The article explores a new way to estimate models for time series data with long memory. Traditional methods have limitations, so researchers are investigating a method called conditional-sum-of-squares (CSS) estimation. While other methods have been proven to work well for long memory data, CSS estimates have not been fully studied. The researchers found that the truncation of the infinite autoregressive representation in CSS estimation may not have a negligible effect, unlike in short memory models. This suggests that CSS estimation may have different properties than previously thought for time series with long memory.