Wavelet-Based Inference for Long-Memory Processes

A long-memory process may be characterized by its corresponding wavelet variance, an analogue of the spectrum, which decomposes the variance of a process with respect to a variable called scale. In this paper, we derive the variance of the logarithm of the maximal-overlap estimator - a relatively efficient estimator of the wavelet variance. We use this to obtain a weighted-least-square estimator and a test for the long-memory parameter. We show that this weighted-least-square estimator is more statistically efficient than the one based on the wavelet-transform estimator of the wavelet variance. Finally, we apply these estimators and tests to determine the long-memory parameter of the Nile river data, a well-known long-memory process