma and the Genetic Expression algorithms do remarkably well and outperform all other models before and after the application of the trading strategy. It seems that their unique architectures allow them to outperform in our forecasting competition compared with the more ‘traditional’ MLP, RNN and GP models. This empirical evidence can be added to the limited bibliography on the superiority of GEP and Psi Sigma in pattern recognition in Science more generally.

We also note that after the application of time-varying leverage, all our models except ARMA demonstrate a substantial increase in their trading performance in terms of annualised return.

The rest of the paper is organised as follows. In Section 2, we present the limited extant literature relevant to the Psi Sigma and the Genetic Expression algorithm. Section 3 describes the dataset used for this research and its characteristics. An overview of the different neural network models, genetic programming algorithms and statistical techniques is given in Section 4. Section 5 gives the empirical results of all the models considered and investigates the possibility of improving their performance with the application a sophisticated trading strategy while Section 6 provides some concluding remarks.

Section snippets

Literature review

The motivation for this paper is to apply two of the most promising artificial intelligence models, the Psi Sigma Neural Network and the Gene Expression algorithm (GEP) which have been developed recently with the purpose to overcome the numerous limitations of more classic neural and genetic programming architectures and to assess whether they can achieve a higher performance in a trading simulation of the EUR/USD ECB fixing exchange rate. Although the unique architecture and characteristics of

The EUR/USD exchange rate and related financial data

The European Central Bank (ECB) publishes a daily fixing for selected EUR exchange rates: these reference mid-rates are based on a daily concentration procedure between central banks within and outside the European System of Central Banks, which normally takes place at 2.15 p.m. ECB time. The reference exchange rates are published both by electronic market information providers and on the ECB’s website shortly after the concentration procedure has been completed. Although only a reference rate,

Benchmark models

In this paper, we benchmark our Psi Sigma and Genetic Expression models with MLP, RNN and genetic programming algorithms, an ARMA model and a naïve strategy.

Trading and statistical performance

The in-sample statistical performance of our models is presented in Table 4 below while Appendix A.3 describes the statistical measures. For all four of the error statistics retained (RMSE, MAE, MAPE and Theil-U) the lower the output, the better the forecasting accuracy of the model concerned.

We note that GEP presents the best in-sample statistical performance with PSI Sigma having the second most accurate forecasts in the in-sample period. GP, MLP and RNN have the second, third and fourth more

Concluding remarks

In this paper, we apply two promising classes of artificial intelligence models, the Psi Sigma neural network (PSI) and the Gene Expression algorithm (GEP) to a one-day-ahead forecasting and trading task of the Euro/Dollar (EUR/USD) exchange rate using the European Central Bank (ECB) fixing series. Their trading performance is benchmarked with those of two more standard neural network designs, the Multi-Layer Percepton (MLP) and the Recurrent Neural Network (RNN), with a genetic programming

References (25)

  • C.L. Dunis et al.

    Optimal trading frequency for active asset management: evidence from technical trading rules

    Journal of Asset Management

    (2005)
  • C.L. Dunis et al.

    Advanced frequency and time domain filters for currency portfolio management

    Journal of Asset Management

    (2006)
  • Cited by (0)

    View full text