Finite Sample Comparison of Long Memory Inference under Short-term Noise
Lu Deng · 2014
This paper makes research on two log-periodogram methods of long memory,that is the Geweke,Portert,Hidak(1983,GPH) and Andrews,Guggenberger(2003,AG) method;and takes importance to the short-term noise effects on the estimates and its mechanism under finite sample,including mean and precision of the estimates,test performance.It turned out that although AG method has a better asymptotic property,its finite sample performance under short-term noise is not robust.Mostly,the modification effects of AG estimator are evident only when the short noise has small negative roots.However,in the presence of large negative roots and positive roots,it appears over-modification and under-modification respectively.In addition,bandwidth choice under short-term noise is also studied by comparing the finite sample property in presence of different sample.Experiments found that the property of two methods is very sensitive in small bandwidth,and is not convergent in large bandwidth.The author recommends that the minimum RMSE criterion is a relatively acceptable choice.