Auto Regressive Integrated Moving Average Modeling and Support Vector Machine Classification of Financial Time Series
With cloud computing offering organizations a level of scalability and power, we are finally at a point where machine learning is set to support human financial analysts in FOReign EXchange (FOREX) markets. Trading accuracy of current robots, however, is still hard limited.
This paper deals with the derivation of a one-step predictor for a single FOREX pair time-series. In contrast to many other approaches, our predictor is based on a theoretical framework. The historical price actions are modeled as Auto Regressive Integrated Moving Average (ARIMA) random process, using maximum likelihood fitting. The Minimum Akaike Information Criterion Estimation (MAICE) yields the order of the process. A Support Vector Machine (SVM), whose feature space is spanned by historical price actions, yields the one-step ahead class label UP or DOWN.
Backtesting results on the EURUSD pair on different time frames indicates that our predictor is capable of achieving high instantaneous profit but on long-term average, is only profitable when the the risk-to-reward ratio per trade is around 1:1.2. The result is inline with related studies.
KeywordsMachine learning Support Vector Machine ARIMA FOREX Probability theory
- 6.Ozaki, T.: On the order determination of ARIMA models. J. R. Stat. Soc. Ser. C 26(3), 290–301 (1977)Google Scholar
- 7.Platt, J.C.: Sequential minimal optimization: A fast algorithm for training support vector machines. Technical report MSR-TR-98-14, Microsoft Research (1995)Google Scholar
- 8.Yu, L., Wang, S., Lai, K.K.: Mining stock market tendency using GA-based support vector machines. In: International Workshop on Internet and Network Economics (WINE). LNCS, vol. 3828, pp. 336–345 (2005)Google Scholar