Shrinkage estimation in time series using a bootstrapped covariance estimate

Abstract

Concurrent time series with strong inter-series dependence occur in several areas of application where each may be modeled by the same Box-Jenkins ARIMA model. We present a method for explicitly incorporating the inter-series dependence in constructing shrinkage estimators of the model parameters by bootstrapping the covariance matrix of marginal parameter estimates. We also study improved estimation for the scale parameter. We present simulation studies to verify the amount of improvement in terms of expected mean squared error and Pitman nearness. © 1995, Taylor & Francis Group, LLC. All rights reserved.