The local partial autocorrelation function and its use in forecasting
Monday, April 23, 2018 - 11:00am - 11:30am
The classical regular and partial autocorrelation functions are powerful tools for stationary time series modelling and analysis. However, it is increasingly recognized that many time series are not stationary and the use of the classical autocorrelations can give misleading answers. We introduce two estimators of the local partial autocorrelation function, establish their asymptotic properties, and demonstrates their utility for both simulated and real time series. We also propose using the local partial autocorrelation to select locally the number of past observations that inform a linear time-varying predictor, which can improve the forecasting of locally stationary time series. Our new forecasting method shows convincingly better predictive interval coverage on a set of simulated nonstationary time series, without much loss on stationary series.