Abstract
In this paper, authors apply feed-forward artificial neural network (ANN) of RBF type into the process of modelling and forecasting the future value of USD/CAD time series. Authors test the customized version of the RBF and add the evolutionary approach into it. They also combine the standard algorithm for adapting weights in neural network with an unsupervised clustering algorithm called K-means. Finally, authors suggest the new hybrid model as a combination of a standard ANN and a moving average for error modeling that is used to enhance the outputs of the network using the error part of the original RBF. Using high-frequency data, they examine the ability to forecast exchange rate values for the horizon of one day. To determine the forecasting efficiency, authors perform the comparative out-of-sample analysis of the suggested hybrid model with statistical models and the standard neural network.
Highlights
Predicting time series using statistical analysis started in the 60s years of 20th century
The weakness of ARMA models is the inability to model non-constant variance. As this type of variance is very common in currency pairs, constant volatility is not able to capture some of the basic properties of heteroscedastic volatility present in financial time series such as stochasticity of volatility, volatility clustering, mean reversion and existence of fat tails
The model of artificial neural network based on human neural system is an universal functional black-box approximator of non-linear type [23], [24], [25] which are especially helpful in modelling non-linear processes having a priori unknown functional relations or this system of relations is very complex to describe [26]
Summary
Predicting time series using statistical analysis started in the 60s years of 20th century. The first techniques of machine learning applied into time series forecasting were artificial neural networks (ANN). Today, according to some studies [11] ANNs are the models having the biggest potential in predicting financial time series. Our machine learning application to exchange rates forecasting is novel in two ways – we use the standard neural network hybridized with simple moving averages to form a whole new hybrid model for forecasting. The weakness of ARMA models is the inability to model non-constant variance As this type of variance is very common in currency pairs, constant volatility is not able to capture some of the basic properties of heteroscedastic volatility present in financial time series such as stochasticity of volatility, volatility clustering, mean reversion and existence of fat tails. In [6] Engle suggested the solution by creating so-called ARCH (Autoregressive Conditional Heteroscedastic) models which assume heteroscedastic variance of εt
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
