CFE 2015: Start Registration
View Submission - CFE
A1377
Topic: Contributed on Macroeconomic forecasting Title: Adaptive forecasting in the presence of structural change and long memory persistence Authors:  Mohaimen Mansur - The University of Manchester (United Kingdom) [presenting]
Abstract: Spurious relationship and consequent confusion between long memory and structural change have been widely documented in econometrics literature and pose critical challenge for forecasting time series that are subject to them. Failure to appropriately account for one or both the features that are truly present in the data can lead to large forecast errors. Given that it is difficult to distinguish between the two it is desirable to develop forecasting strategies that are robust to presence of both structural change and long range dependence. Such a strategy is proposed where future forecasts are weighted average of past data with the data down-weighting parameter selected adaptively through cross-validation. Detailed theoretical proofs of asymptotic optimality of such forecasts have been provided by considering discrete and continuous structural changes and both weak and strong long memory. Practical usefulness of the adaptive forecasts has been illustrated through an extensive Monte Carlo study and an application to many economic and financial time series of the UK.