Improving forecasting by estimating time series structural components across multiple frequencies
Wednesday 6 March 2013, 15:00
LT5, Management School
Dr Nikos Kourentzes & Dr Fotios Petropoulos
Management Science Dept, LUMS
Abstract: Good forecasting performance in terms of bias and accuracy is important. Yet, how to select the appropriate time series model to achieve the desired good performance is not straightforward. We propose a novel algorithm that aims to mitigate the importance of model selection. From the original time series, using temporal aggregation, multiple time series are constructed. These derivative time series highlight different aspects of the original data, as temporal aggregation helps in strengthening or attenuating the signals of different time series components. For of these time series the appropriate exponential smoothing is fitted and its respective time series components are forecasted. Subsequently, the time series components from each aggregation level are combined, and the resulting combined components are used to construct the final forecast. We argue that this framework aids the estimation of the different time series components, through temporal aggregation and mitigates the importance of model selection through forecast combination. Empirical evaluation of the proposed framework demonstrates significant improvements in forecasting accuracy and bias, especially for the long-term forecasts.