Long-memory analysis of time series with missing values

Abstract

The estimation of long memory is often restricted by missing data. We examine the effects on the estimation of long memory of three simple gap-filling techniques: interpolation, random, and mean filling. Numerical simulations show that the gap-filling techniques introduce significant deviations from the expected scaling behavior for both persistent and antipersistent time series. For persistent time series the interpolation method provides a reliable estimation of long memory for scales longer than the largest likely gap.