This paper describes how modern machine learning techniques can be used in conjuction with statistical methods to forecast short term movements in exchange rates, producing models suitable for use in trading. It compares the results achieved by two different techniques, and shows how they can be used in a complementary fashion. The paper draws on experience of both inter-and intra-day forecasting taken from earlier studies conducted by Logica and Chemical Bank Quantitative Research & Trading (QRT) group's experience in developing trading models. In evaluating different models both trading performance and forecasting accuracy are used as measures of performance. Rule induction is a method for deriving classification rules from data. Logica's data exploration toolkit DataMariner™, which combines rule induction with statistical techniques, has been used successfully to model several exchange rate time series. An attractive feature of this approach is that the trading rules produced are in a form that is familiar to analysts. We also show how DataMariner™, can be used to determine the importance of different technical indicators and to understand relationships between different markets. This understanding can then be used to assist in building models using other analytical tools. Neural networks are a general technique for detecting and modelling patterns in data. We describe the principles of neural networks, the data pre-processing that they require and our experience in training them to forecast the direction and magnitude of movements in time series.
[1]
Ken-ichi Funahashi,et al.
On the approximate realization of continuous mappings by neural networks
,
1989,
Neural Networks.
[2]
Jadzia Cendrowska,et al.
PRISM: An Algorithm for Inducing Modular Rules
,
1987,
Int. J. Man Mach. Stud..
[3]
A. Harvey.
Time series models
,
1983
.
[4]
Richard C. Thomas,et al.
Rule Induction in Investment Appraisal
,
1988
.
[5]
A. J. Collins,et al.
Introduction to Multivariate Analysis
,
1982
.
[6]
Kurt Hornik,et al.
Multilayer feedforward networks are universal approximators
,
1989,
Neural Networks.
[7]
Patrick Sevestre,et al.
The Econometrics of Panel Data
,
1993
.
[8]
A. J. Collins,et al.
Introduction To Multivariate Analysis
,
1981
.
[9]
James L. McClelland,et al.
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations
,
1986
.
[10]
George Cybenko,et al.
Approximation by superpositions of a sigmoidal function
,
1992,
Math. Control. Signals Syst..
[11]
Christopher K. I. Williams,et al.
On the relationship between Bayesian error bars and the input data density
,
1995
.
[12]
R. E. Kalman,et al.
A New Approach to Linear Filtering and Prediction Problems
,
2002
.
[13]
Ian T. Nabney,et al.
Rule induction for data exploration
,
1991
.
[14]
Ian T. Nabney,et al.
Rule induction in finance and marketing
,
1993
.
[15]
Yoh-Han Pao,et al.
Adaptive pattern recognition and neural networks
,
1989
.