579 KB, PDF document
Research output: Working paper
Research output: Working paper
}
TY - UNPB
T1 - Modelling Deterministic Seasonality with Artificial Neural Networks for Time Series Forecasting
AU - Kourentzes, N
AU - Crone, S
PY - 2010
Y1 - 2010
N2 - This study explores both from a theoretical and empirical perspective how to model deterministic seasonality with artificial neural networks (ANN) to achieve the best forecasting accuracy. The aim of this study is to maximise the available seasonal information to the ANN while identifying the most economic form to code it; hence reducing the modelling degrees of freedom and simplifying the network’s training. An empirical evaluation on simulated and real data is performed and in agreement with the theoretical analysis no deseasonalising is required. A parsimonious coding based on seasonal indices is proposed that showed the best forecasting accuracy.
AB - This study explores both from a theoretical and empirical perspective how to model deterministic seasonality with artificial neural networks (ANN) to achieve the best forecasting accuracy. The aim of this study is to maximise the available seasonal information to the ANN while identifying the most economic form to code it; hence reducing the modelling degrees of freedom and simplifying the network’s training. An empirical evaluation on simulated and real data is performed and in agreement with the theoretical analysis no deseasonalising is required. A parsimonious coding based on seasonal indices is proposed that showed the best forecasting accuracy.
KW - neural networks
KW - deterministic seasonality
KW - input variable selection
M3 - Working paper
T3 - Management Science Working Paper Series
BT - Modelling Deterministic Seasonality with Artificial Neural Networks for Time Series Forecasting
PB - The Department of Management Science
CY - Lancaster University
ER -