Recurrent Support Vector Regression for a Nonlinear ARMA Model with Applications to Forecasting Financial Returns

Shiyi Chen, Kiho Jeong and Wolfgang Karl Härdle

No SFB649DP2008-051, SFB 649 Discussion Papers from Humboldt University, Collaborative Research Center 649

Abstract: Motivated by the recurrent Neural Networks, this paper proposes a recurrent Support Vector Regression (SVR) procedure to forecast nonlinear ARMA model based simulated data and real data of financial returns. The forecasting ability of the recurrent SVR is compared with three competing methods, MLE, recurrent MLP and feedforward SVR. Theoretically, MLE and MLP only focus on fit in-sample, but SVR considers both fit and forecast out-of-sample which endows SVR with an excellent forecasting ability. This is confirmed by the evidence from the simulated and real data based on two forecasting accuracy evaluation metrics (NSME and sign). That is, for one-step-ahead forecasting, the recurrent SVR is consistently better than the MLE and the recurrent MLP in forecasting both the magnitude and turning points, and really improves the forecasting performance as opposed to the usual feedforward SVR.

Keywords: Recurrent Support Vector Regression; MLE; recurrent MLP; nonlinear ARMA; financial forecasting (search for similar items in EconPapers)
JEL-codes: C45 F37 F47 (search for similar items in EconPapers)
New Economics Papers: this item is included in nep-ecm, nep-ets, nep-for, nep-ifn and nep-opm
Date: 2008-07
View list of references

Downloads: (external link)
http://sfb649.wiwi.hu-berlin.de/papers/pdf/SFB649DP2008-051.pdf (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: http://EconPapers.repec.org/RePEc:hum:wpaper:sfb649dp2008-051

Access Statistics for this paper

More papers in SFB 649 Discussion Papers from Humboldt University, Collaborative Research Center 649
Contact information at EDIRC.
Series data maintained by Janine Tellinger ().