Use this URL to cite or link to this record in EThOS:
Title: Self-organising local regressive models for nonstationary financial time series modelling
Author: He, Ni
Awarding Body: University of Manchester
Current Institution: University of Manchester
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
This thesis presents financial time series modelling methods based on the Self-Organising Map (SOM). The principal modelling technique is the local model approach, which aims to fit simple models to localised segments ofthe data. The simple model uses parametric linear regression to describe the variation of the data. The local model approach combines the simplicity of a parametric method with the flexibility of a non-parametric method. Temporal SOM, in addition, integrates temporal context in normal SOM algorithm for a more appropriate analysis of sequential data. The proposed methods are presented in three forms. The first is the temporal SOM with local support vector regressive models. This method uses the recurrent self-organising map to partition the original data space into several disjointed regions and then support vector machines are used to build predictive regressive models. The second approach is the self-organising mixture autoregressive network (SOMAR), the key proposed method in this thesis. It aims to describe and model non-stationary time series by means ofmixture autoregressive local models constructed from topologically clustered time-series segments. The SOMAR network uses the sum (of the absolute value) of autocorrelation coefficients as the similarity measure to identify the winning local autoregressive model. The last approach uses a hybrid system formed by a mixture ofregressive models and economical indicators, which generate either sell or buy signals by monitoring the overbought and oversold status of trading assets. Extensive experiments are presented and the performance ofthe proposed methods are analysed. It is shown that the proposed method, in particular, the SOMAR network, can yield better than global regressive models, econometric time series models, and random walk models.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available